Ergodicity

The last module showed that zero is a place you can never leave. This one shows why a bet that looks wonderful "on average" can still march you straight toward it — and why the word average hides one of the most expensive confusions in all of investing.

Here is the confusion, in a sentence. There are two very different questions you can ask about a risky bet. One: if a thousand different people each took this bet once, how would the group do on average? Two: if one person took this bet again and again, over and over, how would that single person's money do over time? Our instinct treats these as the same question with the same answer. They are not. And the number the market, the ads, and the back-tests usually quote you is the first — the crowd's average — while the life you actually live is the second.

The word for whether those two answers agree is ergodicity: whether the average across many people equals the average for one person over time. When a system is ergodic, the two match and you can trust the crowd average. Wealth compounding through a single life is not ergodic — the two answers can point in opposite directions — and mistaking one for the other is how people take bets that are "positive on average" all the way down to nothing.

Give the two questions their proper names, because keeping them apart is the whole skill.

The ensemble average is the result across many people who each take the bet once — the crowd, all playing in parallel, their outcomes added up and divided by how many there are. The time average is the result one person gets by taking the bet repeatedly, one round after another, each round landing on whatever the last one left them. The ensemble spreads the risk across many separate lives; the time path piles every round onto the same single account.

For some things these two are equal, and our intuition — built on those things — trusts them to always be. Your height, the temperature, the average age in a room: measure across many people or track one person's readings over time, and you land in roughly the same place. Those systems are ergodic. But money does not add up the way height does. Money multiplies. Next year's return lands on this year's balance, not on your starting balance. A 40% loss does not subtract a fixed rupee amount you can simply earn back elsewhere — it shrinks the base that every future return has to work on. When outcomes multiply like that, the time average and the ensemble average come apart, and they can come apart violently.

This is not a mathematical curiosity. It is the reason a bet can be genuinely, honestly positive on average and still be certain ruin for the person who keeps taking it. The crowd, playing once each, walks away ahead. The individual, playing forever, walks toward zero. Both statements are true at the same time, about the same bet — and only one of them is your life. This is why the average across many people is not your outcome across time, and why survival, not the expected value, is the number a single investor must protect.

Work the classic example slowly, because once you have seen it you cannot un-see it. illustrative

Here is the bet. You flip a fair coin. Heads, your money grows by 50%. Tails, it shrinks by 40%. Start with ₹100: heads takes you to ₹150, tails takes you to ₹60.

Now ask the ensemble question first. A thousand people each flip once. About half get heads and end at ₹150; about half get tails and end at ₹60. The group's average ending balance is (150 + 60) ÷ 2 = ₹105 — a tidy +5% per person. As a crowd, playing once each, they came out ahead. The expected value — the probability-weighted average payoff — is genuinely positive. Every advertisement for this bet would be true.

Now ask the time question, the one that is actually your life. You keep flipping the same pot. The rounds do not average — they multiply. One heads and one tails, in either order, gives ₹100 × 1.5 × 0.6 = ₹90. You are down 10% after two flips, and down again after the next two, and again after the next. Keep flipping and your single account grinds toward zero with near-certainty, even though every individual flip had that lovely +5% expected value. The crowd got richer; the repeat player went broke. Nothing about the bet changed — only whether the outcomes were spread across people or stacked through time.

The engine underneath is the multiplying, and its cruel partner is the absorbing state from the last module. Because losses cut the base that gains must rebuild, a string of tails does lasting damage that a matching string of heads cannot fully undo — 1.5 and 0.6 do not cancel to 1, they multiply to 0.9. Over enough rounds the bad runs compound faster than the good ones repair, and once the pot is small enough, ruin finishes the job.

Take this off the coin table and into a real book. illustrative

An investor is drawn to a high-octane strategy — heavy concentration, a dose of leverage, the kind of thing that can return 50% in a good year. The pitch quotes an average annual return of, say, 20%, and the number is real: averaged across many accounts running it, or across many simulated paths, it genuinely comes to about 20% a year. On the strength of that average he puts his whole ₹15 lakh portfolio into it and plans to run it for a decade.

Here is what the average hides. Among the many paths that produce that 20% figure, some soared and some were wiped out, and the survivors' triumphs are large enough to lift the group average even though a real share of individual accounts hit zero along the way. The 20% is an ensemble number — the crowd of possible outcomes, averaged. But this investor is not the crowd. He is one account, running the same leveraged, concentrated bet year after year, each year landing on whatever the last one left him. For him the returns multiply, the bad years cut the base that the good years must rebuild, and a single bad enough stretch — the kind the 20% average quietly absorbs — takes his one account toward the absorbing state, from which no future 20% year can retrieve him. The average was true for the group and irrelevant to his survival.

The repair is not to refuse every positive-average bet — it is to size and structure the bet so the multiplying cannot carry you to zero. That is precisely what position sizing and avoiding leverage do: they keep any single round, or bad run of rounds, from removing the capital the next round needs.

The two readings are not really in conflict once you see the frame. The 22% can be perfectly real and deadly to one repeat player. The question is never just how large the average is — it is whether the number is measuring the crowd or measuring you.

The idea is powerful, and like every powerful idea it is easy to over-swing.

It does not tell you to avoid all risk or never take a positive-average bet. Most sound investing is taking positive-average bets. The lesson is not "positive average is a lie" — it is "size and structure the bet so the multiplying through time cannot carry you to zero." A well-sized position in a good business is a positive-average bet you can survive taking repeatedly; that is the whole aim.

It does not, by itself, tell you which strategy is good. Ergodicity is a warning about how to read an average, not a method for finding an edge. It tells you to distrust a crowd number quoted to a single life; it does not tell you what to own instead.

And it is not a claim that averages are useless. Expected value is a real and useful number — for a one-shot decision, or for genuinely independent bets small enough that no single one dents the base, the ensemble and the time path stay close, and the average can be trusted. The danger is specific: repeated, compounding bets large enough that a bad run damages the pot the next bet stands on. That is where the two averages divorce, and that is the only place this warning bites.

The gap between the crowd average and your own path fools people in a few reliable ways. Named once, they are easier to catch.

1. Reading a crowd average as a personal promise. "It returns X% on average" describes many players or many paths, not the one account you will actually carry through time. The average is theirs; the sequence is yours.

2. Forgetting that returns multiply, not add. A run of losses does lasting harm because each one shrinks the base the next return must rebuild. 1.5 and 0.6 do not cancel — they multiply to 0.9. Sequences of compounding losses are worse than their average suggests.

3. Trusting an average taken only over survivors. When ruin removes players from the sample, the reported average flatters the strategy and hides the graveyard. Ask who dropped out before believing the number.

4. Confusing a one-shot bet with a repeated one. Expected value is trustworthy for a single, independent, small bet. It becomes a trap the moment the bet is repeated at a size where a bad run can damage the pot the next round depends on.

5. Thinking the fix is to avoid positive-average bets. The fix is never to stop taking good bets — it is to size them so the multiplying through time cannot reach zero. Survival is the setting that makes a positive average finally safe to keep taking.