Test Strategies for Discriminating Between Autocorrelation and Misspecification

Abstract

The main aim of the Durbin-Watson (DW) test is to detect an AR(1) process among the disturbances in a linear regression model. However, it is well known that a significant value of the DW test statistic, dw, can be caused by various types of misspecification, viz., AR(1) disturbances, non-AR(l) autocorrelated disturbances (NAR) and misspecification of the regression function (MS), i.e., explanatory variables are wrongly omitted or the functional form of the model is incorrect. Since a wrong diagnosis of the problem may lead to a wrong correction procedure and, consequently, to poor estimation results it is of great interest to have available a test strategy which is capable of discriminating between the three types of misspecification mentioned above.