Ergodicity

The deepest idea comes last, because it justifies almost everything else. Remember that a positive expected value is necessary but not sufficient. Here's why, and it has nothing to do with temperament or prudence. It's mathematics.

Why can a positive expected value still ruin you?

The multiplicative trap. Investment returns compound. This period's wealth is last period's wealth multiplied by (1 + return). Multiplicative processes behave very differently from additive ones. In particular, the average of the outcomes (the ensemble average) is not what a single investor experiences over time (the time average). Take a coin-flip game offered each period. Heads, your wealth grows 50% (×1.5). Tails, it falls 40% (×0.6). Work out the ensemble average by imagining a thousand people each playing one round. The expected multiplier is 0.5×1.5 + 0.5×0.6 = 1.05, a +5% expected gain. On that number the game looks wonderful, and a naive EV calculation says "play with everything."

Now compute the time average, which is what happens to one person playing round after round. Over many rounds you get roughly half heads and half tails. So your wealth multiplies by about (1.5 × 0.6) = 0.90 every pair of flips, which is (1.5 × 0.6)^0.5 = 0.9487 per round. That's a compound growth rate of −5.1% per round. The single player almost surely goes broke, even though the average across players rose 5% each round.

What does non-ergodic mean, and what is volatility drag?

How can both be true? A vanishingly small number of astronomically lucky paths, the all-heads ones, drag the ensemble average upward. They grow exponentially large and dominate the mean, while the typical path decays to zero. A process where the time average and the ensemble average split apart like this is called non-ergodic, and multiplicative wealth is the textbook case. That matters to you because you live one life, in sequence, so you experience the time average and never the ensemble average. The quantity to maximise is therefore the compound, or geometric, growth rate. The arithmetic expected return is the wrong target, and any strategy that risks ruin destroys the geometric rate however good its arithmetic EV looks. That's what Kelly maximises. The expected log of wealth is the time-average growth rate, so overbetting past Kelly can leave your arithmetic EV positive while your geometric growth goes negative. The coin game is just an over-bet in disguise. The ergodicity economist Ole Peters and, in his own idiom, Nassim Taleb (Skin in the Game, Fooled by Randomness) have pushed this idea back into finance, where it belongs.

Volatility drag. The gap between the arithmetic and geometric means is a tax you pay for volatility, and you can measure it. To a good approximation, geometric return ≈ arithmetic return − variance/2. In the coin game the arithmetic mean is +5% and the variance of the multiplier is 0.45² = 0.2025. So geometric ≈ 0.05 − 0.2025/2 = 0.05 − 0.101 = −5.1%, which matches the exact time average we computed. The lesson generalises. Volatility erodes compound returns, so two strategies with the same average return and different volatility deliver different wealth. The smoother one wins, sometimes by a lot. Cutting the depth of your drawdowns does more than comfort you. It directly raises your long-run compound rate.

Why does a 50% loss need a 100% gain to recover?

Why a 50% loss needs a 100% gain, and why ruin is forever. The same multiplicative asymmetry shows up in recovery math, and every investor should have this table burned in:

Losses and gains aren't symmetric under compounding. A loss of fraction L needs a later gain of L/(1 − L) to recover, and that explodes as L grows. A 50% loss demands a double just to get back to even. A 90% loss demands a ten-bagger. And the limit case is absolute: a 100% loss requires an infinite gain, and you never come back. Ruin is an absorbing state. It ends the game and shuts off all future compounding. An expected-value calculation that treats it as "just another outcome" on the distribution doesn't do it justice. So "avoid ruin" can't be traded off against return in the usual way. Compounding, the force everything here is built to harness, only exists for players who are still in the game. Buffett's "Rule No. 1: never lose money; Rule No. 2: never forget Rule No. 1" sounds like folksy caution. It's really a recognition that the geometric mean is unforgiving and zero is a trapdoor.

How does ergodicity tie the other tools together?

Now the twelve tools fit together. Expected value tells you which bets have an edge. Ergodicity adds that a positive-EV bet can still ruin you if it's sized to risk too much of your one, sequential, compounding life. Kelly is the sizing rule that maximises the time-average growth rate, so it refuses to bet enough to risk ruin. Fractional Kelly stays even further from the cliff, because your edge is uncertain. Base rates keep your probability estimates honest so the EV and Kelly numbers mean something. Bayes revises them as the world speaks. The circle of competence is machinery for finding the ways you could be ruined before it happens. So are inversion and the pre-mortem, and so is second-order thinking. And putting process over outcome is what lets you keep improving across the long run of decisions that a compounding life actually is. Survive first. Be right on average. Size so variance can't kill you. And grade the process, because it's the only part luck can't touch. That's the operating system.