Hello everyone,
I'm working on a difference-in-differences analysis in Stata 18.5 to estimate the impact of a Russian invasion of Ukraine on firms in the neighbouring country - Latvia. Specifically, as a dependent variable, I am looking to profit growth, turnover growth and employment growth. My whole period of analysis is 2018-2023. I divide firms into 2 groups: ever Russian-affiliated and non-affiliated firms. In total I have 7330 firms in the control group and only 230 firms in the treatment group, after I drop all those firm which have 0 at any time in any of three outcome variables.
My goal is to match treated firms (ru_ever1==1) with control firms (ru_ever1==0) based on pre-treatment profit, turnover, employment and sector ensuring parallel trends before the shock year. However, even after matching, my treatment and control groups still show noticeably different pre-treatment trends. I have already tried different ways of matching using psmatch2. I looked to average values over all period before shock, then I tried to look on yearly level as well. I also included or excluded profit, turnover and employment (as a level) and left only growth changes and other way around.
This is my data and the code I tried as last, however, I got the error message option exact() not allowed. If I try the code without exact() command, my results have no parallel trend.
Could you suggest, what am I doing wrong or what better alternatives I could use to try to get results.
Thank you in advance!
I'm working on a difference-in-differences analysis in Stata 18.5 to estimate the impact of a Russian invasion of Ukraine on firms in the neighbouring country - Latvia. Specifically, as a dependent variable, I am looking to profit growth, turnover growth and employment growth. My whole period of analysis is 2018-2023. I divide firms into 2 groups: ever Russian-affiliated and non-affiliated firms. In total I have 7330 firms in the control group and only 230 firms in the treatment group, after I drop all those firm which have 0 at any time in any of three outcome variables.
My goal is to match treated firms (ru_ever1==1) with control firms (ru_ever1==0) based on pre-treatment profit, turnover, employment and sector ensuring parallel trends before the shock year. However, even after matching, my treatment and control groups still show noticeably different pre-treatment trends. I have already tried different ways of matching using psmatch2. I looked to average values over all period before shock, then I tried to look on yearly level as well. I also included or excluded profit, turnover and employment (as a level) and left only growth changes and other way around.
This is my data and the code I tried as last, however, I got the error message option exact() not allowed. If I try the code without exact() command, my results have no parallel trend.
Code:
clear
input float firmID int year long(sector_id profit) double turnover int employment float(profit_growth turn_growth employment_growth ru_ever1)
1 2018 11 565777 2918703 11 . . . 0
1 2019 11 540931 2854582 11 -24846 -64121 0 0
1 2020 11 436467 2681479 12 -104464 -173103 1 0
1 2021 11 545892 2875439 15 109425 193960 3 0
1 2022 11 561049 3350082 15 15157 474643 0 0
1 2023 11 391275 3164302 16 -169774 -185780 1 0
2 2018 19 52204 623623 3 . . . 0
2 2019 19 66982 980624 3 14778 357001 0 0
2 2020 19 14825 442861 3 -52157 -537763 0 0
2 2021 19 47823 1147230 4 32998 704369 1 0
2 2022 19 66297 1413847 4 18474 266617 0 0
2 2023 19 11627 1080578 4 -54670 -333269 0 0
21 2018 19 165780 1595618 3 . . . 1
21 2019 19 8885 1637555 9 -156895 41937 6 1
21 2020 19 -306173 500182 6 -315058 -1137373 -3 1
21 2021 19 64207 257900 2 370380 -242282 -4 1
21 2022 19 60710 27296 2 -3497 -230604 0 1
21 2023 19 0 0 0 -60710 -27296 -2 1
75 2018 11 -43965 2532136 51 . . . 1
75 2019 11 17920 2966754 48 61885 434618 -3 1
75 2020 11 2820 2583578 45 -15100 -383176 -3 1
75 2021 11 356060 4336728 51 353240 1753150 6 1
75 2022 11 -38350 3245693 38 -394410 -1091035 -13 1
75 2023 11 -364821 1603637 38 -326471 -1642056 0 1
preserve
drop if year > 2021
keep if sample_1 == 1
collapse (mean) employment profit turnover employment_growth profit_growth ///
turn_growth (first) ru_ever1 (first) sector_id, by(firmID year)
psmatch2 ru_ever1 employment profit turnover, neighbor(1) noreplacement logit ate
save "psm2_weights.dta", replace
restore
psmatch2 ru_ever1 employment profit turnover, outcome(profit_growth) neighbor(1) noreplacement logit ate
sort firmID year
preserve
keep if !missing(_weight)
collapse (mean) profit_growth, by(year ru_ever1)
twoway ///
(line profit_growth year if ru_ever1==1, sort lcolor(blue) lwidth(medium)) ///
(line profit_growth year if ru_ever1==0, sort lcolor(red) lwidth(medium)), ///
xline(2022, lpattern(dash) lcolor(black) lwidth(thin)) ///
legend(order(1 "Treatment" 2 "Control")) ///
ytitle("Average Profit Growth") ///
xtitle("Year") ///
title("Trend in Profit Growth for Matched Firms (Shock=2022)")
restore
Thank you in advance!
Comment