Overlapping data and hedge funds

Abstract

We consider the overlapping data problem. The conventional estimation approach with overlapping data is to use the Newey-West estimation procedure. When the standard assumptions hold generalized least squares is asymptotically efficient. Monte Carlo results show that the Newey-West procedure has considerably larger variances of parameter estimates and lower power than GLS. Hypothesis tests using the Newey-West procedure also have incorrect size even with sample sizes as large as one thousand. We also discuss possible estimation approaches when overlapping data occurs in conjunction with some other econometric problem. With lagged dependent variables or errors in the explanatory variables, GLS is no longer the preferred estimator.