6.3 Testing for Autocorrelation

P&R present two tests for autocorrelation: the Durbin-Watson test and the Durbin-h test. TSP calculates these two tests automatically when performing OLS. The Durbin-Watson test is printed in the output and its value is stored under @DW. The Durbin-h and the Durbin-h alternative are not printed normally, but they are saved under @DH and @DHALT. When a lagged dependent variable is included in the regression these two tests are automatically included in the OLS output, since in that case, the Durbin-Watson is biased.

6.4 Correcting for Autocorrelation

P&R introduce two methods for correcting for serial correlation: the Cochrane-Orcutt procedure and Hildreth-Lu . TSP's AR1 can correct for autocorrelation using these methods as well as several others. In general, the estimates are efficient if the errors in the equation are homoskedastic and have first order serial correlation, regardless of the estimation method used. The methods supplied by TSP are either two-stage procedures (Cochrane-Orcutt and Hildreth-Lu) or maximum likelihood. The format of the AR1 statement follows:

AR1(METHOD=CORC or HILU or ML or MLGRID) dep-var list-of-indep-vars;

CORC stands for Cochrane-Orcutt, HILU for Hildreth-Lu, ML for maximum likelihood, and MLGRID for maximum likelihood using the grid method.

AR1 output shows the estimated autocorrelation coefficient rho (), as well as its standard error. These values are stored under the names @RHO and @SRHO.