Corporate credit rating systems have been an integral part of expert decision making of financial institutions for the last four decades. They are embedded into the pricing function determining the interest rate of a loan contact, and play crucial role in the credit approval process. However, the currently employed intelligent systems are based on assumptions that completely ignore two key characteristics of financial data, namely their heavy-tailed actual distributions, and their time-series nature. These unrealistic assumptions definitely undermine the performance of the resulting corporate credit rating systems used to inform expert decisions. To address these shortcomings, in this work we propose a novel corporate credit rating system based on Student’s-t hidden Markov models (SHMMs), which are a well-established method for modeling heavy-tailed time-series data: Under our approach, we use a properly selected set of financial ratios to perform credit scoring, which we model via SHMMs. We evaluate our method using a dataset pertaining to Greek corporations and SMEs; this dataset includes five-year financial data, and delinquency behavioral information. We perform extensive comparisons of the credit risk assessments obtained from our method with other models commonly used by financial institutions. As we show, our proposed system yields significantly more reliable predictions, offering a valuable new intelligent system to bank experts, to assist their decision making.
Engineering and Technology
Electrical Engineering - Electronic Engineering - Information Engineering
Cyprus University of Technology
