Ricardo Gonçalves Silva schrieb: > Hi, > > Can Gretl forecast 1 to 3 periods ahead an estimated panel > data(fixed-effects with no iterations) model? > sorry this is not an answer to your question: I wanted to try it out myself and got a crash with this little script (on today's cvs version): <script> open abdata smpl +0 -3 panel n 0 w --fixed-effects fcast </script> I'm 99% sure there are bugs (probably several) in this mini-script, so this likely doesn't have anything to do with your question. But still it's a bug if gretl crashes. thanks, sven _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users

Nenhum vírus encontrado nessa mensagem recebida. Verificado por AVG - www.avgbrasil.com.br Versão: 9.0.791 / Banco de dados de vírus: 271.1.1/2760 - Data de Lançamento: 03/20/10 16:33:00

On Sat, 20 Mar 2010, Sven Schreiber wrote: > Ricardo Gonçalves Silva schrieb: > > Hi, > > > > Can Gretl forecast 1 to 3 periods ahead an estimated panel > > data(fixed-effects with no iterations) model? > > sorry this is not an answer to your question: I wanted to try it out > myself and got a crash with this little script (on today's cvs version): > > <script> > open abdata > smpl +0 -3 > panel n 0 w --fixed-effects > fcast > </script> Forecasting via the "fcast" command in association with panel models (other than pooled OLS) is not implemented at this point, sorry. The crash from the script above is now fixed in CVS and the snapshots, simply by flagging a "not implemented" error. Note that the effect of doing smpl ; -3 with panel data is to discard the last 3 panel units, not the last 3 time-series observations for each unit (which could not be done simply by moving the sample end-points). It is, however, possible to generate a forecast manually based on fixed-effects estimates. The following script illustrates. <script> open greene14_1.gdt # number of cross-sectional units nunits = max($unit) # number of time periods T = $nobs/nunits logs C Q PF list X = l_Q l_PF LF # estimate on data up to 1979 smpl year < 1980 --restrict panel l_C 0 X # retrieve fitted values series yh = $yhat # plus the per-unit intercepts series ah = $ahat # and the coefficients on X matrix b = $coeff[2:] # restore full time-series length smpl --full # expand the per-unit intercepts loop i=1..nunits loop t=11..T k = T*(i-1)+t ah[k] = ah[k-1] endloop endloop series Xb = {X}*b + ah print -o yh Xb </script> This script shows that the constructed forecast series "Xb" agrees with the fitted values obtained via $yhat within sample. Allin Cottrell _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users

Nenhum vírus encontrado nessa mensagem recebida. Verificado por AVG - www.avgbrasil.com.br Versão: 9.0.791 / Banco de dados de vírus: 271.1.1/2762 - Data de Lançamento: 03/21/10 16:33:00

On Thu, 21 Aug 2008, [iso-8859-1] Ricardo Gonçalves Silva wrote: > Thanks so much. I will try the script now. > Only a newbie question: the fcasts generated are out-of-sample, right? Yes, the William Greene dataset I used has data for 6 firms from 1970 to 1984. My script estimates the fixed-effects model over the range 1970 to 1979, then generates forecasts for 1970-1984. I produced "forecasts" for all years just to show that the in-sample values agree with $yhat, but presumably it's the out of sample values that are of most interest. Allin Cottrell _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users

Nenhum vírus encontrado nessa mensagem recebida. Verificado por AVG - www.avgbrasil.com.br Versão: 9.0.791 / Banco de dados de vírus: 271.1.1/2762 - Data de Lançamento: 03/21/10 16:33:00

On Thu, 21 Aug 2008, [iso-8859-1] Ricardo Gonçalves Silva wrote: > Thanks so much. I will try the script now. > Only a newbie question: the fcasts generated are out-of-sample, right? Yes, the William Greene dataset I used has data for 6 firms from 1970 to 1984. My script estimates the fixed-effects model over the range 1970 to 1979, then generates forecasts for 1970-1984. I produced "forecasts" for all years just to show that the in-sample values agree with $yhat, but presumably it's the out of sample values that are of most interest. Allin Cottrell _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users

On Tue, 9 Aug 2011, Leon Unger wrote: > I want to calculate$V[\widehat{u_{ijt}}]$ for *Fixed Effect *models. > In a first step I reproduce the forecast error variance of a pooled OLS > model. > Using the following formula > > $V[\widehat{u_{ijt}}] = V[{\epsilon}] > +{{x_{ijt}}^{'}*V[\hat{\beta}]*{x_{ijt}}$ > > which provided the same values as GRETL produces with $fcerr. > Can I apply the same calculation after the Fixed Effect? That would, I think, ignore the uncertainty due to estimation of the fixed effects. One option would be to estimate the fixed effects model manually, by adding the unit dummies to the dataset and applying OLS (gretl estimates the FE model by de-meaning the data). Allin Cottrell _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users

after investigating further - because the results were quite unsatisfactory - I found out that: 1) there is no memory problem with creating the unit dummies if done via 'dummify' (took millisecs for 865 dummies). Using 'genr unitdum' even did not succeed when shrinking down the dataset to only include 30 units. I guess there is some bug.