Nassim Nicholas Taleb on Skin in the Game
What this covers
Nassim Nicholas Taleb speaks with Russ Roberts in this EconTalk episode about a principle Taleb developed with Constantine Sandis: the argument that decision-makers must bear the consequences of their own choices. The conversation centers on why skin in the game—the exposure of decision-makers to downside risk—works as both a practical tool for preventing reckless behavior and a moral requirement. Taleb's core claim is that when someone can profit from a choice while leaving others to absorb losses, they have no natural incentive to be cautious. Only personal exposure deters the kind of risk-taking that destabilizes systems. This logic applies across banking, policy, forecasting, and professional advice.
The discussion moves through several distinct territories. Taleb distinguishes fat tails and tail events from ordinary occurrences, arguing that forecasters and economists systematically underestimate rare but catastrophic outcomes and should therefore never advise on harm without bearing their own losses. He rejects the notion that frequent small wins prove sound decision-making in fat tails, where a single massive loss can erase years of gains. The law of large numbers reveals truth much faster at local, contractual scales than at government scales, making centralized policy ill-suited to the task. Historical examples—Hammurabi's Code, ancient military leadership, Adam Smith's Impartial Spectator—anchor the principle across domains. Taleb also examines optionality and asymmetric payoffs in hedge funds and banking, showing how compensation structures can hide true risk, and argues that Antifragility systems depend on making mistakes local rather than catastrophic—a structural feature, not a regulatory one.
Taleb argues that 'skin in the game'—forcing decision-makers to share in the downside of their decisions—is both the most effective mechanism for managing tail risk and a moral imperative, and that modernity has stripped it away, especially for bureaucrats, bankers, and forecasters who impose risks on others without bearing them.
- Asymmetric payoffs let agents hide tail risks while collecting steady gains, so only personal downside exposure deters reckless risk-taking.
- The law of large numbers reveals the true mean far faster at the micro (local, contractual) level than the macro (governmental) level, favoring small-scale skin-in-the-game systems.
- Forecasts and predictions are immoral in the harm domain unless the forecaster is personally exposed to being wrong.
Skin in the game is a moral imperative that binds ethics, risk management, and civil society into one inseparable principle.
- It is immoral to make a forecast or give risk-relevant advice unless you stand to be harmed if it is incorrect, because people take risks based on others' advice; this extends skin in the game from contracts into a moral imperative confined to the harm (negative-tail) domain.
“I should not make a forecast unless I am harmed by it... it is immoral for me to say, well, the market is going up or the market is going down or this will happen, unless I stand to lose on that advice. Because people take risks based on other people's advice.”
- You cannot disentangle risk management from ethics, nor economics from moral philosophy, because the same skin-in-the-game principle is simultaneously the foundation of moral philosophy, ethics, civil society, and risk management.
“This is why you cannot disentangle risk management from ethics, you cannot disentangle economics from moral philosophy. Some people try it; they can't.”
- Across history, prestige and honor have been proportional to the risks one takes for the sake of others—heroes in war, medieval lords obligated to protect peasants and die first, Hannibal and Julius Caesar first in battle—and a society cannot function with an imbalance favoring those who make others take risks for them over those who take risks for others.
“the prestige that we have gotten has almost always been proportional to the risks that you take for the sake of others.”
- People who talk about events without facing harm—writing papers without being active—don't understand the real texture of reality, whereas having skin in the game makes you more scientific and more rigorous about risk.
“people who talk about events without harm with impunity, write papers and do this, people who are not active, don't really understand the real texture of reality. When you have skin in the game you are much more in a way scientific about things.”
- Skin in the game means you cannot take risks that may harm others without being subjected to some personal harm yourself—not necessarily equal harm, but enough to act as a deterrent.
“Meaning I cannot take risks entailing that might harm others without being subjected to them myself. That's what I meant by skin in the game.”
Without skin in the game, decision-makers game binary reputation systems and pursue high-probability, low-impact gains while accepting rare catastrophic losses.
- If you have skin in the game your outcome is never binary because what matters is the amount; if you don't have skin in the game you will game the reputation system, which is based on binaries because others' perception is binary—connecting the two papers.
“If you don't have skin in the game, you are going to game the reputation system, and the reputation system is based on binaries. Perception of others is more binary.”
- Without skin in the game, the optimal strategy for a money manager is to pursue high-probability, low-impact gains and accept low-probability, high-impact losses—picking up nickels in front of a bulldozer—because the steady wins attract more capital and reputation until an eventual blowup wipes everything out, with the manager keeping the bonuses.
“So someone is shooting of course for a high probability, low impact benefits and low probability, high impact losses. If he doesn't bear the losses, that's the optimal strategy.”
- In asymmetric, fat-tailed processes (like a Pareto 80:20 distribution), roughly 90% of observations fall above the true mean, so an agent shooting for frequent small gains and rare large losses can show profits for nine years in a row while the true mean is negative—the statistical properties cannot be observed from a small sample.
