A COMPARISON OF ALTERNATIVE COVARIANCE MATRICES FOR MODELS WITH OVER-LAPPING OBSERVATIONS

This paper investigates the relative performance of alternative covari ance matrices for models with ever-lapping observations commonly used in the finance literature. The alternative covariance matrices used ar e those of Hansen (1982), Newey and West (1987) (Bartlett and Quadrati c Spectral (QS) weights) and Andrews and Monahan (1992) (QS weights). All matrices produce standard errors which are too small, yielding emp irical size probabilities above their corresponding theoretical values , even in large samples. Empirical examples, such as testing efficienc y in the foreign exchange market and mean reversion in stock prices, s how that the choice of covariance matrix can affect the outcome of a h ypothesis test. Copyright (C) 1996 Elsevier Science Ltd