Sparse and Distributed Gaussian Processes for Modeling Corporate Credit Ratings

Credit rating agencies are tasked with determining an entity’s credit rating. The agencies perform thorough, often expensive, and time-consuming investigations to obtain these ratings. There has been an increase in the literature on the use of automated data-driven and machine learningMachine learning technologies for predicting the credit rating of corporations. In this chapter, we present a first-in-literature use of sparse and distributed Gaussian processes for this task. Using financial ratio data from US companies and rating information from the major credit rating agencies, we build models for predicting credit ratings for US companies. Our results show that distributed Gaussian processes, which are variations of product-of-expert models, outperform the sparse Gaussian processes—with the benchmark model being logistic regression. However, we find that the performance of the distributed Gaussian processes degenerates when more experts are utilized. The results from the sparse and distributed Gaussian processes also produce the most important variables through the automatic relevance determination kernel. We find that the debt ratio, profit margin-related metrics, and return on capital ratio are the most important financial metrics in determining the corporate rating. These results could prove useful for various stakeholders that make use of the credit ratings.