The mistake of using White SEs on autocorrelated errors

Question

An analyst runs a time-series regression, notices the errors look heteroskedastic, and dutifully reports White heteroskedasticity-robust standard errors. But the residuals are also strongly positively autocorrelated, which the analyst ignores.

Diagnose what is wrong with this choice, which way the reported standard errors are likely biased, and what the analyst should have done.

Accepted Answer

The blind spot. White standard errors are robust to heteroskedasticity but not to autocorrelation. They correct the variance for unequal spread across observations while still assuming the errors are uncorrelated over time. When the residuals are serially correlated, that assumption fails, and White standard errors are just as invalid as the classical ones on this dimension. Direction of the bias. With positive autocorrelation, effective sample information is smaller than the raw count suggests: neighboring observations partly repeat each other, so there is less independent evidence than n po…

The mistake of using White SEs on autocorrelated errors

Your answer

Discussion

Learn the concepts

Related questions