Dear all,
I have quarterly data of firms with many i (some thousands) and a relatively small number of time periods (maximum 6 per observation) and I want to estimate a dynamic probit. I use the Chamberlain/Mundlak approach to handle the correlation between the heterogeneity and the covariates. I also use Wooldridge’s (2005) solution for the endogeneity of the initial conditions.
My question is whether I should use a pooled probit or a random effects probit. I am not sure if the assumptions of the random effects probit, mainly that of serial independence, hold given the small number of time periods and that they are quarters rather than years. I am more interested on the marginal effects rather the coefficients but there is a significant difference between the marginal effects obtained from the two models and I don’t know which one I should trust.
Any suggestion is welcomed.
Regards,
Nikos
I have quarterly data of firms with many i (some thousands) and a relatively small number of time periods (maximum 6 per observation) and I want to estimate a dynamic probit. I use the Chamberlain/Mundlak approach to handle the correlation between the heterogeneity and the covariates. I also use Wooldridge’s (2005) solution for the endogeneity of the initial conditions.
My question is whether I should use a pooled probit or a random effects probit. I am not sure if the assumptions of the random effects probit, mainly that of serial independence, hold given the small number of time periods and that they are quarters rather than years. I am more interested on the marginal effects rather the coefficients but there is a significant difference between the marginal effects obtained from the two models and I don’t know which one I should trust.
Any suggestion is welcomed.
Regards,
Nikos
