Logit estimation with Panel data

For technical questions regarding estimation of single equations, systems, VARs, Factor analysis and State Space Models in EViews. General econometric questions and advice should go in the Econometric Discussions forum.

Moderators: EViews Gareth, EViews Moderator

Aktar
Posts: 74
Joined: Thu May 21, 2009 5:08 am

Logit estimation with Panel data

Postby Aktar » Fri Aug 27, 2021 7:53 am

Hello,

I have a panel dataset (T=157 and N=7) where the dependent variable is a continuous probability bounded between 0 and 1. I want to explain these probabilities by a set of economic variables (specific and common).

I have transformed the dependent variables as binary variables (0 and 1) since these probabilities are very close to 0 or 1. Then, I would like to estimate a logit model with the BINARY option. I will remove the constant and individual effects by demeaning all the variables.

I am not sure how Eviews is handling logit models with panel dataset. Is it possible to use to proceed like this ?

I am using the last EVIEWS's version.

Thank you very much

EViews Glenn
EViews Developer
Posts: 2671
Joined: Wed Oct 15, 2008 9:17 am

Re: Logit estimation with Panel data

Postby EViews Glenn » Fri Aug 27, 2021 10:09 am

Binary estimation in a panel structured workfile is just pooled estimation obtained by stacking all of the cross-sections and performing the estimation.

I am not sure of the rationale for transforming to a binary logit model. Unless you believe there is recording error, this would appear to throw away the exact probability information to no benefit. Given that you continue along these lines, I do note that removing the constant and individual effects by demeaning is not the standard way of controlling for the effects in this setting as the binary logit model is nonlinear.

For your original data, the natural approach would be to simply estimate the linear probability model with a set of dummy variables corresponding to your cross-sections. The usual fixed effects estimator could be used to do this automatically. There are, of course, issues with the LPM that most econometric textbooks will discuss.

Hope this helps.

Aktar
Posts: 74
Joined: Thu May 21, 2009 5:08 am

Re: Logit estimation with Panel data

Postby Aktar » Fri Aug 27, 2021 2:22 pm

Thank you very much !

So the binary option is no longer the good choice…

My original choice was to use a S shaped functional form to link my continuous probabilities and the explaining variables.

Can I fit a logistic parametric form to my model and deal with fixed effects (with Panel options) at the same time in a panel workfile ?

The functional form would be something like
y = 1 / (1+ exp -(c(1)+c(2)*var1+c(3)*var2))

Thank you very much once again

EViews Glenn
EViews Developer
Posts: 2671
Joined: Wed Oct 15, 2008 9:17 am

Re: Logit estimation with Panel data

Postby EViews Glenn » Fri Aug 27, 2021 2:28 pm

Anything nonlinear like this will require dummy variables, not demeaning. You don't have many and a reasonable T, I would just add them to the spec.

Aktar
Posts: 74
Joined: Thu May 21, 2009 5:08 am

Re: Logit estimation with Panel data

Postby Aktar » Fri Aug 27, 2021 9:49 pm

Ok thanks,

So if my understanding is correct, I need to write the following equation specification without selecting "Fixed effects" in "Panel options":

y = 1 / (1 +exp(-(c(1)*dum_1+c(2)*dum_2+c(3)*dum_3+c(4)*dum_4+c(5)*dum_5+c(6)*dum_6+c(7)*dum_7+c(8)*x1+c(9)*x2)))

Thanks

EViews Glenn
EViews Developer
Posts: 2671
Joined: Wed Oct 15, 2008 9:17 am

Re: Logit estimation with Panel data

Postby EViews Glenn » Wed Sep 01, 2021 10:11 am

Sounds right.

EViews Glenn
EViews Developer
Posts: 2671
Joined: Wed Oct 15, 2008 9:17 am

Re: Logit estimation with Panel data

Postby EViews Glenn » Wed Sep 01, 2021 10:13 am

Sounds right. Though I think you might want to look at using GLM for this specification.

Aktar
Posts: 74
Joined: Thu May 21, 2009 5:08 am

Re: Logit estimation with Panel data

Postby Aktar » Thu Sep 02, 2021 7:16 am

Yes I was interested by GLM, but seems we can't perform the estimation in a Eviews Panel framework.

Don't want to make a confusion, but Am I right ?

One other issue is to deal with cross section dependence which is obviously existing in my dataset. Eviews offers interesting options for estimating the standard errors in a panel ols framework (white...). I guess glm mixed methods could work but I didn't check.

What do you think ?

Thank you for your insights !

EViews Glenn
EViews Developer
Posts: 2671
Joined: Wed Oct 15, 2008 9:17 am

Re: Logit estimation with Panel data

Postby EViews Glenn » Fri Sep 03, 2021 1:23 pm

GLM may be estimated but it doesn't use the panel structure. In your case, that's not an issue since you are handling the dummy variables yourself.

You are correct that GLM in a panel setting won't have the ability to handle cross-section dependence.

Aktar
Posts: 74
Joined: Thu May 21, 2009 5:08 am

Re: Logit estimation with Panel data

Postby Aktar » Wed Sep 08, 2021 4:02 am

Thank you,

I would have a last question regarding the estimator of the standard errors.

My errors are auto-correlated and cross-section dependent.

My dataset includes countries that are neighborhood countries and I can assume the variance to be the same for the cross-variant regressors and the dependent variables. However, I have only 7 cross-sections and T is not asymptotics.

I guess the White cross section estimator is not suitable in this case, so I am wondering if I need to use the cross-section SUR estimator...

Which estimators to favor ? Noting that in many cases, the results are very close.

May I have your thoughts on this matter please ?

Thank you once again for your help


Return to “Estimation”

Who is online

Users browsing this forum: No registered users and 23 guests