Coefficient of Variation for Return Comparisons

Why Normalize Return Dispersion?

Standard deviation measures how far returns tend to vary around their expected value, in the same units as the returns themselves. The coefficient of variation expresses that dispersion per unit of expected return. It is useful when the question is not only how variable returns are, but how large that variability is relative to a positive expected return.

The ratio requires care: its usual interpretation depends on expected return being positive. A zero expected return makes the ratio undefined, while a negative expected return does not support the standard interpretation of dispersion relative to an expected gain. The coefficient is not itself a probability or a return forecast.

Definition: For a return $R$ with finite standard deviation $\sigma_R$ and positive expected return $\mu_R=E[R]>0$, the coefficient of variation is $$ \operatorname{CV}(R)=\frac{\sigma_R}{\mu_R}. $$ It is a dimensionless measure of standard deviation per unit of expected return. The positive-denominator condition is required for the usual relative-dispersion interpretation.

Calculating the Coefficient from Scenarios

For a return distribution with possible returns $r_j$ and probabilities $p_j$, the probabilities must be nonnegative and sum to 1. The expected return is the probability-weighted mean, and the variance is the probability-weighted squared deviation from that mean:

\mu_R=\sum_j p_jr_j,\qquad \sigma_R^2=\sum_j p_j(r_j-\mu_R)^2,\qquad \operatorname{CV}(R)=\frac{\sqrt{\sum_j p_j(r_j-\mu_R)^2}}{\sum_jp_jr_j}

These expressions assume that the return outcomes refer to the same investment and time horizon, the probabilities describe the return distribution being analyzed, the variance is finite, and $\mu_R>0$ for the usual coefficient-of-variation interpretation. Use either decimals throughout or percentage points throughout the calculation; do not divide quantities expressed on inconsistent scales.

Worked Example: A Three-Scenario Return Distribution

A hypothetical infrastructure fund has a one-year return distribution with a 25% probability of a 10% loss, a 50% probability of a 10% gain, and a 25% probability of a 30% gain. Calculate its expected return, standard deviation, and coefficient of variation. Assume the stated probabilities and returns describe the complete one-year distribution.

Method. First verify the probabilities: $0.25+0.50+0.25=1.00$. Calculate the expected return using decimal returns:

\mu_R=0.25(-0.10)+0.50(0.10)+0.25(0.30) =-0.025+0.050+0.075=0.100

The expected return is $0.100=10.00\%$, which is positive. The deviations from the mean are $-0.10-0.10=-0.20$, $0.10-0.10=0$, and $0.30-0.10=0.20$. Thus the variance is

\sigma_R^2 =0.25(-0.20)^2+0.50(0)^2+0.25(0.20)^2 =0.25(0.04)+0+0.25(0.04) =0.020

The standard deviation is $\sigma_R=\sqrt{0.020}=0.141421\ldots$, or approximately $14.14\%$. The coefficient of variation is

\operatorname{CV}(R)=\frac{0.141421\ldots}{0.100} =1.41421\ldots\approx 1.4142

Check. The scenario probabilities sum to 1, the expected return is positive, and the standard deviation is larger than the expected return. The ratio is therefore greater than 1, as calculated. On this distribution, standard deviation is about 1.4142 times expected return. The method avoids treating the 30% outcome as the expected return: all scenarios must be weighted by their probabilities.

Why the Ratio Is Unit-Free

The coefficient is unchanged if the same return is expressed in another consistent unit. For example, $0.10$ and $10\%$ are different numerical representations of the same return. Both the standard deviation and the mean change by the same scale factor, so their ratio does not change.

A related justification shows why the ratio describes dispersion relative to the size of the expected return. Suppose each return is multiplied by a positive constant $c$, as when both the dollar gain and its expected value scale with a positive investment amount. By the expectation and variance rules,

E[cR]=c\mu_R,\qquad \operatorname{Var}(cR)=c^2\operatorname{Var}(R),\qquad \sigma_{cR}=c\sigma_R

The last equality uses $c>0$, so $\sqrt{c^2\sigma_R^2}=c\sigma_R$. Therefore,

\operatorname{CV}(cR)=\frac{c\sigma_R}{c\mu_R} =\frac{\sigma_R}{\mu_R} =\operatorname{CV}(R)

This derivation explains two properties: consistent changes of units do not affect the coefficient, and scaling both the expected return and its dispersion by the same positive factor leaves their relative size unchanged. The result does not imply that every other feature of the investment or distribution is unchanged.

Worked Example: Comparing Scenario Dispersion with a Positive Mean

A hypothetical renewable-energy project has one-year returns of $-5\%$, $15\%$, and $25\%$, with probabilities $0.20$, $0.50$, and $0.30$, respectively. Calculate the expected return, standard deviation, and coefficient of variation. Assume these are the complete outcomes for the same one-year horizon.

