st: Multinomial regression using panel data with case-specific regressors only

[Rohit Sonika <net@rsonika.me>]

Hi Statalist,

I have a cross-sectional, time series dataset which has information on
over 500 companies repeated over 10 years from 1991 to 2000. During
the time period, firms have a choice of decisions to make:

1. Stay in their state of paying no dividends or buybacks
2. Pay dividend, or
3. Buyback shares

Once a decision is made, it tends to follow the trend for the rest of
the sample period. For instance, if firm A decides to pay dividend in
1994 then the decision continues to occur through 2000. The equation I
am trying to model is the likelihood of decision 2 or 3 occurring with
decision 1 as the base, regressed on a set of firm-level
characteristic variables.

Question 1: I understand the decision a firm has to make is
mutinomial, but unordered. However, the regressors I am using are
case-specific, and not alternative-specific. I have gone through the
statalist archive to understand that models like that of -gllamm- and
-mixlogit- require alternative-specific regressors in each decision
group, but in my case none exist. Can you suggest how best can I
tackle this multinomial issue with a panel data?

Question 2: Would it make any sense to subsample the data and do
separate -xtlogit- regressions in the two subsamples, comparing 2 with
1, and 3 with 1?

I am trying to do this in Stata 12. Thank you for all your help.

Best,
Rohit
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/