Ergodicity & the Time Average

Here is a bet a casino would sell its soul for. Flip a fair coin. Heads, your money grows by 50%. Tails, it shrinks by 40%. The expected value is gorgeous: half the time you multiply by 1.5, half the time by 0.6, and the average outcome each round is 1.05 — a clean, mouth-watering +5% per round. Play it forever and get rich, obviously.

Except you won’t. Play that “obviously winning” bet round after round and you will, almost certainly, go broke. Not unluckily-broke — systematically, mathematically, the-house-always-wins broke. The +5% is real, and it is a lie, both at once. Resolving that contradiction is the entire course, and the resolution has a name most people who quote “expected value” have never heard: ergodicity.

The two averages that everyone assumes are the same

There are two completely different ways to average a gamble, and almost every mistake in this course comes from confusing them.

The ensemble average lines up a huge crowd — a thousand people, each playing the bet once, all at the same instant — and averages across the crowd. That is the +5%. It is the number the casino’s brochure quotes, the number expected value computes, the number your intuition reaches for. And it is honest about the crowd.

The time average takes one person and follows them through time, round after round after round, and asks how their wealth actually grows along the way. That is the number that decides your fate, because you are not a crowd. You do not get to be the average of a thousand strangers. You get one sequence, multiplied together, and you have to survive every step of it to collect the next.

A process is ergodic when those two averages agree — when one person’s long journey through time looks like a snapshot of the whole crowd at an instant. For the tidy, additive risks your intuition was trained on, they do agree, which is why the confusion is so easy to miss. But most things that actually compound — wealth, a portfolio, a business, a career, your health — are non-ergodic: the time average and the ensemble average come apart, and when they do, expected value is computing the wrong number for the decision you are actually making. The crowd’s average is not your average. Bet as if it were and the gap between them is exactly where you disappear.

This is an expert-tier synthesis, and it stands on three models you have already built:

• From fat tails, the shape of the danger: rare, enormous outcomes dominate the mean, so a single unrepeatable outlier can prop up an ensemble average that no individual will ever touch.
• From compounding, the engine: growth is multiplicative, so order and path matter, one bad factor poisons the whole product, and a zero anywhere is forever.
• From asymmetry & optionality, the survival logic: a capped, irreversible downside is not just “a big loss” — it is an absorbing barrier that ends the game, and no upside pays out after it.

If any of those feels shaky, shore it up first. Everything here is built on them.

Feel the two averages split apart

Inside the very first lesson you will find the whole course as one instrument: an ergodicity engine that plays a multiplicative gamble across 400 parallel players at once and, in the same picture, follows the typical player living the sequence over time. Run the default coin and watch the two lines tear apart — the ensemble average climbing toward the sky while the typical player is quietly ground toward zero. Same bet, opposite fate. That divergence is non-ergodicity, and there is no way to un-see it once you have watched it happen.

Then you will find the one knob that fixes it: drag the bet fraction down — stake a sliver instead of everything — and watch the typical player’s path bend from a slide into ruin to a genuine climb. That is Kelly betting, and it falls out of the mathematics the moment you optimise the average you actually live instead of the average you only imagine.

Two things to carry with you. First, the ensemble average is not lying — it is answering a different question than the one you asked. It faithfully reports what happens to the crowd. It is you who quietly assumed the crowd’s average was yours. Second, nothing is rigged. The coin is fair, the edge is real, the arithmetic is honest. The ruin comes entirely from the difference between averaging across parallel worlds and living one path through time.

Why this earns a place in the latticework

Because “on average it works out” is one of the most confident, most respectable, most ruinous sentences in decision-making — and this is the model that tells you exactly when it is false. Expected value is a genuinely great tool, and it is computed by someone who quietly assumes they survive every round to collect the average. The moment a loss can be irreversible — the moment there is an absorbing barrier — that assumption breaks, and the glittering positive average becomes a fantasy that lives only in the handful of players who got lucky early and never compounded down. A whole class of disasters — blown-up traders, bankrupt “high-EV” gamblers, over-leveraged funds, the person who bet the house on a sure thing — is invisible to someone holding only the ensemble average, and obvious to someone who has internalised the time average.

The map of the course

Six teaching lessons build the model from its core out to its limits, then one exam locks it in:

1. Ergodic vs Non-Ergodic — the central distinction made plain: when the time average equals the ensemble average (and you can trust expected value) and when it doesn’t (and you can’t).
2. The Flagship Coin Flip — the +50%/−40% bet worked out in full: why the ensemble says +5%, the time average says −5%, and a vanishing fraction of lucky paths drags the mean up while almost everyone busts.
3. Additive vs Multiplicative — the fork that decides everything: why averaging is safe for a small, bounded, repeatable side-bet and fatal for your whole bankroll.
4. The Kelly Fix — optimise the average you live: expected log-return, the geometric mean, and how betting to maximise time-average growth automatically sizes your bets to survive.
5. Ruin & the Real World — absorbing barriers, gambler’s ruin, position sizing, insurance as buying back ergodicity, and why a strategy must survive before it is allowed to compound.
6. Where the Model Lies — the honest capstone: not everything is non-ergodic, ergodicity is about the dynamics and not a mood of doom, and why it is not the same thing as plain risk-aversion.

Then a Final Exam — graded, one question at a time, one-way: once you answer, it locks. No back button, no retries, 70% to pass — fittingly irreversible, like the barrier at zero that the whole course is about.

How to use this course

One habit does most of the work: before trusting any average, ask whether you are the crowd or the path. If the outcome merely adds up and no single step can wipe you out, the two averages agree and expected value is your friend. If it compounds and a step can be ruinous, stop — the ensemble average is quietly computing someone else’s future, not yours. Keep returning to the engine above; drag the bet fraction across the point where the typical path flips from ruin to growth until the difference between “the crowd’s average” and “your path” feels like a property of the world, not a technicality — because that single split is what the entire model is built on.

Next up: lesson 1, Ergodic vs Non-Ergodic — the one distinction that decides whether the average you were handed is a map of your future or a map of somebody else’s.