“in any given year, you will have 90% of the observations above the true mean.”
- Bankers are long an option—their volatility either breaks even or makes them money without harm—while society is short that option: if bankers make money society gets nothing, but if they lose money society pays via taxes, more regulation, and regulatory burden, making the bank-society relationship mathematically identical to shorting volatility.
“the class of people who are long an option, namely bankers, their volatility is either break even or make money, they are not harmed; and society is short an option--basically, if the bankers make money we don't get anything; if they lose money, we end up paying for it via taxes, more regulation.”
Localized, decentralized systems naturally enforce skin in the game through direct enforcement and community shame, whereas large centralized governments lose this mechanism.
- The goal of antifragility is to build a system where mistakes can occur but do not threaten the system itself, which requires relying less on centralized public policy and more on localized systems that naturally enforce skin in the game.
“my whole idea in fragility, in Antifragility, was to build a system where mistakes can occur but do not threaten the system.”
- The larger and more convoluted government becomes, the more it attracts people who have everything to benefit from causing you harm—lobbyists—who have interest in the game while bearing none of the harm.
“The larger and more convoluted government is, the more the [?] in the game; and the more they have interest in the game, is that people who have everything to benefit from causing you harm, namely the lobbyists.”
- Skin in the game is a heuristic that can be enforced between two individuals in a contract without recourse to the state (e.g., 'I can eat your food if you taste it in front of me'), which fundamentally distinguishes it from regulation that requires a centralized state.
“That heuristic can be enforced between two individuals engaging in a contract. Regulation requires a state. This is the main difference here, is that we are bypassing the state.”
- Small systems naturally produce skin-in-the-game rules (helicopter mechanics forced to ride, parachute folders asked to jump), but there is a size effect: when systems get large they lose skin in the game, lose warped-incentive corrections, and prefer regulation over simple heuristics.
“when things get large there is a size effect. Large is ugly. And large doesn't have skin in the game.”
- The law of large numbers works much faster at the micro level than the macro level: locally there are thousands of experiments per day allowing quick identification of cause and effect, while at the macro level it takes thousands of years to determine whether a repeated-event decision was right—making skin in the game a micro concept and Washington a macro concept.
“the law of large numbers works much faster at the micro level than at the macro level. At the macro level you need thousands of years to figure out if someone, if for a repeated event someone made the right decision.”
- Successful localized governments (Switzerland, Sweden) work because decision-makers are embedded in a community where their skin in the game is shame from being socialized members, whereas a Washington bureaucrat faces only a spreadsheet, not a community member in church making him feel shame.
“people making the decisions are usually embedded in a community. And their skin in the game is typically shame. Because they are socialized by the community.”
Fat-tailed distributions mean single extreme events dominate outcomes, making magnitude and expectation matter far more than probability for forecasting and strategy.
- In a fat-tailed series, the single extreme observation (the maximum) eventually corresponds to the totals—everything is dominated by one observation—so for someone playing the tail game over thousands of days, a single event determines the entire outcome.
“in a fat-tailed series, eventually the maximum you will have will be equal to the mean. To the totals.”
- Binary events take a value of 0 or 1 (the event happens or it doesn't), while vanilla (natural/ecological) events have a magnitude dimension—the market can go down 0.1% or 20%, a war can kill 5 people or 5 billion—and the critical error is judging forecasters on how frequently they are right when, for fat-tailed variables, only the magnitude (expectation) matters.
“Most predictions are binary, in the sense that binary can take a 0 or 1. The event can happen or not happen. In real life we have a third dimension, the [?] event.”
- The real world is about expectation (probability times magnitude), not probability alone, so it can be rational to be positioned short the market even when the market is more likely to go up—because if it goes up it goes up small, but if it goes down it goes down big.
“The real world is about expectation, not probability. In other words, probability times how much it goes down.”
- Black swans are not more frequent than people think—they are less frequent, with lower probability than expected—but when they happen their effect is much deeper, which is precisely what makes them vicious.
“people thought that I was saying that black swans are more frequent. My point, no. They are not more frequent; they are less frequent. But when they happen, they are deeper.”
- Economists understand the agency problem and moral hazard but not moral hazard with respect to fat tails, because they treat outliers as a 'peso problem' to be ignored rather than as a central component of life, and there is a mismatch between compensation periods and the time the law of large numbers requires for a variable to reveal its true properties.
“while economists understand the agency problem, they understand moral hazard. But they don't understand moral hazard with respect to fat tails because they never really gave too much thinking to fat tails other than thinking, well, it's a peso problem.”
Financial institutions and money managers operate with asymmetric payoffs where they keep profits but externalize losses, demonstrating the 2008 crisis was not about overpaying for protection but about absent accountability.
- The 'long shot bias'—people overpaying for lottery tickets and insurance—is real only for bounded payoffs; it is a methodological error to generalize from experiments with bounded payoffs to natural settings with open-ended payoffs, as proven by banks losing roughly $5 trillion in 2008, which shows they were not overpaying for financial protection.
