Dynamic Panel Data with multiple markets and a different benchmark firm per market

[Louis Belisle]

Hi,
I am using Stata 13 on Windows 10.
I have an unbalanced dynamic panel dataset of 90,000 firms operating in 50,000 markets (some in multiple markets, some in just one) across 13 years in which I want to use one firm per market as the benchmark for that market.
An example of the data for 3 firms in 2 markets over 3 years is as follows:

id	Market	Year	Production
1	1	1998	6.5
1	1	1999	7.3
1	1	2000	3.6
1	2	1999	4.8
1	2	2000	6.8
2	1	1998	1.2
2	1	1999	1.6
2	2	1998	4.4
2	2	1999	4.6
2	2	2000	5.8
3	1	2000	7.6

I want to be able to subtract the Production value of the benchmark firm from the other firms' Production in the corresponding market and year. My benchmark should be a firm that is active throughout the period. In the example above, firm 1 would be the benchmark in Market 1, and firm 2 the benchmark in Market 2. Therefore, I want to create an extra variable Benchmark as such

id	Market	Year	Production	Benchmark
1	1	1998	6.5	6.5
1	1	1999	7.3	7.3
1	1	2000	3.6	3.6
1	2	1999	4.8	4.6
1	2	2000	6.8	5.8
2	1	1998	1.2	6.5
2	1	1999	1.6	7.3
2	2	1998	4.4	4.4
2	2	1999	4.6	4.6
2	2	2000	5.8	5.8
3	1	2000	7.6	3.6

What I have at the moment is a variable to identify which firm should be my benchmark. I am missing the part where I create the Benchmark with the appropriate values.

/*Generate panel structure*/
egen panelid = group(id market), label
xtset panelid year

/*Create indicator for firm activity*/
egen prod_duration = count(production), by(panelid)
gen negprod_duration = -prod_duration
sort market year negdepo_duration panelid
gen prod_benchmark = production if prod_duration == 3
sort year id negprod_duration

/*Assign Benchmark production value to corresponding firm within market-year*/
bysort year (market negprod_benchmark panelid) : replace prod_benchmark = prod_benchmark[1] // This is what doesn't work

What this code does is it puts the value of only one firm across these years for all markets. Like this:

id	Market	Year	Production	Benchmark
1	1	1998	6.5	6.5
1	1	1999	7.3	7.3
1	1	2000	3.6	3.6
1	2	1999	4.8	7.3
1	2	2000	6.8	3.6
2	1	1998	1.2	6.5
2	1	1999	1.6	7.3
2	2	1998	4.4	6.5
2	2	1999	4.6	7.3
2	2	2000	5.8	3.6
3	1	2000	7.6	3.6

Which is not what I want.

My goal is to then use xtabond2 to proceed to a DPD analysis. I want to define a benchmark for the variable Production, as well as for 3 other variables in the same way.

How would I generate such variable?

Thanks a lot for your help.

Cheers,

[Louis Belisle]

A potential solution I found:

egen numt_prod = sum(production>0 & production~= .), by(id market)
sort market numt_prod
gen tmp = 0
replace tmp = 1 if market~=market[_n-1]
egen longestdum = sum(tmp), by(id market)
/*To populate benchmark's values in others*/
egen production_base_mkt = sum(production*longestdum), by(market year)

Another possible one

/*Create indicator for firm activity*/
egen prod_duration = count(production), by(panelid)
egen max_duration = max(prod_duration), by(panelid)
gen tmp = production if prod_duration == max_duration gen
rand = runiform()
sort panelid year tmp rand
replace tmp = . if !(_n == 1 | panelid[_n] != panelid[_n-1] | year[_n] != year[_n-1])
egen prod_benchmark = max(tmp), by(panelid year)
drop tmp rand