When the Coefficient of Variation Misleads

Why a Familiar Risk Ratio Can Mislead

The coefficient of variation (CV) compares the standard deviation of returns with their expected return. When expected returns are positive and comfortably away from zero, the ratio can help describe variability relative to the return level. But that interpretation has limits: as the expected return approaches zero, small changes in the denominator can produce large changes in the ratio. If expected return is negative, the usual interpretation as “risk per unit of expected return” becomes difficult to apply.

This does not make the coefficient of variation mathematically incorrect. It means a comparison can be numerically correct but still be unhelpful for judging investment outcomes. In such cases, report the expected return and standard deviation separately. Standard deviation measures absolute variability in the same units as return; it does not, by itself, say whether the expected return is attractive.

Definition: For a return random variable \(R\) with expected return \(\mu_R=E[R]\) and standard deviation \(\sigma_R\), the conventional coefficient of variation is \(CV_R=\sigma_R/\mu_R\), provided \(\mu_R\ne 0\). It is a dimensionless ratio. For a positive expected return, it describes standard deviation relative to expected return. When \(\mu_R=0\), it is undefined; when \(\mu_R<0\), its usual positive-return interpretation does not apply.

Some presentations use \(\sigma_R/|\mu_R|\), the standard deviation divided by the absolute value of the mean, so the reported ratio is nonnegative. This changes the sign but not the core limitation: with a negative mean, a larger expected loss can increase the denominator and make the ratio look smaller. Always state which convention is being used. Neither version should replace the expected return and standard deviation when the mean is near zero or negative.

What Happens Near Zero Expected Return?

The instability follows directly from the ratio. If standard deviation is positive and stays approximately fixed while the expected return gets smaller and remains positive, dividing by the smaller mean makes the CV larger. At a zero mean, division is impossible. On the negative side of zero, the conventional signed ratio is negative; moving farther from zero can make it closer to zero even when the standard deviation has not changed.

CV_R=\frac{\sigma_R}{\mu_R},\qquad \lim_{\mu_R\to 0^+}\frac{\sigma_R}{\mu_R}=+\infty \quad\text{for fixed }\sigma_R>0

This limit is a mathematical property, not evidence that the asset suddenly became proportionally more volatile. It tells us that a ratio using a near-zero denominator is sensitive and potentially hard to interpret. For instance, holding standard deviation at 2%, reducing the expected return from 0.10% to 0.05% doubles the CV from 20 to 40. The standard deviation itself has not changed.

Worked Example: Comparing Two Returns With Small Positive Means

Two hypothetical investments each have two equally likely annual returns. Investment A returns -1.9% or 2.1%; investment B returns -0.8% or 1.2%. Calculate each expected return, standard deviation, and conventional CV. Explain why the resulting ratios need care.

Method. Use the probability-weighted mean and population variance for each stated scenario distribution. With two equally likely outcomes, each outcome has probability 0.5.

Investment A. Its expected return is \(E[R_A]=0.5(-1.9\%)+0.5(2.1\%)=-0.95\%+1.05\%=0.10\%\). The deviations from the mean are \(-2.00\%\) and \(2.00\%\). Thus its variance is \(0.5(-2.00\%)^2+0.5(2.00\%)^2=4.00\) percentage-points-squared, and its standard deviation is \(2.00\%\). Its CV is \(2.00\%/0.10\%=20\).

Investment B. Its expected return is \(E[R_B]=0.5(-0.8\%)+0.5(1.2\%)=-0.4\%+0.6\%=0.20\%\). The deviations are \(-1.00\%\) and \(1.00\%\), giving variance \(0.5(-1.00\%)^2+0.5(1.00\%)^2=1.00\) percentage-points-squared and standard deviation \(1.00\%\). Its CV is \(1.00\%/0.20\%=5\).

Interpretation and check. Both CVs are correctly calculated, but A’s mean is only 0.10%. If A’s mean were 0.05% while its 2.00% standard deviation stayed fixed, its CV would be 40. The result would change substantially due to the small denominator. The standard deviations, 2.00% and 1.00%, give a direct comparison of absolute variability. The common mistake to avoid is treating a large CV caused by a near-zero mean as a stable, self-sufficient ranking of risk.

