
Nassim Taleb on Black Swans, Fragility, and Mistakes 05/03/2010
What this covers
This is a one-on-one conversation between Russ Roberts and Nassim Taleb, recorded in 2010 in the aftermath of the financial crisis. It extends the ideas of The Black Swan into a broader inquiry about what makes systems break and what lets them survive. Much of it is organized around convexity, which Taleb describes as risking a penny to make a dollar, or as an option: knowing where your ignorance will lead you. He ties it to evolution as a "free option" and to trial and error, where many small failures produce knowledge.
The middle of the conversation is about forecasting. Taleb separates tractable, roughly linear processes, such as tomorrow's weather, from complex nonlinear domains where errors compound over long horizons. He argues that a larger sample doesn't help when the generating process has changed, since new data mostly reveal fat tails. He also holds that a zero-probability estimate is paradoxical, because it implies certainty about the model. Roberts brings in Ed Leamer's critique of econometrics, which concerns hidden specification searches. The human side comes up through dread risk, where evolved paranoia leads people to fear sharks while ignoring driving.
The more contentious turns concern institutions. Taleb says optimization strips out redundancy, that bailouts and regulation remove the natural feedback loops that punish error, and that large banks survive only because of the state. He pushes back on Vernon Smith's view that markets discipline mistakes, and he describes moral hazard through bonus-earning bankers whose losses fall on ordinary taxpayers. He then widens the argument to medicine, claiming that interventions historically increased mortality and that gains came mainly from antibiotics and nutrition. He makes a similar claim about science, that knowledge flows from business and technology to science rather than the reverse. The description also lists exercise, debt, insurance, and the burdens of fame as topics.
Taleb argues that modern systems are fragile because optimization, forecasting, and state intervention suppress the small errors that would otherwise keep them robust, so people should build redundancy and convex, trial-and-error structures rather than rely on forecasts.
- Forecasts of rare events are unreliable, and errors grow explosively in the fourth quadrant where consequences are severe
- Optimization removes redundancy, while redundancy and buffers keep systems stable
- Government bailouts and regulation remove the feedback loops that would punish mistakes and teach lessons
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
Errors in an option pricing model usually raise the value of the option, because uncertainty about the model creates a second-order effect that is convex: if the model is wrong the option gains, so model error is an asset for option holders.
“the errors in pricing an option usually help you so there's a second order effect always raises the value of not of any option”
Evolution is a free option: nature produces variation through trial and error, with many failures such as spontaneous abortions, and growth of knowledge and anything that works comes from the same process of small failures.
“evolution works because it's a free option is compacts you know that a mother nature doesn't produce it knows it can't produce a perfect baby”
Forecasting based on a larger sample does not reduce error when the process generating the data has changed; the apparent narrowing of standard errors is an illusion, and what the new data reveal is more fat tails.
“we don't have anymore data because the process that created the data is different than the process that existed before”
Convexity is the central idea of Taleb's work: losing a penny to make a dollar, or being wrong costing less than being right gains; equivalently, knowing where lack of knowledge will lead you, which is what an option is.
“convexity is when you lose a penny to make a dollar and convexity is when you're being wrong doesn't cost you as much as you get being right”
Humans are wired for dread risk: paranoia and overestimating rare, high-consequence events has survival value, but in modern environments it misfires, making people fear sharks at San Diego beaches while driving is far riskier.
“people are afraid of sharks in San Diego not realizing that trips to the beach is vastly riskier”
Any estimate of zero probability for an event is paradoxical, because a zero estimate implies certainty about the model; with stochastic parameters the expected value of a zero-probability event rises above zero, so more uncertainty about the model raises the expectation.
“if someone tells you I have zero probability that event is has zero probability there some of us as you probably know yes”
Medicine has historically had a bad track record: interventions such as bloodletting and doctor visits increased mortality, and the medical establishment's bias toward acts of commission over omission (therapeutic nihilism) led to harmful action.
“people called it therapeutic nihilism as I was saying it was always better to do some things and do nothing”
People seek comforting experts because they want someone to take care of them, the same way patients seek doctors, even though seeing a doctor statistically worsened outcomes in the era before penicillin.
“seeing a doctor statistically was not a good thing ... increased your chance of death”
What matters for decisions is the expected consequences of outcomes, not the probability that a claim is true or false; hyperinflation can destroy civilization, so the consequences of rare outcomes dominate the probabilities.
“it's the consequences of x and y that matter”
Forecasting errors grow monstrously over longer horizons: a company can reasonably forecast inventory a few days or months out, but a ten-year projection has an error rate several billion times larger, so the question is what is worth forecasting.
