Article ID: | iaor20042116 |
Country: | United States |
Volume: | 33 |
Issue: | 5 |
Start Page Number: | 4 |
End Page Number: | 21 |
Publication Date: | Sep 2003 |
Journal: | Interfaces |
Authors: | Nair Suresh K., Trench Margaret S., Pederson Shane P., Lau Edward T., Ma Lizhi, Wang Hu |
Keywords: | financial, markov processes, programming: dynamic, optimization |
We developed a method for managing the characteristics of a bank's card holder portfolio in an optimal manner. The annual percentage rate (APR) and credit line of an account influence card use and bank profitability. Consumers find low APRs and high credit lines attractive. However, low APRs may reduce bank profitability, while indiscriminate increases in credit-lines increase the bank's exposure to credit loss. We designed the PORTICO (portfolio control and optimization) system using Markov decision processes (MDP) to select price points and credit lines for each card holder that maximize net present value (NPV) for the portfolio. PORTICO uses account-level historical information on purchases, payments, profitability, and delinquency risk to determine pricing and credit-line changes. In competitive benchmark tests over more than a year, the PORTICO model outperforms the bank's current method and may increase annual profits by over $75 million.