Banks PD shadow rating approach
-
Hi RB
In building shadow rating models for Low Default Portfolios (LDPs) (e.g., Banks):
-
Do peers use the full external population from external rating agencies in their development sample, or only the co-rated population (i.e., just the entities that are also the bank’s active customers)?
-
What are the different types of external ratings that you have seen used in modelling?
-
How are banks that receive bailouts dealt with in calculating a long-run average default rate?
-
-
The guidelines for development data should be less around whether only the internal data is used vs. e.g., the full S&P ratings universe. Instead, it should be around sufficiency of internal data and representativeness of development data to application portfolio
-
E.g., if a bank has insufficient internal observations, then they should use a maximal universe
-
We also see an example where a bank lacked financials internally and had to use external financials, but the intersection of (1) being an internal client and also having both (2) external financial data and (3) external ratings resulted in too small a development dataset. This led to them using the full externally rated universe, rather than just their internal observations
-
Using standalone rating from ECAI as a target
-
Include failures in definition of default (i.e., banks that would have defaulted had they not received direct/indirect support from the government, including money or government schemes; e.g., Merrill Lynch in 2008)
-