[Taiba Chau]

I do not really have any control variables currently. But within the dataset there is an extra variable that could be used as a control. The control variable is the dummy for size of firm, either big or not.

[Taiba Chau]

Thanks for the help andrew. I just mentioned that I could add a control variable which is the size of the firm.
Could I then run:
regress movement i.date##i.treated size,cluster(firm) Also I was wondering since this is a balanced panel, would you give any tips in creating a graph to indicate parallel trends for the treated and non-treated firms. I graphed the following: lgraph movement_ch date, by(treated) I was wondering if there is any other graph? I have read some information on a event-study figures Thank you!

[Andrew Musau]

regress movement i.date##i.treated size,cluster(firm)

Yes. If you are able to find more control variables, the better.

Also I was wondering since this is a balanced panel, would you give any tips in creating a graph to indicate parallel trends for the treated and non-treated firms. I graphed the following: lgraph movement_ch date, by(treated) I was wondering if there is any other graph? I have read some information on a event-study figures Thank you!

I will repeat one more time that this is not DID. There is no notion of parallel trends as you do not have a pre-period. With parallel trends, you want to show that the treated and control units were observationally equivalent in the pre-period, but none exists in your case.

[Taiba Chau]

Thank you for clarifying that up. So the graph that I did manage to create what could I infer from that? Also what is coefplot doing then? DiD won't work here, so the parallel trend assumption cannot hold in this case. Is there a way of describing this through a diagram as well?

[Andrew Musau]

I would not waste my time trying to interpret the graphs. Now you know that it's not DID, move on from parallel-trends graphs.

[Taiba Chau]

Okay. Thanks for the help! I really do appreciate it. What would you advise me to do next? Because this is balanced firm-level data. Should I plot some graphs to indicate the trend across the 96 days for the outcome variable? Run any additional regressions?

Something I have also come across is a state-specific time trend. How would that relate?

[Andrew Musau]

I would recommend that you only draw graphs if there is a need for them. I do not see a compelling reason in this instance.