As new bank loan portfolio management tools receive increased attention, many institutions have begun formal work in attempting to compare the models. This article offers considerations banks can use when deciding the most appropriate model for their use. The three models compared here include KMV Portfolio Manager[TM]; J.P. Morgan Credit Manager[TM]; and Credit Suisse Financial Products Credit Risk+. Just as important as choosing the most appropriate model, say the authors, resources must be dedicated to evaluating the adequacy of loan systems, data warehouses, recovery databases, and usage statistics to provide the type and quantity of consistent and reliable data that is required by these models.
Banks and other financial institutions are now contemplating, experimenting with, or actively using credit risk management tools. Three of the best-known models are KMV Portfolio Manager[TM] (KMV), J.P. Morgan Credit Manager[TM] (JPM), and Credit Suisse Financial Products Credit Risk+ (CSFP). A comparative analysis of these three models was completed by the authors to highlight differences in terms of the key portfolio credit risk measurement variables the models provide as output, subsequent to processing of input data.
The comparison was completed by specifying comparable assumptions, completing input requirements, and processing the data through the models. The analysis does not compare the functionality of the models or present a treatise on their logic, modeling, and math. Therefore, the impact of using the models is discussed without preferring the analytical approach or risk and performance measurement features of one model vis-avis the other model.
For the purpose of this analysis, the key portfolio credit risk management variables are:
* Expected or average (mean) loss, or the product of expected default probability for a specified time horizon and expected loss given default (LGD). This risk is to be compensated by pricing in theory.
* Unexpected loss, or the standard deviation of the loss.
* Capital, the amount to be maintained in the form of equity for a specified confidence interval corresponding to the desired debt rating. The unexpected loss amount and required confidence level determine, along with the underlying mathematical approach for the loss distribution, the amount of capital required.
The first step in the comparison was to develop a benchmark portfolio suitable for each of the models and representative of a real-life portfolio in terms of return, risk, usage, and business considerations. The benchmark portfolio used in the comparative analysis consisted of more than 1,000 loan facilities. The borrowers are all publicly traded entities or are subsidiaries of or otherwise guaranteed by publicly traded entities. These commitments are a sample from an actual commercial loan portfolio of a bank consisting only of loans, thereby excluding treasury and other products that would require credit equivalent exposure calculations.
A direct comparison of the various portfolio credit management models is complicated due to the many assumptions used in all the products. Unless there is some consistency in assumptions used across the analysis, differences in the results of the models may be caused by differences in assumptions rather than fundamental differences in the models themselves. Analysts should exercise care in preparing any comparative study. It is, therefore, worthwhile to discuss the approach used for each assumption in this analysis on an individual basis.
One area that is particularly complex is correlation. KMV uses its option-theoretic model to arrive at individual market asset values for each obligor's business. Relationships among these underlying business asset values are then broken down into a set of common and specific factors. These relationship factors represent the basis of KMV's Global Correlation Factor Model. …