Performance of the Durbin-Watson Test and WLS Estimation When the Disturbance Term Includes Serial Dependence in Addition to First-Order Autocorrelation

Abstract

Monte Carlo simulation is used to study the power of the Durbin-Watson test and the properties of the corresponding weighted least squares (WLS) estimates when there is serial correlation in the disturbance term, in addition to first-order autocorrelation. The results indicate that the Durbin-Watson test detects first-order autocorrelation, even when other forms of serial dependence are also present. However, routine use of WLS estimation when the Durbin-Watson test is significant may result in inaccurate and inefficient parameter estimates. Therefore, this procedure should be used with caution unless there is a priori knowledge concerning the nature of any serial dependence in the disturbance terms.

Key Words:

• Durbin-Watson test
• Linear regression
• Robustness
• Serial dependence
• Simulation
• WLS estimation