Host: The Japan Society for Management Information
Binominal logit model is probabilistic model of non-linear type and model parameters are usually estimated by maximum likelihood estimation. In maximum likelihood estimation, parameters are estimated to fit whole data maximally but does not necessarily reflect characteristic of each data. On the other hand, it is natural to think that the management condition of companies varies according to the qualitative factors such as the length of management or the ability of managers even if they have same financial settlements.
Therefore, in this study, we try to improve prediction power of credit risk model by supposing prior distribution for model parameters and estimating parameters by hierarchical Bayesian method.
