A persistent regressor and serially correlated errors

Question

You run a predictive regression of next-month returns on the dividend yield, a highly persistent, slow-moving variable. The residuals show clear serial correlation.

Explain why ordinary standard errors mislead here, why Newey-West HAC standard errors are the standard remedy, and what additional small-sample bias you should be aware of.

Accepted Answer

Why ordinary standard errors mislead. The regressor moves slowly and the residuals are serially correlated, so successive observations are not independent draws. Classical (and White) standard errors assume no autocorrelation, so they understate the true uncertainty of the slope. You get an overstated t-statistic and can wrongly conclude the dividend yield predicts returns. The HAC remedy. Newey-West standard errors are heteroskedasticity- and autocorrelation-consistent: they widen the robust variance to include weighted covariances between residuals at nearby dates, using a declining (Bartle…

A persistent regressor and serially correlated errors

Your answer

Discussion

Learn the concepts

Related questions