Ergodicity in Trading: Why Average Returns Don't Tell You What Happens to You

A coin-flip game: if heads, you double your money; if tails, you lose 50%. Expected value per flip: +25%. Across many simultaneous players, the average ends up positive. Across many flips for a single player, the average ends up at zero. This contradiction is ergodicity, and it's the deepest reason traders blow up despite having "positive expected value" strategies. Understanding it changes how you size and how you think about edge.

The setup, ensemble vs time averages

Ensemble average. Average outcome across many simultaneous players (or instances). What happens "on average" across a population.

Time average. Average outcome for a single player over many sequential plays. What happens to one specific player over time.

For some processes, these two averages are equal, the process is ergodic. For others, they differ, the process is non-ergodic.

Most trading processes are non-ergodic. The ensemble average can be positive while the time average is negative. Knowing the difference is critical.

The classic example

Coin flip: heads doubles your money, tails loses 50%.

Ensemble average (across 1000 players, one flip each):

• 500 doubled to $200; 500 lost to $50
• Total: 500 × $200 + 500 × $50 = $125,000
• Average per player: $125
• Up 25% on average

Sounds great. But...

Time average (one player, 1000 flips):

• After heads: $100 → $200
• After tails: $200 → $100
• After heads-tails (or tails-heads): back to $100
• After H-T-H-T-H-T... pattern: stays around $100 - but actually decreases over time due to compounding (each H gain is smaller than the previous T loss in dollar terms, and each T loss cuts the larger base from the H gain).

Run the time average over many flips: you converge to zero over time, even though the ensemble average says +25% per play.

The two averages disagree because the process is non-ergodic. The math of compounding means multiplicative losses dominate multiplicative gains.

Why this matters in trading

Trading is non-ergodic. The "expected value" of a strategy (ensemble average) doesn't tell you what happens to any specific trader (time average) over many trades.

A strategy with +0.5R per trade in ensemble might have negative time average if the variance is high enough relative to the mean.

The fix is to size positions so that single losses don't hurt your time average too much. The 1% rule, Kelly fractions, position-sizing discipline, all of these are mechanisms to make the trader's time average converge to the strategy's ensemble expected value.

Without this discipline, a positive-EV strategy can still lose money for the trader executing it.

The asymmetry of compounding losses

The math at the core of non-ergodicity in trading:

Multiplicative gains and losses don't cancel.

A 50% loss followed by a 50% gain doesn't return you to break-even. It returns you to 75% of starting capital ($100 × 0.5 × 1.5 = $75).

A 90% loss followed by a 90% gain returns you to 19% of starting ($100 × 0.1 × 1.9 = $19).

The general principle: large multiplicative losses require disproportionate gains to recover. Catastrophic losses (90%+) effectively can't be recovered.

In trading, this means: avoiding catastrophic losses matters much more than maximizing expected gains. The trader who never has a catastrophic loss has a much better time average than the trader with higher expected value but occasional catastrophes.

The "risk of ruin" concept

Risk of ruin: probability that, given your strategy and sizing, you'll eventually hit zero.

A strategy with positive expected value can still have non-trivial risk of ruin if sizing is too aggressive. The math:

For a strategy with W = win rate, R = win/loss ratio, position size as fraction of capital f:

Risk of ruin grows rapidly as f approaches the Kelly fraction. At full Kelly, risk of ruin is ~50% over an infinite horizon. Over finite horizons, it's lower but still substantial.

At fractional Kelly (e.g., quarter), risk of ruin drops dramatically. At very small fractions (e.g., the 1% rule), risk of ruin is effectively zero for any reasonable strategy.

The trade-off: more aggressive sizing produces higher expected growth (ensemble) but higher risk of ruin (time-path issue). Fractional sizing produces lower expected growth but much lower risk of ruin.

For long-horizon trading, the fractional approach wins because it preserves the ability to keep trading, and over many years, the compounding of consistent small returns dominates the occasional large gains followed by occasional ruin.

