Time vs Ensemble Average: Two Kinds of Mean

Key Takeaways

• Time average vs ensemble average compares one path followed over time against many paths averaged at a single instant.
• For multiplicative, compounding returns the two means differ, and only the time average reflects one investor's experience.
• The common mistake is using the ensemble mean, the ordinary expected value, when the process is non-ergodic.
• The distinction underlies position sizing, since real wealth follows the time average, not the expected value.

What It Is

The ensemble average of a random process is the mean across many independent copies measured at the same time. If a thousand investors each make the same bet today, the ensemble average is the average of their thousand outcomes. This is what the ordinary expected value computes.

The time average is the mean of a single copy followed over a long stretch of time. One investor making the same bet repeatedly, period after period, experiences the time average of the process.

When these two averages are equal, the process is called ergodic, and expected value is a safe guide. When they differ, the process is non-ergodic, and you must decide which average is relevant. For an individual living through a sequence, that is the time average.

The Intuition

Picture a casino game. The casino cares about the ensemble average: across thousands of players tonight, what is the average result? The house edge guarantees that average tilts its way, and it pockets the difference.

A single player cares about something else: what happens to their money as they keep playing the same game over time? That is the time average. For additive bets with no compounding, the two line up. For multiplicative bets, where each result scales your remaining stake, they part ways.

The key realization is that you are always the single player, never the ensemble. You have one bankroll moving forward through time. So when the time average and ensemble average disagree, the ensemble average can be a dangerous illusion: technically correct about the crowd, wrong about you.

How It Works

For an additive process, where outcomes add to wealth, the time average and ensemble average of the growth coincide. For a multiplicative process, where outcomes scale wealth, they differ. The ensemble-average growth follows the arithmetic mean of returns:

g_ensemble = ln(1 + arithmetic mean of r)

The time-average growth follows the expected logarithm of the growth factor:

g_time = E[ ln(1 + r) ]

Where:

r       = single-period return, a random variable
E[ ]    = expected value across outcomes
ln      = natural logarithm

Because the logarithm is concave, Jensen's inequality guarantees g_time is less than or equal to g_ensemble. They are equal only when there is no variability. The larger the volatility, the wider the gap, which is why high-variance compounding strategies can look great on an ensemble basis yet shrink a single bankroll over time.

Worked Example

Consider an investment that each year either gains 80 percent or loses 50 percent, with equal probability.

The ensemble average return is the arithmetic mean: average of plus 80 and minus 50, which is plus 15 percent per year. A thousand investors, on average, do well, and the ensemble average says the bet is attractive.

The time average tells a different story for one investor compounding through both outcomes. A gain then a loss multiplies wealth by 1.8 then 0.5, equal to 0.9, a 10 percent loss over the two years. The expected log growth is the average of ln(1.8) and ln(0.5), which is the average of plus 0.588 and minus 0.693, equal to minus 0.053 per year. The single trajectory loses about 5 percent a year and erodes toward zero, even though the ensemble average is a healthy plus 15 percent.

Common Mistakes

1. Defaulting to expected value. Expected value is the ensemble average. For a non-ergodic, compounding process it overstates what one investor will actually earn over time.

2. Assuming the averages always agree. They coincide only for ergodic processes. Multiplicative returns with volatility are non-ergodic, so the two means diverge.

3. Reading the ensemble result as personal. A favorable average across many investors does not mean a favorable path for you. You only ever ride one trajectory.

4. Ignoring volatility's drag. The gap between the two averages grows with variance. A strategy can win on the ensemble metric purely because a few extreme paths drag the average up.

5. Forgetting compounding changes the math. Additive bets do not show the gap. The divergence appears specifically because returns multiply, scaling each bet to the current stake.

Frequently Asked Questions

What is time average vs ensemble average in simple terms? Time average vs ensemble average compares what happens to one investor over many periods against what happens across many investors in one period. For compounding bets, the single investor's long-run result can be far worse than the crowd average.

How does time average vs ensemble average affect investment decisions? It tells you to plan around your own path, the time average, not the expected value. Since real wealth compounds along one trajectory, you size positions so volatility does not erode your long-run growth rate.

What is a real-world example of the two averages diverging? A bet gaining 80 percent or losing 50 percent on equal odds has an ensemble average of plus 15 percent a year. Yet one investor compounding through both outcomes loses about 5 percent a year. The crowd looks fine; the individual shrinks.

How can investors use this distinction effectively? Optimize the expected logarithm of wealth, which is the time-average growth rate, rather than the arithmetic expected return. Keep volatility in check, because the gap between the two averages widens as variance rises.

How is the time average different from the ensemble average? The ensemble average means across many parallel copies at one instant, which the ordinary expected value computes. The time average follows one copy through time. They match only for ergodic processes and diverge for compounding, non-ergodic ones.

Sources

1. Peters, O. (2019). "The ergodicity problem in economics." Nature Physics, 15, 1216-1221. https://www.nature.com/articles/s41567-019-0732-0
2. Ergodicity Economics. "Ergodicity, jail, and time scales." https://ergodicityeconomics.com/2019/05/16/ergodicity-jail-and-time-scales/
3. Peters, O., and Adamou, A. "The Two Growth Rates of the Economy." arXiv:2009.10451. https://arxiv.org/pdf/2009.10451
4. ScienceDaily. "This 'fix' for economic theory changes everything from gambles to Ponzi schemes." https://www.sciencedaily.com/releases/2019/12/191202113024.htm

Disclaimer

This article is educational content only and is not financial advice. Nothing here is a recommendation to buy, sell, or hold any security. Consult a licensed advisor before making investment decisions.