Skip to main content

Ergodicity, Gently Introduced

Ergodicity investing is a concept that separates what should happen on average (across many investors or many scenarios) from what actually happens to any single investor over time. Most textbooks teach that if an asset has an expected return of 10%, you should expect to earn 10%. But that's an ensemble average—the average across all investors, all market paths, all hypothetical worlds. What you as an individual investor actually experience is a single path through time, and that single path has a different mathematical structure than the average of all possible paths.

This distinction sounds academic but is profoundly practical. It explains why investment strategies that look good "on average" often disappoint individual investors, why diversification reduces not just risk but also volatility of your actual returns, and why compounding returns are better described by geometric means than arithmetic means. Ergodicity is the hidden mathematics that reconciles theoretical finance with observed investor experience.

Quick Definition​

Ergodicity: A system property where the time average (what happens to an individual over time) equals the ensemble average (what happens on average across many individuals or scenarios).

Non-ergodic: A system where time averages and ensemble averages diverge. Most financial markets are non-ergodic.

Ensemble average: The statistical average of outcomes across many parallel scenarios (or many investors). In finance: the expected value computed from probability distributions.

Time average: The actual cumulative outcome an individual experiences along a single path through time. In finance: the compound return you actually earn.

Geometric mean: The average growth rate over time, properly accounting for compounding. Smaller than arithmetic mean when volatility is present.

Arithmetic mean: The simple average of returns, ignoring compounding effects. Always larger than geometric mean (by Jensen's inequality).

Key Takeaways​

  • Most investment textbooks focus on ensemble averages (expected returns). Real investors experience time averages (compound growth on a single path).
  • For non-ergodic systems, the ensemble average return can be positive while the time-average (individual) return is negative. This is mathematically possible and empirically common.
  • A portfolio with a +5% ensemble average annual return and 30% volatility has a geometric mean return of only ~4.5%, the true long-term growth rate for an individual investor.
  • The Kelly Criterion and optimal portfolio leverage depend on ergodic assumptions. Ignoring non-ergodicity leads to over-leverage.
  • Diversification increases the probability that time-average returns approach ensemble-average returns, which is why it's so powerful.

Ensemble Average vs. Time Average: A Concrete Example​

Imagine a simple asset: each year, it returns +50% or −25%, each with 50% probability.

The Ensemble Average View​

Expected annual return = 0.5 × (+50%) + 0.5 × (−25%) = +12.5%

Over 10 years, an ensemble average model predicts: $100 × 1.125^10 = $325.92

A textbook would say, "This asset returns 12.5% per year. Invest in it."

The Time Average View (What Happens to You)​

Now follow a single investor through 10 years, observing one actual sequence of returns. There are 2^10 = 1,024 possible sequences. Let's follow one:

Year 1: +50% → $100 × 1.50 = $150 Year 2: −25% → $150 × 0.75 = $112.50 Year 3: +50% → $112.50 × 1.50 = $168.75 Year 4: −25% → $168.75 × 0.75 = $126.56 Year 5: +50% → $126.56 × 1.50 = $189.84 Year 6: −25% → $189.84 × 0.75 = $142.38 Year 7: +50% → $142.38 × 1.50 = $213.57 Year 8: −25% → $213.57 × 0.75 = $160.18 Year 9: +50% → $160.18 × 1.50 = $240.27 Year 10: −25% → $240.27 × 0.75 = $180.20

Final value: $180.20 (total return: 80.2%, or 6.09% annualized)

This investor earned 6.09% annualized, not the ensemble-expected 12.5%. The time-average return fell short by over 6 percentage points per year.

What's the geometric mean? The compound growth rate?

Geometric mean = (0.5 × 1.50 × 0.5 × 0.75)^0.5 = (1.50 × 0.75)^0.5 = 1.125^0.5 = 1.0609 - 1 = 6.09%

Actually, let me recalculate. Each year, the investor faces a 50–50 coin flip. The geometric mean is:

Geometric mean = E[log return] = 0.5 × log(1.50) + 0.5 × log(0.75) = 0.5 × 0.405 + 0.5 × (-0.288) = 0.0585 = 5.85%

So the time-average (geometric mean) return is ~5.85%, while the ensemble-average return is 12.5%. The difference is 6.65 percentage points—more than half the ensemble average. A typical investor following this asset for 10 years would earn less than half the "expected" return.

Why the Divergence?​

The ensemble average treats each year as independent and parallel: "On average, the asset returns 12.5% per year." It computes the expected value of the linear return.

But compounding is multiplicative, not additive. When you earn +50% one year and −25% the next, you don't average to +12.5%:

  • 1.50 × 0.75 = 1.125 (net after two years, not 1.125^2)
  • Geometric mean: √1.125 ≈ 1.0609

The order of returns matters. Even though the sequence has the same number of +50% and −25% years, the compounding is multiplicative. Losses hit a smaller base (because of prior gains), and gains hit a smaller base (because of prior losses). The net effect is a drag compared to simple averages.

