On this page
Variance measures the average squared distance of returns from their mean, and standard deviation takes its square root so dispersion is expressed in return units. A normal distribution gives those measures a convenient interpretation, but the calculation of volatility alone does not establish that returns are normal. For CAIA® Level I, compute the measure correctly and treat any normal-curve conclusion as conditional on the shape of the return distribution.
Begin with the mean and the distance from it
A return series can have a clear average yet vary greatly around that average. Subtracting the mean from each observation shows the distance for each period. Adding those signed distances is not a useful dispersion measure because positive and negative differences cancel. Squaring each distance removes that cancellation and gives unusually distant observations more influence. Variance is the average of those squared distances, with the precise denominator depending on whether the observations are an entire population or a sample.
If the observations are the complete population being described, divide the sum of squared deviations from the population mean by the number of observations. If they are a sample used to estimate a wider return process, divide the sum of squared deviations from the sample mean by one fewer than the number of observations. The sample adjustment matters because the same data have been used to estimate the centre; using the sample size itself tends to understate the underlying variance. Identify the data description before choosing a denominator.
Variance has squared return units. When returns are percentages, the variance is in percentage-points squared, which is awkward to interpret against a return. Standard deviation is the square root of variance, restoring the original units. It indicates typical scale of variation around the mean, but it does not say whether positive and negative departures have the same frequency or severity. Two funds can have identical standard deviations and very different downside shapes.
The CAIA Association includes statistical foundations in its Level I curriculum overview and explains in its own published material that mean and variance can describe a normal distribution compactly. It also warns that alternative investment returns may call for attention to asymmetry and other shape features. That is the reason to learn calculation and interpretation as separate tasks rather than treating a volatility estimate as a complete risk description.
Work the sample calculation
Take four invented quarterly returns of negative two percent, zero percent, four percent and six percent. They are illustrative values, not observed fund performance. Their sum is eight percentage points, so the sample mean is two percent. The deviations from that mean are negative four, negative two, positive two and positive four percentage points. Their squares are sixteen, four, four and sixteen, giving a total of forty percentage-points squared.
Because the four observations are a sample, divide forty by three. Sample variance is thirteen and one-third percentage-points squared. Taking its square root gives a sample standard deviation of approximately 3.65 percentage points. If the task instead declared those four outcomes to be the entire population of interest, divide forty by four. Population variance would be ten percentage-points squared and population standard deviation approximately 3.16 percentage points. The observations have not changed; the statistical question has.
Keep units consistent throughout. If returns are entered as decimals, the same sample observations are negative 0.02, zero, 0.04 and 0.06. The sample variance is approximately 0.001333 in decimal-return units squared, and the standard deviation is approximately 0.0365, or 3.65 percentage points. A candidate who calculates with decimals and reports 0.0365 percent has confused a decimal proportion with a percentage. Writing the unit beside the mean, variance and final square root prevents this slip.
A useful reasonableness check is that the standard deviation should be expressed in the same unit as the input returns and should reflect the visible spread. The illustrative returns range from negative two to six, centred at two. A standard deviation of roughly 3.65 percentage points is plausible. A reported value of 13.33 percent would be a variance mistaken for a standard deviation; a value of 0.13 percent would be a unit conversion error. Neither error needs a formula sheet to detect.
Normality is a further assumption
A normal distribution is symmetric around its mean and has a specific bell shape. If a return model is genuinely normal, mean and standard deviation determine the whole distribution. That makes it possible to translate a threshold into a standardised distance from the mean and use a normal probability model. The translation is a statement about the assumed model, not a finding established by the standard deviation computation itself.
Consider an illustrative model with mean quarterly return of two percent and standard deviation of three percentage points. A return of negative four percent is two standard deviations below its mean. Under a normal model, one can use the normal curve to discuss how unusual a result at or below that threshold is. But if the actual return process has a heavy left tail or a lopsided payoff, the normal curve can assign too little probability to severe losses. The threshold calculation is still arithmetically correct; its probability interpretation may be misleading.
Alternative investments give concrete reasons for caution. Infrequently traded assets can be valued from estimates rather than fresh transactions. Reported returns may look smooth even though an eventual sale price can move sharply. Structured or option-like exposures can have an asymmetric payoff, with many modest outcomes and a smaller set of large losses or gains. Such features affect shape without necessarily making the standard deviation look extraordinary. The CAIA Association’s discussion of statistical methods explicitly connects alternative investment structures with non-normal returns and the need to examine more than mean and variance.
What to inspect beyond one volatility number
Ask first whether returns are balanced around the mean. A long left tail means severe negative outcomes can be more pronounced than positive outcomes of similar frequency. Ask next whether extreme observations appear more often than a normal model would suggest. Finally, consider how the values were produced: stale or appraisal-based valuations can alter the apparent pattern. These checks do not replace variance; they tell you what variance does and does not communicate.
A histogram or set of ordered returns can reveal an obvious asymmetry, although a small sample cannot settle the distribution with confidence. Skewness and kurtosis summarise shape, while a formal test addresses whether departures from normality are large relative to sample size. Those tools answer a different question from sample variance. The candidate should avoid declaring a series normal merely because its standard deviation is finite, small, or easy to calculate.
Time horizon also matters. A standard deviation measured from quarterly returns is a quarterly dispersion measure. Translating it to another horizon requires assumptions about how returns relate across periods. Correlation between successive returns or changes in volatility can break a simple square-root scaling shortcut. For an appraisal-smoothed series, a neat annualised number can disguise rather than resolve uncertainty about the underlying economic movements. State the horizon and the assumptions whenever a scaled figure is presented.
How to answer an exam item cleanly
Start by identifying whether the data are a sample or population and whether returns are percentages or decimals. Compute the mean, then the squared deviations, the appropriate variance and its square root. If the question introduces a normal model, state that it is an assumption before drawing a tail inference. If it describes an illiquid or nonlinear strategy, explain why a normal interpretation may miss relevant risk even when the arithmetic is sound.
The CAIA programme page places this subject within a wider exam path, and Omni’s learning approach can help organise practice. The core distinction remains simple: variance and standard deviation measure dispersion in the observed or modelled returns; normality is a claim about distribution shape. Keeping that distinction visible is more useful than memorising a volatility number without asking how it was generated.
Sources
- Explore the Curriculum | CAIA Association · caia.org · Retrieved
- Curriculum and Study Tools | CAIA Association · caia.org · Retrieved
- What Is an Alternative Investment? | CAIA Association · caia.org · Retrieved



