Before selecting KMV, a handful of banks evaluated its performance along with that of other third-party models, including CreditRisk+, Credit MetricsTM, products from Algorithmics, and CreditPortfolioView. The evaluating banks built prototype portfolio models and benchmarked the performance of third-party options against them. They chose KMV products for their ability to represent corporate portfolios accurately. This accuracy is made possible by KMV products’ strong capability in modeling concentration risk in low-default portfolios.
Feedback on the performance of KMV’s RiskfrontierTM with retail portfolios was mixed. Some banks were dissatisfied with the methodology, and some found that the results were not in line with intuition. They have moved away from KMV’s products and developed a proprietary model based on default intensity for their retail and SME portfolios.
Banks find KMV products’ capital allocation to business units/obligors in line with their intuitive rank ordering of risk in their portfolio. Several of our interviewees using KMV products commended their performance in identifying portfolio concentrations and allocating appropriate capital during the current crisis.
Banks felt that while CreditRisk+ accurately estimates the overall capital required, its marginal capital allocations are not in line with the individual risk contribution of obligors/industries.
Proprietary models are clearly gaining popularity across banks because they:
Increase transparency in portfolio risk estimation (as preferred by regulators);
Can model risk profiles of complex instruments;
Can incorporate region-specific correlation structures;
Can provide levers necessary to perform comprehensive stress testing.
Developing a proprietary model is a fairly complex process. Exhibit 1 on the following page illustrates the three fundamental choices in designing a long-term model: the estimation methodology, the treatment of rating transitions, and the correlation structure.
Of the three design choices, the most difficult is the correlation structure. Most retail-focused banks have developed proprietary models based on default intensity, while corporate-focused banks have favored asset-return-based models. Next in importance is the treatment of rating migrations – specifically, deciding between mark-to-model (MtM) and default mode (DM). Most models (both third-party and proprietary) provide flexibility to recognize losses in both