“forecasting something 10 years down the road is has an error rate that is like several billion times your error rate you may have for a five-day forecast”
Economist Edward Leamer's 'Let's Take the Con out of Econometrics' recommends testing the range and sensitivity of coefficients across all model specifications, because researchers' hidden search across variables means the published model is not the true one.
“what Lima recommends is looking at the range and sensitivity of the coefficients when we try all the different combinations”
Forecasting models are unreliable for small probabilities because such probabilities cannot be measured empirically, and even a small calibration error explodes when applied to small-probability events, so people must be protected from them.
“small probabilities are not measurable did not measurable empirically and even if you have the right model small calibration error explodes”
The state has been tampering with the functioning of the market, so that large companies that would otherwise self-destruct survive while small robust units and cautious companies, which learned from failure, are the ones that should survive.
“the state has been tampering with the functioning of the system ... the ones who survives are the small robust units”
Optimization-based economics reduces the set of considered risks to a narrow class, so it makes systems more fragile to perturbations, and its simple rules of thumb teach specialized 'optimal' behavior that removes the redundancy needed for robustness.
“optimizing economics whether it takes a very very reduced class of things and they optimize so it makes you a lot more fragile on those perturbations”
Vernon Smith argues that markets punish people who make mistakes, but Taleb argues that government has removed these natural feedback loops, so people keep repeating the dangers of debt.
“we've stopped markets from punishing people we've stopped the natural feedback loops that would encourage people to learn from the dangers of debt”
Government-sponsored research has led nowhere except marketing; historical evidence (McCloskey, Edgerton) suggests the arrow of knowledge runs from business and technology to science, not from science to technology.
“it's not science technology business it's business technology science”
Modern medicine has improved outcomes mainly through crime reduction, better nutrition, and antibiotics such as penicillin, rather than through a deep understanding of the human body, which the medical establishment has never fully grasped.
“we never understood the human body mother nature has been better than biologists at understanding the human body”
Government intervention has prevented the market from destroying large firms: large banks like Citi would have been destroyed in the 1982-83 crisis but survived thanks to the state, and the vices of firms are compounded under state protection.
“the same thing with banks city banks would have been destroyed in 1982-83 yeah easily and it survived thanks to the state”
Modern medicine and technology extend lifespan but make people sicker through diseases of civilization, because humans have eaten unnatural foods and lived in unnatural smooth environments, so interventions tend to correct symptoms rather than fix the underlying mismatch.
“technology makes us live longer planet but it punishes us by making making us sicker”
Trial and error, with many small failures, generates knowledge and growth, while top-down design and government-sponsored research tend to inhibit trial and error and have produced little beyond marketing.
“government funding of research has led nowhere except for marketing”
Regulation makes lawyers rich and encourages banks to take hidden risks, and banks have a vested interest in growth, so letting banks fail would shrink the role of financial institutions organically without regulation.
“regulators you know allowed actually encourage banks to take his at risk was an antic less visible risks”
Because the structure of randomness has become less constrained by natural limits, quantitative forecasting models are impossible to use for tail events, except for vague 'Gray Swan' type generalities.
“it's impossible to use really models to forecast that tails okay except for very vague things I called grey laws”
Forecasting simple linear or mechanical processes is reliable (tomorrow's weather from clouds, seasonal temperature patterns), but in complex nonlinear domains like climate the same forecasting is unreliable, since the errors grow in a vicious way.
“a lot of nonlinearities I'm very good I'm very confident that June July and August will be on average warmer”
Monetary policy under nonlinearities is dangerous: policymakers keep adding money with no result until hyperinflation, and governments should not be given toys they do not understand.
“you keep adding money with no result until there is hyperinflation or nothing government should not be given toys they do not understand”
Moral hazard in finance is severe: someone earning 200 million dollars in bonuses can blow up while the hairdresser and schoolteacher pay for the losses, and the models used showed no risk.
“someone who makes 200 million dollars in bonuses blows up and the hairdresser and the schoolteacher pays for it”
Humans are poorly wired to reason about the third dimension of uncertainty, which combines the probabilities of outcomes with their consequences and expected values, but we have a survival-advantageous mechanism for dread risk that overestimates rare, high-consequence events.
“our brains don't we really struggle as Homo sapiens to deal with that third dimension of uncertainty”
Hunter-gatherer energy and food intake were not steady bell-curve distributions but lumpy and feast-or-famine; humans are adapted to irregular inflows of food and periodic stress, so steady modern diets remove stressors the body evolved to handle.
