Kurtosis

What is Kurtosis?

Kurtosis is the fourth statistical moment of a return distribution, sitting one level beyond the mean (first moment), variance (second), and skewness (third). Where volatility tells you how wide the typical spread of returns is and skewness tells you whether that spread leans left or right, kurtosis tells you how much of the action lives in the extremes. It answers a specific question: compared with a normal bell curve of the same width, does this strategy produce more frequent and more severe outlier months, or fewer?

The reference point is the normal distribution, which has a kurtosis of exactly 3. A distribution with kurtosis above 3 is called leptokurtic, the technical word for fat-tailed: its center is more peaked and its tails are heavier, so calm months and violent months both happen more often than normal, at the expense of the moderate middle. A distribution with kurtosis below 3 is platykurtic, with thinner tails and more of its outcomes clustered in a middling range. Because the normal benchmark is 3, many tools report excess kurtosis, defined as kurtosis minus 3, so that zero means normal, positive means fatter tails, and negative means thinner tails.

Financial returns are almost always fat-tailed. Daily and monthly equity returns routinely show kurtosis well above 3, which is the statistical fingerprint of crashes and melt-ups: events that a normal model would deem nearly impossible show up several times a decade. Reading kurtosis on a backtest tells you how prone the strategy is to those outliers. Unlike skewness, kurtosis does not care which direction the surprises lean; it counts the weight in both tails together. A high reading is a flag that the strategy's average and volatility understate how wild its worst and best months can get.

Formula

The formula standardizes each return by subtracting the mean and dividing by the standard deviation, raises the result to the fourth power, and averages. The fourth power is what makes kurtosis a tail statistic. Because it is an even power, sign is discarded, so both large gains and large losses contribute positively; and because the power is so high, a deviation of three standard deviations contributes eighty-one times as much as a one-standard-deviation move. The number is therefore dominated by the handful of most extreme observations, which is exactly the point: kurtosis is built to be sensitive to outliers.

The version shown is the population (biased) estimator, which divides by N and is the natural companion to the population skewness used elsewhere in these results. There is also a sample-adjusted estimator that applies a small-sample correction, the form many spreadsheet functions report by default; the two converge as the number of observations grows. SledgeKey computes kurtosis on the strategy's monthly returns over the test window using the standardized fourth-moment definition above, and the figure is read against the normal benchmark of 3: a value near 3 means tails close to normal, while a value clearly above 3 flags a fat-tailed strategy with an elevated chance of extreme months. If you prefer the excess-kurtosis convention, simply subtract 3 from the reported number, after which zero is the normal reference.

Why Kurtosis Matters in Backtesting

Kurtosis informs how much faith to place in the rest of the risk numbers. Volatility, Value at Risk, and the Sharpe ratio all behave most reliably when returns are roughly normal. The moment a distribution is fat-tailed, those measures quietly understate the real danger, because they assume extreme months are rarer than they actually are. A backtest can show a tame volatility and a comfortable VaR while sitting on a kurtosis of 8 or 10, which means the strategy will occasionally deliver a month several times worse than its volatility alone would suggest. Reading kurtosis is how you catch that mismatch before it catches you.

The failure mode of ignoring kurtosis is calibrating position sizing and risk limits to a normal world that does not exist. If you size a strategy assuming a 5% monthly loss is a once-a-decade event, but the true distribution is fat-tailed enough to deliver that loss every few years, your risk budget is wrong in the direction that hurts. This is the lesson behind a long line of blowups: returns that looked manageable under a bell-curve assumption produced "twenty-five standard deviation" moves, in the words of one bank's CFO during the 2007 quant crisis, only because the bell-curve assumption was never appropriate. Fat tails are not a rare anomaly in markets; they are the normal condition, and kurtosis is the number that quantifies them.

Kurtosis is most useful read alongside the metrics that locate and size the tail it counts. Skewness tells you which direction the extremes lean; Value at Risk and Conditional VaR tell you how deep the worst months go; maximum drawdown tells you what the worst peak-to-trough stretch actually was. Kurtosis adds the missing dimension of how often the extremes arrive. A strategy with high kurtosis and negative skewness is the combination to respect most: frequent extremes that lean toward the loss side.

How SledgeKey Implements Kurtosis

Kurtosis appears on the backtest results page as a single dimensionless number within the risk metrics, computed from the same monthly returns the other statistics use. There is no unit and no time horizon attached, because kurtosis is a shape statistic rather than a return or a rate: it simply reports how heavy the tails of the monthly returns are relative to a normal bell curve. The number is read against the normal reference of 3, so a reading near 3 indicates tails close to normal, while a reading well above 3 indicates a fat-tailed strategy that produces extreme months more often than a normal model would expect.

The benchmark's kurtosis is computed the same way over the identical window, so the comparison is the useful reading. A strategy with markedly higher kurtosis than the index is more outlier-prone than the benchmark, even if its volatility looks similar, and its other risk numbers should be treated as floors rather than worst cases. Because kurtosis depends on deviations raised to the fourth power, it is dominated by the single most extreme month and is therefore noisy over short windows, where one outlier can inflate the figure dramatically. Read it as a directional signal about tail heaviness rather than a precise figure, and lean on longer windows that span more than one market regime before drawing firm conclusions.

Common Pitfalls

The first pitfall is forgetting which convention you are reading. Raw kurtosis uses 3 as the normal benchmark, while excess kurtosis uses 0, and the same distribution will show up as either 3 or 0 depending on the choice. If you compare a kurtosis figure from one tool against an excess-kurtosis figure from another, you will misjudge the tails by exactly 3 units. Always confirm whether a reported value already has the 3 subtracted before comparing across sources.

The second pitfall is trusting a kurtosis estimate from a short sample. Because the calculation raises deviations to the fourth power, it is even more sensitive to a single extreme observation than skewness is. A backtest of only two or three years can show a kurtosis of 9 that is really just one crash month, and the figure can collapse toward normal if the window shifts to exclude that month. Kurtosis from a short window is a hint, not a measurement; the longer the record and the more regimes it spans, the more the number can be trusted.

The third pitfall is treating kurtosis as a measure of direction or of overall risk. High kurtosis says the extremes are frequent, not which side they fall on and not whether the strategy is good. A symmetric, fat-tailed strategy can have high kurtosis with zero skewness, and a strategy with thin tails can still lose money steadily. Kurtosis answers "how often do extremes happen," while skewness answers "which side are they on" and volatility answers "how wide is the typical spread." Reading any one of them in isolation tells you only part of the distribution's shape.

Related Concepts