Validation of Scores and PDs
9. July 2009 07:27
Main objectives:
discriminatory power: relative assessment of the PDs
calibration: absolute assessment of the PDs
Other:
– stochastic dominance
– monotonicity
Model Performance
Model performance is important for ensuring high-quality pooling and approval decisions.
Are the estimated PDs consistent with the observed default rates?
Backtesting and Stresstesting
discriminatory power: relative assessment of the PDs
calibration: absolute assessment of the PDs
Other:
– stochastic dominance
– monotonicity
Model Performance
Model performance is important for ensuring high-quality pooling and approval decisions.
- Do the bad customers have low scores?
- Do the good customers have high scores?
- To what degree do the score distributions of good and bad customers overlap?
- How well does the model separate the good from the bad customers?
- How many of all bad customers can you find within the low-scoring customers?
Are the estimated PDs consistent with the observed default rates?
Backtesting and Stresstesting
- is an internal rating system that has been developed several periods ago still applicable for todays’ data?
- how stable is the rating system?
- simulations of different scenarios
Email 
