Re: [Gretl-users] test for autocorrelation in panel estimation
On Tue, 12 Apr 2011, Hakim, Sam wrote:
I am estimating demand elasticities using panel data and I am
using effect model in Gretl. I suspect that the independent and
dependent variables are correlated due to a simulataneous
equation bias.
Is there a simple test for autocorrelation I can run? For
whatever reason, the DW test option is not available for this
model.
Testing for autocorrelation doesn't tell you anything about
simultaneous equations bias.
But if the DW test is not reported that suggests that either the
time-series length is too short or there are missing values within
the sample. If the problem is missing values and you're willing to
ignore that issue, you could save the residuals (they're available
as $uhat) and construct a test (Durbin-Watson or otherwise)
manually -- e.g. by setting the missing residuals to zero.
Allin Cottrell