Heteroskedasticity means that variance is not constant.

Heteroskedasticity means that variance is not constant. In financial time series, that usually means returns alternate between quieter periods and more volatile periods rather than fluctuating around a single stable level of dispersion. This time varying variance is common in markets and matters because many basic statistical procedures assume equal variance across observations or through time.

For strategy research, the mechanism is straightforward. If return volatility clusters, then the uncertainty around estimated means, ratios, and test statistics also changes through time. A procedure that treats every observation as if it came from the same variance regime can understate or misstate uncertainty. That creates a gap between the assumptions of a naive test and the data generating behavior of market returns.

Sonar Sciences describes this issue directly in its strategy validation research. The framework emphasizes that financial backtests are vulnerable to statistical distortions when researchers rely on conventional significance checks without adjusting for the realities of market data, including non normality and unstable variance. In that setting, an apparently strong result can reflect properties of the sampling process rather than durable evidence about a strategy. This is one reason robust validation needs more than a simple in sample summary statistic or a naive hypothesis test.

The same practical concern appears in Sonar Sciences' backtest overfitting audit. The tool is designed to stress test backtest evidence rather than accept headline metrics at face value. Its documentation frames overfitting and exaggerated significance as risks that arise when researchers do not properly account for the statistical structure of returns. That is consistent with the broader problem of heteroskedasticity. When variance changes over time, unadjusted estimates of precision become less reliable, so a backtest can look more statistically convincing than the underlying evidence supports.

A related correction appears in the deflated Sharpe ratio methodology described by Sonar Sciences. The glossary explains the deflated Sharpe ratio as a way to assess whether an observed Sharpe ratio is likely to be genuinely significant after accounting for multiple testing and non normal return behavior. This matters in heteroskedastic settings because changing variance can destabilize the inputs behind naive risk adjusted performance summaries. A plain Sharpe ratio calculation is easy to compute, but if return variance is unstable, interpreting it with standard textbook assumptions can be misleading. The deflated Sharpe ratio is presented as a more conservative statistical lens for that reason.

The key implication is simple. Heteroskedasticity is not a minor technical nuisance in market data. It is a standard feature of returns, and it breaks the equal variance assumption behind many naive tests. For quantitative traders and research analysts, the practical response is to use validation methods that recognize unstable variance instead of relying on statistical procedures that implicitly assume a constant volatility environment.

Covered in depth in the Strategy validation & overfitting pillar hub.