Fooled by Randomness

Nassim Nicholas Taleb · 2001

Thesis

People systematically mistake luck for skill and noise for information, and in markets and careers the damage comes from rare events that a record of past success cannot reveal.

The main ideas

Judge a result by the full set of outcomes that could have happened, not the one that did. Most of those alternative histories are invisible.
A large enough population of people taking random bets will produce a few spectacular records by chance alone. Survivors are what gets seen.
How often you win matters less than how much you win or lose when you do. A strategy can be right most of the time and still lose money.
The more often you look at results, the more of what you see is noise. At high frequency, even a strong strategy looks like a coin toss.
No number of past observations proves that a rare event will not happen. Traders blow up by treating "it has never happened" as "it cannot happen."
Human intuition about probability is poor even among trained professionals, and it fails in predictable directions.
What luck delivers, luck can take away. Over a long enough time, a lucky strategy tends to revert to its true expectation.

Alternative histories

Taleb argues that a result should be judged by the costs and outcomes of the histories that did not happen as well as the one that did. His case: an eccentric, bored tycoon offers $10 million to play Russian roulette with one bullet in six chambers. There are six equally likely histories. In five, the player walks away rich; in one, the result is an obituary. Anyone who played and won has $10 million, and an observer who sees only the winner sees wealth and nerve. Taleb holds that $10 million won this way is not worth the same as $10 million earned through a career in dentistry, even though the bank balances are identical, because the dentist's alternative histories are nearly all good and the gambler's include death.

He calls the tool for thinking this way a Monte Carlo engine: generate many possible sample paths of a strategy and look at the whole distribution, not the single path reality produced. In life the other five chambers are never shown, so the reasoning has to supply them.

Implication: before crediting a person or a strategy for a result, ask what the same decisions would have produced across the other plausible histories, and weigh the worst of them, not the average.

Survivorship bias

A population of people taking random bets will throw up some who look like geniuses. Taleb's thought experiment: 10,000 fund managers, each with a 50% chance each year of making $10,000 or losing $10,000, with no skill at all. Those who lose in a year are removed. After one year 5,000 remain, then 2,500, 1,250, 625, and after five years about 313 managers have made money every single year. Those 313 are the ones who give interviews and attract money, and their records are pure chance.

The mechanism is that success is sampled from the survivors and failures leave no trace. The fewer the losers anyone counts, the more the winners' records look like skill. Taleb argues that the question to ask about a record is not how good it is but how many people started out and how likely a record this good would be among them by chance.

Implication: a track record is evidence only in proportion to the number of people who took similar bets and failed. Before paying for a manager, a founder's advice or a strategy, estimate the size of the starting population you are not seeing.

Frequency is not expectation

The probability of being right and the size of the payoff when right are separate quantities, and people collapse them. Taleb's example from his own trading: he believed the market was more likely to rise than fall, 70% against 30%, and he was short. A 70% chance of a 1% rise and a 30% chance of a 10% fall gives an expected result of minus 2.3%, so the bet on a fall was the better one even though a fall was the less likely outcome.

The same asymmetry separates Nero Tulip, the book's careful trader, from John, a high-yield bond trader who earned far more than Nero for years with a strategy that won often and in small amounts. In the summer of 1998 John lost everything. A strategy that wins small and often while carrying a rare large loss produces a smooth record that looks like skill until the loss arrives. The record of such a strategy is least informative exactly when it looks best.

Implication: evaluate any bet, trade or policy by its payoff in each outcome times the probability of that outcome, and distrust strategies whose record is smooth because their losses are rare.

Scale and noise

Over short intervals, variance swamps return, so frequent observation mostly shows noise. Taleb's case is a retired dentist with a portfolio expected to earn 15% a year above Treasury bills with 10% volatility. Checked once a year, the portfolio shows a gain 93% of the time. The book's table gives the probability of a gain at shorter intervals: 77% for a quarter, 67% for a month, 54% for a day, 51.3% for an hour, 50.17% for a minute and 50.02% for a second. Watching the market by the minute over an eight-hour day, the dentist gets roughly 241 pleasant minutes and 239 unpleasant ones.

