Journal No 242, October 2013
- As part of the comprehensive capital analysis and review (CCAR) process, banks are required to maintain both champion and challenger models for stress testing.
- Bank managers, likewise, should favor a competitive modeling framework, with the champion model being that which delivers the most strategic insight and the most accurate portfolio projections at the lowest possible cost.
- Most banks, however, are choosing not to adopt a competitive stress test modeling structure. Instead, the standard process seems to be that banks will arbitrarily nominate, in advance, a certain model development initiative labeled the “champion” and a separate, more poorly funded and resourced project known as the “challenger.”
- A dynamic approach allows for mitigating actions and quantifies the expected risk profile and size of future booking
- A static approach freezes the balance sheet figures at the end of the sample and assumes that the rate of origination and expected quality of new bookings will be the same as prior to the stress test
- With the exception of mortgages all other retail products have a relatively short time frame causing bulk of defaults to occur early on.
- For short lived products, who in the portfolio will be causing defaults and losses in the second, third etc… year of a stress test?
- The objective of stress testing is to link in a robust and coherent way macro economic scenarios to credit performance
Macro economic data tend to be smooth capturing systemic movement while granular risk data reflects the idiosyncratic nature of individual loans. Need to aggregate portfolio data to get rid of idiosyncratic noise
- Proposed way is to group accounts by cohorts or vintage according to origination date
- Tendency in the credit industry has been to leverage existing/legacy retail credit risk models by stressing the drivers of existing formulas or by adding a stress testing layer to the outputs of the model(cycle adjustment, macro adjusting factors)
- The next big step in the industry is to start developing first generation of stress testing models that aims at explaining the shape and dynamics of credit risk parameters at the tail of their distribution