How to cluster standard error at country level?

[Edward Wu]

Hey!

I tried to run panal data regression, and worked for cluster standard error at firm level(permno). (ym is year and month)

xtset permno ym
xtreg f6ret PF POSTPF POSTT PFT POSTPFT i.year, fe cluster(permno)

but when I try
xtreg f6ret PF POSTPF POSTT PFT POSTPFT i.year, fe cluster(country)

it says 'panels are not nested within clusters'.

Sample:

* Example generated by -dataex-. To install: ssc install dataex
clear
input double permno float(ym f6ret year PF POSTPF POSTT PFT POSTPFT country)
10207 368     .06632764 1990 .7 0 0 0 0 4
10207 369     -.0794038 1990 .9 0 0 0 0 4
10207 370   -.003264269 1990 .9 0 0 0 0 4
10207 371    -.05579877 1990 .9 0 0 0 0 4
10207 372    -.20440815 1991  1 0 0 0 0 4
10207 373    -.05142857 1991 .8 0 0 0 0 4
10207 374   -.024201704 1991 .6 0 0 0 0 4
10207 375     .10524093 1991 .2 0 0 0 0 4
10207 376     .10358141 1991 .4 0 0 0 0 4
10207 377     .10524093 1991 .2 0 0 0 0 4
10207 378     .17792796 1991 .1 0 0 0 0 4
10207 379      .1547932 1991 .1 0 0 0 0 4
10207 380     .16414873 1991 .5 0 0 0 0 4
10207 381      .1838134 1991 .7 0 0 0 0 4
10207 382     .07760347 1991 .7 0 0 0 0 4
10207 383     .10672788 1991 .8 0 0 0 0 4
10207 384     .15785037 1992 .7 0 0 0 0 4
10207 385      .1630513 1992 .7 0 0 0 0 4
10207 386    .033496205 1992 .7 0 0 0 0 4
11077 366   -.59090906 1990 .2 0 0 .2 0 22
11077 367          -.5 1990 .2 0 0 .2 0 22
11077 368    .20833334 1990 .1 0 0 .1 0 22
11077 369    .08333334 1990 .1 0 0 .1 0 22
11077 370           .5 1990 .1 0 0 .1 0 22
11077 371      .888889 1990 .1 0 0 .1 0 22
11077 372     .7777778 1991 .1 0 0 .1 0 22
11077 373      .631579 1991 .1 0 0 .1 0 22
11077 374  -.034482747 1991 .8 0 0 .8 0 22
11077 375    .03846155 1991 .4 0 0 .4 0 22

11845 449   .16019306 1997 .5 0 0 .5 0 12
11845 450   .11007067 1997 .2 0 0 .2 0 12
11845 451   .06033811 1997 .4 0 0 .4 0 12
11845 452   .16861244 1997 .5 0 0 .5 0 12
11845 453    .2673978 1997 .5 0 0 .5 0 12
11845 454     .395664 1997 .6 0 0 .6 0 12
11845 455    .4124138 1997 .7 0 0 .7 0 12
11845 456   .49556565 1998 .7 0 0 .7 0 12
11845 457    .5543182 1998 .8 0 0 .8 0 12
11845 458  .017565502 1998 .7 0 0 .7 0 12
11845 459  -.21058515 1998 .9 0 0 .9 0 12
end
format %tm ym

Thank you!

[Carlo Lazzaro]

Edward:
I was not able to replicate the problem you reported:

. xtset permno ym
       panel variable:  permno (unbalanced)
        time variable:  ym, 1990m7 to 1998m4
                delta:  1 month

. xtreg f6ret PF POSTPF POSTT PFT POSTPFT i.year, fe cluster(permno)
note: POSTPF omitted because of collinearity
note: POSTT omitted because of collinearity
note: POSTPFT omitted because of collinearity
note: 1998.year omitted because of collinearity

Fixed-effects (within) regression               Number of obs     =         40
Group variable: permno                          Number of groups  =          3

R-sq:                                           Obs per group:
     within  = 0.1445                                         min =         10
     between = 0.6243                                         avg =       13.3
     overall = 0.0041                                         max =         19

                                                F(2,2)            =          .
corr(u_i, Xb)  = -0.9174                        Prob > F          =          .

                                 (Std. Err. adjusted for 3 clusters in permno)
------------------------------------------------------------------------------
             |               Robust
       f6ret |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
          PF |  -.1052988   .1717658    -0.61   0.602    -.8443473    .6337497
      POSTPF |          0  (omitted)
       POSTT |          0  (omitted)
         PFT |  -.5526351   .7961855    -0.69   0.559    -3.978345    2.873075
     POSTPFT |          0  (omitted)
             |
        year |
       1991  |   .2083706    .236178     0.88   0.471    -.8078211    1.224562
       1992  |   .2339173   .1588834     1.47   0.279    -.4497026    .9175372
       1997  |  -.1795912   .1875728    -0.96   0.439    -.9866517    .6274692
       1998  |          0  (omitted)
             |
       _cons |   .2445696   .0988045     2.48   0.132    -.1805518    .6696909
-------------+----------------------------------------------------------------
     sigma_u |  .38582298
     sigma_e |  .28449884
         rho |  .64778088   (fraction of variance due to u_i)
------------------------------------------------------------------------------

. xtreg f6ret PF POSTPF POSTT PFT POSTPFT i.year, fe cluster(country)
note: POSTPF omitted because of collinearity
note: POSTT omitted because of collinearity
note: POSTPFT omitted because of collinearity
note: 1997.year omitted because of collinearity

Fixed-effects (within) regression               Number of obs      =        40
Group variable: permno                          Number of groups   =         3

R-sq:  within  = 0.1445                         Obs per group: min =        10
       between = 0.7422                                        avg =      13.3
       overall = 0.0003                                        max =        19

                                                F(2,2)             =         .
corr(u_i, Xb)  = -0.8551                        Prob > F           =         .

                                (Std. Err. adjusted for 3 clusters in country)
------------------------------------------------------------------------------
             |               Robust
       f6ret |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
          PF |  -.1052988   .1717658    -0.61   0.602    -.8443473    .6337497
      POSTPF |          0  (omitted)
       POSTT |          0  (omitted)
         PFT |  -.5526351   .7961855    -0.69   0.559    -3.978345    2.873075
     POSTPFT |          0  (omitted)
             |
        year |
       1991  |   .2083706    .236178     0.88   0.471    -.8078211    1.224562
       1992  |   .2339173   .1588834     1.47   0.279    -.4497026    .9175372
       1997  |          0  (omitted)
       1998  |   .1795912   .1875728     0.96   0.439    -.6274692    .9866517
             |
       _cons |    .195182   .0883056     2.21   0.158    -.1847666    .5751305
-------------+----------------------------------------------------------------
     sigma_u |  .29336188
     sigma_e |  .28449884
         rho |  .51533407   (fraction of variance due to u_i)
------------------------------------------------------------------------------

Perhaps your data excerpt does not include the critical issues.
That said, to avoid the problem, you may try to create a new panel_id, that takes into account both -permno- and -country-:

egen new-panel_id=group(permno country)

[Edward Wu]

Sure thanks!