Difference in Difference with etherogenous treatment over time (xtreg vs xthdidregress)

[Beatrice Aiello]

Hi all,
I recently created a panel dataset covering companies (id) over 11 years (YEAR). I would like to evaluate what is the effect of implementing a given policy at company level on the number of negative news about the company (NegObs). Policy is introduced in different years in different companies. So far I've created two variables: 1. HRP_anyyear that distinguish the treatment and control group as it takes value 1 for all the years of observation over the 11 years if the company has introduced the policy, 0 otherwise; 2. HRP treatment that takes value 1 from the year of policy introduction and subsequent years for the companies that introduced a policy, 0 otherwise. NegObs counts the number of negative news the company has been involved in the given year.

input long id int YEAR float(HRP_treatment HRP_anyyear) byte NegObs

So far I've tried two model of dif in dif and I try to explain my doubts.

1. xtreg NegObs i.HRP_anyyear##i.HRP_treatment covariates, fe. In this case my doubt is: it is correct to have the post treatment variable specified only for the treatment group and not for the control group? The results I got so far omit the interaction terms due to collinearity.
2. As my treatment is eterogenous over years and cohort I've also tried to use: xthdidregress ra (NegObs covariates ) (HRP_treatment), group (id) time (YEAR). For this model I wonder if it is correct to use the ra option or should I use the tfwe estimator, in the latter case I face a memory problem unless I specify the option hettype(time)/hettype(cohort)

Do you have any suggestion on how to best design a dif in dif model with this type of dataset? I've also considered matching the control and treatment group before running model 1 to get more reliable results, but read some cons about matching for dif in dif models.

* Example generated by -dataex-. For more info, type help dataex
clear
input long id int YEAR float(HRP_treatment HRP_anyyear) byte NegObs
 1 2013 0 1  0
 1 2014 0 1  0
 1 2015 0 1  0
 1 2016 0 1  0
 1 2017 0 1  0
 1 2018 0 1  0
 1 2019 0 1  0
 1 2020 0 1  2
 1 2021 0 1  0
 1 2022 0 1  2
 1 2023 1 1  5
 2 2013 0 1  4
 2 2014 0 1  0
 2 2015 0 1  4
 2 2016 0 1  0
 2 2017 0 1  2
 2 2018 0 1  0
 2 2019 0 1  4
 2 2020 0 1  7
 2 2021 1 1  4
 2 2022 1 1  6
 2 2023 1 1  3
 3 2013 0 0  0
 3 2014 0 0  0
 3 2015 0 0  0
 3 2016 0 0  0
 3 2017 0 0  0
 3 2018 0 0  0
 3 2019 0 0  0
 3 2020 0 0  2
 3 2021 0 0  1
 3 2022 0 0  2
 3 2023 0 0  0
 4 2013 0 1  1
 4 2014 0 1  0
 4 2015 0 1  0
 4 2016 0 1  0
 4 2017 0 1  0
 4 2018 0 1  0
 4 2019 0 1  0
 4 2020 0 1  2
 4 2021 0 1  7
 4 2022 1 1  0
 4 2023 1 1  0
 5 2013 0 1  0
 5 2014 0 1  0
 5 2015 0 1  0
 5 2016 0 1  3
 5 2017 0 1  2
 5 2018 0 1  2
 5 2019 0 1  5
 5 2020 0 1 11
 5 2021 1 1  7
 5 2022 1 1  3
 5 2023 1 1  5
 6 2013 0 1  4
 6 2014 0 1  1
 6 2015 1 1  4
 6 2016 1 1  0
 6 2017 1 1  1
 6 2018 1 1  2
 6 2019 1 1  0
 6 2020 1 1  2
 6 2021 1 1  0
 6 2022 1 1  3
 6 2023 1 1  0
 7 2013 0 1  7
 7 2014 0 1  6
 7 2015 0 1  9
 7 2016 0 1  8
 7 2017 0 1 14
 7 2018 0 1  4
 7 2019 0 1  8
 7 2020 0 1 15
 7 2021 0 1 14
 7 2022 1 1 23
 7 2023 1 1 16
 8 2013 0 0  1
 8 2014 0 0  0
 8 2015 0 0  0
 8 2016 0 0  0
 8 2017 0 0  2
 8 2018 0 0  1
 8 2019 0 0  7
 8 2020 0 0  7
 8 2021 0 0  6
 8 2022 0 0  3
 8 2023 0 0  6
 9 2013 0 1  2
 9 2014 0 1  3
 9 2015 0 1  2
 9 2016 0 1  0
 9 2017 0 1  0
 9 2018 0 1  0
 9 2019 0 1  0
 9 2020 0 1  2
 9 2021 1 1  1
 9 2022 1 1  4
 9 2023 1 1  0
10 2013 0 1  2
end
label values id id
label def id 1 "AEON", modify
label def id 2 "ALDI", modify
label def id 3 "ANTA", modify
label def id 4 "ASICS", modify
label def id 5 "ASOS", modify
label def id 6 "Abercrombie & Fitch Co", modify
label def id 7 "Adidas", modify
label def id 8 "Amazon", modify
label def id 9 "American Eagle Outfitters", modify
label def id 10 "Armani", modify
label var id "Name" 
label var YEAR "YEAR" 
label var NegObs "NegObs"

Thanks in advance
B