High p-value after xtdidgregress

[Hwanseong Pak]

Hello!

I have panel data, and I conducted difference in differences to evaluate the effect of a policy on innovation (Patent Count and R&D Expenses).
However, after all the commands, I have a high p-value.
Could someone please look at my codes and see what could cause the high p-values?
I appreciate the help in advance.

Code:

***Import Excel File
import excel "C:\Users\hwanseong.pak\Desktop\STATA Data_v0.2.xlsx", sheet("STATA") cellrange(A1:M3137) firstrow
destring Ticker, replace force

***Set Panel Data
xtset Ticker Year

***Summary Statistics
summarize

***Create DiD Variables
gen post2=1 if Year >= 2018
replace post2=0 if Year < 2018
gen phase2_treated = post2 * Phase2

gen post1=1 if Year <=2017
replace post1=0 if Year > 2017
gen phase1_treated = post1 * Phase1

replace PatentCount = PatentCount+1
gen f_lnPC=ln(f.PatentCount)

replace RDExp = RDExp+1
gen lnRDE = ln(RDExp)

***Lag Control Variables
gen l_lnTA = ln(l.TotalAssets)
gen l_lnTR = ln(l.TotalRevenue)
gen l_HHI = l.HHI
gen l_ROA = l.ROA
gen l_DE = l.DE
gen l_RG = l.RevenueGrowth

drop if Year == 2015
drop if Year == 2022

***Run Fixed Effects DiD Regressions
xtdidregress (f_lnPC phase1_treated l_lnTA l_lnTR l_ROA l_DE l_RG l_HHI Industry) (phase2_treated), group(Ticker) time(Year) aequations
xtdidregress (lnRDE phase1_treated l_lnTA l_lnTR l_ROA l_DE l_RG l_HHI Industry) (phase2_treated), group(Ticker) time(Year) aequations

[Maxence Morlet]

Maybe the treatment simply does not have a statistically significant effect. You should however not only focus on p-values, but also on economic significance, confidence intervals, etc.

Also, instead of taking the log of the dependent variable and adding one, in line with Santos Silva and Tenreyro (2006), you should use PPML (ssc install ppmlhdfe)

[Maxence Morlet]

I am however confused by your two phase treatments leading to two treatment variables. Is your treatment staggered? If so you should employ xthdidregress

[Hwanseong Pak]

Thank you for the insights Maxence Morlet .
I will try the PPML method.
My treament is somewhat staggered as the policy includes a few more companies in the 2nd phase.
However, I wanted to evaluate the effect of the 2nd phase only, controlling the 1rst phase of the policy.
Is this approach in appropriate, and should I employ xthdidregress instead?
Again, thank you for the help in advance.

[Maxence Morlet]

I don't understand this "evaluate the effect of the 2nd phase only, controlling the 1rst phase of the policy.". From what I gather, this is staggered treatment adoption, so a staggered DiD.

[Hwanseong Pak]

Dear Mr. Maxence Morlet,

Thank you for the continued support.
I tried the xthdidregress method, but my results are not statistically significant.
I assume the prevalence of 0s in my dependent variable is causing this.
Thus, I decided to use the ppmlhdfe method.
However, the output from this ppmlhdfe method is similar to a classic regression model.
As I have mentioned earlier, I want to know the effect of the 2nd phase only, which is why you suggested I use staggered DiD.
Is there any method of knowing the effect of the 2nd phase only using ppmlhdfe method?

Kind regards,
Pak