Estimation of Firm Survival

[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'm not sure whether using a Probit model or a Cox regression is more appropriate. Do you have any suggestion?

Thank you very much.

Regards,

Josh

[Stephen Jenkins]

What you can do partly depends on whether you know in which year each firm entered the market (and hence became at risk of exiting the market). Date of market entry is not necessarily the same concept as date first observed in your panel (but might be). In short, more information about your data is required.

[Joshua Smith]

Dear Stephen,

Thank you for your reply. In my data, I don't have left censoring, i.e. I only consider firms that are in the market at time 0. Then, in the following 10 years some firms exit the market and I want to estimate the probability of exiting, given firm characteristics such as size, revenues, sales, etc...

[Stephen Jenkins]

From what you write, you may in fact have left censored or left truncated data. ("I only consider firms that are in the market at time 0"). Let me rephrase things: of the firms that you observe in the first wave of your panel, how many entered the market in exactly that year? Are there firms observed in the first year of your panel that entered the market before the panel? [Be clear(er) in your particular context about the difference between survival time since first at risk of exit from the market, and calendar time. Put differently, when you say you track firms for 10 years, do you track them for 10 years since market entry -- so firms are observed for different calendar year decades -- or you observe a collection of firms over a specific 10 calendar year period?]

[Joshua Smith]

Sure, let me also be clearer. The firms in my sample are firms that received government subsidies. The subsidy is received in different years, so I normalized this year to be zero for all firms. All the firms that received the subsidy were in the market from at least 4 years.

[Maarten Buis]

But more than 4 years could be 5, 6, 10, or 100 years. If you only know that they are older than 4 years at point 0, then you have left censoring (and a rather selective sample).

[Joshua Smith]

I know how old firms are an, exactly, it could be either 5 or 100. However, from my understanding of the survival analysis, I start from time 0 and see whether the risk of exiting the market changes after that time, but I might be wrong

[Maarten Buis]

I don't understand your data: Do you have the date at which they received the subsidy or date at which the firm was founded, or both?

[Joshua Smith]

I have both