Method. Confirm $0.20+0.50+0.30=1.00$. In percentage points, the expected return is

\mu_R=0.20(-5)+0.50(15)+0.30(25) =-1+7.5+7.5=14

Thus the mean is $14\%$, a positive return. The deviations from the mean are $-5-14=-19$, $15-14=1$, and $25-14=11$ percentage points. The variance in squared percentage points is

\sigma_R^2 =0.20(-19)^2+0.50(1)^2+0.30(11)^2 =0.20(361)+0.50(1)+0.30(121) =72.2+0.5+36.3=109

Therefore, the standard deviation is $\sqrt{109}=10.440306\ldots$ percentage points. The coefficient is

\operatorname{CV}(R)=\frac{\sqrt{109}}{14} =0.745736\ldots\approx 0.7457

Check. Expressed as decimals, the standard deviation is approximately $0.104403$ and the mean is $0.14$; their quotient is again approximately $0.7457$. The calculation avoids mixing percentage points in the numerator with a decimal mean in the denominator.

Estimating from Historical Returns

When a return distribution is not specified and a set of historical returns is used instead, the arithmetic sample mean estimates expected return and the sample standard deviation estimates return dispersion. For observations $r_1,\ldots,r_n$, use the sample standard deviation with denominator $n-1$, then divide it by the sample mean if that mean is positive. This produces an estimate of the coefficient; it is not a guarantee about future returns.

Worked Example: Coefficient from Six Historical Returns

A hypothetical asset had six annual returns, in chronological order, of $8\%$, $-4\%$, $12\%$, $4\%$, $10\%$, and $0\%$. Estimate the coefficient of variation from these observations. Treat the returns as observations for the same annual horizon, and use the sample standard deviation.

Method. The arithmetic sample mean, in percentage points, is

\bar r=\frac{8+(-4)+12+4+10+0}{6} =\frac{30}{6}=5\%

The deviations from $5\%$, in percentage points, are $3,-9,7,-1,5,-5$. Their squared values sum to

$$ 3^2+(-9)^2+7^2+(-1)^2+5^2+(-5)^2 =9+81+49+1+25+25=190 $$

There are $n-1=5$ degrees of freedom, so the sample variance is $190/5=38$ squared percentage points. The sample standard deviation is $\sqrt{38}=6.164414\ldots$ percentage points. Since the sample mean is positive, the estimated coefficient is

\widehat{\operatorname{CV}}=\frac{\sqrt{38}}{5} =1.232882\ldots\approx 1.2329

Check. The six returns sum to $30$ percentage points, confirming the $5\%$ mean. The sample standard deviation uses the specified $n-1$ denominator, and dividing $6.164414\ldots$ by $5$ gives the stated estimate. The method avoids using a single realized return, or a geometric average, in place of the arithmetic sample mean for this calculation.

Conditions and Common Traps

The denominator is central to the interpretation. If expected return is zero, the quotient $\sigma_R/\mu_R$ is undefined. If expected return is negative, the quotient can be computed algebraically, but the usual coefficient-of-variation interpretation as dispersion per unit of a positive expected return does not apply. Do not describe such a result as a standard measure of relative risk around an expected gain.

A frequent arithmetic error is to mix units. For instance, suppose a return is equally likely to be $5\%$ or $15\%$. The mean is $10\%$, and the standard deviation is $5$ percentage points. The incorrect mixed-unit calculation is

\frac{5}{0.10}=50

The numerator is in percentage points, while the denominator is in decimal form. Use consistent units instead:

\frac{5}{10}=0.5 \qquad\text{or}\qquad \frac{0.05}{0.10}=0.5

Both correct forms give the same dimensionless coefficient. Before interpreting a result, check that the returns use a consistent horizon, the mean is positive, and the numerator and denominator are represented in the same units. A positive coefficient is a ratio, not a percentage unless it is explicitly converted to one; for example, $0.5$ may be stated as $50\%$ of the mean.

Key takeaway: For a positive expected return, divide standard deviation by expected return to obtain dispersion per unit of expected return. Keep the units consistent, and do not use the usual interpretation when the expected return is zero or negative.

Check Your Understanding

For each problem, keep the returns on a consistent horizon and use the stated distribution or sample convention.

A hypothetical equipment lessor has annual returns of $5\%$, $15\%$, and $25\%$, with probabilities $0.40$, $0.40$, and $0.20$. Find the expected return, standard deviation, and coefficient of variation. (Answer: mean $13\%$, standard deviation $\sqrt{56}=7.4833$ percentage points, coefficient $0.5756$.)
A hypothetical storage venture has possible annual returns of $0\%$, $8\%$, and $16\%$, with probabilities $0.25$, $0.50$, and $0.25$. Calculate its expected return, standard deviation, and coefficient of variation. (Answer: mean $8\%$, standard deviation $\sqrt{32}=5.6569$ percentage points, coefficient $0.7071$.)
A hypothetical fund has historical annual returns of $6\%$, $10\%$, and $14\%$. Estimate the mean, sample standard deviation using $n-1$, and coefficient of variation. (Answer: mean $10\%$, sample standard deviation $4$ percentage points, estimated coefficient $0.4000$.)
A return is equally likely to be $-10\%$ or $10\%$. Calculate its expected return and standard deviation. Is the usual coefficient-of-variation interpretation available? (Answer: mean $0\%$, standard deviation $10\%$, quotient undefined, so the usual interpretation is unavailable.)
A hypothetical investment is equally likely to return $-20\%$ or $4\%$. Find its expected return and standard deviation, and explain whether the usual coefficient-of-variation interpretation applies. (Answer: mean $-8\%$, standard deviation $12$ percentage points; the quotient is $-1.5$, but the usual interpretation does not apply because the mean is negative.)

What You'll Learn