Central Limit Theorem
The Central Limit Theorem (CLT) states that the sum (or average) of a large number of independent, identically distributed random variables with finite mean and variance converges in distribution to a normal distribution, regardless of the shape of the underlying population distribution.
Key takeaways
- CLT provides the theoretical justification for using normal distribution-based statistical tests (Z-tests, confidence intervals) on sample means, even when the underlying population is non-normal.
- In finance, CLT underlies the assumption that portfolio returns approximate normality when portfolios are well-diversified across many independent assets.
- CLT requires independence, finite variance, and sufficient sample size (typically n ≥ 30 for moderately non-normal distributions; much larger n for heavy-tailed distributions).
- Financial return distributions have fat tails (leptokurtosis) and negative skewness, violating CLT's assumptions and leading to systematic underestimation of tail risks by normal-distribution models.
- Monte Carlo simulation uses CLT when aggregating many simulated scenarios to estimate expected values and confidence intervals for complex financial instruments.
Explanation
The Central Limit Theorem is arguably the most important theorem in statistics, enabling inference about population parameters from sample statistics. Formally: if X_1, X_2, ..., X_n are i.i.d. random variables with mean μ and variance σ², then as n → ∞, the distribution of (X̄ - μ) / (σ/√n) converges to the standard normal distribution N(0,1). In plain terms: the sampling distribution of the sample mean becomes approximately normal for large samples, regardless of the original distribution shape.
In financial applications, CLT provides the foundation for many standard models. The Black-Scholes model assumes that the log-return of an asset over a period T is the sum of many small independent log-returns over sub-periods: ln(S_T/S_0) = Σ(r_i), and by CLT, this sum is approximately normally distributed for large T and many sub-periods. This justifies modeling log-returns as normally distributed (equivalently, asset prices as log-normally distributed), which is the foundation for Black-Scholes option pricing and most VaR models.
Portfolio diversification leverages CLT directly. The portfolio variance of n equally-weighted, uncorrelated assets with individual variance σ² is σ²/n — declining to zero as n → ∞. More generally, when assets have correlation ρ, portfolio variance is: σ²_p = (σ²/n) + [(n-1)/n] × ρσ², which converges to ρσ² as n → ∞. This proves that diversification eliminates idiosyncratic risk but not systematic (correlated) risk. CLT underpins this convergence argument.
However, the CLT's applicability to financial return data is compromised by several empirical features. First, financial returns exhibit fat tails (higher kurtosis than the normal distribution): the probability of extreme moves is much higher than the normal model predicts. The 2008 financial crisis produced '25 standard deviation' daily moves according to normal-distribution models — statistically impossible under CLT assumptions. Second, returns are not i.i.d.: they exhibit autocorrelation (momentum and mean-reversion), time-varying volatility (GARCH effects), and correlation clustering during crises. These violations mean risk models relying purely on CLT-based normality assumptions (standard VaR, Basel II models) systematically underestimate tail risk.
The Generalized CLT extends to heavy-tailed distributions (stable distributions, Lévy processes) where variance may be infinite, in which case the limit distribution is a stable distribution rather than a Gaussian. This mathematical framework is more appropriate for modeling financial tail risks, though it sacrifices the analytical tractability of the normal distribution.
Formula
CLT: √n × (X̄ − μ) / σ → N(0,1) as n → ∞; Portfolio Variance with correlation: σ²_p = (σ²/n) + [(n−1)/n] × ρσ²
Example
A quantitative risk manager calculates one-day 99% VaR for a portfolio of 200 stock positions assuming normally distributed returns. Using CLT, the portfolio's daily return distribution is approximated as N(0, σ²_p), where σ²_p = (individual stock variances + cross-correlations). The 1-day 99% VaR is estimated at -2.33 × σ_p per dollar invested. However, backtesting reveals that actual daily losses exceed the VaR estimate on 18 out of 1,000 trading days (1.8%) rather than the expected 10 (1.0%) — consistent with fat-tailed empirical return distributions. The manager supplements the normal VaR with expected shortfall (CVaR) and historical simulation to better capture tail risk, reducing reliance on CLT normality assumptions.
Related terms
Annuity Autocorrelation Backtesting Black Scholes Model Convergence Copula Correlation Diversification Expected Shortfall Fat Tails Financial Crisis Idiosyncratic Risk