Why Negative Expected Return Changes the Interpretation

A negative mean is not just a positive mean with a minus sign. The conventional CV is then negative, so ordinary comparisons of positive ratios no longer carry over. If the absolute-mean convention is used instead, the result is nonnegative, but dividing by a larger expected loss can make the ratio smaller. A smaller ratio in that case does not mean the investment has a better expected outcome.

The issue can also be seen by shifting every possible return by the same constant \(c\). The mean changes by \(c\), while the deviations from the mean—and therefore the standard deviation—stay unchanged. The CV changes because its denominator changes. A measure whose value changes when all outcomes receive the same constant shift may not be a useful stand-alone comparison when the location of the return distribution is central to the decision.

E[R+c]=E[R]+c,\qquad (R+c)-E[R+c]=R-E[R]

The second equality justifies the claim about standard deviation: adding \(c\) to every outcome leaves each deviation from the mean unchanged, so the probability-weighted variance and its square root are unchanged. But the mean—and hence the CV—does change. This is one reason to show both expected return and standard deviation rather than relying on a ratio alone.

Worked Example: Equal Standard Deviations, Different Negative Means

Two hypothetical assets have equally likely annual returns. Asset C returns -5% or 1%; asset D returns -9% or -3%. Calculate their means, standard deviations, and CVs using both the conventional signed ratio and the absolute-mean version. What comparison is safe to make?

Method. First calculate the mean and standard deviation from the two-point distributions. Then calculate both ratios, keeping their conventions explicit.

Asset C. Its mean is \(0.5(-5\%)+0.5(1\%)=-2\%\). The deviations are \(-3\%\) and \(3\%\), so its variance is \(0.5(3\%)^2+0.5(3\%)^2=9\) percentage-points-squared and its standard deviation is \(3\%\). Its signed CV is \(3\%/(-2\%)=-1.5\); its absolute-mean version is \(3\%/2\%=1.5\).

Asset D. Its mean is \(0.5(-9\%)+0.5(-3\%)=-6\%\). Again the deviations are \(-3\%\) and \(3\%\), so its variance is \(9\) percentage-points-squared and its standard deviation is \(3\%\). Its signed CV is \(3\%/(-6\%)=-0.5\); its absolute-mean version is \(3\%/6\%=0.5\).

Interpretation and check. Both assets have the same 3% standard deviation, but D has a substantially more negative expected return. The signed CVs are negative, and the absolute-mean version is smaller for D. Neither fact makes D’s expected outcome better. The safe statements are that the standard deviations are equal and that D’s mean is lower. The common mistake is to read the smaller absolute CV as proof that D is preferable or less risky in every relevant sense.

Using Standard Deviation as an Alternative

Standard deviation is often the clearer alternative when the mean is near zero or negative. It reports the typical scale of deviations around the expected return, in the same units as the returns. For example, a standard deviation of 3% means the distribution’s standard deviation is three percentage points of return. It does not mean that every outcome will be within 3% of the mean, nor does it provide the probability of a loss without further information about the return distribution.

A useful comparison therefore keeps two questions separate: What is the expected return? and How variable are returns around that expectation? Standard deviation addresses the second question, while the mean addresses the first. If two assets have the same mean, comparing their standard deviations is a direct comparison of dispersion. If their means differ, report both statistics and explain the difference rather than compressing them into a potentially unstable ratio.

Worked Example: Reporting Mean and Standard Deviation Together

Two hypothetical investments have the following annual return scenarios. Investment E returns -4%, 2%, or 8% with probabilities 0.25, 0.50, and 0.25. Investment F returns 0%, 2%, or 4% with the same respective probabilities. Calculate the expected returns and standard deviations, and compare their variability.

Method. For each investment, compute \(E[R]=\sum_j p_jr_j\), then \(\operatorname{Var}(R)=\sum_j p_j(r_j-E[R])^2\), and take the square root for standard deviation.

Investment E. Its mean is \(0.25(-4\%)+0.50(2\%)+0.25(8\%)=-1\%+1\%+2\%=2\%\). The deviations are \(-6\%,0\%,6\%\). Its variance is \(0.25(36)+0.50(0)+0.25(36)=18\) percentage-points-squared. Its standard deviation is \(\sqrt{18}\%=4.2426\ldots\%\), or 4.24% rounded to two decimal places.