This is the essence of non-ergodicity in finance: the time path and the ensemble average are fundamentally different.

Arithmetic Mean vs. Geometric Mean​

Every investor has intuited this: a stock that returns +50% one year and −25% the next doesn't have a 12.5% average annual return, even though that's the arithmetic mean.

The geometric mean (the true time-average return) is always ≤ arithmetic mean, with equality only when there's no volatility.

Geometric mean = Arithmetic mean - (σ² / 2)

where σ is the standard deviation of returns.

Numerical Example​

Asset returns: +10% and −10% in alternating years (arithmetic mean = 0%)

Year 1: $100 → $110 Year 2: $110 × 0.90 = $99

Geometric mean: √(1.10 × 0.90) − 1 = √0.99 − 1 = −0.005 = −0.5%

The investor loses money despite the arithmetic mean return being 0%. The drag is −0.5%, arising from volatility.

Using the formula: σ for ±10% returns is approximately 10% (specifically, √0.01 ≈ 0.1). Drag ≈ −(0.10)^2 / 2 = −0.005 = −0.5%. ✓

For a portfolio with arithmetic mean return of +8% and volatility of 15%:

  • Geometric mean ≈ 8% − (0.15)^2 / 2 = 8% − 1.125% = 6.875%

Over 30 years:

  • Ensemble (arithmetic mean) prediction: $100 × 1.08^30 = $1,006.27
  • Time average (geometric mean) prediction: $100 × 1.06875^30 = $747.70

The difference is $258.57 (25.7% lower wealth) purely from the volatility drag on compounding.

Ergodicity and Portfolio Construction: The Kelly Criterion​

The Kelly Criterion tells you the optimal fraction of capital to wager on a favorable bet. It's a non-ergodic result masquerading as an ensemble-average formula.

For a simple bet with probability p of winning b dollars and probability (1−p) of losing $1, the optimal fraction of capital to wager is:

f* = ((p × b) - (1-p)) / b = (pb - 1 + p) / b

Example: A Favorable Coin Flip Bet​

You can bet on a coin flip where heads pays 2:1 (you win $2 for every $1 wagered) and tails loses $1. It's a fair-looking ensemble bet: E[return] = 0.5 × 2 − 0.5 × 1 = 0.5 per dollar, or 50% expected return.

The Kelly Criterion says: wager f = [0.5 × 2 − 0.5] / 2 = 0.25 = 25% of your capital.

Why not 100%? Because in a time-average sense (what actually happens to you), overbetting leads to ruin. If you bet 100% of capital on every flip:

  • Sequence HT: $100 × 3 × 0 = $0 (wiped out on second flip)

You can't recover from a loss of −100%. If you bet only 25%:

  • Sequence HT: $100 × 1.25 × 0.75 = $93.75 (you lose money but survive)
  • Sequence HTHT: $100 × 1.25^2 × 0.75^2 ≈ $87.89
  • Over many iterations, your geometric mean return is maximized at 25% betting

The Kelly Criterion emerges from the requirement that time-average growth (geometric mean) is maximized, not ensemble-average wealth.

Key insight: If you ignore ergodicity and over-leverage based on ensemble-average expected returns, you will eventually face ruin on your time path, even if the ensemble average is positive.

Diversification Through an Ergodic Lens​

Diversification is often justified on ensemble grounds: "It reduces risk." But the deeper truth is ergodic.

A single stock might have +25% annualized returns and 50% volatility (arithmetic mean). Its geometric mean is 25% − (0.50)^2 / 2 = 25% − 12.5% = 12.5%.

A diversified portfolio of 100 stocks, each uncorrelated with the same 25% arithmetic mean and 50% volatility, has:

  • Arithmetic mean: still 25% (unchanged)
  • Volatility: 50% / √100 = 5% (volatility of the portfolio mean returns)
  • Geometric mean: 25% − (0.05)^2 / 2 = 25% − 0.125% ≈ 24.875%

Diversification reduces the volatility drag on compounding. The time-average (geometric) return improves from 12.5% to 24.875%—a doubling.

This is why diversified portfolios tend to deliver closer to their ensemble-average returns than undiversified portfolios: diversification reduces volatility, which reduces the gap between arithmetic and geometric means.

Non-Ergodic Outcomes: Positive Ensemble Average, Negative Time Average​

Here's a shocking result: it's mathematically possible to have a strategy with a positive ensemble-average return but a negative time-average return that an individual investor will experience.

Example: A Terrible Leveraged Trade​

Suppose you have a strategy that's right 60% of the time, earning +10%, and wrong 40% of the time, losing −25%.

