Kurtosis
Kurtosis is a statistical measure that describes the shape of a probability distribution's tails relative to a normal distribution, indicating the likelihood of extreme outcomes. In finance, high kurtosis (leptokurtosis) signals fat tails and a greater probability of large gains or losses than a normal distribution would predict.
Key takeaways
- Excess kurtosis above 3 (leptokurtosis) indicates fatter tails than a normal distribution, meaning extreme events occur more frequently than standard models assume.
- Standard Value-at-Risk models that assume normality systematically underestimate tail risk in assets with high kurtosis.
- Hedge fund return distributions routinely exhibit excess kurtosis of 2–6, particularly in strategies that sell optionality such as volatility arbitrage.
- Historical simulation VaR partially captures kurtosis by using actual past returns, making it superior to parametric methods in non-normal environments.
- Kurtosis works alongside skewness to give a complete picture of a distribution's departure from normality.
Explanation
Kurtosis derives from the Greek word for 'curved' and measures the concentration of observations in the tails and peak of a distribution relative to a normal (Gaussian) distribution. Mathematically, kurtosis is defined as the fourth standardized moment of a distribution: E[(X − μ)⁴] / σ⁴. A normal distribution has a kurtosis of 3, so practitioners typically work with excess kurtosis (kurtosis minus 3), also called the kurtosis coefficient. A distribution with excess kurtosis greater than zero is called leptokurtic, has heavier tails and a sharper peak than a normal distribution, and implies that outlier events are more probable than Gaussian models predict. Negative excess kurtosis (platykurtic) indicates thinner tails.
In risk management, kurtosis is critical because most standard frameworks—including parametric Value-at-Risk, the Black-Scholes option pricing model, and classical mean-variance portfolio optimization—assume normally distributed returns. When asset returns are leptokurtic, these models underestimate the frequency and magnitude of extreme losses. The 2008 financial crisis, the 1987 stock market crash, and the 2020 COVID selloff all represented tail events that had far higher kurtosis than standard models anticipated.
Different asset classes and strategies exhibit characteristically different kurtosis profiles. Equity index returns typically display excess kurtosis of 3–5 over daily horizons; individual stocks can be much higher. Fixed income instruments tend toward lower kurtosis in normal regimes but spike during credit events. Hedge fund strategies that sell optionality—such as short volatility, convertible arbitrage, or merger arbitrage—often embed short-gamma positions that generate steady positive returns punctuated by catastrophic losses, producing extremely high kurtosis in their return streams.
Risk managers combat the kurtosis problem through several methods. Historical simulation VaR implicitly captures kurtosis by using the empirical distribution of actual returns rather than fitting a parametric distribution. Extreme Value Theory (EVT) focuses specifically on modeling tail behavior. Stress testing and scenario analysis examine performance under specific high-kurtosis events rather than relying on distributional assumptions. Some practitioners use the Cornish-Fisher expansion to adjust VaR estimates for both skewness and excess kurtosis, creating a modified VaR that accounts for non-normality.
Kurtosis interacts importantly with other portfolio construction decisions. Correlation estimates between assets can change dramatically in high-kurtosis environments, as extreme market moves tend to cause correlations to spike toward 1.0. This 'correlation breakdown' during crises is one reason diversification often fails precisely when investors need it most. Understanding kurtosis alongside skewness and correlation gives risk managers a more complete picture of the true risk profile of a portfolio.
Formula
Excess Kurtosis = E[(X − μ)⁴] / σ⁴ − 3
Example
Consider a hedge fund running a short-volatility strategy on the S&P 500. Over a three-year period, the fund records monthly returns with a mean of +1.2%, a standard deviation of 1.5%, and an excess kurtosis of 5.8. A parametric VaR model assuming normality estimates a 1% monthly VaR of approximately 2.3% (≈ 1.2% − 2.326 × 1.5%). However, adjusting for the excess kurtosis using the Cornish-Fisher expansion increases the 1% VaR estimate to roughly 4.1%—nearly 80% higher. In February 2018 (the 'Volmageddon' event), the fund suffers a single-month loss of 8.3%, a move that the Gaussian model implied had roughly a 1-in-10,000 probability but the kurtosis-adjusted model suggested was a 1-in-500 event—still rare, but plausible within a multi-decade investment horizon.
Related terms
Arbitrage Breakdown Convertible Arbitrage Correlation Diversification Equity Equity Index Fat Tails Financial Crisis Gamma Hedge Fund Historical Simulation Var