“the distribution of energy expenditure for a hunter-gatherer that's not steady that's not mediocre Stan”
Convex trial-and-error, where losses are capped and gains are unbounded, lets nature exploit options, and this is why systems that rely on trial and error do better than designed plans that look smart.
“convex Locke is trial and error what cost you hurry that rules you exploiting the option of nature”
In the fourth quadrant, which combines extreme consequences with fat-tailed, complex conditions, one cannot use any map and must instead rely on robustness, because small errors can cause catastrophe.
“this is the fourth quadrant this place monsters lie exactly and in the fourth quadrant you cannot afford to use a map wrong map because small little errors are going to crash a plane”
The right goal is a society in which human error is not penalized severely, because small mistakes function like the price of an option paid for information and discovery, while large mistakes are crippling.
“incur small mistakes is like options it's our discovery you don't even know if it's a mistake”
Economists are appealing despite repeated failure because people desperately want comfort and insurance, and someone with a model feels like someone is taking care of them, just as patients seek doctors even when the care is harmful.
“it's this desperate desire to find some comfort some insurance some someone's taking care of us someone has a model”
Experts' seductive appeal is linked to a preference for top-down design over trial and error, and politically this preference for plans is deep, even though letting losers be weeded out via trial and error is often superior.
“the appeal and the seductiveness of experts is also the appeal of design rather than trial and error”
A pilot who cannot find a map of New York should use a map of Chicago instead of none, so the claim that some map is better than no map is false when the map is wrong for the terrain, and a wrong map can crash a plane.
“you would cancel your flight no you have to train anybody equipped with a brain would do that”
A company's projected value rises when its sales forecast is made stochastic, because uncertainty around a correct average estimate raises its expected value; the more uncertainty one has about the model, the higher the expected value.
“the value of the company may rise dramatically because error around that estimate may raise the value of the company”
Dread risk aversion has survival value: paranoia about certain hazards, such as the Large Hadron Collider, can serve future survival even when specific fears are overestimated, so the bias is an adaptive mechanism.
“so paranoia has a survival advantage so does that explain”
Professors and education focus on 'know-what' (the epistemic, codifiable content) rather than 'know-how,' and history consists of small stories and tricks that cannot be compressed into an equation, so people forget it.
“they don't focus on the know-how they focus on know what and history is an accumulation of small stories and tricks you can't compress into an equation”
Overconfidence in forecasts of returns causes accumulation of debt: a person who believes they will earn 8 percent while borrowing at 5 percent will leverage heavily, and because error rates are left out of the calculation, leverage looks optimal.
“overconfidence translate[s] into accumulation of debt”
The random variables in the world have become fatter-tailed since roughly the rise of the Internet, because interdependence between people means independence assumptions break down, so central limit reasoning no longer applies and extreme events become more likely.
“random variables became a lot more fat-tailed thanks to the Internet”
Human cognitive biases are not harmful in their natural environment; they become harmful because governments create an unnatural background in which the feedback that normally corrects them is absent.
“I don't think that human biases are a harmful the National Environment I think they exist they just don't exist in an unnatural environment”
Large companies become disproportionately more fragile to unseen shocks as they grow: a rogue trader's 50 billion dollar position had to be liquidated at much higher per-unit cost than a five million dollar position, so concentration makes errors vastly more costly.
“when you concentrate companies errors because a squeeze squeezes are vastly more costly to the company”
The body should not have all stressors eliminated; living on smooth surfaces with air conditioning, cars, and elevators removes good stressors and makes people weaker than their ancestors, which is very risky.
“we have had the reduction of good stressors and and that is that is very risky because we are weaker than our ancestors”
Like the human body having two kidneys, a robust system needs redundancy and insurance; an economist-designed single 'central kidney' would be efficient but fragile.
“an economist would just have one central kidney and people would borrow it because it's vastly more efficient”
Redundancy is the opposite of leverage: to protect against uncertainty one needs buffers, such as personal savings exceeding spending or government surpluses, rather than relying on taxpayers as a backstop.
“you need to create some redundancy you need to create some sort of backstop other than the taxpayer”
Forecasters should be asked for their past forecast record before their current forecasts are taken seriously, since an economist who was wrong for years is still being listened to.
“anyone in a room who's listening to this forecast okay without asking for what his forecasts for 2008 and 2009 were”
A barbell strategy, with about 80 percent of capital in very safe holdings and about 20 percent in highly risky bets, produces asymmetric upside while keeping downside limited, the opposite of a leveraged bet on moderate risk.
