Every market figure here was measured from a stored price panel of 384 NEPSE symbols covering 1997 to 2026, or from the filing named beside it. Where a number could not be verified it is dated and attributed, or it is not here. Nothing in these pages is investment advice.

Fifteen Years, and the Calendar That Made Them

In which a man in Baltimore does something impossible; a coin-flipping contest in 1984 explains why it was not; and a tea shop in New Road turns out to contain the identical error, in Nepali.

I have a rule about people who describe their own successes, and I arrived at it the expensive way. The rule is this: listen carefully to the whole story, thank them warmly, and then go and find out how many other people had the same story and are not talking to you.

Nobody does this. Nobody has ever done this. It is, so far as I can determine after some years of looking, the single most reliably neglected operation in all of finance, and its neglect is worth several percentage points a year to whoever is willing to perform it.

Let me show you what it looks like when you do.

The impossible thing

Between 1991 and 2005 a fund manager in Baltimore did something that had never been done in the history of the American mutual fund industry and has not been done since.

Bill Miller ran the Legg Mason Value Trust, and in each of fifteen consecutive calendar years his fund returned more than the S&P 500. Not on average across fifteen years. In every one of them, separately, without a miss. Morningstar named him fund manager of the decade. The financial press wrote about the improbability of the run in the tone normally reserved for planetary alignments. Business schools built cases around him. He was, for roughly a decade, the proof-of-concept that a human being could beat a market.

And here is the inconvenient part, the part that spoils the easy version of this chapter: he was not a charlatan. I want to be emphatic, because the lazy reader is already reaching for the conclusion that Miller was a fool or a fraud, and the lazy reader is going to have a bad time with this book. Miller had done graduate work in philosophy at Johns Hopkins before he came to money, and he talked about markets in the vocabulary of epistemology rather than the vocabulary of momentum — which, if you have spent any time around the profession, marks a man out the way a violin marks a man out at a wedding band. He held concentrated positions in businesses he had studied for years. He bought Amazon when the consensus held it to be a bookshop with delusions of grandeur, and he held it, and he was right, and he was right for the reasons he had written down in advance.

If you set out to build, in a laboratory, a specimen of the intelligent, disciplined, genuinely insightful investor, you would produce something very close to Bill Miller.

In 2006 the streak ended. In 2007 the fund trailed badly. In 2008 the Value Trust fell roughly fifty-five per cent against an index that fell about thirty-seven, because Miller had loaded into financials — Bear Stearns, Freddie Mac, AIG, Countrywide, a roll call that reads today like a casualty list — on the perfectly coherent reasoning that they were cheap against their own history and that the market was panicking.

Which was the same reasoning that had made him famous.

By the time he left the fund in 2012, the whole of the fifteen-year advantage had been handed back. A man who invested at the start of the streak and simply left his money alone had done no better than a man who bought the index and went fishing. The fishing man, I note, had a better decade.

Fifteen years. One of the longest, cleanest, most scrutinised track records ever assembled by a human being in public with real money at risk — and at the end of it we do not know whether Bill Miller could pick stocks.

I would like that sentence to strike you as obvious by the end of this chapter rather than as a provocation. If it still feels like a provocation, I will have failed, and you will go on losing money in a manner I can predict.

The grid nobody looked at

Begin with the smallest fact, because the smallest fact is usually where the body is buried.

The streak was measured in calendar years. January to December. Fifteen times.

Michael Mauboussin — who was chief investment strategist at Legg Mason, which is to say a man with every professional incentive to polish the record rather than scratch it — noticed something about that measurement which had escaped essentially everyone, Miller included. The streak was an artifact of where you put the boundary. Measure instead on a rolling twelve-month basis — February to January, March to February, and so on through the year — and the run of consecutive wins breaks. There were twelve-month stretches inside those fifteen years when the fund trailed. The calendar simply did not happen to cut them out.

Miller, to his very considerable credit, said so in public. He described the streak as an accident of where the year happens to end.

Now, I want you to sit with this rather than nod at it, because nodding is what the profession did.

The most celebrated investment record of a generation rested on a convention adopted, in a different century, for reasons of tax administration. Nobody chose that convention. Nobody gamed it. It was simply the grid that happened to be laid over the data, and the grid was doing a substantial part of the work that everybody — journalists, allocators, the man’s own investors, and for a while the man himself — attributed to a human being’s judgment.

There is a name for this and if there is not there ought to be, so I will supply one, since it recurs on every page of this book and in every conversation you will ever have about money:

The omitted parameter. The detail on which a claim entirely depends, left out innocently. Not concealed. Not spun. Simply not the sort of thing that survives being told, because a story is optimised for vividness and a diagnosis requires the opposite.

Nobody hid the measurement window. It was printed on every fact sheet in eight point type. It was merely never identified as the variable that decided the answer, and so it sat in plain view for fifteen years while a reputation was constructed on top of it.

So the first question to put to any record, any boast, any anecdote, any research paper, any fund advertisement, and (I insist on this) any account you give of your own life, is not whether it is true.

It is: which number would have settled this, and why is it absent?

Buffett, at Columbia, 1984

In 1984 Warren Buffett gave a talk at Columbia to mark the fiftieth anniversary of Graham and Dodd’s Security Analysis. It was printed as “The Superinvestors of Graham-and-Doddsville,” and I hold it to be the most sophisticated piece of statistical reasoning ever performed by a man who makes a point of disliking statistics.

He opens by conceding his opponents’ case with more vigour than his opponents had managed. Imagine, he says, a national coin-flipping contest. Two hundred and twenty-five million Americans each wager a dollar and call a flip; losers drop out, winners take the pot and flip again. After twenty mornings of this there will be something like two hundred and fifteen survivors, each having turned one dollar into roughly a million.

And these survivors, Buffett observes with evident pleasure, will write books. They will give seminars titled How I Turned a Dollar into a Million in Twenty Days Working Thirty Seconds a Morning. Some of them will corner reluctant academics at parties and demand to know how, if it cannot be done, they did it.

(I have met the Nepali version of this man. He does not write books. He does something worse: he starts a Viber group.)

Then Buffett turns, and the turn is the entire point of the essay.

Suppose, he says, that of those two hundred and fifteen winners, forty came from the same small village. You would not shrug and observe that somebody has to win. You would go to the village. You would look at the water, the schooling, the local theory of coin-flipping. The concentration is the signal. And he then argues — this being the real content of the speech — that a wildly disproportionate share of the outstanding investment records of his generation came out of one intellectual village, the students and readers of Benjamin Graham, men who agreed about almost nothing except a single procedure: buy a business for meaningfully less than a conservative estimate of what it is worth, and treat the gap as your protection against your own stupidity.

That procedure is the margin of safety, the oldest surviving idea in this business, and we will spend a great deal of Part Three on it — including the awkward question of what a margin of safety can possibly mean in a market where the audited accounts turn up five months after the year has ended.

But it is not Buffett’s conclusion I want here. It is the shape of his argument. In 1984, in front of a room of academics, without writing down a single formula, he performed the operation that nobody performs.

The two questions, and the one that is never asked

When a man presents a record intended to demonstrate his skill, there is an instinctive question. It is the wrong question. It is always the wrong question, and I have watched intelligent people ask it in boardrooms, on television, and in their own heads at three in the morning.

The instinctive question is: is this consistent with him being skilled?

The answer is nearly always yes. A skilled manager would indeed beat the index fifteen years running. A skilled investor would indeed turn six lakh into forty. The evidence fits the hypothesis, the mind registers a small satisfying click, and the file is closed. This is the mechanism by which a country fills with oracles.

The correct question has two halves, and the second half is where the money is:

How likely is this record if the man is skilled? And how likely is this same record if he is not?

The ratio between those two numbers is what a record is worth. Large ratio, real evidence. Ratio near one, nothing at all — regardless of the size of the number attached, the sincerity of the teller, or the quality of his suit.

Statisticians have a word for that ratio. You do not need the word. You need the habit, which is to demand the second number: the one describing the people who are not in the room, who did not publish, who do not appear on the panel, and whose absence is the sole reason the survivor looks miraculous.

I call it the denominator, and this book is named after it, because I have come to believe that the whole of investment competence is downstream of the willingness to compute it.

The denominator is harder to get. It is unflattering. It requires data about failure, which nobody collects, because failure is quiet and success is loud and the ratio of loudness to quietness is roughly the ratio at which the world misleads you.

So let us compute it. For Bill Miller. Now.

What chance predicts

Treat beating the index in a given year as a coin flip. This is generous to the manager — after fees, the historical share of American diversified equity funds beating the S&P 500 in a given year has typically run below half rather than at it — but take the fair coin as a starting point and let the manager have the benefit.

Fifteen consecutive wins is one half to the fifteenth power. One in 32,768.

Which looks like a miracle until you ask the second question, and the second question is: how many people were flipping? There were on the order of a thousand diversified American equity funds through that era. A fifteen-year streak could have begun in any year — call it thirty overlapping windows across the post-war record. Thirty thousand fund-windows in which a streak was available to occur.

At a fair coin, chance alone predicts 0.92 such streaks in the history of the American fund industry.

Exactly one occurred.

Read those two lines again, slowly, because between them sits the reason that essentially every conversation you have ever had about a successful investor was uninformative.

Now let me anticipate the objection, since I would rather you distrusted this calculation in the right places than in the wrong ones. Fund returns are not independent coin flips; a manager with a persistent style wins and loses in correlated runs, and correlation makes long streaks more likely than my table says, not less — so the objection cuts against me, and I raise it myself because that is what one does. The count of funds and windows is an estimate. And the strongest form of the objection is that the fifteen years are not the only evidence about Miller: there is also Amazon at thirty dollars, bought for stated reasons, held through a ninety per cent drawdown. No coin has ever done that.

Fine. The headline survives all of it and is not sensitive to the assumptions:

A record that chance produces about once, occurring once, is not evidence. It is precisely what the absence of skill predicts.

And an entire industry of people whose professional function is inference — I mean this literally, it is what they are paid for — treated it as the opposite, for fifteen years, on the cover of magazines.

The same error, in Nepali

There is a man in New Road who turned six lakh rupees into forty lakh in seventeen months, and I have heard him account for it perhaps two dozen times.

I call him Ramesh dai, since he is not one man. He is a composite of nine or ten people I have sat with and a considerably larger number I have overheard from the next table, which in Kathmandu amounts to nearly the same thing. You have met him. You liked him — this is important, he is genuinely likeable, and if he were not likeable he would not be dangerous. He is generous with tea. He remembers the state of your mother’s knee. He possesses that rare and commercially valuable gift of appearing wholly absorbed in what you are saying during the interval in which he waits to say the thing he has been waiting to say.

The story opens in the spring of 2077 with the exchange shut. Not slowed. Shut. NEPSE closed on the twenty-second of March 2020 and did not reopen for fifty-one days, transacted for a single session on the twelfth of May, and closed again for another forty-seven. In the whole of that year it held one hundred and eighty-one sessions against a normal year’s two hundred and thirty.

He bought into that silence. Six lakh, everything he held outside land. He names four or five companies. Then both hands open, palms upward — the universal gesture of a man describing something done to him rather than by him — and: and then it went up.

It did. On the second of January 2020 the index closed at 1,166.21; on the eighteenth of August 2021 at 3,198.19. A hundred and seventy-four per cent in five hundred and ninety-four days.

Now. Buffett’s second question. How likely is that outcome if the man has no skill whatever?

Unlike almost everything else ever asserted about this market — and I include in that the entire output of the evening television panels, about which I will have things to say in Chapter Seven that I have been looking forward to — this is computable. I have the data. Let us have the denominator.

One hundred and eighty-five names, and not one of them fell

I took every symbol in my price database that traded through the window from the reopening on the twenty-ninth of June 2020 to the peak on the eighteenth of August

• One hundred and eighty-five names, excluding the index series and anything

too thin to carry a continuous price. Then I asked the least sophisticated question available: what did each of them do?

Not one of a hundred and eighty-five listed securities lost money over fourteen months. The worst company on the entire exchange returned two per cent. The median returned three times your capital, beating the index, because the index leans on the large banks and the large banks were the laggards.

One caveat, offered before you find it yourself: the window opens at a reopening after a national shutdown, which is close to the most flattering start date available anywhere in the series. I chose it because it is the date the story implies. Start in January 2020 instead and a handful of names do fall. The argument survives either way, and I would rather hand you the knife than have you find it.

Consider what a market of that description does to the concept of stock selection.

In that window, selection could not fail. There was no wrong answer. The man who threw a dart, the man who chose by the length of the ticker, and the man who spent four hundred hours on audited accounts all made money — and the third had no means whatever of knowing whether the four hundred hours contributed anything, because there existed no sample of his own decisions that lost.

One variable mattered. How much of the dispersion a man exposed himself to.

Which is to say: concentration. Hold the word.

Building five thousand men with no thoughts in them

Suppose that at the reopening, five thousand people in the Kathmandu valley each committed six lakh rupees. Suppose further — and this is the cruel clause, and I put it in deliberately — that not one of them had an idea in his head. They chose names off a screen. One bought a hydropower company because the river runs past his village. One bought a bank because his sister works there. One bought a microfinance institution because the ticker had four letters and he is fond of four-letter words.

How many of them end up with Ramesh dai’s story?

This is not a rhetorical question and I am not making a point. I have the actual returns of all one hundred and eighty-five names, so I can build the thoughtless men and count them, which is what I did. Two hundred thousand random equal-weighted portfolios at each level of concentration, asking how often one reached 6.67×, which is what six lakh becoming forty lakh requires.

I had expected a flatter table, and I will tell you exactly how wrong I was because it is instructive. In an earlier draft of this chapter I asserted that a tide of that size would manufacture “several hundred” men with a sixfold return, and I intended to leave the assertion standing, unaccompanied, in the confident tone you have been reading for eight pages. At five names the true figure is twenty-eight. I was out by a factor of twelve, in my own field, on my own data, in the direction that flattered my argument.

That is what assertion is worth. Mine included. Especially mine.

Now read what the table actually says. If he held one company, three hundred and fifty thoughtless men in a single city share his outcome and his story is worth nothing whatsoever. If he held ten, not one thoughtless man in five thousand reaches him, and the story becomes remarkable evidence of something real.

Six hundred-fold, across that column. The entire evidential content of the anecdote rests on one parameter.

And the story never says.

Two dozen tellings. Four establishments. Five years. I have heard the companies named and renamed, the reasoning explained, the fear of others described, the moment of purchase dramatised with hand gestures. I have never once heard how the six lakh was divided.

It is the only number that would settle the question. It is the only number nobody mentions. Not because it is hidden — because it is boring, because it carries no narrative weight, and because he does not know it is the important one either.

Which is Bill Miller’s calendar. Precisely. In Nepali.

A reputation resting on a parameter that was never concealed and never identified, in Baltimore and in New Road, in a fund with billions under management and in a tea shop with four plastic chairs. The defect is not a matter of sophistication. It is not a matter of development. It is structural, it is in the shape of narrative itself, and no amount of financial literacy training will touch it. The only defence is the habit, and the habit is one sentence long: which number is missing?

Bernoulli, and why waiting will not help you

The reader will now object — I can hear him, he has been shifting in his seat for two pages — that this is a problem of insufficient data, and that a man with a longer record could be judged properly.

Very well. How long?

In 1713, eight years after his death, the Ars Conjectandi of Jacob Bernoulli was published at Basel, and in it he asked a question nobody had put in that form before: how many observations does a man require before he may claim to know the composition of an urn he cannot see into?

He took a specific case. An urn of white and black tokens in a ratio of three to two. He wanted to be morally certain — his phrase, defined as odds of a thousand to one — that his estimate lay within a fiftieth of the truth. He turned the crank.

Twenty-five thousand five hundred and fifty trials.

Bernoulli was embarrassed by the number. It exceeded the population of Basel. It implied that moral certainty about a rather simple question demanded more observation than a life conveniently affords — not the conclusion a man hopes for at the end of twenty years of work, and among the reasons the book sat in a drawer while he was alive to be disappointed by it.

I regard it as the most useful number in finance and would have it painted above the door of every brokerage in this country, in gold, in Devanagari.

Bill Miller had fifteen annual observations. Ramesh dai has one.

And you — I am talking to you now, the person holding this book, who has a demat account and some opinions — will make perhaps a dozen genuine decisions a year. Decisions about what a business is worth, not trades. Across a forty-year career that is under five hundred, and they will not be independent, because they will be made by the same man in the same market under the same conditions with the same blind spots.

Bernoulli needed twenty-five thousand independent draws to settle an urn with two colours in it. You will get five hundred correlated ones to settle a question considerably harder than an urn.

If you intend to learn anything about your own judgment, it cannot come from your results. There will never be enough of them. Not for you, not for Miller, not for anyone who has ever lived. It must come from somewhere else, and the remainder of this section is a long argument about where.

Thirty-four years, for those relying on time

There is a consolation people reach for at exactly this moment, and I would rather remove it early than let you carry it for two hundred pages. The consolation is that markets recover, so a sufficiently long horizon dissolves the problem.

On the twenty-ninth of December 1989 the Nikkei 225 closed at 38,915.87. Japan was then the second economy on earth and the most admired one; American business schools taught Japanese management technique; the land beneath the Imperial Palace was said, in a comparison everybody enjoyed making, to be worth more than California.

The Nikkei next closed above that level on the twenty-second of February 2024.

Thirty-four years and two months.

A Japanese saver who bought the index at thirty-five and held with perfect discipline through every intervening crisis returned to his starting point at sixty-nine. He did nothing wrong. He did not overtrade, did not panic, did not chase tips from a group chat. He executed, flawlessly, the one instruction every book gives — and it consumed his entire working life to arrive back where he began.

I do not raise Japan to frighten you out of equities; I own equities, I intend to go on owning them, and the alternative in Nepal is a deposit account that has historically paid rather less than inflation in the years you most needed it. I raise Japan because “it always comes back” is an empirical claim, and the empirical record contains a thirty-four-year counterexample in a rich, stable, technologically formidable country with excellent accounting.

Nepal’s own index peaked on the eighteenth of August 2021 at 3,198.19 and has not seen it in the five years since. It has approached three thousand three times and failed three times.

Five years is not thirty-four. But neither was year five in Tokyo.

The sequel, which is the only test that matters

There is one experiment that would cut through all of the above, and in Nepal I can actually run it, which is the advantage of a small market with a complete price history and the reason I built the database in the first place.

Skill, if it is skill, recurs. That is what the word means. A property that appears once and never again was not a property; it was an event. So: did the companies that performed best during the boom go on to perform best afterwards?

I took the hundred and fifty-five names for which I hold a full record on both sides of the peak, ranked them by boom performance, and measured what they did over the following five years — the eighteenth of August 2021 to the end of July 2026.

Minus nought point nought four five. Zero, with a rounding error stapled to it.

The companies that went up fifteenfold and the companies that barely twitched proceeded to do the same thing, which was to shed about a fifth of their value over five years. Thirteen of the boom’s best fifteen sit below their peak today. The second-best performer of the entire boom, a finance company that returned 12.37×, now trades at 0.44× its peak.

This is Miller’s 2008, arrived at from the opposite end. There, one man’s method stopped working. Here, an entire market’s worth of winners turn out to have carried no information about the future at all — not positive, not reliably negative, nothing. A rank correlation of −0.045 is what you get from ranking companies by the alphabet.

Two men in a courtyard

The defect is neither modern nor financial, and Nepal supplies the cleanest illustration of it I know anywhere.

Bhimsen Thapa governed this country for thirty-one years. He rose in the disorder following the assassination of Rana Bahadur Shah in 1806, in a season of purges — the Bhandarkhal killings, some ninety rivals removed — and held the office of mukhtiyar longer than any man before or since. He modernised the army. He built the tower at Sundhara. He kept the East India Company out for a decade and then did not, and after the war and the treaty his position eroded; in 1837 his enemies brought a charge against him involving the death of a child. Imprisoned, released, imprisoned again. In 1839 he opened his own throat with a khukuri in a cell and took nine days to die. His body was dragged through the streets.

Jung Bahadur Kunwar, on the night of the fourteenth of September 1846, was one of several armed and ambitious men standing in the Kot courtyard when an argument became a massacre. Several dozen of the kingdom’s nobility died there within hours. Jung Bahadur did not. By morning he was mukhtiyar. He took the title Rana, made the office hereditary, and his family governed Nepal for a hundred and four years.

Open the histories and you will find two entirely different species of man. Bhimsen is tragic: over-reaching, undone by hubris and British artillery. Jung Bahadur is a force of nature, possessed of precisely that genius for survival which the founding of a dynasty is understood to demonstrate.

They used the same instrument. A courtyard, an accusation, a night, and a willingness. Both competent. Both ruthless. Both exactly as intelligent as the position required. One of them was standing in a different part of the room when the shooting started.

We have no vocabulary for a man who was capable and unlucky. We have only the word for what happened to him.

I am not claiming Jung Bahadur was merely fortunate. He was plainly formidable, and formidable men do survive courtyards at better than the base rate. I am making the narrower and considerably more irritating claim: that the historical record contains no mechanism whatever for separating the formidable man who survived from the formidable man who did not, because the second one generates no record. The officer of equal capacity cut down in the first ten minutes, appearing in no chronicle, is unavailable to us as evidence — permanently, by construction, for as long as there are histories.

This is the condition under which the folklore of every stock market on earth has been assembled, including the one you are about to put your money into.

The gap between a fund and the people in it

One more foreign fact, and it is the one with the most immediate cash value, so if you have been skimming, stop.

Peter Lynch ran Fidelity Magellan from 1977 to 1990 and compounded at about twenty-nine per cent a year. This is the finest sustained record in the history of public mutual funds and it makes Miller’s streak look like a warm-up.

The investors in Magellan did substantially worse than Magellan did.

There is a folkloric version of this claim which asserts that the average Magellan investor lost money, and I do not repeat it, because I have never seen the study it supposedly rests on and neither has anybody else who has looked. The rigorous version comes from Morningstar, which for years has published a comparison between a fund’s reported return and the return actually earned by the money inside it, weighted by when that money arrived and departed. Across the industry the gap runs on the order of one to one and a half percentage points a year depending on the period examined, and it is reliably negative. Money arrives after good years. Money leaves after bad ones.

A fund’s return is a property of the fund. An investor’s return is a property of the investor’s behaviour. They are different numbers, and only the second one buys anything.

Which means that even in the presence of demonstrable world-class skill — and Lynch’s skill was real, nobody seriously disputes it — the typical participant converted that skill into something markedly smaller, by doing the most natural thing available to a human being, which was to feel more confident after gains and less confident after losses.

Hold that for the next fifteen chapters. It is the entire reason this section comes before the analysis, and the reason I will not let you at the valuation models until you have read it.

The chronicles are recopied

I promised to come back to the shortening list, and I keep my promises within a chapter, if not always across one.

The first time I heard Ramesh dai’s story, in 2079, it contained seven names. By 2081 it was five. This year it was four, and one of the four was — I am fairly confident, though of course I cannot prove it, which is itself the point — not among the original seven at all, being a company that did extremely well in 2081, two years after the events under description.

This is not deception and it would be lazy to call it that. It is what memory is for. Memory is not a filing cabinet; it is a workshop in which a usable self is manufactured and continuously repaired, and every retelling is another pass of the file, keeping what is load-bearing and shedding the rest — where “load-bearing” is defined, inescapably and circularly, as the part that turned out well.

Nepal offers an unusually clean illustration at national scale. The vamshavali, the dynastic chronicles, were recopied and extended by generation after generation of court scribes, and they were never neutral records. Each recopying was performed under a particular arrangement of power, and each one made the arrangement then obtaining look like the necessary consequence of everything preceding it. Lines were straightened. Inconvenient claimants thinned. By the time a chronicle reached its final form, the accession of whoever presently sat on the throne had acquired the quality of inevitability — and not one scribe had told a single lie to produce that effect.

Run that process on your own history for five years and you will have a man who honestly recalls having bought only the winners. He is not deceiving you. He was deceived first, by an operator of exceptional competence: himself.

The remedy is the first practical instruction in this book. Write the reasons down — not the positions, the reasons — dated, somewhere you cannot revise them. Then leave them alone for two years.

The experience of reading them afterwards is difficult to convey to anyone who has not done it. It is the closest thing available to an introduction to the man you actually were, as distinct from the man you have since become on his behalf. I keep such a file. I open it rarely, and never willingly. It has improved my judgment more than every book on my shelf, and the mechanism is not insight. The mechanism is humiliation, administered at intervals, by a witness who cannot be argued with.

The verdict, which is not the satisfying one

I have spent a chapter dismantling two men’s accounts of themselves and it would be cheap to stop there, because in Ramesh dai’s case the arithmetic refuses to convict him and I am obliged to say so.

Take the six lakh from the reopening in June 2020 to the end of July 2026 — six full years, through the boom and the whole of the long erosion after it — and put it in three places.

He did not put his money in a deposit. And that single decision — taken for reasons that were probably wrong, in a market he did not understand, at a moment when it felt frightening to everyone including the people who understood it — is worth more than every refinement of technique in the remaining five hundred pages of this book. He showed up. He bought businesses when businesses were cheap. He did not sell them all.

So the verdict is not that he was a fool who got lucky. It is stranger, and more useful, and I have come to think it is the most important sentence in the chapter:

He was right about the large thing for reasons he cannot articulate, and he believes he was right about the small thing, which he was not, and it is the small thing he tells people about.

The transmissible part of his experience — own businesses, buy them when others are frightened, do not sell everything — is the part he considers too obvious to mention. The part he actually teaches, the four or five clever names, carried no information at all.

Miller is the identical case at a higher altitude. The transmissible part of his record is the method: buy what the market has priced for permanent decline, and do the work required to know whether the decline is permanent. The part that got taught was the streak.

This is the ordinary shape of financial wisdom everywhere on earth. The valuable lesson is present in nearly every success story, and it is never the moral the teller draws.

A confession, without which this is intolerable

I am going to spend five hundred pages pointing at men and describing errors they cannot see, and to do that from a position of claimed innocence would be insufferable. So.

In 2078 I owned a hydropower company I had thought about carefully, using methods I still hold to and will teach you in Chapter Thirty-Seven. I had read the licence. I had read the power purchase agreement, which almost nobody does. I knew the remaining licence years, I had worked out what the plant ought to earn across a normal hydrological cycle, and I carried a number in my head that sat comfortably above the price.

The share rose a great deal. I was extremely pleased with myself.

And the reason it rose had nothing to do with the licence, the agreement, the hydrology, or the number. It rose because in 2078 everything with the word jalvidyut in its name rose — including three companies with no operating plant at all, and one whose entire reported profit consisted of interest earned on its own unspent public offering sitting in a fixed deposit.

I was right, and I was right for reasons that were not operative. My analysis and my outcome were both correct and entirely unrelated, like two men arriving at the same wedding from different districts.

Here is the disagreeable part, and I have thought about it more than is healthy. Had the share fallen, I would have concluded the analysis was faulty, gone hunting for the error, and certainly found one — there is always one. I would have learned something false. Instead it rose, I concluded the analysis was sound, and I learned something equally false in the opposite direction.

There existed no available outcome from which the true thing could have been learned. None. The year offered me two roads and both of them led away from the truth, which is that in 2078, analysis and outcome had come entirely uncoupled.

That is not an unusual predicament. In this market, in most years, for most people, it is the ordinary one. And a man who cannot distinguish between being right and being paid will spend a career being trained — patiently, thoroughly, and at considerable expense — by noise.

Bill Miller, it should be recorded, came back. He built a new firm, made an enormous concentrated bet on Amazon and another on Bitcoin, and by the early 2020s was rich again and being interviewed again, and the interviews described a man vindicated.

Perhaps he was. He may well be one of the finest investors of his generation; on balance I lean toward thinking he is, though I notice that I have no method for this and am simply reporting a feeling.

But observe that we are no nearer to knowing than we were in 2005, that the evidence has now moved twice, and that on both occasions the verdict of the entire profession moved obediently with the most recent observation.

Ramesh dai bought me tea last Falgun. He is purchasing a second property in Bhaisepati, and explained that the market is shortly to break out and the timing is therefore excellent.

Then he said the thing he always says at the end, which he delivers as a joke, laughing, wholly unaware that it is a confession:

”Ke garne — market ma ta sabai expert hunchha.”

What to do. In the market, everyone is an expert.

He is right. That is the difficulty.

The Newspaper from Next Year

In which 118 people are handed tomorrow’s headlines and half of them lose money anyway; a Japanese-scale disaster produces a record high in Kathmandu; and we establish that knowing what will happen is a different thing entirely from knowing what it is worth.

Every investor has had the same fantasy, and I have never met one who admits it, which tells you something about investors.

The fantasy is the newspaper. Somebody hands you tomorrow’s front page — or next month’s, or next year’s — and you take it into a quiet room and become unimaginably rich. It is the purest form of the thing everybody in this business is actually chasing, stripped of the professional vocabulary: not analysis, not process, not risk management. Foreknowledge.

In 2023 two men decided to find out what would actually happen, and I regard the result as the single most useful experiment ever run on the human relationship with markets.

The crystal ball, and what people did with it

Victor Haghani is a name worth knowing before we go further, because he has stood on both sides of this question. He was one of the founding partners of Long-Term Capital Management — the fund built around two Nobel laureates, which returned something over forty per cent a year for three years and then, in the autumn of 1998, lost almost everything and required the Federal Reserve Bank of New York to convene a rescue. If there is a man alive with a personal, expensive, unforgettable education in the difference between being right and surviving, it is him.

Haghani, with James White, ran the following experiment. They recruited a hundred and eighteen young adults trained in finance — students at good programmes, junior professionals at real firms, the sort of people who would go on to manage other people’s money. Each was given a stake and fifteen trading days drawn from the period 2008 to 2022, and before each day they were shown the front page of the Wall Street Journal from thirty-six hours in the future.

The actual headlines. Before the fact. They could see the news before they had to trade it. They were permitted to buy or sell the S&P 500 and long-dated US Treasury bonds, and to use leverage.

Consider what you would expect. These are people who know what finance is. They have been handed the fantasy, in laboratory conditions, with a real cash prize.

About half of them lost money. Roughly one in six went bust entirely.

Not underperformed. Bust — wiped out, stake gone, in fifteen days, holding tomorrow’s newspaper.

I want to give this its full weight before explaining it, because the explanation tends to defuse the shock and the shock is the valuable part. If you had asked me, before I read this, what fraction of finance-trained adults with thirty-six hours of perfect foresight would lose money, I would have guessed something under five per cent, and I would have been wrong by a factor of ten. In my own field. About people very like me.

Two ways to be right and still be finished

The experiment failed in two distinct places, and the distinction between them organises everything that follows in this book.

The first failure is that the news does not tell you the price.

A headline announcing that unemployment rose is not a signal to sell. It is a signal to sell only if unemployment rose by more than the market already expected, and the market’s expectation is not printed anywhere on the front page. Prices do not move on events. Prices move on the difference between events and what was already believed about them — and that second quantity, the belief, is invisible, unrecorded, and precisely the thing a newspaper cannot contain.

In August 2022 the participants would have read a headline saying American inflation had come in at 8.5 per cent, and reasoned, sensibly, that this was bad. The S&P 500 rose more than two per cent that day — because 8.5 was below the 8.7 everybody had feared. They had the fact and were missing the denominator. Again the denominator.

The second failure is worse, because it afflicted the people who got the direction right.

Haghani’s participants who correctly called the market’s move still, in many cases, destroyed themselves, and they did it through sizing. Given a view they believed in, and access to leverage, they bet too much. One adverse day inside a correct multi-day view was sufficient to remove them from the game before the view paid off.

This is the oldest lesson in speculation and it is never learned in advance: being right about direction and being right about size are separate skills, and only the second one keeps you alive. We will do the arithmetic of it properly in Part Five, including the uncomfortable fact that in Nepal the growth-optimal bet size is not merely hard to compute but not computable at all, for reasons specific to this market that I will demonstrate rather than assert.

For now, note the shape: a hundred and eighteen people were handed the thing every investor wishes for and a sixth of them were destroyed by it. Foreknowledge is not the constraint. It never was.

The pandemic, for those who think the experiment was artificial

If a laboratory does not persuade you, take the largest natural experiment of our lifetime.

It is January 2020. Somebody hands you a document describing, accurately, the next twenty-four months. A respiratory virus will spread from Wuhan to every country on earth. Millions will die. Governments will shut their economies by decree. Air travel will approach zero. Unemployment in the United States will reach levels not seen since the Depression, in a matter of weeks rather than years. Entire industries — cinemas, cruise ships, aviation, hotels — will simply cease to operate.

You have this document. What do you do with your portfolio?

Everyone sells. I have put this question to a good number of people and the answer has never once varied. Only a lunatic holds equities into a global pandemic.

The S&P 500 bottomed on the twenty-third of March 2020. By the end of 2021 it had more than doubled from that low, and finished the period at an all-time high. The NASDAQ did better. A man who sold in January 2020 on perfect information about the pandemic and returned when it felt safe — which is to say when the vaccines were distributed and the newspapers had calmed — bought back substantially higher than he sold.

The document was accurate. The document was ruinous.

And it is not a one-off. In June 2016, if you had known the Brexit referendum result the night before, you would have sold British equities. The FTSE 100 fell about three per cent on the day, fell again the next session, and was back above its pre-referendum level inside a week — because sterling had collapsed and most of the index’s earnings are in foreign currency, so a national humiliation arrived at the index as a translation gain. A man who had shorted it on perfect information would have been right for two days and wrong for a decade.

In November 2016, index futures fell sharply overnight on the American election result and then the market rose for three years. In each case the event was correctly foreseen and the price went the other way, because price is not a function of events. Price is a function of events minus expectations, and you were only given the events.

Now hand the newspaper to a man in Kathmandu

It is the twentieth of April 2015. You have forty lakh rupees in NEPSE, a reasonable book for a serious person, and somebody gives you the next six months.

The headlines say this. On the twenty-fifth of April there will be an earthquake of magnitude 7.8 centred in Gorkha. Around nine thousand people will die. Whole districts will lose most of their houses; Kathmandu will lose temples that have stood for five centuries. There will be a second large shock in May. The stock exchange will close and remain closed for thirty-one days. And then, in September, before the country has finished counting its dead, the southern border will close, and fuel will vanish, and people will cook on firewood in the capital city, and a schoolteacher will queue eleven hours for a cylinder of gas.

You hold the paper. What do you do with the forty lakh?

Everyone sells. Of course everyone sells.

Here is what happened.

The market closed on the twenty-third of April 2015 at 938. It reopened on the twenty-fourth of May at 910. Three days later, on the twenty-seventh of May, it touched 838. That was the bottom.

Eighty-four days later, on the nineteenth of August 2015, the NEPSE index closed at 1,192.

Sit with that number, because it is not merely a recovery. The previous high — the peak of the great boom of 2008, the level that had stood untouched for seven years while a generation of Nepali investors gave up on shares and bought land instead — was 1,175.4, set on the thirty-first of August 2008.

Four months after the worst earthquake in eighty years, NEPSE broke its seven-year high. It reached 1,205.84 on the fifteenth of September.

Then the border closed. From the twenty-second of September 2015 to the eighth of February 2016 — the blockade, the worst five months the Nepali economy has endured in my lifetime, no fuel, no medicine, no cooking gas, factories dark, the index went from 1,156.1 to 1,263.8.

Up nine point three per cent.

And it did not stop. By the twenty-seventh of July 2016 the index stood at 1,881.45: a gain of a hundred and twenty-four per cent from the post-earthquake low, straight through a national catastrophe and an economic siege.

You had tomorrow’s newspaper and it would have ruined you.

Why it happened, which is the useful part

The lazy reading is that markets are irrational. I dislike this phrase, and not because I think markets are rational — I think the word is simply doing no work. It is what people say when they wish to end an inquiry while sounding as though they have concluded one.

The actual mechanism is specific and instructive, and it is the reason I have put this chapter second.

You were given perfect information about events. You then made a forecast about prices. Between those two operations sits a step you did not notice yourself taking: you assumed you knew what everybody else already expected, and you did not.

In April 2015 Nepali equities were cheap and Nepali savers had almost nowhere else to put money. The earthquake, terrible as it was, destroyed houses and temples and lives — and did not destroy the banking system, which is most of what the index is made of. It arguably helped it. Reconstruction money arrived. Remittances rose, sharply, because Nepalis working abroad sent money home to families whose houses had fallen. Deposits swelled. Banks lend deposits, and a bank with more deposits and government-directed reconstruction lending in front of it is a bank with a better year coming.

None of that was in your head when you read the headline. What was in your head was a feeling, and the feeling was entirely reasonable, and reasonable feelings are not prices.

I keep this example close because it is the cleanest I know. Ordinarily, when a forecast fails, the forecaster may claim he lacked information. Here you had all of it, early and exact — and it was worse than useless, because it conferred enormous confidence in a conclusion that was wrong.

The thousand Nepals

Now the idea that makes this way of thinking operational, because I am going to use it for the rest of the book and I would rather hand it over properly than smuggle it in.

We saw one 2015. One. But 2015 could have gone a great many ways, and the one we got is a single sample from a much larger set of years that were available.

Imagine you could run 2015 a thousand times. The same country, the same starting point, the same people, the same banks, the same quantity of money looking for a home — and then let chance do what chance does. In some of those thousand Nepals the earthquake is centred thirty kilometres further east, beneath the Kathmandu valley itself, and the death toll is not nine thousand but a figure I do not care to write, and the banking system genuinely breaks, and the index does not recover for a decade. In some, the border never closes. In some, the monsoon fails on top of everything and the rural economy goes down with the urban one. In some, nothing much happens and the index drifts sideways and nobody writes a book about it.

We got one draw. Humanly it was a bad one. Financially it was a good one. Those two facts are unrelated, and the failure to see that they are unrelated is where most investment reasoning goes to die.

The error is to treat the history that happened as the only one that could have. Having seen it, we work backwards and construct a reason. The reason feels solid because it explains the thing we know. It would have explained the opposite outcome equally well, with different words, and the words would have been available in the same newspaper.

You cannot build the machine; Nepal does not come with a reset button. But you can ask the question the machine would answer, and it is short enough to keep in your head:

How many of the ways this could have gone would have been kind to me?

Not: did it work. Did it have to work. And if it did not have to work, then in how many of the versions where it failed would I still have been in a position to continue?

That second clause is the whole of risk management, and everything in Part Five is a footnote to it.

The dentist and the two brothers

There is a further consequence of thinking in terms of many possible paths rather than one actual one, and it is the most immediately profitable idea in this chapter, so I will make it concrete.

Two brothers each place twenty lakh into the same portfolio of Nepali shares on the same morning. Identical holdings, identical everything. One difference: a habit.

The elder checks his portfolio every day at four o’clock. The younger checks his once a year, at Dashain, because that is when his wife asks.

Nepali equities swing at something like twenty-two per cent a year, which is more than double what a fixed deposit pays, arriving as noise. Day to day such a portfolio is close to a coin toss — slightly better than a coin toss over long periods, because the market drifts upward, but close.

So the elder brother sits through roughly two hundred and thirty sessions a year. Call it a hundred and twenty green days and a hundred and ten red ones. A hundred and ten times a year he experiences the specific unpleasantness of having lost money — and losing hurts appreciably more than winning the same amount pleases, which is not a character defect but a well-documented asymmetry in how people are built.

The younger brother has one experience a year.

The same portfolio. The same return. A hundred and ten bad days against less than one.

This is not merely a point about happiness, and here is where it costs money. A man with a hundred and ten bad days a year does something about them. He trims. He rotates. He takes a profit to feel better after a bad week. He describes this as managing risk. What he is doing is paying commission, depository charges and capital gains tax for the privilege of converting noise into fees.

The formal version of this belongs to Shlomo Benartzi and Richard Thaler, who proposed in 1995 that the puzzle of why people demand such a large premium to hold equities at all is explained by how often they evaluate them — the more frequently you look, the more losses you experience, and the more compensation you require to tolerate the asset. In a later experiment with Kahneman and Tversky, investors shown returns aggregated over long periods took substantially more equity risk than those shown the same returns broken into short intervals.

The identical portfolio, presented at two resolutions, produced two different people.

I have watched frequency of observation destroy more Nepali capital than any bad company ever has. Not stock selection. Not fraud. Not even the 2078 top. The habit of looking.

Why this market is unusually cruel about it

Every market has this problem. Ours has it worse, for a reason peculiar to Nepal.

We have a daily price limit. A stock may move ten per cent in a session and then it stops — fifteen since April 2026. That sounds protective and partly is. But it also means a Nepali investor’s screen is unusually legible. In a market where a stock can gap thirty per cent overnight, everyone learns quickly that the screen is chaos and cannot be read. Here the moves are small and orderly enough to resemble information.

They are not. A stock closing up 2.4 per cent on a Tuesday in Poush is not telling you anything. It is the sound a market makes.

And we have no short selling and no derivatives, which means there is exactly one way to express any opinion: buy, or do not buy. When the only available tool is a hammer, every daily fluctuation begins to look like a nail — and each swing costs you brokerage, a flat depository fee, and eventually tax.

The trade you would actually have made

Let me close the loop on April 2015 honestly, because there is a version of this argument that is too clever and I do not wish to make it.

Had you sold everything on the twenty-second of April 2015 at 938, and bought back when the news finally felt safe — say the middle of 2016, the blockade well over, the papers no longer frightening — you would have repurchased somewhere around 1,700 or 1,800.

You would have avoided a fall from 938 to 838. One hundred points.

You would have missed a rise from 838 to 1,881. One thousand and forty-three points.

Ten rupees of missed gain for every rupee of avoided loss, in exchange for perfect foresight about the worst year this country has had since the war.

That is not an error at the margins. That is the entire enterprise standing on its head.

And observe what it would have done to your beliefs, which is the part that compounds. You sold before an earthquake and you were right that an earthquake was coming. You would have spent the remainder of your life telling people you called it. You would be a man with a story — and, as we established in the previous chapter, extremely convincing at a wedding.

Three habits, which is all I actually have

I do not want to leave you with only a warning, so here is what I do differently because of this chapter. None of it is sophisticated.

One. I write down what I expect the price to do, and why, before the event. Not after. The whole value is in the timestamp. When the event arrives and the price does something else, I hold a dated record that my model of the link between news and price was wrong, and I cannot talk my way out of it. This has happened more often than I would like to describe and every instance was worth money.

Two. I ask what must be true for me to lose, not what must be true for me to win. The winning case writes itself; you already believe it or you would not be looking. The losing case has to be dragged out of you. And in a market with no short selling, no hedging, and a settlement cycle that traps your cash for two days, the losing case is the only one that can actually reach you.

Three. I look less often. I have not solved this and will not pretend to. But I have moved from something like the elder brother toward the middle, and the improvement in my results from that single change exceeds the improvement from any refinement of analysis I have ever made.

Which is a humbling thing to write in a book that is mostly about analysis, and I have left it in for that reason.

There is an idea underneath all of this that I want to state plainly before moving on, because the rest of Part One grows out of it.

You will live one version of your own life. You will never see the others. But the others were real — they were available, they had probabilities, and the one you received is not truer than they were. Only more visible.

Judge yourself against all of them.

The man who stakes his house on a coin and wins is not a wealthy man who was clever. He is a man who is still at the table, and who has just been taught, thoroughly and at no cost to himself, exactly the wrong lesson.

A Fifth of Everything Is Missing

In which a Hungarian statistician tells the American navy to armour the places with no bullet holes; the original Dow Jones companies are counted and found absent; and one in five of everything that has ever traded in Kathmandu turns out not to be on your screen.

In 1943 the United States Navy had a problem it believed to be a matter of engineering, and it took a refugee mathematician to notice it was a matter of inference.

Bombers were returning from Europe with holes in them. The Navy did what any sensible organisation would do: it counted. Where the holes were, the aircraft was vulnerable; where the holes were, add armour. The damage clustered on the fuselage and the wings, so that is where the armour should go. Armour is heavy and fuel is finite, so the question of where to put it was serious and quantitative and the Navy had good data on it.

They took the analysis to the Statistical Research Group at Columbia, where Abraham Wald — who had fled Vienna in 1938, and whose family was murdered at Auschwitz — looked at the distribution of holes and told them they had the problem exactly inverted.

Armour the places with no holes.

The aircraft they were examining were the aircraft that had come back. A plane struck in the engine housing did not come back, so no engine housings appeared in the sample. The clean areas on the returning planes were not the safe areas; they were the fatal ones, and their cleanliness was a report not about where bullets went but about which planes were available to be studied.

I have thought about Wald a great deal over the years, and what strikes me is not the cleverness. It is that the Navy’s data was perfect. Nobody had miscounted. Nobody had lied. Every hole in the dataset was a real hole, meticulously recorded by competent people, and the entire analysis was nonetheless upside down, because the sample had been assembled by the very process under investigation.

That is the condition you are in when you form an opinion about the stock market. Not sometimes. Always. And the whole of this chapter is an attempt to make you feel it rather than merely agree with it, because everybody agrees with it and almost nobody feels it.

The best advice in Nepal, and where it comes from

The most common piece of investment advice in this country is also the best, and I want to spend a chapter taking it apart — not because it is wrong, but because of where it comes from.

The advice is: buy good companies and never sell.

It arrives with the authority of an example. There is always an example. An uncle bought two hundred shares of a bank in 2050, at a hundred rupees, and did nothing at all for thirty years — no charts, no news, no TMS login — and through the compounding of bonus shares he now holds some large number of shares worth more money than he can spend. He is not clever. That is the entire point of the story. He is proof that cleverness is unnecessary.

I have heard this told with the name of nearly every surviving Nepali bank in it, and I believe every version. The arithmetic is real. Bonus shares compound viciously in your favour when a company grows for three decades, and Nepali banks, for structural reasons I will get to in Part Four, have issued them the way a man hands out sweets at a mundan.

Now the question that turns the story into something usable.

How many other uncles bought two hundred shares of a bank in 2050?

The dead funds, and the returns that were never earned

Before Nepal, the international version, because it is measured and enormous and almost nobody outside the academic literature knows about it.

If you look up the historical performance of American equity mutual funds, you are looking at a database. That database contains the funds that exist. Funds that performed badly were closed or merged into better-performing siblings, and when a fund is merged its record commonly goes with it — the surviving fund’s history is the one that persists, and the dead one’s is not.

The consequence has been measured many times since Burton Malkiel’s 1995 study, and the estimates range from a few tenths of a point to about one and a half percentage points a year, rising with the length of the window — purely from the removal of the dead. Over a decade, something like a third of American equity funds disappear. S&P’s own persistence studies report survival rates alongside performance for exactly this reason.

One and a half points a year does not sound like much until you compound it across a career, at which point it is the difference between a comfortable retirement and a modest one, and it was never real. It was an artifact of who was still in the room.

The same defect, differently dressed: consider the Dow Jones Industrial Average, which is the oldest continuously reported industrial share index in the world and, in the popular mind, a statement about the enduring strength of American enterprise.

It began in May 1896 with twelve companies. American Cotton Oil. American Sugar. American Tobacco. Chicago Gas. Distilling and Cattle Feeding. General Electric. Laclede Gas. National Lead. North American. Tennessee Coal and Iron. United States Leather. United States Rubber.

Read the list again. Of those twelve, exactly one — General Electric — was still in the index in the twenty-first century, and in June 2018 it too was removed.

Zero of the original twelve remain. The index that is held up as evidence of the durability of great businesses has, over its life, replaced every single one of its founding constituents, most of them because they failed or shrank into irrelevance. The index endured. The companies did not. And the returns you read about are the returns of a process of replacement, not of a portfolio anybody could have held by falling asleep.

If you want the same lesson closer to home, run the exercise on the thirty constituents of India’s Sensex when it was constituted in 1986 and count how many are still in it. The answer is small and the survivors are not the ones a 1986 investor would have named.

Now count the Nepali dead

I keep a database of Nepali share prices. Three hundred and eighty-four symbols, five hundred and forty-three thousand daily bars, July 1997 to July 2026. I built it because I wanted to test things, and you cannot test anything on a list of companies that happen to exist today. Wald’s problem is not academic when you are the one assembling the sample.

Here is what is in it that is not on your TMS screen.

Fifty-four symbols with a real trading history — four hundred sessions or more, which is close to two years of genuine listed life — simply stop. They print a price one day and never print again. Not a gap. Not a suspension. The series ends.

Sorted by the year they ended:

Twenty in a single year. And if you remove my four-hundred-session filter and count everything, the figure is seventy-eight out of three hundred and eighty-four.

One in five of the companies that have ever traded on this exchange is gone.

Not “did badly”. Gone. The ticker does not exist. You cannot buy it and you cannot sell it, and — this is what matters for the uncle’s story — it appears on no screen consulted by anybody forming an impression of what Nepali shares do over time.

Two ways to disappear, and only one of them hurts

I want to be honest here, because there is a lazy version of this argument that overstates the case, and I dislike it when other people make it, so I am obliged not to make it myself.

A company can leave an exchange in two very different ways.

It can be absorbed. Somebody merges with it, and your shares become shares of something else at an agreed ratio. Nepal did an enormous amount of this between 2019 and 2024, largely because the central bank wanted fewer and larger banks and used capital requirements to obtain them.

It can be struck off. It stops meeting listing requirements, or it fails, and what you hold becomes a piece of paper with a story attached.

The first is far more common here, and I have looked carefully at whether it is where the damage lies. It is not. Here are the nine bank consolidations I have been able to source properly, with terms checked against the last price each dying ticker actually printed.

Nepal Bangladesh Bank went into Nabil on the seventeenth of January 2022 at a hundred for forty-three. Its last close was 399. Forty-three per cent of a Nabil share that day was worth 402.8. Shareholders received, within one per cent, exactly what the market said they held.

Civil Bank went into Himalayan Bank in February 2023 at a hundred for 80.28. Last close 214; the consideration was worth 228. They received six and a half per cent more than the market price.

This is what you should expect, and it is a modest piece of evidence that this market is less broken than people enjoy saying. By the time a merger completes, the price of the disappearing share already embeds the terms. The swap is not the robbery.

The robbery happened years earlier, and it was not a robbery. It was a decline.

Look at the dates in the middle column. Six of the nine peaked in the same week — the last week of July 2021 — because that is when the whole market peaked, and these were ordinary banks moving with everything else. Then they fell for eighteen months and were absorbed.

A man who bought Laxmi Bank in Shrawan 2078 because it was a well-known commercial bank with a long history did not lose money in a merger. He lost fifty-seven per cent over two years, and then the ticker vanished from his screen and from everybody’s memory, and today, when you pull up a chart of “Nepali bank shares”, he is not in it.

He is the plane that did not come back.

The insurance cull

If you want the effect at industrial scale, look at 2022 and 2023 in insurance.

From my own price series, these stopped printing within about fourteen months: EIC, HGI, SLICL, GIC, SGI, PIC, SIL, LGIL, SIC, RLI, GLICL, PLIC, PICL, AIL, UIC.

Fifteen insurance companies. Several had traded since 2012 with more than two thousand sessions behind them. Two of them — GIC and SGI, both of which listed in late 2020 — printed four hundred and some sessions each and were gone by July 2022. They existed as public companies for under two years.

The cause was a capital requirement. The regulator raised the minimum paid-up capital for insurers; companies that could not raise it merged with companies that could. This is entirely sensible policy, it produced a stronger insurance industry, and I am not complaining about it.

I am pointing at something else. If you sit down today with a list of Nepal’s listed insurers and study their five-year returns, you are studying the ones that had enough capital. The ones that did not are absent from your sample — and not having enough capital was correlated with everything else that was wrong with them.

Your study will conclude that Nepali insurers did well. Your study is a machine for producing that conclusion regardless of the truth. It is the Navy counting holes.

The graveyard nobody visits

Go back further and it is worse. The tickers in my list that stopped between 2017 and 2021: SYFL, HAMRO, PURBL, KMFL, NBBL, BHBL, NCDB, KNBL, SBBLJ, SFFIL, NNLB, KADBL, DBBL, SDESI, MSMBS, UFL, SLBS, RRHP.

Development banks and finance companies, mostly. Small institutions in small towns. Some merged upward, some were quietly wound into other things, and one or two were not healthy at all.

Nobody remembers them. There is no folklore. Nobody’s uncle bought two hundred shares of one of these in 2050 and tells the story at weddings, because the story has an ending nobody wants to hear, and stories are selected for endings.

Which makes the advice hard to use

I said the advice is not wrong, and I meant it. Let me finish the thought.

The uncle’s strategy worked. Buying a strong Nepali commercial bank in the 1990s and holding it through thirty years of bonus issues was one of the great trades available in this country, and it beat land over most of the periods in which people believe land beat it.

The trouble is the word good.

The uncle did not select a good company. He selected a company, and it turned out good, and the ones that turned out otherwise did not produce uncles. Ask him how he chose and he will tell you something — a manager he trusted, a branch near his shop, a name that sounded solid — and whatever he tells you is useless, because five hundred other men used the same criterion and are not being asked.

So the correct response to the advice is not to reject it. It is to ask the one question that converts it into something you can act on:

Which characteristic, visible in advance, separated the banks that survived thirty years from the ones absorbed at a fifty-seven per cent discount to their own high?

That question has an answer. It is not short, it occupies most of Part Four, and it involves where the deposits come from, what the loan book is actually made of, how much capital sits above the regulatory floor, and whether the returns arise from a franchise or merely from leverage. But it is answerable, and answerable in advance, which is the only kind of answer worth anything.

What is not answerable is the version most people run, which is: hold good companies, where good companies are the ones that did well.

The most expensive filter in the country

I want to leave you with the general shape, because it recurs and I would rather name it once than gesture at it fifteen times.

Everything you know about this market reached you through a filter, and the filter selected on outcome.

The companies you have heard of are the ones that survived. The investors you have met are the ones who could afford to keep investing. The strategies you have been told about are the ones that worked recently. The brokers still in business are the ones whose clients did not all leave. Even the historical price charts on the terminals — and I checked this specifically, which is why my own database is built the way it is — generally begin from today’s listed universe and walk backwards, which quietly deletes everything that died.

None of these filters is anybody’s fault. Nobody designed them. They are the natural consequence of failure being quiet and success being loud, and they compound, and by the time you sit down to form a view of what Nepali equities do, you are looking at a curated exhibition and calling it a census.

I have no clean solution. What I have is a habit, and it is the only one I know that works: before believing a claim about this market, ask what it was measured on, and then ask what is missing from that measurement.

In Nepal the answer is usually twenty per cent. One in five. Whatever you are looking at, one in five of the things that ought to be in it are not.

Sunrise Bank was founded in 2007. It traded for eleven years in my series. It reached 415.5 in Bhadra 2073 and never saw that price again in the seven years that followed. It merged into Laxmi Sunrise on the twenty-ninth of Asar 2080, one for one.

A man who bought at the high and held to the end — who did exactly what the uncle did, with exactly as much patience, in a bank whose branches you have walked past — received 41.7 rupees on the hundred.

He also has a story.

Nobody has ever asked him for it.

Paid and Right Are Two Different Words

In which a man makes four billion dollars in a year and is never right again; a football coach calls a defensible play and is called an idiot for a decade; a biologist explains why nobody hits .400 any more; and we discover that the better everyone gets, the more luck decides.

In 2007 John Paulson made more money in a single year than any individual in the history of Wall Street.

He had spent 2006 studying the American mortgage market and had reached a conclusion that was, at the time, close to eccentric: that loans were being written to people who could not repay them, that these loans were being bundled into securities rated as though they were safe, and that the instruments used to insure against their failure were absurdly cheap because nobody believed failure was possible. So he bought the insurance. Enormous quantities of it, across years, while it bled premium and while his investors asked what he thought he was doing.

The trade earned his firm something like fifteen billion dollars. He personally took home about four billion, a figure that had no precedent and has had few since. Books were written. The trade acquired a definite article: the greatest trade ever.

Now.

In 2011 his flagship Advantage Plus fund fell roughly fifty-one per cent. His gold fund, launched on a thesis about currency debasement that read, on paper, exactly as coherent as the mortgage thesis had, lost enormous sums across the following years. Investors withdrew. In 2020 Paulson closed the firm to outside money and converted it into a family office, which is the polite form of retirement in that business.

So: was John Paulson skilled?

I want to answer this one carefully, because it is not the answer people expect and it is the hinge of the chapter.

Yes. Almost certainly yes. The mortgage trade was not luck. He did work that others had not done — he and his analysts went through loan-level data, they understood the difference between the tranches, they identified an instrument whose price did not reflect its risk, and they structured a position that could survive being early. That is skill in its purest available form, and I would not insult it.

The error was not his. The error was everybody else’s, and eventually his too, and it was this: from one correct answer, the world inferred a general faculty.

He had found one mispricing. That is not the same thing as possessing an ability to find mispricings, any more than a man who correctly predicts one monsoon possesses meteorology. And when the world hands a man four billion dollars and calls him a genius, he will look for the next one, and he will find something that resembles it, and the resemblance is the trap.

Resulting

There is a name for the error and it comes from poker, which has thought about this more clearly than finance has, for the excellent reason that poker players get their feedback in minutes rather than decades.

Annie Duke, who won a great deal of money at poker before writing about decisions, calls it resulting: judging the quality of a decision by the quality of its outcome.

Her opening example is not from a card table. It is the final minute of the 2015 Super Bowl. Seattle, trailing by four, has the ball on New England’s one-yard line with twenty-six seconds left and one timeout. They have Marshawn Lynch, among the most effective short-yardage runners in the game. Everybody in the stadium expects a run.

The coach, Pete Carroll, calls a pass. It is intercepted. Seattle loses the Super Bowl.

He was, within hours, the worst coach in living memory. The call was described as the most idiotic play in the history of the sport, and it is still described that way, and Carroll has been answering for it for a decade.

Duke’s point is that the play was defensible. Given the clock and the timeout situation, a pass on that down preserved the possibility of three attempts rather than two; the historical interception rate on that kind of pass in that situation was around two per cent. A decision with a two per cent disaster rate produced the disaster. That is what a two per cent rate means. It does not mean the decision was wrong; it means that one time in fifty you find out what the tail feels like.

