Dear all,
I had posted my question before but the post was very long so I want to write it in a more compact form. I'm interested in testing whether beta_market in my gls-panel-regression (for 20 periods) significantly differs between the subperiods 1-10 and 11-20. I have done 3 xtgls regressions without & 3 with the options panels(heteroskedastic) corr(ar1).
In the first case (colums 1-3) I got b_market = 0.106 for the period 1-10 and b_market = 0.192 for the period 11-20, regardless if I look at the subperiods regressions (column 2-3) or the full pooled regression (1). Also - because beta_1.dummy#c is significant - i would conclude, that the subperiod beats are significantly different.
This is not the case when i do the same 3 regressions with the options panels(heteroskedastic) corr(ar1)(column 3-6). Here the market betas depend on whether I look only at the pooled regression in column 4 or the subperiod regressions in column 5 and 6. So my question is: to which market betas the insignificant beta_1.dummy#c in column 4 refers, the ones resulting in column 4 -> 0.099 and 0.099 + 0.020 or the ones resulting from column 5 and 6-> 0.083 and 0.112.
I tried to copy the table as CODE, but it looked very bad (in the preview) so I added a screenshot of the table.
I had posted my question before but the post was very long so I want to write it in a more compact form. I'm interested in testing whether beta_market in my gls-panel-regression (for 20 periods) significantly differs between the subperiods 1-10 and 11-20. I have done 3 xtgls regressions without & 3 with the options panels(heteroskedastic) corr(ar1).
In the first case (colums 1-3) I got b_market = 0.106 for the period 1-10 and b_market = 0.192 for the period 11-20, regardless if I look at the subperiods regressions (column 2-3) or the full pooled regression (1). Also - because beta_1.dummy#c is significant - i would conclude, that the subperiod beats are significantly different.
This is not the case when i do the same 3 regressions with the options panels(heteroskedastic) corr(ar1)(column 3-6). Here the market betas depend on whether I look only at the pooled regression in column 4 or the subperiod regressions in column 5 and 6. So my question is: to which market betas the insignificant beta_1.dummy#c in column 4 refers, the ones resulting in column 4 -> 0.099 and 0.099 + 0.020 or the ones resulting from column 5 and 6-> 0.083 and 0.112.
I tried to copy the table as CODE, but it looked very bad (in the preview) so I added a screenshot of the table.
Code:
. use http://www.stata-press.com/data/r12/invest2.dta . xtset company time panel variable: company (strongly balanced) time variable: time, 1 to 20 delta: 1 unit . . gen dummy = 0 if inrange(time,1,20) . . replace dummy = 1 if inrange(time,11,20) (50 real changes made) . . quiet xtgls invest c.market##i.dummy i.time . . estimates store full_wo_opt . . quiet xtgls invest market i.time if inrange(time, 1, 10) . . estimates store firstperiod_wo_opt . . quiet xtgls invest market i.time if inrange(time, 11, 20) . . estimates store secondperiod_wo_opt . . quiet xtgls invest c.market##i.dummy i.time, panels(heteroskedastic) corr(ar1) . . estimates store full_w_opt . . quiet xtgls invest market i.time if inrange(time, 1, 10), panels(heteroskedastic) corr(ar1) . . estimates store firstperiod_w_opt . . quiet xtgls invest market i.time if inrange(time, 11, 20), panels(heteroskedastic) corr(ar1) . . estimates store secondperiod_w_opt . . estout full_wo_opt firstperiod_wo_opt secondperiod_wo_opt full_w_opt firstperiod_w_opt secondperiod_w_opt, cells(b(star fmt(3)) se(par fmt(2)))
Attached Files
Comment