A common mistake: optimizing for ensemble expected value

A trader sees a strategy with high expected value. They size aggressively to capture it. They blow up. The expected value was real (in the ensemble), but the trader's specific path didn't produce it.

The fix: size for time-average optimization, not ensemble-average optimization. The 1% rule, Kelly fractions, conservative sizing all serve this.

A common mistake: comparing your performance to "the average trader"

A trader sees "the average crypto trader earned X% last year." They feel they should have earned the same. They take more risk to "catch up." They blow up.

But "the average trader" is the ensemble average. Many traders earned much more, many earned much less (or were wiped out). The average is a mathematical construct, not what any individual trader experienced.

The fix: don't size based on aspirations to match ensemble averages. Size based on what's sustainable for your specific time path.

A common mistake: using ensemble math for a single decision

A trader faces a single bet with 60% probability of doubling money, 40% probability of losing half. Expected value: +20%. They take the bet with all their capital.

But for one play, you don't get the average, you get one outcome. With 40% probability, you lose half your capital. The "positive expected value" doesn't help if the bad outcome materializes.

The fix: for single high-stakes decisions, the expected value isn't the only consideration. Variance and tail outcomes matter. Don't bet large on individual high-EV propositions that have significant downside variance.

A common mistake: ignoring path-dependent risks

A strategy might have +0.3R per trade expectancy but have specific paths that produce ruin. A 50-trade losing streak (improbable but possible) might bankrupt the account before the strategy's positive EV can play out.

The fix: stress-test sizing for unfavorable paths, not just for expected paths. The sustainable sizing is one that survives a 99th- percentile bad streak, not the average streak.

Implications for thinking about trading

Several deeper implications:

1. "Maximum expected value" is wrong objective. The right objective is "maximum sustainable growth", accounting for the time path. This usually means smaller positions than EV maximization suggests.

2. Survival is the precondition. You can't earn returns from capital you've already lost. Sizing that preserves capital through worst cases is the foundation that returns can compound on.

3. Aggregate returns of "the market" don't apply to you individually. "Crypto outperformed in the last cycle" is true in aggregate. Whether YOU outperformed depends on YOUR specific path, your sizing, your timing, your survival.

4. Diversification helps. Multiple uncorrelated bets average out variance, making time-average converge to ensemble-average faster. Diversification isn't just about lower risk; it's about converting non-ergodic into more-ergodic dynamics.

5. Compounding requires consistency. Small consistent returns compound; volatile returns (with same average) compound much less. Smoothing your equity curve has real value beyond psychological comfort.

Mental model, ergodicity as the difference between the lottery winner average and the lottery player average

Take 1 million lottery players, each spending $10 on a ticket. One wins $10M. The ensemble average is positive, total winnings ($10M) exceed total spending ($10M) plus you have some additional small prizes.

But for an individual player, the time average is clearly negative, you spend $10 to expected win of less than $10. Across many tickets over time, you're certainly losing money.

The lottery is non-ergodic in a particularly extreme way. Trading is less extreme but the principle is the same: ensemble math and time- path math diverge.

You experience time-path math, not ensemble. Size for that.

Why this matters for trading

Ergodicity is the deepest reason for conservative sizing in trading. It's why Kelly-fraction discipline matters. It's why the 1% rule isn't just a heuristic, it's a defense against the non-ergodic nature of compounding returns. Hex37's position sizing infrastructure is calibrated for this, small per-trade risk fractions because that's what produces sustainable time-path returns.

Takeaway

Ergodicity is the property where ensemble average equals time average. Trading is non-ergodic, the expected value across many traders isn't the return any single trader gets. Multiplicative losses asymmetrically punish recovery. Optimal sizing isn't expected-value-maximizing, it's sustainable-growth-maximizing across your specific path. This is why the 1% rule, fractional Kelly, diversification, and conservative sizing matter: they convert your time-path returns toward what the ensemble math suggests. Survive the bad paths; the good ones take care of themselves.