The mechanism is that return grows in proportion to time while volatility grows only with its square root. Because losses hurt more than equal gains please, the dentist who checks every minute ends each day drained by a portfolio that is working. The same arithmetic applies to news: the more frequently information arrives, the higher the share of it that is noise, and the more a journalist must supply a cause for each random move.

Implication: set the review interval for any strategy to match the horizon over which it is expected to show a result. Daily figures for a strategy meant to pay off over years are information about noise, and acting on them turns noise into trades.

The problem of induction

No number of observations can confirm a general claim, but a single observation can refute it. Taleb takes this from Hume, through the image of white and black swans, and from Karl Popper's view that theories can be falsified but never verified. He describes himself as a skeptical empiricist: he accepts evidence against his views readily and treats evidence for them as provisional.

Traders who reason from "it has never happened" are exposed to the event that has not happened yet. Victor Niederhoffer, an accomplished speculator, blew up in 1997 selling puts on the stock index, a strategy that paid steadily as long as the market did not do something it had not done in his experience. Carlos, the book's emerging-market bond trader, made about $80 million over several years by buying dips, then lost about $300 million in the summer of 1998, mostly in Russian bonds, when a dip did not recover.

Implication: for any position, identify the event that would break it and ask whether it is survivable, whether or not it has ever happened. A strategy that is ruined by a single unprecedented event will eventually meet one.

Probability blindness

Humans are poor intuitive statisticians, including those trained in statistics. Taleb draws on the heuristics-and-biases research of Kahneman and Tversky and gives a medical test as the example. A disease strikes 1 person in 1,000. A test for it has a 5% false-positive rate and detects every true case. A patient tests positive. Most doctors in the study the book cites said the chance the patient has the disease is 95%. The correct answer is about 2%: of 1,000 people tested, about 50 healthy people test positive alongside the 1 who is sick, so a positive result means disease roughly 1 time in 51.

The error is neglecting the base rate: the rarity of the disease dominates the test's accuracy, and intuition ignores it. The same blindness makes people treat a vivid rare risk as common and a boring common one as rare.

Implication: when a signal suggests a rare condition, whether a disease, a fraud or a star employee, start from how rare the condition is and work out the false-positive count before believing the signal.

Solon's warning

The first part of the book takes its name from Solon, who told the rich king Croesus that no one's life can be called happy until it is over. Croesus lost his kingdom to Cyrus. Taleb's reading: what came with luck can be taken away by luck. He connects this to ergodicity, the mathematical property under which very long sample paths come to resemble each other. Over enough time a lucky strategy regresses to its true expectation, and the fool of randomness who was paid by chance tends to give the money back.

The traders in the book fail because they infer from early wins that their method works and then scale it up.

Implication: do not increase the size of a bet because it has paid; increase it only when you can show why the edge exists and why it will persist.

Limits

The dentist table's figures are exactly what a normal distribution gives. The book's central claim is that market returns are not normal and that rare large moves dominate, so its own illustration of noise at different scales uses the thin-tailed model it argues against.
The book implies that sellers of rare-event insurance are the fools of the story. The empirical record for index options points the other way on average: Bondarenko ("Why Are Put Options So Expensive?", 2003) found that S&P 500 puts were historically overpriced relative to their payoffs, so buyers of crash protection lost money on average over his sample.
The survivorship argument shows that chance can produce any given record; it does not show that a particular record is chance. The book gives no method for separating skill from luck in a real manager beyond estimating the population of competitors, which is rarely known.

The Black Swan + Antifragile + Skin in the Game + The Bed of Procrustes — Taleb's later books extend rare events into a theory of fragility and of who should bear risk.
Thinking, Fast and Slow — the research on heuristics and base-rate neglect that this book relies on.
Poor Charlie’s Almanack — a long, concentrated investment record to test against the survivorship argument.