Cluster Standard Errors by firm and time

[Kouvas]

Hello there

I am trying to estimate a regression with clustered standard errors by firm and time following the study of Thomson (2011). The author proposes a simple method by adding the estimated variances from a regression with errors clustered by firm and a regression with errors clustered by time and then subtracting the standard errors from a regression using heteroscedasticity robust standard errors. In order to calculate this in eviews should I run the regression using white cross-section (clustered by period), then white period (clustered by firm) and then white diagonal (heteroscedasticity robust)? In this case the squares of the standard errors obtained for each coefficient will be the desired variances? Thanks in advance.

[EViews Glenn]

http://forums.eviews.com/viewtopic.php? ... ust#p15983

[Kouvas]

Thank you very much, the post clarifies the whole procedure, however, in some of the regressions I run the square root of the variances cannot be estimated due to having a negative number when calculating the sum of the variances of regs with clustering by firm and time and then subtracting the variance of the reg with stacked white. Any possible solutions? Thanks in advance.