diagnostic checks

[johnsmith]

k

[startz]

The Drubin-Watson suggests you have significant serial correlation.

[johnsmith]

thanks for your reply

Is there a way to sort this out?

[startz]

You probably have a variety of omitted slow-moving variables that would help explain the spread. The best thing to do would be to include such variables. The second best might be to make a statistical adjustment. Look at AR(1) in the EViews help.

[johnsmith]

k

[startz]

Probably.

The remaining issue is whether this has picked up a real lagged effect or just masked serial correlation. Unfortunately, in the latter case all the coefficients are screwed up. Normally, one would test this last model for serial correlation. Unfortunately, EViews doesn't seem to implement a test for panel data.

The other issue is to realize that this new model introduces a dynamic response that you have to take into account.