Had the pass been caught, Carroll would have been a tactical genius who outfoxed the obvious. Same decision. Same information. Same reasoning. Opposite reputation, determined entirely by a football’s trajectory.

We do not grade decisions. We grade outcomes, and then we invent the grade for the decision to match.

The test: can you lose on purpose?

If skill and luck are tangled, we need a way to pull them apart, and the cleanest one I know is Michael Mauboussin’s, from The Success Equation. It is a single question and it takes about four seconds.

Can you lose on purpose?

In chess, obviously yes. I can lose deliberately, immediately, and convincingly. That establishes that chess contains skill, because a thing you can deliberately do badly is a thing that can be done well.

At a roulette wheel, no. I cannot lose on purpose. I can bet on red instead of black, but my expected outcome is identical and my actual outcome is not mine to determine. Roulette contains no skill, and the test detects it in one question without any statistics at all.

Now apply it to investing.

Can you lose on purpose in the Nepali stock market? Of course you can. Buy the most expensive company on the exchange with borrowed money, trade it daily, pay the depository fee on every round trip, and concentrate in a business whose only customer does not pay its bills. You will lose reliably. So investing contains skill; the question was never binary.

But note what the test also tells you, which is the part people skip. The speed at which you can deliberately lose is a measure of how much skill the activity contains. In chess I can lose in four moves. In investing it takes months, and in a rising market it may take years, because the tide will bail out even a determined idiot for a surprisingly long time.

That gap — between how quickly you can lose on purpose and how quickly you can win on purpose — is where all the confusion lives.

Why the better everybody gets, the more luck decides

Here is the most counterintuitive idea in this chapter, and it comes to finance by way of baseball and evolutionary biology, which is the sort of route I find irresistible.

In 1941 Ted Williams batted .406. Nobody in Major League Baseball has hit .400 since. The obvious reading is that hitters have declined — that the giants are gone, that the modern player is soft, that something has been lost. This is the reading every generation of sportswriters has offered.

Stephen Jay Gould, the palaeontologist, demolished it in Full House. Hitting has not declined. Everything else improved. Pitching got better, fielding got better, training, scouting, nutrition, the analysis of a batter’s weaknesses. And crucially, the variation between players narrowed: as the whole population approaches the outer limit of what a human body can do with a bat, the gap between the best and the average shrinks. The .400 average was never a measure of Williams alone. It was a measure of the distance between Williams and everybody else, and that distance has closed.

Mauboussin took this and gave it its financial name: the paradox of skill.

As absolute skill in a field rises and becomes more uniform, luck becomes more important in determining outcomes.

Sit with what that does to fund management. The people managing money today are, by any measure, more capable than the people managing money in 1960 — better trained, better informed, with data that would have seemed like sorcery. And precisely because of that, the dispersion of skill among them has collapsed, and the share of their results attributable to chance has risen.

This is why beating the market has become harder in developed markets even as the average manager has become better. Both facts are true simultaneously and neither causes the other; they are the same fact, viewed from two ends.

And it tells you exactly where to look for an edge. Not where everybody is excellent — there, luck rules. Where everybody is mediocre, inattentive, or absent. Which is a description of the Nepali market, and the single most hopeful sentence in this book, and I will spend Part Four showing you where the inattention actually sits.

The four boxes

The whole business fits in a small table. I did not invent it and cannot identify who did.

Two boxes are honest. Think well and get paid, you earned it; think badly and get hurt, you earned that too.

The top-right — good decision, bad outcome — is painful and survivable. It teaches nothing false provided you know which box you are in, which is the whole problem.

The dangerous box is bottom-left. Lucky. Bad decision, good outcome.

It is dangerous for reasons that have nothing to do with the money, and money you can recover. What the lucky box does is teach. It emits a signal indistinguishable from the signal emitted by the deserved box, and the human nervous system has no mechanism for telling them apart. It updates. It strengthens whatever behaviour produced the reward. It is doing precisely what it evolved to do, and in a market it is doing it on noise.

You cannot switch this off. It is not a bias you can reason your way out of; I have tried, in full knowledge of the literature, and it makes no difference at all. The only defence is to record the decision separately from the outcome, before the outcome exists.

My own bottom-left box

I described the hydropower company in Chapter One. Let me give you the other one, because it is worse and more useful.

In 2079 I bought a development bank. Small, regional, not well known. My reason was specific and I still think it was decent: it traded below book value, its non-performing loans were low against its peers, and its cost of funds was better than its size implied, because it had an unusually sticky deposit base in a district where it was the only real bank.

Six months later I was up about forty per cent and felt like a man who had done some work and been paid for it.

Then I went and looked at why it had gone up.

It had announced a merger. A larger bank was acquiring it, and the price had jumped to the swap value.

Now — was I right?

I had bought a bank below book value in a consolidating industry and been paid, and the mechanism of payment was exactly what you would predict for a cheap small bank in a consolidating industry: somebody bought it. In one sense that is the thesis working perfectly.

Except that merger was not in my thesis. Not in any form. I had never written the word. I believed I was buying an under-priced earnings stream that would re-rate as the market noticed it. What occurred was a corporate event I had not considered, could not have predicted, and which would have paid me identically had my analysis of the deposit base been complete nonsense.

The result validated a thesis I did not hold. And the effect on me — I watched it happen from the inside and could not stop it — was to raise my confidence in the analysis of sticky deposits in district towns, about which the outcome said nothing whatsoever.

I caught it only because I had written the thesis down and went back and read it. I want to be honest about that too: I did not catch it through insight. I caught it through paperwork.

Why this market is unusually good at training you wrong

Every market confuses decisions with outcomes. Nepal’s has four features that make it worse than most, and they compound.

The feedback is slow and the sample is tiny. As established in Chapter One, a long-horizon investor here makes perhaps a dozen real decisions a year. Over a decade that is a hundred and fifty data points across four or five regimes, most of them correlated. You will never accumulate enough outcomes to learn from outcomes.

Everything moves together. In 2077–78 essentially every listed share in Nepal rose; not most, essentially all — one hundred and eighty-five of one hundred and eighty-five. In such a year the correlation between the quality of your thinking and your return is approximately zero, not because thinking is worthless but because the common factor swamped it. And that year was not exceptional in kind, only in degree. This market has a powerful tendency to move as one object, for reasons I set out in Part Two.

There is no way to be wrong quickly. No short selling. No options. If you believe a company is overvalued, the only available expression is not owning it, and not owning something produces no record, no position, and no feedback. In a market where you can short, being wrong is expensive and immediate and you find out within weeks. Here, the entire negative half of your judgment evaporates without trace.

The consequence is worth stating plainly: a Nepali investor’s track record consists exclusively of things he thought were cheap. He never learns anything about the other half of his judgment, which is the half that would tell him whether he can judge at all.

And the tide is large relative to the differences. Which is the paradox of skill running in reverse: here, dispersion between companies is enormous, so a concentrated bet produces spectacular outcomes with no skill required — as the ladder in Chapter One demonstrated.

Scoring the decision instead

So what do you actually do. I keep this small, because an elaborate system will not survive contact with a busy year and an abandoned system is worse than none.

One file. Every position gets an entry when opened, with four lines.

What I am buying and at what price. Obvious, but dated, because the date is the entire point.

What I think it is worth, as a range, and how I got there. A range, never a number. If I cannot write the derivation in three sentences I do not understand it well enough to own it. This line has stopped more purchases than any other.

What would have to happen for this to be wrong. Specific and falsifiable. Not “if the market falls” — that is not a thesis failure, that is weather. If I am buying a bank because credit costs are normalising, then the non-performing ratio is still above three per cent in two years is the falsifier. If I am buying a hydropower company for its contracted tariff escalations, then the escalations turn out to be exhausted is the falsifier.

What would make me sell that has nothing to do with price. Because if the only sell trigger is price, then price is the whole model, and you do not have a thesis. You have a hope with a decimal point.

Twice a year I score the entries, not the returns. Did the thing I said would happen, happen? Was the falsifier triggered? Did I act on it when it was?

The scores and the returns disagree constantly, and that disagreement is the most valuable information I have ever gathered about myself.

On tuition fees

There is a phrase I hear often, usually from people who have just lost money: it was my tuition fee.

I like the humility and dislike nearly everything else about it, because a tuition fee purchases an education and most market losses purchase nothing.

A loss teaches only if you can identify what was wrong. Bought at a bad price for a good reason: the lesson is about price. Bought at a good price for a bad reason: the lesson is about reasoning. Bought at a good price for a good reason in a year when the country had a catastrophe: there is no lesson, and manufacturing one is worse than useless. It is negative learning, because you will now avoid a correct behaviour.

Most people, faced with a loss, change something. Anything. They swear off small companies, or resolve to always sell at twenty per cent down, or conclude that fundamentals do not work in Nepal. The change feels like learning. It is usually a reaction to the most recent draw.

Before changing anything after a loss, ask one question: would this change have helped in all the other years, or only in this one?

If only in this one, it is not a lesson. It is a scar, and you are about to institutionalise it.

When the outcome is the right teacher

I do not want to leave you believing outcomes never matter. They do, in exactly one circumstance, and it deserves naming precisely.

Outcomes are informative when they are numerous, independent, and fast.

Nothing about long-horizon investing in Nepal is any of the three. But some things are. The quality of your execution is — if you place a hundred orders a year and are consistently filled worse than the price you saw, that is a real signal from a large sample and you should act on it. The accuracy of your earnings estimates is, if you make them often and score them. The reliability of a data source is.

So the rule is not “ignore outcomes”. It is: use outcomes where the sample is large, and use process everywhere else. In practice this means judging your plumbing by results and your judgment by reasoning — and if you do it the other way round, which is what nearly everyone does, you will end up with magnificent convictions and a leaky pipe.

Paulson gave an interview some years after the closure in which he was asked about the mortgage trade, and he explained the reasoning, and the reasoning was excellent. It was excellent in 2006 and it remained excellent, and it made four billion dollars, and none of that tells us whether he should have been given the next twenty billion to manage.

What we can say is narrower and duller and true: he was right once, spectacularly, about something real. Then a great many people, including him, drew from that single observation a conclusion about a general faculty that the observation could not support.

Which is the same error the market makes about you, every year, in both directions — and unlike Paulson, you do not even have four billion dollars to show for it.

Four Numbers a Day, One of Them Invented

In which the greatest tape reader who ever lived shoots himself in a hotel cloakroom; a famous academic vindication of chart-reading is destroyed by counting how many rules were tried; a fund in Long Island proves that price does contain information and then closes its doors; and Nabil Bank opens at exactly the previous close on one hundred and sixty-four consecutive days.

On the twenty-eighth of November 1940, Jesse Livermore went into the cloakroom of the Sherry-Netherland Hotel on Fifth Avenue and shot himself.

He left a note for his wife. Among the things it said was that his life had been a failure.

I want to establish who this man was before we go any further, because if you have read anything at all about trading you have absorbed his methods without knowing his name. Livermore was the finest reader of price action in the history of markets — not one of the finest, the finest, by acclamation of everyone who watched him work. He began as a boy chalking quotations in a bucket shop in Boston and made his first fortune before he was thirty by noticing, purely from the tape, how prices behaved around certain kinds of activity. He shorted the market into the panic of 1907 and was asked by J. P. Morgan to stop. He shorted it again into 1929 and came out with something around a hundred million dollars, which in that year was an amount of money it is difficult to describe.

Reminiscences of a Stock Operator, published in 1923 and thinly disguised as fiction, is his account of how he did it, and it remains the most widely read book on trading ever written. Every rule you have heard in a Nepali technical analysis course descends from it. Cut your losses. Let your winners run. Do not average down. Trade with the trend. The market is never wrong, opinions are.

He went bankrupt four times.

He was worth a hundred million dollars in 1929 and by 1934 he was declared bankrupt again, and when he died six years later the estate was a fraction of what he had made. The man who wrote the rules could not survive by them, and he knew it, and the note in the cloakroom says so.

I do not raise this to be cruel to a dead man. I raise it because Livermore is the patron saint of an entire industry of instruction, and the industry teaches his methods and does not teach his ending, which is the single most informative fact about him. The ending is the omitted parameter.

The honest case for the other side

Now let me do the thing that almost no fundamental investor is willing to do, which is to make the best available argument against my own position, using the strongest evidence rather than the most convenient.

In 1992 three respectable academics — William Brock, Josef Lakonishok and Blake LeBaron — published a study in the Journal of Finance testing twenty-six simple technical trading rules against ninety years of the Dow Jones Industrial Average, from 1897 to 1986. Moving average crossovers. Trading range breakouts. The kind of thing on Bikash’s screen.

They found the rules had predictive power. Buy signals produced higher subsequent returns than sell signals, the differences were large, and the statistical tests looked solid. The paper was, and remains, the most serious empirical support that technical analysis has ever received, and it came from people with no stake in the answer.

For seven years it stood.

In 1999 Ryan Sullivan, Allan Timmermann and Halbert White published the correction, and it is one of the most instructive papers in finance for reasons that have nothing to do with charts.

They pointed out that Brock and his colleagues had tested twenty-six rules — but that those twenty-six had not descended from heaven. They were the rules that the technical analysis community had arrived at after a century of collective experimentation. Thousands upon thousands of variants had been tried by practitioners over the decades, and the twenty-six that got tested were the survivors of that search.

So Sullivan, Timmermann and White constructed the universe. Nearly eight thousand trading rules — every plausible parameter combination of the families in question — and asked a different question: given that this many rules were available to be searched, how good does the best one have to look before we should be impressed?

The answer is more uncomfortable than a debunking, and I report it against my own case because I had it backwards until I checked.

Brock’s result survived. Over the original 1897 to 1986 sample, once the full universe of 7,846 rules was priced into the test, the best rule was still statistically superior. The data-snooping correction did not kill it.

What killed it was the future. Applied to the decade after the sample ended — 1987 to 1996 — the same rules produced no superior performance at all. And on the S&P 500 futures contract, once the search was accounted for, there was no evidence of outperformance in the first place.

Which is the more damning pair of findings, and the one worth carrying: a rule can be genuinely significant in the data it was discovered in and worthless the day after that data stops. Not a false positive. A true positive with no future.

This is a specific and generalisable statistical idea and it is worth more to you than any chart pattern. It is called data snooping, or the multiple comparisons problem, and it is this: if you test enough hypotheses, some will pass by chance, and the pass rate tells you nothing unless you know how many were tried.

Test one rule at the five per cent threshold and a false positive is unlikely. Test eight thousand and you will find four hundred that clear it while being pure noise. And crucially, you will only be shown the four hundred. The 7,600 failures are not written up, not taught, not sold in a three-month course. They are the bombers that did not come back.

This applies to every backtested strategy you will ever be shown, in every market, by every person with a slide deck. The first question is not “what were the returns.” The first question is ”how many did you try?” — and if the answer is not recorded, the returns are uninterpretable, not merely unimpressive.

I will come back to this in Chapter Thirty-Two, where I have to apply it to my own work and it costs me several findings I liked.

And yet: Long Island

Here is the honest complication, and I will not pretend it away.

There is a fund on Long Island called Medallion, run by Renaissance Technologies, founded by Jim Simons, who was a distinguished mathematician — a Chern-Simons theory distinguished, not a business-school distinguished — before he was anything else. Medallion has been reported to compound at something like sixty-six per cent a year gross of fees, and roughly thirty-nine per cent after them, over three decades.

That is the greatest investment record ever produced by anyone, by a distance so large that the second-place holder is not visible from it.

And Medallion is a technical operation. It does not read annual reports. It looks for statistical structure in prices, volumes and related series, and it trades on that structure at horizons measured in days.

So price contains information. That is settled, and I am not going to argue with the arithmetic of thirty years.

But look at the conditions attached, because the conditions are the entire message. Renaissance hires astrophysicists, signal processing specialists, and computational linguists — famously, almost nobody from finance. It maintains data infrastructure of a scale ordinary institutions cannot approach. Its edge per trade is tiny and it survives only because the edge is applied millions of times with ferocious cost control. And in 1993 it began closing to outside money, and by the mid-2000s Medallion was managing essentially only the partners’ own capital, because the strategies have limited capacity and the partners would rather have the returns than the fees.

The people who have demonstrably extracted information from price charts responded by locking the door and refusing your money.

Which tells you what to conclude. Not “technical analysis does not work.” The honest conclusion is narrower and more useful: whatever exploitable structure exists in price series is small, is competed for by people with resources you do not have, and is not available through the vocabulary of hammers and doji taught in a three-month course. The activity Renaissance performs and the activity being sold in Kathmandu share a subject matter and nothing else, in the way that surgery and butchery share a subject matter.

Bikash, and his framed certificate

Bikash is twenty-four. He has a laptop, a TMS login, and a certificate from a three-month course in technical analysis that cost him twenty-two thousand rupees. The certificate is framed. I have seen it.

He showed me his screen once, and it was a beautiful thing — I mean that sincerely, the way a well-made watch is beautiful. Candlesticks in red and green. Two moving averages crossing. A histogram in a second panel. Fibonacci lines drawn from a low in Poush to a high in Chaitra. He talked me through it for twenty minutes, articulately and precisely, and I did not understand a word of it in the sense of being able to predict anything from it, and understood every word in the sense of knowing what he meant.

He is not stupid. He is among the more diligent people I know. He rises at seven to review charts before the market opens at eleven. He has spent, by my estimate, something like fifteen hundred hours on this.

Let me now tell you what I think those fifteen hundred hours have been spent on, and I will start not with whether technical analysis works in general — I have just given you the state of that argument, honestly, at more length than most of its critics manage — but with something considerably narrower and more embarrassing.

The candles are made up

A candlestick chart is built from four numbers a day: open, high, low, close. The body of the candle is the distance between the open and the close. That body is the whole content of the picture. Everything Bikash was telling me — the hammers, the doji, the engulfing patterns, the entire vocabulary — is a vocabulary about the relationship between the open and the close.

So I went and checked whether NEPSE has ever recorded an open.

Take Nabil Bank. Not an obscure name: the largest and most heavily traded commercial bank on the exchange for most of its life, the closest thing this market has to a blue chip. Here is the fraction of trading days on which Nabil’s recorded open was exactly equal, to the paisa, to the previous day’s close.

One hundred per cent. In 2013, on every one of a hundred and sixty-four trading days, Nabil Bank opened at precisely the price at which it had closed the day before.

That does not happen. It cannot happen. A stock with real buyers and real sellers does not open at the previous close a hundred times consecutively, let alone a hundred and sixty-four. Nepal Telecom shows the same: 146 of 146 in 2013. Himalayan Bank: 119 of 119.

What happened is that NEPSE did not publish a session open in those years, and the data vendors filled the empty column with the previous close, because a chart requires four numbers and they possessed three.

Across my whole database of three hundred and eighty-four symbols the pattern is unmistakable. In 2012, 2013 and 2014, between seventy-five and eighty-two per cent of all daily bars carry an open identical to the previous close. By 2025 it is seven per cent, which is roughly what genuinely thin trading produces.

Every candlestick pattern anybody has ever identified on a Nepali stock chart before about 2015 was drawn on a number that nobody recorded.

The bodies are all the same shape because they were generated by the same rule. A bullish engulfing pattern in Nabil in 2013 is not a description of two days of trading psychology. It is a description of an arithmetic operation performed by a vendor’s import script.

I want to be fair about one thing: this is not a scandal and nobody cheated. The exchange did not publish the field, the vendors needed the field, and they filled it in the least misleading way available, which is to carry the last price forward. That is standard practice and defensible. What is not defensible is drawing conclusions from it. And I should say plainly that after about 2016 the opens become real, and by 2025 they are as real as anything here — so the fabrication has an end date.

But anybody who backtested a candlestick system on ten years of Nepali history, and people do this, and there are courses that teach it, was for most of that history testing a rule against a field somebody invented.

The stock you are not permitted to buy

Set the data aside; suppose the numbers were perfect. There is a second problem, and being structural, it does not improve with better data.

NEPSE has a daily price limit. A stock may move ten per cent from the previous close and then it stops. Fifteen since April 2026, but for almost all the history that matters, ten.

Now consider what a breakout system is. Every version of it — moving average crossovers, Darvas boxes, fifty-two-week highs, “buy strength” in any dialect — says the same thing: when price moves up decisively, buy.

In a market with a hard limit, the most decisive upward move available is a limit move. And a stock locked at its upper limit has, by definition, buyers at that price and no sellers. That is what limit-up means: a queue of people wanting to buy and nobody willing to sell.

The strongest signal your system can generate is precisely the signal you cannot act on.

This is not a friction and it is not a worse fill. It is a structural inversion: the better the signal, the less available the trade. On the days your system is most confident, your order sits unfilled behind two thousand other orders from two thousand other people whose systems fired on the same morning for the same reason. And the days you are filled on a breakout are the days the move was weak enough for a seller to remain — which is to say, the days the signal was worst.

I do not know a way around this and I have looked. You can chase the next day, ten per cent higher, which destroys the arithmetic. You can try to anticipate the breakout, at which point you are not running a breakout system but a prediction system with extra ceremony. Or you can accept that in this market, strength is not purchasable.

Half the toolbox is missing

There is no short selling in Nepal. No options, no futures. One direction.

I do not think people who have never traded elsewhere appreciate how strange this is. Every technical system in every book was developed in markets where a negative view is expressible. When the head-and-shoulders top completes, you sell short. That is the trade. That is what the pattern is for.

Here, the completion of a bearish pattern produces one of two actions: sell what you own, or do nothing. If you do not own it, the pattern is information you cannot use. And since you can own only a few dozen things, the overwhelming majority of bearish signals your system generates are unactionable by construction.

So: a system whose long half is blocked by the circuit and whose short half does not exist. What remains is the middle — moderate signals on things you already hold.

That is not a system. It is a mood with a subscription fee.

The two-hundred-day average and the fifty-one-day hole

One further mess, and this is my favourite, because it is trivial to verify and universally ignored.

A two-hundred-day moving average is a claim about two hundred trading days. It assumes those days are roughly evenly spaced in time, because if they are not, “two hundred days ago” means something different every time you compute it.

NEPSE closed on the twenty-second of March 2020 and reopened on the twelfth of May. Then closed again, reopening on the twenty-ninth of June. In the whole of 2020 it held one hundred and eighty-one sessions against a normal year’s two hundred and thirty.

For most of a year afterwards, every two-hundred-day average in Nepal was reaching back further in calendar time than it previously had — by about two months. Nobody adjusted anything. Nobody could have; the indicator has no way to know.

It had happened before. Thirty-one days closed after the Gorkha earthquake in April

• Thirty-three in early 1998. Thirty-one more later that year.

And then there is the week itself. I counted every session in the index series by weekday, and the trading week in this country has changed four times.

Which means a “weekly” chart of NEPSE is not one thing. It is four different things stitched together and the stitches are invisible. A weekly candle in 2003 covers Monday to Friday. A weekly candle in 2015 covers Sunday to Thursday. If you have ever drawn a trendline across those two eras you drew it across a change in the definition of a week.

So why does everybody do it

Here I want to stop being clever, because the interesting question is not whether Bikash’s charts work. It is why fifteen hundred hours went into them, voluntarily, by an intelligent person, with nobody forcing him.

I think technical analysis solves a real problem, and the problem it solves is not prediction.

It tells you what to do today.

Sit with that. You have four lakh rupees. The market is open. You do not know whether Nepal Telecom is worth eight hundred rupees or six hundred, and finding out properly would take a week of reading filings you find boring, and at the end of it you would have a range rather than an answer. Meanwhile the price moves on your screen and doing nothing feels like bleeding.

A chart supplies a rule. The rule produces an action. The action produces relief. And relief is worth a great deal — more than anybody admits — because the alternative state, sitting in uncertainty with money at stake and no procedure, is genuinely unpleasant to inhabit.

Fundamental analysis, done honestly, cannot compete on this dimension. It is slow. It produces ranges instead of answers. It very often concludes I do not know, or this is roughly fairly priced, do nothing — and “do nothing” is the one output that provides no relief whatsoever.

So the demand for technical analysis is not really a demand for forecasts. It is a demand for permission — permission to act, issued by something that feels external and objective. Which is why the certificate is framed. The frame is not about the knowledge. It is about the authority.

I have some sympathy for this. I have considerably less for the people selling the courses.

What I concede

I would be a hypocrite to claim that nothing observable in price and volume carries information, since I use several such things, so let me be exact about which and why.

Liquidity is real and visible. How much of a stock trades, on how many days, in what rupee amounts, tells you something extremely concrete: how large a position you can build and unwind. That is not a forecast, it is a constraint — and it is the most under-used piece of information available to a Nepali retail investor, most of whom hold positions they could not exit in a week without moving the price against themselves.

The limit is real. Knowing a stock is at its limit tells you the order book is one-sided today. A fact about execution, not a prediction about tomorrow.

Long silences are real. A stock that has not printed a trade for many sessions is telling you something true about who owns it and who wants it.

Notice what the three have in common. They are statements about the present mechanics of trading, not about the future direction of price. They tell you what you can do, not what will happen.

Price data describes the market’s plumbing. It does not describe the market’s future. The plumbing is worth knowing.

What I actually said to him

I did not say most of this to Bikash, because it would have been cruel and because he had not asked.

What I said was: show me your record.

He did not have one. He had remembered trades — a very clear memory of a hydropower company where the averages crossed and he made thirty-one per cent, and a much vaguer memory of some others.

So I asked him for one thing, and it is the only thing I would ask of anyone reading this who trades on charts. Not to stop. Not to argue.

Write down every signal your system generates, including the ones you do not take. Then write down what happened. Do it for one year.

The ones you do not take are the entire experiment. Everybody remembers the trades they made. Nobody records the fifty signals skipped because the setup “did not feel right” — and did not feel right is where all the actual decision-making lives, which means the system is not the system. You are the system, and the charts are a costume.

A year of honest records will tell you which of you is doing the work. It is also, and this is not a coincidence, exactly what Sullivan, Timmermann and White did to Brock: it counts the rules that were tried and not just the ones that were reported.

He said he would. That was in 2081. I asked him last month and he said he had been meaning to start.

One Fact, Fifty Thousand Voices

In which 787 people at a country fair guess the weight of an ox and are collectively almost exactly right; Solomon Asch shows what happens when they can hear each other; Keynes explains that you are not picking the best company; and a short calculation reveals that a Viber group of forty thousand contains, on a generous estimate, ten opinions.

In 1906 Francis Galton went to a livestock fair at Plymouth and found a competition in progress.

A fat ox had been put on display, and for sixpence you could buy a ticket and write down what you thought it would weigh once slaughtered and dressed. The best guess won a prize. Eight hundred people entered — butchers and farmers who knew what they were looking at, but also clerks and children and men who had come for the day and had no idea whatever.

Galton, who was eighty-four, a cousin of Darwin’s, and possessed of a temperament that made him want to measure absolutely everything, obtained the tickets afterwards and analysed them. Seven hundred and eighty-seven were legible.

The middlemost guess was 1,207 pounds.

The ox weighed 1,198.

Nine pounds out. Under one per cent. No individual expert came that close, and the crowd — including the children and the men who had come for the day — had between them located the answer with a precision that would have satisfied a scientific instrument.

Galton published this in Nature under the title “Vox Populi,” and it is the foundation stone of every argument you have ever heard that a market price contains wisdom. The reasoning is straightforward: individual errors point in different directions, and when you average them, the errors cancel and the signal survives.

The argument is correct. It also has a condition attached, and the condition does all the work, and essentially nobody who cites Galton mentions it.

The guesses were written privately, on slips, by people who could not see each other’s answers.

What happens when they can hear each other

In 1951 Solomon Asch sat a student down at a table with seven other people and showed the group a card with a line on it, and a second card with three lines of obviously different lengths, and asked which of the three matched the first.

The answer was not ambiguous. It was the sort of question a child answers correctly.

The seven other people were confederates, and on certain trials they all, one after another, gave the same wrong answer, out loud, before the real subject’s turn.

About three-quarters of subjects conformed to the obviously incorrect majority at least once. Across all the critical trials, roughly a third of responses went along with the group.

These were not weak-minded people, and they were not confused about the lines. Asch interviewed them afterwards; most had seen the lines correctly and answered wrongly anyway, some because they assumed they must be missing something, some because they simply did not want to be the one man in the room who disagreed with seven others about a line.

Now put Galton and Asch side by side, because between them they describe the entire difference between a market that works and one that does not.

Galton’s crowd was wise because it was silent. Asch’s crowd was stupid because it spoke in order.

The calculation that ought to end the argument

Let me make this quantitative, because the qualitative version is easy to nod at and forget, and the quantitative version is genuinely shocking.

When you average a number of independent estimates, the uncertainty of the average falls in proportion to the square root of the number of estimates. Four independent guesses are twice as precise as one; a hundred are ten times as precise. This is why Galton’s ox worked.

When the estimates are correlated — when they influence one another — that stops being true, and it stops being true violently. The precision of the average no longer improves without limit; it converges to a floor set by how correlated the opinions are, and beyond a certain point adding more people accomplishes nothing at all.

There is a standard way to express this: the effective number of independent opinions in a group of n people whose views are correlated by a factor ρ is

n / (1 + (n − 1)ρ)

Take a Nepali share-market Viber group with forty thousand members and ask how many independent opinions it contains.

Read the fourth row. At a correlation of just five per cent — which is to say, if members’ opinions overlap only a twentieth as much as they might — forty thousand people contain twenty opinions.

At ten per cent, ten opinions.

And a Nepali share group is not at ten per cent. Everyone reads the same two or three portals. Everyone is in the same groups; the eleven groups I belong to are largely the same people eleven times. The same twenty or thirty prominent voices are forwarded into all of them. The evening television panels rotate a cast of about a dozen, and those dozen read the same portals as everybody else. If the correlation is fifty per cent — and I would guess higher — then forty thousand people contain two opinions.

This is not a metaphor and it is not rhetoric. It is what the arithmetic says.

And here is what makes it dangerous rather than merely disappointing. An average of two correlated opinions does not look like two opinions. It looks like forty thousand. It arrives with the full social weight of forty thousand people agreeing, and it produces in you the confidence that forty thousand people ought to produce, and it contains the information content of a conversation between two men who read the same newspaper.

The queue outside the restaurant

The mechanism by which a crowd becomes correlated has a formal treatment — Sushil Bikhchandani, David Hirshleifer and Ivo Welch called it an informational cascade in 1992 — but the intuition is better delivered as a story, and I have moved theirs to Thamel.

Two restaurants, side by side. Identical menus, identical prices, both empty. The first person to arrive has no information at all, so he picks one at random. Say the left.

The second person arrives. She also has no information — but now she has one piece: there is a man sitting in the left restaurant. That is weak evidence. It is better than nothing. She goes left.

The third arrives and sees two people on the left and none on the right. The evidence appears stronger. He goes left.

By the tenth person the left restaurant looks obviously better and everyone who arrives can see both a queue and a reason.

But there is exactly one person’s worth of information in that entire queue, and it was a coin flip.

Everyone after the first is reading the crowd rather than the restaurant. Each behaves completely rationally. Each adds one more body to the evidence without adding a single bit of information. And the queue is now self-reinforcing, completely uninformative, and indistinguishable from the outside from a queue caused by good food.

That is a share-market group chat. That is, near enough exactly, what happens to a Nepali stock between nine and eleven in the morning.

You are not picking the best company

There is a further layer, and it belongs to Keynes, who put it in Chapter Twelve of the General Theory in 1936 and has never been improved upon.

He described a newspaper competition in which readers were shown a hundred photographs of faces and asked to pick the six prettiest. The prize went to whoever’s selection came closest to the average selection of all competitors.

So, Keynes observed, you do not choose the faces you find prettiest. You choose the faces you believe other people will find prettiest. But everyone else is doing the same, so really you must choose the faces you believe other people believe other people will find prettiest. And there are, he noted drily, some who practise the fourth, fifth and higher degrees.

This is what a market with a thin float and a loud group chat becomes. Nobody is estimating what the hydropower company is worth. Everybody is estimating what everybody else will conclude about what everybody else will conclude, and the underlying business has been quietly excused from the proceedings.

I want to be precise about the practical implication, because it is not “therefore ignore other people.” It is this: there are two different games available in any market, and you must know which one you are playing. One is estimating the value of a business and waiting to be paid. The other is estimating the crowd’s next destination. Both can make money. They require entirely different skills, entirely different holding periods, and entirely different temperaments — and the reason most people lose is that they begin the first game, become impatient, and finish the second, without ever noticing the substitution.

GameStop, and what a cascade does at industrial scale

In January 2021 the shares of an American retailer of video games — a declining business, in shopping malls, in the year the malls were shut — went from under twenty dollars to an intraday four hundred and eighty-three.

The mechanism was a forum. Several million people on Reddit, reading each other, posting screenshots of their gains, and buying. There was a genuine underlying observation at the start, about the size of the short interest, and it was made by a small number of people who had done real work. What followed was almost entirely the crowd reading the crowd.

I raise it because the aftermath is the instructive part and it never gets discussed. The people who bought at forty dollars did very well. The people who bought at four hundred, on the same reasoning, from the same forum, with the same conviction and the same memes, did not. They were not less intelligent. They were not less committed. They were later, and in a cascade the only variable that matters is when you joined the queue, because the queue’s information content was exhausted at position one.

Every Nepali investor who bought a hydropower company in Bhadra 2078 because it had gone up for four days is the man at position four hundred. He read the same signal as the man at position four. The signal was identical. Its value was not.

The vacuum that rumour fills

Now the structural piece, and it is specific to Nepal, and I think it explains more about our market’s behaviour than any amount of psychology.

Official information in this country arrives slowly and on a schedule.

A listed company must publish quarterly results within thirty days of the quarter’s end — Schedule 14 of the Securities Registration and Issuance Regulations — and companies use most of the thirty days. Audited annual results get five months. Schedule 15.

Consider what that means. The Nepali fiscal year ends at Asar-end, in mid-July. The audited accounts for that year may legally appear as late as mid-December. In the meantime a great many people know how the year went — the bank’s own staff, the auditor, the board, the people who drink tea with the board — and the market does not.

Between the moment a company knows something and the moment it is obliged to say it, there can be five months of silence. Silence is not empty. It fills.

What fills it is rumour, and rumour in Nepal is not a random process. It flows along social lines: district, community, business network, family, the specific set of people who have known each other since school. A piece of true information about a hydropower company will reach a particular set of people in Kathmandu three weeks before it reaches anybody in Butwal, and by the time it arrives in Butwal it will have acquired a target price and shed its source.

I am not making an accusation about insider dealing. I am describing a communication topology. Even under perfect legal compliance, information in a small country with dense social networks does not diffuse evenly. It travels in channels.

The consequence is that the retail investor at the far end of a channel is not receiving news. He is receiving the residue of news — what remains after the people closer to the source have already acted.

Compare Bear Stearns, which in March 2008 went from a functioning investment bank to a forced sale at two dollars a share in about ten days, on a rumour of illiquidity that became true by being believed. Rumour is not a developing-market pathology. It is simply faster where the channels are shorter, and Nepal’s channels are very short.

The anatomy of one forwarded message

Let me slow down on a single message, because its anatomy repays inspection.

Somebody posts: ”XYZ Hydro — big news coming. Accumulate below 420. Target 600.”

Observe what is communicated and what is not.

There is no source. “Big news coming” is unfalsifiable: if news comes, he called it; if it does not, it has not come yet. There is a number, 420, which creates false precision and, more importantly, an anchor — your brain will now evaluate every subsequent price against 420 whether you consent or not, and the anchoring literature is unambiguous that knowing about the effect does not remove it. There is a target, 600, with no horizon and no method, which is not a prediction, because a prediction that cannot be scored is not a prediction.

And there is one more thing, the important one.

The message is free.

The person posting bears no cost if it is wrong. He is not identified, not tracked, not recorded. If XYZ Hydro reaches 600 he will post a screenshot; if it goes to 300 the message scrolls away and is replaced by another about a different company, and the group will remember the screenshot and not the silence.

It is the same filter as the wedding in Chapter One, running at higher frequency. The group is a machine that maintains a permanent, self-updating record of successful calls and deletes the rest, without anybody intending it, and the members experience this as evidence that the group contains skill.

There is a simple remedy and almost nobody applies it. Keep your own scoreboard on the loudest three voices in your groups. Every call, with the date. Do not judge, just record. Six months is sufficient.

What you will find is not that they are wrong. Some will be right about half the time, which is roughly what the market offers anybody. What you will find is that their hit rate is unremarkable and their remembered hit rate is extraordinary, and the gap between those two numbers is where your money goes.

The float makes it worse

One final amplifier, the most Nepali of all, sketched here and given a full chapter in Part Two.

In most Nepali listed companies a majority of shares — typically fifty-one per cent — are promoter shares, which do not trade in the ordinary market. What changes hands is the public portion, and within that a good deal is held by people who never trade at all.

So the genuinely floating quantity of many Nepali stocks is small. Small enough that a few hundred motivated buyers can move a price ten per cent in a day and hit the limit.

Put that beside the group chats and you have a closed loop. Forty thousand people receive one weak signal containing, as we computed, perhaps ten opinions’ worth of information. A few hundred act. The float is thin enough that a few hundred moves the price. The price move is then screenshotted and posted, which constitutes a new signal, apparently confirming the first — and it is entirely endogenous. It is the market reacting to itself with a delay.

This is why Nepali stocks move as they do: nothing, nothing, nothing, then four limit-up days in a row, then nothing for two months. That pattern is not information arriving. It is a cascade running to completion and exhausting itself.

The only defence I have found

I have tried several and most did not survive contact with real life. Leaving the groups failed; I missed genuine notices and came back. Muting worked better, but I still checked.

What actually works, for me, is a rule about sequence.

I am not allowed to form a view on a company after seeing what other people think of it. Only before.

In practice: when a name comes up and I am curious, I do not read the thread. I open the company’s own filings, do the work, and write down a range. Then I go back and read what everybody was saying. Sometimes they had a fact I lacked, and I incorporate it. Usually they had a mood.

It sounds fussy. It is fussy. But the alternative is that I read forty opinions and then “do my own analysis,” and I have done that too, and what emerges is always — every single time — a number remarkably close to the consensus I had just absorbed, arrived at by a route I would have sworn was independent.

You cannot un-see a number. Once 420 is in your head your valuation will land near 420, and you will experience this as confirmation.

Galton’s ox worked because the tickets were written in private. That is not a historical detail. That is the entire finding, and every serious investor’s process is, in the end, an elaborate apparatus for reproducing the conditions of a Plymouth livestock fair in 1906: form your estimate before you hear anyone else’s.

The Telegram group had a good week last Mangsir. Somebody posted about a manufacturing company on Sunday and it moved eleven per cent over three sessions, and by Wednesday there were two hundred messages, and by Thursday somebody had made a chart with arrows on it.

The company published its second-quarter results eighteen days later. Profit was down thirty-one per cent.

There were four messages about that.

The Man Who Cannot Be Wrong

In which a king is told he will destroy a great empire and does; 284 experts make 82,361 forecasts and perform about as well as chance; the most distinguished economist in America announces a permanently high plateau; an exchange-traded fund is created for the sole purpose of betting against a television presenter and loses money doing it; and a man in Kathmandu explains what the market did today.

Croesus, king of Lydia, was the richest man in the world and about to make a decision.

Cyrus of Persia was expanding westward and Croesus wished to know whether to attack first. Being a careful man he did not rely on his own judgment. He tested the oracles — sending messengers to each of the famous ones with instructions to ask, on an appointed day, what the king of Lydia was doing at that exact moment, so that he could establish which of them was genuine. Delphi answered correctly, describing a peculiar dish of lamb and tortoise that Croesus had prepared specifically because nobody could have guessed it.

Satisfied, he sent to Delphi with the real question. Should he march against Persia?

The Pythia answered that if Croesus made war on the Persians, he would destroy a great empire.

He marched. He was defeated. The great empire destroyed was his own.

Herodotus tells us that Croesus later sent his chains to Delphi and asked the god whether it was customary for Greek deities to be so ungrateful. The oracle replied that the prophecy had been perfectly accurate, and that Croesus ought to have asked which empire was meant.

I open with this because it is two and a half thousand years old and I have heard the identical structure on Nepali television within the last month. The oracle at Delphi was not a fraud; it had, as Croesus’s own test established, a genuine capability. What it had additionally developed, across centuries of practice, was the professional art of saying a true thing that cannot be scored.

That art has not changed. It has only acquired a studio.

The grammar of safety

Every evening at around seven, on one channel or another, a man sits in a chair and explains what the stock market did that day.

I will call him Mr. Pradhan. He is well dressed, he speaks well, and he is introduced as a senior market analyst — a title to which no examination is attached. He has been doing this for years and he is respected.

He has never been wrong.

I do not mean he has been right. I mean something more remarkable: in several years of watching I have not identified a single statement he has made that could have been wrong. Not one sentence which, at the moment of utterance, carried an outcome capable of embarrassing him.

This is not incompetence. It is a highly evolved professional skill and it took me a long time to appreciate how good he is at it. Here is roughly what he says, written flat on a page, because the structure only becomes visible when you can see it all at once:

”The market has corrected from its recent high and we are seeing some profit booking at these levels. If the index sustains above the support zone, we may see a recovery towards the previous resistance. However, if selling pressure continues, further correction cannot be ruled out. Investors should be cautious and invest in fundamentally strong companies for the long term.”

Count the escape hatches. May see a recovery. If it sustains. Cannot be ruled out. Every clause is conditional on something that has not happened, and the two branches of the condition cover, between them, the complete set of possible futures. Up, down and sideways have all been pre-approved.

Then the closing sentence, which is the finest part: invest in fundamentally strong companies for the long term. Unimpeachable. Also unactionable — it names no company, no price and no horizon — and possessed of the additional elegance that if you follow it and lose money, you evidently failed to select a fundamentally strong company, which is your fault.

A forecast that cannot fail is not a forecast. It is a mood set to music.

Eighty-two thousand forecasts

We do not have to argue about whether this class of person is useful, because somebody spent twenty years finding out.

Philip Tetlock, a psychologist, began in 1984 to collect forecasts from people who were paid to have opinions about politics and economics — academics, government advisers, journalists, think-tank fellows. Two hundred and eighty-four of them. He did not ask for vague pronouncements; he required answers in a form that could be scored, with probabilities attached and outcomes that would definitely resolve.

By the time he published in 2005 he had eighty-two thousand three hundred and sixty-one forecasts.

The headline result, which Tetlock delivered with a memorable and much-repeated image, was that the average expert performed about as well as a dart-throwing chimpanzee. More precisely: the experts barely outperformed simple statistical baselines, and on many questions they were beaten by an algorithm that just extrapolated recent trends.

That is bad. But it is not the finding that matters here.

The finding that matters is that fame ran the wrong way. The experts with the greatest media presence — the ones invited onto programmes, the ones whose names you would recognise — were, on average, less accurate than their obscure colleagues. Not equally accurate. Less.

Tetlock explained this using a distinction Isaiah Berlin had borrowed from a fragment of Archilochus: the fox knows many things, the hedgehog knows one big thing. Hedgehogs have a single organising theory and apply it everywhere with confidence. Foxes hold many partial models, contradict themselves, hedge, and update.

Foxes forecast better. Hedgehogs get invited back.

And that is not a paradox — it is a selection mechanism, and once you see it you cannot watch a panel discussion the same way again. Television requires someone who will say a clear, memorable, confident thing in ninety seconds. The fox says “it depends on three factors and I would put it at about sixty per cent.” The hedgehog says the market will reach five thousand. Only one of these is bookable.

The medium selects for the trait that anti-correlates with accuracy. Not occasionally. Structurally, every time, in every country.

The permanently high plateau

Lest this seem like a problem of pundits rather than of genuine experts, consider Irving Fisher.

Fisher was not a broadcaster. He was arguably the finest economist America had produced — the man who gave us the distinction between real and nominal interest rates, whose work on index numbers and the quantity theory of money is still taught, and whom Milton Friedman would later call the greatest economist the United States ever had.

On the fifteenth of October 1929 he announced that stock prices had reached what looked like a permanently high plateau.

The market broke nine days later. By 1932 the Dow had fallen about eighty-nine per cent from its peak. Fisher lost his own fortune and his wife’s and his sister-in-law’s, and spent the rest of his life explaining, in work that was itself excellent and which laid the foundation for how we now think about debt deflation, why he had been wrong.

I raise Fisher not to mock him — his subsequent work on debt deflation is more valuable than the plateau remark was harmful — but to close off an escape route. The problem is not that television books shallow people. Fisher was the deepest man in the room. The problem is that forecasting a market is not a thing that expertise makes you good at, and expertise supplies confidence anyway.

You cannot even bet against him

Here is my favourite item in the entire literature, and it is not from a journal.

In March 2023 an American asset manager launched an exchange-traded fund whose stated strategy was to do the opposite of whatever Jim Cramer, the loudest stock-picking presenter on American financial television, recommended. It was called the Inverse Cramer Tracker ETF. It had a ticker: SJIM.

The reasoning was that if a famous pundit is reliably wrong, fading him should be reliably profitable, and a good many people found this obviously true and funny.

The fund lost money and closed in February 2024, in under a year.

I regard this as the most instructive failure in modern finance, and the lesson is subtle enough that most people who tell the story miss it.

A pundit is not a contrary indicator. He is noise.

A contrary indicator would be useful — it would contain information, merely sign-flipped, and you could extract it by reversing the sign. What the experiment demonstrated is that there was no information there of either sign. You could not profit by following him and you could not profit by opposing him, because the content was zero, and zero has no sign to reverse. Multiply zero by minus one and you still have nothing, and you have paid the expense ratio for the privilege.

Hold that when you are tempted to think that the man on the evening panel is at least useful as a signal to do the opposite. He is not. He is weather.

The swap test

I want to hand you a tool, because it takes ten seconds and will permanently alter how you read every market report published in this country.

Take tomorrow’s wrap-up from any Nepali portal. Find the sentence that explains the day. Then reverse the direction and see whether the explanation still works.

”NEPSE gained 14 points as investors showed confidence following the central bank’s monetary policy.”

Now: ”NEPSE lost 14 points as investors booked profits ahead of the central bank’s monetary policy.”

Both are fine. Both would have been printed without anybody raising an eyebrow. The explanation is not attached to the outcome in any way that constrains it. It is selected afterwards, from a menu of available reasons, of which there are always at least six on any given day in a country where something is always happening.

The classics run in matched pairs. Market rises: value buying at lower levels. Market falls: profit booking at higher levels. These are the same sentence. They mean “people bought” and “people sold” — which is what a market is, described as though it were a cause.

My own favourite, and I collect these, is ”the market moved sideways as investors remained on the sidelines awaiting clarity”, which says that nothing happened because nothing happened.

I am not mocking the journalists. They have a column to fill by six and no information, because as established in the last chapter, on most days there is none. They are performing a ritual, and the ritual has a function: it reassures the reader that the market is a place where things happen for reasons.

That reassurance is precisely what costs you money. It teaches you, five days a week, that daily movements are explicable — and a man who believes daily movements are explicable will try to explain them, and then act on his explanation.

The free option

Let me set out the incentive structure plainly, because once seen it cannot be unseen and it accounts for the entire industry.

A public forecaster holds a free option.

If the call works, he is a man who called it. There will be a screenshot. It will circulate. His fee rises, his invitations multiply, his audience grows — which mechanically increases the number of people available to witness the next correct call.

If the call fails, nothing happens. There is no counterparty. Nobody comes to collect. The audience has moved on and the forecast has dissolved.

Unlimited upside, zero downside, no capital at risk. Offer that structure to any trader alive and he will take it and run it as hard as he can, and the optimal way to run it is exactly what you observe: make many bold calls, be loud about the winners, and never, under any circumstances, keep a list.

Compare the position of the listener. He has capital at risk. His downside is not zero. He is on the other side of an asymmetry so severe it would be considered scandalous in any regulated transaction, and it is conducted in public, for free, as a service.

I do not think Mr. Pradhan is a bad man. He is a man responding to incentives in a role constructed around him by a media economy that requires somebody in the chair at seven o’clock. If he began saying “I do not know” he would stop being invited within a month, and somebody else would occupy the chair and say the thing that gets you invited back.

The job is not forecasting. The job is filling the chair. He is excellent at his actual job.

What makes one worth listening to

There are people in this market whose work I genuinely value, and I can tell you exactly what separates them, because it is a single property.

It is not accuracy. Accuracy is unobservable in small samples; that was Chapter One.

It is this: they say things that could be wrong, and they say when.

A useful statement looks like: ”I think this bank is worth between 480 and 560 a share. It trades at 390. I expect that gap to close over about two years, and the thing that would tell me I am wrong is if the non-performing loan ratio has not come back below three per cent by the end of the next fiscal year.”

Look at what that does. A range, which is honest, because nobody knows a number. A horizon, so the claim expires. And a falsifier — a specific, checkable, publicly disclosed fact that would demonstrate the reasoning wrong independent of what the price does.

The last is rarest and worth the most, because offering it means that in eighteen months somebody may return and say: the ratio is four per cent, you were wrong. Which is unpleasant, and which is why the sentence is almost never spoken.

The willingness to be caught is the entire signal. Everything else is production values.

This is also, incidentally, what Tetlock’s superforecasters turned out to look like when he went hunting for the people who could forecast: not the most knowledgeable, but the ones who assigned numerical probabilities, updated them in small increments as evidence arrived, and kept score on themselves.

When you meet someone who talks this way, follow them for years, because they have handed you the means to grade them. When you meet someone who does not, it does not matter how impressive they are. There is nothing there to grade.

Now let me implicate myself

It would be comfortable to end here, having established that I stand outside all this. I do not, and the mechanism from the inside is worth showing.

I run a valuation process. It produces a range for a company — a number with a band around it — through models I trust and have tested. That process is dramatically better than a television panel and I will not pretend otherwise.

But I have caught myself, more than once, doing something with exactly Mr. Pradhan’s shape, wearing a better suit.

It looks like this. I value a company. The number comes out below the market price. I do not like this, because I already own it, or want to own it, or because the last three companies I examined were also expensive and I am bored. So I go back into the model and find an assumption that is “conservative” — the growth rate, the normalised return, the fade period — and adjust it to something still defensible.

Every one of those adjustments is defensible. That is the point. I never once have to write down a number I believe to be false. I simply choose, repeatedly, from the defensible end of the range in the direction I want, and the output arrives where I intended.

It is the same free option. Unlimited flexibility on the inputs, and no counterparty checking whether I applied it consistently.

The only remedy I have found is procedural and unglamorous: fix the inputs before you look at the price. Write down the growth rate, the required return, the normalisation, all of it — and only then compute the value, and only then look at what the market says. If you look at the price first you are no longer valuing a company. You are constructing a justification, and you will construct an excellent one, because you are clever and the entire defensible range is available to you.

I break this rule perhaps one time in five. That is my honest estimate and it is probably flattering.

The most expensive question in Nepal

There is one question asked in every group, at every wedding, on every panel, and it is the question this chapter exists to disarm.

”Sir, market kata jancha?” — where is the market going?

There is no answer. Not “the answer is difficult.” There is no answer, in the sense that the question has no content: the market is going up and down simultaneously, in different amounts, over different horizons, for different people, and the only honest completion of the sentence is a probability distribution that nobody wants.

But it will be asked of you, and you will feel the pull to answer, because saying “I do not know” in Nepali company sounds like an admission of not having done your homework.

Here is what I have started saying, and it is not a dodge:

”I do not know where the market is going. I know roughly what four or five companies are worth, and I know what they cost, and that is all I need.”

The first time you say it aloud it feels like a defeat. It is the opposite. It is the only sentence in this chapter that anybody could be held to.

Mr. Pradhan was on last week. The index had fallen for three sessions and he explained that this was a healthy correction and that accumulation at these levels was advisable for the long-term investor, though caution was warranted in the short term.

The host thanked him for his valuable analysis.

Then they went to the weather, where a woman gave a number and a probability and stood behind it.

The Pendulum Does Not Rest in the Middle

In which a man spends sixteen months raising eleven billion dollars and does nothing with it until the world breaks; Warren Buffett is publicly declared finished; a magazine puts the question on its cover; and we establish that knowing where you stand tells you a great deal and knowing when tells you nothing at all.

Between January 2007 and May 2008 Howard Marks raised the largest distressed debt fund in history — $10.9 billion — and then sat on it.

He did not deploy it. Its investment period did not begin until the raise closed, and for the length of that raise, while credit markets remained buoyant and his investors’ capital earned very little and his competitors were putting money to work at spreads that looked adequate, Oaktree waited. Marks had been writing memos to clients since the early 1990s and the memos of that period say plainly what he thought: that risk was being priced as though it had been abolished, that lenders were accepting terms they would regret, and that this condition was unsustainable without specifying when it would end.

Lehman Brothers failed on the fifteenth of September 2008.

In the fifteen weeks that followed, Oaktree deployed more than half of what that fund would ever draw — some $5.3 billion of an eventual $9.8 billion — into assets that were being sold by people who had no choice about selling. The returns from that deployment are the reason Marks is a figure in this business rather than merely a successful one.

I want to be exact about what he did and did not do, because the story is routinely told wrong.

He did not predict the crash. He has said so repeatedly and in writing, and his own memos confirm it: he did not know what would break, or when, and had he attempted to time it he would have been wrong like everybody else. What he did was observe that the price of risk had become absurd — an observation about the present, verifiable at the time, requiring no clairvoyance — and then arrange his affairs so that he would have money when other people did not.

That distinction is the entire content of this chapter and almost nobody holds it properly, so let me state it in its harshest form.

A forecast and a position are different objects

A forecast says: the market will fall next year.

A position says: the market is currently expensive, people are behaving as people behave when things are expensive, and therefore the distribution of what can happen to me from here is worse than it was two years ago — so I will own less and hold more cash.

The first is a claim about the future and is almost always wrong. The second is a claim about the present and is frequently right, because the present is observable.

This distinction is easy to state and extraordinarily hard to keep hold of, because the moment you say “the market is expensive,” the person you are talking to hears “you think it will fall,” and twelve months later, when it has risen forty per cent, he informs you that you were wrong.

And you were not wrong. You made a statement about the present and he graded it as a statement about the future.

Knowing where you are in a cycle tells you about the distribution of what can happen. It tells you nothing whatsoever about when.

If you cannot hold those apart, cycles will be useless to you — worse than useless, because you will act on them as timing signals, be punished, and conclude that they do not work.

The pendulum

The organising image is Marks’s and I make no claim on it. Markets swing between excessive optimism and excessive pessimism, and they spend remarkably little time at the sensible midpoint. They pass through it on the way from one extreme to the other, the way a pendulum passes through the bottom of its arc without pausing there.

John Templeton put the same observation in a sentence that has never been improved: bull markets are born on pessimism, grow on scepticism, mature on optimism, and die on euphoria.

Templeton is worth a moment, because he did the thing rather than merely describing it. In 1939, with Europe going to war and the American market still broken from the Depression, he borrowed ten thousand dollars and bought a hundred shares of every company on the New York and American exchanges trading below one dollar. There were a hundred and four of them. Thirty-four were in bankruptcy at the time of purchase.

Only four became worthless. He roughly quadrupled his money over the following four years.

Note what that operation required, because it was not insight into any of the hundred and four companies — he could not possibly have analysed them all, and did not try. It required two things: a correct reading of the general temperature, and the ability to act while the newspapers said that Europe was ending.

Being early is indistinguishable from being wrong

Now the cost of thinking this way, which the books do not adequately warn you about.

If you become skilled at reading temperature, you will become cautious before the top. Not at the top. Before it. And the gap between “before” and “at” is where the largest gains of any cycle are made.

The canonical demonstration is Warren Buffett in 1999.

In July of that year he gave a talk at Sun Valley, later expanded into a Fortune article, in which he laid out with great care why the returns investors were anticipating from American equities were arithmetically implausible. His argument was not a market call; it was an observation about corporate profits as a share of the economy, interest rates, and the price being paid for the combination. It was substantially correct and it has held up beautifully.

At the time, Berkshire Hathaway shares fell about twenty per cent over 1999 while the NASDAQ rose eighty-six per cent.

In December 1999, Barron’s put him on its cover under the headline ”What’s Wrong, Warren?” The article suggested, in the courteous manner of financial journalism, that the man had lost his touch and failed to understand the new economy.

The NASDAQ peaked in March 2000 and subsequently fell about seventy-eight per cent.

Jeremy Grantham lived the same episode with less protection. He had refused to hold technology stocks on valuation grounds and a very large part of his client base left him for it — not because they disagreed with his analysis, in many cases, but because underperforming for three consecutive years is not a thing an institutional committee can defend to its own board.

A man who is early and a man who is wrong look identical, and they look identical for exactly as long as it takes to find out.

There is no version of this in which you escape that period. Nobody has ever solved it, and anyone who claims to has described one instance and concealed four. If you position for a cycle you will spend some interval — sometimes a year, sometimes three — being demonstrably, visibly, publicly wrong, while your friends make money and are kind about it, which is worse.

The only thing that survives that interval is having written down why. Not conviction; conviction erodes under social pressure faster than anyone expects. Written reasons. Dated. So that in month fourteen you can go back and check whether the reasons have changed or only the price. If the reasons have changed, change your mind — that is the job, not a weakness. If only the price has changed, then nothing has happened except that you are uncomfortable, and discomfort is not information.

What the starting price actually tells you

There is one respectable piece of evidence that valuation carries information about returns, and it deserves stating precisely because it is usually overstated.

Robert Shiller’s cyclically adjusted price-earnings ratio — earnings averaged over ten years, to smooth the cycle out of the denominator — has a reasonably robust relationship with subsequent long-run returns in American data. Buy the market when that ratio is low and your following decade tends to be good; buy it when the ratio is high and your following decade tends to be poor. The relationship is not tight, but it is there, and it has survived out of sample.

Now the two things everybody gets wrong about it.

It says nothing about next year. The correlation with one-year returns is essentially nothing. A market can be expensive and become far more expensive, which is what 1996 to 2000 consisted of.

