10-15-2012 02:15 PM
As a part of my master’s thesis, I am currently undertaking some statistical analysis of corporate turnaround. I have a panel dataset consisting of approximately 200 firms for 6 years each gathered within a 15-year time-period and with large cross-sectional dimensions (various industries and countries).
Based on a set of criterions, I have characterized each firm as either being a turnaround case or non-turnaround case, why I have a binary dependent variable taking the value 1 if the firm succeeds, otherwise 0. This implies that my independent variable does not differ for the individual firm in the 6 observations linked to the given firm. .
Due to unobserved fixed-effects specific to the individual firm, which do not vary over time, I have tried to apply a fixed-effect logit model (for your information, I have already applied fixed-effect and dynamic panel models to an alternative sample with a continuous dependent variable, which also is the reason for me wanting to use fixed-effects). Based on several examples available online, I have tried the following code/approach in SAS:
PROC LOGISTIC DATA=mydata;
CLASS YEAR /PARAM=REF;
MODEL TURN (Event = '1')= YEAR OC DOMI COSTR ASSETR SIZE;
(where all variables are continious except DOMI, which is a dummy)
However, the estimation cannot succeed or yield any results (i.e. convergence is not achieved and I get the error "all strata are noninformative"), which seem to be a result of the incidental parameters problem as my dependent variable takes on the same value for all observations per firm. As this is my first experience with panel data, I have looked around online for information but with no luck.
Therefore, I am seeking advice/help on how to estimate a fixed-effect logit model in SAS given my variables, or an alternative approach.