Staggered Dif-in-Dif and Post periods

[Mahtab Karimi]

Hi all,

I have a question about a staggered DiD and would appreciate your help:

I want to see the effect of an industry shock on firms in the industry. The shock happens in different years for different industries so it is a staggered structure. I am interested in changes in a specific variable after the shock for the year shock happens and the following year.
My question is: what should I do with the years after these two years?
I have a few options
First: I can treat them like zero years for treatment firms:

gen Dit=1 if T==1 & fyear>=shock & fyear<shock+2
replace Dit=0 if missing(Dit) 

reg Y Dit  i.fyear i.x2digitsic i.cik, vce(cluster cik)

Linear regression                               Number of obs     =      1,149
                                                F(13, 115)        =          .
                                                Prob > F          =          .
                                                R-squared         =     0.8933
                                                Root MSE          =     .09168

                                  (Std. err. adjusted for 116 clusters in cik)
------------------------------------------------------------------------------
             |               Robust
Y | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
         Dit |   .0299991    .012173     2.46   0.015     .0058867    .0541115

Second, I can include them in the model with another indicator:

gen Dit=1 if T==1 & fyear>=shock & fyear<shock+2
replace Dit=0 if missing(Dit)  

gen D_Af=1 if T==1 &  fyear>=shock+2
replace D_Af=0 if missing(D_Af)


reg Y Dit D_Af  i.fyear i.x2digitsic i.cik, vce(cluster cik)

Linear regression                               Number of obs     =      1,149
                                                F(14, 115)        =          .
                                                Prob > F          =          .
                                                R-squared         =     0.8935
                                                Root MSE          =     .09163

                                  (Std. err. adjusted for 116 clusters in cik)
------------------------------------------------------------------------------
             |               Robust
Y | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
         Dit |   .0220018   .0180591     1.22   0.226    -.0137698    .0577735
        D_Af |  -.0175593   .0262822    -0.67   0.505    -.0696193    .0345007

Third, I can exclude them from the model:

gen Dit=1 if T==1 & fyear>=shock & fyear<shock+2
replace Dit=0 if missing(Dit)  
replace Dit=. if T==1 & fyear>shock+2


reg Y Dit i.fyear i.x2digitsic i.cik, vce(cluster cik)


Linear regression                               Number of obs     =        985
                                                F(13, 115)        =          .
                                                Prob > F          =          .
                                                R-squared         =     0.8969
                                                Root MSE          =     .09096

                                  (Std. err. adjusted for 116 clusters in cik)
------------------------------------------------------------------------------
             |               Robust
Y | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
         Dit |   .0216025   .0142029     1.52   0.131    -.0065307    .0497357

Obviously, I prefer the first approach because this is just an industry shock and it is not a permanent chagne in the firm. But, I am not sure if treating the periods after two like before treatment is acceptable or not.

Thank you for your help,
Mahtab