Daily equity index returns typically show autocorrelation near zero, often slightly negative at lag one. Squared or absolute returns, by contrast, show strong positive autocorrelation lasting weeks, which is volatility-clustering stated numerically.
It matters for testing as much as for signals. Overlapping observations, such as 20-day forward returns sampled daily, are heavily autocorrelated, which inflates apparent significance because the effective sample is far smaller than the row count. Corrections like Newey-West standard errors or a block-bootstrap handle this.
Worked example: 1,000 daily rows of 20-day overlapping returns contain roughly 50 independent observations. Treating them as 1,000 understates the standard-error by about sqrt(20), or 4.5 times, which can turn noise into a p-value of 0.001.
Related: volatility-clustering, block-bootstrap, embargo-period, mean-reversion-half-life