Estimation of Firm Exit Rate

[Joshua Smith]

Dear All,

I need to estimate firm exit from the market for panel data from India. I have 582 firms that I track for 10 years. Some of them exited the market within these 10 years. I would like to estimate how the probability of exit varies according to some firm specific information, for example, size, revenues, sales, etc...
I thought running a probit model, but Stata runs into an error since when the indicator for firm exit is equal to 1 (firm exited the market) all the covariates are no longer observed, then missing.
I would like to try a Cox regression, but I have a question: Could I use time varying covariates in a Cox model? Size, revenues and sales are observed annually and change each year. I already set up the data in a st format and use the command stdescribe or strum to get to know my data.

Thank you very much.

Regards,

Josh

[Carlo Lazzaro]

Joshua:
welcome to this forum.
As you might have already read from FAQ, you question should have been adressed to the General forum.
Moreover, as per FAQ #12, please report exactly what you typed and what Stata gave you back, as this approach is likely to increase your chance of getting helpful replies.
That said:
- why do not considering a -xtlogit- panel data regression insofar your dependent variable seems a categorical one?
- as far as your question about time-varying covariates in Cox regression is concerned, you're probably looking for -tvc- option, which is covered under -stcox- entry in Stata 13.1 .pdf manual, page 122.

[William Lisowski]

Joshua - You might find this recent discussion in the General forum addressing an issue similar to your problem with the probit model to be helpful.

[Joshua Smith]

Dear All,

Thank you for your kind replies and sorry if I wasn't precise enough in my first post.

I have a dummy, entry, that equals 1 if a firm exits the market in year i, and 0 otherwise. This is my dependent variable. My independent variables are firm revenue, sales, and size. I tried to run a Probit model, but the problem is that my covariates are 0 if a firm exits the market, though I got the following error message:

probit exit sales revenues size, no constant
outcome sales=0 predicts data perfectly

This makes me think that probit model might not be the right model for my data, since I observe a value for the covariates conditional on having a dep. var. equal to 1. From this respect a Cox regression might be more appropriate. However, I'm not sure whether a full panel of data could be fitted by this model. I tried to read the Stata guide but I'm not entirely sure if this is the case.

Thank you again!

Josh

[Rich Goldstein]

please read the FAQ and then re-post in the "general" forum; that is the correct forum for questions about either Stata or statistics; this forum is for questions about how to use the forums

[William Lisowski]

Your problem is not related to whether probit is or is not the right model for your data. Following the discussion linked to in #3 above suggests that for a probit model, your dependent variable should be exit in year t+1 as a function of sales, revenue, and size in year t. Or, equivalently, exit in year t as a function of sales, revenue, and size in the previous year.