The Open Curriculum

The Mind · Level 4

The Black Swan + Antifragile + Skin in the Game + The Bed of Procrustes

Nassim Nicholas Taleb · 2007–2018

Thesis

The events that matter most are rare, extreme and unpredictable, so instead of trying to forecast them, people and institutions should be built to survive or gain from them, and anyone who makes decisions for others should bear the cost when those decisions go wrong.

The main ideas

Black Swans

In The Black Swan (2007), Taleb defines a Black Swan by three properties: it is an outlier, outside the realm of regular expectations; it carries an extreme impact; and human nature makes people concoct explanations for it after it occurs. His short form is rarity, extreme impact, and retrospective, though not prospective, predictability.

The mechanism is the problem of induction. Taleb's illustration is a turkey fed every day for 1,000 days. Each feeding raises the turkey's confidence that people look after it, and that confidence is at its highest the day before Thanksgiving, when it is killed. The turkey's evidence was consistent and long, and it said nothing about the event that mattered. Taleb takes the case from Bertrand Russell's chicken.

The retrospective explanation is what keeps the error alive. Once an event has happened, a story makes it seem foreseeable, which persuades people that the next one will be foreseeable too. Taleb calls this the narrative fallacy: the habit of fitting a causal story to a sequence of facts, which makes the world seem more predictable than it is.

Implication: a long, stable record is not evidence of safety when the risk that matters is rare. Ask what event the record could not have shown, and what it would cost.

Mediocristan and Extremistan

Taleb divides randomness into two kinds. In Mediocristan, no single observation can change the total much. In Extremistan, one observation can dominate it.

His thought experiment: line up 1,000 people at random and weigh them. Add the heaviest person alive, at between four hundred and five hundred pounds, and he adds about half a percent to the total. Now total the same group's net worth and add Bill Gates. He may account for 99.9 percent of it. Height, weight and calorie intake belong to Mediocristan. Wealth, book sales, city populations, casualties in wars and moves in financial markets belong to Extremistan.

The mechanism is that physical limits cap the size of observations in Mediocristan, while in Extremistan nothing caps them, so the largest event can be many times larger than any seen before. The bell curve describes Mediocristan well, and Taleb calls its use in Extremistan the Great Intellectual Fraud. For those domains he follows Benoit Mandelbrot's fractal randomness, in which large deviations are far more frequent than the bell curve allows.

Implication: before using an average, a standard deviation or a historical maximum, decide which kind of domain the quantity lives in. In Extremistan, plan around the largest possible deviation, not the typical one.

Silent evidence

The record that survives is biased toward the cases that survived. Taleb's example comes from Cicero. Diagoras, a nonbeliever, was shown painted tablets of worshippers who had prayed and then survived a shipwreck, offered as proof that prayer protects. He asked where the pictures were of those who prayed and then drowned.

The mechanism is that losers leave no record. Successful writers, entrepreneurs and gamblers are visible, and the larger number who followed the same methods and failed are not, so the traits of the winners look like causes of success when they may be common to the losers too. The same bias runs through history, which records the cities, species and civilizations that lasted.

Silent evidence also hides the cost of risk. People who survived a risky course of action conclude it was less dangerous than it was, because the cases that would show the danger are the ones that are missing.

Implication: before inferring a cause from a set of successes, ask how many others did the same thing and failed. When you have survived a risk, do not take the survival as evidence that the risk was small.

Epistemic arrogance and the Procrustean bed

People overestimate what they know. Taleb cites a test in which subjects gave ranges they were 98 percent sure contained the true value of various quantities. The true value fell outside the range 45 percent of the time.

The ludic fallacy, from the Latin for game, is the error of modeling real uncertainty on games of chance, where the odds are known. Taleb's case is a casino that spent heavily on surveillance and gambling-risk models; its four largest losses came from none of them, including a white tiger mauling a performer in its star show, at a cost of about $100 million.

In The Bed of Procrustes (2010; expanded 2016), a book of aphorisms, Taleb names the general habit after the host of Greek myth who stretched or cut his guests to fit his bed. Facing the unknown, he writes, people squeeze the world into crisp commoditized ideas, reductive categories and prepackaged narratives. His targets include defining intelligence as what classrooms can test and treating a salary as safety: "The three most harmful addictions are heroin, carbohydrates, and a monthly salary."