And it is a statement about a starting price, not a signal to act. The useful translation is not “sell” but “expect less” — which changes how much you save, what you promise yourself about retirement, and how much risk you need to take, and changes none of those things this afternoon.

The honest summary of a century of evidence is this: valuation predicts returns over a decade and predicts nothing over a year. Anyone using it for the second purpose is misusing a real tool, and anyone dismissing it because it failed at the second purpose is discarding a real tool.

The room is small

The pendulum swings everywhere. What I want to add is what happens to it in a room this size.

Nepal has a few hundred listed companies, of which perhaps a hundred and fifty trade with any regularity. The genuinely floating shares are a fraction of those. There is no institutional money to speak of — no pension funds running equity mandates, no foreign investors, almost nothing that behaves differently from everybody else. There is no way to bet against anything. And there is a single dominant alternative to owning shares, which is a bank deposit.

Put that together and you have a market in which nearly all participants are the same kind of participant, doing the same kind of thing, at the same time, on the same information, with only one direction available.

The pendulum still swings. It simply swings further, and it strikes the walls.

The evidence is in the index and it is stark. Over twenty-nine years NEPSE has had four proper collapses:

A three-quarters decline is not a correction. That is what happens when everybody in a small room decides the same thing on the same afternoon and there is nobody on the other side.

The two questions

Only two questions about the cycle are worth asking, and both are in the present tense.

Where are we?

And what is that paying me, or costing me, to own things here?

The second is the one people skip, and it is the more important, because it converts an observation into an action.

Suppose you decide the market is hot. Fine. What do you do? If the answer is “sell,” you are back to forecasting, and you will sell in 2019 and watch the index triple. The better answer is a question: given that things are hot, what am I being paid to take risk right now, and is it enough?

In a cold market you are typically paid a great deal. Shares trade below what the businesses are worth on any reasonable calculation, dividend yields are high against price, and the fixed deposit rate — the thing you are giving up — is often low, because the central bank is trying to revive the economy. You are paid handsomely to be brave and it costs you little to be.

In a hot market, the reverse. Shares trade above any defensible calculation, dividend yields compress, and the deposit rate is often high, because the central bank is trying to cool things down. You are paid very little to take risk, and the safe alternative is paying you more than usual.

Notice that this is not a forecast. Every quantity in it is observable today. The cycle does not tell you what will happen. It tells you what you are being paid, and you can compute that this afternoon.

Taking the temperature without a thermometer

The index level alone tells you almost nothing. Two thousand five hundred is neither high nor low; it depends on what the companies earn and what the alternative pays, and Part Four is largely about doing that arithmetic properly.

But there is a second reading, behavioural, and in my experience at least as reliable and considerably faster. You are not measuring the market. You are measuring the people.

Who is talking about shares. The oldest indicator in the world and it still works. In a cold market nobody mentions the market at a wedding. In a hot one you will be asked for a tip by someone who has never held a share — and at the very top you will be given one by that person.

What happens to new issues. When an ordinary offering is oversubscribed many times over and people are opening accounts specifically to apply for allotments, something has changed about who is in the market. Not the valuation: the composition.

How fast new accounts are opening. This is published. When the number grows very fast, the market’s marginal buyer has been investing for under a year, and a market whose marginal buyer is inexperienced behaves differently from one whose marginal buyer is not.

Whether people are borrowing to buy. Margin lending against shares is the cleanest signal available, because it converts a decline into forced selling. In a cold market almost nobody borrows to buy shares. In a hot one, the loan is described as leverage rather than as debt.

The vocabulary. This one is subtle and I trust it more than the others. Listen for what people believe the risk is. Near a bottom, the risk everybody discusses is losing money. Near a top, the risk everybody discusses is missing out. It is the same word performing opposite work, and when you hear the second version used without irony by people who are otherwise sensible, you are late in a cycle.

How stories about companies are told. In a cold market people talk about dividends and book value. In a hot one they talk about potential, about government plans, about what a sector could become. When the median conversation about a company contains no number, the temperature is high.

None of these is a timing signal. Every one can be present for two years before anything happens. That is not a flaw in the method; that is the method. You are describing where you stand, not when you will fall.

The arithmetic that makes it matter

There is one respect in which cycles are structurally, unavoidably asymmetric, and it is the most practically useful paragraph in the chapter.

Losses and gains are not symmetric in their effect on capital.

Down fifty per cent requires a hundred per cent gain to recover. Down seventy-five per cent — which Nepali holders were in 2011 — requires a quadruple. This is not theory. It is division.

Now set that beside the history. From the 2008 peak of 1,175.4 the market fell to 292.0 by June 2011 and did not see its old high until August 2015: seven years, peak to peak. From the 2021 peak of 3,198.19 it fell to 1,815.10, and as I write, five years later, it has not recovered.

Of the six thousand five hundred and fifty-one sessions NEPSE has ever held, only seventy-two closed at or above 2,900. One point one per cent of all the days there have ever been.

That statistic is the whole argument for caring about cycles, and it has nothing to do with prediction. It is about the fact that the top of a market is a thin place. Very few days occur there. If you are fully invested at the top you were not unlucky; you were standing on a narrow ledge that almost nobody stands on, and the odds of being there when the music stopped were far higher than they felt.

The bottom is thin too, and being fully invested at a bottom is the single most valuable thing that can happen to a portfolio in a lifetime. Which is why the useful objective is not to time anything.

It is to still have money when the market is cheap.

That is a far lower bar than prediction, and almost nobody clears it — because having money when the market is cheap requires not having spent it when the market was expensive, which requires reading a temperature and acting mildly on it, years early, while looking foolish.

Which is precisely what Marks did from 2007, and precisely what the eleven billion dollars was for.

What I actually do, and modestly

I do not go to cash. I have never gone to cash and do not intend to, because going to cash is a forecast wearing a costume, and it requires a second correct decision — when to return — which is harder than the first and which almost nobody makes.

What I do is tilt. When the temperature is high I hold more cash than usual and let it accumulate rather than deploying it; I stop adding to positions that have run; I let dividends sit. When the temperature is low I deploy, and into the things the market has abandoned rather than the things that held up.

The tilt is small. Something like fifteen or twenty per cent of the book moves between the extremes, not the whole thing. That is enough to matter over a full cycle and small enough that being wrong about the temperature does not destroy me.

And I do the arithmetic. Every time I feel the pull to announce that the market is expensive, I make myself compute what I am actually being paid — the earnings yield against what a fixed deposit pays this year, the dividend yield, the price against a defensible estimate of value. Sometimes the arithmetic disagrees with the mood.

When it does, I go with the arithmetic. The mood is downstream of the same group chats as everybody else’s.

The pendulum is not a metaphor about the future. It is a description of where the weight is right now.

You cannot know when it will swing back. You can know, with fair confidence, which side of the middle you are standing on — and that is enough to change what you own, which is the only decision you actually get to make.

The Narrowing

In which an economist nobody read until he died explains every bubble in history; the American market tops out two years before its index does; a shoeshine boy gives Bernard Baruch a stock tip, or possibly does not; and on the sixth of September 2021, a quarter of the Nepal Stock Exchange locks limit-up while the market is already falling.

Hyman Minsky spent his career at Washington University in St. Louis being largely ignored, and died in 1996 having been cited by almost nobody outside a small circle. Twelve years later, in the autumn of 2008, every financial journalist on earth suddenly wanted to talk about the Minsky moment, which is a phrase he never used and would probably have disliked.

His argument, stripped to its bones, is that stability is what causes instability.

He divided borrowers into three kinds. Hedge finance: the borrower’s cash flow covers both interest and principal. He is fine, and he is boring. Speculative finance: the cash flow covers the interest but not the principal, so the debt must be rolled over, and the borrower is fine as long as somebody will keep rolling it. Ponzi finance: the cash flow covers neither, and the borrower is solvent only if the asset’s price keeps rising, because the only way out is to sell it to somebody else.

Minsky’s insight was that an economy does not sit in one of these states. It migrates between them, and it migrates in one direction. A long period without losses convinces everybody — lenders, borrowers, regulators, all of them behaving reasonably at each step — that the previous standards were too cautious. Terms loosen a little. Nothing bad happens, which appears to vindicate the loosening. So they loosen again.

The absence of disaster is taken as evidence that disaster is unlikely, and that belief is what produces it.

Charles Kindleberger took this and applied it to four centuries of financial history in Manias, Panics and Crashes, and found the same sequence every time: a displacement (some genuine change that makes the future look different), a boom funded by expanding credit, euphoria in which new entrants arrive and price ceases to relate to anything, distress as the informed begin to leave, and finally revulsion.

I have read a great many descriptions of the Nepali bull market of 2077 and 2078, and none of them is as accurate as a framework written by a dead American about Dutch tulips.

The displacement

Every mania begins with something real. This is the part people forget when they use the word “bubble” as an insult.

Nepal’s displacement was not one thing but three, arriving simultaneously, and any one of them alone would have mattered.

The exchange went online. For the entire previous history of this market you had to physically visit a broker or telephone one, and that friction was a filter — it excluded the casual, the young, the geographically distant and the merely curious. The trading management system removed the filter.

The country was locked in its houses. NEPSE closed on the twenty-second of March 2020, reopened for one session on the twelfth of May, closed again, and finally resumed on the twenty-ninth of June. In the whole of 2020 it held a hundred and eighty-one sessions against a normal year’s two hundred and thirty. During that closure several million people sat at home with a telephone and nothing to do.

And the money had nowhere to go. This is the part that was genuinely counterintuitive and that almost nobody predicted: remittances did not collapse. Nepalis working abroad, facing uncertainty, sent more money home rather than less. Meanwhile domestic spending stopped — no weddings, no travel, no construction, no shops. Deposits accumulated in the banking system by default, because the ordinary channels for spending had been switched off, and deposit rates fell accordingly.

Displacement, in Kindleberger’s sense, exactly. Something real had changed. The people who bought in the middle of 2020 were not fools; they were responding to a genuine alteration in the world.

That is always how it starts. A mania is not a mistake at the beginning. It is a correct observation held past its expiry.

Stage one: disbelief, which is the profitable one

The early part of a boom does not feel like a boom. It feels like an error.

Through the second half of 2020, as the index passed its old levels, the dominant emotion among people who had been in this market for a while was suspicion. It did not make sense. The economy was in real trouble. Businesses were shut. Everybody knew somebody who had lost work. How could shares be rising?

I heard the phrase yo ta bubble ho — this is a bubble — in Ashoj 2077, with the index around 1,500.

It went to 3,198.

This is the reliable feature of stage one and it deserves to be understood properly. Early in a boom, the most experienced participants are the most sceptical, their scepticism is well-founded, and it is completely unprofitable. They have seen 2008. They know what a market that has run looks like. They are correct about all of it and early by two years.

Meanwhile the newest entrants — no memory, no scars, no framework — buy, and are rewarded, and are rewarded again.

There is a cruelty in the design. The market spends the first phase of every boom systematically punishing judgment and rewarding its absence, and it does so for long enough that the punishment begins to feel like a verdict on the judgment.

Stage two: the arrival

Somewhere around the turn of 2077 into 2078 the composition of the market changed. You could feel it before you could measure it.

Demat accounts opened at a rate nobody in the country had seen. People who had never owned a share were opening accounts from their telephones, at home, during a lockdown, at an age when the alternative entertainments were unavailable.

The vocabulary appeared next. Circuit. Support. Breakout. Words that had existed among a few thousand people became words that existed among a few hundred thousand, used with total fluency by people who had acquired them three weeks earlier.

And the market began to behave the way a market behaves when its participants are all new. Here is that statement converted into a measurement.

In 2019 — a cold, ordinary, forgettable year — the average session had 0.45 per cent of listed stocks closing at the ten per cent daily limit.

During the run-up from June 2020 to August 2021, that figure was 2.22 per cent.

Five times as many stocks locked at the limit, every day, for over a year.

Stage three: when doubt becomes expensive

This is the part that is hardest to convey to somebody who was not there.

At a certain point in a boom — and it is not gradual, it happens across a few weeks — being cautious stops being a respectable position and becomes a social liability.

In Baisakh 2077 you could say “I think this is expensive” and people would nod. By Falgun 2077 you could say it and people smiled at you the way one smiles at an elderly relative who has said something out of date. By Jestha 2078 you did not say it, because it made the room uncomfortable, and because the man you were saying it to had made eleven lakh rupees since you last saw him and you had not.

This is not weakness of character, and “discipline” is the wrong frame entirely.

A boom converts a financial question into a social one. The question stops being is this share worth 640 rupees and becomes am I the sort of person who understands what is happening, or the sort who is being left behind by it. Nobody in the history of the world has answered that second question in the negative about himself.

I watched genuinely careful people — with more experience than me, whose analysis I respect — buy things in mid-2078 they would not have glanced at eighteen months earlier. They had reasons. The reasons were elaborate. And the reasons had all been constructed in the preceding four months.

I bought things too. I am not writing this from outside.

The shoeshine boy, and a warning about him

The folkloric marker for this stage is Bernard Baruch’s shoeshine boy: the financier who supposedly sold out before the 1929 crash because the boy polishing his shoes offered him stock tips, and he concluded that when the shoeshine boy is in the market, the market is finished. The same story is told of Joseph Kennedy.

I repeat it because it captures something true, and then I want to spoil it, because spoiling it captures something truer.

The story is almost certainly retrospective. It appears in memoirs written after the crash, by men whose reputations depended on having seen it coming. There were shoeshine boys with stock tips in 1926 and 1927 as well, and the men who sold then missed three years of the greatest bull market in American history, and none of them wrote memoirs about it.

The shoeshine boy story is itself a survivor’s story — the exact object dissected in Chapter One, deployed as evidence for a claim about tops. It tells you what a top looks like using a sample selected on the outcome.

Which does not make the underlying observation useless. It makes it a thermometer rather than a clock, which is the distinction from the last chapter, and which is why I gave you the last chapter first.

What it looked like on a screen

The mechanics of the frenzy were visible in a way I have not seen before or since.

A stock would open and be immediately limit-up. Not rise to the limit — open there. There would be a queue of buy orders and no sellers at all, and the quantity queued would be several times the number of shares that had traded in the preceding month. Nothing would trade. The stock would sit at plus ten per cent for four hours, untouchable, and close there.

And do the same thing the next day.

You would sit with money in your account watching a price rise, unable to buy any of it. There is no experience in finance quite like it. The thing you want is on the screen, priced, apparently available, and it is not available at all, and the only action open to you is to join tomorrow’s queue behind two thousand people who thought of it first.

What this does to a person is specific. It converts the desire to own into an urgency to secure, and urgency is fatal to price discipline. When you are finally filled, three days and thirty per cent later, you feel you have won something.

The top was not an event

Here is the thing I most want you to take from this chapter.

The eighteenth of August 2021 was a Wednesday. The index closed at 3,198.19, the highest it has ever closed, and has not returned in the five years since.

Nothing happened that day.

No announcement. No policy. No scandal. No crisis. Read the market reports from that week and they say what market reports always say — some profit booking, some caution, some investors awaiting clarity. There is nothing in them that marks the day. Nobody rang a bell.

The next session it closed at 3,180.86 and everybody assumed it was a pause.

Tops are not made by news. They are made by the exhaustion of buyers, and the exhaustion of buyers is invisible, because a buyer who has run out of money looks exactly like a buyer who is waiting.

The narrowing, measured

And now the thing I did not understand at the time and consider one of the most useful facts I know.

The index turned on the eighteenth of August. The frenzy did not.

On the sixth of September 2021 — nearly three weeks after the top, with the index already 6.8 per cent below its high at 2,980.79 — forty-seven of the one hundred and ninety-five stocks trading that day closed limit-up.

Twenty-four per cent of the market. Locked at the limit. In a single session. After the top.

That was the most speculative day of the entire cycle, and it occurred while the market was already falling.

This is what a top actually looks like from inside. The large, sensible, heavily weighted companies stop rising first, because the people who own them are the people who look at valuations. The index — mostly those companies — turns. And the money does not leave. It migrates, downward, into smaller and smaller and more speculative names, where the float is thin enough to still produce the sensation of a rising market.

If you held big banks in September 2021 you experienced a market that had stalled. If you held small hydropower companies you experienced the best three weeks of your life. You were in the same market on the same days, and one of you was watching the top happen and the other was celebrating it.

Narrowing is the tell. When fewer and fewer things are rising, and the things rising are those furthest from any calculation of value, and volume in the largest companies is drying up — that is the sound a market makes when it is running out of buyers.

And this is not a Nepali phenomenon. It is the most reliable regularity in the history of market tops.

In the United States, the advance-decline line — a simple running count of how many stocks rise against how many fall — peaked in April 1998. The S&P 500 peaked in March 2000. For nearly two years the index climbed while the majority of its constituents did not; in 1999 the index rose about twenty per cent while more NYSE issues fell than rose. The market had already topped. The index had not, because the index is an average and the average was being held aloft by a shrinking handful.

The same thing ran in 2021, in real time, alongside Nepal’s. Speculative American assets — the special-purpose acquisition companies, the high-growth technology funds — peaked in February 2021. The S&P 500 peaked in January 2022. Eleven months of index strength on top of a speculative market that had already died.

Narrowing precedes the top by months in a large liquid market and by weeks in a small one. It is descriptive rather than predictive, and it is only clear afterwards to the extent that you were counting at the time.

Which is the whole argument for counting.

The honest part

I do not want to leave the impression that a boom is nothing but a trap, because that would be a lie by omission.

A great deal of real money was made between 2077 and 2078, and much of it stayed made. People who bought in the disbelief phase, when buying felt stupid, and who sold something — anything — on the way up, are permanently better off. Some of them bought land with it. Some educated children with it. That money is real and it is still there.

The tragedy is not that the boom happened. It is in the arithmetic of when people arrive.

The largest number of new participants entered this market in the twelve months before August 2021. That is what a boom is — it draws people in proportion to how far it has already run. Which means the quantity of rupees that entered near the top vastly exceeds the quantity that entered near the bottom, which means that the average participant in the greatest bull market in Nepal’s history bought somewhere near the end of it.

The index rose 174 per cent over that period. The average person who lived through it did not.

Both sentences are true at once. Holding them together is the entire skill.

I was at a tea shop in Putalisadak in Bhadra 2078, a week or so after the top, though of course nobody knew that yet. A man at the next table was explaining to two younger men that the index was going to five thousand, and that the only mistake available now was selling too early.

He was not a fool. His arithmetic about the deposit rate and the money supply was better than most of what I heard on television that year, and I remember thinking he had understood something real.

He had. Minsky would have recognised him immediately: a man correctly describing the displacement, at the exact moment the displacement had finished paying.

Five thousand. The index has not seen three thousand two hundred since.

The last thing I heard him say, as I was paying, was that the correction we had just had was healthy.

No Bad Day

In which the American market falls a fifth in a single afternoon; Japan loses four fifths of its value without ever having a bad day at all; ten thousand brokerage accounts reveal that people sell the wrong things; and NEPSE destroys three quarters of itself in instalments of half a per cent.

On Monday the nineteenth of October 1987, the Dow Jones Industrial Average fell 22.6 per cent.

Not over a month. In one session. A fifth of the value of American corporate equity ceased to exist between breakfast and dinner, with no war, no bank failure, no announcement, and to this day no agreed explanation. The largest single-day decline before it had been 12.8 per cent, in October 1929, and everyone had assumed that was a historical curiosity produced by an era with no safeguards.

That is a crash. It is a specific thing and it has specific properties. It arrives on a date. It is on the front page. Everyone you know remembers where they were. And afterwards — this is the part nobody notices — everybody has a shared reference point. People say “I sold after the crash” or “I bought after the crash,” and the crash is a fixed object in the calendar that conversations can be organised around.

The United States responded by installing circuit breakers, which halt trading when the index moves too far too fast, on the reasoning that a market pausing to catch its breath makes better decisions than one in free fall. India has them. Most exchanges now do. Nepal has them too, at five and eight per cent on the index, plus a limit on each individual stock.

I want to spend this chapter on what those safeguards actually purchase, because the price is not what anybody expects.

The crash is the small part

First, an inconvenience for the popular history.

The Dow peaked at 381.17 on the third of September 1929. The famous days — the twenty-eighth of October at minus 12.8 per cent, the twenty-ninth at minus 11.7 — are the ones in every documentary.

The Dow bottomed on the eighth of July 1932 at 41.22.

That is a decline of 89 per cent, and the two famous days account for under a quarter of it. The rest arrived across thirty-four months of ordinary sessions in which nothing in particular happened, and by the end there was nobody left to photograph. The crash was the overture. The destruction was a slow movement.

The same shape in our own time. The NASDAQ peaked at 5,048.62 on the tenth of March 2000 and bottomed at 1,114.11 on the ninth of October 2002. Down 78 per cent — over thirty-one months, in a long series of rallies and relapses, each rally sufficiently convincing to draw money back in. There was no single day on which it became obvious. There were about eight days on which it became obvious, spread over two and a half years, and each was followed by a recovery that made the previous obviousness look premature.

And then Japan, which is the pure case, and which I raised in Chapter One for a different purpose.

From 38,915.87 at the end of 1989, the Nikkei fell to 7,607.88 by April 2003. Eighty per cent, over thirteen years. There was no Japanese Black Monday. There is no date. No photograph. No documentary. An entire generation’s savings were destroyed at a rate of roughly half a per cent a month, and because it never happened on any particular day, it never quite happened at all.

Ask a Japanese saver of that generation when the bad thing occurred and he cannot answer, because there was no bad thing. There were four thousand ordinary days.

Nepal has never had a crash

Now to the local case, and it is more extreme than Japan’s.

Across six thousand five hundred and fifty trading sessions from 1997 to 2026, exactly twenty closed more than five per cent below the session before. Twenty, out of six and a half thousand. Three days in a thousand.

The worst single day in the history of this exchange was the third of July 2000, at minus 9.25 per cent. Nothing since 2007 has come near it; the worst of the modern era is about minus six.

Now hold that beside the other fact. Between August 2008 and June 2011 the NEPSE index fell from 1,175.4 to 292.0.

Seventy-five per cent of the market’s value disappeared, and the worst single day of the entire three years was minus 4.17 per cent.

Six hundred and thirty-nine sessions. Four hundred and one of them down. Fourteen worse than minus three per cent. The average down day was small enough that describing it to somebody would elicit a shrug.

The most recent one has the same signature. From the peak on the eighteenth of August 2021 to the trough on the twenty-fifth of September 2022: minus 43.2 per cent over two hundred and seventy-one sessions, a hundred and sixty-two of them down, eight worse than minus three per cent, the worst being minus 4.06.

What the safeguard actually costs

The mechanism is the daily price limit, and its effect is genuinely double-edged in a way that deserves more thought than it gets.

A stock cannot fall more than ten per cent in a session — fifteen since April 2026. The index halts at five per cent and again at eight. So a panic, a real one, everybody out at once, is not expressible in this market. The price simply stops moving and the sellers form a queue.

The intention is protective and I understand it. Nobody wants a market in which a family’s savings can fall thirty per cent between lunch and closing.

But consider what it does to the experience of a decline.

In a market without limits, a bust delivers a day. One horrible, unforgettable day when everything falls at once and the telephone rings and people you know are ruined. That day is terrible. It is also a marker.

In Nepal, the decline arrives in slices of half a per cent.

Every individual day is survivable. No individual day requires a decision. And the sum of them takes three quarters of your money.

There is never a moment. That is the whole of it. There is no afternoon on which it becomes obvious something has broken, because nothing ever breaks — it merely weighs, a little more each week, for two or three years, and every week your position is slightly worse and there has still never been a reason to act.

The circuit breaker does not prevent the loss. It prevents the event. It converts a catastrophe into a climate, and human beings are equipped to respond to catastrophes and are almost entirely unequipped to respond to climates.

Ten thousand accounts

At this point the argument requires evidence about what people actually do, rather than what I think they do, and fortunately somebody went and looked.

Terrance Odean obtained the records of ten thousand accounts at an American discount brokerage covering 1987 to 1993 — real trades by real people with real money — and asked a simple question: when an investor sells, is he selling something that has risen or something that has fallen?

The answer was that investors realised their gains at a rate roughly fifty per cent higher than they realised their losses. Confronted with a winner and a loser and a need for cash, people sold the winner.

Hersh Shefrin and Meir Statman had named this the disposition effect in 1985, and its root is in Kahneman and Tversky’s observation that losses are experienced roughly twice as intensely as equivalent gains, and that a loss is not a loss until you have realised it. Holding the loser keeps the question open. Selling it closes the question with the wrong answer.

Now consider what this does to a portfolio across a full cycle, mechanically, without anybody making a single conscious decision.

The positions that fell a little come back to your purchase price and you sell them, relieved. The positions that fell a great deal never come back, so you hold them. Repeat across a decade and your portfolio has been distilled, slowly and automatically, down to precisely the things you were most wrong about.

I have never met a Nepali investor who does not do this. I do it. The only defence I have found is to force the question into a different form: if I held cash instead of this share today, would I buy this share at this price? If the answer is no, then holding is the same decision as buying, and I have just made it by default.

The stages, in order

Two of these I have now been through, and the sequence is remarkably consistent — because it is not really about markets. It is about how a person defends a picture of himself.

Stage one, three to six months in: this is a correction.

The word does an enormous amount of work. It implies the previous level was correct and this is a temporary deviation, which is exactly backwards. Nobody says “this is a correction of my belief.” They say “the market is correcting,” as though the market were a student who had misbehaved.

At this stage people buy more. It is called averaging down and it feels like courage. Sometimes it is: in Mangsir 2078, with the index near 2,700, adding was reasonable. It is still under water five years later.

Stage two, six months to a year: I will sell when it comes back to my price.

The single most expensive sentence in Nepali investing, and worth dismantling properly.

The price you paid is a fact about your history. It is not a fact about the company and certainly not about the market, which does not know your name. But it operates in your head as a level — a place the price ought to return to — and the resolution to sell there feels like a plan.

It is not a plan. It is a hostage negotiation with yourself, and it is the disposition effect wearing a suit. It guarantees you will hold your worst positions longest.

Stage three, year one to year two: silence.

People stop talking about the market. The group chats thin. The television panels continue but nobody watches. The man who explained the boom to you at the tea shop is still there and does not mention shares.

Volumes fall, dramatically. A market that traded on excitement trades on nothing, and stocks that used to lock limit-up sit untouched for days. Liquidity, which felt infinite in 2078, turns out to have been a phase rather than a property.

Stage four: the account is not opened.

The most common terminal state of a Nepali investor is not selling at a loss. It is not opening the application. The shares sit in the demat account, dividends arrive and go to the bank, the password is forgotten, and the position persists outside of thought, sometimes for a decade.

I do not think this is stupid. It is a rational response to an unbearable situation: as long as you do not look, the loss is not realised in the only ledger that matters, which is the one in your head. People do the same with medical tests.

It is also extremely expensive, and the expense is invisible, because the alternative use of the money is invisible too.

Where we actually are

The index peaked at 3,198.19 on the eighteenth of August 2021 and bottomed at 1,815.10 on the twenty-fifth of September 2022.

Then it recovered, after a fashion. It reached 3,000.8 in August 2024. It reached 3,002.1 in July 2025. It reached 2,960.4 in March 2026.

Three times to the door. Three times back.

As I write, at the end of July 2026, the index stands at 2,672.6. From the 2021 peak that is a compound rate of minus 3.56 per cent a year over four years and eleven months.

Anyone who bought the Nepali market at its high has now spent five years compounding backwards, while a fixed deposit paid them between five and eight per cent a year for doing nothing at all.

That gap — not the 43 per cent fall, the gap against the alternative — is the real cost, and almost nobody computes it, because a brokerage statement shows you your loss and never shows you the deposit account you did not open.

The comparison that ought to frighten you

There is a version of the last five years people tell themselves: it fell, it recovered most of the way, it will get there.

Look at 2008.

That peak was 1,175.4 on the thirty-first of August 2008. The index did not close above it again until the nineteenth of August 2015.

Seven years.

And the falling part was only the first three. The other four were the flat part — the part where nothing happened, where the market drifted between 300 and 900, where the newspapers stopped covering it, where a generation of Nepali savers concluded that shares were a fraud and put their money into land. Which is one reason the land in your neighbourhood costs what it costs.

The 2000 peak took six and a half years to recover. The 2016 peak took four and a half.

There is no rule that a market returns within a career. Japan has just demonstrated that it may not return within a working life. What there is, in Nepal, is a fact: it has come back every time so far, and every time it took between four and seven years, and every time the people most invested at the peak did worst.

What actually separated the survivors

I have paid attention to who came out of the last two busts intact and who did not, and I expected the answer to be about stock selection.

It is not. It is about whether they were still adding.

A person who kept putting money in through 2079 and 2080 — small amounts, monthly, into things that were demonstrably cheap — has a book today whose average cost is far below the index, and he never had to call a bottom to get it. He had income and a habit.

A person who was fully invested in Bhadra 2078 with no further savings arriving had nothing to add. His only available actions were to hold or to sell. Holding was correct. Holding produced nothing. And after two years of producing nothing he sold some of it to pay for a wedding.

The difference between those two men is not intelligence and it is not analysis. It is that one had unspent capacity when prices were low and the other had used all of his when prices were high.

Which is, turned around, the entire practical content of the previous chapter. Reading the temperature is not about avoiding the fall. It is about arriving at the bottom with something in your hand.

On being told it is a good time to buy

At the bottom of a bust — the real bottom, Ashoj 2079, index at 1,815 — nobody says it is a good time to buy.

That is not irony and not coincidence. The bottom is defined by the absence of anyone willing to say it. If people were saying it and acting on it, there would be buying, and it would not be the bottom.

So the instruction, easy to write and nearly impossible to execute: the moment when buying feels most obviously stupid, when your family believes you have lost your judgment, when the asset class itself seems discredited — that moment is not a warning. It is the fee. It is what you are paying for the price.

And the reason almost nobody pays it is not courage. It is that at the bottom, the people most convinced are also the most depleted, because they have been buying all the way down.

Which brings me back to the only thing I actually recommend from any of this, and it is unglamorous enough that I will simply state it and stop.

Keep something back. Not because you can time anything. Because the bottom of a Nepali bust arrives four years after the top, and you will need money on that day, and you will not be able to earn it then.

The man at the tea shop in Putalisadak — the one who said five thousand — I saw him again in

• Different shop, same neighbourhood.

He was talking about land in Chitwan. He said the share market is not for ordinary people, that it is manipulated, and that anyone who tells you otherwise is selling something.

I did not disagree with him. He has more evidence for his position than I have for mine, and all of it is his own.

The Price You Paid Is Not a Fact About the Company

In which two psychologists rig a wheel of fortune and demonstrate that a number you know to be meaningless will still decide what you believe; a coffee mug is worth twice as much the moment you own it; the most respectable company in America falls ninety per cent while its owners explain why they are holding; and we identify the sentence that costs Nepali investors more money than any other.

In the early 1970s Amos Tversky and Daniel Kahneman put a wheel of fortune in a room, numbered from zero to a hundred, and rigged it to stop only at 10 or at 65.

They brought in subjects, spun the wheel in full view, and then asked two questions. First: is the percentage of African nations in the United Nations higher or lower than the number that just came up? Second: what is the actual percentage?

The subjects who saw the wheel stop at 10 gave a median estimate of 25 per cent. The subjects who saw it stop at 65 gave a median estimate of 45 per cent.

They had watched the wheel. They knew it was a wheel. Nobody suggested it had any connection to African membership of the United Nations, and no subject would have claimed it did. And the number moved their answers by twenty percentage points.

Tversky and Kahneman called it anchoring, and in the half-century since, it has proved to be among the most robust findings in the whole of psychology. It works on experts. It works on people who have been warned about it. It works when the anchor is obviously absurd, when the subject is explicitly told to ignore it, and when the subject is paid for accuracy. Judges have been shown to hand down different sentences after rolling dice — though I should say that this particular experiment has had a harder time replicating than the anchoring literature as a whole, which has held up very well.

You cannot defend against it by knowing about it. I want to say that plainly at the start, because everything else in this chapter follows from it, and because the usual response to a chapter like this one is yes, but I would not do that.

You would. So would I. The only question is what you build around yourself so that it costs less.

The most powerful anchor you will ever meet

Now consider the number you carry around for every share you own.

The price you paid.

It has all the properties of a perfect anchor and several worse ones. It is precise. It is emotionally significant, because you chose it. It is yours in a way the wheel of fortune never was. And it is entirely, absolutely, without exception, a fact about your own history rather than about the company.

The business does not know what you paid. The other shareholders do not know. The future cash flows are indifferent to it. If a hydropower company is worth 340 rupees a share, it is worth 340 to the man who paid 200 and to the man who paid 600 and to the man who has never heard of it. There is one value and there are many purchase prices, and only one of those numbers is about the company.

And yet. Ask any investor about a holding and he will tell you where he is relative to his cost, before he tells you anything about the business. It is the first fact in his mind and frequently the only one. He does not think this company is worth 340 and trades at 290. He thinks I am down twenty per cent.

One of those statements can lead to a decision. The other can only lead to a mood.

The mug

The purchase price does something worse than anchoring you. It changes what the thing is.

In 1990 Kahneman, Jack Knetsch and Richard Thaler ran an experiment of almost insulting simplicity. They gave coffee mugs to half the participants in a room and nothing to the other half, then opened a market: mug owners could sell, non-owners could buy, at any mutually agreeable price.

Standard economics says the mugs should end up distributed according to who likes mugs, with roughly half changing hands. What happened is that owners demanded about twice what non-owners were willing to pay, and almost no trades occurred.

Minutes earlier, none of these people had owned a mug. Ownership itself — with no time to form attachment, no story, no sunk cost worth the name — had approximately doubled the perceived value.

This is the endowment effect, and its implication for a portfolio is severe. The shares you already hold are not being evaluated on the same scale as the shares you do not. You are systematically demanding more to part with what you have than you would pay to acquire it — which means that the composition of your portfolio is substantially determined not by your view of the companies but by the accident of which ones you happened to buy first.

Every portfolio has an invisible line down the middle. On one side, things being judged. On the other, things being defended.

Two kinds of regret, and only one of them is visible

Now the asymmetry that makes all of this so expensive, and which I think is the least appreciated idea in retail investing.

There are two ways to regret an investment decision. You can regret acting — you sold, and it went up. Or you can regret not acting — you held, and it went down further.

These two regrets are not equally available to your mind, and the reason is mechanical rather than psychological.

If you sell a share at 400 and it goes to 900, you watch that happen. The stock is still on your screen. Every quotation is a fresh report on your error, delivered daily, in a form you cannot avoid. The counterfactual is vivid, precise, and continuously updated.

If you hold a share from 400 to 150, you also lose money — considerably more money — but the alternative you rejected is invisible. There is no screen showing what the 250 rupees would have become in a different company, or in a fixed deposit, or in your daughter’s education. The counterfactual has no ticker.

Selling is punished by a mechanism. Holding is punished by arithmetic. And the mind responds to mechanisms.

The consequence is that every investor is biased toward inaction, not because inaction is wise — sometimes it is, and Chapter Fifteen is about when — but because inaction is the option whose regret cannot be rendered on a chart.

This is also, I think, the real reason the sentence exists.

The sentence

”I will sell when it comes back to my price.”

I have heard this from more Nepali investors than I can count, from sophisticated ones and naive ones, and I have said it myself and caught myself saying it and said it again afterwards.

Take it apart.

It sets a target that is derived from your history rather than from the company’s prospects. It has no time limit, so it cannot expire and cannot be evaluated. It requires no analysis, so it cannot be updated by new information. It provides psychological relief, because the loss remains provisional until the sale, and a provisional loss is not a loss. And it is completely unfalsifiable, which makes it a cousin of everything in Chapter Seven, only self-administered.

Above all: it converts your purchase price into a plan, and a purchase price is the one number in the entire situation that carries no information about the future.

Now watch what it does over a full cycle, mechanically, without your ever making a conscious choice. The positions that fell a little return to your cost and you sell them, relieved and slightly pleased with yourself. The positions that fell a great deal never return, so you hold them indefinitely. Run this for ten years and your portfolio has been distilled, automatically, into precisely the things you were most wrong about — and every individual decision along the way felt disciplined.

The most respectable company in America

The international illustration is General Electric, and I choose it deliberately over some obvious fraud, because nothing about it was fraudulent.

GE was, for most of the twentieth century, the company you bought if you did not want to think about it. It was the only original member of the Dow Jones Industrial Average still in the index after a hundred years. It had a triple-A credit rating. It paid a dividend that had not been cut since the Depression. American financial advisers put widows into it. Under Jack Welch it was the most admired corporation on earth and its chief executive was on the cover of everything.

The shares peaked around sixty dollars in the summer of 2000.

By 2018 they were around six or seven. The dividend was cut in 2009, cut again in 2017, and cut to a penny in 2018. In June of that year GE was removed from the Dow after a hundred and eleven years.

Ninety per cent, over eighteen years, in the bluest of blue chips.

And the whole way down, millions of retail holders explained to each other why they were holding — because it was GE; because it had always come back; because selling at these levels would be locking in a loss; because the dividend was still coming; because the new chief executive was going to fix it.

Every one of those is a statement about the past or about the holder. Not one is a statement about the future cash flows of an industrial conglomerate whose finance arm had quietly become the largest and least understood part of the business.

The people who lost the most money in GE were not speculators. They were the most patient, most loyal, most long-term-oriented investors in the country, running exactly the strategy every book recommends, on a company that had earned that loyalty for eighty years and then stopped.

I raise this because the advice in Chapter Three — buy good companies and never sell — has a failure mode, and this is it. “Never sell” is a policy about you. Whether the company is still good is a question about the company, and it must be re-asked, and the purchase price must play no part whatsoever in the re-asking.

The Nepali version, with the arithmetic

Return to the table from Chapter Three, because it is populated by people who held.

A man bought Laxmi Bank at 403.7 in Shrawan 2078. Perfectly respectable decision: a long-established commercial bank, in a country where commercial banks had made the uncles rich.

By 2080 it traded at 173, and merged into Laxmi Sunrise one for one.

At every point on that journey he told himself he would sell when it came back to

• It never came back. The ticker itself stopped existing.

Now the arithmetic that the sentence conceals. To return from 173 to 404 requires a gain of 134 per cent. He was not waiting for a recovery; he was waiting for the share to more than double, in an institution that was being absorbed precisely because it could not compete. Had he asked himself in 2079 — would I buy this bank today at 240? — the answer would almost certainly have been no. But he was never asked that question, because he had asked himself a different one, and the different one had 404 in it.

Meanwhile Nepal SBI, Standard Chartered and Everest — the dull, unfashionable names that had barely moved in the boom at all — went on to be among the few things that rose over the following five years.

Mental accounting: why the portfolio is invisible

There is one more mechanism and it is Thaler’s again.

People do not hold one portfolio. They hold a collection of separate mental accounts, one per stock, each with its own history, its own purchase price and its own emotional balance sheet. The hydropower company that is up sixty per cent lives in one account; the finance company that is down forty lives in another; and the two are almost never evaluated together.

This is why an investor will tell you, quite sincerely, that he has “made money in hydropower and lost it in finance,” as though he had run two businesses. His net position — the only number that has any claim on reality — is often something he has never computed.

And it interacts with everything above. Because each stock has its own account, the decision to sell is framed as closing an account at a loss rather than as moving capital from a worse use to a better one. Framed the first way it is an admission. Framed the second way it is Tuesday.

There is a further Nepali wrinkle, which I mention here and take up properly in Part Two. Capital gains tax steps down at three hundred and sixty-five days of holding, from 7.5 per cent to 5. This is a genuine reason to delay a sale you already intend to make, and it is worth real money. It is not a reason to hold something you would not buy. But it is an extremely convenient thing for a man to remember when he is looking for a reason not to face a decision, and I have watched an honest tax consideration do a great deal of dishonest work.

The one question

I have tried a number of remedies and only one has survived contact with my own psychology. It is a replacement question, and it works because it strips out the anchor by construction.

If I held the cash instead of this share today, would I buy this share at this price?

That is it. Note what it does. It contains no purchase price, because a man holding cash has no purchase price. It contains no history. It cannot be answered by reference to what you have suffered. It converts every holding decision into a buying decision, which is the same decision, and which is a thing you already know how to do.

If the answer is yes, you hold, and you hold with a reason rather than with a hope.

If the answer is no, then holding is the same act as buying, and you have just performed it by default, and the fact that it does not feel like a purchase is precisely the illusion this chapter is about.

Two refinements make it stronger. Ask it on a schedule rather than when you are upset, because when you are upset you will find a reason. And ask it against a named alternative — not “would I buy this” in the abstract, but “would I buy this rather than the specific other company I am currently interested in,” which is the real choice and which forces the portfolio to become visible.

What to do about the fact that none of this works

I said at the start that knowing about anchoring does not defend you against it, and I meant it, so let me be honest about what the replacement question actually achieves.

It does not remove the anchor. It routes around it. You will still know what you paid. You will still feel the twenty per cent. What the question does is create a second number — a value estimate, arrived at separately — and put it in the same room, so that the decision has something to argue with.

That is the general shape of every practical remedy in this book, and it is worth naming now because it recurs. You do not fix a bias by resolving to be unbiased. You fix it by changing the procedure so the bias has less to grip. Write the reasons down before the outcome. Fix the inputs before you look at the price. Ask the replacement question on a schedule. Every one of these is a piece of machinery built by a man who does not trust himself, and the distrust is the qualification.

Tversky and Kahneman ran a version of their wheel experiment on people who were offered payment for accuracy.

It made no difference. The subjects wanted the money, tried harder, and anchored just the same.

There is no amount you can be paid to stop seeing a number that is in front of you. The only thing available is to arrange matters so that a better number is in front of you as well.

The Country Will Export Electricity

In which the fastest-growing economy in modern history rewards its shareholders with almost nothing; a hundred years of aviation produces a cumulative loss; four letters coined by a Goldman Sachs economist attract a great deal of money to two countries that did not deserve it; and we discover why a true story about a country is not an investment thesis.

Every Nepali investor has heard the sentence, and most of them believe it, and it is true.

Nepal has enormous hydroelectric potential, domestic demand is growing, India will buy the surplus, and the country’s rivers are its one great endowment.

I want to be clear before anything else that I am not going to argue with this. The water is real. The demand is real. The export agreements are real. The engineering is real. On the facts, the story is correct, and anyone who tells you otherwise is selling pessimism, which is a commodity with its own market.

The question this chapter asks is narrower and considerably more useful:

Does a true story about a country tell you anything about what its shares will return?

The answer, established across a century of data in twenty-one countries, is no. Not “weakly.” Not “less than you think.” The relationship is, if anything, slightly negative, and the reason it is negative is a mechanism you can see operating in the Nepali hydropower sector every time you open a company notice.

The most expensive true story of our lifetime

Between roughly 1990 and 2020, China did something no country had ever done. Real GDP compounded at close to ten per cent a year for three decades. Six hundred million people left poverty. It went from a peripheral economy to the second largest on earth, and it built more physical infrastructure in twenty years than Europe had built in a century.

Every part of that story was known in advance. It was not a secret; it was on the cover of every magazine, and any investor in 1993 could have told you China would grow faster than the West for a generation, and he would have been exactly right.

Chinese equities were, over most of that period, a poor investment. An investor who bought the Chinese market when it opened to foreigners and held for the next two and a half decades earned a return that lagged not only the American market but most of the developed world, and did so while enduring drawdowns of seventy per cent.

Correct about the country. Correct about the growth. Correct about the direction, the magnitude and the duration. And poorly paid for all of it.

The measurement

This is not an anecdote and it is not an argument about China. It has been measured properly, more than once, and the result is one of the most under-known facts in finance.

Elroy Dimson, Paul Marsh and Mike Staunton at London Business School maintain the longest and cleanest cross-country record of equity returns in existence, covering more than twenty markets back to 1900. When they correlate a country’s real per-capita economic growth with the real return earned by its stock market over the same period, the relationship comes out negative — around minus 0.3.

Jay Ritter of the University of Florida ran the same test independently over 1900 to 2002 for sixteen countries and obtained about minus 0.37.

One dataset, two ways of measuring the growth side, and the sign is not merely absent. It points the wrong way. I should be precise about the independence, because I overstated it at first: Ritter used Dimson, Marsh and Staunton’s own return series and substituted a different set of per-capita income figures. That is a robustness check on one body of evidence rather than two bodies of evidence agreeing — which makes the finding less overwhelming than I first wrote, and still the opposite of what everybody assumes.

I remember the first time I read this properly and having to sit with it for some days, because it contradicts the single most intuitive belief in investing, which is that you should put your money where things are growing.

Why: the two per cent that goes somewhere else

The mechanism is not mysterious once stated, and it is the reason this chapter belongs in a book about Nepal.

Growth requires capital, and capital comes from issuing new shares.

A country that is growing quickly is a country building things — factories, roads, transmission lines, power plants. Building requires money that the existing earnings cannot supply. So companies raise it: they issue new equity, and every new share issued divides the same pool of future profits among more claimants.

The aggregate economy grows. The profits grow. The profits per existing share grow by much less, because the denominator has been growing too.

William Bernstein and Robert Arnott measured the size of this leak and called it dilution. Across long periods and many markets, net new share issuance runs on the order of two percentage points a year. Which means that in a country whose corporate earnings grow at eight per cent, the earnings attributable to a share you already own grow at about six.

Two points a year, compounded across a career, is roughly half your money.

And there is a second leak, which is that the benefits of growth do not accrue principally to shareholders at all. They accrue to consumers, who get cheaper electricity; to workers, who get wages; and to new investors, who supply the capital and take their cut. Rapid growth in a competitive industry drives prices down and reinvestment up, and both of those are transfers away from the incumbent owner.

Buffett put this better than any economist. Reflecting on the airline industry — a century of extraordinary, genuine, world-changing growth in air travel, and a cumulative return to shareholders that is approximately zero or worse — he remarked that a farsighted capitalist at Kitty Hawk would have done his successors a favour by shooting Orville Wright down.

Air travel grew beyond anyone’s imagining. The people who financed it were not paid for it.

The four letters

If you want to see the narrative operating on money in real time, consider the BRICs.

In 2001 Jim O’Neill, then at Goldman Sachs, published a paper observing that Brazil, Russia, India and China would collectively become a large share of world output by mid-century. The observation was sound and largely accurate.

What followed was not investment analysis. It was branding. The acronym was irresistible — it sounded solid, it fitted in a headline, and it converted a demographic projection into a product. Funds were launched. Enormous sums moved.

The acronym was coined in 2001 and the markets then had a spectacular decade: Brazilian and Russian equities were among the best performers on earth through 2007. It was the decade after that was dismal, and the money mostly arrived for it. Goldman’s own BRIC fund was folded in 2015 after losing the great majority of its assets, and was quietly merged into a broader emerging markets fund.

Note that the original thesis was not falsified. Those economies did grow. India and China did become enormous. The forecast was substantially correct and the investment was substantially bad, which is the same sentence as China’s, and the same sentence as aviation’s, and the same sentence as the one I am about to write about hydropower.

Robert Shiller — who has a Nobel Prize and, more relevantly, a book called Narrative Economics — argues that stories spread through populations the way epidemics do, with a contagion rate and a recovery rate, and that they drive real economic events rather than merely describing them. The BRIC story was a pathogen with a four-letter name. So is ours.

Now Nepal

There are one hundred and eight listed hydropower companies on the Nepal Stock Exchange.

One hundred and eight. In a country of thirty million people with a stock exchange of a few hundred names, more than a third of the listed universe is one industry, and that industry sells a single undifferentiated commodity to a single buyer at a contracted price.

Every one of those listings was justified by the story, and the story is true.

Here is what happened to the money.

I took every company for which I hold a complete price record on both sides of the 2021 peak, grouped them by sector, and measured the boom and the five years that followed.

Read the last two columns together, because separately they mislead.

Hydropower had a magnificent boom — 3.88× median, third best on the exchange — and then gave back nineteen per cent. Look at the individual names at the top of that boom and the pattern is starker: Union Hydropower up 11.43× and then down to 0.68×. Upper Chapagaun 9.10× and then 0.67×. Dolti Power 8.74× then 0.68×. Modi Khola 7.78× then 0.36×.

Now look at the bottom row. Hotels — three companies, the dullest and least narrated corner of the exchange, a sector nobody was writing about in 2078 — had the worst boom of any sector and the only positive sequel on the board, at 1.93×. Over the full six years they beat hydropower.

Nobody made a WhatsApp group about hotels.

The specific mechanism, in Nepali

Now let me show you Bernstein’s two per cent operating in front of you, because in Nepali hydropower it is not two per cent. It is enormous, and it is disclosed, and almost nobody reads it as dilution.

A hydropower company is a construction project that becomes an annuity. Before the plant operates there is no revenue at all — none, not a rupee — and the construction must be funded. It is funded by debt and by right shares: new equity offered to existing shareholders, again and again, through the construction period.

If you do not subscribe to a right issue, your ownership is diluted directly. If you do subscribe, you are supplying fresh capital, which means the return on your total investment is being calculated on a much larger base than the shares you originally bought. Either way, the growth of the company and the growth of your holding are different quantities, and the gap between them is exactly Ritter’s negative correlation, arriving one right issue at a time.

There is a further trap peculiar to the pre-operational stage, and I referred to it in Chapter One. A hydropower company that has raised money and not yet built anything is sitting on a large pile of cash. That cash earns interest in a fixed deposit. The interest appears in the profit and loss account.

The company reports a profit. The profit is the interest on your own money.

It has generated not one unit of electricity. Its earnings per share are real in the accounting sense and entirely meaningless in the economic sense, and I have watched those earnings per share get put into a price-earnings ratio and compared across companies on television.

I will treat all of this properly in Part Four, where hydropower gets a full chapter, including the thing that actually decides the value of one of these businesses — the number of years remaining on its generation licence, which is the denominator of everything and appears on the first page of nothing.

The general form

Strip the specifics away and the error has a shape you can carry into any sector, any country, any decade.

A story about a country describes the size of a future pie. An investment thesis must describe the size of your slice. These are different questions, they have different answers, and the second one is decided by things the story never mentions: how much new capital must be raised to capture the growth, how many competitors will arrive to compete it away, what price the regulator will permit, and what you paid.

Consider the hierarchy of what you can be right about, in ascending order of difficulty and descending order of availability:

You can be right that the sector will grow. This is usually easy, frequently public, and worth nothing, because everybody else knows it too and it is in the price.

You can be right that a particular company will grow faster than its sector. This is genuinely hard, requires real work, and is worth something.

You can be right that the growth will accrue to existing shareholders rather than being competed away or funded by dilution. This is the rarest and most valuable judgment in investing, and it is the one nobody makes, because it is not a story about the future. It is a question about capital structure, competition and regulation, and it is deeply boring, and boring things are not forwarded.

Notice that the hydropower story only ever operates at the first level. Nepal will export electricity — a claim about the sector, freely available, priced.

What the story is actually for

I want to end with something more sympathetic, because I have been dismantling a belief that a great many sincere people hold, and because the dismantling is not the whole truth.

A narrative is not merely an error. It performs a real function: it is how capital gets mobilised in a country that needs capital. Nepal genuinely required a mechanism to move household savings into the construction of generation capacity, and no rational discounted cash flow model was ever going to persuade a schoolteacher in Butwal to fund a run-of-river project in Dolakha. The story did that. The plants exist because people believed something slightly too enthusiastically, and the country has electricity that it did not have in 2072, and the load-shedding that defined my childhood is gone.

The story was economically productive and individually expensive. Those are compatible, and the person who understands they are compatible is the person who can own the sector at the right price rather than at the moment of maximum belief.

So the discipline is not cynicism. It is sequence. Hear the story. Grant that it is true. And then ask the three questions the story cannot answer: what must be paid to participate, how much new capital will be demanded of me before it pays, and what happens to my slice while the pie is being baked.

If the answers are good, the story becomes a reason to act rather than a substitute for one.

At a wedding in Chitwan last year a man told me, with complete confidence and a fair grasp of the detail, that Nepal’s installed capacity would triple within a decade and that India had committed to purchase ten thousand megawatts.

I said that sounded right and asked him which company he owned.

He named it, and then he told me what he had paid, and then — before I had asked anything at all — he explained that he had also subscribed to the last two right issues, which he described as a good opportunity to increase his position at a discount.

He had put in four times his original investment. He was computing his gain against the first tranche.

Ruin Is Not a Probability

In which a twenty-eight-year-old destroys a bank that had financed the Napoleonic Wars; a man loses twenty billion dollars in two days; a game with a positive expected value takes ninety-nine per cent of your money; and a retired schoolteacher in Butwal owns four microfinance companies.

Barings Bank was founded in 1762. It financed the Louisiana Purchase. It was banker to the Queen. The Duc de Richelieu once remarked that there were six great powers in Europe: England, France, Prussia, Austria, Russia, and Barings.

In February 1995 it was sold to a Dutch bank for one pound.

The proximate cause was a twenty-eight-year-old named Nick Leeson, running the Singapore futures operation, and the popular account of what he did is wrong in a way that matters for this chapter. He was not, at the beginning, a rogue speculator seeking enormous gains. He was a man trying to erase a mistake.

It began with a small error — a junior colleague’s trade booked the wrong way, costing perhaps twenty thousand pounds. Rather than report it, Leeson placed it in an error account numbered 88888, intending to trade his way back to flat. This is not an exotic impulse. It is the most ordinary impulse there is, and every reader of this book has felt it in some domain.

The position moved against him. So he added to it, because a larger position would recover the loss faster. That moved against him too. By the end of 1994 the hidden loss was around two hundred million pounds and he was, by any reasonable description, running an enormous unhedged bet on the Nikkei with a bank’s balance sheet, in order to get back to a number he had passed two years earlier.

On the seventeenth of January 1995 an earthquake struck Kobe. The Nikkei fell. Leeson responded by buying more, on the reasoning that the market would rebound.

The final loss was £827 million, roughly twice the bank’s available capital.

The impulse is not greed

I want to insist on this, because the moralising version of the Leeson story is useless and the accurate version is directly applicable to a man in Kathmandu with four lakh rupees.

Leeson was not trying to get rich. If he had wanted to get rich there were easier and safer routes available to a man in his position. He was trying to return to zero.

Every increase in the position was a rational-feeling response to the previous one, in a chain where each link followed from the last, and the whole chain led off a cliff. And at no point did the situation present itself to him as should I risk the bank. It presented itself as should I close this at a loss now, or hold a little longer for the recovery I am confident is coming — which is the identical question a Nepali investor asks himself about a hydropower company at 6 p.m. on a Sunday.

The difference between Leeson and that investor is not psychological. It is structural, and the structure is called leverage, and this chapter is about what leverage does to arithmetic.

A game you should refuse

Let me put a proposition to you.

I offer a coin flip. Heads, your wealth increases by fifty per cent. Tails, it falls by forty per cent. Fair coin. You may play as many rounds as you like.

Compute the expected value of a single round. Half the time you multiply by 1.5, half the time by 0.6:

(0.5 × 1.5) + (0.5 × 0.6) = 1.05

A five per cent expected gain per round. Positive. Substantially positive — better than any fixed deposit in Nepal, offered repeatedly, with no fee.

Every finance textbook, and every reasonable person, says take it.

Do not take it.

Here is what actually happens to a person who plays. I simulated two hundred thousand players at each horizon.

Read the hundred-round row twice.

The average player ends with eighty-four times his money. The typical player ends with half of one per cent of it. More than half of all players lose over ninety-nine per cent.

Both numbers are correct. They are not in conflict. And the gap between them is the single most important thing in this book that is not about Nepal.

Why the average is a lie

The arithmetic is simple once you see where to look.

The expected value calculation adds up what happens across many parallel players. Some of those players get an extraordinary run of heads and end up with millions of times their stake, and those few carry the entire average. The mean of 83.75 is real — it is genuinely what you would get if you could pool the outcomes of everybody who plays and share the proceeds.

But you are not everybody. You get one path, sequentially, through time. And what matters to a single path is not the average of the multiplications but their compound effect, which means you must multiply, not add.

Multiply one gain and one loss: 1.5 × 0.6 = 0.90. You are down ten per cent after a head and a tail, in either order. The per-round geometric growth rate is the square root of 0.90, which is 0.9487 — a loss of 5.1 per cent per round, every round, almost surely, forever.

The mathematician Ole Peters has spent years arguing that this distinction — between what happens across an ensemble and what happens along one trajectory through time — is not a technicality but the source of a great many wrong conclusions in economics. For most of the things a human being cares about, the ensemble average is a fiction. You do not get to be all the players. You get to be one, in order, with your actual money.

The general form of the leak has a name: variance drag. The rate at which your capital actually compounds is below the average return by approximately half the variance. Two portfolios with identical average returns and different volatility do not end in the same place; the volatile one ends lower, mechanically, with no misfortune required.

Volatility is not merely discomfort. It is a subtraction.

This is the arithmetic reason behind the observation in Chapter Ten that a seventy-five per cent decline requires a quadruple to undo. It is the same fact wearing a different hat.

Twenty billion dollars in two days

If Barings feels historical, consider March 2021.

Bill Hwang ran a family office called Archegos. He had built, through total return swaps arranged with a number of investment banks, positions of enormous size in a concentrated handful of stocks — leverage of perhaps five times, spread across counterparties in such a way that no single bank could see the whole picture.

The positions moved against him. The margin calls came. He could not meet them, and the banks liquidated.

Approximately twenty billion dollars of personal wealth ceased to exist inside two days. Credit Suisse lost over five billion on the unwind and never recovered its footing as an institution; it was absorbed by UBS two years later.

Hwang had been genuinely, spectacularly right for years. The stocks he owned had made him one of the wealthiest people on earth. His analysis was not the failure. His size was.

And note the same structural fact as Barings: at no point was there a decision that looked like should I risk everything. There was a series of decisions, each of which looked like this position is working, and I have collateral, and I could hold a little more.

