ivregress looped on individuals of panel data

[Julia Martinsen]

Hi all,

I have a panel data and I want to run an instrumental variable regression on each individual. The code I am using is the following:

ivregress 2sls y (x = z) if person_id==7, vce(hac bartlett 12)

But I get the following error:

sample may not include multiple panels

Shouldn't the "if person_id==7" solve the issue?
How can I solve this?

Alternatively, do you know a code that can do the trick, i.e. run an IV regression on each individual of a panel, with HAC s.e.?

Thank you!

[Julia Martinsen]

It works with vce(robust) but not with vce(hac bartlett 12)

[Andrew Musau]

vce(hac bartlett 12) is for use with time series and not panel data. Even if you use the -if- qualifier, you have already xtset your data and declared it a panel to Stata. That is why you get the error. One solution is to use -preserve- and -restore- repeatedly.

  
preserve
keep if person_id==7
ivregress 2sls y (x = z), vce(hac bartlett 12)
restore  
preserve
keep if person_id==8
ivregress 2sls y (x = z), vce(hac bartlett 12)
restore

Alternatively, if your panel is balanced, you can tsset your data using the observation number and use the -if- qualifier

xtset, clear
sort id year
gen t= _n
tsset t

ivregress 2sls y (x = z) if person_id==7, vce(hac bartlett 12)

[Julia Martinsen]

Hey Andrew,

Nice work arounds!

I instead used ivreg2 with if and HAC s.e. - it should give the same results, right?

[Andrew Musau]

Absolutely the same. Remember HAC= Heteroskedasticity and Autocorrelation Consistent, so remember to include robust option in -ivreg2- (otherwise you have AC covariance estimation).

. clear

. webuse grunfeld

. keep if company==1
(180 observations deleted)

. tsset year
        time variable:  year, 1935 to 1954
                delta:  1 year

. ivregress 2sls mvalue invest kstock, vce(hac nw 5)

Instrumental variables (2SLS) regression          Number of obs   =         20
                                                  Wald chi2(2)    =     164.64
                                                  Prob > chi2     =     0.0000
                                                  R-squared       =     0.6396
                                                  Root MSE        =     529.14

------------------------------------------------------------------------------
             |                 HAC
      mvalue |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      invest |   4.664183   .4846287     9.62   0.000     3.714329    5.614038
      kstock |  -1.456692   .2845878    -5.12   0.000    -2.014474   -.8989102
       _cons |   2442.498   180.3164    13.55   0.000     2089.085    2795.912
------------------------------------------------------------------------------
(no endogenous regressors)
HAC VCE:       Bartlett kernel with 5 lags

. ivreg2 mvalue invest kstock, bw(6) kernel(bartlett) robust

OLS estimation
--------------

Estimates efficient for homoskedasticity only
Statistics robust to heteroskedasticity and autocorrelation
  kernel=Bartlett; bandwidth=6
  time variable (t):  year

                                                      Number of obs =       20
                                                      F(  2,    17) =    69.97
                                                      Prob > F      =   0.0000
Total (centered) SS     =  15537575.83                Centered R2   =   0.6396
Total (uncentered) SS   =  391181828.9                Uncentered R2 =   0.9857
Residual SS             =  5599712.818                Root MSE      =    529.1

------------------------------------------------------------------------------
             |               Robust
      mvalue |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      invest |   4.664183   .4846287     9.62   0.000     3.714329    5.614038
      kstock |  -1.456692   .2845878    -5.12   0.000    -2.014474   -.8989102
       _cons |   2442.498   180.3164    13.55   0.000     2089.085    2795.912
------------------------------------------------------------------------------
Included instruments: invest kstock
------------------------------------------------------------------------------

[Julia Martinsen]

Thank you!!

(yes I type both bw and robust)