The mechanism is that a model can only price the risks its builders thought of, and the large losses come from outside it.

Implication: treat any risk estimate as conditional on the model that produced it, and ask what the model leaves out.

The barbell

Because Black Swans cannot be predicted, Taleb recommends arranging exposure to them instead of forecasting them. His barbell strategy puts 85 to 90 percent of assets in extremely safe instruments and the remaining 10 to 15 percent in many small, highly speculative bets with large potential payoffs. Nothing goes in the middle.

The mechanism is asymmetry. The safe end limits the worst case: whatever happens, losses are capped at the speculative portion. The speculative end gives exposure to positive Black Swans, rare events with very large payoffs, such as a venture that succeeds beyond any forecast. Middle-of-the-road investments look moderate, but in Extremistan their risk is badly estimated, so they can carry hidden exposure to large losses while offering little upside.

Implication: set a hard limit on what can be lost, then spread the remainder across many small bets where the gain is open-ended and the loss is limited to the stake.

Antifragility

Antifragile (2012) names the opposite of fragile. Fragile things are harmed by shocks, robust things resist them, and antifragile things gain from them. Taleb maps the three onto myths: Damocles, under a sword hung by a single horsehair, is fragile; the Phoenix, reborn the same from its ashes, is robust; the Hydra, which grows two heads for each one cut off, is antifragile.

His case for how variability protects is two brothers in London. John has been a clerk in a large bank's personnel department for 25 years and is paid the same every month. George drives a black taxi, and his income varies from day to day. John's income looks safer, but it is exposed to one large shock, being laid off, which would cut his income to zero at once. George's daily fluctuations keep him adapting and expose him to no single event that could end his income.

The mechanisms Taleb draws on include hormesis, in which a small dose of a harmful stressor strengthens an organism, and iatrogenics, harm done by the healer. From these follow via negativa, improving a system by removing things, such as unnecessary interventions, rather than adding them, and the Lindy effect: for nonperishable things such as books and ideas, every additional period of survival raises the expected remaining life.

Implication: prefer arrangements that reveal small losses often over those that hide a large one. When choosing between an old idea and a new one, weight the one that has survived longer.

Skin in the game

Skin in the Game (2018) argues that a person who makes decisions for others should share their downside. Taleb's model is the Code of Hammurabi: if a builder builds a house and it collapses and kills its owner, the builder shall be put to death, and if it kills the owner's son, the builder's son is put to death. The rule does not depend on anyone predicting which buildings are unsafe. It transfers the risk to the person best placed to know it.

The mechanism is that decision makers who keep the upside and pass the downside to others will take hidden risks. Taleb's case is what he calls the Bob Rubin trade. Rubin, a former Treasury Secretary, received about $120 million from Citigroup in the decade after he left the Treasury, and kept it when the bank was rescued by taxpayers in 2008.

Taleb extends the argument to what he calls the Intellectual Yet Idiot: credentialed experts, commentators and policymakers who are rewarded for sounding right and never pay for being wrong. Their errors persist because nothing in their position filters them out.

Implication: before acting on advice or a forecast, ask what the adviser loses if it is wrong. Give more weight to those who bear the consequences of their own recommendations.

Ruin and ergodicity

Taleb distinguishes two situations that people treat as the same. One hundred people each go to a casino once. Some lose, some win, and the average return for the group can be calculated; if one of them is ruined, the others are unaffected. One person goes to the same casino one hundred days in a row. If he is ruined on day 28, there is no day 29. The group average says nothing about the single person's outcome, because the sequence ends at ruin.

The mechanism is that ruin is absorbing: once it happens, no later gains can offset it. A strategy with a positive average outcome but any repeated chance of ruin will eventually reach it. For this reason Taleb argues that cost-benefit analysis, which weighs average costs against average benefits, does not apply to risks of ruin, whether for a person, a firm or a species.

Implication: separate risks that can be recovered from and risks that cannot, and refuse any exposure to the second kind regardless of the expected return. A small chance of ruin taken repeatedly is not small.

Limits

Related