Bootstrap standard errors

[Dario Maimone Ansaldo Patti]

Dear All,

I created the following program to bootstrap standard errors:

cap program drop myboot
program define myboot, eclass
preserve

xtreg lnfdipccurrd lic Lgwtpol_corr legor_uk legor_fr legor_sc legor_ge Llnpoliticaldf Llntradedf Llntradedj Llnpop Lgrowthpwt Llneys i.time, r
predict f, xb

gen period = 5*floor(time/5)

collapse pcgdp gcf f ndg pcgdpin, by(iso3 period)
egen time = group(period)
egen id = group(iso3)
tsset id time

gen lnpcgdp = ln(pcgdp)
gen s = ln(gcf)
gen lnndg = ln(ndg)
gen lnf = ln(f)
gen lnpcgdpin = ln(pcgdpin)

xtreg lnpcgdp s lnndg lnf lnpcgdpin i.time, r 


matrix b=e(b)
ereturn post b
ereturn local cmd="bootstrap"
restore
end

bootstrap _b, seed(12345) reps(500): myboot

According to my understanding (which is clearly wrong, due to the results I get), Stata is supposed to estimate the model in red, then it should collapse the dataset according to the part of the code in green and run the second estimation and let me have my bootstrapped standard errors. After collapsing I should have about 950 observations. Instead, the run of the program produces the following table:

Bootstrap results                                        Number of obs = 4,659
                                                         Replications  =   500

------------------------------------------------------------------------------
             |   Observed   Bootstrap                         Normal-based
             | coefficient  std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
           s |   .0788844   .0161333     4.89   0.000     .0472637    .1105051
       lnndg |  -.0142946   .0036668    -3.90   0.000    -.0214814   -.0071078
         lnf |   .0128075   .0064819     1.98   0.048     .0001033    .0255117
   lnpcgdpin |   .9769231   .0031725   307.94   0.000     .9707051    .9831411
             |
        time |
          1  |          0  (empty)
          2  |   .0056588   .0110584     0.51   0.609    -.0160153    .0273329
          3  |  -.0119457   .0110423    -1.08   0.279    -.0335882    .0096968
          4  |   .0034954   .0119264     0.29   0.769    -.0198799    .0268707
          5  |  -.0176491   .0126874    -1.39   0.164    -.0425159    .0072177
          6  |  -.0337107   .0128047    -2.63   0.008    -.0588076   -.0086139
             |
       _cons |  -.0087348   .0603189    -0.14   0.885    -.1269576    .1094881
------------------------------------------------------------------------------

where the number of observations corresponds to the number of observations before collapsing (pink bold). I am bit lost about how I can achieve what I need. Any suggestion?

Thanks for any help you may provide.

Dario

[George Ford]

why not just vce(bootstrap)?

where is your myboot program?

[Felix Bittmann]

I think your Code is not necessarily wrong. When you use bootstrap in the last line of your code, Stata reports the number of obs in the original dataset. It does not care or know that you use collapse later in the program.

[Dario Maimone Ansaldo Patti]

George Ford Thanks for your reply. The program myboot is in the code I posted above. It starts in the first line. But eventually I opted for what you suggested (vce(bootstrap)).
Felix Bittmann I see. Actually, I do not want to use all the available observations. That is why I abandoned that program and use the suggestion offered by George.

Thanks both for your help.

Dario