F stat in dynamic model

[viviancwj]

I noticed that Eviews does not report F-stat in dynamic model estimation results. Could anyone tell me why?

Thank you.

[EViews Gareth]

Could you define what you mean by dynamic model estimation?

[viviancwj]

Thanks for your reply, Gareth.

When I use dlog(Y) as left hand side variable, or use lagged dependent variable as right hand side variable, F stat is not reported in the regression output. I am using simple OLS estimation.

[EViews Gareth]

Code: Select all

create u 100 series y=nrnd equation eq1.ls y c y(-1) show eq1

That displays an F-stat.

[startz]

Gareth meant to ask you what version of EViews are you using and what's the build date. :)

[viviancwj]

I am using Eviews Enterprise Version 9. And my sample is quite short: quarterly data from 2004 to 2015.

By looking at your example, I think I know why I don't have F-stat: I don't have constant in my regression.

Thank you for your help.

[EViews Glenn]

EViews won't produce an F-statistic if you don't have an intercept in the model as the two are non-nested.

[startz]

EViews won't produce an F-statistic if you don't have an intercept in the model as the two are non-nested.

They are perfectly well nested. The F-stat is
((y'y-ssr)/k)/(ssr/(n-k))
There is, however, a risk of confusion when the RHS includes dummies.

[EViews Glenn]

For better or worse, EViews always computes the main F-statistics relative to the intercept only model. The OP indicated that she didn't have an intercept in her model which is why EViews doesn't produce the F-stat as her model doesn't nest the former.

In a general world in which one is comparing the LR for a model with regressors versus one without regressors, your comment about a valid nested test is certainly true. We could switch to a system in which one computed against an intercept only null if present under the alternative and against an empty model if not, but given that the output doesn't have space for description of the null, I'm not crazy about having a null hypothesis that changes. And as you point out, things get a bit tricky in the dummy variable case.

[startz]

I agree that the confusion isn't worth it

[EViews Glenn]

That was our thought, but we're always willing to hear arguments to the contrary.