Standard errors calculation in fixed effects by Eviews

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

Wiz_Ven1
Posts: 2
Joined: Thu Jan 28, 2021 2:30 am

Standard errors calculation in fixed effects by Eviews

Postby Wiz_Ven1 » Thu Jan 28, 2021 2:33 am

Dear Eviews programmer.

I am a student. I am writing you with respect to the Cross-section fixed (dummy variables) in Eviews. While is clear to me as the estimated coefficient of constant is obtained (mean of the estimated coefficient of the dummies of the least square dummy variables estimator), less clear is the computation of the standard errors of the estimated coefficient of the constant. Please can you kind tell me how are computed (what formula is used for that). Thank you in advance for your help.

Wiz_Ven1
Posts: 2
Joined: Thu Jan 28, 2021 2:30 am

Re: Standard errors calculation in fixed effects by Eviews

Postby Wiz_Ven1 » Sun Jan 31, 2021 2:08 am

Could anyone please help me? Tks!

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

Re: Standard errors calculation in fixed effects by Eviews

Postby EViews Glenn » Tue Feb 02, 2021 9:07 am

There are many standard errors in panel data models so I'm answering this with respect to the simplest form. The basics apply to the other methods as well.

In the usual way of computing fixed effects estimators and the associated standard errors, we orthogonalize all of the data with respect to the dummy variables and compute standard errors from the "transformed" model. If this orthogonalization matrix is defined as Q, the standard errors are proportional to (X'QX)^{-1}.

Colloquially, we demean the data and then compute the usual OLS covariance matrix, adjusting the d.f. for the extra implied coefficients that are partialed out in the transformation.

In EViews we do roughly the same thing, but instead of the former orthogonalization matrix, we use the idempotent matrix Q* which projects onto the Q subspace but adds back in the overall means. This allows us to estimate the dummy variables as differences from overall means. The remainder of the computation is the same using (X'Q*X)^{-1}.


Return to “Estimation”

Who is online

Users browsing this forum: No registered users and 34 guests