Overlapping returns and the need for HAC standard errors

Question

You regress 12-month-ahead returns on a predictor, but you sample the data monthly, so consecutive observations share 11 months of overlapping return window. The residuals are therefore strongly autocorrelated, and may be heteroskedastic too.

Explain why White (heteroskedasticity-robust) standard errors are not enough here, what Newey-West HAC standard errors do differently, and how you would pick the lag length.

Accepted Answer

Why the residuals are autocorrelated. Because each 12-month return overlaps the next by 11 months, adjacent observations share most of their underlying data. Their errors move together, inducing strong positive autocorrelation that decays over roughly 12 lags (the overlap length). This is a mechanical feature of overlapping windows, not a modeling mistake. Why White standard errors are not enough. White standard errors correct for heteroskedasticity, non-constant variance across observations, but they still assume the errors are uncorrelated across time. With overlapping returns that assumpti…

Overlapping returns and the need for HAC standard errors

Your answer

More Statistics questions