[Abstract]

Machine Learning, Fuzzy Logic and Chance Discovery: Applications in Credit Risk Modeling With Special Reference to Indian Banking

Surendra Barsode, Lokesh Naga, and Sameer Rege
Shailesh J Mehta School of Management , Indian Institute of Technology, India



Basel-II norms for the regulatory capital of commercial banks require banks to use Internal Rating Based system to enable them take benefit of lower and risk-sensitive measurement of capital requirement. Implementing such a system is a tall order even for bigger banks in G-10 countries. In the context of developing countries like India, the problem is even more intractable due to lack of data and systems in place. We present a Basel-II compli ant model by designing a 3-stage Intern al Rating System (IRS) for assessing creditworthiness of borrowers. In first stage, taking publicly available data on credit rating on Indian manufacturing companies from 1995 to 2003 , we use statistical and machine learning techniques along with meta-learning for multi-valued classification of borrowers. In 2nd stage, we use a fuzzy logic based assessment system for incorporating qualitative factors and introduce further granularity in the classification. In the 3rd stage, we integrate the results from previous two stages and introduce Chance Discovery framework , an iterative process of anticipating risks and opportunities with human and computer interaction, by examining relationships between risk factors and their causes and identifying chance events . This framework can be used by any commercial bank to install not only a Basel-II compliant IRS but also to improve its capital adequacy and profitability .