Credit Risk Evaluation Using Neural Networks

Credit risk evaluation and credit default prediction attract a natural interest from both practitioners and regulators in the financial industry. The Bank for International Settlements has been reporting a continuous increase in corporate borrowing activities.1 In the first quarter of 2006 alone, syndicated lending for mergers and acquisitions sharply exceeded the 2005 levels. In the euro area for example, corporate demand for credit rose from 56% of international claims on all non-bank borrowers at the end of December, 2005, to 59% at the end of March, 2006. These heightened borrowing activities naturally imply increased risk related to credit default. A study by Office of the Superintendent of Bankruptcy Canada and Statistics Canada2 reveals that while the number of Canadian firms going bankrupt has declined, the average size of losses has significantly risen. In 2005 only 0.7% businesses failed, a sharp decline from the 1992 rate of 1.54%. However, over the last quarter of the century net liabilities from business failures increased dramatically. In 1980 the losses represented 0.32% of Canada’s net assets, while in 2005 they rose to 0.52%. Both trends, the acceleration in corporate borrowing and the related risks of credit defaults, command the need for a reliable and effective risk management system on part of financial institutions in order to improve their lending activities. Moreover, the new international standard on capital adequacy outlined in Basel II, a regulatory requirement for financial services institutions, promotes the active involvement of banks in assessing the probability of defaults. Therefore, the accuracy of any predictive models constituting the foundation of a risk management system is clearly essential. Any significant improvement in their predictive capabilities will be worth billions of dollars and therefore deserves serious attention.