t-stats grouped DOLS estimator

[funkymind]

Hello,

I recently estimated some relationships using the grouped version of the Dynamic Ordinary Least Square Estimator (DOLS). I run the estimations using panel data, but I also calculated the coefficients as well as the t-stats from individual countryspecific estimates. I noted that these manually calculated t-stat values strongly deviate from that ones shown in the panel regression table. I base my own calculations on the formula provided by Pedroni (2001) t_b = N^(-1/2) Sum(t_b_i). How does E-Views calculates the panel t-Statistic?

Best,

K.

[EViews Glenn]

I don't have the docs in front of me, but I am almost certain that this is an area where we differ by choice. The documentation should describe our calculation...I believe that we don't sum the t-stats, but rather average the coefficients in the numerator and sum the variances and take roots for the denominator.

[funkymind]

You are right. You use the sum of the individual coefficients and divide it by the root of the sum of variances. However, this method (at least in my case) is much more restrictive and leads to lower values of the t-statistic.

[EViews Glenn]

We differ in the computation since EViews needs to be able to compute a full covariance matrix for the coefficients since that's the way all of our equation objects work. Obviously, the Pedroni averaging of the t-statistics doesn't provide that information. Sounds like you were able to do the calculation yourself (I'm not sure what method you used), but I want to note that if you need to replicate his computation using the EViews estimator, you can save the individual coefficient results matrix and covariances matrix using the Options page, then compute the relevant t-statistics yourself. Note that the covariances are saved in vech form, one to each row.