Predicting corporate investment/non-investment grade by using interval-valued fuzzy rule-based systems—A cross-region analysis

Abstract

Systems for predicting corporate rating have attracted considerable interest in soft computing research due to the requirements for both accuracy and interpretability. In addition, the high uncertainty associated primarily with linguistic uncertainties and disagreement among experts is another challenging problem. To overcome these problems, this study proposes a hybrid evolutionary interval-valued fuzzy rule-based system, namely IVTURS, combined with evolutionary feature selection component. This model is used to predict the investment/non-investment grades of companies from four regions, namely Emerging countries, the EU, the United States, and other developed countries. To evaluate prediction performance, a yield measure is used that combines the return and default rates of companies. Here, we show that using interval-valued fuzzy sets leads to higher accuracy, particularly with the growing granularity at the fuzzy partition level. The proposed prediction model is then compared with several state-of-the-art evolutionary fuzzy rule-based systems. The obtained results show that the proposed model is especially suitable for high-dimensional problems, without facing rule base interpretability issues. This finding indicates that the model is preferable for investors oriented toward developed markets such as the EU and the United States.