Every distribution in the catalog described a single random variable drawn from a population. Now take a genuinely new idea: draw a random sample of size n from that population, compute some statistic from the sample (most often the sample mean, x̄), and ask, if you repeated this whole sampling-and-computing exercise over and over, what distribution would the resulting statistics themselves follow? That distribution is the statistic's sampling distribution, and it is the single idea that everything in point and interval estimation rests on. For the sample mean specifically, if the population has mean μ and variance σ², the sampling distribution of x̄ (over repeated samples of size n) has:
E[x̄] = μ Var(x̄) = σ² / n SD(x̄) = σ / √n ("the standard error")Notice what this says: the sample mean is, on average, right (it targets the true population mean), and its own spread shrinks as n grows, at the rate of 1/√n, quadruple your sample size and you halve your standard error (Practice Problem 14 drills exactly this scaling). If the underlying population is itself normal, x̄'s sampling distribution is exactly normal for any n at all. If the population is not normal, and most real financial data (fund expense ratios, deal sizes, daily returns) is not, you need the next idea to justify treating x̄ as approximately normal anyway.
Advertisement
The Central Limit Theorem: the single most useful fact in this module
Here is the fact that quietly licenses almost every confidence interval and hypothesis test used in finance, stated precisely: for a population with any distribution shape at all (provided it has a finite mean μ and a finite variance σ²) the sampling distribution of the mean of a sufficiently large random sample is approximately Normal(μ, σ²/n), regardless of the population's own shape. "Sufficiently large" is a rule of thumb, not a bright line, n ≳ 30 is the commonly cited threshold, smaller is fine if the population is already roughly symmetric, and larger is safer if the population is heavily skewed or has extreme outliers.
Sit with why this is remarkable rather than just useful. You almost never know, and often cannot easily check, the true shape of a real financial population, fund expense ratios are typically right-skewed, deal sizes span orders of magnitude, corporate bond spreads cluster near zero with an occasional distress-driven spike. The Central Limit Theorem says: it does not matter. Once you are reasoning about the sample mean of a large-enough sample rather than about individual observations, the normal-family machinery, z-scores, t-distributions, the confidence intervals built from both, applies safely, even though not one of those individual observations was ever close to normally distributed. Worked Example 5 shows this concretely: a genuinely skewed population of fund expense ratios still produces a nearly normal sampling distribution for its sample mean once n reaches 40.