Skewness and Kurtosis
The higher statistical moments that describe the real shape of financial return distributions — what they are, why they matter, and how they apply in options trading.
The Four Moments of a Distribution
Any probability distribution can be characterised by an infinite sequence of statistical "moments". The first four moments are the most important and the ones used universally: (1) the mean (first moment) — the mathematical expectation or "centre" of the distribution. (2) the variance (second moment) — dispersion around the mean; its square root is the standard deviation. (3) skewness (third moment) — the asymmetry of the distribution. (4) kurtosis (fourth moment) — the "fatness" of the tails relative to the centre. The normal distribution is characterised entirely by its first two moments (mean and variance), with skewness = 0 and kurtosis = 3. Any deviation from those values indicates non-normality. Financial returns typically have negative skewness and elevated kurtosis (>3), a clear sign that the normal is a poor approximation for tail risk. Understanding these two moments beyond mean and variance is critical for: (a) pricing options correctly, especially OTM puts; (b) sizing positions with tail events in mind; (c) evaluating strategies with very different distributions (buying versus selling options, trend following versus mean reversion); (d) detecting regime changes (shifts in skew and kurtosis precede shifts in volatility).
Kurtosis: The Fatness of the Tails
Kurtosis (from the Greek kurtos, "convex") measures the fatness or heaviness of a distribution’s tails relative to a normal. Mathematically it is the fourth standardised moment: kurt = E[(X − μ)⁴] / σ⁴. The normal distribution has a kurtosis of exactly 3, which is the reference value. Excess kurtosis, defined as kurtosis − 3, is often reported instead, putting the normal at 0. A leptokurtic distribution (kurt > 3, excess > 0) has fatter tails and a higher peak than the normal: extreme events are more frequent than expected, and "typical" outcomes are also more frequent — but "intermediate" outcomes are less so. Almost all financial returns are leptokurtic. Excess kurtosis of daily S&P 500 returns typically ranges between 7 and 20, against 0 for the normal. Cryptocurrency returns run even more extreme, above 30. A platykurtic distribution (kurt < 3, excess < 0) has thinner tails and a flatter peak: extreme events are rare and intermediate outcomes more common. Examples: uniform distributions, returns in some tightly regulated markets. A mesokurtic distribution (kurt ≈ 3) matches the normal. High kurtosis explains why lognormal and normal models systematically underestimate tail risk — events "impossible" under those models occur regularly, and Black Monday 1987 (+22σ under a normal) is the paradigmatic example.
Real Returns: Negative Skew Plus High Kurtosis
The characteristic combination in real financial returns is negative skewness plus high kurtosis. That combination is treacherous: (1) negative skew means large losses are more likely than equivalent large gains; (2) high kurtosis means extreme events (on both sides, but especially the downside given the skew) are more frequent than the normal model predicts. The practical result is that historical maximum drawdowns almost always far exceed what an analysis based on σ alone would predict. Concrete examples: the Dow Jones lost 89% between 1929 and 1932; under a normal with the historical σ, something like that should happen once every many millions of years. The S&P 500 fell 57% between October 2007 and March 2009. And bitcoin has recorded declines of 70% to 85% in each of its cycles — 2014, 2018, 2022 — magnitudes any constant-σ model grossly underestimates. The implication for traders: always add a margin of safety to any position sizing derived from models assuming normality. Nassim Taleb’s "Black Swan" popularised these ideas for a general audience; in the academic literature, Mandelbrot had already rigorously documented — in 1963 — that returns follow distributions with far heavier tails than the normal. Even so, the industry’s standard models (CAPM, traditional VaR) still assume normality, because the alternative introduces considerable complexity.
Black Monday as a 20σ Outlier
On 19 October 1987 — the famous Black Monday — the Dow Jones Industrial Average fell 22.6% in a single day. Under the assumption of normal returns with the historical volatility measured over the preceding months, that move represented roughly 20 to 22 standard deviations. The probability of a 22σ event under a normal distribution is literally 10⁻¹⁰⁷, a number so small it has no physical analogue — less likely than any specific event in the age of the universe (13.8 × 10⁹ years, with roughly 10⁸⁰ atoms in the whole of it). And yet it happened. Nor was it an isolated case: (1) October 2008 had several days with 7-10σ moves under a normal; (2) the flash crash of 6 May 2010 saw inexplicable instantaneous moves; (3) the March 2020 crash strung together four sessions with ±8% moves, each nominally 4 to 5σ; (4) the February 2018 "Volmageddon" destroyed funds that sold volatility with a move their models considered impossible. The fundamental lesson: do not trust models assuming normality for tail-risk decisions. In real markets, the probability of an extreme event is between a hundred and ten thousand times higher than the normal predicts. Modern alternatives — Student-t, Lévy stable, jump models — all acknowledge that observed reality of fat tails, though they treat it mathematically in different ways.
Implications for Options Pricing and Trading
Skewness and kurtosis have direct and important implications for options pricing. The key empirical observation: real markets quote options with a volatility smile or volatility skew. In equities, out-of-the-money puts trade at higher implied volatility than equivalent calls — negative skew — reflecting demand for crash protection and consistent with the empirical negative skewness. Puts further out, at three standard deviations or more, quote at higher implied volatility still, reflecting the elevated kurtosis: the market knows the normal underestimates tail risk and charges an additional premium for it. Sophisticated models that capture this: (1) SABR, widely used for fitting the smile; (2) Heston, a stochastic volatility model with negative correlation between price and volatility; (3) Merton jump-diffusion, which adds jumps to geometric Brownian motion; (4) variance gamma, flexible and able to capture both skew and kurtosis. For the retail trader not using sophisticated models, the conclusion is that out-of-the-money option prices already incorporate the tail risk premium: buying OTM puts as a hedge is expensive precisely because the market knows they work. Strategies that claim to "arbitrage" the failure of normality — selling far OTM puts on the assumption that the normal underestimates risk — have blown up repeatedly in real events. The professional advice: assume the market already has fat tails priced in, and if you are going to trade against the smile by selling tail insurance, do it at very conservative size.