Ensemble average: 0.6 × 10% − 0.4 × 25% = 6% − 10% = −4%

Okay, that ensemble average is already negative. Let me adjust:

Right 70% of the time, earning +5%, and wrong 30%, losing −10%.

Ensemble average: 0.7 × 5% − 0.3 × 10% = 3.5% − 3% = +0.5%

Geometric mean: 0.7 × log(1.05) + 0.3 × log(0.90) = 0.7 × 0.04879 + 0.3 × (−0.10536) = 0.03415 − 0.03161 = +0.00254 ≈ +0.254%

Both are positive, so this isn't a counterexample. Let me try extreme leverage.

Suppose you have a bet that's right 55% of the time (gain +20%) and wrong 45% of the time (lose −15%).

Ensemble average: 0.55 × 20% − 0.45 × 15% = 11% − 6.75% = +4.25%

Geometric mean: 0.55 × log(1.20) + 0.45 × log(0.85) = 0.55 × 0.1823 + 0.45 × (−0.1625) = 0.10027 − 0.07313 = +0.02714 ≈ +2.71%

Still both positive. Let me consider path ruin.

Actually, a more relevant example: consider a strategy that with 51% probability returns +0.1% and with 49% probability loses −100% (wiped out).

Ensemble average: 0.51 × 0.1% − 0.49 × 100% = 0.051% − 49% = −48.949% (negative)

But suppose we modify: 51% return +100%, 49% return 0% (break-even).

Ensemble average: 0.51 × 100% − 0.49 × 0% = +51% (positive!)

Geometric mean: 0.51 × log(2) + 0.49 × log(1) = 0.51 × 0.6931 = 0.3534 ≈ +35.34%

Both positive.

The honest non-ergodic example is harder to construct without leverage. But here's the intuition: if a strategy can lead to ruin (−100%) with any probability, the time-average is pulled down toward negative, even if the ensemble average is positive. A single catastrophic loss wipes out all prior gains due to the multiplicative nature of compounding.

A Clearer Example: Bankruptcy Risk​

A risky venture has:

  • 90% chance of +5% annual return
  • 10% chance of bankruptcy (−100%, total loss)

Ensemble average: 0.9 × 5% − 0.1 × 100% = 4.5% − 10% = −5.5% (already negative ensemble)

But imagine a modified scenario:

  • 95% chance of +8% annual return
  • 5% chance of bankruptcy (−100%)

Ensemble average: 0.95 × 8% − 0.05 × 100% = 7.6% − 5% = +2.6% (positive!)

Geometric mean: 0.95 × log(1.08) + 0.05 × log(0) = undefined (or −∞)

The geometric mean is undefined because you have a non-zero probability of ruin. Over a long time horizon, you will almost certainly experience bankruptcy, making the time-average return −∞.

This is the ergodic trap: strategies with positive ensemble-average returns but non-zero bankruptcy risk will eventually result in ruin for an individual investor with a long enough time horizon.

The Kelly Criterion and Over-Leverage​

The Kelly Criterion is perhaps the most important non-ergodic insight in finance. It says: the optimal leverage depends not on ensemble-average returns, but on time-average (geometric mean) maximization.

Many investors and firms use the Sharpe ratio or expected return to size positions, ignoring the Kelly insight. They over-lever, betting the strategy's ensemble-average return is attractive enough to justify high risk.

The 1998 collapse of Long-Term Capital Management is the archetypal example: brilliant Nobel laureates maximized ensemble-average Sharpe ratios, massively over-leveraging a convergence-arbitrage strategy. When a non-correlated shock hit (Russian default), the ensemble average exploded. The time average was catastrophic: $4.7 billion became billions in losses, wiping out investors and requiring a Fed-orchestrated bailout. See Federal Reserve's LTCM Case Study and SEC Financial Crisis Resources for further reading on systemic risk and leverage.

The lesson: ensemble averages can be deeply misleading about what will happen to your actual portfolio. Non-ergodicity is not theoretical; it's operational reality.

Graphical Illustration​

Real-World Examples​

Example 1: Tech Stocks in the 1990s​

In the 1990s, tech stocks had high arithmetic mean returns (25%+) and high volatility (60%+). The ensemble average said: invest heavily. But the time average depended on the path.

An investor who bought in 1990, held through 2000, and sold had volatility drag. The geometric mean fell from +25% to perhaps +15%. Then came the 2000–2002 crash.

Investors who started in 1995, right before the bubble peak, and held for 5 years, experienced a time-average negative return, despite the ensemble average still being positive (in forward expectations, pre-crash).