Ruin is a boundary

Here is the conceptual move that I think matters more than any of the arithmetic, and it is why the chapter is titled as it is.

People treat ruin as a low-probability outcome, to be weighed against other outcomes in the ordinary way. A two per cent chance of losing everything against a ninety-eight per cent chance of doing well sounds, framed like that, like a good trade. Expected value is positive. Take it.

But ruin is not one outcome among many. It is an absorbing state. Once you reach it you stop playing, and every subsequent good outcome that would have come to you is deleted — not reduced, deleted. The man who is wiped out in year three does not participate in years four through forty, and the forty-year compounding you were counting on when you sized the position was computed on the assumption that he would.

So a two per cent annual chance of ruin is not a two per cent problem. Over twenty years it is a thirty-three per cent chance of never seeing the outcome you planned. Over forty years it is fifty-five per cent. The small number compounds into the certainty of your absence.

This is why the correct question about any position is never “what is the expected return.” It is:

In how many of the ways this can go am I still able to act afterwards?

That is the question from Chapter Two, arriving in its most consequential form. And it reorganises priorities completely, because it means that avoiding the small number of catastrophic outcomes is worth more than optimising the large number of ordinary ones.

Sunita ma’am

She is a retired schoolteacher in Butwal, she is a composite of three people, and she has done nothing that anybody would call reckless.

She has no debt. She has never borrowed to buy a share and would be horrified at the suggestion. She does not trade. She reads carefully, she asks sensible questions, and she made her decision on the basis of a real observation.

The observation was that microfinance companies were paying very large dividends. Twenty, twenty-five, thirty per cent on the face value of the share, year after year, in a country where a fixed deposit paid seven. She worked out that a portfolio of these would produce something that behaved like a salary, which is precisely what a retired person needs, and she put essentially the whole of her retirement into four of them.

Four. Out of fifty listed microfinance companies. All in one sector, all subject to the same regulator, the same interest-rate cap, the same borrowers, and the same weather.

Now, she is not Nick Leeson. She has no leverage. She cannot receive a margin call and nobody can liquidate her. In the narrow sense she cannot be ruined.

But look at what she has actually constructed. Her entire retirement income depends on one regulatory decision. If Nepal Rastra Bank caps the spread microfinance institutions may charge — which it has done before, and which it may do again for entirely respectable reasons of consumer protection — then all four of her holdings are impaired on the same morning. Not one of them. All four, because they were never four bets. They were one bet, purchased four times.

This is the most common form of ruin in Nepal and it does not involve borrowing at all. It is concentration disguised as diversification by the simple device of counting tickers instead of counting risks.

And there is a further sting. The dividend that attracted her is the thing that made the sector fragile. A company distributing twenty-five per cent of face value annually is not retaining capital. The distribution is the vulnerability, and it looked like the attraction, and there was no way to tell the difference from the outside without reading the capital adequacy position — which is Part Four, and which is exactly why Part Four exists.

The concentration ladder, read the other way

Return to the table in Chapter One, because it has a second reading I withheld at the time.

I showed that a thoughtless investor holding one Nepali stock in 2077 had a seven per cent chance of a sixfold return, and one holding ten had essentially none. I used it to argue that a concentrated success carries no information.

Turn it over. Concentration is the machinery of an extreme outcome, and it operates in both directions with perfect impartiality.

The same one-stock portfolio that produced three hundred and fifty spectacular stories in a rising market produces, in a falling one, the corresponding number of people who do not attend weddings. In 2077 nothing fell, so the left tail was empty and invisible. That was a property of the year, not of the strategy.

The strategy has not changed. Only the weather has.

The right-share ratchet

There is one uniquely Nepali mechanism that produces Leeson’s escalation without requiring Leeson’s psychology, and I want to name it because it is nearly invisible.

You own a hydropower company. It calls a right issue — new shares offered to existing holders, to fund construction.

Consider your position. If you subscribe, you must put in more money. If you do not subscribe, your stake is diluted and, in Nepal, the price typically adjusts in a manner that makes non-subscription feel like an immediate loss. The right is described as an opportunity to buy at a discount, and it is framed as a benefit.

So you subscribe. And the following year, the same. And the year after.

At no point did you decide to increase your position. You decided, three times, not to accept a loss. And your exposure to one construction project in one valley has quadrupled through a sequence of individually defensive decisions.

That is the 88888 account, built out of corporate actions rather than futures contracts, and operated by a man who has never heard of Nick Leeson and is doing nothing wrong at any single step.

The three rules I actually keep

I do not have a formula for position sizing and I distrust people who present one with too much confidence, for reasons I will set out properly in Part Five — where I have to demonstrate that in this market the growth-optimal bet size is not merely difficult to compute but not estimable at all, which is an unusual thing to be able to prove.

What I have are three rules, and they are crude, and crude is the point.

One. No position may be large enough that its permanent impairment changes my life. Not my year. My life. This is a lower bar than it sounds and almost nobody in Nepal clears it, because the largest position in a typical Nepali portfolio is not a share at all — it is the house, and the second is the family business, and both are already concentrated in the same economy as the shares.

Two. Count risks, not tickers. Before adding anything, ask what single event would impair it, and then ask how many of my existing holdings that same event would impair. Four microfinance companies are one holding. Nineteen commercial banks are, to a first approximation, one holding, since they share depositors, a regulator, and a credit cycle.

Three. Never let a decision not to accept a loss become a decision to increase a position. These are different acts and they feel identical. A right issue is a fresh investment decision and must be evaluated as one, from a blank page, with the replacement question from Chapter Eleven: if I held this cash today, would I buy this?

Leeson was sentenced to six and a half years in a Singapore prison and served four. He has since had a career as a speaker, explaining what he did.

In interviews he returns repeatedly to a single point, and I have come to think it is the most honest thing anybody in this book says. He does not describe a moment of recklessness. He describes the opposite: an intense, sustained, exhausting effort to avoid admitting a small mistake, which required a slightly larger position, which required a slightly larger one after that.

He has said that if he had reported the original twenty thousand pounds, nothing would have happened to him at all.

Twenty thousand pounds. Against two hundred and thirty-three years.

Waiting Is an Accounting Entry

In which a widow’s portfolio outperforms her husband’s because he died; sixty-six thousand American households are watched for six years and the busiest of them earn the least; and doing nothing turns out to be worth twenty-three per cent of your money, for reasons that have nothing whatever to do with character.

In 1984 Robert Kirby published a short piece in the Journal of Portfolio Management describing something that had happened to him a decade earlier, and it has become the most quoted anecdote in investment writing that is also, as far as anybody can establish, true.

Kirby had managed money for a woman for about ten years. Her husband, it emerged, had been quietly following along: every time Kirby recommended a purchase for the wife’s account, the husband bought the same stock for himself, in a roughly equal amount.

The difference was that he ignored every sell recommendation.

When he died and his holdings came to light, Kirby found a portfolio that was dramatically larger than the one he had been actively managing next door. It contained a number of small, forgotten positions worth very little. It also contained one holding worth over eight hundred thousand dollars — a company called Haloid, which had bought the rights to a process called xerography and renamed itself Xerox.

Kirby called the idea the coffee can portfolio, after the tin that American frontier families kept under the mattress: you put things in and you do not open it.

Now, before we extract the wrong lesson — and almost everybody does extract the wrong lesson, which is roughly “be patient, it is a virtue” — I want to look at the mechanics, because the mechanics are the transferable part and the virtue is not.

What the dead husband actually did

He did three things, none of which required insight.

He did not pay tax. Every sale in the wife’s account crystallised a gain and handed a portion of it to the state, permanently removing that portion from the compounding. His account compounded on the whole amount.

He did not pay costs. Every round trip in her account paid commission twice.

And he did not truncate his winners. This is the largest of the three and the least obvious. A portfolio’s long-run return is not the average of its holdings; it is dominated by a small number of positions that become enormous. Selling on a rule — after a double, at a target, on a valuation trigger — systematically removes exactly those positions, because a stock on its way to a fortyfold return passes through every sell trigger you have ever set, one after another, on the way.

The husband’s Xerox position was worth more than most of the rest of the portfolio combined. Any sensible rebalancing discipline would have trimmed it a dozen times.

Note that none of the three is a statement about patience as a personal quality. They are, in order, a tax fact, a fee fact, and a distributional fact. The coffee can worked for accounting reasons.

Which is why the chapter is titled as it is, and why I am going to be careful in the second half about where the argument stops.

Sixty-six thousand households

The anecdote is charming and an anecdote. Fortunately somebody did the study.

Brad Barber and Terrance Odean obtained the complete trading records of 66,465 American households at a large discount brokerage over 1991 to 1996, and published the results under a title that saved everyone the trouble of reading further: Trading Is Hazardous to Your Wealth.

The market returned 17.9 per cent a year over the period. The average household in their sample earned 16.4 per cent. And the households in the most active quintile — the ones who turned their portfolios over most aggressively — earned 11.4 per cent.

Six and a half percentage points a year, given up by the busiest participants, in a sample of ordinary people using the same information and the same market as everybody else.

The authors then decomposed it, which is the part that matters. The gap was not principally caused by bad stock selection. When they measured the performance of the stocks these investors bought against the stocks they sold, the purchased stocks underperformed the sold ones — but by a modest amount. The dominant term was cost. Commission and the bid-ask spread, applied repeatedly, took the rest.

In a companion paper the same authors observed that men traded about forty-five per cent more than women and underperformed them accordingly, a finding I mention chiefly because I have never met a Nepali investor who believed it applied to him.

The Nepali arithmetic, in rupees

Now let me do the calculation for this market, because Nepal’s fee structure has a peculiarity that makes turnover more expensive here than almost anywhere, and it is worth seeing in rupees rather than percentages.

Take a hundred thousand rupees, compounding at twelve per cent a year for twenty years. Apply the actual Nepali charge schedule: broker commission of roughly 0.36 per cent, the SEBON fee, the flat depository charge of twenty-five rupees per scrip per settlement on both sides, and capital gains tax at 7.5 per cent under a year and 5 per cent over.

Three investors, identical stock-picking ability, identical returns before charges.

Two hundred and thirteen thousand rupees, on a one-lakh starting stake, is the price of the habit. It is 23 per cent of the patient outcome, and it was paid by a man whose analysis was exactly as good.

Two mechanisms are doing that work and they are worth separating.

The tax deferral is the larger. An investor who never sells is compounding on money the government has a claim to but has not yet collected. That deferred liability works for you the entire time, and it is settled once, at the end, at the lower rate. An investor who sells annually pays the tax each year and compounds on what remains. Over twenty years the difference between compounding at twelve per cent and at something nearer eleven is not a rounding error; it is the last several years of the sequence.

The flat depository fee is the peculiar one, and it deserves a moment.

Twenty-five rupees per scrip per settlement is not a rate. It is an absolute amount, and absolute amounts behave very differently from percentages. At the bottom commission tier the variable charges come to roughly 0.375 per cent on one side, so the flat fee equals the variable charge at a trade value of

25 ÷ 0.00375 = NPR 6,667

Below that threshold, more than half of what you are paying is a fixed toll that does not care how small your trade is. A trade of two thousand rupees in Nepal is made mostly in order to pay for itself.

This single arithmetic fact has consequences that run through the whole of Part Five — it sets a minimum economic trade size, which sets a minimum rebalancing band, which determines how many positions a book of a given size can sensibly hold. It is derived from a published fee schedule rather than estimated from data, which makes it one of the few things in this book that cannot be wrong.

And I should report, because it belongs here rather than anywhere else, the one finding that survived my own research programme into this market intact. When I measured the drag from charges across holding horizons, it fell monotonically from about seven per cent a year at very short holding periods to under one per cent at long ones; and lengthening the interval between portfolio rebalances was worth a median 1.8 percentage points of annual compounding, positive in every sub-period I tested it on. It survives, I think, because it is not a forecast about anything. It is an accounting identity over a fee schedule.

Where the argument stops

I have now spent most of a chapter arguing for inaction, and if I leave it there I will have written something dangerous, because the coffee can has a failure mode and Chapter Eleven was about it.

General Electric was a coffee can holding. So was Kodak. So was Nokia, so was Sears, so was every Nepali bank in that table in Chapter Three that peaked in Bhadra 2078 and was absorbed at a discount two years later. Doing nothing is not a strategy. It is the absence of one, and it happens to coincide with a good strategy about eighty per cent of the time.

The distinction that resolves this is between two objects that the word “patience” conflates:

Patience about a price is almost always correct. The share has fallen and nothing about the business has changed; the market is offering you a worse price for the same thing; there is no information here and you should do nothing, and the fee schedule agrees with you.

Patience about a business is a decision, and it must be re-made. The share has fallen because the loan book is deteriorating, or the licence is running down, or the only customer has stopped paying, or the regulator has changed the rules. That is not weather. That is the thesis failing, and holding through it is not patience. It is the 88888 account from the last chapter, operated slowly.

Which means the operative rule is not hold. It is:

Hold, unless a specific pre-written condition has occurred.

That is why Chapter Four insisted on writing down the falsifier at the moment of purchase. The falsifier is what converts inaction from a default into a decision. A man who holds because he wrote down, two years ago, that he would sell if the non-performing loan ratio exceeded four per cent, and it has not, is being patient. A man who holds because selling would hurt is being something else, and the two are indistinguishable from the outside and from the inside.

The husband in Kirby’s story got a wonderful result. It is worth noticing that he also had no mechanism whatever for detecting a Xerox that had become a Kodak, and that in a different decade the same behaviour produces a very different anecdote, which nobody would have published.

The dead-investor study that probably does not exist

There is a claim that circulates constantly, and you have encountered it: that Fidelity conducted an internal review of its accounts and found the best-performing ones belonged to investors who were dead, followed by those who had forgotten they had an account.

I have looked for this study on several occasions and cannot find it, and neither, so far as I can establish, has anybody else. Fidelity has never produced it. It appears to be an artefact — a story that is too good to check, propagating because it compresses a true idea into a memorable form.

I raise it for two reasons. First, because the true idea underneath it is genuinely supported by Barber and Odean, so the story is directionally honest even if it is not real. And second, because it is a small demonstration of the entire first section of this book: an unverifiable anecdote, selected for its punchline, repeated by intelligent people, in support of a conclusion that happens to be correct.

Being right for the wrong reason is not confined to markets. It is how most financial folklore is transmitted, including the parts that are true.

What this actually implies for you

Three things, and they are the practical residue of the chapter.

Set the bar for action high, and make it explicit. Not “I will trade less,” which is a resolution and will fail. A written rule: no position is sold unless a pre-recorded falsifier has triggered, or the money is needed, or something demonstrably better has been found and the improvement exceeds the round-trip cost — which, given the flat fee and the tax step, is a larger hurdle than almost anybody assumes.

Cross the 365-day line deliberately. The tax step from 7.5 to 5 per cent is free money for a sale you already intend to make. It is not a reason to hold something you would not buy — Chapter Eleven warned about exactly this rationalisation — but where the decision is genuinely marginal, the calendar has a vote and it is worth about two and a half per cent of the gain.

Do not confuse a small portfolio with a small fee. The flat depository charge means that an investor with two lakh rupees spread over fifteen names is paying a materially higher percentage cost than an investor with twenty lakh doing the same thing. If your book is small, hold fewer things, and hold them longer, and understand that this is not a compromise forced by poverty. It is what the arithmetic recommends.

Kirby, writing in 1984, was honest about something that gets left out of the retellings.

He noted that the coffee can portfolio’s success depended on the initial selections being reasonable, and that the approach would be a disaster in the hands of someone who filled the tin with rubbish. The husband, remember, was buying what a professional had recommended. He simply declined to accept the professional’s second thought.

The tin does not confer quality. It only stops you interfering with it.

The Desk That Mostly Says Nothing

In which the last man to hit .400 divides the strike zone into seventy-seven squares; the most successful investor of the century closes his fund because he cannot find anything to buy; another gives four billion dollars back to people who wanted him to keep it; and we consider the possibility that the most valuable thing an analyst produces is a refusal.

Ted Williams hit .406 in 1941 and no one has done it since, which we established in Chapter Four was a fact about the narrowing of variance rather than about the decline of hitting.

But Williams himself had a theory, and he set it out in The Science of Hitting in 1970, and it contains a diagram that has been reproduced in more investment presentations than any other image in sport.

He drew the strike zone as a grid of seventy-seven squares, each the width of a baseball, and wrote in each square the batting average he believed he could achieve on a pitch arriving there. In his best zone — high and inside, belt height — he reckoned he could hit .400. In the low outside corner, .230.

His discipline, which teammates found maddening, was to wait for pitches in the good squares and let the others go by, even at the cost of taking strikes and even at the cost of walking rather than swinging.

Buffett has used this diagram for decades, and always with the same addition, which is the entire point:

In investing there are no called strikes.

Williams had to swing eventually. Three pitches in the poor squares and he was out, which means his discipline had a hard limit imposed by the rules. You have no such limit. You may stand at the plate for a year, or five, letting every pitch go past, and nothing happens to you at all. Nobody rings a bell. Your capital sits in a deposit account earning something, and you wait.

This is the single largest structural advantage available to a private investor, and it is the one that almost nobody uses, for reasons that are entirely social and that I will get to.

The punch card

Buffett has offered a thought experiment for this, more than once, and it is the most useful thing he has ever said about portfolio construction, which he did not intend it to be about.

Imagine, he says, that you were issued a card with twenty punches on it, representing every investment you would be permitted to make in your entire life. Each purchase uses a punch. When they are gone, you are done.

You would think very hard about each one. You would decline a great many things that were probably fine. And your results, he argues, would be dramatically better than they otherwise would.

I want to draw out what the punch card is actually doing, because “be selective” is a platitude and this is not a platitude.

It converts the cost of a bad decision from monetary to structural. In real life, a mediocre purchase costs you the difference between its return and a better one — an amount you cannot observe, arriving over years, easily rationalised. On the punch card it costs you a slot, visibly, immediately, and forever. The card makes the opportunity cost of an ordinary idea concrete, and opportunity cost is the thing human beings are worst at feeling.

And it does something else, arithmetically. Twenty positions over forty years is one purchase every two years. Sit with that. Not one a month, which is roughly what an engaged Nepali investor manages. One every two years. If that number sounds absurd, it is worth asking why — because the honest answer is not that better opportunities are available, but that our tolerance for inactivity is low.

The two men who stopped

Two episodes make the case better than any argument.

In 1969 Warren Buffett closed the Buffett Partnership. He was thirty-eight, had compounded at roughly thirty per cent a year for thirteen years, and was by any measure the most successful money manager in America. His letter to partners explained that he could no longer find investments that met his standards, that the market had become a place where his approach did not work, and that he was therefore returning their capital.

Consider what that required. He was not underperforming. Nobody was demanding an explanation. He could have continued for years, charging fees, buying second-best ideas, and nobody would have thought worse of him until much later. He shut it because the opportunities were not there.

The American market peaked shortly afterwards and the Nifty Fifty era ended in the 1973–74 collapse, in which the S&P fell about forty-eight per cent.

The second episode is Seth Klarman, who runs Baupost and wrote Margin of Safety, a book so out of print that second-hand copies sell for more than most people’s monthly salary. Klarman has held cash balances of thirty, forty, sometimes fifty per cent of his fund when he could not find things worth owning — an act that is far harder for an institution than an individual, because clients pay fees on the cash and ask, reasonably, what they are paying for.

In 2010 he handed back five per cent of the fund’s capital, and at the end of 2013 he went further and returned about four billion dollars to investors — roughly a seventh of the fund — on the grounds that he had more money than ideas.

I have thought about that a great deal. Fund managers are compensated on assets. Handing back four billion dollars is handing back the fee on four billion dollars, every year, forever. It is the most expensive sentence a fund manager can utter, and he uttered it because the alternative was to deploy capital into things he did not believe in and report that he had been busy.

The third basket

Charlie Munger described his method for handling opportunities as three baskets: in, out, and too hard.

The third basket is the one worth talking about, because the first two are obvious and the third is where the discipline lives.

“Too hard” is not the same as “bad”. A company in the too-hard basket may be excellent. It may be about to double. The claim being made is narrower and more honest: I cannot evaluate this to a standard I would be willing to act on, and the fact that it might work out is not a reason to pretend otherwise.

This connects to the idea Buffett calls the circle of competence, and I want to quote the formulation carefully because the popular version drops the crucial half. He wrote, in 1996, that you do not have to be an expert on every company, only on those within your circle of competence — and then: the size of that circle is not very important; knowing its boundaries, however, is vital.

The size is not the point. A man who genuinely understands four industries and knows that he understands four industries will do better than a man who half-understands twenty and believes he understands twenty. The failure is never insufficient breadth. It is a boundary drawn in the wrong place, or not drawn at all.

How many companies can you actually know?

Now let me make this uncomfortable and specific, because in Nepal the arithmetic is brutal and nobody does it.

There are something over two hundred and fifty listed securities on the Nepal Stock Exchange, of which perhaps a hundred and fifty trade with enough regularity to be worth considering. One hundred and eight of them are hydropower companies. Fifty are microfinance institutions.

Now ask what it takes to genuinely know one.

For a bank, that means reading the annual report and at least the last eight quarterly statements; understanding where the deposits come from and how sticky they are; knowing the loan book by sector and the concentration within it; tracking non-performing loans and, more importantly, the restructured loans that are not yet non-performing; knowing the capital position against the regulatory floor and what happens to it if credit costs normalise; and having a view on management that is based on what they have done rather than what they have said.

That is perhaps twenty to forty hours for the first pass and eight to ten a year to maintain, assuming the filings are available and legible, which — and this is a fact about Nepal rather than a complaint — they frequently are not. I maintain a database of fundamentals extracted from primary filings for this market, and after several years of sustained effort it covers seventy-nine companies. Not because I lost interest, but because the documents for the rest are inconsistent, late, in scanned images, or in formats that fight extraction.

Seventy-nine, with machinery. A person doing it by hand, alongside a job, might genuinely cover eight to fifteen.

So the effective universe of any honest Nepali investor is about ten companies, and everything outside that is the too-hard basket.

I find that people react badly to this number, and the reaction is always the same: it feels like a confession of inadequacy. It is the opposite. It is the recognition that the alternative — holding thirty things you have not read about — is not diversification. It is the appearance of diversification, purchased by abandoning the only advantage a small investor has, which is that he is allowed to be selective in a way no fund manager is.

A system that is permitted to refuse

I want to describe something I built, because it made this concrete for me in a way that reading about Munger never did.

I run a valuation engine over Nepali companies. When I first built it, it always produced a number. Every company, every time. Feed it a ticker and out came a value, because that is what a valuation system does, and a system that returns nothing feels broken.

That was the defect, and it took me an embarrassingly long time to see it.

A company whose disclosures are inadequate does not have a value that is merely uncertain. It has a value that cannot be responsibly computed, and reporting a number for it — even a number with a wide band around it — is a lie with error bars. The band communicates “we are unsure of the level.” The truth was “we are unsure whether the inputs mean what they say.”

So the engine now refuses. If the growth estimate does not have enough observations behind it, it declines to produce one and states why, by name. If a filing’s figures fail a plausibility check, the record is held rather than stored. If the required inputs are absent, the answer is not a number, it is a named reason.

The number of companies on which it will render a full verdict is much smaller than the number in the market, and the difference between those two numbers is the most useful output it produces.

Because here is the thing I did not expect. The refusals turned out to be information. A company that cannot be valued from its own disclosures is telling you something about itself — about the quality of its reporting, the attentiveness of its board, and in several cases about what it would prefer you not compute. It is not a null result. It is a finding, and it is available to anybody willing to treat “I cannot determine this” as an output rather than a failure.

Why nobody does it

If silence is so advantageous, why is it so rare? The answer is not intellectual.

Silence is socially expensive. This is the theme of Chapter Nine returning in a different costume. A man who owns nothing new for two years has nothing to say at a wedding. He cannot participate in the conversation, which is entirely about what people have bought. He appears to be doing nothing, because he is doing nothing, and in a culture where activity is the visible proxy for competence he looks like a man who has lost his nerve.

Silence provides no feedback. A purchase gives you something to watch. Cash gives you a number that does not move. The mind experiences the second as an absence rather than a position, even though a decision to hold cash is exactly as much a decision as a decision to hold shares, and frequently a better-considered one.

And silence is punished by the industry. A broker earns nothing from your patience. A television programme cannot fill an hour with a man who says there is nothing to do today. An analyst who publishes four notes a year saying “still too expensive” will be replaced by one who publishes forty saying something. Every institution surrounding a retail investor is optimised to convert his attention into transactions, and every one of them is sincere about it.

Against all of that, the punch card is a piece of paper in a drawer.

What silence is not

Two clarifications, because this argument is easily bent into something lazy.

It is not the same as being permanently bearish. The man who has not bought anything since 2078 because “the market is overvalued” is not exercising discipline; he has one opinion, held indefinitely, immune to price. Silence means waiting for a pitch in your squares, and a pitch in your squares arrives at some point, and when it does you must swing hard. Klarman held fifty per cent cash and then deployed it. Buffett shut the partnership and then spent decades buying. The waiting is instrumental. If it is terminal, it is not discipline, it is paralysis wearing discipline’s clothes.

And it is not the same as inattention. The husband in Kirby’s coffee can did nothing because he did nothing. Buffett in 1969 did nothing because he had examined everything and found it wanting. From the outside the two are identical; from the inside they are opposite. Silence is expensive to produce properly: it requires you to have done the work and reached the conclusion that the work does not justify a purchase, which is the least satisfying possible reward for effort.

The practical form

Here is what I actually keep, and it is small.

A written list of what I am waiting for. Not a watchlist of tickers — a list of conditions. This bank at this price. That hydropower company once the construction is funded and the escalations are visible. It is short, it is specific, and it means that when a price finally arrives I am executing a decision I made calmly rather than making one in the presence of a moving number.

A count of purchases per year, written down. I do not set a limit, because I would break it. I simply record the number, and looking at the record is sufficient. In a year when the count climbs, something is usually wrong with me rather than with the opportunity set.

And a standing question before every purchase: is this better than adding to something I already own and understand? Most of the time the answer is no, and the question has therefore prevented far more transactions than any resolution ever did. A new name requires a new circle. Adding to an existing one requires only a price.

Williams was, by the accounts of everyone who played with him, an infuriating man to watch. He walked more than almost anyone. Fans booed him for not swinging at pitches they thought he could have hit. Managers wanted him to be more aggressive with runners in scoring position.

He hit .406 and nobody has done it since.

He also, and this is the part the investment presentations always leave out, played nineteen seasons and never won a World Series.

Which is the honest ending, and I will leave it there rather than explain it.

What You Are Permitted to Believe

In which a chicken forms a well-supported theory about a farmer; an economist in Chicago separates two things that everybody confuses; Keynes admits in four words what most of the profession will not admit in a career; and Part One is reduced to a list short enough to be useful.

Bertrand Russell, writing in 1912 about how we come to believe things, offered a chicken.

The chicken is fed every morning by the farmer. This happens on the first day, and the second, and every day for a considerable time. The chicken, being a reasonable animal and an empiricist, forms a theory: the farmer’s approach means food. The theory has excellent predictive power. It is confirmed hundreds of times. It has never once failed.

Then comes the morning the farmer wrings its neck.

Russell’s point — he was restating David Hume, who had made it more austerely in 1748 — is that no quantity of confirming observations logically establishes a general rule. The chicken’s evidence was real, its reasoning was sound, and its conclusion was wrong, and there was nothing available within its experience that could have told it so.

This is the problem of induction, and every investor in every market is the chicken about something. The only question is which thing, and you cannot find out by gathering more of the same evidence, because more of the same evidence is precisely what the chicken had.

Two things that are not the same

In 1921, Frank Knight of Chicago published a distinction that ought to be taught in the first hour of every finance course and is usually not taught at all.

Risk is a situation where you do not know the outcome but you do know the distribution. A roulette wheel. A fair die. An insurance company’s expectation of how many houses will burn in a city of a million, which it does not know for any particular house and knows with considerable precision in aggregate.

Uncertainty is a situation where you do not know the distribution either.

Almost everything that matters in investing is the second, and is universally treated as the first. When a model reports that a portfolio has a 5 per cent chance of losing more than a given amount, it is describing a distribution it has assumed. The assumption is usually estimated from history — which is to say, from the period during which the farmer was feeding the chicken.

Keynes put the same point in a 1937 paper more bluntly than any economist has managed since. Writing about the prospect of a European war, the price of copper twenty years out, and the obsolescence of an invention, he said that about such matters there is no scientific basis for any calculable probability at all.

Then four words: we simply do not know.

I would like that sentence to be available to you. Not as defeatism — the rest of this book is four hundred pages of things I think can be worked out — but as a legitimate output. In most professions “I do not know” is an admission of failure. In this one it is frequently the correct answer, and the inability to say it is what produces the evening panel in Chapter Seven.

The only claim worth making

If confirmation cannot establish a theory, what can?

Karl Popper’s answer, and it is the most useful idea in the philosophy of science for anyone who handles money, is that theories are never proved. They are only ever not yet refuted. And what separates a claim worth taking seriously from one that is not is whether it forbids anything — whether there exists some observation that, if it occurred, would kill it.

A theory that is compatible with every possible outcome tells you nothing about the world, however impressive it sounds and however often it turns out to be consistent with events.

Read Mr. Pradhan’s paragraph from Chapter Seven again with that in your hand. If the index sustains above support we may see a recovery; if selling pressure continues, further correction cannot be ruled out. It forbids nothing. It is unkillable. And unkillability, which feels like strength, is the definitive marker of emptiness.

This is why Chapter Four asked you to write down a falsifier at the moment of purchase, and why Chapter Seven said the willingness to be caught is the entire signal. It is the same principle appearing three times: a belief you cannot lose is a belief you do not hold.

The hierarchy of what can be known

Let me organise the whole of Part One into a hierarchy, because I think the confusion that costs people money is a confusion of levels.

Level one: facts about the present. What a company reported. What its capital ratio is. What price something traded at. How many shares exist. These are knowable, and they are where all serious work begins.

But — and Part Two is largely about this — they are less knowable than they look. In Chapter Five we established that Nabil Bank’s opening price for the whole of 2013 was manufactured by a data vendor’s import script. That is a level-one fact, published, authoritative, and false. The first discipline is not scepticism about opinions. It is scepticism about data.

Level two: relationships. Banks tend to earn more when credit costs normalise. Cheap markets tend to produce better decade-long returns than expensive ones. Companies funding growth through repeated equity issues tend to dilute their existing owners. These are probabilistic, they hold on average, and they have exceptions that will find you personally.

Most of the useful content of this book lives here. Not certainty — tendency, with a stated mechanism, and a rough sense of how often it fails.

Level three: the future. What the index will do. Whether a particular company will succeed. When the cycle turns.

Not knowable. Not by me, not by anyone, not with more data, not with a better model. Chapter Two handed you tomorrow’s newspaper and it ruined you. Chapter Seven counted eighty-two thousand expert forecasts and found chance. Chapter One computed that a fifteen-year record contains no information about the sixteenth year.

Nearly all financial conversation consists of level-three claims delivered in the grammar of level one. That is the disease. And the cure is not to know more. It is to notice which level a sentence is on, including your own sentences, particularly your own sentences.

What you are permitted to believe

So, at the end of Part One, here is the list. It is short because it is what survived.

You may believe that you cannot read your own results. One outcome, or fifteen, carries almost no information about your ability. Bernoulli wanted twenty-five thousand trials for an urn with two colours in it; you will accumulate five hundred correlated decisions in a career.

You may believe that your sample is filtered. One in five of everything that has traded on the Nepal Stock Exchange has disappeared. The investors you have met are the ones who could afford to keep investing. The strategies you have heard of are the ones that worked recently.

You may believe that events do not determine prices. Prices move on the gap between events and expectations, and expectations are not printed anywhere. A hundred and eighteen people were given tomorrow’s headlines and a sixth of them went bust.

You may believe that a crowd’s confidence is not a crowd’s information. Forty thousand people, correlated at a mere ten per cent, contain ten opinions.

You may believe that valuation predicts decades and not years. The starting price tells you a great deal about the next ten years and essentially nothing about the next twelve months, and the person who uses it for the second purpose will abandon it before it pays.

You may believe that concentration produces extremes in both directions, that volatility is a subtraction rather than merely a discomfort, and that a positive expected value can still ruin you, because you get one path and not the average of many.

You may believe that costs and taxes are certain while returns are not — which is why the flat twenty-five rupees, and the boundary at three hundred and sixty-five days, deserve more of your attention than the next forecast you hear.

And you may believe that a true story about a country is not an investment thesis. Across sixteen countries and a full century, the correlation between per-capita economic growth and real equity returns comes out negative rather than positive.

That is the whole of it. Eight things. None of them tells you what to buy.

What you are not permitted to believe

Shorter still.

That you know where the market is going. That your good year was skill or your bad year was misfortune. That the man on television has information. That a pattern which survived a search you did not observe is a pattern. That the price you paid is a fact about the company. That because a thing has never happened it will not. That holding is the same as deciding. That you are the exception to a chapter you have just read and agreed with.

I include the last because I have watched it happen — in others, and in myself, within minutes of finishing a book like this one.

Fragments

I close Part One the way I would close a notebook rather than an argument, because a summary would suggest that the preceding chapters resolve into a doctrine, and they do not. They resolve into a posture.

1. The market pays you for being right about a business. It does not pay you for being right about the market. These are different jobs and only one of them has an employer.

2. A man who has never said “I do not know” in public has told you the most important thing about himself, and it took him no words at all.

3. You cannot tell a good decision from a good outcome by looking at your account. You can only tell by looking at what you wrote down before, which is why the writing down is not administration. It is the instrument.

4. Every strategy that has ever been sold to you was selected from a set whose size you were not shown.

5. In a rising market, courage and stupidity produce identical statements and identical returns, and are distinguishable only in the year that follows.

6. The most dangerous position in any portfolio is the one you have stopped examining because it is working.

7. Anyone can be patient with a stock that is rising. Patience is only a decision when it is unpleasant, which means you have never tested yours in a bull market.

8. A forecast without a date is not a forecast. A thesis without a falsifier is not a thesis. A range without a method is not a range. Each of these can be produced by anybody, instantly, at no cost, and each is therefore worth what it cost.

9. The reason to compute what a business is worth is not that the number is right. It is that the act of computing it is the only defence you have against the number somebody else puts in your head.

10. Ruin is not a bad outcome. It is the end of outcomes.

11. If you cannot say what would change your mind, you are not holding a position. You are holding a wound.

12. Everybody in this market is an expert, as Ramesh dai observed. He was right, and he meant it as a joke, and it was the only falsifiable thing he said all year.

Richard Feynman, addressing graduating scientists in 1974, gave them a principle he described as the first one, which was that they must not fool themselves — adding that they were the easiest people to fool.

He was talking about physics, where the experiment eventually tells you.

Here there is no experiment. There is a price, which is somebody else’s opinion, and a statement of account, which is a record of the weather. Neither will ever inform you that you were wrong. You will have to arrange that yourself, in advance, in writing, and then be honest enough to read it.

Part Two is about the machine that produces the price.

A Market Built by Accident

In which twenty-four brokers under a buttonwood tree agree first of all to fix their prices; a Dutch company invents the tradeable share and immediately invents the short seller; and we discover that most Nepali companies are listed not because they wanted your money but because somebody told them they had to be.

On the seventeenth of May 1792, twenty-four men signed an agreement at 68 Wall Street, and every history of American finance begins with this scene, generally in a tone of reverence and generally with the men standing under a buttonwood tree. The tree was where they had been in the habit of trading, and it gave the agreement its name; the signing itself, so far as the record goes, happened indoors.

The operative text is a single sentence carrying two promises. It is worth knowing what they were.

The first: that they would charge a commission of not less than one quarter of one per cent on the specie value of anything they traded. The second: that they would give preference to each other in their negotiations.

That is the founding charter of the New York Stock Exchange. It is a cartel agreement. Twenty-four brokers agreeing not to undercut one another and to keep outsiders out, which is to say that the greatest capital market in history began as a restraint of trade, and the price-fixing clause survived, in one form or another, until 1975.

I begin here because there is a habit of thinking about exchanges as though they were designed — as though somebody sat down and engineered an institution to allocate a nation’s savings to its best uses. Almost none of them were. They are accretions. They start as a convenience for a small number of people who want to trade with each other, and they acquire their public purpose later, usually because a government notices them and attaches obligations.

Amsterdam got there earlier and more completely. The Dutch East India Company was chartered in 1602 with a feature nobody had tried at scale: subscribers could transfer their claim to somebody else. The charter still promised a general reckoning after ten years, at which a subscriber could ask for his money back; in 1612 the company declined to liquidate, and the capital became permanent. The two halves together — permanent capital for the company, liquidity for the investor — are the invention on which everything since rests. Within a few years Amsterdam had a secondary market, forward contracts, options, and a man named Isaac Le Maire running an organised bear raid on the company’s own shares. The world’s first short-selling ban followed in 1610.

Four hundred years. The instruments have not changed much and neither have the people.

What an exchange is for

Strip away the ceremony and a stock exchange performs two functions, and they are genuinely distinct.

The primary function is to raise capital. A company that needs money to build something sells a share of its future to people who have money and want a share of a future. This is the function that appears in the textbooks and in the speeches at listing ceremonies.

The secondary function is to let owners change their minds. Once the shares exist, they trade. No new capital reaches the company; the money moves between investors. This is the function that generates all the noise, all the commentary, and all of Part One.

The two are connected by a single link, and it is worth being precise about it, because it is the entire justification for the second function’s existence: nobody would supply permanent capital in the first place if they could never get out. The secondary market is what makes the primary market possible. That is Amsterdam’s insight and it is not a small one.

Now hold those two functions up against the Nepal Stock Exchange, and ask which one it actually performs.

How Nepal got an exchange

The chronology is short and the shape of it matters more than the dates.

There was no securities market in Nepal until the 1970s. The Securities Exchange Centre was established in 1976 under the Nepal Industrial Development Corporation and the Ministry of Finance, and for its first fifteen years it was not really an exchange — it was a government body that handled government bonds and a handful of share issues, in a country with very few companies of the kind that issue shares.

The reforms of the early 1990s converted it. The Securities Exchange Act was amended, the Securities Board of Nepal was created as a regulator, and the Centre was reorganised into the Nepal Stock Exchange, which began floor trading in January

• Brokers were licensed. An open-outcry floor operated in Kathmandu, and for the

next two decades that floor, and later a screen-based system reachable through those brokers, was the market.

Now the detail that a foreign reader always finds surprising and that a Nepali reader has never thought about, because it has always been true.

NEPSE is majority owned by the government.

The exchange itself is not a private company competing for listings. Its shareholders are the Government of Nepal, Nepal Rastra Bank, and a handful of state-linked institutions. It is regulated by SEBON, a government body. And the companies listed on it are, in overwhelming proportion, financial institutions supervised by Nepal Rastra Bank or the insurance regulator.

Consider the diagram that produces. A state-owned exchange, overseen by a state regulator, listing companies licensed by a state central bank. There is no adversarial party anywhere in the structure, no competing venue, no short seller, no activist fund, and until very recently no meaningful independent research. Every institution in the room has broadly the same interests, and the one participant with different interests — the outside minority shareholder — has no representative at the table.

I am not alleging conspiracy. Nothing about this structure requires bad faith and I do not think there is much. But a market’s architecture determines what kind of information gets produced, and an architecture with no adversarial participants produces very little.

Why the companies are actually listed

Here is the observation that reorganised how I think about this market, and I have never seen it stated plainly anywhere.

Most Nepali listed companies did not list in order to raise capital. They listed because a regulator required it.

Nepal Rastra Bank requires commercial banks, development banks and finance companies to maintain a public shareholding — the familiar promoter-to-public split, typically fifty-one to forty-nine. Insurance companies operate under a comparable requirement from their own regulator. A bank cannot simply remain private and well-capitalised; the licence carries an obligation to distribute a portion of its equity to the public and list it.

So run through the exchange in your head. Nineteen commercial banks. Sixteen development banks. Fifteen finance companies. Fifty microfinance institutions. Twenty-eight insurers. That is well over half the listed universe, and essentially none of it is there because a chief executive concluded that public equity was the cheapest available source of funding for an expansion plan.

They are there because of a rule.

Hydropower is the interesting exception and it goes the other way. A hydropower company genuinely needs the money — construction is enormously capital-hungry — and the state has deliberately used the exchange as the mechanism for mobilising household savings into generation capacity, including through local-shares provisions that reserve allocations for residents of the affected districts. Those companies wanted your capital. That is why there are a hundred and eight of them.

So the exchange has two populations with entirely different relationships to their own shareholders, and this single fact explains more about disclosure quality, dividend behaviour, and management attitude than any amount of governance theory.

A bank that listed because it was told to regards its public shareholders as a compliance obligation. It will publish what Schedule 14 and Schedule 15 require, on the last permitted day, in the format that costs least. It will not hold an earnings call. It will not explain a deterioration. Why would it? The shareholders did not choose it and it did not choose them.

A hydropower company that listed to fund a turbine has a genuine ongoing need for the market’s goodwill, because it will be back for a right issue within two years. Its communication is better, and its incentive to keep the share price up is real, and both of those facts cut in more than one direction, as Chapter Twelve showed.

The consequence: a market that is mostly secondary

Now return to the two functions.

In a developed market both operate, and the primary function is enormous — American companies raise and retire equity constantly, and the whole apparatus is oriented around that flow.

In Nepal the primary function is small, sporadic, and largely regulatory. New capital does get raised — the initial offerings, the right issues, the further public offerings — but the great bulk of what happens on any given day is the secondary function: existing shares changing hands between existing savers, with no rupee reaching any company.

That is not a criticism. It is a description, and it has three consequences that run through the rest of this book.

The market is a machine for reallocating household savings, not for allocating capital to projects. When the index rises 174 per cent, no factory has been built. Wealth has been transferred and re-marked, and the country’s productive capacity is exactly what it was.

Price discovery has few professional participants. In a market where companies are listed by obligation, nobody is being paid to work out what they are worth. There are no sell-side analysts of consequence covering the smaller names, no institutions running fundamental mandates at scale, no short sellers with an incentive to find the overstatement. Which means that a private person willing to do the work is not competing with a hundred well-resourced professionals, as he would be in New York, or even with ten, as he would be in Mumbai. This is the single most encouraging fact in this book, and Part Four is built on it.

And the exchange has almost no competitive pressure to improve. A private exchange competes for listings and order flow, and that competition drives down costs and drives up data quality. NEPSE has no competitor. When I come to what the data actually looks like, in Chapter Thirty-One, this will be the explanation for most of what I find.

What was built, and what it grew into

There is a way of reading the last thirty years that I find more useful than the triumphal version and more useful than the cynical one.

Nepal built a securities market in the 1990s as part of a general liberalisation, on templates borrowed from India and elsewhere, at a moment when the country had perhaps two dozen companies of a size that made listing meaningful. It was an institution constructed slightly ahead of its economy — a reasonable thing to do, since the alternative is to build it too late.

Then the banking sector expanded enormously, the regulator required those banks to float shares, and the exchange filled up with financial institutions. Then the state decided that hydropower should be funded domestically, and it filled up further. Then, in 2020, online trading arrived at the same moment as a lockdown and a deposit glut, and several hundred thousand people who had never owned a share discovered that they could buy one from a telephone.

At no point did anybody design this. The exchange we have is the accumulated residue of regulatory decisions taken for unrelated reasons over fifty years, which is exactly what the New York Stock Exchange is, and exactly what the Bombay Stock Exchange is, and exactly what every exchange is.

A market is not an instrument built for a purpose. It is a sediment. And the practical consequence is that you must never reason about it from what it ought to be. You must reason from what it is, layer by layer, which is what the remaining sixteen chapters of this part are for.

What is coming

Part Two is the least glamorous section of this book and the one I would keep if I could keep only one, because everything in Part Four depends on it.

The next chapter asks who actually owns the shares, and establishes that the answer determines the price more than the businesses do. Then the history: four booms, four busts, and the single cause underneath all of them, which is not the one anybody names. Then the plumbing — settlement, charges, the daily limit, the tax boundary, corporate actions, the calendar — six chapters of rules that decide what strategies are even expressible in this market. Then who is on the other side of your trades. Then what the data is really like, and one honest chapter on what happened when I spent a year trying to beat this market and could not.

None of it will tell you what to buy. All of it determines whether what you buy can be bought, held, and sold at a price that leaves you with anything.

The buttonwood tree is not there any more. It came down in a storm in 1865.

The commission floor those twenty-four men agreed on lasted a hundred and eighty-three years, until the American regulator abolished fixed commissions on the first of May 1975 — a date the industry called Mayday, because a great many firms did not survive it.

The thing they built to protect themselves outlived all of them and was destroyed by the only force that has ever reliably improved a market, which is somebody being allowed to charge less.

Fifty-One Per Cent

In which a German car maker becomes, for about an hour, the most valuable company on earth; two thirds of the Chinese stock market turns out not to be for sale; and we measure how much money it actually takes to move a Nepali share, and find that the answer changes depending on the year you ask.

On the twenty-eighth of October 2008, in the middle of the worst financial collapse since the 1930s, Volkswagen briefly became the most valuable company in the world.

Its shares had closed near €210 the previous Friday. Within two sessions they touched something over €1,000, which valued a European car manufacturer, during a global recession that was destroying car manufacturers, at roughly three hundred billion euros — more than ExxonMobil, more than anything.

Nothing had happened to the business. Volkswagen sold no more cars on the twenty-eighth of October than on the twenty-fourth.

What had happened was arithmetic about ownership. Porsche had been quietly building a position, and on the Sunday it disclosed that between shares held outright and cash-settled options it controlled about seventy-four per cent of the company. The state of Lower Saxony held a further twenty per cent and was not selling. It was not obliged to hold — the Volkswagen Law does not forbid a sale — but that law requires an eighty per cent supermajority for major resolutions, which turns any stake above twenty per cent into a blocking minority, and a blocking minority is not a thing a state gives up.

Which left, for anybody who actually wished to buy a Volkswagen share, something under six per cent of the company.

And approximately twelve per cent of the company had been sold short.

Every one of those short sellers now had to buy shares that did not exist in a quantity sufficient to satisfy them. They bid against each other into a supply of nearly nothing, and the price went where a price goes when demand is finite and supply is zero.

The market capitalisation of Volkswagen that day was not a statement about Volkswagen. It was a statement about how many shares were available.

The number that is not the number

This is the most under-appreciated idea in market structure, so let me state it plainly.

A company’s market capitalisation is the price of the last share traded multiplied by every share in existence. But the price of the last share traded was set by the balance of buyers and sellers among the shares that can trade, and in most of the world that is a fraction — sometimes a small fraction — of the total.

The gap between those two quantities is where an enormous amount of trouble lives.

The world’s index providers eventually noticed. Until the early 2000s the major global indices weighted companies by full market capitalisation, which meant that a company whose shares were mostly locked up in a founding family or a government got an index weight far larger than the amount of stock an investor could actually buy. Funds tracking the index were obliged to purchase quantities that were not for sale, which pushed prices up, which increased the weight further.

MSCI reconstructed its entire index family onto a free-float basis in two phases across 2001 and 2002 — a rebalancing so large it moved markets for weeks. The principle it enshrined is the one this chapter is about: the investable size of a company is not its size.

Two thirds of China was not for sale

The most extreme case in modern history is China, and it is worth knowing because Nepal’s structure is a cousin of it.

When Chinese companies listed in the 1990s, the state retained large blocks that were legally non-tradable. State shares and legal-person shares could not be sold on the exchange at all. At the peak of the arrangement, roughly two thirds of the shares in Chinese listed companies could never be bought by an ordinary investor.

So Chinese prices were being set on the remaining third. And every participant knew that the other two thirds existed, overhanging the market like a held breath: if the state ever converted them, the effective supply of stock would triple.

The split-share structure reform began in 2005 and did exactly that, with compensation paid to public shareholders for the dilution of their scarcity. The market fell in anticipation, then rose enormously once the uncertainty resolved.

A comparable, quieter version has been running in Japan for thirty years. Post-war Japanese companies held large cross-shareholdings in one another — the keiretsu structure — and those blocks never traded. They have been unwound slowly since the 1990s, and each unwinding adds supply that was previously invisible.

The general rule: the number of shares that can be sold is a policy variable, and policy changes.

Nepal’s version

Now the local structure, which is unusually rigid even by the standards above.

Nepali financial institutions are required to maintain a split between promoter shares and public shares, and the customary proportion is fifty-one to forty-nine. The promoter portion belongs to the founders and institutional sponsors. It is not merely a description of who happens to own it; it is a separate class of security, recorded separately, subject to lock-in periods, requiring regulatory approval to transfer, and trading — when it trades at all — in a distinct market at a distinct price, customarily at a discount to the public shares of the same company.

Read that again, because it is genuinely strange and Nepalis have stopped noticing it. Two claims on the same stream of earnings, in the same company, with the same voting economics in most respects, trade at two different prices, permanently, because one of them is easier to sell than the other.

That price gap is the value of liquidity, made visible. Almost nowhere else in the world can you observe it so directly, and a serious analyst should look at it, because when the promoter–public discount widens, it is telling you something about how much the market is currently paying for the ability to change its mind.

So the maximum float in a Nepali bank is forty-nine per cent. And the actual float is considerably less, because within that forty-nine sit institutions that never sell, employees holding allotments, and a large population of investors who bought in an initial offering years ago, put the certificate in a drawer, and have not looked since.

So how much money does it take to move a share?

At this point the conventional thing to say is that NEPSE is desperately thin and that a modest sum can move any price. I believed this myself and repeated it, and then I went and measured it, and the measurement is more interesting than the cliché.

I took every symbol in the exchange’s sector universe, computed the rupee value traded in each session over the twelve months to the end of July 2026, and took the median.

The median name on the exchange trades about NPR 8.3 million a day. Of two hundred and sixty-one names with a reasonable trading record, only nine trade under a million rupees a day, and none trades under two hundred and fifty thousand.

That is not a thin market. That is, for a private investor in Nepal, an entirely adequate market.

Work it through with a participation rule — the sensible convention that you should not be more than about five per cent of a day’s volume, or you are no longer trading in the market, you are the market.

So an investor building a five-lakh position in the median Nepali stock is not constrained by liquidity at all. At one crore in a single name he is — a month of patient buying — but that is a constraint about him, and it arrives at a size most readers will never reach in a single security.

I report this because it contradicts something I expected to find, and because the practical implication runs the opposite way from the folklore: for most Nepali investors, liquidity is not the binding constraint. Knowledge is. You are far more likely to be hurt by owning something you have not understood than by being unable to sell it.

Except that liquidity is a phase

Now the qualification, and it is the important half of the chapter.

I ran the same measurement across three regimes.

From boom to bust the median stock’s daily turnover fell by about sixty-eight per cent.

This is the fact that matters, and it is the one that the comfortable numbers above conceal. Liquidity in this market is not a property of a security. It is a property of the mood, and it evaporates precisely when you need it — because the circumstances in which you urgently want to sell are, by construction, the circumstances in which everybody else does too and nobody wants to buy.

Which produces the only rule from this chapter worth writing on a card:

Size your position against the liquidity of the worst year, not the current one.

If a name trades eight million rupees a day today, assume three million when you need to leave. If you would need forty sessions to exit at that rate, you do not have a position. You have a commitment.

The float and the cascade

One last connection, back to Chapter Six.

I argued there that a Nepali share can be moved ten per cent by a few hundred motivated buyers, and that the resulting price move gets screenshotted and forwarded as though it were new information, producing a self-feeding loop.

The float is the reason the loop closes. In a market where the true tradeable quantity of a mid-sized company is a modest fraction of a modest company, the amount of buying required to hit the daily limit is well within the reach of a group chat. The cascade in Chapter Six is a psychological mechanism. The float is the physical mechanism that lets it move a price. Neither works without the other.

And the same arithmetic explains the pattern noted in Chapter Nine — nothing, nothing, nothing, four limit-up days, nothing for two months. A thin float does not absorb a burst of demand gradually. It absorbs nothing, hits the limit, and then, once the burst is exhausted, there is no natural buyer to hold the level, so the price simply stops.

What to actually do with this

Look at the promoter–public discount. It is published, it is free, and it is the market’s own estimate of what liquidity is worth in that company at that moment. A widening discount in a name you own is information.

Check the turnover, not the price, before you buy. Two minutes. Median rupee turnover over the last few months, then divide your intended position by five per cent of it. If the answer exceeds ten sessions, halve the position or choose a different company.

And treat any company where the float is being changed as a different company. A right issue, a further public offering, the release of a lock-in, a promoter selling down — each of these alters the supply of stock, and the price effect of a supply change has nothing to do with the value of the business. Chapter Twenty-Seven is about what these actions do to your per-share arithmetic, which is a separate and equally neglected question.

Porsche’s manoeuvre worked, in the sense that it made a great deal of money at the expense of a great many hedge funds, at least one of whose principals is reported to have taken his own life shortly afterwards.

It did not work in the longer sense. Porsche had borrowed enormously to build the position, the credit markets closed, and within a year the hunter had been swallowed by the prey: Volkswagen acquired Porsche’s automotive business, and the man who had engineered the squeeze left the company.

The float he had cornered was worth exactly what it was worth on the day the shorts had to buy, and nothing at all the following spring.

Four Booms, One Cause

In which we look at every major move the Nepal Stock Exchange has made in twenty-nine years, ask what caused each of them, and find the same answer four times — an answer that has almost nothing to do with the companies, and which turns out to be a thermometer rather than a clock.

The single most repeated sentence in American financial commentary is don’t fight the Fed, and like most repeated sentences it is a compression of something real.

Here is the real thing underneath it, and it is not about America.

A share is a claim on money that arrives later. To decide what that claim is worth today you must compare it with the alternative — what you could earn by not owning it and holding something safe instead. When the safe alternative pays very little, the distant money is worth more today. When the safe alternative pays a great deal, the same distant money is worth less.

This is not a theory about investor psychology. It is arithmetic, and it operates on every asset in every country simultaneously, and it is the reason a global bond market and a global equity market that have nothing to do with each other move together.

The extreme demonstration is Japan in the 1980s, where the Bank of Japan held rates down through the second half of the decade and the Nikkei tripled while land under Tokyo reached prices that made the whole of California look cheap. The clearest modern demonstration is 2009 to 2021, when American policy rates sat near zero for most of thirteen years and equity multiples expanded to levels that no earnings forecast justified and no earnings forecast needed to.

And the most instructive demonstration for our purposes is Turkey, where real interest rates were driven deeply negative in 2021 and 2022 and the Istanbul index rose several hundred per cent in nominal terms while the currency collapsed — an enormous bull market that made its participants poorer.

Now let us look at Nepal, which has no foreign investors, no derivatives, a currency pegged to the Indian rupee, and a domestic savings pool with essentially two places to go.

Four booms

Here is the complete record of what this market has done since the index series begins in July 1997.

Four complete cycles in twenty-nine years. Roughly seven years each — three to five years up, two to three years down, then a flat period nobody counts because nothing happens in it.

Now, what caused them?

The conventional answers are the ones you will hear on any panel: political stability, the peace process, remittance growth, the reconstruction after the earthquake, government policy, investor awareness. Each of these was cited at the time, in newspapers I have read, as the reason for whichever move was then in progress.

I want to test them against a single number.

The number

Nepal Rastra Bank publishes the weighted average deposit rate of commercial banks — not a headline poster rate on some fixed-deposit product, but the rate actually earned across the entire deposit book of the banking system. That is the correct number for our purposes, because it is what a Nepali saver genuinely receives for choosing not to own shares.

I could obtain it in machine-readable form for eight fiscal years, observed at mid-July.

Look at the two lowest rates in the series.

3.28 per cent, in July 2016. The index had peaked ten days earlier, on the twenty-seventh of July 2016, at 1,881.5 — the top of Boom III.

4.65 per cent, in July 2021. The index peaked four weeks later, on the eighteenth of August 2021, at 3,198.2 — the top of Boom IV.

The two cheapest years of money in the observed record sit at the two market tops. Not near them. At them.

And the two highest rates, 7.41 and 7.86 per cent in 2022 and 2023, sit at the bottom of the subsequent collapse.

The mechanism, which is not what people think

The tempting reading is that low rates cause the market to rise, and it is not wrong, but it is missing the interesting half.

In Nepal the causation runs through the banking system in a specific, physical way.

Deposits are the raw material. Nepali banks lend what they take in. Their capacity to lend is bounded by deposits and by the regulatory ratios that govern the relationship between the two. When deposit growth is fast — because remittances are surging, or because nobody is spending, or because the government has released capital expenditure into the system at year end — banks find themselves with more funds than lending opportunities.

Surplus deposits push the price of money down. Banks compete less hard for deposits, so the rate they pay falls. Interbank rates fall. The system is described, in the local vocabulary, as having “excess liquidity.”

And the money goes looking. Some of it becomes credit, some of it becomes margin lending against shares, and a great deal of it simply becomes a comparison in a saver’s mind: my fixed deposit is paying four and a half per cent, and that hydropower company paid a fourteen per cent dividend last year.

Then the sequence reverses. Credit growth outruns deposit growth, banks compete for funds, the deposit rate rises, the regulator tightens the ratios, margin lending becomes expensive, and the same saver runs the same comparison in the opposite direction.

The cycle in Nepali share prices is the cycle in Nepali bank funding, transmitted almost undamped, because there is nothing else in the room.

This also explains something that puzzles people about the composition of the index — that a market more than half made of financial institutions moves as a single object. It is not merely that banks are correlated with each other. It is that the thing that moves the banks is the same thing that moves everybody’s willingness to own anything. A hydropower company in Dolakha with a twenty-five-year power purchase agreement has no operational connection whatsoever to interbank liquidity in Kathmandu, and its share price is nonetheless governed by it.

Now the honest part

I have just presented an attractive story with a mechanism and a table, and this is precisely the point in a book where the author normally stops.

So let me do the thing Part One spent sixteen chapters demanding.

