Coefficient of Variation and Relative Dispersion

Stock A trades around $10 with a daily standard deviation of $0.50. Stock B trades around $500 with a daily standard deviation of $10. Which one is "more volatile"? By raw standard deviation, B looks twenty times riskier. But that comparison is meaningless — B is simply a higher-priced stock, and a $10 swing on a $500 base is proportionally smaller than a $0.50 swing on a $10 base. The coefficient of variation exists exactly for this situation: it strips out the scale of the underlying quantity so that dispersion can be compared fairly across things measured on different scales.

An analogy: comparing a mouse's stride to an elephant's

An elephant's stride length varies by 30 cm from step to step; a mouse's stride varies by 1 cm. In raw terms the elephant looks "more variable." But the elephant's average stride is 2 meters and the mouse's is 5 cm — relative to their own size, the mouse's stride is actually far more erratic (1/5 = 20% of its stride length) than the elephant's (30/200 = 15%). You can't compare raw variability across things of different scale without first dividing by the scale itself. The coefficient of variation is that division, applied to any measurement with a meaningful, non-zero average.

The mechanics, one piece at a time

The coefficient of variation (CV) of a variable with standard deviation $\sigma$ and mean $\mu$ is

$$CV = \frac{\sigma}{\mu},$$

often expressed as a percentage. In plain English: it's the standard deviation measured in units of the mean — "how big is the typical wobble, relative to the typical level." A CV of 0.05 means the standard deviation is 5% of the mean; a CV of 1.5 means the standard deviation is 150% of the mean, i.e. dispersion so large relative to the average that the average itself is a poor summary of typical outcomes. Because CV is a pure ratio, it's scale-free — multiplying every observation by a constant (say, converting currency or changing units) leaves CV unchanged, unlike raw standard deviation.

Worked example 1: stock A vs stock B, correctly compared

Stock A: mean price $10, standard deviation $0.50, so $CV_A = 0.50/10 = 0.05$, or 5%. Stock B: mean price $500, standard deviation $10, so $CV_B = 10/500 = 0.02$, or 2%. Despite B's raw dollar swings being twenty times larger, B is actually less volatile relative to its own price level — its CV is less than half of A's. Anyone sizing risk by raw dollar standard deviation alone, without normalizing by price, would have drawn exactly the wrong conclusion about which stock carries more relative risk.

Worked example 2: comparing two strategies' P&L consistency

Strategy X has an average monthly P&L of $50,000 with a standard deviation of $15,000: $CV_X = 15{,}000/50{,}000 = 0.30$. Strategy Y has an average monthly P&L of $8,000 with a standard deviation of $6,000: $CV_Y = 6{,}000/8{,}000 = 0.75$. Even though X's dollar standard deviation ($15,000) is more than double Y's ($6,000), X is the more consistent strategy relative to its own average earnings — Y's P&L is far noisier relative to what it typically makes, meaning an investor allocating based on smoothness of returns, not absolute size, should prefer X despite its larger absolute swings.

Adjust the mean and spread below and watch how the same standard deviation looks totally different in relative terms depending on where the distribution is centered — that shift is exactly what CV quantifies.

What this means in practice

CV is used to compare volatility across assets priced on different scales, across strategies with different capital bases, and across time periods where the mean itself has drifted. It's a close cousin of the Sharpe ratio's logic (excess return per unit of risk) but applied to the raw level of a series rather than to returns net of a risk-free rate. Risk teams use CV to flag when a low-priced or small-notional instrument is being under-weighted in a raw-dollar risk report despite being proportionally the riskiest thing in the book.