Simulation-based vs. analytical estimation of portfolio loss. Simulation-driven models provide flexibility to incorporate fat-tail distributions of default probability and are more representative of underlying loss distributions than analytical models, which use normal approximations. They also allow easier extension of models to capture PD-LGD correlations and mark-to-model effects on portfolio losses.
Mark-to-model vs. default mode recognition of portfolio loss. Corporate-focused banks recognize loss in value due to deterioration in credit quality (mark-to-model), whereas retail-focused banks consider losses only in the event of realized default (default mode).
The ideal VaR model would be a simulation-based model in mark-to-model mode. The corporate exposure correlation would be driven from an asset-based correlation model and retail exposure correlation from a default-intensity correlation. The correlation between corporate and retail, though still being researched, could be based on the historical observed correlation of asset returns and default rates.
Almost all our interviewees agreed that a VaR model estimate of economic capital does not afford sufficient protection in the event of a downturn and that a comprehensive stress-testing methodology is needed to complement the VaR approach. For stress testing to play this role, however, more work is needed to refine the stress-testing governance model, the scope of the stress tests to be conducted, and the approach to incorporating stress test results into decision making.
Though most of the banks we interviewed estimate economic capital, they choose to keep and allocate the more conservative of economic or regulatory capital (usually the latter) to their business units, using a mix of methodologies to do so. About half of the group allocates capital based on marginal contribution, while some use heuristics to drive business objectives and others use models to allocate excess capital over economic capital.
1. Choosing a value-at-risk model
Banks may choose either to purchase third-party VaR models for estimating economic capital or to build their own proprietary model. Moody’s KMV suite of portfolio products is clearly the most popular choice among the 22 banks we surveyed; no other products have such broad acceptance. The KMV products’ high global penetration makes them the benchmark portfolio model; most risk management teams have chosen KMV products in order to be in line with this benchmark and because with KMV it is easier to defend both model choice and outputs to their leadership and colleagues.
Our survey also found that more than 60 percent of retail-focused banks rely on one model, while the majority of corporate-focused banks (45 percent) opt for two models. Of those banks running one model, 55 percent use KMV products.