Eight observations. That is the entire sample. And they are not eight independent observations — they are one series with about four turns in it. The rate rose through the 2016–19 bear, fell through the 2019–21 bull, rose through the 2021–24 collapse, and the earlier period is extrapolated rather than measured.

Four turns. An effective sample size of roughly four, whatever the number of days suggests. Chapter One’s arithmetic applies without mercy: Bernoulli wanted twenty-five thousand trials to settle an urn with two colours in it, and I am proposing a relationship on the basis of four.

Worse, look at the middle of the table. July 2020: the deposit rate was 6.01 per cent — unremarkable, near the middle of the range — and the following twelve months returned 111.6 per cent. The single largest move in the series came out of an ordinary rate. If you had been running a rule keyed to the deposit rate, you would have missed the greatest bull market in the country’s history.

I have tested a rule of exactly this kind — hold less equity when the prevailing rate is high — against the price panel. It scored about two thirds of a percentage point a year in its favour. And I will not use it, and I do not recommend it, because a signal that turns four times in eleven years cannot be distinguished from luck by any test that exists. It is an unfalsified hypothesis, not a finding, and the distinction is the whole of Chapter Four.

A thermometer, not a clock

So what is the deposit rate actually good for?

Read the table one more time, but read it as a description of the present rather than a forecast of the future.

When the deposit rate is at a multi-year low, you are already late in a boom. You are not being told to buy. You are being told where you are standing: in a market that has been lifted by cheap money, at a moment when the safe alternative is paying you almost nothing to be patient, which is exactly when patience is least fashionable and most valuable.

When the deposit rate is at a multi-year high, you are being paid handsomely to wait, and shares are being priced against a hostile alternative, and the businesses underneath them have not changed.

This is Chapter Eight arriving with a number attached. The rate does not forecast the market. It measures the temperature of the room, and it does so in a published, objective, unfakeable figure rather than in the anecdotal thermometers I offered earlier — who is talking about shares at weddings, whether people are borrowing, what vocabulary they use.

And it tells you the thing that actually matters for a decision, which is not what will happen but what am I being paid. In July 2021 you were being paid 4.65 per cent to hold cash while the index sat at an all-time high. In July 2023 you were being paid 7.86 per cent to hold cash while the index sat forty per cent below that high. Those are two completely different propositions, and neither of them required a forecast.

The four booms, retold

With the mechanism in hand, the history reads differently.

Boom I, 1999–2000. A tiny market — a few dozen listed companies, most of them banks, in a country with a nascent private financial sector. It tripled and then lost two thirds as the insurgency intensified and the royal massacre of 2001 removed whatever confidence remained. I have no deposit-rate data for this period and will not pretend the pattern is established here.

Boom II, 2002–2008. The conflict ended in 2006. Remittances, which had been modest, began the extraordinary climb that made them a quarter of national income. Deposits flooded into a banking system that was simultaneously being expanded by licensing, and the index went up more than fivefold. Then it fell 75 per cent — the largest decline in Nepali market history — through a global financial crisis and a domestic liquidity squeeze that arrived together.

Boom III, 2011–2016. Reconstruction after the earthquake, continued remittance growth, and a banking system that by 2016 was paying its depositors 3.28 per cent. The index peaked in July 2016 at 1,881.5 and did not recover that level for four and a half years.

Boom IV, 2019–2021. Covered in Chapter Nine as an experience; here is the mechanism. The exchange went online. The country was locked down and could not spend. Remittances, against every expectation, rose. Deposits accumulated with nowhere to go and the deposit rate fell to 4.65 per cent by July 2021. The index reached 3,198.2 on the eighteenth of August.

Four booms. In every case for which I have data, the same fuel.

What this means for you

Do not attribute the market’s direction to the news. The news is the story people tell about the liquidity. In 2078 the story was hydropower and the awareness of a new generation; in 2073 it was reconstruction; in 2064 it was peace. The stories were all true and none of them was the cause.

Watch the deposit rate, and do not trade on it. Check it once a quarter. Use it to answer the question in Chapter Eight — what am I being paid to take risk right now — and let it inform how fast you deploy rather than whether you participate.

And understand that when the cycle turns, it turns for everything. There is no sector in Nepal that is insulated from the banking system’s funding position, because there is no pool of capital in this country that is insulated from it. Chapter Twenty-Two takes up what that does to the value of diversification, and the answer is unwelcome.

The next chapter does something simpler and more uncomfortable. It takes each of those four peaks, puts a man at the top of it with money, and asks what happened to him.

Three of the four are still alive, in the sense that the money eventually came back.

One of them is you.

The Arithmetic of Buying a Peak

In which the unluckiest investor in America puts money in only at market tops and retires comfortably; a Japanese saver does the same and does not; and we compute what happened to four men who bought the Nepal Stock Exchange on the four worst days in its history.

There is a parable that circulates among American financial writers, and its protagonist is usually called Bob.

Bob begins saving in 1970 and has the worst timing of any investor in history — not bad luck, perfect bad luck. He saves patiently in cash and then invests his entire accumulated balance at the exact top of the market. Every time. He puts money in just before the 1973 collapse, again just before 1987, again just before the dot-com peak in 2000, and again in October 2007.

Four investments, all at the four worst possible moments across four decades.

He retires wealthy.

The reason is not subtle and it is the entire point of the parable: Bob never sold. He had no ability to time the market and he did not attempt to, and because he never realised any of the losses, the eventual recoveries reached him in full. The lesson usually drawn is that time in the market beats timing the market, and it is generally drawn with a certain smugness.

I want to take the parable seriously, because it is either the most reassuring fact in investing or a survivorship story of exactly the kind Chapter Three was about, and which it is depends entirely on the market you run it in.

Bob in Tokyo

Run the same man in Japan.

He invests his accumulated savings at the peak of the Nikkei on the twenty-ninth of December 1989, at 38,915.87. He never sells, because that is his defining characteristic.

He gets his money back in February 2024.

Thirty-four years and two months. If he was thirty-five when he invested he was sixty-nine when he broke even, in nominal terms, before considering what a third of a century of even modest inflation did to the purchasing power of the money he recovered.

Same man. Same discipline. Same virtue. One market rewarded it and one did not, and he had no way to know in advance which he was standing in.

So the honest version of the parable is not never selling works. It is: never selling converts a timing problem into a duration problem, and whether that is a good trade depends on how long you have.

Which means the question is not whether you can survive buying a peak. It is what buying a peak actually costs, in the specific market you are in, measured properly.

So let me measure it here.

Four men in Kathmandu

Take four investors, each of whom put a lump sum into the Nepali market on the exact day of one of its four great peaks, and each of whom never sold.

I have compounded each of them to the thirtieth of July 2026.

Now the comparison that matters. The realised weighted average deposit rate across the period I have measured properly was 5.31 per cent.

So: the man who bought the 2000 peak beat a deposit account. The man who bought the 2008 peak roughly matched it, slightly behind. The man who bought 2016 lost to it by nearly two points a year for a decade. And the man who bought August 2021 has spent five years compounding backwards while his neighbour, who did nothing at all, earned between five and eight per cent.

One large caveat, stated before you use these numbers

The NEPSE index is a price index. It does not include cash dividends.

This matters more in Nepal than in most markets, because Nepali companies — particularly the banks and microfinance institutions that dominate the index — have historically distributed a meaningful portion of earnings in cash, alongside the bonus shares which the index does capture mechanically.

I do not have a reliable long-run cash dividend yield series for this market, and I am not going to invent one. But if the average cash yield across these periods was of the order of two and a half per cent, the true picture becomes:

On that adjustment, three of the four peak-buyers beat a deposit account despite having chosen the worst day of a generation to invest. Which is a genuinely encouraging result and I want it on the record, flagged as resting on an assumption I could not verify.

The peak-buyer’s punishment in Nepal is real, and it is a few percentage points a year for a decade, not a wipeout. That is worth knowing, because the fear of buying a top keeps a great many people in deposits permanently, and the fear is disproportionate to the measured penalty.

The man who kept going

Now the calculation that reorganises the whole chapter, and it is the reason I built the price panel in the first place.

The four men above were finished. Each invested a lump sum at a peak and then had nothing further to add — which is a very particular assumption, and an odd one, since most people who invest at a market top are doing so because they have income and are saving from it.

So consider instead a fifth kind of investor. He also begins on the worst possible day. But he has a salary, and every month, without judgment or timing or opinion, he puts a thousand rupees into the index.

Here is what happened to him, computed as a proper money-weighted return on the actual index series.

Compare the columns.

The man who bought the 2008 peak with a lump sum earned 4.69 per cent. The man who started buying on the same day and kept going earned 10.95 per cent — more than double, on a price index, with no skill, no timing, and no view about anything.

And look at the last row, which is the one that should reach you. The August 2021 peak-buyer is down 3.56 per cent a year and miserable. The man who started on that same catastrophic morning and simply kept putting in a thousand rupees a month is up 5.49 per cent a year, which is roughly a deposit account, after the worst five-year stretch in recent Nepali market history.

He bought the top of the greatest bubble in the country’s history and came out level. Because he was not finished when he started.

Why this works, mechanically

The reason is not mystical and it is not the folk explanation about “averaging” being inherently clever.

It is that a declining market increases the number of units your money buys while the price falls, so your average cost is dragged toward the bottom of the range rather than sitting at the top of it. The peak purchase becomes a smaller and smaller fraction of your total position with every month that passes.

The 2021 investor’s first thousand rupees bought units at 3,198. His thirtieth thousand bought them at around 1,900. His most recent bought them at 2,672. His average cost is nowhere near the peak, and the peak — the thing that ruined the lump-sum man — is now about one-sixtieth of his book.

Note carefully what this does and does not claim. It is not an argument that staged buying beats a lump sum in general; the evidence there runs the other way, because markets rise more often than they fall and delaying is usually costly. It is a narrower and more useful claim:

The damage from bad timing is a function of what fraction of your lifetime capital was deployed at the bad moment. If it was all of it, the damage is permanent. If it was two per cent of it, the bad moment is an anecdote.

Which reframes the entire anxiety. The question is never is this a good time to buy. The question is how much of what I will ever have am I committing right now — and that is a question you can actually answer, today, without a forecast.

The Nifty Fifty, for balance

There is a well-known counterweight, and honesty requires it.

In the early 1970s American investors bid a group of about fifty large, high-quality growth companies — the Nifty Fifty — to extraordinary valuations, some of them at eighty or ninety times earnings, on the reasoning that these were “one-decision” stocks you bought and never sold. The market peaked in December 1972 and the group collapsed in the 1973–74 bear market, some names losing eighty and ninety per cent.

It became the standard cautionary tale about paying any price for quality.

Then Jeremy Siegel went back and measured it properly, from the December 1972 peak through to the late 1990s, and found that the group as a whole had roughly matched the S&P 500 over the subsequent quarter-century. The great businesses in it — the ones that really did compound for decades — carried the ones that did not.

Two lessons sit inside that, and they point in opposite directions, which is why I find it useful.

The first is that paying a foolish price for a genuinely great business is survivable over a long enough horizon, which is a real and underappreciated point.

The second is that the average concealed enormous dispersion. Some of those companies were superb and some were Polaroid, and the investor who bought five of them rather than fifty had a completely different experience — which is Chapter Thirteen’s concentration arithmetic arriving again, in a different decade and a different country.

What to take from this

Buying at a peak is a few points a year, not a catastrophe — provided the market you are in eventually recovers, which Nepal’s has, four times out of four, in four to seven years, and which Japan’s took thirty-four.

Being finished when you buy is the actual risk. The lump-sum peak-buyer and the monthly peak-buyer started on the same day with the same view and ended six percentage points a year apart. Nothing separated them except the presence of future savings.

Which means the most valuable financial asset most readers of this book have is not in their demat account. It is their future income, and the ability to keep directing it into a falling market — and that ability is destroyed not by markets but by over-commitment, by leverage, and by having spent it all when things felt good.

That is the same conclusion as Chapter Ten’s, arrived at from a different direction, and I did not arrange for that. It is simply what the arithmetic keeps saying.

Bob, in the American parable, retires wealthy on four perfectly mistimed investments.

The detail the parable never mentions is that Bob was employed for forty-five years, saved continuously in cash between his investments, and lived in a market that rose about six per cent a year in real terms across the whole period — a very good run, though not the best on record: on the long cross-country data, South Africa and Australia both beat the United States over the twentieth century.

Change any one of those three and Bob is a different story with a different moral, and nobody writes it down.

Liquidity Is the Market

In which a Victorian journalist works out what a central bank is for; we trace every rupee that enters and leaves the Nepali banking system; and we measure how much diversification a Nepali portfolio actually provides, and find that in the year you need it, a twenty-stock portfolio is worth about one and a half bets.

In 1873 Walter Bagehot, who edited The Economist and had no formal training in anything relevant, published a book describing how the London money market actually worked as opposed to how everyone assumed it worked.

Lombard Street is still read, and its famous prescription — that in a panic the central bank should lend freely, at a penalty rate, against good collateral — is the foundation of central banking practice everywhere.

But the prescription is not the important part. The important part is the diagnosis underneath it, which was genuinely novel and which almost nobody had articulated before: that a modern economy sits on a pyramid of promises, that the base of the pyramid is a quantity of actual reserves far smaller than the promises resting on it, and that the whole structure is therefore governed not by the wealth of the nation but by the funding position of its banks.

Bagehot’s insight is that the banking system’s liquidity is not one economic variable among many. It is the weather in which every other variable occurs.

That is true in London and it is true in New York, where it is obscured by the existence of bond markets, foreign investors, insurers, pension funds and a dozen other pools of capital that can act independently of the banks.

In Nepal it is not obscured by anything at all, and this chapter is about what that looks like.

Why Nepal is the pure case

Consider what does not exist here.

There is no corporate bond market of consequence. A Nepali company that wants long-term money borrows from a bank or issues shares; it does not issue paper to a pension fund.

There are no foreign investors in the secondary market. Nothing arrives from outside seeking a return, and nothing flees when sentiment turns elsewhere. Whatever happens to Nepali share prices is done by Nepalis with Nepali savings.

There are no derivatives, no short selling, and no securities lending, so there is no mechanism by which a participant can express a view without moving actual money.

And there is one dominant alternative to owning shares: a bank deposit.

Put those together. The entire pool of capital that can bid for Nepali shares is the domestic savings pool, and that pool sits inside the banking system, and the price of staying inside the banking system is the deposit rate.

There is one tank. Everything drinks from it.

The four taps

Money enters the Nepali banking system through four channels, and it is worth knowing each, because their rhythms explain most of what the market does.

Remittances. The largest by a distance. Nepalis working in the Gulf, Malaysia, Korea and India send home an amount equivalent to roughly a quarter of national income — among the highest ratios in the world — and it arrives, overwhelmingly, through the banks. This is not a trade flow that adjusts to interest rates or sentiment. It is a demographic fact, and it makes the Nepali banking system unusually deposit-rich for a country at this income level.

Government spending. The state collects revenue steadily through the year and spends it very unevenly, with capital expenditure clustering into the final months of the fiscal year. When the government finally releases money in Jestha and Asar, it lands in contractors’ accounts, which are bank deposits.

Foreign aid and concessional borrowing, which behave like government spending with a longer lag.

And repayment of existing credit, which returns funds to banks continuously and invisibly.

The four drains

Imports. This is the big one and the one that catches people out. Nepal imports far more than it exports, and imports must be paid for in foreign currency. When import demand surges, foreign exchange leaves the country, and the rupee liquidity used to buy that currency is withdrawn from the system. An import boom is a liquidity drain, and because the Nepali rupee is pegged to the Indian rupee, the central bank cannot let the exchange rate absorb the pressure. It must manage the quantity instead — which means restricting imports, or letting reserves fall, or tightening credit.

Festival cash. In the weeks around Dashain and Tihar an enormous quantity of currency is withdrawn from banks and circulates as physical notes. It comes back afterwards, but for a period the banking system’s deposits are genuinely smaller.

Credit growth. When banks lend aggressively, their capacity to lend further is consumed, and they must compete for deposits to fund it.

And revenue collection, which pulls money out of accounts and into the treasury, in a mirror image of the spending tap.

The thermometers

Three published numbers tell you where the system stands, and a Nepali investor who checks them quarterly knows more about the market’s near-term direction than anybody reading charts.

The interbank rate. What banks charge each other to borrow overnight. When the system is flush this collapses toward almost nothing; when it is tight it rises sharply. It is the fastest and most honest indicator available and it is published daily.

The credit-to-deposit ratio. The regulator caps how much of its deposit base a bank may lend, and the aggregate ratio tells you how much headroom the system has left. When banks are pressed against the cap, credit stops regardless of demand, and the marginal borrower — including the man who wanted to borrow against his shares — is refused. The specific permitted ratio and its precise definition have been changed by Nepal Rastra Bank more than once; verify the figure in force before relying on it, and treat the direction of travel as the signal rather than the level.

The deposit rate, which was Chapter Nineteen’s subject and which is the slowest of the three but the one that reaches savers’ decisions.

The 2021 case, traced end to end

The cleanest illustration is the collapse that began in August 2021, because every link in the chain is visible.

Through 2020 and early 2021 the country was locked down. Imports fell because nobody was building or buying. Remittances rose, contrary to every forecast, because migrant workers facing uncertainty sent money home rather than holding it abroad. Deposits accumulated. Credit demand was weak because businesses were shut.

The result was a banking system with more money than it could deploy. Interbank rates fell to almost nothing. The weighted average deposit rate reached 4.65 per cent by mid-July 2021, the second-lowest observation in the record.

Money went looking for a return and found the exchange.

Then the country reopened. And a year of postponed consumption and postponed construction arrived at once. Imports surged — vehicles, machinery, consumer goods — and every one of them was paid for in foreign currency. Reserves fell sharply enough to become a political subject. The central bank responded as a central bank with a pegged currency must: it restricted imports of selected goods, and it tightened credit.

Simultaneously, credit growth had outrun deposit growth, so banks were pressed against their lending ratios and began competing for deposits. The deposit rate went from 4.65 per cent to 7.41 and then 7.86.

Now put yourself in a Nepali saver’s position in Ashoj 2079. Your shares are down forty per cent. Your bank is offering you close to eight per cent, guaranteed, on a fixed deposit. Margin lending has become expensive and in some cases unavailable.

The index bottomed at 1,815.10 on the twenty-fifth of September 2022.

No company failed. No earnings collapsed. Nothing happened to the businesses. A tank filled and then drained, and the price of everything followed.

Which is why nothing here is diversified

Now the consequence, and it is the most practically important measurement in this chapter.

If every asset in the country is drinking from the same tank, then the things you own are not independent of each other. They are versions of the same bet.

I measured this. For each of four periods I took the sixty most consistently traded names on the exchange, computed daily returns, and calculated the correlation of every pair — 1,770 pairs in each period.

Read the first and last rows together.

In the boom, Nepali stocks were correlated at about 0.31 — respectable, roughly what you would see in a developed market. In the bust, the average pair moved together at 0.64, more than double.

Correlation rises exactly when you need it to fall. This is a well-documented international phenomenon, but I had never seen it measured for NEPSE, and the magnitude here is severe.

Now convert it into something usable, using the formula from Chapter Six. The effective number of independent positions in a portfolio of n holdings correlated at ρ is n ÷ (1 + (n−1)ρ).

Look at the last column, and then look down it.

A twenty-stock Nepali portfolio, in the year it matters, is worth about one and a half independent bets. So is a forty-stock portfolio. So is a ten-stock portfolio.

Beyond roughly ten holdings, adding names to a Nepali portfolio does essentially nothing for your risk in a bad year. The correlation floor has been reached. You are not diversifying; you are collecting.

This has two immediate consequences, and they run in opposite directions from what most people assume.

Owning thirty Nepali stocks is not safer than owning twelve. It is merely more work, more depository fees, and — as Chapter Fifteen argued — a guarantee that you have not read the filings of most of them.

And genuine diversification for a Nepali investor cannot be found inside the Nepali stock market. It has to come from assets that do not drink from the same tank: fixed deposits, which pay most exactly when shares pay least; property, with its own problems; and, for those able to access it, assets outside the country. That is not a comfortable conclusion and I am aware it is easier to state than to act on given the restrictions on Nepalis holding foreign assets. But the measurement says what it says.

The one exception worth naming

There is a partial exception and it appeared in the sector table in Chapter Twelve.

Hotels had the worst boom of any sector — 2.09× while hydropower did 3.88× — and the only positive sequel on the board, at 1.93×. Over the full six years they beat everything.

Why? Because their fortunes were driven by something outside the tank: the collapse and then the recovery of international tourism. Their cycle was out of phase with the domestic liquidity cycle, so they fell when everything rose and rose when everything fell.

Three companies. It is not a strategy and I would not build a portfolio on it. But it is a demonstration that the correlation of 0.64 is an average, and that the exceptions to it are the businesses whose revenues come from outside the Nepali banking system’s weather — which is a genuinely useful thing to look for, and which I will return to when the sectors get their own chapters.

What to do

Check the interbank rate and the deposit rate quarterly. Not to time anything — Chapter Nineteen established the sample is too small for that — but to know what you are being paid to wait, which is the only question from Chapter Eight that has an answer.

Stop counting tickers and start counting taps. Before adding a holding, ask what would impair it. If the answer is “a tightening in the banking system,” you already own that risk many times.

Cap your holdings at what you can actually read. The correlation table removes the main argument for owning more, and Chapter Fifteen’s arithmetic removes the rest.

And treat the tank as the cycle. When you next hear an explanation of why the market is rising that involves awareness, or reforms, or a sector’s prospects, ask what the deposit rate is doing. The explanation may be true. It is very unlikely to be the cause.

Bagehot wrote that money will not manage itself, and that Lombard Street had a great deal of money to manage.

Nepal has a great deal of money to manage too — a quarter of national income arriving every year from people who left in order to send it — and one channel through which all of it flows.

The stock exchange sits at the end of that channel, and calls what happens to it sentiment.

The Index Is Not the Market

In which the most famous number in finance turns out to be computed by a method nobody would choose today; the average American company and the American index part company for a decade; and the NEPSE index is revealed to be a report on nineteen banks wearing the costume of a nation.

The Dow Jones Industrial Average is the most quoted number in the world, and it is computed by adding up the share prices of thirty companies and dividing by a constant.

Not their market values. Their share prices.

Which means that a company whose stock happens to trade at nine hundred dollars has roughly nine times the influence on the Dow as a company trading at a hundred, regardless of whether the second is ten times larger. When a company splits its stock — an event of no economic consequence whatever — its influence on the index falls proportionately. When Apple split four-for-one in 2020, its weight in the Dow was quartered overnight, and nothing at all had happened to Apple.

Charles Dow devised this in 1896 because he was computing it by hand with a pencil, and dividing a sum of prices was what a man with a pencil could do before dinner. It has never been fixed, because the number has become a cultural object rather than an analytical one.

Japan’s Nikkei 225 has the same defect, and it produces the same absurdity: a handful of high-priced stocks dominate a benchmark that is supposed to describe an economy.

I begin here to establish the general point, which people nod at and then immediately forget:

An index is not a measurement of a market. It is a construction, built by a particular method, and the method has an opinion embedded in it.

Cap weighting owns more of whatever went up

Almost every modern index — the S&P 500, and NEPSE’s — weights by market capitalisation. Bigger company, bigger weight. This is far more defensible than price weighting, and it has one enormously convenient property: it is the only weighting scheme that everybody can hold simultaneously, since it is the market portfolio.

But it has a consequence that is rarely stated plainly. A company’s weight in a cap-weighted index rises as its price rises. You automatically own more of whatever has already gone up and less of whatever has gone down. Momentum is not a strategy layered onto a cap-weighted index; it is the index’s construction.

In a market whose largest constituents are excellent, this is wonderful. The American index’s returns over the last fifteen years came overwhelmingly from letting a handful of technology companies grow into an enormous share of the whole — by the mid-2020s the ten largest names were around a third of the S&P 500, a concentration not seen since the early 1970s.

In a market whose largest constituents are mediocre, the same mechanism is a trap. And that is the situation in Nepal.

What the NEPSE index actually contains

The exchange publishes several indices and I have the full history of four of them. Here is what they did.

The decade from the July 2016 peak to July 2026:

Read the last row.

Over ten full years, the Nepali banking sector index lost money. Not underperformed — declined, in nominal terms, in a country with meaningful inflation, over a decade in which bank balance sheets grew enormously.

And now recall from Chapter Seventeen why the exchange is composed the way it is: because Nepal Rastra Bank requires financial institutions to float shares. Nineteen commercial banks, sixteen development banks, fifteen finance companies, fifty microfinance institutions, twenty-eight insurers. More than half the listed universe by count and a much larger share by market capitalisation.

So when you look at the NEPSE index and conclude that Nepali equities returned 3.6 per cent a year over the last decade, you have not learned what Nepali companies did. You have learned, mostly, what happened to a set of banks whose shares are listed because a regulator said they must be.

The average company did something else entirely

I wanted to know what the typical listed Nepali company had done, as distinct from the capitalisation-weighted average, so I built an equal-weighted index — every company counting the same, rebalanced monthly — and ran it over the same decade.

The first version I built returned 5.29×.

Against the index’s 1.42×. An enormous gap, and my immediate instinct was that I had found something important.

I had found a bias, and it is precisely the bias of Chapter Three, and I want to show the whole decomposition because catching yourself is more instructive than being right.

Three separate defects were inflating the first number.

Daily rebalancing. Rebalancing every session captures a large mechanical bonus from mean reversion in noisy prices, and it is completely unachievable — Chapter Fourteen’s flat twenty-five rupees per scrip per settlement would consume it many times over. Moving to monthly took the result from 5.29× to 3.64×.

New entrants. Using today’s listed universe means the index silently acquires companies at the moment they list — including, in Nepal, at their post-listing prices after the customary opening surge. A 2016 investor could not have owned a company that listed in 2022.

And survivorship. Today’s universe excludes everything that was absorbed or delisted along the way.

Freezing the universe to the names actually available in July 2016 gives 2.10×; including the names that were available then and subsequently died gives 2.24×. Those two are not a clean isolation of survivorship, because the second list is larger and therefore differently composed — I mention it rather than dress it up.

The honest answer is that the typical Nepali listed company returned somewhere around 2.1 to 2.2 times over the decade — roughly eight per cent a year — against an index that returned 3.6 per cent.

The gap is real and it is large. It is simply half the size of what my first calculation claimed, and the missing half was made of exactly the errors I spent Chapter Three warning about.

The same pattern in the boom

Look at the 2020–21 boom through the same lens.

The index rose 169 per cent and the median company rose 200 per cent, because the index is anchored to the largest banks and the largest banks lagged. Anybody who compared his own portfolio to the index that year and felt pleased was comparing himself to the wrong thing.

And in the five years since the peak: NEPSE 0.84×, Banking 0.70×. The banks fell harder on the way down as well as rising less on the way up — which is what happens to an asset whose earnings are a leveraged function of the very liquidity cycle described in the last chapter.

The benchmark you cannot buy

There is a further and more subtle problem, and it caught me directly.

When I first built a proper backtesting engine for this market, I needed a benchmark — something to measure strategies against. I used a buy-and-hold portfolio of the investable universe, which seemed obviously right.

It returned 5.0 per cent a year over thirteen years, and every strategy I tested beat it by between five and eleven points. I was, briefly, delighted.

The tell was the trade count: seventeen fills, in a universe of a hundred and eighty-nine names. The rule required 250 sessions of history before a name could be bought, so in February 2013 only seventeen companies qualified, and it never bought any of the other one hundred and seventy-two that listed afterwards. It was not the market. It was a frozen 2013 vintage.

Replacing it with a monthly-rebalanced equal-weight book that picks up new listings moved the benchmark from 5.0 per cent to 16.8 per cent — and turned every apparent victory into a defeat.

A backtest is only ever as honest as the thing it is compared against, and an implausible trade count is worth more than a plausible return.

But there is a sting in the correction, and it is the point of this section. The 16.8 per cent benchmark is also not buyable. It rebalances monthly across a hundred and eighty-nine names, which under Nepal’s flat depository charge would cost a fortune, and it acquires new listings at prices a real investor could not have obtained.

So the honest position is that the “market return” for Nepal is not a single number, and every candidate for it is either unachievable or unrepresentative. The index understates what companies did; the equal-weight construction overstates what an investor could have captured; and the truth sits in between, closer to the equal-weight figure than to the index, and unavailable in a form you can put in a table.

What to compare yourself against

This matters practically, because Part Five is about measuring yourself honestly, and you cannot do that without a benchmark that is neither flattering nor impossible.

Not the NEPSE index. It is a bank index. If you own hydropower and microfinance, comparing yourself to it tells you almost nothing, and in a decade like the last one it will flatter you substantially.

Not the equal-weight construction. You could not have held it.

What I use, and recommend, is a deliberately conservative and buyable alternative: the return you would have achieved by putting the same money, on the same dates, into a small basket of the largest and most liquid names in the sectors you actually invest in, held without trading. It is unglamorous, it is achievable, it embeds the same liquidity cycle you are exposed to, and it will not flatter you.

And alongside it, always, the number from Chapter Nineteen: what a fixed deposit paid over the same period. That is the alternative you genuinely had, available to anyone, requiring no skill. If a decade of work has not beaten it, the work has a problem, and no benchmark that says otherwise is doing you a service.

The sub-indices are more useful than the headline

One practical note. NEPSE publishes sector sub-indices, and they are far more informative than the headline number, because they let you separate the liquidity cycle from the sector story.

When banking is falling and hydropower is rising, that is information about the composition of demand. When everything falls together — the 0.64 correlation of the last chapter — that is the tank draining, and no amount of sector analysis will help you.

Watch the spread between the sub-indices rather than the level of the main one. A narrowing spread means the market is being driven by liquidity; a widening spread means it is being driven by something specific, and something specific is where an analyst can actually earn his keep.

The Dow’s divisor — the constant you divide the sum of prices by, adjusted every time a component splits or is replaced — is now a number less than one.

Which means that a one-dollar move in any of thirty share prices moves the world’s most-watched financial indicator by more than one point, and has done for years, and essentially nobody who quotes the number is aware of it.

Indices are not measurements. They are conventions that have outlived the conditions that produced them, and the only defence is to know how yours is built.

Two Days

In which Wall Street drowns in its own paperwork and has to close on Wednesdays; a brokerage stops its customers from buying a stock and is accused of conspiracy when the real culprit is a settlement rule; and we count how many times a strategy in Nepal simply could not afford to do what it had decided to do.

Between 1967 and 1970 the American securities industry nearly destroyed itself with paper.

Trading volumes had risen faster than the back offices could process them, and every transaction still involved a physical certificate that had to be located, transferred, recorded and delivered by hand. The failures accumulated. Brokerages lost track of what they owned. At the worst of it, the New York Stock Exchange shortened its trading hours and closed entirely on Wednesdays so the clerks could catch up.

More than a hundred member firms failed or were absorbed. The crisis is why the Securities Investor Protection Corporation exists, and it is why the world eventually dematerialised share certificates into electronic entries.

I begin here because settlement is the most boring subject in this book and the one that has, more than once, decided what was possible in a market. When people say the plumbing does not matter, they are describing a period in which the plumbing happened to be working.

What settlement actually is

When you buy a share, two things must happen: you must deliver money, and somebody must deliver the share. The interval between agreeing the trade and completing that exchange is the settlement period.

It exists because the exchange, the depository and the banks need time to confirm that both sides can perform, and because for most of history the confirmation involved moving objects.

The direction of travel has been relentlessly downward. The New York Stock Exchange settled at T+5 until 1995, moved to T+3, then to T+2 in 2017, and to T+1 in May 2024. India got there first among large markets, completing its move to T+1 in January 2023 and launching an optional same-day settlement in 2024.

Nepal settles at T+2.

The reason everybody has been shortening it is counterparty risk: the longer the interval, the longer the clearing house is exposed to the possibility that somebody fails between the trade and the delivery, and the more collateral it must demand against that possibility.

Which brings us to the most misunderstood episode in recent market history.

Why Robinhood stopped the buying

In January 2021, at the peak of the GameStop episode described in Chapter Six, the American brokerage Robinhood abruptly prevented its customers from buying the stock while continuing to permit selling.

The reaction was immediate and furious. It was described as market manipulation, as a conspiracy on behalf of hedge funds, as evidence that the game was rigged. There were congressional hearings.

The actual cause was settlement arithmetic.

Under T+2, a clearing house carries two days of exposure to every broker’s unsettled trades, and it demands collateral sized to the risk. The risk is a function of volatility and concentration. GameStop’s volatility was extraordinary and Robinhood’s customer positions in it were enormously concentrated, so the deposit demanded of Robinhood by the National Securities Clearing Corporation rose overnight into the billions — an amount the firm did not have.

It could not post the collateral, so it had to stop adding to the exposure, so it restricted buying.

The most controversial trading decision of the decade was caused by a settlement convention, and almost nobody involved in the public argument understood that.

Two days does not sound like much. It is enough to close a brokerage.

What T+2 costs a Nepali investor

Now to the specific consequences here, and there are three, in ascending order of importance.

First: you cannot sell what you have just bought. Shares purchased on Sunday are not in your account until Tuesday, and cannot be sold before then. This makes intraday and next-day trading structurally impossible in Nepal, which is a considerable mercy and eliminates an entire category of ruin available elsewhere.

Second: the proceeds of a sale are not yours for two days. You sell on Sunday, and the money is available to you on Tuesday. If you wish to sell one holding in order to buy another — a rotation, which is the most common transaction an active investor makes — you must either wait two sessions out of the market, or hold enough spare cash to fund the purchase before the sale settles.

Third, and this is the one nobody thinks about: it converts a portfolio decision into a financing decision. Any strategy that involves regularly moving between holdings requires a permanent cash buffer that earns a deposit rate rather than an equity return. That buffer is not a choice. It is a tax on rotation, levied in the form of foregone return, and it is invisible because it never appears on a statement as a cost.

How often it actually binds

I can put a number on the third point, because when I built a simulation engine for this market I enforced settlement as a hard constraint rather than assuming it away — which, as far as I know, no off-the-shelf backtesting tool does, because they are all built for markets where this does not bite.

Running a trend-following strategy across the investable Nepali universe over thirteen years, the engine refused 1,728 orders for insufficient cash. Not because the strategy was leveraged or badly designed, but because it had sold something in order to buy something else and the proceeds had not arrived. A momentum strategy was refused 755 times.

Those are not rounding errors. Those are seventeen hundred occasions on which a strategy that had made a decision could not execute it, and by the time the cash arrived the price had moved.

No backtest built outside this market would have shown that, and any strategy evaluated without it is being credited with trades it could not have made.

The interaction nobody notices

There is a nastier version of this and it only appears in the situation you care most about.

Suppose the market is falling hard and you decide to sell a holding and move to cash.

You place the order. The stock is at its lower limit, so it does not fill — Chapter Twenty-Five is about that mechanism. The next day you fill, at a price ten per cent lower. And now your money is not available for two sessions.

By the time you actually have cash in hand, four sessions and perhaps twenty per cent have passed since you made the decision. The decision and its execution are separated by a gap that widens exactly when the market is moving fastest, because a fast market produces limit locks and limit locks produce delays.

This is why “I will sell if it drops below X” is a much weaker plan in Nepal than it appears. The stop is not executable at X. It is executable somewhere below X, on a day determined by the queue, with proceeds arriving two days after that.

Anyone building a risk-management rule around a price trigger in this market must model the gap or accept that the rule is decorative.

The one place it helps you

I have described T+2 entirely as a cost, and there is a case on the other side that deserves stating, because I think it is genuine.

Settlement friction is the main reason Nepali retail investors have not destroyed themselves through frequency.

Chapter Fourteen established that turnover is the most reliable way to convert a decent return into a poor one, and that the busiest quintile of American households earned six and a half points a year less than the market. Those households could trade as often as they liked, instantly, with no impediment. American brokers then removed commissions entirely and added the rest of the apparatus of engagement, and the frequency went up again.

In Nepal, a round trip takes a minimum of four sessions of your capital’s time, costs a flat depository fee on each leg, and cannot be reversed within the day. The market is structurally hostile to churn.

That hostility is a genuine protection, and it is worth recognising as one before complaining about it. A great many Nepali investors have been saved from themselves by a rule they resent.

What to actually do about it

Four things, and they are mechanical.

Keep a settlement buffer. If you intend to rotate holdings at all, hold enough cash to fund a purchase without waiting for a sale. Something of the order of one intended position size. It will earn a deposit rate and it will feel like dead money, and it is the price of being able to act on a decision in the session you make it.

Never plan a same-week round trip. Any strategy whose logic requires being out and back inside a few days does not survive contact with this market. Ask of any rule you are considering: how many sessions of my capital’s time does one cycle of this consume? If the answer is more than about four, and the cycle repeats monthly, the strategy is spending a substantial fraction of the year in transit.

Sell before you need the money, not when. Because of the two-day lag, an obligation on Thursday requires a sale by Tuesday, and a sale by Tuesday means the decision was made Sunday — and a decision made under a deadline is made at whatever price the deadline permits. This is the single most avoidable way Nepali investors sell badly, and it is entirely a calendar problem.

And treat the buffer as part of the portfolio, not as a leftover. It has a return — the deposit rate — and it has a purpose, and Chapter Fifty-One will show that in a market where the growth-optimal exposure cannot be computed, holding some cash is not a concession but a position.

The move to T+1 in the United States in 2024 was estimated to reduce clearing collateral requirements by something around forty per cent, and the reason regulators pushed it through was, explicitly, the Robinhood episode.

A retail mania in a video game retailer changed the settlement architecture of the largest capital market on earth.

Nepal will get there eventually, and when it does, the constraints in this chapter will soften and the protective friction described in the middle of it will soften with them, and it will be very interesting to see which of the two effects turns out to have been larger.

Twenty-Five Rupees

In which a cartel that had stood for a hundred and eighty-three years is abolished on a single morning; commissions in America fall to nothing and something worse takes their place; and a flat charge of twenty-five rupees turns out to determine how many companies a Nepali investor is permitted to own.

The first of May 1975 is known on Wall Street as Mayday, and it was the day the agreement signed under the buttonwood tree finally died.

Until that morning, American brokerage commissions were fixed by the exchange. Every firm charged the same, competition on price was prohibited, and the arrangement — the one those twenty-four men wrote down in 1792 — had survived a civil war, two world wars, the Depression, and the arrival of the computer.

The Securities and Exchange Commission abolished it. Rates were now negotiable.

Institutional commissions collapsed almost immediately, by something like half. A number of old firms did not survive the year. And a man named Charles Schwab, who had been running a newsletter, opened a brokerage that did no research, gave no advice, and simply executed trades cheaply — which had been illegal the week before.

The story continued in one direction for the next forty-five years. Discount broking became online broking; online broking became Robinhood; and in October 2019 the major American brokerages went to zero commission within a fortnight of each other, because one of them moved and the rest could not afford not to.

Now the part that matters, and it is the lesson I want you to carry into the Nepali schedule.

The commission did not disappear. It changed form.

American retail orders are now largely sold to wholesale market makers, who pay the broker for the right to execute them — payment for order flow — and who make their money on the spread between what they buy and sell at. The customer sees a commission of zero and pays in a currency he cannot observe.

Which is the first principle of transaction costs and the reason this chapter exists: the fee you can see is rarely the whole fee, and the fee you cannot see is not smaller for being invisible.

Two kinds of fee, and why the difference is everything

There are only two ways to charge for a transaction.

A percentage fee is scale-neutral. Half a per cent on a thousand rupees and half a per cent on a crore are the same proportion. It costs a small investor exactly what it costs a large one, measured in the only unit that matters.

A flat fee is regressive. Twenty-five rupees on a crore is nothing. Twenty-five rupees on a thousand-rupee trade is two and a half per cent, before anything else.

Almost every market’s visible charges are percentages, which is why almost nobody thinks about this. Britain levies stamp duty at half a per cent on share purchases; India levies a securities transaction tax; both are proportional, both reduce turnover somewhat, and neither changes what a small investor can do.

Nepal has a flat fee sitting in the middle of its schedule, and it changes what you can do.

The Nepali schedule

Every trade, on each side, pays approximately the following. The rates are set by regulation and have been adjusted before; verify the figures in force rather than trusting a book.

The depository charge is the structurally important one, and I want to derive its consequence rather than assert it.

At the bottom commission tier the variable charges come to about 0.375 per cent of the trade on one side. The flat fee is twenty-five rupees. Those two are equal when

25 ÷ 0.00375 = NPR 6,667

Below a trade value of six thousand six hundred and sixty-seven rupees, more than half of what you pay is a fixed toll that does not care how small your trade is.

This number is not estimated from data and it cannot be wrong. It is arithmetic performed on a published fee schedule, which makes it one of the very few things in this book that will still be true if everything else in it is mistaken — until the schedule changes, at which point you redo the division.

What a round trip actually costs

Here is the complete round-trip cost, buying and selling, at various trade sizes, excluding capital gains tax.

A thousand-rupee trade costs seven per cent to enter and exit. The stock must rise seven per cent before you have broken even on the mechanics — and then capital gains tax applies to whatever is left.

At a hundred thousand rupees the round trip is 0.80 per cent, and by a million it has converged to 0.76 and stops improving, because at that point the flat fee is irrelevant and you are paying the commission.

So the entire cost curve in Nepal is decided between one thousand and twenty-five thousand rupees. Above that, size buys you almost nothing. Below it, size costs you everything.

The result that decides your portfolio

Now the consequence nobody works out, and it is the most useful thing in this chapter.

If a trade must be worth at least NPR 6,667 to be economic, and your positions are worth (book ÷ number of names), then a position must drift by 6,667 ÷ position size before rebalancing it is even worth doing.

Read the top-left cell.

An investor with two lakh rupees spread across twenty companies cannot rebalance at all. A position would have to move sixty-seven per cent relative to the rest of the book before an adjusting trade covered its own costs. His portfolio is not a portfolio in any managed sense; it is twenty small piles that will drift wherever they drift, and the diversification he believes he has purchased has cost him the ability to maintain it.

Move down the same column and the constraint dissolves: at fifty lakh in twenty names the band is 2.7 per cent, which is a normal, workable tolerance.

And move across the top row: the same two-lakh investor holding five names has a 16.7 per cent band, which is wide but survivable.

There is a second cost running alongside, which is the pure fee drag from holding many names at all.

That is the depository fee alone, assuming each position is touched twice a year. A per cent a year, on a small book, for the privilege of owning twenty things.

The instruction that follows

Your book size determines how many companies you may own. Not your conviction, not your appetite for diversification, not what a textbook written in another country recommends.

As a working rule derived from the tables above:

And notice that this collides with nothing. Chapter Fifteen argued that you can only genuinely understand about ten Nepali companies. Chapter Twenty-One measured that beyond about ten holdings, the correlation floor means additional names add essentially no diversification in a bad year.

Three completely independent lines of reasoning — what you can read, what diversification actually delivers, and what the fee schedule permits — converge on the same answer, which is a portfolio of roughly eight to fifteen names.

I did not arrange that. I noticed it while assembling this chapter, and it is the strongest argument in the book for a conclusion I had previously held for weaker reasons.

The fee that is not on the schedule

One more, and it is the American lesson arriving in Nepali form.

Everything above is the visible cost. There is an invisible one: the spread between what you must pay to buy immediately and what you would receive to sell immediately, plus whatever your own order does to the price while it executes.

Nepal has no payment for order flow and no wholesale market makers, so the mechanism is different from the American one. But the effect is present. In a name trading a few million rupees a day, an order of any size walks up the book, and the difference between the price you saw and the average price you got is a real cost that appears on no statement.

Chapter Fifty-Two deals with execution properly. For now the point is only this: when you compute whether a switch is worth making, add something for the spread, and add more for a thinly traded name. If the decision is marginal after the visible charges, it is negative after the invisible ones.

And the reason the flat fee is defensible

I have spent a chapter describing the depository charge as a regressive imposition, so let me finish honestly.

It is regressive because the cost of providing the service is largely fixed. Recording a transfer of shares costs the depository roughly the same whether the transfer is worth a thousand rupees or a crore. A flat fee reflects a real cost structure. Charging a percentage would mean large investors subsidising small ones, which is a policy choice rather than an obvious improvement.

What is not defensible is investors failing to compute what it does to them. The fee is published. The arithmetic takes four minutes. And the conclusion — that a small book must hold few names and trade rarely — is not a hardship. It is the same conclusion that reading, correlation and Chapter Fourteen’s tax deferral all arrive at independently.

The schedule is not your enemy. It is a rather good adviser that charges nothing and is never wrong, and almost nobody reads it.

Charles Schwab’s firm, founded in the wreckage of Mayday to charge less than everybody else, announced in November 2019 — one month after commissions went to zero — that it would buy TD Ameritrade for twenty-six billion dollars. The deal closed in October 2020, merging the two largest discount brokerages in the country.

The business built on undercutting a fixed price ended up buying its largest rival at the moment there was no price left to undercut, and the industry’s revenue moved somewhere the customer could not see.

Every fee that is abolished reappears. The only question is whether you can find it.

The Circuit

In which China introduces a safety mechanism on the first of January and abolishes it on the seventh; the American market falls a thousand points in five minutes and recovers; and we measure three hundred and fifty thousand Nepali trading days to discover that this market locks upward six times more often than it locks downward, and that the stock you most want to buy runs away from you the following morning.

On the fourth of January 2016, China’s new circuit breaker was activated for the first time. The CSI 300 fell five per cent, trading halted for fifteen minutes, and when it resumed the index fell to seven per cent and the market closed for the day.

On the seventh of January it happened again, but faster. The market opened, fell five per cent, halted, reopened, fell to seven, and closed permanently for the session. Total trading time: twenty-nine minutes. It remains the shortest trading day in Chinese history.

That evening the mechanism was suspended. It had existed for four trading days.

The failure is now the standard teaching example, and the diagnosis has a name. It is called the magnet effect, and the theory had been articulated years earlier: a price limit does not merely stop a move, it attracts one. As the price approaches the threshold, anybody who wants to sell realises that if the halt triggers he will be unable to, and so he sells now, at whatever price is available, in order to get out before the door closes. The rush to beat the halt is what causes the halt.

A device installed to slow a decline had accelerated it, twice, in a week.

Why every market has them anyway

The counter-case is real, and the American experience is the reason.

After Black Monday in 1987 — twenty-two and a half per cent in a session, described in Chapter Ten — the United States installed market-wide circuit breakers, on the reasoning that a market falling that fast is not processing information but transmitting panic, and that a pause allows both.

They were triggered for the first time in October 1997, and then not again for twenty-three years, and then four times in eleven days in March 2020.

Separately, the flash crash of the sixth of May 2010 — when the Dow fell about nine per cent and recovered most of it within minutes, and individual shares briefly traded at a cent — produced single-stock limits, on the reasoning that a stock printing a price that bears no relation to anything is an error rather than a market.

So the honest summary is that circuit breakers protect against a specific pathology (disorderly collapse, erroneous prices, cascading margin liquidation) and create a different one (the magnet effect, and the inability to transact when you most need to). Whether they are net positive depends on which pathology your market is prone to.

Nepal’s version is unusually strong, and it produces consequences that I have never seen written down.

Nepal’s rules

Two mechanisms operate.

An individual stock may move ten per cent from its previous close and then it stops. This was widened to fifteen per cent in April 2026, in the same window as the change in the trading week — so essentially all of the history that matters was governed by the ten per cent rule, and the fifteen per cent regime is young enough that nothing can yet be concluded about it.

The index itself halts at five per cent and again at eight.

The individual limit is the one that shapes behaviour, and its critical property is this: a stock closing at its upper limit has buyers and no sellers. That is the definition. There is a queue of orders at the limit price and nothing coming the other way. An order placed into that queue does not execute; it waits, behind everybody else who had the same idea earlier in the morning.

What I measured

I took every Nepali stock in the exchange’s sector universe from January 2016 to the widening of the circuit in April 2026 — 352,236 daily bars — and counted every close at the limit, in both directions, and then looked at what happened the next day.

Read that ratio. This market locks upward six times as often as it locks downward.

I did not expect a six-fold asymmetry and spent some time checking it before believing it. The explanation, once you see it, is the whole of Part One arriving in the order book.

There is no short selling. Buying pressure can come from anybody with money — several hundred thousand people, coordinating loosely through the group chats of Chapter Six, concentrated onto one thin float. Selling pressure can only come from the comparatively small number of people who already own the stock.

And those owners do not sell. Chapter Eleven established the disposition effect with Odean’s data and Chapter Ten traced its Nepali form: the holder of a falling share resolves to sell when it comes back to his price, and therefore does not sell at all.

So demand concentrates and supply disperses. The market has an accelerator available to everybody and a brake available only to the people who are least inclined to press it.

And then the cruel part

Here is what happens on the day after a limit lock.

Take the top row slowly, because it is the most hostile fact in this book for anybody who trades on price strength.

A stock that locks limit-up rises, on average, another 3.17 per cent the following session — and has close to a one-in-three chance of locking limit-up again.

Now recall what a limit-up close means: you could not buy it. The queue was ahead of you. So the market’s behaviour is:

The stock you most want to own is the one you cannot have, and tomorrow it will be three per cent further away, and one time in three it will be ten per cent further away and still unbuyable.

Chapter Five argued on structural grounds that breakout strategies are close to inexpressible in this market. This is the measurement behind that argument, and it is worse than the argument was. It is not merely that you cannot buy strength. It is that strength persists and continues to be unbuyable, so the gap between the signal and your entry widens for as long as the signal is valid.

By the time a limit-up name is available to you, it is available because the demand has been satisfied — which is to say, at the moment the thing that made it attractive has finished happening.

The good news, which nobody believes

Now the bottom row, and it corrects a fear that is widespread and largely wrong.

Nepali investors talk about limit-down as a trap: the market falls, your stock locks, you cannot get out, and it locks again the next day and the day after until you are destroyed.

The data does not support it.

Limit-down closes are rare — 0.18 per cent of bars, less than one day in five hundred. After one, the average next day is −0.09 per cent, which is nothing, and the median is −0.18 per cent. Only 11.9 per cent lock down again.

So the asymmetry runs in your favour on the exit side. Locks upward cluster and persist; locks downward do not. Whatever else is true about Nepali market structure, the multi-day trapdoor that people fear is not a feature of the historical record.

I want to add the necessary caution. This is measured across a period containing the 2021–22 decline, which was the second-largest in the market’s history, so it is not a sample of quiet years only. But it does not contain 2008–11, and a genuinely systemic event — a bank failure, a currency crisis — could produce a sequence with no precedent in this sample. The circuit does not prevent that. It only ensures it would take several weeks rather than a single afternoon, which is Chapter Ten’s argument exactly.

What the widening changes

The move from ten to fifteen per cent in April 2026 is too recent to evaluate, and I will not pretend otherwise. But the direction of the effects is predictable from the mechanism.

A wider band means fewer locks, which means more days on which a determined buyer can actually transact, which reduces the queueing problem above. It also means a larger possible single-day loss, and a larger overnight gap for anybody using price triggers.

The magnet effect, if it operates here, should weaken: a threshold fifteen per cent away exerts less pull than one ten per cent away, because the urgency to beat it is lower.

And the six-fold up-down asymmetry should persist, because it is caused by the absence of short selling and by the disposition effect, and neither of those has changed.

The practical rules

Never chase a limit-up. The data says it will rise another three per cent and may well lock again — which sounds like a reason to chase and is the opposite. You are being invited to buy at the end of a queue whose front you cannot see, in a name whose float is thin enough to lock, on demand that is by definition transient. The times you succeed in getting filled are disproportionately the times the demand had already exhausted itself.

Use limit orders, always. A market order in a Nepali stock during a fast session is an instruction to accept whatever price the queue delivers, in a market where the queue can be several times the day’s normal volume. There is no execution urgency in a long-horizon strategy that justifies it.

Do not build a plan around a price trigger. Chapter Twenty-Three showed that a stop is not executable at the level you set it; this chapter shows why. Combine the two and any rule of the form “I will sell if it falls below X” should be understood as “I will sell somewhere below X, on a day I do not choose, with the money arriving two sessions later.”

And if you want to own something that is running, wait for the pause. A limit-up sequence ends, and it ends at a price that is genuinely available, which is the first price at which anybody was willing to sell you the thing. That price is a fact. The sequence before it was an auction you were not admitted to.

There is one more thing worth saying about the Chinese episode, because it is the part everybody forgets.

The mechanism was not badly designed by fools. It was designed by capable people who had studied the American system and adapted it — and their error was a single parameter. The American thresholds are seven, thirteen and twenty per cent, and the Chinese ones were five and seven, in a market whose ordinary daily volatility was far higher.

They had set the halt inside the range of normal behaviour. So it triggered on an ordinary bad day, and the knowledge that it would trigger made the day worse.

A safety device calibrated to the wrong distribution is not a safety device. It is a schedule of panics, and the market found it in four days.

The 365-Day Line

In which several rich countries decide not to tax capital gains at all; America creates the strongest incentive in the history of taxation to hold an asset until you die; and a small piece of algebra reveals that the Nepali tax boundary everybody arranges their selling around is worth, at its absolute theoretical maximum, 2.63 per cent — and is a coin flip.

Singapore does not tax capital gains. Neither does Hong Kong. New Zealand largely does not. Switzerland does not tax them for private investors, only for people it deems to be trading professionally.

These are not tax havens in the pejorative sense; they are ordinary rich countries that concluded, for various reasons, that taxing the realisation of a gain does more harm than the revenue justifies.

The harm has a name. It is called the lock-in effect, and it was formalised in the American public finance literature around 1980: a tax that falls only when you sell converts every sale into a taxable event, which means investors hold assets they would otherwise dispose of, purely to postpone the liability. Portfolios become misallocated. Capital fails to move from worse uses to better ones. And the distortion is largest exactly where the gain is largest — that is, in the positions that have grown most, which are often the ones most in need of trimming.

The extreme case is the United States, which combines a capital gains tax with a provision called the step-up in basis at death. When an American dies, the cost basis of his assets resets to their market value, and the accumulated gain of a lifetime escapes tax entirely.

Consider what that instructs a rational person to do. It says: never sell anything, ever, under any circumstances, until you are dead. It is the most powerful lock-in incentive any tax system has ever created, and it is why wealthy Americans borrow against portfolios rather than selling them.

India offers the cleanest natural experiment. Long-term gains on listed equity were entirely exempt from tax in India until 2018, when a ten per cent levy was reintroduced — subsequently revised — after a fourteen-year holiday. The reintroduction was announced in the budget and the market’s reaction was immediate and visible, and Indian investors have been arranging their holding periods around the twelve-month boundary ever since.

Which brings us to what Nepalis do around theirs.

Nepal’s step

An individual pays capital gains tax on listed shares at 7.5 per cent if the holding period is under three hundred and sixty-five days, and 5 per cent if it is three hundred and sixty-five days or more. Institutions are treated differently, and the rates have been changed before; verify what is in force.

The step is two and a half percentage points, and it falls on the gain rather than on the proceeds.

Every Nepali investor knows about it, and a great many of them manage their selling around it. The reasoning is obvious: if I am at day three hundred and forty with a large gain, surely I should wait twenty-five days and save the tax.

This is, in my experience, the single most confidently held piece of practical tax folklore in the Nepali market.

I worked out what it is actually worth, and the answer surprised me enough that I checked the derivation three times.

The algebra

Set it up properly. You hold a position with cost basis B, currently worth V. If you sell today you pay tax at 7.5 per cent on the gain; if you wait past day 365 you pay 5 per cent. Selling expenses e are deductible, and the flat depository fee appears identically in both branches and therefore cancels.

Net proceeds are (1−t)(V−e) + tB, where t is whichever rate applies.

Equate the two branches and solve for how far the price may fall before waiting stops being worthwhile. Writing g for the embedded gain and c for the round-trip cost rate, the breakeven decline is

d\*(g) = (0.025 ÷ 0.95) × [ 1 ÷ ((1−c)(1+g)) − 1 ]

Now look at what is and is not in that expression.

There is no *t* in it. No days. No holding period. The number of days remaining until the boundary does not appear at all.

This is the first surprise, and once seen it is obvious: the size of the tax saving does not depend on how long you must wait. Only the probability of the price falling below the boundary depends on the wait. The threshold itself is a function of the embedded gain alone.

The number

That last row is the finding.

No position, however enormous its gain, can justify riding out a decline of more than 2.63 per cent in order to capture the tax saving.

The reason is that the saving is capped. It is two and a half per cent of the gain, and the price risk you are accepting applies to the whole position value. As the gain grows, the saving grows in absolute terms — but so does the position, and the two converge to a hard ceiling of 2.63 per cent of the position’s value. There is no gain large enough to escape it, because the ratio is bounded.

Now the second half, and it is the part that closes the question.

I measured, across the full price panel, the probability that a Nepali stock sits below that boundary on day 366 — that is, the chance the decline while you waited exceeded the saving.

Forty-seven to fifty-two per cent, at every level of embedded gain.

A coin flip. You are accepting a fifty-fifty chance of losing more than you save, to capture at most 2.63 per cent, and typically far less than that — under one per cent for a position sitting on a fifty per cent gain.

And when I measured the value of the entire capital-gains-deferral lever across a thirteen-year simulation — deferring every eligible sale across the boundary whenever the arithmetic favoured it — it contributed +0.01 per cent to the annual compound return.

One hundredth of a percentage point. It is not a strategy. It is a rounding error that a great many people have organised their portfolios around.

Correcting myself

In Chapter Fourteen I wrote that crossing the 365-day line is free money for a sale you already intend to make, and that where a decision is genuinely marginal the calendar has a vote worth about two and a half per cent of the gain.

That was too generous, and the algebra above is the correction.

The calendar’s vote is worth at most 2.63 per cent of the position and usually much less, and it is exposed to a coin flip on price. The honest statement is:

If you are indifferent between selling on day 300 and day 366, wait — it costs nothing and saves a little. If you have any reason at all to sell now, sell now, and do not let the boundary detain you. The tax tail is far too small to wag the investment dog, and I overstated it.

