Utilizing historical data for corporate credit rating assessment

摘要

Corporate credit rating assessment is one of the crucial problems of credit risk management; it will help the financial institutions and government decide whether to issue debts. Recent studies focusing on the prediction of credit rating by using artificial intelligence (AI) techniques have shown impressive results compared to traditional statistical methods. Although the AI techniques can be used to assess credit risk, the prediction accuracy is still worth improving further, as even a small improvement in credit rating prediction accuracy leads to significant loss reduction in the industry. In this paper, we propose new learning analytic methods to enhance the prediction accuracy of credit rating. First, we devise the metrics based on the credit rating history of the firms, and expand the feature space with new input variables. This approach can be applied to any conventional AI methods for improvement of prediction accuracy. Second, we develop a novel learning algorithm that is designed to take into account historical financial data. We propose the parallel artificial neural networks (PANNs) ensemble model that creates several independent artificial neural networks (ANNs); each ANN deals with financial performance of the firms for each year, and the final output of PANNs is aggregated by ensemble learning. In our experiment, three different real-world datasets are used to validate the performance of our proposed approach. Consequently, the experimental results show that our proposed approach achieved competitive results compared to conventional AI techniques.