The Inappropriate Use of Serial Correlation Tests in Dynamic Linear Models.

A survey of several economic journals reveals that very often the Durbin-Watson and the portmanteau (Box-Pierce or Ljung-Box) tests are inappropriately applied to linear models with lagged dependent variables and exogenous regressors. Sampling experiments indicate that the Durbin-Watson performs poorly in models with more than one lag of the dependent variable, a situation commonly considered in the literature. The experiments also indicate that the portmanteau test is inadequate when applied to dynamic linear models with exogenous regressors. In addition, the performance of Durbin's h and m tests in models commonly used in the literature but not considered by previous studies is evaluated. The results reveal that among the four tests examined, the one which is the least frequently used in practice (the m test) has the best performance. Copyright 1990 by MIT Press.