I leave the original sentence in Chapter Fourteen rather than quietly editing it, because a book that silently repairs its own errors is doing to you exactly what Chapter One accused Ramesh dai of doing to himself.

What the tax genuinely does change

None of the above weakens the deferral argument, which is different and much larger, and which was Chapter Fourteen’s real point.

The distinction is between postponing a sale by twenty-five days and not selling for twenty years.

The first is worth a rounding error. The second was worth two hundred and thirteen thousand rupees on a one-lakh stake over twenty years, in the table in Chapter Fourteen, because a man who never sells compounds on money the state has a claim to but has not collected, and settles once, at the lower rate, at the end.

So the tax system’s genuine message to a Nepali investor is not manage your holding periods. It is have fewer holding periods. Those sound similar and they are opposite: the first is an optimisation performed at the moment of sale, the second is a decision made at the moment of purchase.

The other tax, which people forget

Dividends are taxed too, and the rate for individuals has been five per cent, withheld at source, on both cash dividends and bonus shares. Again: verify the current figure, because this is exactly the sort of number that moves in a budget.

Two things follow that matter more than they appear.

Bonus shares are taxed. A bonus issue is not a gift; it is a distribution, and it is taxed as one, even though you receive no cash with which to pay. Nepali companies — particularly banks — have historically distributed heavily in bonus shares, and a shareholder receiving a large bonus issue may find himself with a tax obligation and no money. This is a real cash-flow problem for retired investors living on their portfolios, and Chapter Fifty-Six deals with it.

And a five per cent tax on dividends alongside a five per cent tax on long-term gains means the tax system is, unusually, close to neutral between the two. In many countries dividends are taxed considerably more heavily than gains, which pushes companies toward buybacks and investors toward growth. Nepal does not create that distortion, which is a genuine merit of the system and one nobody mentions.

The practical residue

Do not manage your selling around the boundary. The arithmetic caps the benefit at 2.63 per cent under the most favourable possible assumptions and it is a coin flip.

Do manage your buying around the intention to hold. The deferral value is real and large, and it is captured at purchase — by buying something you expect to own for a decade — not at sale.

Set aside cash for the tax on bonus shares before it arrives, because it will arrive without any cash attached to it.

And when someone tells you that a tax consideration justifies holding a position, ask what the consideration is worth as a percentage of the position. In nearly every case that computation has never been done, and in nearly every case the answer is smaller than the price risk being accepted to obtain it.

That is not a rule about tax. It is a rule about any argument in which a small certainty is being used to justify a large exposure — and in Chapter Thirteen we saw where that reasoning ends when it is allowed to compound.

The American step-up at death has beaten back every attempt to remove it for fifty years — Congress actually enacted its replacement in 1976, deferred it, then repealed the repeal in 1980 before it ever took effect, and a modified version applied for a single year in 2010 and was reversed — because the constituency that benefits from it is precisely the constituency that funds political campaigns.

Its effect is that the most tax-efficient possible investment strategy in the largest capital market on earth is to buy assets, borrow against them for consumption, and never sell anything.

Which is, purely by accident, also close to the best strategy on the merits — and it is worth noticing that when a tax code and a sound investment principle agree, it is almost always because the tax code was written by people who already owned the assets.

The Illusion of Free Shares

In which the first event study in the history of finance examines stock splits and finds nothing; a bonus issue turns out to be a stock split that charges you tax; and a right share is revealed to be a demand for money dressed as a privilege, with a penalty of up to forty-four per cent for declining it.

In 1969 four economists — Eugene Fama, Lawrence Fisher, Michael Jensen and Richard Roll — published a paper that invented a method as much as it produced a result. They wanted to know what happens to a share price around a stock split, and to find out they built the first event study: line up nine hundred and forty splits by the month each one took effect, average the returns around that month, and see what the market does.

The finding was that prices rose in the months before a split announcement and did essentially nothing afterwards. The market had already worked out that a company about to split its stock was a company that had been doing well, and by the time the split occurred there was nothing left to react to.

Which is exactly what should happen, because a stock split does nothing.

If you own one share worth a thousand rupees and the company splits it into ten, you own ten shares worth a hundred rupees. You have precisely what you had before. No value has been created, no cash has moved, and the company’s earnings, assets and prospects are identical. It is the most famous non-event in corporate finance and it has been demonstrated a hundred times since.

Nepal does not have stock splits. It has something that looks different, is economically the same, and — this is the part nobody computes — costs you money.

A bonus share is a split with a tax attached

When a Nepali company declares a bonus issue, it capitalises reserves and issues new shares to existing shareholders in proportion to their holdings. A hundred per cent bonus doubles your share count.

The price adjusts. It must: the company is worth what it was worth, and there are now twice as many claims on it. A share at 500 before a 1:1 bonus becomes two shares at 250 afterwards, and your position value is unchanged at 500.

Nothing has happened.

Except that in Nepal a bonus issue is treated as a distribution, and distributions are taxed. Individuals pay five per cent, withheld at source, and the bonus is valued at par — a hundred rupees a share — for that purpose. So on a hundred per cent bonus you pay five rupees of tax for every original share you held.

You received nothing. You paid tax on it.

Now the observation I have not seen made anywhere, which falls straight out of the arithmetic: the cost of a bonus issue, as a percentage of your wealth, depends entirely on the share price — because the tax is levied on par value and your wealth is measured at market value.

A hundred per cent bonus in a share trading near par costs you five per cent of your position. The identical bonus in a share trading at twelve hundred costs you four-tenths of one per cent — twelve times less.

Which means that the same corporate action is twelve times more expensive to the holder of a cheap share than to the holder of an expensive one, and neither of them receives anything for it.

And notice which Nepali companies trade near par: the smaller banks, the finance companies, the ones that have already issued heavy bonuses for years and driven their own price down toward it. The bonus mechanism grinds hardest on the shareholders who have already been ground the most.

I want to be careful not to overstate this. Five per cent of par on a hundred per cent bonus is not a catastrophe, and Nepali companies do not typically issue hundred per cent bonuses annually. But over a decade of steady bonus distributions in a low-priced share it becomes a real and entirely invisible leak — invisible because your share count went up, and a rising share count feels like something good is happening.

It is the purest illustration in this market of a rule worth generalising: any corporate action that increases your share count without increasing your wealth is either neutral or negative, and you must find out which.

Why companies do it anyway

If a bonus creates nothing and costs the shareholder something, why is Nepal so fond of them?

Three reasons, and only one is about you.

Regulatory capital. This is the real one. A bonus issue converts retained earnings into paid-up capital, and Nepali financial institutions operate under minimum paid-up capital requirements set by their regulators. When Nepal Rastra Bank raised the required capital for commercial banks, bonus issues were one of the principal mechanisms by which banks got there. The bonus was not a gift to shareholders; it was a compliance instrument that happened to pass through their accounts.

Signalling. A company that can afford to capitalise reserves is telling you it has reserves. This is real information, though it is available more directly by reading the balance sheet.

And price psychology. A share at twelve hundred rupees looks expensive to a retail buyer and a share at three hundred looks cheap, and this is entirely irrational and entirely effective. Lowering the nominal price widens the buyer base. Every market does this; Nepal simply does it more.

Note the asymmetry in that list. Two of the three reasons are about the company’s needs, and the third is about exploiting a cognitive error in the buyer. None of them is a reason for you to prefer a company that issues bonuses, and the widespread Nepali belief that a high bonus is a mark of quality is a confusion between a distribution policy and a business.

The right share is a bill

Now the more consequential one.

A rights issue offers existing shareholders the opportunity to buy new shares, usually at par — a hundred rupees — in some ratio to their existing holding. In Nepal it is the principal mechanism by which hydropower companies fund construction, and Chapter Twelve described what it does to a shareholder’s arithmetic over time.

Here is the mechanics on a single issue.

Read the fourth column. If you subscribe, you gain nothing. You put in more money and you own more shares and your total position is worth exactly what you put in. A rights issue at par is not a discount and not an opportunity; it is a capital call.

Read the fifth column. If you decline, you lose — badly. A 1:1 rights issue in a share trading at 800 costs a non-subscriber 43.8 per cent of his position, because the price falls to the blended level whether or not he participated.

That is the ratchet from Chapter Thirteen, seen at the moment it engages. You are not being offered an investment. You are being offered a choice between putting in more money and accepting an immediate loss, and the loss is engineered to be larger than the cheque.

In several developed markets this problem is solved: rights are renounceable and tradeable. A shareholder who does not wish to subscribe sells his rights to somebody who does, and the proceeds compensate him for the dilution. The British market has done this for a century. Where rights cannot be sold, the non-subscriber’s loss is a pure transfer to the subscribers.

Nepal’s treatment of unsubscribed rights has changed over time and involves an auction process; check the current mechanism and, more importantly, check what a specific company is actually doing before the issue closes, because the difference between a tradeable right and a lapsing one is the difference between an inconvenience and a forty-four per cent haircut.

The practical rule for a right issue

Chapter Thirteen gave it and I will restate it here because this is where it bites.

A right issue is a new investment decision and must be evaluated as one, from a blank page.

Not “should I avoid the dilution” — that framing guarantees you subscribe, every time, forever, and it is how a man ends up with four times his intended exposure to one construction project in one valley.

The correct question is Chapter Eleven’s replacement question, applied to the new money: if I held this cash today, and this company were offering shares at a hundred rupees, would I buy them?

If yes, subscribe, and understand that you are increasing your position.

If no, then decline and accept the loss — and recognise that the loss was already incurred the moment the issue was announced. The dilution is a fact about the company now. Subscribing does not undo it; it merely adds capital to an enterprise you have just decided you would not buy into.

And if the answer keeps coming back “no” for a company you own, that is not a tax problem or a timing problem. That is your own analysis telling you something, and the right issue was merely the occasion on which you finally listened.

Further offerings and auctions

Two more mechanisms, more briefly.

A further public offering sells new shares to the general public rather than to existing holders, usually at a premium to par determined by a valuation exercise. It dilutes existing shareholders without offering them the chance to maintain their proportion — but at a price closer to fair value, so the dilution is economic rather than mechanical. Whether it harms you depends entirely on whether the issue price is above or below what the shares are worth, which is a valuation question and therefore Part Four’s.

Auctions distribute shares that were not taken up — unsubscribed rights, unclaimed allotments, promoter shares being converted. They are one of the few places in this market where supply arrives without accompanying enthusiasm, and prices in them are sometimes genuinely attractive. They are also small, sporadic and poorly publicised, which is precisely why they occasionally offer something.

The data trap

One technical warning that belongs here rather than in Chapter Thirty-One, because it follows directly from everything above.

Every corporate action described in this chapter changes the share count and therefore the price, with no change in value. Which means that any historical price series must be adjusted for them, or it is nonsense.

A share that went from 500 to 250 because of a 1:1 bonus did not fall fifty per cent. But an unadjusted price series says it did, and every calculation you perform on that series — returns, volatility, moving averages, correlations, the value of your own portfolio over time — will be wrong, and wrong in a direction that makes every bonus-paying company look like a disaster.

The panel I have used throughout this book is on an adjusted basis, and it refuses to mix adjusted and unadjusted series, because a silent mix of the two is invisible in every summary statistic and corrupts everything downstream.

If you are looking at a Nepali price chart and it shows a sudden fifty per cent drop in a healthy company, look up the corporate action calendar before you conclude anything. That drop is usually the shareholders receiving something.

Fama and his colleagues, having established that stock splits contain no information, noted that the practice nevertheless persisted and showed no sign of stopping.

More than fifty years later it has not stopped, in any market, anywhere. Companies continue to split shares, shareholders continue to be pleased about it, and the financial press continues to report it as news.

The most robust finding in the history of empirical finance has had no effect whatsoever on the behaviour it describes, which is worth remembering the next time you believe that demonstrating something will change it.

The Calendar

In which an anomaly discovered in 1976 politely dies once everybody knows about it; thirty-six countries out of thirty-seven appear to have the same seasonal quirk; and the Nepali fiscal year, which ends in the middle of July, turns out to leave a mark on the index that survives being tested against sixteen years nobody had looked at.

In 1976 Michael Rozeff and William Kinney published a study of American stock returns by month of the year going back to 1904, and found something that ought not to have been there.

January was different. Returns in January were substantially higher than in other months, and the effect was concentrated in small companies. The proposed explanation was tax: American investors sell losing positions in December to realise losses against their tax bill, depressing prices in exactly the securities that had done worst, and then buy back in January.

The January effect became the most famous seasonal anomaly in finance. It was written up in textbooks. Funds were designed around it.

And it substantially faded. Studies from the 1990s onward found it much weaker in American large caps, and in some later periods reversed. The most persuasive explanation is the simplest one: it was published, people traded it, and the trade consumed the opportunity.

That story — an anomaly documented and then eroded by the documentation — is the normal life cycle of these things, and it is worth holding in mind before I show you one from Nepal.

The other famous one is sell in May and go away, an English market adage of uncertain antiquity, which Sven Bouman and Ben Jacobsen tested in 2002 across thirty-seven countries and found present in thirty-six of them. That result has been argued about ever since, and the argument is instructive, because the objection is always the same and it is always right: if you test twelve months, you have twelve chances to find something, and one of them will look significant by accident.

Any calendar claim must survive that objection before it deserves a sentence.

Nepal runs on a different year

Before the anomaly, the plumbing — because Nepal’s calendar is genuinely different and most of its consequences are ignored.

The Nepali fiscal year runs from Shrawan 1 to Asar end, which in the Gregorian calendar is roughly mid-July to mid-July. Companies report on that year. The government budgets on it. Everything financial in the country is anchored to a boundary that falls in the middle of July.

The quarters end at approximately 17 October, 14 January, 13 April and 16 July — the ends of Ashoj, Poush, Chaitra and Asar. These dates drift by a day or two with the lunar calendar, which is itself a small hazard for anybody automating anything.

And the disclosure obligations attach to those dates. Quarterly results must be published within thirty days of the quarter’s end under Schedule 14 of the Securities Registration and Issuance Regulations. Audited annual results get five months under Schedule 15.

So the information calendar for a Nepali investor looks like this:

Note the gap at the bottom. The audited accounts for a year that ended in July may legally appear in December — five months in which the company knows and you do not, and which Chapter Six identified as the vacuum that rumour fills.

The three seasonal flows

Three things move money around the Nepali calendar with enough force to matter.

The Asar-end government spending flood. The state collects revenue steadily and spends its capital budget very unevenly, with a pronounced rush in the final weeks of the fiscal year as ministries race to avoid surrendering unspent allocations. When that money is finally released it lands in contractors’ accounts, which are bank deposits, which — per Chapter Twenty-One — is the tank filling.

The Dashain and Tihar cash drain. In the weeks around the festivals an enormous quantity of currency is withdrawn as physical notes and circulates outside the banking system. Deposits genuinely shrink. It returns afterwards, but for a period the system is tighter.

And the quarterly reporting rhythm, which concentrates whatever genuine company news exists into four narrow windows a year, each about a month after a quarter closes.

The July effect

Now the anomaly, and I am going to present it the way Part One demands rather than the way it would be presented on a panel.

I measured the NEPSE index’s return in each calendar month, using one realised return per calendar month per year — not pooled daily returns, which inflate the apparent evidence roughly twentyfold by treating twenty-odd correlated days as independent observations. Then I corrected for the fact that twelve months were tested.

Applying a false-discovery-rate correction across all twelve months, exactly one survives: July. It is also the only month whose sign holds across all four blind sub-periods of the sample.

The out-of-sample test is the part that carries the weight, and it cost nothing to run. Every quantitative result in my research programme was developed on data from 2013 onward, while the index series itself begins in 1997 — so sixteen years sat entirely untouched by any decision I had made. Running July against them was a genuine test rather than a re-description.

The effect decays by two thirds out of sample, from +8.64 to +3.00 per cent, which is the normal fate of a discovered mean and exactly what you should expect. But it keeps its sign and its significance.

And there is a mechanism, which matters enormously, because a seasonal pattern without a mechanism is a coincidence with a t-statistic. Asar ends around the sixteenth of July. Splitting the month across the pooled twenty-eight years:

• 1–16 July: +3.45% (t = 3.63)
• 17–31 July: +2.12% (t = 2.31)

Both halves positive, the fiscal-year-end half slightly stronger. Which is what you would predict if the cause is the government spending flood plus fiscal-year-end positioning, and not what you would predict if it were noise.

And why you should not trade it

Here is where I part company with how such a finding would ordinarily be used.

The actionable reading is not “buy in July.”

A July-only rule buys once a year and sells once a year. Each round trip pays roughly 0.75 per cent in variable charges, plus the flat depository fee, plus 7.5 per cent capital gains tax on the gain, because a position held for a month is a short-term holding. Chapter Fourteen established that turnover is the single most reliable way to convert a decent return into a poor one, and Chapter Twenty-Six established that the short-term rate is the punitive one.

So a rule that captures a three per cent seasonal effect by trading twice a year gives most of it back to the exchange and the tax authority, and disturbs a holding it would otherwise have left alone.

The defensible instruction is the passive one:

Do not be out of the market in July.

That costs an already-invested holder nothing at all. It means: if you are planning to sell something, do not schedule it for late Asar. If you are deploying new money, do not sit in cash through the fiscal year end waiting for clarity. If you rebalance annually, do it in a different month.

It is an instruction about timing decisions you were already going to make, not about generating new ones — which is, I think, the correct use of nearly every seasonal finding ever published, and almost never the use they are put to.

The week that changed four times

One more calendar hazard, and it is the one that quietly corrupts data.

I counted every session in the index series by weekday. The Nepali trading week has changed four times.

The most recent change, in April 2026, arrived in the same window as the widening of the price limit from ten to fifteen per cent — two structural changes at once, which will make the next few years unusually difficult to compare with the past.

The consequence for anybody computing anything: never generate a Nepali trading calendar from a weekday rule. It will be wrong somewhere in your sample and it will be wrong silently. Sessions must be observed from the data.

And the market has closed for extended periods more than once: thirty-three days and then thirty-one in 1998, thirty-one after the Gorkha earthquake in April 2015, and fifty-one days followed by another forty-seven in 2020. Any indicator with a fixed lookback in sessions was reaching back a different distance in time for a year afterwards, as Chapter Five described.

A working calendar

For a Nepali investor who wants a rhythm rather than a rule, here is the one I keep.

Mid-August. Unaudited full-year figures arrive. This is the first real look at how the fiscal year went, and it is where I do most of my annual re-reading of the companies I own.

Mid-November, mid-February, mid-May. Quarterly results. Check them against the falsifiers you wrote down at purchase — Chapter Four’s discipline — and do nothing else.

Mid-December. Audited annual accounts. Compare them with the unaudited numbers from August. Where they differ materially, that difference is information about the company’s reporting, and it is one of the few genuinely diagnostic signals freely available in this market.

Around Dashain. Expect thinner liquidity and wider spreads. Do not initiate a large position into a festival week if you can avoid it.

Late Asar. Do not sell, do not sit in cash, and do not schedule your rebalancing here.

That is the whole calendar. It has five entries and none of them is a trade.

Rozeff and Kinney’s January effect was, at the time, one of the strongest pieces of evidence against the idea that markets are efficient.

Its subsequent decay is now used as one of the strongest pieces of evidence for that idea — the market learned, and the opportunity closed.

Both readings cannot be right, and the fact that the same observation has been enlisted on both sides of the argument for fifty years should tell you something about how much weight a single seasonal regularity can bear.

Mine decayed by two thirds the moment I looked at data I had not chosen. I expect it to decay further. I have written it down anyway, with the mechanism and the correction and the reason not to trade it, so that when it disappears entirely there will be a record of what was claimed and by whom.

What Disappears

In which America loses two thirds of its banks without anybody much noticing; the academic literature establishes that buying a company is good for the seller and usually bad for the buyer; and we measure every Nepali bank merger and find that the acquirer’s shareholders have underperformed the banking index by a median of thirteen percentage points.

In 1984 the United States had about fourteen thousand four hundred commercial banks.

By the 2020s it had about four thousand.

Two thirds of the American banking industry vanished in a generation, and there was no crisis that did it — there were crises along the way, but the mechanism was legislative. Until 1994 American banks were largely prohibited from branching across state lines, a relic of nineteenth-century politics that had left the country with thousands of small institutions serving small areas. The Riegle-Neal Act removed the prohibition, and the industry consolidated for twenty-five years.

Almost nobody experienced this as an event. Depositors got a new sign on the branch. Employees got a new letterhead or a redundancy. And shareholders — who are the subject of this chapter — got shares in something else, at a ratio somebody had negotiated.

Nepal has just been through its own version, compressed into about five years, and Chapter Three counted the bodies. This chapter is about what it does to you when your company is one of them, and it reaches a conclusion I did not expect.

The best-documented finding in corporate finance

If you buy a company, who gains?

This has been studied more thoroughly than almost anything else in finance, across thousands of transactions and several decades, and the answer is remarkably stable:

Target shareholders gain. Acquirer shareholders do not.

The target’s holders receive a premium — that is what makes them agree — and they capture it immediately, on announcement. The acquirer’s holders receive the target’s business, the target’s problems, an integration project, and a bill.

The literature is not merely neutral on acquirers; it is frequently negative. One much-cited study of American acquisitions during the late 1990s found that acquiring shareholders lost hundreds of billions of dollars in aggregate, and the pattern is worse for large deals and for deals paid in shares rather than cash.

The proposed explanations are unglamorous and probably all partly true: acquirers overpay because the process is competitive and the chief executive wants the deal; synergies are estimated by the people who want the deal approved; integration is harder than anyone models; and a company that is for sale is often for sale for a reason.

Hold that finding and let us look at Nepal, where consolidation was driven by something the international literature mostly does not contain.

Nepal’s version: consolidation by capital requirement

Nepal Rastra Bank decided, across the 2010s, that the country had too many financial institutions. This was a defensible judgment. A small economy with dozens of commercial banks, dozens of development banks, dozens of finance companies and fifty microfinance institutions has a supervision problem, a governance problem, and a great many institutions too small to invest in systems.

The instrument chosen was minimum paid-up capital. Rather than ordering mergers, the regulator raised the amount of capital an institution was required to hold, by multiples, over a defined period. An institution that could raise it survived. One that could not had to find a partner.

The insurance regulator did the same thing to insurers a few years later.

This is an elegant policy in one respect — it is neutral as to who merges with whom, and it lets the market sort out the pairings — and it has one consequence that matters enormously for the finding below.

In a merger driven by a capital requirement, the acquirer is not necessarily buying something it wanted. It is buying something available, at a moment when it also needs to grow, from a seller whose alternative is not to remain independent but to find somebody else. That is a very different transaction from a strategic acquisition, and there is no reason to expect it to produce a better outcome for the buyer.

What actually happened to the acquirers

I took every Nepali bank merger for which I have sourced dates and terms — nine disappearing tickers producing seven surviving institutions — and measured what the surviving entity’s shares did after unified trading began, against the banking index over the identical window.

Five of seven underperformed the sector in their first year as a combined institution. Median excess return: −6.6 per cent.

And from the merger date through to July 2026:

Median excess: −13.1 per cent.

Four of seven have underperformed the banking index since absorbing their target, two of them severely. Himalayan Bank, which paid a premium of about six and a half per cent over market for Civil Bank, has trailed the sector by nearly forty-one percentage points since. Nabil, the largest and most admired bank in the country, has trailed by twenty-two.

And two have done extremely well — Laxmi Sunrise and Global IME — which is the dispersion you should expect and which is the reason this is a tendency rather than a law.

But the central tendency is clear and it matches the international literature exactly. In Nepal, as everywhere, being the acquirer has been bad for shareholders.

Why, specifically

Three mechanisms, and they are worth separating because they have different implications.

You buy the loan book. A bank is not a factory; it is a portfolio of credit decisions made by other people over a decade. When you absorb one, you absorb every loan its managers wrote, including the ones they had not yet recognised as impaired. In a regulator-driven consolidation, the institutions most likely to be absorbed are the ones least able to raise capital, and the reason a bank cannot raise capital is frequently something in its loan book.

The reported ratios get worse before they get better. Combined entities in Nepal have generally reported deteriorating asset quality in the quarters following a merger, which is partly genuine deterioration and partly the recognition of things that were already there. Either way the market sees a worse bank than either predecessor.

And integration consumes years. Two branch networks, two systems, two credit cultures, two sets of staff on different terms. In a country with a limited pool of senior banking talent, this is not a project that concludes quickly.

The rule that follows

Chapter Three established that the merger swap itself was fair — Nepal Bangladesh shareholders received within one per cent of market value, Civil Bank shareholders received six and a half per cent more. The disappearance was not the robbery.

This chapter adds the other half, and together they give a clear instruction:

If you own the target, the merger is broadly neutral to slightly positive, and you should not panic. You will receive shares in the acquirer at a ratio that approximately reflects what the market said you had.

If you own the acquirer, a merger announcement is a reason to re-underwrite the company from scratch. It is no longer the institution you analysed. Its loan book has changed, its capital position has changed, its management’s attention is now consumed by an integration, and the historical record says its shares will probably lag its sector for a year or more.

That second instruction is the opposite of the instinct, which treats an acquisition as evidence of strength — the strong bank absorbing the weak one. It is evidence of strength in a regulatory sense and frequently a burden in an economic one.

The other way out

Absorption is the common exit in Nepal. There is a rarer and worse one.

Delisting occurs when a company ceases to meet listing requirements, or is wound up, or simply stops complying. The shareholder is left holding a security with no market, and the practical position is that recovery depends on whatever process follows, which in Nepal is slow.

I want to be careful not to alarm here, because outright delisting of an operating Nepali company has been rare — far rarer than merger. The larger practical risk is a trading suspension: a company that fails to hold its annual general meeting, or fails to publish its accounts, may find its shares suspended, and a suspended share is one you cannot sell for as long as the suspension lasts.

Which produces the only genuinely defensive rule in this chapter, and it is cheap: treat a company that is late with its filings as a company with a problem. Not necessarily a fraud, not necessarily an insolvency — but a company that cannot produce its accounts on time is telling you something about its systems, its auditors, or its board, and the cost of acting on that signal is zero because you can simply own something else.

Chapter Fifteen’s engine refuses to value companies whose disclosures do not support a valuation. This is the same rule applied by a human being: the refusal is information.

What the consolidation actually achieved

I should be fair to the policy, because I have spent a chapter on its costs to shareholders.

Nepal now has fewer, larger, better-capitalised financial institutions than it did a decade ago. Supervision of nineteen commercial banks is a tractable problem; supervision of thirty-two was not. The institutions that disappeared were disproportionately the ones that could not raise capital, which is a reasonable proxy for the ones that ought not to have existed.

The system is more robust. That is a public good, and it was purchased partly with the returns of the shareholders of the acquiring banks, who were not consulted and who have collectively underperformed their own sector by a median of thirteen points since.

A policy can be correct and still be paid for by a specific, identifiable group. Recognising which group you are in is not cynicism. It is the entire job.

The Riegle-Neal Act was passed in 1994, and by the time American consolidation had run its course the country had lost about ten thousand banks.

The number of bank branches, over the same period, went up.

Which is worth sitting with. The institutions vanished and the service did not. What disappeared was not banking. It was ownership — ten thousand separate boards, ten thousand separate shareholder registers, ten thousand small towns where somebody’s grandfather had bought two hundred shares and never sold.

Who Is On The Other Side

In which the American market stops being owned by people; India builds a domestic institutional bid out of thirty thousand crore a month of small standing orders; and we ask who, in Nepal, is actually buying the share you are selling — and find that the answer is almost always somebody exactly like you.

In 1950, individuals owned something like ninety per cent of American equities directly. Households held certificates. The market was retail in the most literal sense.

By the 2010s, institutions owned roughly eighty per cent of large American companies. Pension funds, mutual funds, endowments, insurers, and above all the index funds — Vanguard launched the first one for retail investors in 1976 to widespread derision, and by 2019 passive funds held more American equity than active ones.

Three firms now hold something in the region of a fifth of the S&P 500 between them.

This transformation changed the character of the market in ways worth naming, because Nepal has none of them and the absence is the subject of this chapter.

Institutions produce research. A fund with a billion dollars can afford analysts, and analysts read filings, and the reading gets into prices. American mid-cap companies are covered by a dozen people whose job is to find the error.

Institutions impose price discipline. A fund that believes a company is overvalued will not buy it at any price, and can sell what it holds, and — where short selling exists — can bet against it. The presence of large, unsentimental, well-informed capital narrows the range over which a price can wander.

Institutions vote. Governance in developed markets works, to the extent it works, because somebody with three per cent of a company turns up and objects.

And institutions herd, which is the cost. Their incentives are career incentives: underperforming the peer group is far more dangerous than underperforming in absolute terms, so they cluster. Chapter Eight’s Jeremy Grantham lost clients for being right too early, and that is the mechanism.

What India built

The most instructive parallel for Nepal is not America. It is India, and specifically what happened there over the last fifteen years.

Indian equity markets were, for decades, dominated at the margin by foreign institutional flows. When foreign money arrived, the market rose; when it left — as in 2008, or during the 2013 taper tantrum — the market fell hard, and the currency fell with it. India’s market was a leveraged bet on somebody else’s risk appetite.

Then Indian mutual funds built the systematic investment plan: a small monthly standing order, automated, marketed relentlessly, aimed at the salaried middle class. Individually trivial. Collectively, monthly SIP inflows grew into the tens of thousands of crores of rupees a month.

The effect was structural rather than merely additive. India acquired a domestic institutional bid that arrives every month regardless of sentiment — money that buys on the fifth of the month whether the news is good or bad, because it is a standing instruction rather than a decision. When foreign investors sold heavily in subsequent episodes, domestic institutions absorbed it, and the market did not behave as it had in 2008.

A country changed the character of its market by giving a few tens of millions of salaried people an automatic way to buy shares.

Hold that, because it is the single most transferable idea in this chapter.

So who buys in Nepal?

Now the local census, and it is short.

The Employees Provident Fund and the Citizen Investment Trust are the two large domestic institutions with genuine long-term liabilities and correspondingly large asset pools. Both invest in listed equity, and both are constrained — by their own governing rules, by prudential limits, and by the practical difficulty of deploying size into the market described in Chapter Eighteen. They are large relative to Nepal and small relative to what an equity market needs to be priced by institutions.

Insurance companies hold investment portfolios against their policy liabilities, with allocations shaped by regulation. Life insurers in particular accumulate genuine long-duration money, and they are the closest thing Nepal has to a natural, patient, institutional equity buyer.

Mutual funds exist, and I will come back to them, because their structure is unusual and creates something exploitable.

Banks invest, subject to significant restrictions on holding equity.

And that is essentially the list.

There are no foreign portfolio investors in the secondary market. Nothing arrives from outside seeking a return; nothing flees when Istanbul or Jakarta looks more attractive. Whatever happens to Nepali share prices is done entirely by Nepalis with Nepali savings.

There are no index funds, so there is no automatic, price-insensitive bid.

There is no meaningful pension industry beyond the two institutions named above.

And there is essentially no sell-side research on the smaller two thirds of the market, which follows from Chapter Seventeen: in an exchange where companies list because a regulator required it, nobody is being paid to work out what they are worth.

Which means the marginal buyer is a person

Here is the consequence, and it is the most important sentence in this chapter.

In Nepal, the price is set at the margin by an individual.

Not by a fund with a model. Not by an analyst with a spreadsheet. By a person with a demat account and a telephone and, quite often, a group chat — the forty thousand members containing ten opinions from Chapter Six.

This is worth stating as a double-edged fact, because it is genuinely both.

The cost: prices can wander a long way from any defensible estimate of value, in both directions, and stay there for years, because there is no large well-informed capital with an incentive to correct them. The cascade mechanics of Chapter Six operate without an opposing force.

The benefit — and it is the reason Part Four exists: in a market where the marginal price-setter has not read the filings, a person who has read the filings is competing against almost nobody.

Chapter Four introduced Mauboussin’s paradox of skill: as everybody in a field gets better and the variation between them narrows, luck comes to dominate outcomes. That is why beating the American market has become so difficult even as the average manager became more capable.

Nepal is at the opposite end of that spectrum. The variation in skill among participants here is enormous, which means the return to being one of the careful ones is correspondingly large. It will not remain so; every market’s history runs in one direction. But it is so now.

The mutual funds, and the thing about them

Nepali mutual funds deserve a section because their structure is unusual and it creates a measurable, dateable, mechanical opportunity of a kind that barely exists elsewhere in this market.

Nepali funds are predominantly closed-end. A fund raises a fixed amount, issues units at a par of ten rupees, lists those units on the exchange, and runs for a defined term — commonly seven or ten years — at the end of which it matures and distributes its assets to unit holders at net asset value.

Two prices therefore exist simultaneously for the same thing.

The net asset value, published periodically by the fund manager, which is what the underlying portfolio is worth.

The market price of the unit, which is whatever the exchange says.

These diverge, frequently and substantially, and closed-end funds around the world typically trade at a discount to their net asset value — a puzzle the academic literature has argued about for fifty years without fully resolving.

Now the part that makes Nepal’s version interesting. A closed-end fund with a fixed maturity date must converge to its net asset value on that date. It is not a matter of sentiment; the fund liquidates and pays out. So a unit trading at a twenty per cent discount three years before maturity carries a mechanical return of roughly seven per cent a year on top of whatever the portfolio itself earns, provided the discount does not widen further and the published net asset value is accurate.

I am not going to tell you this is free money, and there are three real objections. The published net asset value depends on the manager’s valuation of the holdings. The discount can widen before it closes, which is a two-year problem in a market that moves in the ranges described in Chapter Ten. And the units are frequently thin.

But it is one of the very few places in this market where price and value are both published, side by side, on the same screen — and where a convergence date is written into the fund’s own documents. If you want a place to practise the discipline of buying below value without having to construct the value yourself, it is here.

One data point worth noting: the single best-performing security in the entire 2020–21 boom, across all one hundred and eighty-five names I measured in Chapter One, was not a hydropower company or a finance company. It was a mutual fund, at 15.43×, and a substantial part of that was a deep discount closing at the same time as the market rose.

Nobody attends

One more absence, and it is the governance one.

In a market with institutions, somebody with a meaningful stake attends the annual general meeting, reads the resolutions, and occasionally objects. Nepali annual general meetings are attended by a small number of retail holders, largely for the refreshments and the token gift, and the resolutions pass.

The consequence is that Nepali companies are run by their promoters, for their promoters, with public shareholders as a compliance category — which is exactly what Chapter Seventeen predicted from the way the exchange was populated.

I do not have a remedy to offer and I distrust people who pretend otherwise. What I have is a piece of practical advice that follows directly: since you cannot influence governance, you must price it. A company whose promoter has a record of related-party transactions, or of treating minority holders casually, is not a company you can fix. It is a company you buy cheaper or do not buy. Part Four returns to this repeatedly, because in Nepal governance is not a soft factor. It is frequently the entire difference between two otherwise identical banks.

What would change all of this

Three things, in ascending order of likelihood.

Foreign portfolio investment. Repeatedly discussed, never implemented at scale. It would bring research and price discipline and would also bring the volatility of somebody else’s risk appetite, as India learned before it built its domestic bid.

A real pension system. The most consequential of the three, and the slowest.

And systematic investment plans at scale — Nepal’s version of what India built. The infrastructure largely exists: mutual funds exist, demat accounts exist, banks can automate standing orders. What does not exist is the product, the distribution, and the habit.

If it arrives, two things follow. The market acquires a monthly bid that does not care about news, which dampens the cycle described in Chapter Nineteen. And the amateur advantage described in this chapter begins to close, because money that arrives automatically is money that is not making the errors you are trying to profit from.

Both of those are good for Nepal. Only one of them is good for you.

Vanguard’s first index fund raised eleven million dollars against a target of a hundred and fifty million, and was described within the industry as un-American — the argument being that settling for the average was a betrayal of the whole enterprise.

Its founder spent the next four decades being told that the idea could not work at scale, and then that it worked too well, and that if everybody indexed there would be nobody left to set prices.

That last objection is the interesting one, and it is the one Nepal answers from the other direction. A market needs somebody doing the work. In America the worry is that too few people are left doing it.

Here, almost nobody has started.

The Data You Can Actually Get

In which nobody knows what the American stock market has returned until 1964; a university spends four years building the answer; and we discover that one of Nepal’s two price series is shifted forward by exactly one trading day, which took a year to find and cannot safely be repaired.

Until 1964, nobody knew what the American stock market had returned.

I do not mean that estimates varied. I mean that the question had no answer, because nobody had ever assembled the prices. There were newspaper records going back decades, and index levels of various constructions, and a great deal of folklore — but no continuous, corrected, dividend-inclusive record of what an investor would actually have earned.

In 1960 the University of Chicago received a grant from a brokerage to find out, and the Center for Research in Security Prices was created. It took four years. The work involved reconstructing monthly prices for every stock on the New York Stock Exchange back to 1926, adjusting for splits and dividends and delistings, from paper records, onto punch cards.

When Lawrence Fisher and James Lorie published the result in 1964, the answer was that American common stocks had returned about nine per cent a year.

That number — which every investor now treats as common knowledge, and which underpins every retirement calculation ever performed — did not exist before somebody spent four years building a database.

Empirical finance begins with a data project, and essentially everything you believe about markets descends from one. This chapter is about what happens when the data project has not been done, which is the situation in Nepal, and about what I found when I tried to do it.

The three sources

A Nepali investor has three places to get prices, and they are not equivalent.

Merolagani publishes server-rendered pages that can be read programmatically, which makes it the only genuinely automatable primary source. Its layout changes occasionally, at which point whatever you built stops working.

NepseAlpha is a commercial provider with the deepest history. Its bulk export interface caps at five years; its charting endpoint does not, and serves far more — close to twenty-nine years of index history and up to fourteen and a half years for individual securities. It is Cloudflare-protected against automated access.

ShareSansar is widely used and structured in a way that resists programmatic reading.

And the exchange itself publishes, but not in a form that supports research.

I want to record what it took to assemble a research panel from these, because the difficulty is itself a finding about this market, and because the successful approach is not obvious.

Four transport routes failed before one worked. Bulk automated downloads are refused by browser policy. Clipboard transfer requires operating-system window focus, which an automated window does not have. A direct request from a secure page to a local receiver is blocked by mixed-content and private-network rules. And using the provider’s own data endpoint directly was declined on principle, since it carried a session-scoped token that was not mine to use.

What worked was an HTML form submission to a local receiver — because a form submission is a navigation rather than a subresource request, and browsers permit navigation to loopback addresses where they forbid other traffic. Every payload was written to disk verbatim before being parsed, so that a parsing error could never silently corrupt the record.

The result is a panel of 384 symbols and 543,622 adjusted daily bars, 1997 to 2026, including delisted and merged names.

I report the mechanics because acquisition difficulty is a genuine barrier to research in frontier markets, and because anybody attempting this will otherwise spend a month discovering the same four dead ends.

The one-day shift

Now the defect that took longest to find and that I want to use as the general lesson.

I had two price series for Nepali stocks: one harvested from the charting source, and an older one built from Merolagani. When I compared them, prices disagreed constantly — but not randomly. The disagreement had a structure I could not characterise for a long time.

Then I tested a specific hypothesis: that one series was shifted by exactly one trading day relative to the other.

For Laxmi Sunrise, the two series matched on 97.7 per cent of observations when one was shifted forward a day, against 30.9 per cent when compared on the same date. For Bottlers Nepal, 57.6 per cent against 32.3.

That is not a suspicion. That is a proof. One of the two series carries each day’s price under the following day’s date.

There was a second, cruder confirmation. The newer panel has zero Saturday bars. The older one carried thirteen to fifteen Saturdays a year — sessions that did not happen, because the market was closed, appearing because the dates had been pushed forward across a weekend.

I have not repaired it. The temptation is enormous and I have resisted it, for a reason worth stating: a safe repair requires knowing exactly which segments of the series are affected, and a partial repair — fixing some spans and not others — is far worse than a known, documented, consistent offset. A documented error can be reasoned about. A half-corrected one cannot, because you no longer know which convention any given row follows.

The general rule, and it applies well beyond price data: a known defect that is recorded is safer than an unknown correctness.

The fabricated opens

Chapter Five established this and I restate it here because it belongs in the data chapter.

NEPSE did not publish a session opening price for most of its history. Data vendors, needing four numbers per bar to draw a chart, filled the missing column with the previous close.

Nabil Bank’s recorded open equalled the previous close on 164 of 164 trading days in 2013. Nepal Telecom: 146 of 146. Across the whole panel, between seventy-five and eighty-two per cent of bars in 2012 to 2014 carry a manufactured open, falling to about seven per cent by 2025 as real opens became available.

Which means that any calculation involving the opening price — candlestick patterns, overnight-gap studies, opening-range strategies, intraday volatility estimates — is, for the first half of the available history, a calculation performed on a field that was generated by an import script.

I detected this by comparing the distribution of open-to-close returns across eras: a distribution with a large spike at exactly zero is not a market. My engine now nulls the open field for the affected span rather than using it, and emits a warning that capacity for that period is unknown rather than unlimited.

What point-in-time actually requires

Prices are the easy part. Fundamentals are harder, and the difficulty has a specific name.

If you want to test whether cheap companies outperformed, you must know what was knowable at the time. A company’s results for the quarter ending in mid-July were not public in mid-July; under Schedule 14 they may appear up to thirty days later, and under Schedule 15 the audited annual accounts may take five months.

A database that stores figures against their period rather than their publication date will happily let you construct a strategy that bought a company in July on the strength of numbers published in December. The backtest will look magnificent. It is using information from the future.

This is the most common and most fatal error in quantitative work, and it is invisible in every summary statistic, because nothing about it looks wrong.

My own gating uses statutory disclosure dates rather than period ends, with Bikram Sambat quarter boundaries anchored at approximately 17 October, 14 January, 13 April and 16 July, and no fundamental strategy is permitted to reach period data by any other route.

And the coverage is the real limitation: fundamentals of usable quality exist for about fifty of the hundred and eighty-nine tradeable names.

Not because the others do not file. Because the filings arrive as scanned images, in inconsistent formats, with line items that change names between years, in Preeti-font Nepali in some cases, and with no machine-readable version anywhere. Extracting them is a parsing problem of genuine difficulty, and I have spent more time on it than on any other part of this work.

What simply is not available

An honest register of the holes, because a reader should know the boundary of what any Nepali analysis can claim.

Shares outstanding are not reliably available as a time series. Which means no size factor, no market-capitalisation weighting of a custom index, and no ability to compute a company’s market value historically without reconstructing every bonus and rights issue by hand.

Corporate action detail is not verifiable against the vendors’ own adjustments. I can see that a series has been adjusted; I cannot always confirm that it was adjusted correctly.

Listing dates are unavailable, which means the drift of newly listed shares cannot be studied cleanly.

Mutual fund net asset values are not available as a series, which is why Chapter Thirty could describe the closed-end discount mechanism but not measure it.

And the trading calendar cannot be generated, only observed, because the week has changed four times and the market has closed for months.

The gates I run, and why you should run something similar

Everything above produced a set of checks that I apply to any new data before it enters the store, and the principle behind them transfers to anybody handling numbers.

A calendar gate. A date enters the trading calendar only if a minimum number of symbols printed that day. This is not cosmetic. A calendar containing dates the market never opened on produces no error at all; trailing windows simply come back short and signals quietly stop firing.

A basis gate. Adjusted and unadjusted series are stored separately and the system refuses to mix them, because a silent basis mix is invisible in every summary statistic and corrupts everything downstream.

A continuity gate. Bars that move beyond a multiple of the price limit in force are flagged. Across 543,622 bars, only 132 — 0.024 per cent — breach it, and they are concentrated in funds and promoter lines rather than ordinary shares, which is reassuring about the panel and about the exchange.

And a plausibility gate on fundamentals. A figure that fails a scale check is held for review rather than stored. The reason is a specific incident: a parser once extracted a bank’s figures at roughly a thousand times their true value, and every identity check passed — assets still equalled liabilities plus equity, ratios were still internally consistent — because the error was uniform. Identity checks cannot catch a uniform scale error. Only an absolute plausibility check can.

The practical version for a reader who is not building a database

You are probably not going to harvest half a million bars. Here is what the chapter means for you anyway.

Know which series you are looking at, and whether it is adjusted. If a healthy company appears to have fallen fifty per cent on a single day, look up the corporate action calendar before concluding anything. Chapter Twenty-Seven explains what you will find.

Distrust any chart of a Nepali stock before about 2015 that depends on the opening price. The field is manufactured.

Never compare two sources on the same date without checking the alignment first. One of Nepal’s two main sources is shifted a day. If you are eyeballing a price from one site against a calculation from another, you may be comparing different sessions.

And when a number surprises you, check the number before you build a theory around it. This is the least glamorous discipline in the book and it has saved me more times than any insight. A striking finding is far more often a data defect than a discovery, and the ratio is not close.

Fisher and Lorie’s nine per cent has been revised many times since — for survivorship in the original sample, for delisting returns that were missing, for the treatment of dividends. Each revision moved it, and one paper in 1997 showed that the absence of proper delisting returns had been inflating a whole generation of published results.

The most-cited number in finance has been corrected repeatedly for sixty years by people going back to the source.

Nepal’s equivalent number does not exist yet. What I have offered in this book is a first attempt, built from one panel, with the defects listed above written down where you can see them.

Treat it accordingly.

I Tried to Beat This Market

In which the finance literature discovers that most of the finance literature is wrong; three hundred and sixteen published ways to beat the market turn out to be mostly noise; and I run twenty-six strategies against thirteen years of Nepali data, find nothing, publish three things anyway, and retract all three within days.

In 2016 Campbell Harvey, then president of the American Finance Association, published a paper cataloguing every factor that had been claimed in the academic literature to predict stock returns.

He found three hundred and sixteen.

Momentum. Value. Size. Quality. Accruals. Asset growth. Net share issuance. Idiosyncratic volatility. Three hundred and sixteen distinct published claims, each supported by statistical tests, most in respectable journals, each purporting to describe something real about how prices behave.

His argument was not that they were all false. It was that the profession had been using the wrong threshold for the whole period. When you have tested hundreds of candidates — and the profession collectively has, even if no individual paper tested more than a few — a result that clears the conventional significance bar is not evidence of anything, because with enough candidates a large number will clear it by chance. He proposed that the hurdle be raised substantially, and noted that on that hurdle a majority of published factors do not survive.

Four years later, Hou, Xue and Zhang attempted to replicate four hundred and fifty-two documented anomalies using consistent methods, and found that around sixty-five per cent failed to replicate at conventional significance.

And McLean and Pontiff had already measured the decay: anomaly returns are about twenty-six per cent lower out of sample, and fifty-eight per cent lower after publication. Some of that is the original estimate having been too high — a result selected for being large is a result selected partly for being lucky. But most of it, they concluded, is the market learning: they attribute the larger share of the post-publication decay to investors reading the paper and trading the anomaly away.

This is the state of the field. It is not a scandal and nobody was dishonest. It is what happens when a large number of intelligent people search the same dataset for regularities and only publish what they find.

I bring it up because I did the same thing to Nepal, and I want to tell you what happened, including the parts that make me look foolish, because those are the parts worth transmitting.

What I built

I wanted to know whether anything worked in this market, and I concluded early that no existing tool could answer it, because every backtesting framework in existence is built for markets where the frictions are second-order.

Here they are not. So the engine enforces, during execution rather than as an afterthought:

Settlement at T+2, as a hard constraint on capital availability, exactly as Chapter Twenty-Three describes.

Price limits as a dated regime table — ten per cent until April 2026, fifteen thereafter — under which a bar that closes at the limit is treated as offering no supply, and orders against it are refused rather than filled. Chapter Twenty-Five’s measurement is what justifies this.

Charges in currency, not basis points. The flat twenty-five rupees cannot be expressed as a rate, and any cost model in percentages systematically understates the cost of small trades and therefore overstates every strategy that holds many small positions.

The capital gains boundary at 365 days, with the correct rate on each side.

Participation caps derived from realised volume, so a strategy cannot buy more than a plausible share of what actually traded.

And deposit interest on idle cash, drawn from the observed Nepal Rastra Bank series rather than an assumption — a decision that later destroyed one of my own findings, as you will see.

Look-ahead is prevented structurally rather than by discipline: the engine cannot reach future data because the interface does not expose it, and an adversarial test verifies this rather than trusting it.

The screen: twenty-six strategies, none of them worked

Universe: 189 investable names. Window: 2013 to 2026, thirteen and a half years. One million rupees. Full charges, settlement enforced.

Measured as excess over the benchmark, which is the only question that matters:

Every strategy underperformed. Every information ratio is negative. A formal test across the whole set — White’s Reality Check, which asks whether the best performer in a collection is better than chance given how many were tried — returned p = 0.993.

Not “not significant.” As close to a complete absence of evidence as the test can report.

Look at the charges column, because it contains the mechanism. Momentum paid 1.25 million rupees in charges on a one-million-rupee account over thirteen years. It paid its own starting capital, and a quarter more, to the exchange and the depository, for the privilege of underperforming a portfolio that did nothing.

And short-horizon mean reversion, which I tested separately, destroyed ninety-six per cent of the account. Charge drag fell monotonically from about 7.1 per cent a year at the shortest holding periods to under 1 per cent at the longest — which is Chapter Fourteen’s argument, measured.

The benchmark error, which the protocol caught

Before any of that, I nearly published the opposite conclusion.

My first benchmark was a buy-and-hold portfolio of the investable universe, and against it every strategy beat the market by five to eleven points a year. I was, briefly, delighted.

The tell was the trade count: seventeen fills, in a universe of a hundred and eighty-nine names. The rule required 250 sessions of history before buying, so in February 2013 only seventeen companies qualified, and it never bought any of the hundred and seventy-two that listed afterwards. It was not the market. It was a frozen 2013 vintage, and it returned 5.0 per cent while the real thing returned 16.8.

Replacing it turned every apparent victory into a defeat.

A backtest is only as honest as the thing it is compared against, and an implausible trade count is worth more than a plausible return. I found this because I had committed in advance to checking trade counts, not because I was clever on the day.

The three retractions

I published three findings during this work and retracted all three. Here they are, in the order they died.

One: portfolio construction. Inverse variance, risk parity and minimum variance weighting appeared to add up to 2.51 percentage points a year over equal weight. Then I ran it across five specifications, and every scheme’s sign reversed across them. The reason is Chapter Twenty-One’s correlation measurement: NEPSE’s cross-section is correlated enough that no weighting scheme reduces portfolio volatility below about 19.5 per cent against equal weight’s 22.3. There is not enough independence in this market for construction to have anything to work with.

Two: the exposure result — retracted twice. I found that reducing equity exposure appeared to retain ninety-seven per cent of the compound return while eliminating sixty-three per cent of the drawdown. It looked like the closest thing to a free lunch I had ever measured.

It died in two stages. First, excising the single worst crash showed the effect was concentrated in one episode rather than being a general property. Then I replaced my assumed deposit rate of 8 per cent with the observed Nepal Rastra Bank series, whose realised mean was 5.31 per cent. At the real rate, the optimal exposure moved to a hundred per cent and the finding evaporated entirely.

I had manufactured a result by assuming a risk-free rate 2.7 points above what depositors actually earned. That is not a neutral placeholder; it is a thumb on the scale, and it is exactly the failure that a standing rule against hardcoded market assumptions exists to prevent — a rule I had written myself and then violated.

Three: volatility timing. Scaling exposure by recent volatility scored +0.35 percentage points. Across blind sub-periods it ranged from −1.45 to +1.49. Retracted.

And a defect that voided results I had already reported

Separately from the retractions, I found a bug in a covariance computation that voided two cells of the strategy screen and one result from a sealed test.

The suite was green. Every test passed. The defect was of a kind that unit tests structurally cannot catch: the function computed something internally consistent that was not the quantity intended, so every assertion about its self-consistency held.

I record it because it is the general case. A test that checks whether code agrees with itself will agree forever. The only checks that catch this class are external: does the number look like what it must look like if it is right? Which is a question a human has to ask, and which I did not ask for several weeks.

What actually survived

Two things. Both are unglamorous, and I trust them precisely because of what they are.

Slowing the rebalance clock is worth about 1.8 percentage points a year. Moving from a twenty-one-day rebalance to a two-hundred-and-fifty-two-day one produced a median gain of 1.80 points, positive in all four blind sub-periods, and monotone in the clock within each.

It survives because it is not a forecast. It is an accounting identity over a published fee schedule — the same arithmetic as Chapter Fourteen and Chapter Twenty-Four, arriving from a third direction. Nothing about it requires predicting anything, which is why it did not evaporate when tested.

And the growth-optimal equity weight is not estimable in this market. Across blind sub-periods the implied optimum ranged over thirteen units of exposure — the best weight was a hundred per cent in both bull blocks and zero in both bear blocks.

The reason is a single ratio. Nepali equity volatility is stable: 17.7 to 26.1 per cent across the blocks. The mean return is not: −9.5 to +35.2 per cent. When the numerator of a ratio moves forty-five points and the denominator moves eight, the ratio is not a quantity you can estimate. It is noise with units.

I regard this as the most important thing the whole programme produced, and Part Five is built on it, because it converts a question normally posed as an optimisation into a question of preference. If the optimal exposure cannot be computed, then how much equity you hold is a decision about what you can tolerate, not a calculation. Anybody offering you the calculated answer has not measured the spread.

The register

I kept a log of every error I made during this work. It has twenty-five entries, classified by type and by what detected each one.

The distribution is the useful part. Very few were caught by tests. Several were caught by an implausible number — the seventeen trades, a strategy that turned out to be 26 per cent invested while appearing fully deployed, a beta of 0.4 for nineteen banks against an index they dominate. Several were caught by a robustness check I had committed to in advance and had to run whether I wanted to or not. One was caught by somebody telling me I was wrong about the trading week.

Almost none was caught by re-reading my own reasoning. Reasoning does not audit itself. Only an external check does, and the external checks have to be arranged beforehand, because afterwards you will not want them.

What this means for you

Nothing in the standard toolkit beat this market after costs. Not momentum, not trend, not low volatility, not equal-weight variations, not mean reversion. If a course or a group is selling you one of these for Nepal, ask what it was tested against, over what period, with what charge model, and how many variants were tried.

The costs are the dominant term. Momentum’s problem was not that its signal was wrong; the signal was weakly positive. Its problem was 1.25 million rupees of charges.

Ask “how many did you try?” before “what were the returns?” That is Harvey’s correction, Chapter Five’s Brock-and-Sullivan story, and my own screen, all saying the same thing.

And notice what is left. If systematic strategies do not work here, and the market’s prices are set at the margin by individuals who have not read the filings — Chapter Thirty — then the remaining avenue is not cleverness about prices. It is knowing what a business is worth.

Which is the only thing I have left to offer, and it is the subject of everything that follows.

Harvey’s paper is titled with an ellipsis: … and the Cross-Section of Expected Returns. The joke is that the elided part is the name of whatever factor the author is proposing, because there had been so many that the title had become a template.

He was not attacking the field. He was a distinguished member of it, saying that the collective search had gone on long enough that the results could no longer be read at face value, and that somebody should say so.

Somebody usually has to, and it is generally somebody who has spent years producing the thing he is now questioning.

What Is Left

In which a man loses seventy per cent of his money and then writes the book that founds a discipline; we establish why estimating what a business is worth is a tractable problem when forecasting its price is not; and Part Two closes by naming the only thing it has not ruled out.

Benjamin Graham was nearly destroyed by the crash.

This is the fact that gets left out of the reverential accounts. Between 1929 and 1932 his investment account lost something in the region of seventy per cent — he was not a spectator of the Depression, he was one of its casualties, and he came within reach of being finished entirely. He spent the following years working out what had gone wrong, teaching at Columbia in the evenings, and in 1934 he published Security Analysis with David Dodd.

It is a book written by a man who had just been proved wrong about almost everything, and its central idea is shaped by that experience in a way that a more successful author would never have arrived at.

Graham’s contribution was not a method for being right. It was an admission that you will frequently be wrong, and a procedure for surviving it.

The margin of safety is not a valuation technique. It is a confession. It says: my estimate of what this business is worth is imprecise, my inputs are uncertain, the future is unknowable, and therefore I will only buy when the price is far enough below my estimate that I can be substantially wrong and still not lose money.

That is the intellectual descendant of everything in Part One, arrived at independently in 1934 by somebody who had learned it the expensive way.

What Part Two established

Before going forward, let me put the machine on one page, because sixteen chapters of plumbing are worth nothing if you cannot carry them.

The exchange exists for reasons that have little to do with capital formation. Most Nepali companies are listed because a regulator required it, which explains the disclosure quality, the governance, and the absence of anybody being paid to work out what things are worth.

The float is small and the price is set on it. Fifty-one per cent is locked by construction, and much of the remaining forty-nine never moves.

The cycle is a banking cycle. Four booms, four busts, and in every case for which data exists, the fuel was the funding position of the banking system. The two lowest deposit rates on record sit at the two market tops.

Buying a peak costs a few points a year; being finished when you buy costs everything. The lump-sum peak buyer of August 2021 is down 3.56 per cent a year. The man who started on the same morning and kept adding is up 5.49.

Diversification inside this market runs out at about ten holdings, because pairwise correlation rises from 0.31 in a boom to 0.64 in a bust, and a twenty-name portfolio is worth about one and a half independent bets in the year it matters.

The index is a bank index. It returned 3.57 per cent a year over the decade while the typical listed company returned something closer to eight.

The plumbing forbids things. T+2 traps your capital for two sessions. The flat twenty-five rupees means a trade under NPR 6,667 exists mainly to pay for itself, which sets a minimum position size, which caps how many companies you may sensibly own. The circuit locks upward six times as often as downward, and a limit-up stock rises another 3.17 per cent the next day while remaining unbuyable.

And nothing systematic worked. Twenty-six strategies, zero winners, p = 0.993. Three findings published and retracted. Two survivors, one of which is an accounting identity and the other of which is the statement that a key quantity cannot be estimated.

The negative space

Set all of that out and something becomes visible in what is absent.

