---------------------------------------------------------------------------------------------------------------
       log:  /Users/jwalker/Work/ps2.log
  log type:  text
 opened on:  10 Dec 2007, 15:04:08

. infile caseid year firm grp y using vctest.dat
(15000 observations read)

. compress
caseid was float now int
year was float now byte
firm was float now byte
grp was float now int

. save vctest , replace
file vctest.dta saved

. 
. sum y, detail

                              y
-------------------------------------------------------------
      Percentiles      Smallest
 1%    -3.801925       -6.54169
 5%     -2.68644      -6.328892
10%    -2.109073      -6.068696       Obs               15000
25%    -1.104151      -5.947643       Sum of Wgt.       15000

50%     .0727825                      Mean           .0755198
                        Largest       Std. Dev.      1.715693
75%     1.246118       5.910426
90%     2.312204       5.978642       Variance       2.943602
95%     2.908397       5.980018       Skewness       .0269152
99%     4.039303       6.810177       Kurtosis       2.908424

. *Record variance of y for future use.
. scalar define var_y = r(Var)

. 
. *Now recover estimate of the variance of the transitory component 
. egen mg = mean(y), by(grp)

. gen dg = y - mg

. 
. sum dg, detail

                             dg
-------------------------------------------------------------
      Percentiles      Smallest
 1%    -2.111425      -3.203313
 5%    -1.493781      -3.138394
10%    -1.169652      -3.092287       Obs               15000
25%    -.6004642      -2.986183       Sum of Wgt.       15000

50%     .0014804                      Mean           2.38e-10
                        Largest       Std. Dev.      .9011704
75%     .6005392       3.282832
90%     1.143927       3.308813       Variance       .8121081
95%     1.484322       3.318286       Skewness       .0054814
99%     2.099428       3.594743       Kurtosis        3.00173

. 
. *variance of dg equals Variance of transitory component, epsilon
. *store this estimate for subsequent use
. scalar define var_eps = r(Var)

. 
. * collapse data file to firm person observations, averaging across 
. * time periods
. 
. collapse (mean) y , by(caseid firm)

. 
. * create cross product terms to form Wooldridge's estimator
. * cross products of different firms recovers variance of 
. * individual component. Notice exclude own product
. gen y1 = y[_n-1] if caseid == caseid[_n-1]
(1000 missing values generated)

. gen y2 = y[_n-2] if caseid == caseid[_n-2]
(2000 missing values generated)

. gen a = y*y1
(1000 missing values generated)

. gen b = y*y2
(2000 missing values generated)

. 
. *sum across periods to gather cross products then sum across
. *individuals.  missing values treated as zeros.
. *
. egen ssq = rowtotal(a b)

. sum ssq , detail

                             ssq
-------------------------------------------------------------
      Percentiles      Smallest
 1%    -3.585646      -10.08237
 5%    -1.432561      -9.334759
10%    -.6508919      -8.130392       Obs                3000
25%            0      -7.778816       Sum of Wgt.        3000

50%            0                      Mean           .9432132
                        Largest       Std. Dev.       2.71497
75%     1.136436       19.60847
90%     3.903668       21.24195       Variance       7.371063
95%     6.200994       22.05335       Skewness       2.783689
99%     11.53966       25.73206       Kurtosis       16.14628

. scalar define var_alpha = r(sum)/r(N)

. 
. *Variance of the match specific component is the residual
. scalar define var_mu = var_y - var_alpha - var_eps

. 
. scalar list _all
    var_mu =  1.1882802
 var_alpha =  .94321324
   var_eps =  .81210813
     var_y =  2.9436016

. 
. 
end of do-file

. exit
no; data in memory would be lost
r(4);

. exit, clear