Understanding and predicting sovereign debt rescheduling: a comparison of the areas under receiver operating characteristic curves

This paper extends the existing literature on empirical research in the field of sovereign debt. To the authors' knowledge, only one study in the area of sovereign debt has used a variety of statistical methodologies to test the reliability of their predictions and to compare their performance against one another. However, those comparisons across models have been made in terms of different probability cut-off points and mean squared errors. Moreover, the issue of interpretability has not been addressed in terms of interactions among explanatory variables with their correspondent debt rescheduling threshold level. The areas under the Receiver Operating Characteristic (ROC) curves are used to compare the discrimination power of statistical models. This paper tests logit, MARS, tree-based and neural network models. Analyses of the relative importance of variables and deviance were done. All of the models rank the previous payment history as the most important explanatory variable.