underdispersion

[Univ_Aveiro_util]

Someone could help me to correct underdispersion in Poisson estimation?

I tried to do the Negative Binomial (QLM), but when I enter the fixed variance parameter (which is negative in my case) the estimation results says that the QML parameter used in estimation is 1 (not -0.01907 that I enter)...

Thank you for your help.

[EViews Glenn]

The negative binomial parameterization we use only allows for overdispersion. What you can do is to estimate with a Poisson and use GLM standard errors which should be valid given underdispersion. The Huber-White standard errors should also be robust to underdispersion.

[Univ_Aveiro_util]

Thank you very much for your reply. Could you, please, explain me how i should do that in Eviews?
- In the equation estimation panel for Poisson (integer count), in options, should i select Robust Covariances GLM instead of Huber-White? Is it enough?


Thank you.

[EViews Glenn]

Yes. Either should be robust to underdispersion.