Checking parallel trends

[Giovanni Coppola]

Hi everyone, I need your precious help to check parallel trends assumption. I am studying the impact of a directive on cash holdings of firms. I tried to generate the parallel trends graph but my Professor is not convinced about the results. I launched the following commands on Stata, with NFRD which is 0 when the directive does not apply for a company and 1 when the directive applies for a company; dummyT which is an indicator for whether or not the Directive has been established:


egen sumtreat = sum(NFRD), by(CompanyName)

gen Treated = sumtreat > 0

generate InteractionDID = dummyT*NFRD

drop sumtreat




preserve

collapse (mean) CashandEquivalents, by(Treated YEAR)

reshape wide CashandEquivalents, i(YEAR) j(Treated)

graph twoway connect CashandEquivalents* YEAR

restore




could you please tell me if the procedure is the correct one?

Thanks a lot for the responses!

[George Ford]

ssc install lgraph
lgraph CashandEquivalents year, by(Treated)

[Giovanni Coppola]

Thanks George,
the graph that I obtain is the same that I obtain using my formulas, so I hope they are right. Is there another possibility of checking trend by using a regression? Or how can I do an event study?

[George Ford]

prior to treatment, you can regress ln(outcome) on year with interaction treated*year. see if growth rates differ by much. growth would need to be relatively stable and linear'ish.

[Giovanni Coppola]

Hi George,
just so understand, I should create a variable called lnCash which is the ln(CashandEquivalents) and then launch xtreg ln(CashandEquivalents) YEAR Interaction if the Year is < then the first year of treatment?
By the way the treatment year is different in some countries, so I should just check the year before any treatment right?

Thanks for the help!

[George Ford]

you just need to compute a growth rate and the differential for the treated units in the pre-treatment period.

ln(y) = a0 + a1*year + a2*treat*year

could use fixed effects to center.