Quant Analytics: from scratch to pro · Statistics for markets · Beginner
Averages that mislead: arithmetic and geometric means
The average of a list of returns and the growth rate you actually earned are different numbers. The gap between them is called volatility drag, and it grows with the square of the volatility.
In this lesson
- Compute an arithmetic and a geometric mean of a list of returns
- Explain why the geometric mean is the growth rate an investor actually earned
- Estimate volatility drag with g = mu - sigma^2 / 2 and know where the approximation is rough
- Say which average answers which question
Read first: Returns: simple, log and compounded, Volatility explained
Ask what a fund returned "on average" and you can get two honest answers that disagree. One is the average of the yearly figures. The other is the yearly rate that turns the starting capital into the ending capital. Only the second tells you what an investor earned, and the difference between them depends on how bumpy the ride was.
Two averages
The arithmetic mean adds the returns and divides by how many there are: mu = (R1 + R2 + ... + Rn) / n. It is the right average for a single typical period, for example the expected return of next year if each year is a fresh draw.
The geometric mean is the constant rate that gives the same final wealth: g = ((1 + R1) x (1 + R2) x ... x (1 + Rn))^(1/n) - 1. It is the right average for what happened to money over several periods, because returns multiply (see simple and log returns). The geometric mean is also the average of the log returns, turned back into a simple return.
The geometric mean is never larger than the arithmetic mean. They are equal only when every return is the same.
Worked example
A fund gains 50% in year one and loses 40% in year two.
- Arithmetic mean:
(50% - 40%) / 2= 5.0%. - Total return:
1.5 x 0.6 - 1= a loss of 10%. - Geometric mean:
sqrt(1.5 x 0.6) - 1= a loss of 5.13% a year.
An investor who is told "5.0% average" and ends 10% down has not been lied to. The arithmetic mean answers a different question.
Worked example: five years
Annual returns: +30%, -20%, +15%, -10%, +25%.
- Arithmetic mean: 8.0%.
- Growth factor:
1.30 x 0.80 x 1.15 x 0.90 x 1.25= 1.3455, a gain of 34.55% in total. - Geometric mean:
1.3455^(1/5) - 1= 6.11% a year.
Money grew at 6.11% a year, not 8.0%. The 1.89-point gap is the cost of the swings.
Volatility drag
The gap has a handy approximation: g = mu - sigma^2 / 2, where sigma is the standard deviation of the returns. The term sigma^2 / 2 is called volatility drag.
For the five years above, sigma is 21.97% (sample standard deviation), and the approximation gives 5.59%. The exact figure is 6.11%, so the approximation is rough with only five noisy observations and large swings. It works better when returns are small and there are many of them. It comes from a second-order expansion of the logarithm, so it drops the higher-order terms.
The size of the drag depends on the square of volatility. With a mean of 8%:
| Volatility | Drag | Approximate growth |
|---|---|---|
| 20% | 2.0 points | 6.0% |
| 30% | 4.5 points | 3.5% |
Raising volatility by half more than doubles the drag. This is the same arithmetic as recovery from a drawdown: after a fall of 50%, a gain of 100% is needed to get back.
Which average for which job
- Looking back at what an investment delivered over several years: geometric mean.
- Looking forward over one period, or averaging across assets in one period: arithmetic mean.
- Comparing two histories with different volatility: the geometric mean, because it already includes the cost of the swings.
Projecting long-run wealth with the arithmetic mean overstates it; the drag has to come off first.
Common mistakes
- Quoting the arithmetic average of yearly returns as the growth rate.
- Forgetting that the arithmetic mean of +50% and -40% is positive while the money is lost.
- Using
mu - sigma^2 / 2on a handful of very large moves and treating it as exact. - Comparing two assets by their arithmetic means when one is far more volatile.
- Averaging returns with different lengths (a quarter and a year) without converting to the same period first.
In practice
When you read a return figure, ask which average it is and over what period. Report the geometric mean for growth and the standard deviation next to it so the reader can see the drag. Analytics on FinUties applies measures like these to the markets it covers, so they are at hand when you read a chart.
Approximate growth rate against volatility, for a mean return of 8%
Worked numbers
Each result is recomputed from the formula shown.
| What | Formula | Result |
|---|---|---|
| arithmetic mean of +50% and -40% | (0.5 - 0.4) / 2 | 5.0% |
| two-year total return of +50% then -40% | 1.5 * 0.6 - 1 | 10% |
| geometric mean of +50% and -40% | sqrt(1.5 * 0.6) - 1 | 5.13% |
| year 1 return | 0.30 | |
| year 2 return | -0.20 | |
| year 3 return | 0.15 | |
| year 4 return | -0.10 | |
| year 5 return | 0.25 | |
| arithmetic mean of five annual returns | (r1 + r2 + r3 + r4 + r5) / 5 | 8.0% |
| five-year growth factor | (1 + r1) * (1 + r2) * (1 + r3) * (1 + r4) * (1 + r5) | |
| five-year total return | growth5 - 1 | 34.55% |
| geometric mean of the five returns | growth5 ** (1/5) - 1 | 6.11% |
| sample standard deviation of the five returns | sqrt(((r1 - am5)**2 + (r2 - am5)**2 + (r3 - am5)**2 + (r4 - am5)**2 + (r5 - am5)**2) / 4) | 21.97% |
| approximation mu - sigma^2 / 2 | am5 - sd5**2 / 2 | 5.59% |
| drag-adjusted growth, mean 8%, volatility 20% | 0.08 - 0.20**2 / 2 | 6.0% |
| drag-adjusted growth, mean 8%, volatility 30% | 0.08 - 0.30**2 / 2 | 3.5% |
| gain needed after a 50% fall | 1 / 0.5 - 1 | 100% |
Check yourself
A fund gains 50% in one year and loses 40% in the next. What is the arithmetic average return, and what did an investor actually earn per year?
The arithmetic average is (50% - 40%) / 2 = 5.0%. But 1.5 x 0.6 = 0.90, a total loss of 10%, which is a geometric return of sqrt(0.90) - 1 = -5.13% a year.
An asset has an arithmetic mean return of 8% and a volatility of 20%. What is the approximate geometric return?
g = mu - sigma^2 / 2 = 0.08 - 0.04 / 2 = 6.0%. The drag is 2 percentage points.
The same mean of 8% with a volatility of 30%. What changes?
The drag becomes 0.09 / 2 = 4.5 points, so g = 3.5%. Drag grows with the square of volatility: 1.5 times the volatility gives 2.25 times the drag.
A portfolio falls 50%. What gain gets it back to the start?
1 / 0.5 - 1 = +100%. Losses need larger gains to repair, which is the same effect that drives volatility drag.
Key terms
Sources
- The Econometrics of Financial Markets. Princeton University Press (John Y. Campbell, Andrew W. Lo, A. Craig MacKinlay), 1997. Accessed 9 Nov 2026