“you cannot generalize from an experiment that is not natural to natural settings, and you identify exactly the mistake people make because you are making a statement derived from a variable that has a bounded payoff and generalizing to things that have open-ended payoffs.”
- Banks lost approximately $4.7 trillion during the financial crisis—more than they ever made in the history of banking—demonstrating that a string of profitable quarters is irrelevant when the true mean is negative.
“Banks lost $4.7 trillion during the Crisis.”
- Hedge fund investors get a smaller share of profits than of losses because the manager keeps profits as his own money while the investor pays for losses, and combined with high fees this optionality means investors lose money in the long run—unless corrected by forcing managers to keep much of their own money in the fund.
“So you get smaller share of profits than you do of losses. In the long run you are going to lose money.”
- Sharing losses through skin in the game is liberating for both parties: the manager who has lost 10 to 50 times more of his net worth than the client feels no guilt when reporting losses, and the client sleeps well knowing the manager holds the same portfolio—eliminating the guilt and conflicts that arise without shared exposure.
“you are not uncomfortable because I lost 10 times more, at least 10 times, up to 50 times more, share of my net worth that day than he did from the loss. So, it's liberating on both sides. You don't feel guilt if you have skin in the game.”
- Capitalism has a built-in asymmetry where bankruptcy bounds losses at zero, but skin in the game can still be maintained by forcing decision-makers to lose a painful amount—not unlimited and not matching others' full exposure, but enough to be a real disincentive beyond the cosmetic.
“Meaning I cannot take risks entailing that might harm others without being subjected to them myself. That's what I meant by skin in the game.”
Historical societies honored those who took risks for others, but modern society perversely rewards those who make others take risks for them.
- The decline in willingness to take risks for others is due to technology and modernity disrupting the system, replacing risk-taking with the bureaucrat—a recent, abrupt cultural change rather than a gradual one driven by rising wealth.
“modernity, put the bureaucrat in place of risk-taking.”
- Modern society dishonorably rewards a class of people who make others take risks for them (citing George W. Bush escaping war and engaging others in one, and Petraeus's decorations despite never being in battle) while failing to honor entrepreneurs and risk-takers, instead putting bureaucrats and Harvard grads on the pedestal—an imbalance under which society cannot evolve.
“we have a generation of people who have never had to take risks for the sake of others.”
- Entrepreneurs take risks for the sake of others because their probability of success is much lower than that of venture capitalists, yet collectively society needs them to advance; the Amazon example shows investors and founders bearing all the early losses while customers reaped the gains.
“risk-taking is entrepreneurship... they take risks for society. And they save a lot, but collectively we need them because otherwise we can't advance.”
Economists lack meaningful skin in the game for their forecasts, so reform requires isolating their errors to academia rather than expecting accountability to society.
- Since the system lets economists be wrong with total immunity and they cannot have meaningful skin in the game, the way to reform economics is to build a society that keeps economists' mistakes local on campus—insulating the rest of society from them like an ivory tower—rather than expecting economists to be accountable for forecasts.
“The best way to do it is build a society in which mistakes made by economists stay on campus. That's the idea... if Larry Summers wants to make mistakes, more mistakes, let him make them at Harvard where we are insulated from it.”
- Hammurabi's Code (the architect put to death if his house collapses and kills the owner) embodies skin in the game because the builder can hide delayed risks by cutting corners in the foundation, and no inspector or regulation will ever outperform that simple rule forcing the builder to bear the consequences.
“The architect can hide their risks, delayed blowups, because the cheapest, the place where you can hide the risk is by cutting corners in the foundation... no inspector, no regulation, will ever ever outperform that simple rule.”
- Prediction markets do not hedge you because a prediction market is binary while real exposure is open-ended; events like wars (predicted to last two weeks but lasting years and killing billions) show that in fat-tailed domains the event is not defined by occurrence but by magnitude.
“when we talk about prediction markets, people think that prediction markets hedge you. They don't hedge you. A prediction market is a binary; and an exposure is open-ended.”
- Isocrates produced a version of the Golden Rule applied not to individuals but to states: a stronger state should not do to a weaker state what it would not want a still-stronger state to do to it.
“he applied it to states. And he said: States should not do to other states, weaker states, a strong state should not do to a weaker state what he wouldn't want a stronger state to do to it.”
Child development and institutional design both require periods of exposure to harm proportional to capacity, enabling learning and adaptive resilience.
- Good parenting maps onto fragility and antifragility, captured by a Lebanese expression: in the first 7 years you protect children because they are fragile, in the second 7 years you let them get in trouble because they are antifragile and learn only through skin in the game, and in the third 7 years you advise them on how they got in trouble.
“The first 7 years you protect them, because they are fragile. The second 7 years are antifragile; they need to get in trouble because they never learn unless they have skin in the game.”