With every forward step in global risk regulations, there are hurdles that force you backward. This is true of all industries, of course, but the more restrictive the regulations the more likely you are to trip over these hurdles. With regards to the banking sector, the development, refinement, calibration and validation of internal credit scoring and risk models is an on-going issue, particularly in this intensified regulatory environment.
The key challenges that still keep practitioners on their toes are:
- Data warehousing: This age-old issue has implications to all industries wishing to partake in model development – making the phrase “garbage in, garbage out” well worn. A clear and consistent database is crucial for model development and validation. This is of the upmost importance for the credit risk modelling domain as a comprehensive data warehouse is required for Basel capital adequacy. Data still represents a significant challenge for banks wishing to comply with Pillar I of the Basel Accord because of need for granularity of data.
- Scarce default history: When a portfolio contains a relatively small number of defaults (Low Default Portfolio), accurate predictions are difficult, and methods such as under sampling or oversampling need to be considered. This is important to take into consideration when, for example, modelling PD (Probability of Default) for corporate exposures, as some traditional techniques do not perform as well when a large class imbalance is present.
- Reject inference: There are a number of potential methodologies that can be applied for reject inference (i.e., parceling, fuzzy augmentation and hard cut-off ). There is still much debate however as to which methodology produces the most accurate results and whether other methodologies could still be developed further.
- Forward-looking indicators: The use of forward-looking indicators, such as the deviations from a trend for the ratio of domestic credit to GDP, is still an area which requires more development for rating models and could be incorporated more.
- Accurate models: The accuracy of credit risk models is again another key issue in the modelling of the Basel risk parameters [PD, Loss Given Default (LGD) and Exposure at Default (EAD)]. In these cases, accuracy is paramount as the more accurate and robust the models are, the lower the risk. Several modelling techniques have been discussed before, but improvements can still be made to the modelling process.
There are still several potential pot holes to be filled in the successful implementation of credit risk models. I am interested to know what day-to-day challenges you encounter with regards to your model development and validation, and what improvements would you like to see moving forward?