A "snapshot" of talk by Gary Wilhite, Wells Fargo, on Keeping Capital Models in Perspective:
How Should Capital Models Deal With Uncertainty & Assumption Mismatches?
Why are we interested in this question?
– Affects capital requirement
– Helps us understand the risk that models will mislead
• Tell us to do the wrong thing
• Be used in the wrong way
• Give us a false sense of security
How do we communicate to others?
Choices May Result from Model Objectives
What are your EC models being asked to do?
– How much capital do we need?
– How much loss volatility is inherent in our portfolio?
– Are we being paid for the risk we are taking?
– Can we improve the risk/return ratio for our portfolio?
– Where do we have problematic concentrations?
What’s the benefit if we do something about them?
– Why are our peers pricing differently?
Can one number answer all these questions?
Recognize the Choices or Assumptions Made
Input Parameter Estimation (PD, LGD, EAD)
– Conditional on what?
• Which risk factors do I know/include?
– Reference data set?
• Time / consistency
– Structure of the analysis
Capital Model
– Consistency of inputs and how the model is run
– Data for correlation modeling
– Tail risk assumptions
Assessing the impact
Parameter examples
Through the Cycle / Point in Time PD
– Graded portfolio
• Given that I know the grade…
• … but not current stock price/volatility/economic outlook (except as included in the grade) – Retail equivalent…
• Given I know current Fico and LTV
• …?? what about HPI, unemployment rate ??
• … but not HPI and unemployment forecasts
Downturn LGD
– What’s the downturn?
– How many downturns? How severe?
Relationship to Confidence Level
...
Key Points
– Recognize the assumptions and choices you’re making.
– Understand how the choices affect the meaning of your result
– Be aware of how others may interpret your result and
communicate about the differences in meaning.
– Understand how sensitive your results are to the assumptions you make that are not fully aligned with reality.
– Think through the impact
– Decide how to adjust the result, if necessary
Do this with models and/or with subjective assessments when models don’t address the uncertainty well
– The key point is to do it
– These are tools; must understand them
– Decide for your institution how to answer each question
– There may be a value in running analyses several ways
– INTEGRATE with other tools like stress testing
Communicate
– People want the answer to be easier than it is
– People want a number that assures them
– Risk must educate
Replies