December 16th, 2014, 6:53 am
First of all, consider the EBA Guidelines on IRC. At the end of the day, IRC is a regulatory exercise, and some academically appealing concepts will not be allowed anyway. For instance, you will not be allowed to generate hypothetical spreads (what will be the spread for a CCC-rated Pfandbrief?) from historical data (PD/LGD) but will be required to somehow use "market data". IMHO,- use some very basic CreditMetrics-type portfolio model- forget about liquidity horizons, use constant position assumption instead (the regulatory requirements on liquidity horizons are particularly challenging) -> a one-period portfolio model will do- try to run the model with few parameters (correlations, migration matrices, LGD, spreads) but invest heavily in their validation- take validation very seriously, and be quite formal (have a validation concept, perform validation according to the concept, document validation results accordingly)