I have ruled out forecasting the index. I have ruled out reading price patterns. I have ruled out the standard factor toolkit. I have ruled out seasonal rules, portfolio construction, volatility timing, and every form of trading frequency above the very lowest.

I have not ruled out one thing.

Nowhere in this book have I tested whether it is possible to work out what a Nepali business is worth and to buy it for less.

That question has not appeared, because it is not a question a backtest can answer. You cannot simulate judgment. There is no historical series of “what a careful analyst would have concluded about this bank in 2079,” and any attempt to construct one is circular — you would be testing a formula, and a formula is a factor, and factors are what Chapter Thirty-Two just spent a chapter dismissing.

So the honest position at the end of Part Two is: the systematic avenues are closed, and the judgmental one is untested rather than validated. I am not going to claim more than that. What I can do is give you the reasons to think it is the right place to spend a career, and they are three.

Why valuation is tractable when prediction is not

The first reason is that a valuation has an anchor and a forecast does not.

When you predict a price, you are predicting what a large number of other people will believe at a future date. There is nothing underneath that. It is opinion about opinion, which is Keynes’s beauty contest from Chapter Six, and it has no floor.

When you estimate what a business is worth, you are estimating a quantity that is attached to something: the deposits it holds, the loans it has written, the tariff in its power purchase agreement, the years remaining on its licence, the capital sitting above its regulatory floor. Those things exist. They are reported, imperfectly and late, but they are reported, and they constrain the answer.

An estimate with an anchor can be wrong. A forecast without one cannot even be graded.

The second reason is that nobody else is doing it. Chapter Thirty established the census: no foreign investors, no index funds, essentially no sell-side research on the smaller two thirds of the market, and a marginal price-setter who is an individual with a telephone. Chapter Four introduced the paradox of skill — as everyone in a field improves and the variation between them narrows, luck comes to dominate. Nepal sits at the opposite end of that curve from Wall Street. The variation in effort here is enormous, and the return to being one of the careful ones is correspondingly large.

And the third reason is that the frictions which killed every trading strategy do not touch a valuation strategy. Momentum paid 1.25 million rupees in charges on a one-million-rupee account. An investor who buys eight companies and holds them for a decade pays the flat fee sixteen times in total. Every structural feature of this market — T+2, the flat depository fee, the tax step at 365 days, the circuit that makes strength unpurchasable — punishes activity and is indifferent to patience.

The machine is hostile to trading and neutral toward ownership. That is not a small observation. It means the one remaining approach is also the one the market’s plumbing does not tax.

What valuation is not

Three warnings before Part Three, because “value investing” is a phrase that has absorbed a great deal of nonsense.

It is not buying cheap-looking things. A low price-to-earnings ratio is a description, not an analysis, and Chapter Twelve showed that a company reporting profit made of interest on its own unspent share capital will have an excellent ratio and no business. Part Four is largely about what the ratios mean sector by sector, and about which of them mean nothing at all.

It is not precise. Anyone producing a value of 487.30 for a Nepali company is performing arithmetic on assumptions, and the decimal places are decoration. Every valuation in this book is a range, with a stated width, and the width is not a weakness in the method — it is the method being honest about its inputs. Graham’s whole point.

And it is not fast. A valuation that concludes “roughly fairly priced, do nothing” is a successful valuation, and it is the most common result. Chapter Fifteen’s engine refuses to value companies whose disclosures do not support one, and the refusals turned out to be the most informative output it produced.

What comes next

Part Three is about the schools — the serious methods that people have built for deciding what a business is worth and whether to own it.

Graham’s margin of safety and the deep-value tradition it founded. Buffett’s departure from it toward quality and durability, and what Charlie Munger actually changed. Fisher’s scuttlebutt and the discipline of finding out things that are not in the accounts. Peter Lynch’s categories and his insistence that you can only value what you can describe. William O’Neil’s CAN SLIM, which is a hybrid and which meets an interesting fate in a market with a ten per cent circuit. Greenblatt’s formula and what happens when you mechanise judgment. The quality-growth-longevity-price framework that Indian managers built for a market structurally closer to ours than any Western one.

Each gets a chapter: what it claims, who built it, what evidence supports it, and — the part that matters here — whether it can be run in Nepal at all, given a market where the accounts arrive five months late, the float is locked, there is no short selling, and the flat fee makes a small trade uneconomic.

Some of them cannot. Saying which, and why, is the work.

Then Part Four takes the thirteen sectors one at a time and does the actual valuation — what each business is made of, which metrics rank first and which you may ignore entirely, what the bull and bear cases require, and how to arrive at a range.

Graham ran his fund for another twenty-two years after Security Analysis, and by his own account the single most profitable investment of his career was a large position in an insurance company that he had bought for reasons that were mostly quantitative, and which then grew enormously for reasons that were not.

He described it, with the dryness that runs through everything he wrote, as an irony — that one lucky decision had contributed more to his lifetime results than all of the careful ones combined.

He is the founder of the discipline of not being fooled by that kind of thing, and he was fooled by it, and he wrote it down.

That is the standard. It is not being right. It is noticing.

The Margin of Safety

In which a ruined man invents a discipline; a strategy of buying companies for less than their liquid assets returns twenty-nine per cent a year and then ceases to exist; the founder recants at the end of his life; and we discover that in Nepal today, every single company trading below one and a half times its book value is a commercial bank.

By 1932 Benjamin Graham had lost roughly seventy per cent of the money he managed.

He was thirty-eight. He had been running an investment partnership since 1926 with a respectable record, he was regarded as a capable analyst, and the three years from the autumn of 1929 had taken most of it. He has written that he considered leaving finance entirely. He kept going in part because his partner Jerome Newman’s family put in capital, and in part because he had begun teaching an evening class at Columbia, and teaching forced him to work out what he actually believed.

Security Analysis was published in 1934, co-authored with David Dodd, into a market that nobody wanted anything to do with.

It is a strange and important fact that the founding text of this discipline was written by a man in the middle of the worst experience of his professional life, about what he had got wrong. Almost every subsequent investment book has been written by somebody who had just done well. The difference shows.

Mr Market

Graham’s most durable idea is not a formula and appears in The Intelligent Investor of 1949 rather than the technical book.

Imagine, he says, that you own a small share in a private business, and that one of your partners is a man named Mr Market. Every day, without fail, Mr Market appears and names a price at which he will buy your interest or sell you more of his. He is obliging and he never takes offence.

He is also emotionally unstable. On days when he is cheerful he names a very high price, because he can see nothing but good things ahead. On days when he is frightened he names a very low one. His mood has nothing to do with the business, which is sitting there doing whatever it does regardless.

Graham’s instruction is that you may let Mr Market’s quotation set your view of value only if you happen to agree with him, or if you want to trade with him — which Buffett later compressed into the line everybody now quotes: Mr Market is there to serve you, not to guide you. It is his pocketbook, not his wisdom, that you will find useful. You are free to transact with him or ignore him. What you must never do is take his quotation as information about the value of your business — which is precisely what the whole of Part One of this book demonstrated that everybody does.

I have read a great deal of investment psychology written since, most of it with citations and experimental support, and none of it improves on a paragraph written in 1949 by a man describing his imaginary business partner.

What the margin of safety actually says

The second idea is the one Graham himself nominated as the central concept of investment if he had to reduce it to three words — the title and the opening of Chapter Twenty of The Intelligent Investor, present from the first edition of 1949.

It is not “buy cheap things.”

The margin of safety is the recognition that your estimate of value is wrong, and the construction of a buffer large enough that being wrong does not hurt you.

Consider the difference. An investor who calculates that a company is worth 400 and buys it at 395 has performed a valuation. An investor who calculates the same 400 and refuses to buy above 280 has performed a valuation and made an admission — that his 400 might really be 320, that the inputs are estimates, that the accounts might be optimistic, that the industry might turn.

The buffer is not timidity. It is the only honest response to the epistemics of Chapter Sixteen. You do not know. You cannot know. Therefore you require compensation for not knowing, and the compensation is a lower price.

And the arithmetic of it is asymmetric in a way that matters. If you buy at a thirty per cent discount to value and you are correct, you make forty-three per cent as the gap closes. If you buy at a thirty per cent discount and your valuation was twenty per cent too high, you still make fourteen per cent. The buffer converts a range of outcomes from “some good, some bad” into “some good, some less good,” and doing that repeatedly across a career is the whole enterprise.

The net-nets

Graham’s most famous mechanical application was extreme, and it worked spectacularly until it disappeared.

Take a company’s current assets — cash, receivables, inventory — and subtract all liabilities, including long-term debt. Ignore the factories, the land, the brand, the going concern entirely. What remains is the net current asset value: roughly what you would have if the business were shut down tomorrow and its liquid assets were collected and its debts paid.

Graham’s rule was to buy at two thirds of that number or less.

Consider what such a purchase means. You are acquiring the entire operating business — every machine, every customer, every employee — for less than nothing, with a discount on the liquid assets thrown in.

Henry Oppenheimer tested it properly in a 1986 Financial Analysts Journal study covering 1970 to 1983, and found that a portfolio of net-nets returned around twenty-nine per cent a year against a market return of about eleven and a half.

And then they vanished. In developed markets net-nets have become vanishingly rare outside the depths of a crisis, for exactly the reason Chapter Four’s paradox of skill predicts: the screen is trivial to run, everybody runs it, and a company trading below liquidation value is now noticed within days rather than years.

The strategy did not stop working. It ran out of instances.

The defensive investor’s checklist

For the ordinary person, Graham offered something less exotic — seven criteria in Chapter Fourteen of The Intelligent Investor, which I set out because we are about to test them against Nepal.

• Adequate size. Not a tiny company.
• Sufficiently strong financial condition. Current assets at least twice current

liabilities; long-term debt no greater than net current assets.

• Earnings stability. Positive earnings in each of the last ten years.
• Dividend record. Uninterrupted payments for twenty years.
• Earnings growth. At least a third over ten years, using three-year averages at

each end.

• Moderate price-to-earnings. Not more than fifteen times average earnings of the

last three years.

• Moderate price-to-book. Not more than one and a half times.

With a combining rule: price-to-earnings multiplied by price-to-book should not exceed 22.5, so a company may breach one criterion if it compensates on the other.

Notice what this list is. It is not an attempt to find the best business. Every criterion is a filter against disaster — size against fragility, current ratio against insolvency, ten years of earnings against a fad, twenty years of dividends against a management that has never actually returned anything, and two price limits against paying too much for whatever survives.

Graham was not building a portfolio of winners. He was building one from which the catastrophes had been excluded, and letting the arithmetic do the rest.

The recantation

In 1976, months before he died, Graham gave an interview in the Financial Analysts Journal in which he said something his followers have been managing ever since.

He was, he said, no longer an advocate of elaborate techniques of security analysis for finding superior value opportunities. That approach had been rewarding forty years earlier, when Security Analysis was first published, but the situation had changed enormously since.

He favoured, instead, a simple and mechanical approach applied to a group of stocks.

The founder of fundamental security analysis concluded, at the end of a long career, that the detailed work no longer paid — because too many capable people were doing it.

This is the paradox of skill again, stated in 1976 by a man who had watched it happen to his own discipline over forty years.

And it is the single strongest reason to be interested in a market like Nepal’s, where the number of capable people doing the detailed work is, as Chapter Thirty established, close to zero.

Now: can any of this be run in Nepal?

Take the pieces in turn, honestly.

Net-nets are unavailable. Not scarce — structurally unavailable. More than half the Nepali listed universe by count and far more by market value consists of banks, development banks, finance companies, microfinance institutions and insurers. For a bank, the concept of “current assets less all liabilities” is meaningless: a bank’s liabilities are its business, its assets are loans, and the entire framework Graham built for industrial companies does not translate. You cannot liquidate a deposit franchise.

For the non-financial remainder — manufacturing, hydropower, hotels, trading — the concept applies in principle. In practice, as the P/B table below shows, Nepali manufacturers trade at multiples of book, not fractions of liquid assets.

Criterion four is unavailable. Twenty years of uninterrupted dividends requires twenty years of continuous existence with consistent reporting. Chapter Twenty-Nine counted the consolidation: nine bank mergers, fifteen insurers absorbed, fifty-four tickers ceasing to print. A large part of this market does not have a twenty-year history because the entity did not exist twenty years ago in its present form.

Criterion three is difficult. Ten years of positive earnings is checkable in principle, but Chapter Thirty-One established that fundamentals of usable quality exist for about fifty of a hundred and eighty-nine tradeable names, and reconstructing a decade by hand from scanned filings is the work of a week per company.

Criterion two does not apply to financials for the same reason net-nets do not.

Criteria six and seven transfer completely. Price to earnings and price to book are computable for any company that reports, and they are the two most useful numbers in this market.

So Graham’s checklist mostly fails here, for reasons of data and structure rather than principle. His concept transfers entirely. That is the distinction to hold, and it will recur in every chapter of this part.

What the book value screen finds

Let me run the one criterion that does work.

I took every company in my fundamentals store with a computable book value — total equity divided by shares outstanding, where shares are derived from paid-up share capital at Nepal’s par value of a hundred rupees — and compared it with the current market price. Seventy-two companies.

That is the complete list of Nepali companies passing Graham’s seventh criterion.

Nine companies out of seventy-two. Every one of them is a commercial bank.

And at the other end of the same table:

Two things follow, and they organise a great deal of Part Four.

The Nepali market is not expensive or cheap. It is both, in different rooms. The commercial banks trade near or below their accounting net worth while manufacturers and life insurers trade at eight to twelve times theirs. Any statement about “the valuation of the Nepali market” that does not name a sector is meaningless.

And a low price-to-book in a bank is not automatically a bargain. This is the trap Graham’s own framework warns about and which his followers routinely fall into. A bank’s book value is an accounting number, and it is an accounting number about loans, which are worth what they are worth only if they are repaid. A bank trading at 0.98 times book with non-performing loans at five per cent and rising is not cheap; it is being told, by Mr Market, that the book value is wrong. Sometimes Mr Market is having one of his episodes. Sometimes he has read the loan book.

Distinguishing those two cases is the entire content of Chapter Forty-Five, and it cannot be done from a screen.

A live demonstration of why you check your data

While assembling that table I found one company reporting a price-to-book of 41.5, which would make it the most expensive security in the country by a factor of three.

Its computed book value per share was 18.8 rupees, against a par value of a hundred.

That is not a valuation. That is either a company that has lost eighty per cent of its paid-up capital, or a figure my extraction has taken from the wrong column, and until I have opened the filing and looked I do not know which. It has therefore been excluded from the analysis and flagged rather than reported.

This is Chapter Thirty-One’s plausibility gate operating in real time, on my own work, in the middle of a chapter. A number that should have stopped you is an error message, and the discipline is to treat an extraordinary result as a defect until proven otherwise, because it is a defect far more often than it is a discovery.

The verdict on Graham in Nepal

What survives, and should govern everything you do:

The margin of safety itself. Every valuation in Part Four produces a range, and no purchase is made near the top of it. Given the data quality established in Chapter Thirty-One — accounts five months late, scanned filings, inconsistent line items — a Nepali investor needs a wider buffer than Graham did, not a narrower one.

Mr Market. Chapter Nineteen showed that Nepali prices are driven by banking-system liquidity rather than by company performance. That is Mr Market with a mechanism attached, and knowing the mechanism does not make his quotation any more informative about value.

The insistence on the balance sheet. Graham’s instinct was that the balance sheet lies less than the income statement, because earnings can be managed in ways that assets resist. In a market where auditing standards vary and enforcement is light, that instinct is worth more here than where he formed it.

What does not survive:

Net-nets. Structurally unavailable in a financial-dominated exchange.

The full seven-criterion checklist. Two of the criteria do not apply to financial companies, one requires twenty years of history that this market’s consolidation has destroyed, and one requires data that exists for a minority of listings.

And what Graham himself would probably say about Nepal:

That the conditions here in 2026 resemble the conditions in New York in 1934 far more than those in 1976 — few analysts, poor disclosure, wide dispersion between price and value, and a marginal buyer who has not read anything. In that environment he did not recommend mechanical screens. He recommended the elaborate work, and he did it himself, and it paid.

The recantation was about a market that had filled up with people like him.

This one has not.

Graham’s largest single profit came from a stake in an insurance company he bought in 1948 — a position so large it breached his own diversification rules, taken partly because he liked the business and partly because the numbers were compelling. It subsequently grew to be worth more than everything else his partnership did across its entire life.

He recorded the fact in The Intelligent Investor with a dryness that has always struck me as the most honest sentence in the book: an irony, he called it, that a single fortunate decision counted for more than all the careful ones.

The man who founded the discipline of not being fooled by outcomes, describing his own best outcome, and refusing to take credit for it.

A Great Business at a Fair Price

In which the most successful investor alive describes his own most famous purchase as a two-hundred-billion-dollar mistake; a confectioner bought for twenty-five million dollars generates two billion without needing any more; and we test whether the Nepali market prices its banks according to how much they earn, and find that it half does.

Berkshire Hathaway was a textile business, and Warren Buffett has called buying it the worst investment decision he ever made.

He bought it in the 1960s because it was cheap in exactly the way Graham had taught — trading below its working capital, a classic cigar butt, the metaphor being that you pick up a discarded cigar butt off the street and get one free puff out of it. He intended a quick profit. He ended up in control, largely out of irritation at the management, and then spent the better part of two decades running a New England textile operation into an inevitable decline against foreign competition. The last mills closed in 1985.

In 2010 he estimated what that decision had cost. Because Berkshire became the vehicle for everything that followed, all of the capital deployed through it carried the drag of a failing textile business, and had he instead started with a clean insurance company the compounding would have been substantially larger.

His estimate of the cost was two hundred billion dollars.

The lesson he drew is the subject of this chapter, and it is the largest intellectual turn in the history of investing: a cheap price does not fix a bad business, because time works against you.

What Munger changed

Buffett has been consistent about who was responsible.

Charlie Munger, a Los Angeles lawyer turned investor, had never been a Graham disciple and did not accept the premise that everything reduced to the balance sheet. His argument — and Buffett has said it took a powerful force to move him off Graham’s more limiting views, and that the force was the power of Charlie’s mind — was that a business with genuinely superior economics is worth paying up for, because it does something a statistically cheap business cannot do: it compounds.

The line that came out of it is in the 1989 Berkshire letter, and it is quoted so often it has stopped being heard: it is far better to buy a wonderful company at a fair price than a fair company at a wonderful price.

The test case was See’s Candies, purchased in 1972 for twenty-five million dollars.

At the time it earned about four million a year pre-tax on roughly eight million of net tangible assets — a return on tangible capital of around fifty per cent. Graham’s framework had nothing to say about that number. Graham looked at assets; See’s had almost none worth mentioning. What it had was a brand in California that let it raise prices every December without losing customers.

Buffett reported some decades later that See’s had produced well over a billion dollars of pre-tax earnings, and — this is the part that matters — had required only about thirty-two million dollars of additional capital across the whole period to do it.

Twenty-five million in. One and a third billion out. Thirty-two million of incremental investment.

The arithmetic underneath it

The See’s story gets told as a parable about brands. It is really a story about a number, and the number is worth stating precisely because it governs Part Four.

Over a long holding period, your return converges toward the return the business earns on its own capital.

Munger put it plainly: over the long term it is hard for a share to earn a much better return than the business underlying it earns.

Here is why. Suppose you buy a company at a substantial discount to what it is worth. In year one, that discount is most of your return — the gap closes, and you are up. But the discount closes only once. From then on, your return is whatever the business produces, which is its return on equity, adjusted for how much of the earnings it can sensibly reinvest at that same return and how much it must hand back.

Run two companies for twenty years. One earns eight per cent on equity and you buy it at half of book. The other earns twenty per cent and you pay twice book.

In year one the first looks brilliant and the second looks reckless. By year twenty the first has compounded your capital at something close to eight per cent plus a one-off gain from the discount closing. The second has compounded at something close to twenty per cent, and the price you overpaid has become a rounding error against two decades of compounding.

The discount is a one-off. The return on capital is a rate. Rates beat one-offs over long horizons, and the crossover is a great deal sooner than most people expect.

This is the whole of the quality argument, and it has one enormous condition attached that gets dropped in the retelling.

The condition nobody mentions

The compounding only happens if the business can reinvest at its high rate.

A company earning twenty-five per cent on equity that can redeploy all its earnings at twenty-five per cent is an extraordinary machine. A company earning twenty-five per cent that has no use for the money and pays it all out as dividends is a bond with a good coupon — perfectly nice, but you must then find somewhere to put the dividends, and you will not find twenty-five per cent.

See’s was actually the second kind, and Buffett has said so. It could not absorb more capital; the candy business in California is a fixed size. What made it valuable was that it threw off cash which Berkshire could deploy elsewhere, and Berkshire had elsewhere.

So the quality question has two halves and you must ask both:

How much does it earn on the capital it employs?

And how much additional capital can it employ at anything like that rate?

A high return with no reinvestment runway is an income stream. A high return with a long runway is a compounding machine. They deserve very different prices, and confusing them is the most common error made by people who have absorbed the quality gospel without the arithmetic.

What a moat actually is

The reason a high return persists is that something prevents competitors from competing it away. Absent that, capitalism does its work: high returns attract entrants, entrants compete on price, and returns fall to the cost of capital. This is not a market failure; it is the market functioning.

The durable exceptions have been catalogued and the standard taxonomy has five entries.

Intangibles — brands, patents, and licences. A customer pays more for the label, or a regulator forbids a competitor from operating.

Switching costs — leaving is expensive, disruptive or frightening, so customers stay even when a better offer exists.

Network effects — the product becomes more valuable as more people use it, so the leader’s lead widens by itself.

Cost advantages — scale, location, or process that lets you produce at a price a competitor cannot match.

Efficient scale — a market large enough to support one or two operators profitably and no more, so nobody rational enters.

And the evidence that quality persists is respectable. The academic work on profitability as a return factor — gross profitability, and the broader quality-versus-junk work — has held up across markets and periods better than most published factors, which given Chapter Thirty-Two’s replication crisis is a meaningful endorsement.

So where are the moats in Nepal?

Now the local question, and it needs to be asked sector by sector, which is Part Four. But the shape of the answer is worth establishing here, because it is surprising.

Licences are the dominant moat in Nepal, and they are everywhere.

A commercial banking licence is not available on request. Nepal Rastra Bank has spent a decade reducing the number of institutions, which means the licence you hold is worth more than it was and cannot be replicated by a competitor with capital and ambition. The same is true of an insurance licence, a microfinance licence, and — most starkly — a hydropower generation licence with a signed power purchase agreement, which confers the right to sell a specific quantity of electricity at a contracted price for a defined number of years.

That is as clean a regulatory moat as exists anywhere in the world. Nobody can build a second plant on your stretch of river.

Efficient scale operates in the small sectors. Nepal has two listed trading companies and a handful of hotels for a reason: the addressable market supports few operators.

Brand exists, weakly, and mostly in manufacturing. A few consumer names have genuine pricing power built over decades.

Network effects are largely absent, which is unsurprising in an economy with a limited technology sector.

And switching costs in banking are lower than they look. Nepali depositors move for rate, and the base-rate disclosure regime makes comparison easy.

The important consequence is this. In Nepal, most moats are granted rather than built. They come from a licence, and a licence has two properties a brand does not: it has an expiry date, and it can be changed by the body that issued it.

Which means the durability question in Nepal is not “will competitors erode this?” It is ”what will the regulator do?” — and that is a different kind of analysis, with different sources, and Part Four treats it as a first-order input rather than a footnote.

Does the market pay for quality here?

This is testable, so let me test it.

If the market prices businesses by their economics, then price-to-book should track return on equity: a bank earning fifteen per cent on its capital should trade at a higher multiple of that capital than one earning five. The relationship is close to mechanical in theory — the justified multiple is a function of the return, the growth and the cost of equity.

Across the commercial banks in my fundamentals store, the correlation between price-to-book and reported return on equity is +0.40.

Positive, and much weaker than theory demands. Here is the underlying table.

Read the top and bottom rows together.

NMB earns the highest return on equity in the group and trades on the fourth-lowest multiple of book. NICA earns the lowest return, carries the highest bad loans, and trades at a higher multiple than NMB.

Meanwhile Standard Chartered, with much the lowest non-performing loan ratio, commands the highest multiple in the table — so the market is clearly paying for asset quality even while it appears to ignore returns.

The market is discriminating. It is just not discriminating on the variable a textbook would predict.

And the honest reason it might be right

Before concluding that this is an inefficiency to be harvested, I have to state the objection, and it is serious enough to reshape the rest of the book.

These are single-period returns on equity, and a single period’s return on equity is not the return on equity of the business.

A bank’s reported return in any given quarter contains the credit cycle. A bank that has just taken a large provision reports a poor return — and may be reporting it because it is recognising problems that its peers are still carrying. A bank that has not yet provided reports a fine return and has a worse loan book.

Look at NICA’s row again: return on equity of 1.6 per cent alongside non-performing loans of 9.53 per cent. Those two numbers are not independent. The low return is the high bad-loan ratio, arriving in the income statement as provisions. A market pricing that bank at 1.86 times book may be looking through a bad period at a franchise it expects to recover — which is exactly what an intelligent owner should do.

So the correlation of +0.40 does not straightforwardly show that Nepali investors are ignoring quality. It shows that reported return on equity is a poor measure of a bank’s earning power, and that both the market and my table are working with a contaminated number.

Which is the single most important methodological point in this book, and it is the reason Part Four does not begin with ratios:

You cannot value a bank on the return it printed last year. You must normalise it first — strip out the position in the credit cycle, form a view of what the franchise earns through a full cycle, decide how much of the current damage is permanent and how much is timing, and only then apply a multiple.

That process has a name and a procedure and it occupies Chapter Forty-Five. Everything in this chapter is the argument for why it is worth the trouble.

The verdict on quality investing in Nepal

What transfers:

The core arithmetic. Over a decade your return converges toward the business’s return on capital, and paying a fair price for a superior business beats paying a low price for a poor one. That is arithmetic and it is not country-specific.

The moat question, in its Nepali form. Ask what prevents a competitor from doing this, and expect the answer to be a licence, and then ask who issues the licence and what they might do next.

The two-part reinvestment test. High return plus a runway is a compounding machine; high return without one is an income stream. Nepali hydropower is frequently the second masquerading as the first, since a single plant on a single river has no reinvestment runway at all once it is built — a point Chapter Forty-Six takes up in full.

What needs adapting:

You cannot read quality off the reported numbers here, because the reported numbers are noisier, later and less normalised than in the markets where these methods were developed. The work is heavier.

And a Nepali moat has an expiry date on it. A generation licence runs out. A banking licence survives at the regulator’s discretion. Buffett could assume Coca-Cola would still be selling Coca-Cola in thirty years. You cannot assume the tariff.

Munger was asked, late in his life, what he would do differently.

He said that he and Buffett had spent their early years looking for cheap things and had learned, slowly and expensively, that the price you pay matters far less than the quality of what you buy — and that they had learned this mainly by owning a great many cheap things that stayed cheap.

Then he added the qualification that always gets left out, and which is the one that should stay with a Nepali reader.

Buying a great business at a fair price only works if you have correctly identified a great business.

Most people have not.

Scuttlebutt

In which a man with no research department outperforms for fifty years by asking competitors what they think of each other; a fund manager visits a hundred factories a year and finds what the accounts do not contain; and we ask what a Nepali investor can actually find out about a company by leaving the house.

Philip Fisher opened his investment counselling business in 1931, in San Francisco, in the third year of the Depression, on the reasoning that things were bad enough that people might be willing to try somebody new.

He had no research department and never acquired one. He ran a deliberately tiny practice, refused most clients, and held a very small number of companies for extremely long periods. He bought Motorola in 1955 and still held it when he died in 2004 — forty-nine years.

His book, Common Stocks and Uncommon Profits, appeared in 1958 and is the first serious treatment of a question Graham had largely ignored: what if the most important facts about a company are not in its financial statements?

Fisher’s answer was a method he called the scuttlebutt — a naval term for the gossip that circulates around the water barrel.

The method

It is disarmingly simple and almost nobody does it, which is what makes it valuable.

Ask the company’s competitors about it. Fisher found that executives would speak with startling candour about rivals — partly out of professional interest, partly because criticising a competitor is a socially permitted way of praising yourself, and partly because nobody had ever asked. Ask five competitors which firm in the industry they most respect, and you will get a consistent answer that no annual report contains.

Ask its suppliers and its customers. A supplier knows whether the company pays on time, and whether it is a demanding or a lazy buyer. A customer knows whether the product actually works and whether the service is real.

Ask former employees. Not for grievances, which are unreliable, but for texture: how decisions get made, whether the good people stay, whether the stated strategy resembles what happens in the building.

And then talk to management, last rather than first, having assembled enough independent detail to know whether what you are being told is true.

Fisher’s own checklist ran to fifteen points, and the majority of them are unanswerable from a filing. Does the company have products with sufficient market potential for years of growth? Is there a determination to continue developing products when current lines mature? How good are the research efforts relative to the size of the company? Are labour relations good? Are executive relations good? Is there depth of management? Does the company have a short-range or long-range outlook on profits?

And the one I regard as the sharpest question in all of investment literature:

Does the management talk freely to investors about its affairs when things are going badly?

Anybody will talk when results are good. The information is in what happens the other way.

Why this is not soft

There is a tendency to file Fisher under “qualitative,” meaning optional, and to regard the numbers as the real work.

That is backwards, and the reason is a point Chapter Thirty-Five established. If your return over a decade converges toward the return the business earns on its capital, then the central question is whether that return persists — and persistence is a question about competitive position, management behaviour and reinvestment opportunity, none of which appears in a financial statement.

The accounts tell you what happened. They are, necessarily, a report on the past, and in Nepal a report on a past that ended five months ago.

Fisher’s questions are about whether it will keep happening. That is not softer. It is simply harder to source, which is exactly why it is where an advantage might live.

Lynch’s version

Peter Lynch, whose own school gets the next chapter, ran Fisher’s method at industrial scale.

Managing Fidelity Magellan he visited several hundred companies a year, made thousands of calls, and — the detail that always gets told — famously took investment ideas from shopping trips with his family, on the reasoning that a product his wife and daughters were queueing for was a fact about the business that no analyst in New York possessed yet.

The formulation I would use for it is this: he was after information that was available to anybody and gathered by almost nobody. Lynch, who worked the same seam a generation later, put the same thought as a rule about effort — the person who turns over the most rocks wins the game.

Available to anybody and gathered by almost nobody. That is the important half of it, and it distinguishes scuttlebutt from something it is often confused with.

Scuttlebutt is not inside information. Inside information is material, non-public, and obtained from somebody with a duty to keep it confidential; acting on it is illegal in most jurisdictions and rightly so. Scuttlebutt is public information that requires effort to collect. The difference is not the sensitivity of the fact. It is whether anybody was obliged to keep it secret.

A supplier telling you that a company has stretched its payment terms from thirty days to ninety is not breaching a duty. He is complaining. The fact is available to anybody who asks a supplier, and nobody does.

Now: Nepal

Here is where this method becomes interesting, because Nepal is close to ideal for it and almost nobody exploits that.

The country is small. Everybody in Nepali business is roughly two introductions from everybody else. A determined investor in Kathmandu can, without any special access, reach somebody who has worked at almost any listed company.

The accounts are poor. Chapter Thirty-One established that fundamentals of usable quality exist for about fifty of a hundred and eighty-nine tradeable names, that filings arrive as scanned images with line items that change name between years, and that audited accounts may appear five months after the year ends. When the documents are weak, the non-documentary evidence is worth proportionately more.

And nobody is doing it. Chapter Thirty’s census found no sell-side research covering the smaller two thirds of the market. There is no analyst calling the company’s suppliers. The field is empty.

So what can you actually find out? Here is what is available to a private person in Nepal, in ascending order of effort.

Walk into the branches. For a bank, a development bank or a microfinance institution, the branch is the business. Go to three of them, in different districts if you can. Is anybody there? Are the staff idle or busy? How long is the queue? Is the branch in a location that suggests deliberate placement or an available lease? A bank whose branches are empty at eleven in the morning is telling you something about deposit gathering that its quarterly statement will report six months later.

Read the base rate disclosure and then ask a borrower. Nepali banks publish their base rate. What you cannot read is the premium actually charged to a real borrower, and any businessman you know can tell you what he is paying and by whom he was courted. A bank aggressively winning business by cutting the premium is buying growth, and the growth will appear in the loan book long before the consequences appear in the provisions.

Ask about the promoter. This is the single highest-value enquiry in Nepal, for the reason Chapter Thirty gave: nobody attends the annual general meeting, nobody votes, and governance is not enforced by shareholders. Whether the controlling family treats minority holders decently is therefore not a soft factor. It is frequently the entire difference between two banks with identical ratios. And it is knowable — reputations in Nepali business are well established and freely discussed, and the people who have dealt with a promoter will tell you.

For hydropower, go and look at the river. I mean this literally. A company’s prospectus describes a hydrology study. The people living upstream have watched that river for forty years. Ask them what a bad year looks like. Ask whether the intake silted. Ask whether the road washes out, and how often, and what that does to maintenance. Chapter Forty-Six will show that the difference between a plant’s contracted energy and its delivered energy is the honesty metric for the whole sector — and that difference has a physical cause which is visible from the bank of the river.

For manufacturing, count the trucks. Distribution is the constraint in Nepal, not production. A consumer goods company’s real asset is its dealer network, and the dealers will tell you which company’s van actually turns up.

And for anything with a single large customer, ask about payment. Nepal Electricity Authority is the sole buyer for every hydropower company on the exchange. Receivable days appear in the accounts eventually. The finance director of a plant will tell you now.

The discipline that makes it usable

Scuttlebutt has an obvious failure mode, which is that you collect a great deal of gossip and then believe whichever piece supports what you already thought.

Four rules keep it honest, and they follow directly from Part One.

Write down what you expect before you ask. If you believe a bank has a strong deposit franchise, write that down, then visit the branches. Otherwise you will interpret whatever you see as confirmation, which Chapter Eleven established you will do automatically and without noticing.

Ask about the negative case. Not “is this a good company” but “what would have to go wrong here.” People answer the second question far more informatively, because it does not require them to criticise anybody.

Count sources, and count them by independence. Three people at the same firm are one source. Chapter Six’s arithmetic applies: forty thousand correlated opinions contain ten, and five correlated informants contain one.

And separate what you were told from what you inferred. Keep them in different columns. A supplier said payment terms went from thirty to ninety days; I concluded the company has a liquidity problem. The first is evidence. The second is a hypothesis with a falsifier attached, and the falsifier is the next set of accounts.

The limit, stated honestly

I have been enthusiastic about a method, so let me mark its boundary.

Scuttlebutt tells you about a business. It does not tell you what to pay.

Fisher was famously indifferent to price. He argued that if a company’s prospects were genuinely outstanding, quibbling over a few percentage points of entry price was foolish, since the growth would swamp it. That is defensible for a company compounding at twenty per cent for thirty years and it is a catastrophe applied to anything else, and the Nifty Fifty investors of 1972 in Chapter Twenty found out which they had.

So the honest combination is the one this book keeps arriving at from different directions: Graham decides what you pay, Fisher decides what you buy, and neither is sufficient alone. Buffett has said as much — that he is part Graham and part Fisher, and that the proportions have shifted over his life.

For Nepal, given the data problems, I would weight Fisher higher than one would elsewhere and the price discipline higher still. Poor accounts mean you need more non-documentary evidence and a wider margin of safety, not one instead of the other.

What this looks like in practice

Concretely, for one company, before buying:

Read the last two annual reports and the last four quarterlies, and write down what you think the business is and what would break it.

Visit three points of contact with the real business — branches, dealers, the site, whatever the sector’s physical form is.

Speak to two people who have dealt with the company commercially and one who has worked there.

Ask everybody the same two questions: what would have to go wrong here, and who in this industry do you most respect and why.

Then read what you wrote at the start, and see whether it survived.

That is perhaps fifteen to twenty hours. Chapter Fifteen argued that a Nepali investor can genuinely know about ten companies. This is what the ten hours per company per year are actually spent on, and it is why the number is ten and not thirty.

Fisher stated his own answer to the question of how many good decisions an investor needs, and it was not a large number: I do not want a lot of good investments; I want a few outstanding ones.

By his son Kenneth’s account he meant it literally — the bulk of the money came from a handful of positions held for decades, and the rest of a fifty-year career was noise around them. The handful did not come from cleverness about the market. They came from having found, on a small number of occasions, a business almost nobody had bothered to look at properly.

He had bought Motorola in 1955 after visiting the company and concluding that its management thought further ahead than its competitors did.

That is the entire thesis. He held it for forty-nine years and it made him.

Know What You Own

In which the best mutual fund record in history is built by a man who says he never had a strategy; six kinds of company turn out to require six entirely different decisions; and we discover that the single most useful sorting exercise in investing takes about a minute and almost nobody performs it.

Peter Lynch ran Fidelity Magellan from 1977 to 1990 and compounded at about twenty-nine per cent a year, which remains the finest sustained public record in the history of mutual funds.

He has spent the decades since insisting that he had no system.

This is not false modesty and it is not quite true either. What Lynch had was not a formula but a sorting discipline, and it is the most immediately usable idea in this part of the book, because it costs nothing and it corrects an error that essentially every investor makes without noticing.

The error is treating all shares as the same kind of object.

Six kinds of company

Lynch’s argument, from One Up on Wall Street, is that a portfolio contains several different species and that applying one standard to all of them guarantees you will mishandle most.

Slow growers. Large, mature, expanding at or below the pace of the economy. You do not own these for growth. You own them for dividends, and the only questions that matter are whether the dividend is safe and whether you are paying too much for it.

Stalwarts. Big companies growing at a moderate, dependable rate. These are not going to multiply your money, but they do not collapse either, and their function in a portfolio is protection. Lynch’s rule was to take a reasonable gain and rotate rather than to hold forever, because the compounding is not fast enough to reward infinite patience.

Fast growers. Small, aggressive, expanding rapidly. This is where the enormous returns live, and where the enormous losses live, and the central question is not whether growth is happening — that is visible — but how long the runway is and whether the balance sheet can survive the expansion.

Cyclicals. Companies whose earnings rise and fall with an external cycle. These are the ones that destroy people, and the reason is precise: a cyclical looks cheapest at the top of its cycle and dearest at the bottom. At the peak, earnings are at a maximum and the price-to-earnings ratio is therefore at a minimum, and the screen says buy. At the trough, earnings have collapsed, the ratio is enormous or undefined, and the screen says sell.

Turnarounds. Damaged companies that may recover. Lynch owned some and made a great deal from them, and he was clear that they are a separate discipline with a separate failure rate.

And asset plays. Companies whose value sits in something the market has not noticed — land carried at cost, a stake in another business, cash.

Why the sorting matters more than the analysis

Here is the point that took me embarrassingly long to appreciate.

The same fact means opposite things in different categories.

A price-to-earnings ratio of six is a bargain in a stalwart and a warning in a cyclical. Rapid earnings growth is the thesis in a fast grower and a red flag in a slow grower, where it usually means the company has bought something or changed its accounting. A dividend cut is a catastrophe in a slow grower, held for exactly that dividend, and frequently good news in a turnaround, where retaining capital is the whole strategy.

So the first act of analysis is not to compute anything. It is to decide what kind of thing you are holding, because that decision determines which numbers are relevant and what they mean.

And Lynch’s harder observation is that companies change category without announcing it. A fast grower that has saturated its market is a stalwart, and the owner who continues to value it as a fast grower will pay far too much for a year or two. A stalwart that has been overtaken is a slow grower. A cyclical in a long upswing looks like a fast grower to anybody who arrived recently.

Most permanent losses I have seen in Nepal are of this shape: a company was bought as one thing and became another, and nobody re-sorted it.

Growth at a reasonable price

Lynch’s price discipline sits between Graham’s and Fisher’s, and it has a name.

He was willing to pay for growth, unlike Graham, but not any price, unlike Fisher. His working rule was to compare the price-to-earnings ratio with the growth rate, and to regard a company growing at twenty per cent and trading at twenty times earnings as fairly priced, one at twenty growing on ten times as attractive, and one at ten growing on twenty as expensive.

This ratio — earnings multiple divided by growth rate — is crude, and Lynch said so. It ignores the cost of capital, the durability of the growth, the capital required to produce it, and the quality of the earnings. As a formal valuation it is indefensible.

As a filter, it is superb, and the reason is that it forces the two numbers into the same sentence. An investor who says “this company is growing at thirty per cent” has made half an argument. An investor who says “this company is growing at thirty per cent and trades at seventy times earnings” has made a whole one, and can hear how it sounds.

Its one genuine analytical merit is that it makes explicit what you are assuming. Paying seventy times earnings for thirty per cent growth requires that growth to persist for a long time. How long? That is a question with an answer, and the answer is usually longer than any company has ever managed.

Know what you own

The phrase most associated with Lynch is his instruction to invest in what you know, and it has been mangled into something he did not say and would have disliked.

He did not mean: buy the shares of companies whose products you enjoy.

He meant: your familiarity with an industry is a source of information, and that information is worth something only if you then do the work. Noticing that a shop is always full is the beginning of research, not the end of it. Lynch’s own formulation was that the ability to describe, in a couple of sentences and in plain language, why you own something is a prerequisite — and that if you cannot, you do not own an investment, you own a hope.

I have adopted a harder version of this test and it has saved me money.

Explain the company to somebody who does not invest, in three sentences, without using the words growth, potential, or undervalued.

If the explanation contains no mechanism — no description of who pays this company money and why they will continue to — then there is nothing there. Try it on the last thing you bought. The failure rate is startling and it is highest on the positions people are most confident about.

Now, Nepal, category by category

Lynch’s taxonomy maps onto this exchange better than any framework in this part, and the mapping is not obvious. Let me do it properly, because it reorganises how the sector chapters in Part Four should be read.

Slow growers. Nepal Telecom is the type specimen — large, dominant, growing slowly, generating cash. The bigger commercial banks approach this in the mature phase.

Stalwarts. The established commercial banks. Nabil, Standard Chartered, Everest. Growing with the economy and the credit cycle, unlikely to disappear, unlikely to multiply.

Fast growers. Almost none, and this is a genuinely important observation about the Nepali market that I have not seen stated. A fast grower requires an expanding addressable market and the ability to reinvest at high rates. Microfinance was one, for about a decade, until the regulator capped the spread. Life insurance has been one, as penetration rose from very low levels. But the structural feature of this exchange is that its two largest sectors — commercial banking and hydropower — cannot be fast growers, one because it is bounded by deposit growth in a small economy and the other because a plant is a plant.

Cyclicals — and this is the crucial one. Nepali banks are cyclicals wearing the clothes of stalwarts. Their earnings are a leveraged function of the credit cycle, which is itself a function of the liquidity cycle traced in Chapter Twenty-One. And Lynch’s warning applies with full force: at the top of the credit cycle a Nepali bank reports strong earnings and a low multiple, and looks cheap on every screen, and it is not. At the bottom it reports collapsed earnings after heavy provisions and looks expensive, and frequently is not.

This is the single most expensive misclassification available in this market, and it is why Chapter Forty-Five spends most of its length on normalising a bank’s earnings before applying any multiple at all.

Turnarounds. Nepal has had many, mostly involuntary, and Chapter Twenty-Nine established what happened to the acquirers who bought them.

Asset plays. The most under-explored corner here. Hotels sitting on land carried at historical cost in a valley where land has multiplied. Investment companies holding stakes in listed businesses. The closed-end mutual funds of Chapter Thirty, trading at discounts to a published net asset value with a fixed maturity date — which is as close to a textbook asset play as this market offers.

Hydropower has its own category, and Lynch does not have a word for it

One Nepali sector refuses the taxonomy, and naming the reason is useful.

A hydropower company passes through three distinct lives.

Before commercial operation it is a construction project with no revenue. Any reported profit is interest on unspent share capital, as Chapter Twelve showed. It is not a slow grower or a fast grower; it is not a business at all yet, and no earnings multiple computed on it means anything.

After commercial operation and while the debt is being repaid, it is a leveraged bond. Revenue is contracted, costs are largely fixed, and almost all the operating cash goes to the lenders. Earnings per share are small and rising for reasons that have nothing to do with the plant improving.

After the debt is repaid, it becomes what it always was going to be: an annuity with an expiry date, throwing off cash, with a finite number of years left on its licence and no ability to reinvest.

Three businesses, one ticker, and a valuation method that must change twice during your ownership. Chapter Forty-Six is built entirely around this.

The verdict on Lynch in Nepal

What transfers, and should be the first thing you do with any company:

The sorting. Before any calculation, decide which of the six kinds you are looking at, write it down, and re-check it annually. The whole exercise takes a minute and it determines which of the next forty minutes are worth spending.

The plain-language test. If you cannot describe the mechanism by which the company gets paid, you have not finished.

And the cyclical warning, which in Nepal is not one category among six but the dominant fact about the largest sector on the exchange.

What needs adapting:

The growth-versus-multiple rule is less useful here than elsewhere, because Nepali earnings are noisier, later and more cycle-contaminated than the American earnings Lynch was working with. Use it as a conversation-starter, never as a verdict.

And “invest in what you know” has a specific Nepali hazard. What most Nepalis know is banking, because everybody has a bank account and everybody’s cousin works at one. That familiarity produces comfort rather than insight, and comfort in the largest and most correlated sector on the exchange is precisely the exposure Chapter Twenty-One warned about. Familiarity is a starting point for research, not a substitute for diversification.

Lynch retired from Magellan in 1990 at forty-six, at the top, having compounded at twenty-nine per cent for thirteen years.

He has said since that the job had taken his Saturdays and then the rest of the week with them, and that he stepped down at the age his own father had died, having decided the trade was not worth it. He was not tired of the work. He had priced it.

The finest record in the industry ended because the man who produced it decided it was not worth what it cost, which is a piece of investment advice that no amount of compounding can improve on.

The Formula

In which a hedge fund manager publishes the recipe for beating the market and then watches his own clients fail to follow it; nine yes-or-no questions add seven and a half points a year; and I run four mechanical value signals across eighteen quarters of Nepali data and find that none of them survives, for a reason that no amount of extra data will fix.

Joel Greenblatt ran a hedge fund that compounded at something over forty per cent a year for two decades, and then in 2005 he published a short book explaining exactly how to do it, in language aimed at a teenager.

The Little Book That Beats the Market gives away a formula, and the formula has two ingredients.

Earnings yield: the company’s operating earnings divided by what it would cost to buy the whole business including its debt. In other words, how much you get for what you pay. This is Graham’s half.

Return on capital: the same operating earnings divided by the capital actually tied up in the business — working capital plus fixed assets. How good a business is it. This is Buffett’s half.

Rank every company on each measure, add the two ranks together, buy the thirty at the top, hold for a year, repeat.

That is the entire method. Greenblatt reported that it had returned around thirty per cent a year over the seventeen years he tested, against about twelve for the market.

He called it magic partly as a joke and partly, I think, because he expected nobody to believe that something so simple could work.

What happened next is the interesting part

Greenblatt subsequently set up a business offering the strategy to ordinary investors, and he did something unusual: he offered two versions.

In the first, the client received the list of qualifying companies and chose which of them to own — he could skip the ones he disliked, hold cash when nervous, and time his entries.

In the second, the account traded the formula automatically. No discretion at all.

Over the period he later reported on, the automated accounts substantially outperformed the self-managed ones — and, more damningly, the self-managed accounts underperformed the index while the automated ones beat it.

The clients had the formula. They had paid for it, they believed in it, and they had read the book. And exercising judgment over it destroyed the entire advantage and then some.

Greenblatt’s own diagnosis, which he has repeated in interviews since, is that the self-managers systematically skipped the most frightening names — which are precisely the ones the formula is designed to surface — and stopped following it during drawdowns, which is exactly when a contrarian strategy does its work.

A strategy and the use of a strategy are different objects, and the second is where returns are actually determined. This is Chapter One’s investor-versus-fund gap arriving in its purest experimental form: same method, same period, same market, one variable changed, and the variable is the human being.

Piotroski’s nine questions

The other great mechanisation is narrower and, I think, better constructed.

Joseph Piotroski, an accounting academic, published a study in 2000 asking a simple question. Value portfolios — cheap companies by book-to-market — outperform on average, but the average conceals enormous dispersion: a lot of cheap companies are cheap because they are dying.

Could you separate them using nothing but the financial statements?

His answer was nine binary tests, each worth one point. Is net income positive? Is operating cash flow positive? Is operating cash flow greater than net income — that is, are the earnings backed by cash? Is the return on assets improving? Is leverage falling? Is the current ratio improving? Has the company avoided issuing new shares? Is the gross margin improving? Is asset turnover improving?

Score nine. Buy the high scorers among cheap companies, avoid the low.

Piotroski found the exercise shifted returns by something like seven and a half percentage points a year, and — more usefully — that the improvement came disproportionately from avoiding the disasters rather than from finding the winners.

I like this construction more than the Magic Formula because of what it is made of. Every one of the nine questions is about financial health rather than about price, they use only the statements, and several of them — cash flow exceeding net income, no new shares issued, falling leverage — are direct tests of whether the reported earnings are real.

In a market with weak auditing, that is exactly the right family of questions.

Why mechanisation works, and then stops

Formulas have a genuine advantage and it is not analytical. They remove you.

Every failure mode in Part One — anchoring on your purchase price, the disposition effect, the social cost of buying the frightening thing, resulting, narrative bias — is a failure of the human operating the method. A rule executed without discretion is immune to all of them, which is why Greenblatt’s robots beat his clients.

And then they decay. Chapter Thirty-Two gave the measurement: published anomalies deliver about twenty-six per cent less out of sample and fifty-eight per cent less after publication. The Magic Formula’s realised performance since the book has been considerably more modest than the backtest, which is the ordinary fate of a well-publicised rule, and Greenblatt has been straightforward about it.

Two mechanisms drive the decay, and only one is competition. The other is that a result selected for being large was selected partly for being lucky, and the luck does not repeat.

So: can you run a formula in Nepal?

I wanted to know, so I tested it properly, and the study is the most rigorous thing in this book that produced a negative result.

The setup. Eighteen quarterly cross-sections from early 2022 to mid-2026. Median breadth of forty companies per cross-section. Every fundamental gated through statutory disclosure dates so that nothing was used before it was legally public — the point-in-time discipline of Chapter Thirty-One, without which the whole exercise is worthless.

I measured each signal’s information coefficient: the rank correlation between the signal at the start of a quarter and the returns that followed. A positive coefficient means the signal ranked the winners ahead of the losers.

And I included a placebo — a seeded random score, ranked alongside the real signals, which should produce nothing.

Applying a false-discovery correction across the four real signals: no survivors.

Three things this says

One: the placebo behaved. This is what licenses reading anything else on the page. A random score returned p = 0.73 and ranked nothing. Had the placebo scored, every other number in the table would have been meaningless, and I would have had a broken test rather than a result. I recommend building one into any study you ever run: it costs nothing and it is the only check that catches a whole family of errors at once.

Two: book value ranks returns and profitability does not. Book yield’s coefficient is double earnings yield’s, and plain return on equity is indistinguishable from the random number — negative 0.014 against the placebo’s positive 0.017.

In a market this dominated by financial institutions, where price-to-book is the standard yardstick, that is coherent rather than surprising. But note what it means for anybody importing a quality-plus-value screen from abroad: in Nepal, the quality half contributes nothing measurable. The whole of the (weak, unproven) signal is in the cheapness half.

Three, and this is the finding that matters: more companies will not fix it.

A period’s information coefficient is noisy for two separable reasons. Some of the noise comes from measuring across a finite number of companies. The rest comes from the signal genuinely working better in some quarters than others.

Only the first shrinks when you add names.

I decomposed it, and roughly four fifths of the variance is real time-variation. Doubling the universe from forty companies to eighty would cut the total standard deviation from 0.383 to about 0.367 — a four per cent improvement in precision. Essentially nothing.

What would settle the question is not more companies. It is more quarters — about thirty-eight cross-sections against the eighteen available, which is roughly nine and a half years of overlapping fundamentals and prices.

That cannot be manufactured. The financial statements simply do not exist further back in usable form, for the reasons Chapter Thirty-One set out.

So the honest verdict is: not disproven, not proven, and not resolvable with today’s data. Book yield may well work in Nepal. A mean coefficient of +0.124 would be strong in a developed market, where equity factor coefficients of 0.02 to 0.05 are normal. It is also entirely consistent with luck across eighteen periods, and the confidence interval straddles zero.

I record it as an open question with a date on it, and I re-run it annually, and one day there will be an answer. That is a less satisfying thing to write than a strategy, and it is what the data supports.

The structural obstacles, on top of the statistical one

Even if the signal were proven, three features of this market would obstruct running a formula.

The Magic Formula is not computable here. Its earnings yield requires enterprise value, which requires shares outstanding and debt as a reliable time series. Chapter Thirty-One established that shares outstanding are not available. And its return on capital — operating earnings over working capital plus fixed assets — is meaningless for a bank, which is more than half of this exchange.

Piotroski is partly computable and expensively so. Several of the nine tests apply to financial companies awkwardly, and assembling two consecutive years of clean statements for forty companies is a project rather than a screen.

And the fee schedule forbids the portfolio. Greenblatt’s method holds thirty companies, rebalanced annually. Chapter Twenty-Four’s table says that a thirty-name portfolio is unworkable in Nepal below roughly five million rupees, and that at two lakh across twenty names a position must drift sixty-seven per cent before rebalancing is economic. The formula requires a portfolio the charge structure will not permit.

What to take from the formulas anyway

I have just spent a chapter dismantling mechanisation in Nepal, so let me salvage the parts that are genuinely usable, because they are the most valuable in Part Three.

Use the formula as a checklist rather than a portfolio. Piotroski’s nine questions cost nothing to ask about a single company you are already considering. Is operating cash flow positive and greater than net income? Has the company issued new shares? Is leverage rising? Those three alone would have kept a Nepali investor out of a considerable amount of trouble, and none requires a screen or a universe.

Take the anti-discretion lesson seriously. Greenblatt’s clients failed at the same task their own robots succeeded at. Whatever method you settle on, the value lies in executing it when it feels wrong, and the mechanism for that is not willpower — it is the written falsifier from Chapter Four and the pre-committed conditions from Chapter Fifteen.

And notice what the negative result implies about where the edge is. If mechanical signals cannot be shown to rank Nepali returns — and if, per Chapter Thirty-Two, no price-based strategy beat the benchmark either — then the remaining candidate is the one that cannot be tested by any of these methods: understanding an individual business better than the person on the other side of the trade.

That is not a satisfying conclusion for anybody who wanted a rule. It is where four separate lines of investigation in this book have now arrived, and it is what Part Four is for.

Greenblatt was asked why he had published the formula at all, given that publication is the standard way to destroy an advantage.

He said he had two answers. The first was that he did not think many people would follow it, because following it requires doing badly for two or three years at a stretch and most people cannot.

The second was that his own children were going to need to invest one day, and he wanted something written down.

Then his clients were handed the formula and demonstrated the first answer.

CAN SLIM Meets the Circuit

In which a Los Angeles broker studies every big winner of the previous half century and finds they had seven things in common; a method built entirely on buying strength arrives in a market where strength cannot be bought; and we work out which of its seven letters survive the journey.

William O’Neil bought a seat on the New York Stock Exchange at thirty, having turned a small stake into enough to pay for it, and spent the following decades doing something nobody else had bothered to do systematically: studying the winners.

Not the market. Not a theory. He assembled the characteristics of the biggest stock market winners of the preceding fifty years — the companies that had gone up many times over — and asked what they had shared before the move began.

The answer became an acronym, and the acronym became How to Make Money in Stocks, and whatever you think of the method it is one of the very few in this book derived inductively from a large sample rather than deductively from a principle.

C — Current quarterly earnings. Big winners showed sharp acceleration in the most recent quarter, typically twenty-five per cent or more against the same quarter a year earlier. Not steady growth: acceleration.

A — Annual earnings growth. Backed by several years of growth, so the quarter is a continuation rather than a fluke.

N — Something new. A new product, a new management, a new industry condition, or a new high in the share price. O’Neil regarded this as essential: something has changed, and the change is why the earnings are moving.

S — Supply and demand. Fewer shares outstanding means a given amount of buying moves the price further. He favoured companies with modest share counts and watched volume closely.

L — Leader, not laggard. Buy the best company in a strong industry, not the cheap one nearby. His relative-strength measure ranked every stock against every other, and he wanted the top decile.

I — Institutional sponsorship. Some institutions accumulating, but not too many. You want the buying still ahead of you, not behind.

M — Market direction. Three quarters of stocks follow the general market, so do not buy anything when the market is in a downtrend.

What it really is

Strip the letters away and CAN SLIM is a momentum strategy with an earnings filter and a market-timing overlay.

That is not a criticism. It is a description, and it is worth being precise about it because it determines what evidence bears on the method.

The earnings components — C and A — are fundamental, and there is respectable academic support for something adjacent to them: post-earnings-announcement drift, the tendency of prices to continue moving in the direction of an earnings surprise for weeks afterwards, is among the more durable anomalies in the literature.

The relative-strength component — L, and much of N — is price momentum, which is the most extensively documented factor in finance and which has worked in most markets over most long periods.

The M component is market timing, which the evidence supports considerably less well.

So O’Neil assembled, inductively, a combination that the academic literature would mostly endorse two decades later. That is a genuine achievement, and it makes what follows more interesting rather than less.

The method requires buying strength

Here is the operational core, and everything in the second half of this chapter turns on it.

CAN SLIM does not buy dips. O’Neil was emphatic and repeated the point throughout his career: the method buys a stock as it breaks out of a consolidation pattern to a new high, on volume well above normal, and it buys within a narrow window — a few per cent above the breakout point — because the whole logic is that you are joining a move at its inception.

And it exits on a hard rule: sell at a fixed loss, conventionally seven or eight per cent below the purchase price, no exceptions and no reasoning.

That combination — buy strength in a tight window, cut losses at a fixed threshold — is what makes the method coherent. The small losses are affordable because the winners are supposed to be very large.

