On this page
A population moment is a property of the distribution you are trying to understand; a sample moment is a statistic calculated from the observations you actually have. In an FRM® question, identify which population is meant, which observations were sampled and whether the requested result is a parameter or an estimate. Changing the sample can change the estimate without changing the underlying parameter.
Begin with the question the number answers
Suppose a risk analyst wants to know the long-run average return of a strategy under a stated model. That average is a population mean: it refers to the entire probability distribution of possible outcomes, including outcomes that have not appeared in the analyst's data. The analyst instead has a finite set of realised returns. Their arithmetic average is a sample mean. It is observable and useful, but it is an estimate of the population quantity, not a direct observation of it. The distinction matters even if the numbers happen to match in a particular exercise.
The word population needs care. It may describe every member of a finite collection, such as every trade in a specified file, or every possible outcome of a model, which need not be enumerable. If the exercise explicitly provides all members of a finite population and asks for its mean, dividing by the number of population members is appropriate. If it provides a subset drawn from a larger target, the arithmetic average of that subset is a sample statistic. Decide what the target is before recognising a familiar formula.
The GARP FRM study materials include quantitative analysis in Part I and organise the learning objectives by reading. This population-versus-sample idea underlies later work with volatility, covariance and estimation. The FRM programme overview also places quantitative analysis among the Part I domains. In other words, this is a basic interpretation skill that the candidate carries into risk calculations, rather than a notation exercise isolated from practice.
What is a moment?
A moment summarises a feature of a distribution. The first raw moment is the mean, which locates its centre. The second central moment is the average squared distance from that centre, which describes dispersion. Higher moments can describe asymmetry and tail shape, but a first encounter should focus on whose distribution the calculation describes. A population moment is an expectation under a defined distribution. A sample moment is computed from a finite collection of observed values. Calling both simply the average conceals the object each averages over.
For a discrete model, the population mean weights each possible outcome by its probability. A sample mean weights each observed data point equally in the ordinary arithmetic calculation. Equal weight across observations does not assert that the possible outcomes are equally likely; the realised sample may contain an outcome several times. The sample can overrepresent or underrepresent a part of the underlying distribution. This is why its average can differ from the model's expected value without either calculation being arithmetically wrong.
Notation helps if it is read rather than memorised. A Greek mu generally identifies the population mean, while an x-bar identifies a mean computed from observations. A Greek sigma squared refers to population variance, while an s squared commonly denotes a sample variance estimate. A population parameter is fixed for the population as defined. A sample statistic varies with the draw. The word fixed does not claim that a market is permanently stable; it means that under the particular population or model specified in the question, repeated sampling targets the same quantity.
Work an illustrative model and an observed sample
Take an invented return model with three possible one-period outcomes: a loss of 10%, a return of 0% and a gain of 20%. Suppose the model probabilities are 20%, 50% and 30%, respectively. Its population mean is the probability-weighted sum: negative 2 percentage points plus zero plus 6 percentage points, giving 4%. This result is a property of the stated model. It would remain 4% if an analyst collected a different sample, provided the model itself stayed the same.
Now suppose the analyst observes five independently drawn outcomes from that model: negative 10%, 0%, 0%, 20% and 20%. The sample mean is the sum of these five observed returns divided by five, or 6%. The sample gives two gains of 20%, whereas the model gives that outcome a probability of 30%. A realised frequency of two in five is possible under the model, so the 6% sample average is not evidence of a calculation error. It is one realisation of an estimator.
Imagine a second sample of the same size: negative 10%, 0%, 0%, 0% and 20%. Its mean is 2%. The parameter remains 4%; only the observed mix has changed. This comparison is more revealing than a single calculation because it shows exactly why a statistic has a sampling distribution. Repeating the sampling process would produce a range of sample means. We know the arithmetic of each realised sample, but without knowing the true model in practice we usually do not know the exact centre those estimates target.
This invented example is deliberately transparent. The percentages and probabilities are illustrative, not market observations or forecasts. In an exam item with supplied probabilities, calculate the population expectation from those probabilities. In an item with observed data only, calculate the statistic from the observations and label it as such. If both are supplied, use the difference to discuss sampling variation, rather than quietly substituting one for the other.
Read variance with the same discipline
Population variance measures the expected squared deviation from the population mean. The phrase population mean is essential: the deviations are defined relative to the distribution's own centre. A variance estimated from sample data instead relies on an estimated centre and a finite set of realised deviations. It is tempting to treat the two expressions as interchangeable because they both involve squared distances. They differ in both the object being described and, in common estimator formulas, the divisor.
For an illustrative complete finite population of four values, 0, 0, 4 and 4, the population mean is 2. Each value is two units away from the mean, so every squared deviation is 4 and the population variance is 4. If those four values instead represent a sample from a larger distribution, the sample mean is still 2, but the usual unbiased sample variance estimate divides the sum of squared deviations, 16, by three rather than four. The estimate is therefore sixteen thirds, about 5.33 squared units. The data did not change; the question being answered did.
This article's decision point is whether the stated observations constitute the target population or a sample from a broader target. The separate discussion of estimator bias examines why the sample variance divisor changes. Do not carry the divisor choice back into the definition of the population moment. Nor should you assume that a larger numerical variance estimate proves the actual population is more volatile; the estimate still carries sampling uncertainty.
Why a larger sample helps, and when it may not
Under a stable population and an appropriate sampling process, collecting more observations generally reduces the noise in a sample mean. The OECD's sampling manual demonstrates how means from different samples vary around a population mean and how sample size affects that variation. This does not mean that the sample mean moves steadily towards the population mean with every new observation. A new observation can move it away temporarily. Nor does a large sample repair systematic selection problems: drawing only calm trading days can give a precise estimate of the wrong population for a stress question.
In financial data, defining the population is especially consequential. Is the analyst estimating returns from the present strategy, a historical strategy before a mandate change, or a model of future market states? A long history can mix different regimes. A narrow history can miss rare but material outcomes. The mathematical separation between parameter and statistic does not solve model choice, but it makes the uncertainty visible. The GARP-hosted discussion of volatility estimation illustrates that an equally weighted historical sample can ignore the ordering of observations. That limitation should temper an interpretation of any sample volatility as a timeless truth.
A practical exam reading sequence
First underline words that define the target: all possible outcomes, the complete population, a historical sample or a random draw. Next identify what is supplied. Probabilities over outcomes support an expected value; a list of observations supports a sample statistic. Then write a label before the arithmetic: population parameter or sample estimate. Finally interpret the result in one sentence that acknowledges the scope of the data. This order prevents a correct computation from being attached to the wrong question.
When a question asks what changes across repeated samples, the answer is the statistic, assuming the same population is being sampled. When it asks what a model implies about its entire distribution, the answer is a parameter. If a different sampling design or a different population is introduced, say so; the comparison is no longer merely sampling variation. These distinctions are more useful than an isolated memory aid about Greek and Latin symbols.
The common mistake is to promote an observed average into the true expected return. A related mistake is to see two different sample means and infer two different population means. Both conclusions require more information than those samples provide. The FRM programme page provides the wider study context, and Omni's methodology page can help place concept work alongside practice. On an exam, keep the final statement modest and exact: a sample moment is evidence about a population moment, subject to how the sample was obtained.
Sources
- FRM Study Materials | GARP · garp.org · Retrieved
- FRM Exam | GARP · garp.org · Retrieved
- Volatility estimation discussion | GARP · garp.org · Retrieved
- PISA Data Analysis Manual: SAS | OECD · oecd.org · Retrieved



