TESTING FOR AUTOCORRELATION IN THE PRESENCE OF LAGGED DEPENDENT-VARIABLES - A SPECIFICATION ERROR APPROACH

Hausman's specification error testing procedure is used to develop serial correlation tests in lagged dependent variable models. Properties of the tests are discussed and comparisons are made with existing tests. A reparameterization of the model using the error structure under the alternative hypothesis results in an augmented regression equation; standard tests of exclusionary restrictions on coefficients of the augmented regressors can be regarded as serial correlation tests. Monte Carlo experiments are used to examine small sample properties. Several of the new tests are shown to perform well in comparison with existing tests.