Now recall what Chapter Twenty-Five measured.

The collision

Nepal has a daily price limit. Ten per cent for essentially all of the relevant history, fifteen since April 2026. And I measured, across 352,236 daily bars, what happens at and after the limit.

Closes at the upper limit: 1.11 per cent of all bars — six times more frequent than closes at the lower limit. Buying pressure concentrates because anybody with money can supply it; selling pressure disperses because the only potential sellers are existing holders and, per Chapter Eleven, existing holders do not sell.

The day after a limit-up close, the stock rises another 3.17 per cent on average, and has a 30.8 per cent chance of locking limit-up again.

Now put a CAN SLIM practitioner into that market.

His system fires on a breakout to a new high on heavy volume. In Nepal, a genuine breakout on heavy demand in a thin float locks at the limit, which by definition means there are buyers and no sellers. His order does not execute; it queues behind everybody else whose system fired that morning for the same reason.

The next day the stock is on average three per cent further away, and one time in three it locks again and remains unbuyable.

The signal is real, the move is real, and the trade is unavailable.

And the days he does get filled on a breakout are the days when the demand was weak enough for a seller to remain — which is to say, disproportionately the false breakouts. His executed sample is adversely selected by the market’s own mechanics.

And the stop-loss does not work either

The other half of the method fails for a related reason, and Chapters Twenty-Three and Twenty-Five together explain it.

O’Neil’s seven-per-cent stop assumes you can sell at approximately your stop price.

In Nepal you cannot. A stock falling hard may close at its lower limit, in which case your sell order joins a queue and does not execute. When it does execute, it executes at whatever the queue delivers. And then — Chapter Twenty-Three — the proceeds are not available for two further sessions.

So a stop set at minus seven per cent is realised somewhere below that, on a day you did not choose, with cash arriving two days later. The tight loss control that makes the whole risk-reward arithmetic work is not implementable.

I should be fair and note the mitigation Chapter Twenty-Five found: limit-down closes are rare — 0.18 per cent of bars — and they do not cluster, with only an 11.9 per cent chance of a second one. The multi-day trapdoor people fear is not in the historical record. So the stop is degraded rather than destroyed.

But degraded is enough. A method whose edge depends on cutting losses at seven per cent does not survive cutting them at an uncontrolled number somewhere below.

The letters that do survive

Having dismantled the execution, let me be scrupulous about what remains, because two of the seven letters are genuinely valuable in Nepal and one of them is undervalued.

C — current quarterly earnings acceleration — survives, with a caveat. Nepali quarterly results are published within thirty days of quarter end under Schedule 14, and the market’s reaction to them is slow, because there is no analyst community to disseminate them. A shareholder who actually reads the quarterly the week it appears is ahead of most of the market.

The caveat is severe and specific: Nepali quarterly figures are usually cumulative, not discrete. A third-quarter statement typically reports nine months. Comparing reported quarter to reported quarter without de-cumulating first produces a series that looks smoothly rising because it is an accumulation, and any growth measure computed on it is an artefact. I have made this error myself and it took a re-derivation to catch.

A — annual earnings growth — survives and is the workhorse of Part Four.

N — something new — survives, and it is the most under-used idea in this chapter for Nepal. O’Neil’s insight was that big moves have a cause, and that the cause is usually identifiable and recent. In Nepal the “something new” is very often regulatory: a change in the spread cap for microfinance, a revision to capital requirements, a new tariff regime, the approval of a power purchase agreement, a change in the credit-to-deposit rule. These are published, dated, and their consequences are traceable through a company’s economics. That is a genuinely exploitable category of event and it requires no chart.

S — supply and demand — survives in transformed form. O’Neil wanted small share counts. Nepal gives you something better and stranger: Chapter Eighteen’s promoter and public split, in which fifty-one per cent is locked by construction, and Chapter Twenty-Seven’s corporate actions, which change the share count on announced dates. A right issue is a supply event with a known date, and its effect on the price is mechanical. That is far more tractable than inferring demand from volume.

L — leadership — does not survive as a relative-strength rule, because acting on it requires buying strength, which the circuit forbids.

I — institutional sponsorship — does not survive. Chapter Thirty’s census: there are essentially no institutions to track, no disclosure regime for their positions, and the marginal buyer is a private individual.

M — market direction — does not survive as a timing rule, for the reason Chapter Nineteen established: the Nepali cycle turns roughly four times in a decade, which gives an effective sample of about four, which cannot be distinguished from luck by any test.

The verdict

CAN SLIM as a system cannot be run in Nepal. Its entry mechanism is blocked by the price limit, its exit mechanism is blocked by the limit and the settlement cycle, and two of its seven components have no local data.

This is not a judgment about whether momentum works. It is narrower and harder to argue with: the trades the system generates cannot be executed at the prices the system requires. A strategy whose signal and whose fill are systematically separated by three per cent, in the direction against you, on the occasions it is most confident, has had its edge removed by the market’s plumbing before any question of merit arises.

What to take instead: O’Neil’s method, not his system. He studied a large sample of winners and asked what they had in common in advance. That is a superb research instinct and almost nobody in Nepal has applied it locally.

The Nepali version of that question is open and I have partly answered it in this book — the sector table in Chapter Twelve, the merger study in Chapter Twenty-Nine, the concentration ladder in Chapter One are all instances of it. It is available to anybody with the price panel and the patience, and the answers are not in any book because nobody has looked.

A warning about who is selling this

I would not spend a chapter on CAN SLIM if it were merely inapplicable. I am spending one because it is taught in Nepal, in paid courses, as a method for this market.

The technical analysis certificate on Bikash’s wall in Chapter Five sits on a curriculum that includes breakout trading, relative strength, and stop-losses, imported essentially unmodified from American material.

Not one of those three survives contact with a ten per cent daily limit, T+2 settlement, and a flat depository fee. The people teaching it are, in my experience, sincere. They have read a good book and are transmitting it faithfully.

Faithful transmission of a method whose preconditions are absent is the most common form of financial harm in this country, and it is not fraud, which makes it harder to object to.

The defence is the question this whole part of the book is organised around, and it takes one sentence: what does this method require in order to work, and does this market provide it?

O’Neil’s own funds did not, over their full lives, reproduce the returns his research described, and his firm’s public performance was a subject of periodic argument.

He was consistent in his response, which was that the method required following it exactly, and that almost nobody did — that people bought outside the window, held past the stop, and ignored the market rule.

Greenblatt’s clients said the same thing in the last chapter, from the other side of the table.

Two men, two methods, two decades apart, arriving at the same complaint: the strategy was fine, the users were the problem.

They may both be right. It is also worth noticing that a method which only works when executed perfectly by people who never execute it perfectly has a defect, and that the defect is not entirely the users’.

QGLP, and the School That Fits Us Best

In which an Indian broker begins publishing an annual study of who actually created wealth and keeps doing it for thirty years; two Indian private banks founded a decade apart end up in very different places; the regulator publishes the warning three years early and nobody reads it; and we look for the same warning in Nepal.

Every method in this part of the book so far was built in America, for American companies, by people working in the deepest and most heavily analysed capital market that has ever existed.

Nepal is not that. Nepal is a market with promoter-controlled companies, family ownership, weak minority protection, developing disclosure standards, a dominant banking sector, heavy retail participation, and a regulator that shapes industries by decree.

There is a country next door with all of those characteristics and forty years more history, and its investors have built frameworks for exactly these conditions. It is strange to me that Nepali investors read Buffett and O’Neil and almost never read the Indian practitioners, whose problems are our problems.

Thirty years of asking one question

In 1996 Raamdeo Agrawal, co-founder of Motilal Oswal, published the first of what became an annual Wealth Creation Study: a systematic examination of which Indian companies had created the most shareholder wealth over the preceding five years, and what they had in common.

He has published one every year since. It is, so far as I know, the longest-running continuous body of empirical work on an emerging equity market conducted by practitioners, and it is free.

The method is O’Neil’s — study the winners, inductively, from a large sample — applied to a market that structurally resembles ours far more than America’s does.

The findings have been remarkably stable across three decades, and three of them matter here.

Price follows earnings, over five years and longer. The correlation between shareholder return and earnings growth is weak over one year, moderate over three, and strong over five and ten. This is the single most useful empirical fact an Indian or Nepali investor can hold, because it tells you what to watch and what to ignore.

The biggest wealth creators sustained high returns on equity rather than achieving spectacular one-off growth. Longevity of return beat magnitude of growth.

And the majority of wealth creation was concentrated in a small number of names. The distribution is extremely skewed, which is Chapter Thirteen’s concentration arithmetic appearing again — and which explains why an investor who trims his winners on a rule systematically removes his own results.

The framework

Out of that work came the acronym Motilal Oswal built its equity philosophy on, and it is the tidiest summary of the value-plus-quality tradition I know.

Q — Quality, of two kinds. Quality of business: does it earn a high return on the capital it employs, and is there something protecting that return? Quality of management: is capital allocated sensibly, and are minority shareholders treated as owners or as an inconvenience?

G — Growth in earnings. Not revenue, not assets, not the share price. Earnings.

L — Longevity of both the quality and the growth. This is the letter that does the work, and it is the one most frameworks leave out. A high return that lasts three years is worth a fraction of the same return lasting fifteen, and the difference between them is not visible in any current-year number.

P — Price. A favourable valuation, which reintroduces Graham and prevents the framework from becoming an argument for paying anything for quality.

Their slogan is Buy Right, Sit Tight, which compresses the whole of Chapter Fourteen into four words.

Why I prefer this to the American formulations, for our purposes: it makes management quality an explicit first-order input rather than a footnote. In a market of promoter-controlled companies where minority shareholders have no practical influence — Chapter Thirty’s finding that nobody attends the annual general meeting and the resolutions pass — how the controlling family behaves is not a soft consideration. It is frequently the whole difference between two companies with identical ratios.

Two banks

The clearest demonstration of Q and L in a market like ours is a comparison every Nepali banking investor should know and almost none does.

HDFC Bank was founded in 1994. It grew steadily rather than spectacularly, kept its gross non-performing assets at around one per cent through multiple Indian credit cycles, earned a return on assets consistently near two per cent, and compounded shareholder wealth at roughly twenty per cent a year over the twenty-five years to 2020, though the rate has come down markedly since. Its underwriting was regarded, at various points, as too conservative — it declined business that competitors took.

Yes Bank was founded in 2004. It grew loans at rates far above the banking system’s for a decade. Its reported return on equity was excellent, at times better than HDFC’s. Its reported non-performing assets were low. By 2018 it was among the most admired private banks in India and its shares had multiplied many times over.

The share price closed at its all-time high of about ₹394 in August 2018.

In March 2020 the Reserve Bank of India placed the bank under a moratorium, capped depositor withdrawals, and orchestrated a rescue led by State Bank of India. Additional Tier 1 bonds with a face value of about ₹8,415 crore were written down to zero — a detail worth pausing on, because those instruments had been sold to retail investors as safe. Equity holders were left with a share worth a small fraction of its peak.

Two private banks. Same country, same regulator, same credit cycle, same customers. One compounded for twenty-five years and one did not survive sixteen.

The warning was published three years early

Here is the part that makes this a lesson rather than a story.

The Reserve Bank of India requires banks to disclose divergence — the difference between the bad loans a bank has recognised and the bad loans the regulator’s own inspection found. When the regulator’s number materially exceeds the bank’s, the bank must publish the gap.

Yes Bank disclosed a divergence of roughly ₹4,176 crore in gross non-performing assets for one financial year, and about ₹6,355 crore for the next.

That is a public, audited, mandatory disclosure stating, in effect: our regulator examined our loan book and found thousands of crores of bad loans that we had not recognised.

It was published years before the collapse. It appeared in the annual report. Anybody could read it.

The reported ratios were excellent and the divergence disclosure said the reported ratios were wrong. An investor who read the second and ignored the first had three years’ notice.

This is the general lesson and it is worth stating in the abstract, because the specific Indian disclosure has no exact Nepali twin:

When a company’s own reported quality metrics conflict with an external assessment of the same thing, the external assessment is the information. The reported number is what management chose. The divergence is what somebody else found.

The Nepali version of the same question

Nepal does not publish divergence in the Indian form. So what is available?

The non-performing loan trajectory, not the level. A rising ratio is worth far more than a high one, because the level reflects history and the trajectory reflects what is happening now.

Loan growth relative to the system. A bank growing its book materially faster than the banking sector is winning business somebody else declined, and there are only three reasons that happens: it is better, it is cheaper, or it is less careful. The first is rare.

The provision coverage ratio, which my fundamentals store carries, and which tells you how much has been set aside against the bad loans already recognised. A bank with a rising bad-loan ratio and falling coverage is deferring recognition.

And the gap between the unaudited full-year figures published in mid-August and the audited accounts published around December, which Chapter Twenty-Eight flagged as one of the few genuinely diagnostic signals freely available in this market. When those differ materially, the difference is information about the company’s reporting, and it is the nearest thing Nepal has to a divergence disclosure.

Let me look at what the first two show.

What the Nepali numbers say

From my fundamentals store, for commercial banks with at least two fiscal years of comparable loan data. The windows differ between banks, because coverage in the store begins at different points — so this is illustrative rather than a controlled comparison, and I would not rank banks on it. Read it for the shape.

Two banks anchor the ends of this table and both are instructive.

Standard Chartered grew its loan book by under ten per cent across five years and has the second-lowest bad-loan ratio in the sector. It has been criticised in Nepal for years for exactly this — for declining growth, for being too conservative, for sitting on capital. Chapter Thirty-Five’s table showed it commands the highest price-to-book multiple of any commercial bank at 2.98. The market is paying for the restraint.

Kumari Bank grew its book seventy per cent in three years and its non-performing ratio went from 1.11 to 6.95 — a rise of nearly six percentage points. That is the pattern in the Yes Bank shape, visible in published quarterly statements, in a Nepali bank, now.

And Everest Bank is the honest complication that stops this becoming a rule: it grew loans ninety-six per cent and its bad-loan ratio is 0.61 per cent, the lowest in the sector. Fast growth with a clean book. So growth alone is not the signal.

The signal is growth combined with deteriorating asset quality. Either alone proves nothing. Together they are the thing to be afraid of, and the arithmetic of why is simple: loans written in year one become non-performing in years three to five, so a bank growing fast today has a bad-loan problem that has not arrived yet. The NPL ratio of a fast-growing bank is flattered by its own denominator.

Where the Indian school helps and where it does not

What transfers directly:

The QGLP structure, and especially the L. In Nepal, longevity is mostly a question about a licence and a regulator, per Chapter Thirty-Five — and the QGLP framework is built for markets where that is true, which is why it accommodates it more naturally than the American frameworks do.

Management quality as a first-order input. In a promoter-dominated market with no shareholder enforcement, this is not optional.

The five-year earnings horizon. Agrawal’s repeated finding — that price follows earnings over five years and is noise over one — is the empirical justification for everything Chapter Fourteen argued about patience.

And the Wealth Creation Study method. Nobody has done this for Nepal. Studying, over a decade, which Nepali companies actually created wealth and what they had in common beforehand is an entirely doable project with the data described in Chapter Thirty-One, and it has never been published.

What does not transfer:

India has depth Nepal lacks. Its wealth creation studies draw on hundreds of companies across dozens of industries; Nepal has thirteen sectors of which two dominate. The skewness finding — a handful of names producing most of the wealth — is harder to exploit when the universe is a hundred and fifty names and a third of them are the same hydropower business.

And India’s disclosure regime is materially better than ours. The divergence disclosure that warned about Yes Bank has no Nepali equivalent. What we have instead is the audited-versus-unaudited gap, the NPL trajectory, and Chapter Thirty-Six’s scuttlebutt — which is weaker evidence, requiring more work, and is precisely why the margin of safety here must be wider than in Mumbai.

Rana Kapoor, who built Yes Bank, gave interviews through the years of rapid growth in which he described the bank’s risk management as its core strength, and its technology-driven underwriting as the reason its asset quality was superior to its competitors’.

He was, at the time, one of the most respected bankers in India, and there was no particular reason for anybody to disbelieve him.

The divergence disclosure was in the annual report the whole time, a few pages further in, in a table.

What the Quants Actually Give You

In which two Nobel laureates and the finest bond traders on earth are destroyed by a position they had measured correctly; a physicist’s fund compounds at sixty-six per cent for thirty years and refuses your money; and we extract the four things quantitative finance offers a Nepali investor, none of which is a strategy.

I have already established, in Chapter Thirty-Two, that no systematic strategy I could build beat this market after costs, and in Chapter Thirty-Eight that no mechanical value signal survives correction on the available Nepali data.

So this chapter is not about running a quantitative strategy in Nepal. It is about something more useful and much less discussed: what the quantitative tradition contributes to an investor who will never run a model.

The answer is four things, and I would trade almost any strategy for them.

First: the failure that teaches the most

Long-Term Capital Management was founded in 1994 by John Meriwether, who had run the most successful bond arbitrage desk in Wall Street history at Salomon Brothers. He recruited Robert Merton and Myron Scholes, who would receive the Nobel Prize in 1997 for the option pricing work that underpins modern derivatives.

The fund returned something over forty per cent a year in 1995 and 1996 and about twenty in 1997.

In the summer of 1998 Russia defaulted on its domestic debt. LTCM’s positions — which were largely convergence trades, betting that small pricing discrepancies between closely related securities would narrow — all moved against it at once, and they moved together, in a way its models assigned negligible probability. By September the fund had lost most of its capital and the Federal Reserve Bank of New York convened fourteen institutions to recapitalise it in an orderly wind-down, on the reasoning that a disorderly one would have damaged the market itself.

Two Nobel laureates and the best bond traders alive, destroyed in four months.

The lesson usually drawn is that the models were wrong. That is not quite it, and the accurate version is more useful.

Their measurements were largely correct. Their leverage was not.

LTCM’s positions were individually sensible and its estimates of individual risks were reasonable. What broke it was leverage of roughly twenty-five to one against a correlation assumption — the assumption that its many positions were substantially independent of one another.

They were not. Under stress everything they owned turned out to be the same trade: liquidity is available. When it was not, every position moved together.

That is Chapter Twenty-One’s measurement in a different market. Nepali pairwise correlation runs at 0.31 in a boom and 0.64 in a bust. Diversification is a fair-weather property, everywhere, always, and a portfolio built on the assumption that it persists is a portfolio built on the one assumption that reliably fails at the moment it is needed.

The first gift of quantitative finance is therefore a warning about itself: a correctly measured risk, sized wrongly, is fatal — and the sizing error is almost always a correlation error.

Second: the fund that proves it can be done, and refuses you

Chapter Five introduced Renaissance Technologies’ Medallion fund and I return to it because it establishes the boundary of what is possible.

Jim Simons was a distinguished mathematician — he had done foundational work in differential geometry — before he was anything in finance. Medallion has been reported to compound at something like sixty-six per cent a year before fees over three decades, which is not merely the best record in investing but the best by a distance that makes the second place invisible.

So structure exists in price data and can be extracted.

Now the conditions. Renaissance hires astrophysicists, signal-processing specialists and computational linguists, and famously almost nobody from finance. Its edge per trade is tiny and survives only through millions of repetitions with ferocious cost control. And it began closing to outside money in 1993; by the mid-2000s Medallion managed essentially only its partners’ capital, because the strategies have limited capacity and the partners preferred the returns to the fees.

The people who have demonstrably solved this locked the door.

The honest inference is not that quantitative investing does not work. It is that the version which works requires resources you do not have, operates at horizons you cannot trade at Nepali costs, and is unavailable at any price.

Third: the thing quants actually gave everybody

Set the strategies aside. The quantitative tradition’s real contribution to a private investor is a set of habits of inference, and they are worth more than any signal.

Count your attempts. Chapter Thirty-Two opened with Campbell Harvey’s catalogue of 316 published factors and the argument that the profession’s significance threshold was wrong given how many candidates had been searched. Chapter Five told the same story about technical trading rules: Brock’s twenty-six rules looked significant, and when Sullivan, Timmermann and White reconstructed the universe of 7,846 from which those twenty-six had emerged, the best rule survived the correction over Brock’s own sample — and then earned nothing whatever in the decade that followed.

The transferable habit is a single question, asked before any other: how many things were tried before this one was shown to me?

Use a placebo. In my own value study I ranked a seeded random number alongside the real signals. It returned p = 0.73 and ranked nothing, which is what licensed me to read anything else on the page. Had the placebo scored, I would have had a broken test rather than a result — and I would not have known.

This is the single cheapest quality control in all of empirical work and essentially nobody outside academia does it. Whenever you evaluate anything — a screen, a broker’s recommendations, your own last twenty decisions — score something you know is worthless alongside it, and see what it does.

Separate sampling noise from real variation. When my value signal came back noisy, I decomposed the noise and found that roughly four fifths of it was genuine time-variation — the signal really does work better in some quarters than others — and only one fifth was the finite sample. That decomposition is what told me that adding companies would improve precision by about four per cent, which is to say not at all, and that only more years would settle it.

Without the decomposition I would have spent a year extending coverage for nothing.

And pre-register. Decide what you are testing and what would count as a failure before you look. Every finding I retracted in Chapter Thirty-Two was killed by a robustness check I had committed to in advance and therefore had to run. Not one was killed by re-reading my own reasoning, because reasoning does not audit itself.

Fourth: the sizing question, and the honest Nepali answer

The most practically valuable quantitative contribution is about how much, not about what.

The formal treatment goes back to a 1956 paper by John Kelly at Bell Labs, which derived the bet size that maximises the long-run growth rate of capital given an edge. Edward Thorp applied it to blackjack and then to markets, and it is the reason Chapter Thirteen’s coin-flip game destroys you despite a positive expected value: the arithmetic of compounding punishes over-betting far more severely than intuition suggests.

The formula requires two inputs: the expected return and its variance.

And in Nepal, one of those cannot be estimated.

I measured it. Across blind sub-periods of the Nepali market, volatility was stable — 17.7 to 26.1 per cent — and the mean return was not: −9.5 to +35.2 per cent. The implied growth-optimal equity weight ranged over thirteen units of exposure across those blocks, taking the value of one hundred per cent in both bull periods and zero per cent in both bear periods.

That is not an estimate with wide error bars. That is a quantity which does not exist in any stable form.

The denominator of the ratio is measurable and the numerator is not. When the mean moves forty-five points and the volatility moves eight, no amount of data cleaning or model refinement produces an optimal exposure, because the thing being estimated is not sitting still.

I regard this as the most important single result in my research programme, and its implication is liberating rather than depressing:

How much equity you hold is not a calculation. It is a preference.

There is no correct answer being withheld from you by insufficient sophistication. Anyone who offers you a computed optimal allocation for a Nepali portfolio has not measured the spread, and Part Five is built on this — on the question of what you can tolerate rather than what a formula recommends.

The one quantitative result that does hold

For completeness, and because I have spent this chapter demolishing things: one quantitative finding survived everything I threw at it.

Slowing the rebalancing clock is worth about 1.8 percentage points a year, positive in all four blind sub-periods and monotone within each.

It survives because it is not an estimate of a mean. It is an accounting identity applied to a published fee schedule — the same arithmetic that Chapter Fourteen computed in rupees and Chapter Twenty-Four derived from the flat depository charge.

Which is a pattern worth naming as this part of the book closes. The quantitative results that survive in Nepal are the ones that do not require forecasting anything. Costs are certain. Taxes are certain. The fee schedule is published. Correlation in a crisis is measurable and stable in its instability. Everything that required estimating a future mean — factor returns, optimal exposure, seasonal effects, portfolio weights — either failed or could not be resolved with the data available.

That is not a statement about Nepal being primitive. It is a statement about what is and is not estimable in any market with thirteen years of usable data and four turns of a cycle, and a Nepali investor who understands it will avoid a great many expensive products in the decades ahead, as they arrive — and they will arrive.

Merton and Scholes both continued working after LTCM, and both have been asked repeatedly, over the years, what they would have done differently.

The answers have been consistent and unglamorous. Neither has said the models were wrong. Both have said, in various formulations, that the positions were too large.

Two of the most sophisticated financial theorists who have ever lived, reduced by experience to the oldest piece of advice in speculation, which Chapter Thirteen’s twenty-eight-year-old in Singapore arrived at from the opposite direction and which does not require a Nobel Prize to state:

It was not the analysis. It was the size.

The Scorecard

In which seven schools are marked against a single test — what does this method require in order to work, and does Nepal provide it — and the survivors are assembled into one method that is not any of them.

Part Three has examined seven traditions. Each was built by somebody serious, each has evidence behind it, and each was designed for conditions that are not ours.

The question throughout has been the one that ought to be asked of any imported idea and almost never is:

What does this method require in order to work, and does this market provide it?

Here is the complete answer.

The scorecard

The three reasons things fail here

Read down the failure column and the causes sort into exactly three kinds. This is worth internalising, because any new method you encounter will fail for one of them or for none.

Failure by data. Graham’s seven criteria, the F-score as a screen, and every quantitative strategy fail because Nepal does not have the history, the coverage or the fields. Twenty years of dividends require twenty years of continuous existence, which Chapter Twenty-Nine’s consolidation destroyed. Enterprise value requires shares outstanding, which Chapter Thirty-One established is unavailable. My value study needed thirty-eight quarterly cross-sections and had eighteen.

This kind of failure is temporary. Data accumulates. A reader in 2040 should re-run every one of these, and some will pass.

Failure by structure. CAN SLIM fails because a ten per cent daily limit makes a breakout unbuyable, and because T+2 plus the limit makes a tight stop unexecutable. The Magic Formula’s thirty-name portfolio fails because a flat twenty-five rupee depository charge makes small positions uneconomic. Net-nets fail because a bank has no net current assets.

This kind of failure is permanent while the rules stand — and note that the rules do change. The circuit widened to fifteen per cent in April 2026. Settlement will shorten eventually, as it has everywhere. Each such change reopens a question that this book closes, and the closing should be understood as dated.

Failure by absence of a counterparty. Institutional sponsorship in CAN SLIM fails because there are no institutions to track. This is the rarest kind and the most interesting, because it cuts both ways: the same absence that makes the signal unavailable is what makes Chapter Thirty’s amateur advantage real.

What survives, assembled

The survivors do not constitute any one school. They constitute a method, and here it is in the order the work is actually done.

Sort it first. Lynch. Before any calculation, decide which of the six kinds of company you are looking at and write it down. In Nepal, expect the answer to be cyclical far more often than the company’s stability suggests — the banks are cyclicals wearing the clothes of stalwarts, and misclassifying them is the most expensive routine error available here.

Ask what protects the returns, and expect the answer to be a licence. Buffett and Munger, adapted. Then ask the Nepali follow-up that they never had to ask: who issues that licence, when does it expire, and what has the issuer been doing lately? In a market where the moat is granted rather than built, regulatory analysis is not context. It is the durability assessment.

Ask whether the returns can be reinvested. High return with a runway is a compounding machine; high return without one is an income stream. A single hydropower plant is the second, always, and is routinely priced as the first.

Leave the house. Fisher. Branches, dealers, suppliers, the river, and above all the promoter’s reputation. In a market with weak accounts and no analysts, non-documentary evidence carries proportionately more weight, and the field is empty.

Run the health checklist on the statements. Piotroski, as questions rather than as a screen. Is operating cash flow positive and greater than reported profit? Have new shares been issued? Is leverage rising? Is the bad-loan ratio rising while coverage falls? Is loan growth far above the system’s?

Watch the external assessment where one exists. QGLP’s Yes Bank lesson. Where a company’s own numbers conflict with somebody else’s examination of the same thing, the outside number is the information. In Nepal that means the gap between the unaudited August figures and the audited December ones, and the non-performing trajectory rather than its level.

Then price it as a range, and demand a wide buffer. Graham. Not a number — a range, with a stated width, and no purchase near the top of it. Given the disclosure quality established in Chapter Thirty-One, the Nepali buffer must be wider than the one Graham used in New York, not narrower.

And hold. Chapter Fourteen’s arithmetic: NPR 914,104 held against NPR 700,141 round-tripped, over twenty years, on identical stock selection. Every structural feature of this market punishes activity and is indifferent to patience.

Throughout: count your attempts, use a placebo, and write the falsifier down before the outcome exists. The quantitative discipline, which costs nothing and which is the only defence against the entirety of Part One.

What this method cannot do

Three honest limits, stated before Part Four begins applying it.

It will not tell you when. Nothing in this book forecasts a price or a turn. Chapter Nineteen showed the cycle has a mechanism and an effective sample of four, which is not enough to trade. The method tells you what something is worth and what you are being paid to own it. The timing is not available and I have not found anybody who has it.

It will frequently return nothing. Chapter Fifteen’s engine refuses to value companies whose disclosures do not support a valuation, and the refusals turned out to be its most informative output. A method that always produces an answer is not producing answers.

And it has not been proven to work in Nepal. I want this said plainly at the hinge of the book. Chapter Thirty-Eight’s value study found no signal surviving correction and established that the question cannot be resolved with today’s data. What I have is a method with sound logic, international evidence, and structural arguments for why this market should reward it — not a measured Nepali edge.

The reasons to believe it anyway are the ones Chapter Thirty-Three set out: a valuation has an anchor and a forecast does not; nobody else here is doing the work; and the frictions that destroyed every trading strategy do not touch a method that buys and holds. Those are arguments, not proof, and you should hold them as arguments.

One test to carry

If you take a single thing from Part Three, take the question rather than the answers, because new methods will keep arriving and this book will age.

When somebody presents you with an approach — in a course, a book, a group, a newsletter, or your own head at two in the morning — ask:

What does this require in order to work?

Then check each requirement against the machine described in Part Two. Does it need to buy strength? The circuit forbids it. Does it need to trade often? The flat fee and the tax step forbid it. Does it need many holdings? The fee schedule and the correlation floor forbid it. Does it need clean historical accounts? Fifty of a hundred and eighty-nine names have them. Does it need to short? There is no short selling. Does it need institutional flow data? There are no institutions.

Most imported methods fail two or three of these, and they fail silently, and the person selling them has usually never asked.

Part Four takes the surviving method and applies it to the thirteen sectors, one at a time — what each business is actually made of, which metrics rank first and which you may ignore entirely, what the bull and bear cases each require to be true, and how to arrive at a range you would act on.

It begins with the two questions that must be answered before any sector can be valued: what discount rate to use in a country whose risk-free rate is a fixed deposit, and how to convert a range into a decision.

Neither has a textbook answer here. Both have an answer.

What Rate Do You Discount At?

In which a Nobel-winning formula turns out not to describe returns; nineteen Nepali banks are measured against an index they themselves dominate and produce a beta of 0.12; the same nineteen banks measured correctly produce 0.90; and we build a discount rate for a country whose risk-free asset is a fixed deposit.

Every valuation in the rest of this book requires one number that cannot be looked up.

A company’s value is the money it will hand its owners, discounted back to today. The discounting requires a rate. That rate is the return you require for taking the risk, and it is not printed anywhere, and small changes in it move the answer enormously.

So before any sector can be valued, this has to be settled.

The formula everybody uses

The Capital Asset Pricing Model was developed independently by several economists in the early 1960s, and it earned William Sharpe a share of the 1990 Nobel Prize. It says something elegant: the return you should require from an asset depends only on how much it moves with the market as a whole.

Required return = risk-free rate + β × (market return − risk-free rate)

Beta is the sensitivity. A share that moves one-for-one with the market has a beta of 1. One that moves half as much has 0.5 and, the model says, should be required to return less, because it adds less risk to a diversified portfolio.

The logic is genuinely beautiful. The only difficulty is that it does not appear to be true.

In 1992 Eugene Fama and Kenneth French tested the relationship across the American cross-section and found that the relation between beta and average return was essentially flat. High-beta stocks had not delivered higher returns. What did explain returns was size and book-to-market — the very characteristics the model said should not matter.

Thirty years of subsequent work has not rehabilitated beta as a predictor of returns.

And yet every corporate finance department, every valuation textbook and every analyst report on earth still uses it, for a reason that is worth being honest about: you need a number, and it is the only defensible procedure for producing one. The alternative is to assert a required return with no framework at all, which is worse — not because the framework is right, but because it forces you to state your assumptions where somebody can attack them.

Use it in that spirit and it is useful. Use it as truth and it will mislead you, and in Nepal it will mislead you in a specific and measurable way.

Three problems in Nepal, and the third is the interesting one

First: there is no risk-free rate.

The formula starts with the return on a riskless asset, which in America means a Treasury bill. Nepal issues government securities, but the secondary market is thin and the yields are not a continuously observable, liquid benchmark of the kind the model assumes.

More importantly, the risk-free asset that a Nepali saver actually chooses between is not a government bond. It is a fixed deposit at a commercial bank, and Chapter Nineteen established that the weighted average deposit rate is published, real, and ranged between 3.28 and 7.86 per cent across the observed period, with a realised mean of 5.31 per cent.

That is the genuine alternative. It is what you give up by owning shares. It should be the foundation of the discount rate, and it is a floating foundation — it moves by several percentage points across a cycle, which is a fact the rest of this chapter has to accommodate.

Second: beta against an index you dominate is circular.

Commercial banks are the largest component of the NEPSE index. Measuring a commercial bank’s sensitivity to an index made substantially of commercial banks is not a measurement of anything external; it is close to regressing a variable against itself.

Theory demands that the answer come out near 1.0, and if it does not, the measurement is wrong before the interpretation begins.

Third: it is extraordinarily easy to compute wrongly, and the wrong answer looks plausible.

The demonstration

I computed the beta of every listed commercial bank against the NEPSE index over the five years from August 2021 to July 2026, using daily returns.

Done correctly — that is, using only the dates on which both the share and the index printed a price, and computing returns on that intersection:

Mean 0.898. Median 0.910.

That is exactly what it must look like if it is right. A sector that constitutes much of the index has a beta a little under one, with the conservatively run banks — Standard Chartered at 0.641, Nabil at 0.731 — genuinely less volatile than the aggressive ones, and Nepal Investment Mega Bank at 1.100 genuinely more.

Now the same data, the same period, the same formula, with one thing changed. Instead of intersecting the dates, take each series on its own calendar and line up the returns positionally — the share’s first return against the index’s first return, and so on.

Mean 0.121. Median 0.140.

Nineteen banks, which between them are the index, apparently moving one-eighth as much as the thing they constitute.

Why this matters more than a footnote

The two series do not share a date grid. The index prints on days some individual shares do not trade, and Chapter Thirty-One established that one Nepali price source is shifted forward by exactly one trading day relative to another.

Line up two series positionally and you are correlating Tuesday’s share return against Monday’s index return, at an offset that drifts as the number of missing days accumulates. The correlation collapses toward zero, and since beta is that correlation scaled, beta collapses with it.

The result is not obviously absurd. Zero point one two looks like a low-beta stock. A report stating that Nepali banks are defensive, low-volatility holdings would read perfectly well, and would flow into a cost of equity, and from there into a valuation, and from there into a published recommendation.

I know this because a version of it happened to me. A beta near 0.4 for the bank sector reached a published cost of equity and two live websites before anybody caught it, and what caught it was not a test. It was somebody asking what the number must look like if it were right.

A number that should have stopped you is an error message. Nineteen banks cannot have a beta of 0.12 against an index they dominate, in the same way that a river cannot flow uphill, and the impossibility is visible without any statistics at all.

Before trusting any measurement in this book or anywhere else, ask what it must look like if correct. Then check whether it does.

Building a rate from the ground

Given all of the above, here is the procedure I actually use. It is not elegant and it does not have a Nobel Prize attached, and it has the single merit of making every assumption visible.

Start with the deposit rate. The published weighted average deposit rate of commercial banks, which is what a Nepali saver genuinely forgoes. Use the current figure, and know that it has ranged from 3.28 to 7.86 per cent in the observed record.

Add an equity risk premium. This is the compensation for owning a business rather than a deposit, and it is a judgment. The international evidence is a useful anchor: the long-run world equity premium over bonds, measured across twenty-plus countries and more than a century, is of the order of three to four percentage points — considerably lower than the five to seven that American-centric sources quote, because America was the among the very best-performing markets of the twentieth century and using it alone is survivorship of exactly the kind Chapter Three described.

For Nepal, a premium over the deposit rate rather than over a bond needs to be higher than that, because you are also being compensated for illiquidity, weak disclosure, poor governance enforcement and the absence of any mechanism to express a negative view.

Then adjust for the specific company. Not with beta — beta here is either circular or broken — but with the things that genuinely differentiate risk between two Nepali companies. A concentrated loan book. A single customer. A licence approaching expiry. A promoter with a record. A capital ratio close to the regulatory floor.

And then state the number and defend it, rather than deriving it from a regression that will not survive Chapter Thirty-One’s data problems.

The rule I will not break, and why

One cost of equity per company. Applied to shareholder cash flows. No weighted average cost of capital, anywhere, in any model.

This deserves an explanation because it departs from every textbook.

The weighted average cost of capital blends the cost of equity and the after-tax cost of debt in proportion to their weights, and it is used to discount cash flows available to all providers of capital — after which you subtract net debt to get to the shareholder’s claim.

That procedure is fine for an industrial company. It is incoherent for a bank, because a bank’s debt is not financing; it is raw material. Deposits are what a bank sells its services against. Treating a bank’s deposits as part of its capital structure and computing a blended cost produces a number that means nothing.

And since more than half of this exchange is financial, running two discounting stances — one for banks and one for everybody else — creates exactly the situation Chapter Thirty-Two warned about: two surfaces answering one question two different ways, where the defect is the divergence, whichever answer is nicer.

So: every model in Part Four discounts a shareholder flow at a shareholder rate. Dividends, residual income, free cash flow to equity. One stance, applied everywhere, which means the numbers from different sectors are comparable — which is the whole point of having a stance.

How much does it matter?

Enough that the rest of this part would be worthless without settling it.

Take the standard relationship for what a bank should be worth relative to its book value. Justified price-to-book = (return on equity − growth) ÷ (cost of equity − growth). Set the return on equity at 14 per cent and growth at 8, and vary only the discount rate.

A two-percentage-point error in the discount rate moves the answer by fifty per cent. Nothing else in a valuation is anywhere near this sensitive — not the growth estimate, not the margin assumption, not the terminal value convention.

Which produces three obligations that govern everything that follows.

State the rate explicitly, every time. A valuation that does not name its discount rate has concealed the most important input.

Never fetch it and never pre-fill it. This is a rule I hold absolutely in my own work: the cost of equity, the growth rate and the forecast return on equity are judgments, and a system that supplies them by default is making the analyst’s decision for him while letting him believe he made it. Reference values may be displayed. They must never populate the box.

And always show the range. Since a two-point error moves the answer by half, a single-point valuation is a false precision. Every number in Part Four is a range produced by varying the rate across a defensible span — which is Graham’s margin of safety expressed as arithmetic rather than as advice.

The awkward consequence nobody likes

There is a conclusion that follows from the deposit rate being the foundation, and it is unwelcome.

Your discount rate should move with the cycle, which means the same company is worth different amounts in different years for reasons that have nothing to do with the company.

When the deposit rate is 4.65 per cent, as it was in July 2021, the required return on equity is lower and the same earnings stream is worth more. When it is 7.86, as in 2023, it is worth less.

That feels wrong. It feels like letting the market’s mood into the valuation, which is precisely what a valuation is supposed to exclude.

It is not the market’s mood. It is the alternative. Value is always relative to what else you could do with the money, and when a bank will pay you 7.86 per cent for nothing, a business must clear a higher bar to deserve your capital. That is not sentiment; it is the arithmetic of choice.

The discipline, then, is not to hold the rate constant. It is to be honest about which direction the cycle is pushing it, and to notice — as Chapter Nineteen’s table showed — that the deposit rate is lowest at market tops, which means your discount rate will be lowest, and your valuations highest, at exactly the moment they should be most suspect.

Knowing that does not fix it. It does tell you where to apply the margin of safety, and Chapter Forty-Four is about how.

Sharpe was asked, decades after the Nobel, what he thought of the model’s empirical record.

He said, in effect, that the theory described how prices should be set if investors behaved as the assumptions described, and that the interesting question had always been which assumption failed.

He did not defend beta as a forecasting tool and did not need to. The model’s value was never the number it produced.

It was that it forced everybody to say out loud what they were assuming, in a form somebody else could check.

That is all I am asking of the rate you choose.

The Range and the Verdict

In which investment banks draw a picture that admits they do not know; the average of three methods turns out to be one method wearing a disguise; two models that look different are revealed to be the same algebra voting twice; and we settle how a range becomes a decision.

Walk into any investment bank in the world and ask what a company is worth, and after some weeks you will be handed a chart called a football field.

It is a set of horizontal bars stacked vertically. One bar for the discounted cash flow valuation, spanning the range produced by varying the assumptions. One for comparable companies, spanning the range implied by the peer multiples. One for precedent transactions. One for the fifty-two-week trading range. And a vertical line showing the current price.

The bars overlap somewhat and disagree considerably, which is the point.

I have always found it revealing that the most sophisticated valuation practitioners in the world, working on transactions worth billions, present their conclusion as a picture of disagreement — while a television analyst discussing a company he has never visited gives a single number to two decimal places.

The football field is an institutional admission that a single valuation is a fiction. Everything in this chapter follows from taking that admission seriously.

The flaw of averages

There is a general reason single numbers mislead, and it goes beyond humility.

Sam Savage gave it a name: the flaw of averages — the observation that plans built on average assumptions fail more often than the averages suggest, because the relationship between inputs and outputs is not a straight line.

His illustration is a statistician who drowns crossing a river of average depth three feet.

In valuation the non-linearity is everywhere and it runs in one direction. Chapter Forty-Three’s table showed it: a cost of equity of 12 per cent gives a justified price-to-book of 1.50, and 16 per cent gives 0.75. The midpoint of the rates, 14 per cent, gives 1.00 — which is not the midpoint of the values, 1.125.

Valuing at the average assumption does not produce the average value. It produces a number that sits nowhere in particular, and the direction of the error depends on which input you are averaging.

Aswath Damodaran, who has done more than anybody to make valuation a public discipline, has made this his central practical point for years: he publishes distributions, simulates his inputs, and treats the spread as part of the answer rather than as an embarrassment about it.

And the empirical record on point estimates is poor. The academic work on sell-side target prices finds them systematically optimistic and frequently far from realised prices at the horizon they specify — which is unsurprising once you notice that a target price is a point estimate of a quantity that is not a point.

So how do you combine methods?

Suppose you have valued a company three ways. A dividend discount model says 280. A residual income model says 340. Graham’s formula says 190.

What is the company worth?

The naive answer is to average them, and averaging has a real defect: it lets one extreme method drag the answer. If your third method produced 900 because of a runaway terminal value, the mean moves 200 points and you have published a number driven by the model you trusted least.

So take the median instead, says the next instinct. And here is the problem that took me an embarrassingly long time to see.

At three methods, the median *is* one method’s number.

The median of 280, 340 and 190 is 280 — which is precisely the dividend discount model, unchanged, to the rupee. The other two contributed nothing. You have run three valuations and published one, and the arithmetic has concealed which.

Worse, it is not stable. Nudge the residual income model down to 270 and the published value jumps from 280 to 270 — the whole answer changes because a different method became the middle one, while the underlying analysis barely moved.

And this is not a corner case. Three or four methods is exactly what a real valuation produces. A median at that sample size is a mechanism for silently selecting one lens and discarding the rest.

The trimean

The fix is old and comes from John Tukey, who spent a career building statistics that behave sensibly on small and messy samples.

Trimean = (lowest + 2 × middle + highest) ÷ 4

It weights the middle observation at fifty per cent and each extreme at twenty-five, which means every method contributes and no single method dominates.

On our three: (190 + 2 × 280 + 340) ÷ 4 = 272.5.

Compare: mean 270, median 280, trimean 272.5. On a well-behaved set they agree closely, which is the sign of a healthy valuation.

Now a set that disagrees. Suppose the methods produce 150, 155 and 400.

The median publishes 155 and pretends the 400 does not exist. The mean publishes 235, which no method produced and which sits in empty space between two clusters. The trimean publishes 215 — still influenced by the outlier, still anchored near the cluster, and honest about both.

But look at what that set is actually telling you, which is the more important lesson: your methods disagree by a factor of two and a half. No combining rule fixes that. The correct response is not to publish 215. It is to find out why one model says 400 and two say 150, because one of them contains an error or an assumption you have not examined.

A combining rule is for reconciling methods that broadly agree. When they do not, the disagreement is the finding.

At five or more methods, use the proper quartile form — (first quartile + 2 × median + third quartile) ÷ 4 — which is Tukey’s original construction and which behaves better once there are enough observations to have quartiles worth the name.

The duplicate problem

Here is the subtler failure, and it is one I built into my own work and had to remove.

Suppose you run four methods and one of them is a justified price-to-book model and another is a residual income model.

At the settings a Nepali bank valuation typically uses, those two are the same algebra. Justified price-to-book says value equals book times (return on equity minus growth) over (cost of equity minus growth). The residual income model says value equals book plus the discounted excess of return over cost of equity. Work either through with constant growth and a constant return, and they reduce to the same expression.

They are not two opinions. They are one opinion, computed twice, using the same inputs.

And when you feed both into a combining rule, that single idea votes twice, and the anchor moves toward it, and a cap you have placed on any one model is defeated by running the model twice under two names.

The fix is to collapse duplicates before combining: if two lenses land within a small tolerance of each other and share their inputs, count them once and record which was folded.

The general principle applies well beyond valuation, and it is Chapter Six’s arithmetic in another costume: the number of opinions is not the number of voices. Forty thousand people in a group chat contain ten opinions. Four models sharing two inputs contain two.

Before averaging anything, ask what is actually independent.

How wide should the range be?

Now the band, and the principle governing it is simple to state and unusual in practice.

The published range must contain every method it claims to summarise.

If your lenses produced 190, 280 and 340, and you publish an anchor of 272 with a range of 250 to 295, you have published a range that excludes two of your three methods. The range is then not a summary of your work; it is a claim of precision your work does not support.

So the band is set by the distance from the anchor to the furthest method, with two adjustments.

A floor, because a valuation of three methods that happen to agree closely is not actually more precise than the data permits. If three models on Nepali accounts land within two per cent of each other, that agreement reflects shared inputs rather than genuine convergence, and publishing a range of ±2 per cent would be absurd given Chapter Thirty-One’s data quality.

And a cap, because beyond a certain width a range stops being a valuation and becomes a refusal.

The cap creates an obligation, and it is the part most people would skip. When the cap binds — when your methods disagree by more than the published range shows — you must say so. The reader is entitled to know that the range was truncated and that the underlying lenses spread wider than it appears. A silently capped range is a lie of omission, and it is exactly the sort of quiet flattery that Part One’s chapters were about.

Premium, not upside

Now the verdict, and this is where a great deal of money is quietly lost to arithmetic.

You have a value range and a market price. Which number describes the gap?

There are two candidates and they are not the same.

Upside = (value ÷ price) − 1. How much the price must rise to reach your value.

Premium = (price ÷ value) − 1. How much the price exceeds your value.

Take a company you value at 100 which trades at 50.

Upside is +100 per cent. Premium is −50 per cent.

Both are true and they describe the same situation, and one of them is a great deal more exciting than the other. Now take a company you value at 100 trading at 200. Upside is −50 per cent; premium is +100 per cent.

Notice the asymmetry. Upside is unbounded above and floored at −100 per cent. It makes cheap things look spectacular and expensive things look merely disappointing. It is the number every promotional document uses, and it is not a coincidence.

Judge on premium, always. It is bounded the way your loss is bounded, it treats overvaluation and undervaluation symmetrically, and it does not produce the headline “three hundred per cent upside” that has preceded so many disasters.

The practical form: this share trades at a 35 per cent discount to the middle of my range is a statement you can act on. This share has 54 per cent upside is a statement designed to make you act.

The band endpoints are the triggers

The last piece, and it is what converts analysis into behaviour.

The endpoints of the range are the buy and sell levels. Not the anchor.

If your range is 250 to 340, then:

Below 250, the price is beneath everything your work supports, and that is a buy — not “attractive,” not “worth watching.” Below the bottom of the range, with a margin of safety on top, is where you act.

Between 250 and 340, the price is inside the range and you have no information. This is the most common outcome and the correct response is to do nothing, which Chapter Fifteen established is a position rather than an absence of one.

Above 340, the price exceeds everything your work supports.

The reason to use the endpoints rather than the anchor is Graham’s, and it is the whole of Chapter Thirty-Four: your anchor is an estimate and it is wrong. Buying at the anchor means buying at your central guess, which leaves you nothing if the guess was high. Buying below the band means the market is offering a price beneath your most pessimistic method — and that is a margin of safety expressed in a form you can execute.

Critically: the band belongs to the valuation, not to the analyst. A company whose methods agree closely has a narrow band and a tight trigger. One whose methods spread widely has a wide band and a distant trigger — which is correct, because you should demand a bigger discount from a company you understand less well. Applying one fixed percentage to every company would grant the same confidence to a bank with eight years of clean statements and a hydropower project with a hydrology study nobody has checked.

The refusal

One more output, and Chapter Fifteen argued it is the most informative one.

Sometimes the correct answer is no valuation at all.

If the inputs are absent, if the growth estimate has too few observations to support it, if the filing fails a plausibility check, if two methods disagree by a factor that implies one of them is broken — the answer is not a wide range. It is a named refusal: I cannot value this, and here is which input is missing.

A wide range says “the value is uncertain.” A refusal says “I do not know whether the inputs mean what they say.” Those are different statements and only one of them is true when a company’s disclosures do not support a valuation.

A method that always produces an answer has not been tested against a company that cannot be valued, and in Nepal there are many.

The whole procedure

For any company in Part Four:

One. Run three to five genuinely independent methods, appropriate to the sector.

Two. Collapse duplicates. Two lenses sharing inputs and landing together are one vote.

Three. Combine with the trimean, not the mean and not the median.

Four. Set the band to reach the furthest surviving method, floored and capped, and say so when the cap binds.

Five. Compute the premium — never the upside.

Six. Act only outside the band, and only with a margin of safety beyond it.

Seven. If any of the above cannot be done honestly, refuse and name the reason.

That is the machinery. The next fifteen chapters are about what to put into it, sector by sector — which methods are appropriate, which metrics rank first, which you may ignore entirely, and what the bull and bear cases each require to be true.

Damodaran has said, in various forms, that his valuations are usually wrong and that this does not trouble him, because the purpose was never to be right.

The purpose is to have a number you arrived at yourself, by a route you can retrace, so that when the price moves you can tell whether the world changed or only the price did.

Without one you are left with the only alternative, which is somebody else’s number, and you will not know how they got it either.

Commercial Banks

In which we establish that a bank is a business with the risk on the wrong side of the balance sheet; that its reported profit is a forecast rather than a fact; and we value all nineteen Nepali commercial banks, showing every input, and find them worth between eighty-three rupees and six hundred and fifty.

Nineteen commercial banks. One rulebook. One regulator. One economy.

There is no sector on any exchange in the world more suited to comparison than this one — nineteen shops selling the same commodity under identical rules, which means the excuses that blur every other sector’s comparisons do not exist here. If Bank A earns more than Bank B, it is not because of a different business model, a different market or a different accounting convention. It is because it is better run.

That makes commercial banking the natural place to begin Part Four, and also the most dangerous, because a bank’s accounts do something no other company’s accounts do.

The business, in one paragraph

A bank borrows short and lends long. It takes deposits, which are repayable on demand or at short notice and cost it very little, and it lends them for years at more. The difference is the spread, and the spread multiplied by the size of the balance sheet is most of the profit.

Two smaller engines sit beside it. Fee income — remittance, cards, trade finance, guarantees — which is valuable out of proportion to its size because it consumes almost no capital. And treasury, the return on the securities portfolio, which is mostly a function of what rates did.

That is the whole business. A momo shop with a licence to use other people’s money.

Why a bank is not a normal company

Three features make bank analysis a separate discipline, and every error people make here traces to one of them.

Leverage is not a choice, it is the product. A manufacturer with ten times more liabilities than equity is in distress. A bank with ten times more is normal, and one with five times is wasting its licence. You cannot read a bank’s leverage as a risk signal the way you read anybody else’s, because lending other people’s money is the enterprise.

The balance sheet is the profit and loss account with a delay. A loan written this year appears as an asset. If it is a bad loan, it will appear as a provision in three years. Which means a bank’s reported profit today is partly a statement about loans it has not yet had to assess — and a bank can raise this year’s profit simply by recognising less, with the consequence arriving after the chief executive’s term.

And the risk is on the asset side while the funding is on the liability side. A bank does not fail because it loses money. It fails because depositors want their money back faster than the loans repay. Yes Bank in Chapter Forty had a loan-book problem for years and a crisis only when the withdrawals started.

The central error

Which brings us to the thing this chapter exists to prevent.

You cannot value a bank on the return on equity it printed last year.

Chapter Thirty-Five showed why with data: across Nepali commercial banks, the correlation between price-to-book and reported return on equity is only +0.40, far weaker than the theory demands. And Chapter Thirty-Seven established the classification error underneath it: Nepali banks are cyclicals wearing the clothes of stalwarts.

At the top of a credit cycle, a bank reports low provisions, high profit, and a low price-to-earnings ratio. Every screen says it is cheap. It is not — it is at the point of maximum earnings.

At the bottom, it has provided heavily, its return on equity has collapsed, and the screen says expensive. It frequently is not.

Look at NIC Asia in Chapter Thirty-Five’s table: return on equity of 1.6 per cent alongside non-performing loans of 9.53 per cent. Those two numbers are not independent. The low return is the bad-loan ratio, arriving in the income statement as provisions. Anybody screening on return on equity would rank that bank last, and anybody screening on price-to-earnings would find the ratio meaningless because the earnings have vanished.

The reported number is contaminated by the cycle, and the whole job is to decontaminate it.

The normalisation chain

Here is the procedure, in the order it must be done. Each step exists because doing it in a different order produces a wrong answer.

Step one: normalise the bad loans.

There are two numbers. npl_now is what the bank reports today. npl_normalised is what you judge the book carries through a full cycle.

The gap between them becomes a haircut on returns — but not the whole gap, because some impaired loans cure and some are recovered. The fraction you judge will actually be paid for rather than cured is a weight I call omega, and in the established Nepali bank set it sits at 0.5: half the gap is treated as a real permanent cost, half as timing.

That is a judgment. It is stated, it is visible, and it can be attacked — which is the standard from Chapter Forty-Three.

Step two: normalise the return on ASSETS, not on equity.

This is the step everybody skips and it is the one that makes the rest work.

Strip the one-off items out of the reported return on assets to get a core return. Then apply the haircut to get a normalised return on assets.

Why assets and not equity? Because leverage is a capital structure decision and it should not contaminate the judgment about earning power. Two banks with identical lending skill and different capital ratios will report different returns on equity, and the difference tells you about their capital, not their competence.

Working at the asset level lets a damaged bank and a clean one be compared on the same basis: what does each earn on the assets it holds, through a cycle?

Step three: lever it back up.

Operating return on equity = normalised return on assets × leverage.

This single number is what the entire valuation turns on, and it is the only input worth simulating.

Name the damage, and let the name set the horizon

Two banks with identical ratios today are not the same investment if one is recovering from a shock it did not cause and the other from its own lending decisions.

So each bank carries two labels.

Damage origin — why the book is where it is:

Chapter Twenty-Nine gives the inherited category its force: seven Nepali acquirers, five of which underperformed their sector in the first year, four of which still do. When you absorb a bank you absorb its loan book, and the label records that.

Tier — how bad, which sets how long the franchise can be defended:

The logic is subtle and worth stating carefully. It is not that a damaged bank is worth less — the normalised return on equity already says that, and saying it twice would be double-counting.

It is that a damaged bank has a shorter defensible runway. A franchise you can argue is durable for thirty years is a different asset from one you can only defend for fifteen. Chapter Thirty-Five called this longevity and Chapter Forty called it the L in QGLP; here it is expressed as an explicit horizon rather than smuggled into a growth rate where nobody can see it.

Origin informs tier without mechanically setting it. Self-inflicted damage earns scepticism about management, and scepticism about management is a reason to shorten the runway.

One cost of equity per bank

Chapter Forty-Three established the principle and this sector is where it bites hardest.

The objection that forced it: using one discount rate for nineteen banks means discounting NIC Asia’s 9.53 per cent non-performing book at the same rate as Standard Chartered’s 1.81 per cent. That is not conservatism. It is a refusal to distinguish.

So each bank carries its own rate, built from its own measured beta shrunk toward the market, plus a premium for asset quality and capital position. Across the nineteen the rates run from 8.24 per cent for Standard Chartered to 10.28 per cent for NIC Asia — and that two-point spread is the framework doing its job.

One further note, and it is a confession. The published set recorded the resulting rate but not the risk-free rate, market premium, beta and quality premium that composed it. A rate you cannot re-derive is a rate you cannot audit. Every sector after this one stores the components, and I record the omission here rather than quietly fixing it.

And this is the sector where Chapter Forty-Three’s beta demonstration was not a hypothetical. The original valuations used betas measured across a one-day grid offset between two price sources; Nabil’s beta was 0.44 when it should have been 0.76. Every one of the nineteen was revalued when it was found.

Three lenses, chosen to disagree

They are chosen to disagree. Three variations on one argument would produce a narrow range that means nothing, which is Chapter Forty-Four’s warning about duplicate lenses — and note that the first two here are close relatives, which is exactly why the duplicate-collapse rule exists and why Graham is capped so a rule-of-thumb screen cannot dominate two modelled lenses.

Combined by the trimean. Band set to reach the furthest surviving lens, floored and capped, with the obligation to say so when the cap binds.

And the simulation samples one input only: normalised operating return on equity. Not the cost of equity, which is decided. Not terminal growth, which is a house rule. Sampling decided numbers while holding the undecided one still is theatre. Dispersion comes from the standard error of the bank’s own annual return history — measured, never assumed — and with fewer than three annual observations there is no simulation at all, named rather than silently omitted.

The nineteen

Here is the complete established set, as at August 2026. Every input visible.