Create comparison between firm-years

[Max Dost]

Hi all,

I am trying to create a regression model, but fail to find the right way to modify the data so i can perform my regression correctly.
I have a database of 36.000 firm-year observations, with approximately 10 years of observations per firm. The names with which the firm-years are identified is (FYEAR) and the names for the specific companies are defined by the variable (COMPANYCODE1).

The database contains a binary variable (ICMW) which only presents a 1 for certain firms and firm-years.

The 2 problems i am stuck with are as follows:

1. I want to perform a regression which compares (per company) the firm-years before the variable ICMW presents a 1 and the firm-years after.
But i can't seem to find the code for letting Stata perform a comparison between the yearly data of a company, with the ICMW-variable as border.

2. I only want to perform the regression for firms that actually contain a 1 for ICMW in one or more years. But Stata keeps dropping observations of all yearly-observations without a 1 for ICMW, while i only want to drop all companies (so all the yearly observations of companies) without a 1 for the ICMW. For this problem, i have been jumbling with panel data, but i haven't found the right solution yet.

I would want to provide you with the code for this problem, but seeing as i haven't been able to find it, i can't post it either.

I hope i provided enough information for you guys to understand the problem! If anyone knows a solution to my problem, it would be much appreciated!

Kind regards

Max

[Sergiy Radyakin]

To the extent that I understand your task:

sort idfirm year
by idfirm: egen suitable=total(icmw)
generate byte ba=.
by idfirm : replace ba=cond(_n==1,icmw,ba[_n-1]+icmw) if suitable
regress y x icmw ba

I assumed that

1. icmw event occurs no more than once for each firm in the observation window
2. that icmw takes value of 0 everywhere, where it is not 1.
3. The period where icmw occurs is joined with the periods after the event.

I felt the description is not precise for the case of multiple events. You may want to clarify that and modify the above code.
I am equally not sure if the window may be only a single year or if the event may occur in the first period of observation, in which case the before group is zero-sized.

[Max Dost]

I tried and modified a bit on the code you provided, it works now! Thank you Sergiy. Now i can move on with my research!