Ergodicity — the gap between ensemble and time

The core distinction

A system is ergodic if the average outcome across many people equals the average outcome for one person over time. Most of statistics and economics assumes ergodicity. The problem is that real life — especially investing — is not ergodic.

The coin flip game above makes this concrete. Heads +50%, tails −40%. The expected value is positive: 0.5 × 1.5 + 0.5 × 0.6 = 1.05. A statistician says "good bet." But the geometric mean — what actually happens when you compound — is √(1.5 × 0.6) ≈ 0.949. Less than 1. Every round you lose about 5% on average. Play long enough and you go broke with certainty, regardless of how many heads you flip.

Why the ensemble lies

When economists calculate "expected returns," they average across a hypothetical population of parallel-universe investors. A few get lucky early and compound spectacularly — dragging the mean up. But you are not a population. You are one person, living one sequential path through time. The lucky outliers don't help you.

This is what the two charts show side by side. The ensemble mean looks fine. But the median player is quietly going broke. And the time chart shows what actually happens when you play the game yourself: ruin is the absorbing state — the place you can fall into but never climb out of.

What Taleb does with this

Taleb's argument is that classical economics built its entire framework — utility theory, portfolio optimisation, even much of Kahneman's prospect theory — on the ensemble assumption without questioning whether it applies to individuals living through time. He considers this a civilisational-scale intellectual error.

The practical implications:

• Ruin is categorically different. A 10% chance of losing everything is not just a large expected loss — it's a potential exit from the game permanently. Once you're ruined, the future paths where you would have recovered don't exist for you anymore.
• The barbell follows directly. Keep the bulk of capital in genuinely safe assets where you cannot lose everything. Take aggressive positions with the remainder — those can go to zero without ending the game. The "diversified moderate risk" middle is actually the most dangerous zone: it looks safe statistically while still carrying tail ruin risk.
• Volatility is not risk; ruin is risk. Modern portfolio theory treats variance as the measure of risk. Taleb's counter: variance is irrelevant if it doesn't threaten your ability to stay in the game, and catastrophic if it does.
• Skin in the game is an ergodicity argument. Advisors who recommend strategies they don't personally follow operate in ensemble-world — they see the average across all clients and call it acceptable. They don't live through any individual client's ruin.

The deeper point

Ergodicity breaks down anywhere absorbing states exist — states you can enter but not leave. Bankruptcy. Death. Reputational collapse. The single most important question about any risk is not "what's the expected value?" but "does this path lead to an absorbing state?" If yes, no expected value calculation is relevant, because expected value assumes you'll be around to collect on the average.

The standard economic framework, built for institutions and populations, keeps getting applied to individuals. Institutions are roughly ergodic — they survive individual bad outcomes, pool across many sequential bets, outlast any single event. Individual humans are not. We each get one path. That asymmetry is what Taleb thinks the entire profession missed.