Hazard Fuction - Logit estimation

[glmacha]

I have a hazard function like this Hij(t|Wi) = Wi*Ho(t)exp{Xij(t)'B}, that represents a semiparametric duration model. The data set is an unbalanced panel with 10950 anual firms observations. I assume I have to use a LOGIT model to measure the rate of adjustament on firm's capital structure (main issue of the problem). I need a help on how do I have to run this estimation on EViews. Any tips ? Thanks in advance.

[EViews Glenn]

You'll have to be a bit more specific about what your data look like. When you say "unbalanced panel" do you mean that you have binary responses at different durations for firms? And that firms leave this panel when they transition?

If so, then *one* way to proceed is to stack your data as you have, then to run a dummy variable binary data model where the dummies are not over firms, but rather are over periods. If you choose the extreme value binary choice models, you'll get a semi-parametric proportional hazards model. If you choose one of the other binary models, you'll get something that is valid, but different (it's not accelerated failure time, but rather a different model).

To run one of these, just do a standard binary model with no constant and with an @expand on the indicator for the duration associated with the observation in the panel. This will get you the dummy variables for each duration. Note that the @expand in the model is on the duration (which gets you the semiparametric estimate of the baseline hazard function), not on the firms. The latter is not identified.

There's an article in the Journal of Applied Econometrics, 10(4), 1995, 411-431 that discusses this particular approach in some depth.

[glmacha]

Hi QMS Glenn, thanks for the support.
Being more specific about the data: I have a panel data of 15 proxies from 73 companies over 10 years.
Two of this proxies are dummy variables for each companies per year.
I am trying to estimate the rate of adjustment of a variable (in this case the optimal capital structure leve).
My main doubt is: Do eviews support this kind of regression ? If so, I don't have any clue how to run it in terms of the code I need to use ?
I know it's a basic question but I tried to find an answer/tutorial at Eviews Manual but it was not useful...
Thanks in advance.
Regards.

[EViews Glenn]

You'll have to be a bit more specific still about your data. You say you have data on 73 companies over 10 years. From the 10950 obs, it sounds as though it isn't annual data.

What does the data for a specific company look like? Suppose company A is around for 12 periods and company B is around for 5. What does the data look like?

[glmacha]

Sorry for the misunderstanding. I have annual data for the companies.
73 companies x 10 years = 730 annual data x 15 proxies = 10950 total data panel
For the 15 proxies I mean: company's size, leverage, capex, cash, equity return, sales expense, etc.
The 2 dummies variables indicates when a company made a leverage increase ou a leverage decrease.
I don't know if I answered your doubt....