Improving credit risk measurement and management : a new application of statistical techniques
Abstract
In an ever-growing economy, increased competition and pressures for increased revenue has led financial institutions to search for more effective ways to attract creditworthy clients. Since the 1950s, credit scoring has been widely adopted to guide credit decisions, however literature on credit scoring has been limited. Credit scoring plays a critical role in the banking environment which affects future impairments, capital and profits. In light of the limited re-search and the importance of credit scoring, the need to augment existing techniques and develop new techniques to improve credit risk measurement and management are para-mount. This thesis explores three significant problems in the world of credit scoring that affect credit risk measurement and management.
The first article addresses the issue that no optimal technique exists in building credit score-card models and also provides a solution for the disagreement on the appropriate cut-off score. An optimal Credit Scoring Matrix Model (CSMM) to determine which clients will go bad in the future is proposed. The CSMM gives uplift to the Gini-coefficient compared with a one-dimensional credit scorecard and provides a solution to determine an appropriate cut-off score on a more granular level.
The second paper explores the effect of scorecard implementation on the performance measures for the accept population. In credit scoring, much focus has been placed on model-ling techniques for building credit scorecards. Less focus has been put on credit scorecard implementations. Performance measures on the accept population appear to change after the implementation of the credit scorecard against the development sample. In this paper a statistical technique called swap-set Gini-coefficient provides a more comparable statistic be-tween development and post implementation of the credit scorecard. The swap-set Gini-co-efficient performance measure results indicate significant improvement for monitoring the credit application scorecard for the accepts population.
In the third paper, the procyclicality problem on credit scorecards is investigated. The perfor-mance of a bureau scorecard during a downturn and an upturn period is analysed with the results indicating the necessity of calibration to account for procyclicality. Various calibration scenario results are presented indicating the significant contribution of the proposed calibra-tion model.