The Central Limit Theorem for IID Sums

Why a Normal Limit Appears for Sums

A sum of random observations can have a complicated distribution even when each observation has a simple one. The central limit theorem explains a broadly useful pattern: under appropriate conditions, the shape of a suitably centered and scaled sum approaches the standard normal shape as the number of observations grows. The original observations do not need to be normally distributed.

The theorem is a limit result, not a claim that every finite sum is exactly normal. It is useful because it identifies the distribution approached by the standardized sum and gives a principled basis for large-sample approximations. This tutorial is aligned with the SOA VEE Mathematical Statistics syllabus.

The conditions matter. The variables must be independent and identically distributed, and the common distribution must have finite mean and strictly positive finite variance. Identical distribution means each \(X_i\) has the same probability distribution; independence means learning one observation does not change the distribution of another. If the variance is zero, every observation is constant with probability one and the standardizing denominator is zero, so this stated form does not apply.

Centering and Scaling an IID Sum

Before interpreting the theorem, identify the center and scale of the sum. Linearity of expectation gives the mean of a sum whether or not its terms are independent. Independence is needed for the variance calculation: when independent variables have finite variances, their covariances are zero, so the variance of the sum is the sum of their variances.

The standard deviation of \(S_n\) is therefore \(\sigma\sqrt{n}\), not \(n\sigma\). Subtracting \(n\mu\) centers the sum at zero; dividing by \(\sigma\sqrt{n}\) gives it variance one. That calculation explains the standardization in the theorem. It does not by itself prove that the standardized sum is normal; the normal limiting shape is the additional conclusion of the central limit theorem.

Convergence in distribution has a precise interpretation. If \(Z_n=(S_n-n\mu)/(\sigma\sqrt{n})\), then for every real \(z\),

where \(\Phi(z)=P(Z\le z)\) for \(Z\sim N(0,1)\). In applications with a large but finite \(n\), this limiting relationship motivates approximating probabilities involving \(Z_n\) by standard normal probabilities. The quality of a finite-sample approximation depends on the population distribution and sample size; the theorem does not promise exactness at any particular \(n\).

What the Limiting Normal Distribution Says

The standardized sum has a limiting distribution with mean zero and variance one, regardless of the common distribution’s particular mean and variance, provided the theorem’s assumptions hold. Those population parameters still determine the centering and scaling: the sum is centered at \(n\mu\) and scaled by \(\sigma\sqrt{n}\).

Equivalently, for large \(n\), it is often useful to describe the sum as approximately normal with mean \(n\mu\) and variance \(n\sigma^2\). This is a distributional approximation, not an assertion that \(S_n\) has that exact distribution. A sum may also be restricted to a range of possible values—for example, a sum of nonnegative quantities cannot be negative—even though a normal approximation assigns positive probability to every real number. That is one reason to keep the word “approximately” in view.

Worked Examples

Assumptions and Common Exam Traps

The CLT is powerful, but it is not a universal rule that makes every sum approximately normal. Check the model before using it: identify the random variables being summed, verify that they are IID, and confirm that their common distribution has finite, positive variance. If observations are dependent, or if their variances are not finite, the theorem in the form stated here does not justify the conclusion.

Another trap is to confuse a limiting statement with an exact finite-sample result. A normally distributed parent population produces a normal sum for every finite \(n\). For a nonnormal parent population, the CLT says the standardized sum approaches a standard normal distribution as \(n\) increases; it does not say the sum is already normal for a chosen sample size. Nor does the theorem specify a universal sample size at which an approximation is sufficiently accurate.

Finally, distinguish the theorem’s target from the original variables. The limit is for the standardized sum, not for each \(X_i\). The population distribution may be skewed or bounded. The required centering and scaling remain tied to that population’s \(\mu\) and \(\sigma\), even though the limiting distribution is the same standard normal distribution.

Why the Theorem Matters

The IID central limit theorem is a bridge between population models and normal-based large-sample reasoning. It explains why sums from many different finite-variance populations can share a common limiting shape after the appropriate centering and scaling. In actuarial work, this perspective can help assess aggregate quantities built from many independent contributions, provided the model assumptions are defensible.

The theorem’s usefulness does not remove the need to understand the data-generating model. Dependence, extreme tails, small sample sizes, and discrete outcomes can affect how suitable a particular approximation is. Those are practical accuracy questions beyond the basic limit statement. For this theorem, retain the central distinction: the normalized sum converges in distribution to \(N(0,1)\), while a finite sum need not itself be exactly normal.