Autocorrelated disturbances in the light of specification analysis

The paper reconsiders the problem of autocorrelation among disturbances caused by the omission of some regressors from a single-equation regression model. Regression coefficients and disturbances of the misspecified model have been redefined appropriately. Performance of the OLS method of estimation of these regression coefficients and that of the Durbin-Watson test of randomness of disturbances have been studied. Some of the alternative methods of estimating the regression coefficients in situations where the disturbances are autocorrelated heve been examined. It appears that these methods can no longer be used if the autocorrelation is due to omission of regressors.