Investment F. Its mean is \(0.25(0\%)+0.50(2\%)+0.25(4\%)=2\%\). The deviations are \(-2\%,0\%,2\%\). Its variance is \(0.25(4)+0.50(0)+0.25(4)=2\) percentage-points-squared. Its standard deviation is \(\sqrt{2}\%=1.4142\ldots\%\), or 1.41% rounded to two decimal places.

Interpretation and check. The expected returns are both 2%, so the standard deviations directly show that E has greater variability: 4.24% versus 1.41%. The calculation is consistent with the scenarios: E’s outer outcomes are farther from its mean than F’s. Reporting the mean and standard deviation separately avoids making a near-zero or negative mean the denominator of a headline comparison.

Deriving the Near-Zero Warning

Suppose the standard deviation is a fixed positive number \(s\), and the expected return is a positive number \(m\). The CV is \(s/m\). If the mean is halved, the new ratio is \((s/(m/2))=2(s/m)\): it doubles, even though variability \(s\) did not change. More generally, as \(m\) approaches zero from above, \(s/m\) increases without bound. At \(m=0\), the ratio has no value because division by zero is undefined.

Approaching zero from below gives a signed ratio that decreases without bound rather than a positive measure of relative risk. If the absolute-mean convention is used, the ratio again grows without bound on either side of zero, but it still does not tell whether the expected return is positive or negative. The mean’s sign and size must remain visible in the analysis.

Key takeaway: Use the conventional CV cautiously when expected return is positive and not close to zero. At zero it is undefined; near zero it is unstable; with negative expected return its usual positive-return interpretation fails. In those cases, report expected return and standard deviation separately, with standard deviation interpreted as absolute dispersion rather than a complete measure of investment quality.

Common Mistakes and Exam Traps

Calling a near-zero CV comparison reliable just because the arithmetic is correct. Check the size and sign of the mean. A ratio may be mathematically valid yet highly sensitive or conceptually unhelpful.
Ranking negative-mean assets using signed CVs as if all CVs were positive. Negative ratios do not have the familiar positive-mean interpretation. State the means and standard deviations directly.
Treating the absolute-mean convention as a cure. It removes the negative sign but can make a larger expected loss produce a smaller ratio. It does not restore a complete or intuitive ranking.
Confusing percentage and decimal scales. The ratio must use equivalent units in numerator and denominator. For example, \(2\%/0.1\%=20\), and the same calculation in decimals is \(0.02/0.001=20\). A genuinely mixed-scale calculation such as \(0.02/0.1=0.2\) is wrong here because 0.1 as a decimal return means 10%, not 0.1%.
Assuming standard deviation describes every outcome or the probability of loss. It describes dispersion around the mean; it does not determine the whole distribution.

Check Your Understanding

For each problem, calculate the requested quantities and use the mean and standard deviation together when the CV may mislead.

A hypothetical asset has equally likely returns of -0.9% and 1.1%. Find its expected return, standard deviation, and conventional CV. (Answer: mean 0.10%; standard deviation 1.00%; CV 10.)
A hypothetical fund has expected return 0.04% and standard deviation 1.20%. Calculate its conventional CV, then explain the concern. (Answer: CV 30; it is highly sensitive to the near-zero mean and should not be used alone.)
A hypothetical investment has mean return -3% and standard deviation 6%. Give the conventional signed CV and the absolute-mean version, and state why neither alone describes whether the expected outcome is favorable. (Answer: signed CV -2; absolute-mean version 2; both omit the fact that expected return is negative.)
Asset G has mean -2% and standard deviation 4%; asset H has mean -8% and standard deviation 4%. Calculate the absolute-mean versions of their CVs and identify the safe comparison. (Answer: G 2; H 0.5; both have the same standard deviation, while H has the more negative expected return, so the smaller ratio does not show H is preferable.)
A return has standard deviation 3% and expected return 0.15%. Calculate its conventional CV. If the mean falls to 0.05% with standard deviation unchanged, calculate the new CV and describe the change. (Answer: first CV 20; new CV 60; the ratio triples because the mean is divided by three, although standard deviation is unchanged.)

What You'll Learn