1. What are your thoughts on methods for default correlation estimation? When there are no direct estimates available, sometimes equity or CDS correlations are used, but these seem skewed upward for similar companies. Any thoughts on this?
A. Equity correlations are useful when there is no other source of information such as credit spreads. Equity returns as default predictors have weaknesses that are quite well characterised in the credit portfolio risk literature. Otherwise, CDS correlations are useful but consideration of the potential (common) jumps in CDS spreads is important.
2. One of the problems that we observed with the correlation between interest rates and hazard rates is that it is generally negative, but we have seen some situations in which the correlation disruptively changed sign, for example, during the UK mini-budget crisis where CDS and UK interest rates where surging at the same time. Therefore, the problem is that general correlation estimation may give you the opposite direction; however, the correlations explode very quickly and last momentarily, making WWR very difficult to estimate. How can this be solved in a meaningful way?
A. Non-stable correlations are an indication that the model specification is problematic. I don’t think there is a meaningful way to solve this within a model but the use of stress scenarios, reserving and macro-hedges is important. The general WWR with unstable correlations (other non-credit correlations may also show this behaviour) may tick regulatory boxes but not be meaningful for risk management.
3. In your opinion, what would be good approaches to model jumps? Is a historical based approach justified, for example, for PFE? In terms of modelling, MLE could be a good solution, or do you think there are better ones?
A.Yes, historical data is probably required as the credit options market is limited. Maximum likelihood estimation (MLE), threshold-based methods, or moment matching are the obvious three approaches.
4. Could you provide an overview of the types of backtesting approaches one could consider when testing the performance of WWR methodologies?
A.I think backtesting a WWR approach is problematic, similar to initial margin methodologies in that they are not conditional. I think it would be necessary to try and test against default or major market events specifically, rather than using general unconditional data. One could also try and ask the question of what multiplier converts a non-WWR measure to a WWR one. For example, in the case of Archegos, one can compare a 1.5 billion initial margin (presumably 99% confidence and including excess margin) to a loss of approximately 5.5 billion.
5. What are your thoughts on incorporating Wrong Way Risk to other XVA, for example, FVA?
A. It's probably easier since they do not need to be conditional on default. A hazard rate approach makes more sense for FVA WWR.
6. Would it make sense to use different methods of measuring WWR for different types of counterparties, or would this make it very difficult to compare?
A. Definitely different. The WWR to a corporate counterparty could be general and moderate, whereas to a fund, could be specific and sizeable.
7. Is there any way to measure default risk apart from CDS spread?
A. Ratings and structural models (e.g., Moody’s KMV) are the obvious alternatives. The three approaches together all have strengths and weaknesses.
8. CVA sensitivities: should they be using a WWR-informed model or an unconditional view? Also, are some forms of WWR hedgeable?
A. Being pragmatic, it is unlikely that the sensitivities in a WWR model are going to be materially different and allow hedging of WWR (aka cross gamma). This is why I think the BAU approach may incorporate no WWR or only general WWR, and then stress testing / toy models / reserves can be used to capture more complex WWR.
