Skewness and Kurtosis in Trading Returns — Why Average Performance Lies

Skewness and kurtosis in trading returns describe the shape of the return distribution beyond what its average and standard deviation can show. Two models can share the same mean return and the same volatility, yet behave nothing alike — one steadily winning small and occasionally losing big, the other doing the opposite. The averages will not tell you which is which. The higher moments will.

This matters because almost every popular performance metric is built on the assumption that returns are roughly symmetric and well-behaved. The Sharpe Ratio divides excess return by standard deviation. The standard deviation treats a loss and a gain of equal size the same. That works fine if the distribution is broadly normal. Real trading returns are not.

What skewness measures

Skewness measures the asymmetry of the return distribution. A skewness of zero means losses and gains are mirrored around the mean. A positive skew means the right tail is longer — large outsized winners, smaller more frequent losers. A negative skew means the opposite — small frequent wins and the occasional brutal loss.

This is one of the more practically useful descriptors of a strategy's character. Trend-following models usually run with positive skew: most trades break even or scratch, a handful of runners pay for everything. Premium-selling and mean-reversion models often run with negative skew: small steady gains punctuated by sharp drawdowns when the regime turns.

A strategy's skew tells you what kind of equity curve you are about to live with. It also tells you what to expect when the model misfires. A positively skewed system has its bad days roughly distributed throughout the year. A negatively skewed one can give back six months of profit in a single week. Same mean return, completely different experience.

What kurtosis measures

Kurtosis describes how fat the tails of the distribution are. Higher kurtosis means more extreme outcomes — both large gains and large losses — than a normal distribution would predict. Lower kurtosis means returns cluster tightly around the mean and the extremes are rare.

Real market returns almost always show excess kurtosis. The big move that backtests treat as a once-in-a-decade event tends to show up several times a decade. This is the root of most blown-up strategies. A model sized using standard deviation alone is implicitly betting that the distribution behaves normally — and when the fat tail arrives, the realised loss is several times what the volatility number suggested.

The practical reading is straightforward: high kurtosis means the Standard Deviation is understating risk. Two strategies with the same standard deviation but very different kurtosis are not equally risky. The fatter-tailed one will eventually surprise its owner.

Why averages and Sharpe can mislead

The Sharpe Ratio is the most-quoted single measure in quant finance, and it treats the entire return distribution as if it were symmetric and thin-tailed. Two strategies with identical Sharpe ratios can be nothing alike once you look at skew and kurtosis. One can be a smooth, slightly negatively skewed grinder. The other can be a positively skewed trend rider with long flat stretches. Their performance reports will look almost identical until the day they don't.

This is why darwintIQ does not rely on Sharpe alone. The platform pairs it with the Sortino Ratio, which separates downside volatility from upside volatility, and with distribution-shape metrics like Wasserstein Distance and the KS Statistic, which measure how far a model's live return distribution has drifted from its evaluation distribution. A model whose tails are getting fatter in production is a model whose Sharpe has not noticed yet.

Final thoughts

Skewness and kurtosis in trading returns are not exotic statistics. They are basic descriptors of how a strategy actually behaves, and ignoring them is how respectable-looking systems end up with humiliating tail losses. The mean tells you what a strategy averages. The standard deviation tells you how much it varies around the average. Skew and kurtosis tell you which surprises it has been hiding.

For anyone evaluating a trading model — whether on the darwintIQ dashboard or in a spreadsheet at home — the right instinct is to assume the headline numbers are flattering. The higher moments are where you find out what kind of strategy you actually have.