Overdispersion test

[Nitzan04]

equidispersion tests.docx

(24.36 KiB) Downloaded 574 times

I did both test that are mentioned in the manual Cameron and Trivedi and Wooldridge - is it reasonable that i got underdispersion at the first test and overdispersion at the last test?
Im really confused.

[EViews Glenn]

For the first test, which variable is the original and which variable is the fitted?

[Nitzan04]

Hi,
The first variable is the fit variable which i got by using the command: fit patentfittct . the second variable i got by "forecast".
Doesnt fit has the same meaning as forecast?!
It is really confusing because at the manual it written , and i quote :"first estimate the Poisson and obtain the fitted values of the dependent variable"
What should basically i need to do? can you guid my steps in Eviews so i can get a clear picture of the results of the dispersion.
Thank you in advance,
Nitzan.

[Nitzan04]

This is the result i just got after doing : a. astimate the variables by the Poisson-Count regression model, b. forecast and named the forecasted variable as " patentffinal", c. using the command :
equation testeq.ls (patent-patentffinal)^2-patent patentffinal^2.

Is it o.k?

[EViews Glenn]

I haven't looked at the Cameron and Trivedi stuff in ages, so I haven't verified this, but from the description in the manual, your specification for the original test equation is incorrect.

The original test equation you estimated is
Your original equation was

Code: Select all

equation testeq.ls (patentfittct - patentfct)^2-patentfittct patentfct^2.

when the manual suggests that you should have estimated

Code: Select all

equation testeq.ls (patentfct - patentfittct)^2-patentfct patentfittct^2.

Your second attempt uses

Code: Select all

equation testeq.ls (patent-patentffinal)^2-patent patentffinal^2.

which looks to be of the correct form.