“I hide eighty what I called my barbell flashy in the Black Swan I hide eighty and then I think a lot of risk was twenty”
The real challenge is deciding what is worth forecasting and what to insulate yourself from; since errors grow and tomorrow may not resemble yesterday, one needs redundancy and a backstop rather than giving up on prediction.
“you need to create some redundancy you need to create some sort of backstop other than the taxpayer”
Philosophy and logic have been consequential truncated to two dimensions: they ignore the third dimension of consequence and expected value, making much of formal logic unusable for real-world uncertainty.
“this truncation of dimension in philosophy and epistemology has been very very consequential”
Expert forecasters should be discredited because their predictions damage society; Taleb's position is that people like the IMF forecaster should be made irrelevant so that their errors do not harm innocent people.
“the first thing I gotta do is get rid of expert problem because it makes the society more fragile”
Knowledge of fragility is ancient but gets forgotten: people have understood how fragile systems fail since the Babylonians, and the lesson was codified in Islamic and Roman practice before being lost again.
“we've known about the topic since the Babylonians”
Nutrition science faces the multicollinearity problem: correlated dietary and lifestyle factors make it hard to determine which one is responsible, so researchers can easily pick the wrong cause.
“multicollinearity is one of those words and it just means some things go together and make it hard to figure out which ones responsible”
Convexity is knowing where lack of knowledge will lead you, since averaging over uncertain estimates is not the same as the expectation of the average, which is what Jensen's inequality captures.
“an average of expectations is not the expectation of an average”
Leverage and debt are seductive because they increase returns in good times but leave institutions vulnerable to collapse, and this is the Kahneman view of hubris and myopia applied to finance.
“you talk about the role of leverage and debt and how its seductive and it increases your returns but it leaves you vulnerable to collapse”
The belief that basic research will not be done without government funding is a myth: discoveries such as the steam engine and evolution-era innovations came from businessmen and tinkerers without theory, so incentive for private discovery exists.
“the myth is definitely that basic research won't be done by individuals that there's no money in it”
The AIG bailout sent government money to Société Générale and Goldman Sachs, which were its largest creditors, and AIG's counterparties were protected by a triple-A rating that did not reflect its ability to pay.
“the government bailed out AIG in September of 2008 the number-two creditor was Goldman Sachs”
Standard statistics and forecasting methods work in the first three quadrants, but the fourth quadrant is where the most important mistakes occur, and the lesson for policy is to avoid relying on forecasts in that domain.
“all the mistakes besides forecasting comes out the fourth quadrant”
Instead of telling people their methods fail, the persuasive approach is a map of quadrants showing where conventional statistics and forecasting work well and where they fail, because people are not offended when told where they are competent.
“instead of telling people this is what doesn't work people get offended”
The standard advice of three meals a day has never been tested, and periods of starvation may be healthier for humans, based on his reading of evolutionary and empirical evidence.
“people tell you to us three meals a day and in fact no animal eats to work out you see”
Dietary guidelines such as the food pyramid favor carbohydrates over saturated fat, and the resulting diets may cause disease clusters (heart problems, blood pressure) that the official guidance does not address.
“they want you to eat cards and they don't want to eat saturated fat”
Large corporations capture the state and dictate policy because they claim to employ many people, which leads to socialism-like collusion with the government, and small firms are disadvantaged.
“most corporations eventually become start dictating to the states what they want”
Personal redundancy means spending less than you earn to keep a buffer, and governments should likewise run surpluses instead of letting deficits grow; running deficits just to avoid bigger deficits is sailing too close to the wind.
“governments should suit for positives revenues me and surplus surplus food for surplus”
Efficiency gains from scale create a trade-off: it is more efficient to be large, but large entities are more exposed to Black Swans, so natural selection should destroy large fragile firms like the dinosaurs.
“it's more efficient to be large but Black Swan strong types walks unseen you know are gonna hit you harder than other people”
The belief that more data reduces forecast uncertainty is a deep error, because the sample size advantage assumes stable processes, and it leads institutions to trust forecasters who have repeatedly failed.
“the whole essence of the econometrics priesthood is the idea that the bigger the sample the more reliable the forecast”
Forecasting is misguided in complex domains, but the right response is not to discard all maps: a map of where forecasting techniques work is more useful than a map that pretends to work everywhere.
“I'm plane in Atlanta ... we're going to use the map of Chicago it worked”
Human biological intuition is limited: the medical establishment has not understood the human body, and nature has been better than biologists at designing it, since humans tried to understand organ functions by rationalism rather than evidence.