Example 2: Leveraged Hedge Funds​

Many hedge funds report ensemble-average returns (arithmetic mean of strategies) of 12–15% with volatility of 8–10%, yielding attractive Sharpe ratios. But individual investors in these funds experience time-average returns (geometric means) of 8–10%, depleted by:

  • Volatility drag on compounding
  • Fees (1% + 20%)
  • Leverage drag (daily rebalancing losses)

A 2% annual fee and 20% performance fee alone reduce geometric returns by 2–3 percentage points. Add volatility drag and leverage drag, and a 15% ensemble-average fund becomes an 8% geometric-mean reality.

Example 3: Cryptocurrency (Bitcoin) Volatility​

Bitcoin has a high arithmetic mean return (~30–40% annualized over 2010–2020) and extreme volatility (~80–100%). Its geometric mean is roughly 15–20%, due to volatility drag. An investor who bought the peak (2017, near $20,000) experienced a time-average negative return over 2017–2019, despite Bitcoin's ensemble-average return being positive in long-term analysis.

Common Mistakes​

Mistake 1: Confusing Expected Return with Actual Return​

An investment fund's "expected return" (ensemble average) is not what you'll actually earn. Your realized return (time average) is lower by the volatility drag.

Mistake 2: Ignoring the Probability of Ruin​

Strategies that maximize ensemble-average Sharpe ratios can have non-zero probability of bankruptcy or catastrophic loss. The optimal time-average strategy (per Kelly) is more conservative.

Mistake 3: Over-Leveraging Based on Sharpe Ratios​

The Sharpe ratio is an ensemble-average concept. It doesn't account for the compounding damage of leverage. Optimal leverage by time-average considerations is much lower.

Mistake 4: Assuming Diversification Only Reduces Risk​

Diversification reduces volatility, which increases the geometric mean return for a given arithmetic mean. It's not just about risk reduction; it's about improving realized returns via non-ergodic mechanics.

Mistake 5: Treating Historical Volatility as Irrelevant​

High-volatility periods create large geometric-mean drags. An investor's actual return during a volatile year is much worse than the simple arithmetic return predicts.

FAQ​

What's the difference between ergodic and non-ergodic?​

Ergodic: time averages equal ensemble averages. What happens to you on average over time matches the statistical expected value. Non-ergodic: time averages differ from ensemble averages. Your actual experience diverges from statistical expectations. For investor protection and understanding of risk, see Investor.gov Risk Resources and FINRA Investor Education.

Is the stock market ergodic?​

No. The stock market is non-ergodic because of volatility drag, path dependency, and bankruptcy risk. Diversification pushes it toward ergodicity (reducing the variance of time-average returns), but it's never fully ergodic.

Why do most finance textbooks ignore ergodicity?​

Historical reasons: early finance theory (1950s–1970s) focused on ensemble concepts (CAPM, utility functions, expected returns). Ergodic thinking is more recent and less intuitive. Textbooks are slow to update.

Is the geometric mean the "correct" return measure?​

For long-term compounding and individual investor experience, yes. For short-term risk management and portfolio theory, arithmetic means are standard. Both are relevant; context matters.

Can I use Kelly Criterion for my personal portfolio?​

Yes, but carefully. The full Kelly Criterion is aggressive and can lead to large drawdowns. Many practitioners use "fractional Kelly" (e.g., 0.25 × optimal Kelly) for a more conservative approach.

Does mean reversion make the market more ergodic?​

Somewhat. Mean reversion reduces path dependency and makes time-average returns more stable. But non-ergodicity persists due to volatility drag and tail risk.

How does inflation interact with ergodicity?​

Inflation affects both ensemble and time averages. Real returns have the same ergodicity properties as nominal returns. High inflation periods increase volatility, which increases geometric-mean drag.

Summary​

Ergodicity investing distinguishes what should happen on average (ensemble average, expected return) from what actually happens to you (time average, geometric mean return). These are equal only when volatility is zero—which never happens in real financial markets.

The gap between arithmetic mean and geometric mean is a volatility drag that compounds over time. For a +8% arithmetic mean return with 15% volatility, the geometric mean is only 6.875%, a 1.125% annual gap. Over 30 years, this translates to 25% lower final wealth.

Diversification is powerful because it reduces volatility, narrowing the gap between what theory predicts and what you experience. Leverage is dangerous because it amplifies volatility, widening the gap and creating non-zero bankruptcy risk that can turn positive ensemble-average returns into time-average ruin.

The Kelly Criterion, ergodic theory, and geometric means provide the correct framework for long-term wealth accumulation. They explain why many investors underperform their portfolios' "expected returns," why diversification is so valuable, and why over-leverage based on Sharpe ratios leads to catastrophic losses.

For the individual investor compounding wealth over decades, non-ergodic mathematics is not an academic curiosity—it's the operating system of actual portfolio outcomes.

Next Steps​

Explore Time Average vs Ensemble Average for a deeper examination of path dependency, the empirical evidence of non-ergodicity in markets, and practical strategies for aligning your portfolio with time-average optimization.