Fat tails describe a simple but important empirical fact: large market moves occur much more often than a normal distribution would suggest.

Fat tails describe a simple but important empirical fact: large market moves occur much more often than a normal distribution would suggest. For quantitative traders, that matters because many standard risk intuitions are built around averages, variances, and normal-theory confidence bands. When return distributions are fat-tailed, those intuitions can fail precisely where risk management matters most: in the extremes.

The core issue is distribution shape. Under a normal distribution, extreme observations become vanishingly rare as you move farther into the tails. In financial data, however, realized returns often show heavier tails than the normal benchmark. Sonar’s fundamentals material explicitly highlights non-normality, noting that market returns commonly exhibit skewness and excess kurtosis rather than the symmetry and thin tails implied by Gaussian assumptions. In practical terms, excess kurtosis means more mass in the tails and a higher incidence of large deviations than a normal model would predict. That is the statistical basis for saying tail risk is understated when normality is assumed.

This breaks intuition trained on averages in two ways. First, mean and variance summarize the center of a distribution better than the extremes. Second, if a researcher implicitly maps volatility to risk using normal assumptions, they may underestimate how often outsized losses or gains can occur. A backtest can therefore look stable in ordinary periods while still being exposed to rare but consequential moves that are not well captured by Gaussian-style expectations.

Sonar’s research materials frame this as part of a broader caution about inference from historical results. The fundamentals page discusses how financial return series depart from idealized textbook assumptions, including through heavy tails and regime dependence. That matters because many familiar summary statistics remain easy to compute even when their interpretation becomes fragile. If the data-generating process is not close to normal, then z-score style reasoning and tail-probability estimates derived from the normal distribution can be materially misleading.

One implication is visual and diagnostic rather than purely theoretical.

Heavy tails amplify this concern because tail events can dominate realized outcomes while remaining poorly represented in finite samples. A model that looks attractive under simplistic assumptions may not be robust once non-normality and selection bias are taken seriously.

This connects directly to the deflated Sharpe ratio. Sonar’s glossary explains the deflated Sharpe ratio as an adjustment designed to evaluate whether an observed Sharpe ratio is genuinely significant after accounting for non-normal returns and multiple trials. That is highly relevant in fat-tailed settings. If returns are skewed or leptokurtic, a plain Sharpe ratio can overstate the strength of evidence because its usual interpretation is often tied to assumptions that the observed return stream does not satisfy. The deflated Sharpe ratio is therefore not a cure for tail risk, but it is a framework that explicitly acknowledges two realities common in quant research: returns are often not normal, and many ideas are tested before one is reported.

For risk modeling, the practical lesson is modest but important. Fat tails do not mean summary statistics are useless; they mean those statistics should not be treated as complete descriptions of risk. A normal model can still be a convenient approximation, but if it is used without checking distributional shape, tail exposure may be understated. Sonar’s materials support a more careful workflow: examine whether returns appear non-normal, treat backtest evidence cautiously, and use evaluation tools such as the deflated Sharpe ratio that are designed for settings where skewness, kurtosis, and data-mining effects matter.

So the claim is this: financial returns are often non-normal, with skewness and excess kurtosis, and that makes tail outcomes more important than normal-theory intuition would imply. Standard risk interpretations that rely on normality can therefore understate exposure to extreme events.

Covered in depth in the Strategy research fundamentals pillar hub.