Application of the count panel data models to credit event (default/delinquency) of consumer loans is discussed. Model selection criteria is reproductive property of distribution and overdispersion property, and a negative binomial model, NB1 (variance is a linear function of the mean), is selected. Empirical study of the model is conducted by using insurance surrender data. Model parameters are estimated through panel data analysis. Macroeconomic indices as well as policyholders’ attributes are taken as explanatory variables. It is found that surrender of the insurance contract is predominantly explained by change of unemployment rates, time elapsed from the contact data, and seasonality.
|