An estimation model for country risk rating

Several statistical models, such as multiple regression, logistic regression (LOGIT), classification and regression tree (CART), etc. have been suggested in the literature and used in practice to explain and predict country risk ratings as a function of some selected social, political, but mostly economic factors or indicators. Such models, however, seem to have some important shortcomings. First, the same set of parameter values are usually assumed to apply to all countries, regardless of whether a particular political-economic factor carries more weight in the case of some countries. Second, the level of data adjustability is achieved only within the context of the optimization techniques inherent in the statistical models chosen a priori. This paper proposes, and applies to a group of 70 countries, a procedure that employs a generalized logit model (G-LOGIT) to link country risk rating and political-economic indicators. The estimates for the parameters of the G-LOGIT model are obtained through an independently developed mathematical programming model, rather than relying on classical optimization techniques as do most statistical models. The performance results of the proposed procedure are compared with two widely used statistical models: LOGIT and CART. The results indicate that the new procedure is superior to the statistical models, with respect to both estimation and validation errors.