“mother nature has been better than biologists at understanding the human body”
Family-owned companies survive in the long run and natural systems survive because they are robust to perturbations, and this robustness comes from redundancy that can be reproduced by copying nature.
“the things that are robust on perturbations ... look at mother nature”
Evolution does not give second chances: nature destroys what fails and does not provide do-overs, and Taleb does not want to be given a second chance either after surviving a crisis.
“evolution does not preach you know what it does it destroys”
Taleb is trading hyperinflation as a convex bet: he knows his model has error that would be disproportionately amplified in a hyperinflation, so the trade pays off if the rare event occurs.
“I'm a trader of hyperinflation for example okay and my idea of hyperinflation stem from that”
Traditional wisdom such as a grandmother's advice to keep redundancy captures the core robustness lesson, so people need no advanced theory to see that fragile arrangements make life unstable.
“anyone who has a grandmother knows about the ills of debt that makes you fragile”
Government programs such as cash for clunkers and tax breaks for struggling firms transfer costs to taxpayers and reward those who lost money, creating moral hazard.
“those who bought a bad car get three thousand dollars back from the administration but those who bought a good car got nothing”
Strength training with heavy objects overhead is better than gym machines because machines isolate muscles, while natural movement engages the body as a whole and the emotional drive of effortful tasks matters for adherence.
“I don't use machines because you don't isolate a muscle or body part”
It is a mystery why the number-two official at the IMF is still listened to for forecasts when he has never been right, and why demand for economists' services has risen after the crisis despite their poor record.
“how can people still listen to the number two men of the IMF talking about his forecast in the future”
Government programs and bailouts reward the wrong actors: the cash-for-clunkers program paid those who bought bad cars but not those who bought good ones, and large companies end up dictating to the state, which Taleb calls socialism.
“those who bought a bad car get three thousand dollars back from the administration but those who bought a good car got nothing”
Taleb has avoided corruption by walking out of meetings, including a White House delegation where Bob Rubin was present, because he believes he has a mission and should not compromise his statements.
“I was supposed to be in a delegation to go see to go to the White House and what I looked around I saw Bob Rubin in it”
Regulation creates revenue for lawyers and consultants rather than safety, and the growth in Taleb's former illustrator business came from regulation that created revenues to comply with it.
“regulation makes lawyers rich I was at the illustrator okay it was most of the business we had came from a regulation”
Economics and forecasting is an 'n of one' problem in which the sample of nature is large, so a single observer's experience can be informative when the sample spans nature's full range of states.
“people tell you you have an N of one okay that but the end of nature is not an end of one and is the largest N you can get”
Government spending on war and cancer has produced less than spontaneous discoveries made outside that funding, and universities are better at public relations than at generating knowledge.
“universities are much better public relations”
The 2008 crisis taught Taleb little, while the 1987 stock market crash taught him more, because the 2008 crisis was not a novel event and history contains many comparable or larger shocks that people forget.
“I learned more from the crisis of 1987 from the stock market crash of 1987 than anything else”
The Congressional Budget Office projected in 2006 that Social Security payroll tax revenue would first fall below payouts in 2016, but the shortfall occurred in 2010, only six years off, showing forecasts are unreliable.
“they were only off by six years”
Taleb's central idea is convexity rather than the Black Swan, and the economics concept of emergent behavior shows that simple ideas can have deep implications that take a lifetime to grasp.
“the Black Swan is not the central idea the central idea is convexity”
A year and a half after the crisis, the system became worse, with larger financial institutions, larger state roles, more deficits, more fragility, and more dependence on economists who never got forecasts right.
“we're worse off because all the ills are compounded larger financial institutions larger state roles more deficits more fragility”
Government deficits persist because budget projections are marred by overconfidence: France has run deficits for 39 years, and forecasters at institutions such as the IMF project low deficits while Greece's actual debt-to-GDP is far higher.
“the French government has had now a deficit for 39 years”
Taleb's personal writing productivity improved when he made little effort most of the time and only applied intense effort to tasks that were natural to him, paralleling the stress-and-rest principle of homeostasis.
“I wrote in 2 or 3 weeks close to homeostasis so that editions of Black Swan”
Hospitals' advice to ice an injury is unsupported by evidence because Mother Nature heals injuries without ice, and the advice mainly benefits the ice makers.
“there's absolutely no evidence that I think is good for you it's good for the ice makers”
Taleb's own health practices, such as walking about 20 hours a week, lifting heavy objects overhead, and avoiding machines and smooth surfaces, produced large benefits, though he worries the results may reflect confirmation bias.
“I started walking a lot about 20 hours a week”