CREDIT RISK MODELLING FOR INDIAN DEBT SECURITIES USING MACHINE LEARNING

We develop a new credit risk model for Indian debt securities rated by major credit rating agencies in India using the ordinal logistic regression (OLR). The robustness of the model is tested by comparing it with classical models available for ratings prediction. We improved the model’s accuracy by using machine learning techniques, such as the artificial neural networks (ANN), support vector machines (SVM) and random forest (RF). We found that the accuracy of our model has improved from 68% using OLR to 82% when using ANN and above 90% when using SVM and RF.

European Puts, Credit Protection, and Endogenous Default

In a default corridor [Formula: see text] that the stock price can never enter, a deep out-of-the-money American put option replicates a pure credit contract (Carr and Wu, 2011, A Simple Robust Link between American Puts and Credit Protection, Review of Financial Studies 24, 473–505). Assuming discrete (one-period-ahead predictable) cash flows, we show that an endogenous credit-risk model generates, along with the default event, a default corridor at the cash-outflow dates, where [Formula: see text] is given by these outflows (i.e., debt service and negative earnings minus dividends). In this endogenous setting, however, the put replicating the credit contract is not American, but European. Specifically, the crucial assumption that determines an endogenous default corridor at the cash-outflow dates is that equityholders’ deep pockets absorb these outflows; that is, no equityholders’ fresh money, no endogenous corridor.

Hybridisation of Feature Selection and Classification Techniques in Credit Risk Assessment Modelling

In recent years, the use of artificial intelligence techniques to manage credit risk has represented an improvement over conventional methods. Furthermore, small improvements to credit scoring systems and default forecasting can support huge profits. Accordingly, banks and financial institutions have a high interest in any changes. The literature shows that the use of feature selection techniques can reduce the dimensionality problems in most credit risk datasets, and, thus, improve the performance of the credit risk model. Many other works also indicated that various classification approaches would also affect the performance of the credit risk assessment modelling. In this research, based on the new proposed framework, we investigated the effect of various filter-based feature selection techniques with various classification approaches, namely, single and ensemble classifiers, on three credit datasets (German, Australian, and Japanese credit risk datasets) with the aim of improving the performance of the credit risk model. All single and ensemble classifier-based models were evaluated using four of the most used performance metrics for assessing financial stress models. From the comparison analysis between, with, and without applying the feature selection and across the three credit datasets, the Random-Forest + Information-Gain model achieved a better trade-off in improving the model’s accuracy rate with the value of 96% for the Australian credit dataset. This model also obtained the lowest Type I error with the value of 4% for the German credit dataset, the lowest Type II error with the value of 2% for the German credit dataset and the highest value of G-mean of 95% for the Australian credit dataset. The results clearly indicate that the Random-Forest + Information-Gain model is an excellent predictor for the credit risk cases.