Comparing regression coefficients across groups in presence of fixed effects

[Kedar Kelkar]

Hey I have some company data and i run a panel regression with firm fixed effects. I have a list of firm level controls. I run the regressions across 2 groups as given below

 reghdfe pm p_diff l1.pm dd ddXp_diff rnd f_vol ind_vol mkt_conc ind_tang sales_growth net_ppe if sd==1, absorb(co_code)
reghdfe pm p_diff l1.pm dd ddXp_diff rnd f_vol ind_vol mkt_conc ind_tang sales_growth net_ppe if sd==0, absorb(co_code)

Now i am interested in the comparing the "ddXp_diff" coefficient. Hence i did the following after each of the regressions

est store coeff1
est store coeff2
suest coeff1 coeff2

I get the error

unable to generate scores for model coeff2
suest requires that predict allow the score option

How do i use suest in reghdfe?

[Bruce Weaver]

I assume you are referring to this user-written program (please see item 12.1 in the FAQ):

ssc describe reghdfe

I am not familiar with it. Does it support factor variable notation? If so, I would include net_ppe and its interaction with ddXp_diff in the model. The following assumes that net_ppe is categorical with only two possible values (0 and 1).

reghdfe pm p_diff l1.pm dd rnd f_vol ind_vol mkt_conc ind_tang ///
sales_growth net_ppe c.ddXp_diff##i.net_ppe, absorb(co_code)

HTH.

[Andrew Musau]

You should not use suest after reghdfe. See

help reghdfe

for a work-around. A general approach to including interactions, as Bruce does in #2, is to do a joint estimation and subsequently use the test command. The result should be the same but here you can compare multiple coefficients..

gen gr= cond(sd==1,1, cond(sd==0,0,.))
gen co_code1= co_code*(gr==1)
gen co_code0= co_code*(gr==0)

reghdfe pm i.gr#(c.p_diff ...), a(co_code1 co_code0)