Situation
A leading US credit card issuer wanted to conduct stress tests on its portfolio to determine impact on model performance and RWAs
Approach
A dedicated team of 6 specialists conducted thorough stress testing of portfolio encompassing risk driver stress testing, default rate stress testing and scenario stress testing (macro-economic)
IDENTIFY STRESS PERIOD
- Stress Period Identification: Plotted historical default rates for the portfolio
- Identified Peak Historical Default Rate as stress period
- For all segments identify top model predictors
STRESS INDEPENDENT VARIABLES
- Stress factor computation: For all predictors index them against risk* grades as of the stress period
- For all such grades (without changing grade bounds) create index as of the “scoring” month
- For each grade-variable cell compute the stress factor, floor it to 1
- Stress variable creation: Apply this factor to all independent variables as of scoring month at segment – risk grade level
RWA AND MODEL PERFORMANCE
- Model Performance: Score the model using stressed model variables
- Compute Gini, K-S, and other metrics to determine model performance, Characteristic’s analysis
- Compute RWA for stressed capital requirement and comparison with economic capital
Results
- Conducted stress tests within 3 months’ time frame for a single portfolio
- Around 16 segment PD model, 4 segment LGD model, and 6 segment EAD model
- Led to enhanced understanding of model strength and predictive power
- RWA computation report for various scenarios provided for benchmarking and business use