Review of Internal ratings methodologies

SUMMARY

The project focused around the full review of the credit rating methodologies for non-financial counterparties, specifically models for sovereigns, corporates, funds and asset managers. The objective of the project was to test and document the credit rating methodologies to meet regulatory requirements and start the model validation process. Avantage Reply started the review with a pre-assessment of the rating model based on the best practices in order to identify the models’ gaps in different areas. Following the initial review, the project followed two work streams to address the identified weaknesses:

• Execution a full model review for each of the five credit rating models,
• Statistical testing of the models.

CUSTOMER GOAL

The client needed the models to be adequately documented and to address the weaknesses identified by the bank’s Supervisory Authority as well as to be able to perform a model validation.

CHALLENGES

The client had the models in use without them being fully tested or documented, as they related exposures were seen as non-significant in the overall risk portfolio. The model development and assumptions were not adequately documented and, as consequence, the models had to be reverse-engineered. Missing structured and long-term dataset for the model input and outcomes was making the statistical testing of the models a challenge.

SOLUTION

Avantage Reply started the assignment with a pre-assessment of the rating model based on the best practices laid out in the ECB’s Targeted Review of Internal Models (TRIM) guidelines, as well as the other applicable regulations, in order to identify the model’s gaps in data availability, methodology suitability, soundness of applied assumptions and documentation completeness and quality.

Following the initial review, the project followed two work streams to address the identified weaknesses:

• The team performed a full model review for each of the five non-financial credit rating models and reverse-engineered the models in order to address the issues with completeness and quality of model documentation. Based on this analysis, the documentation was amended so that it could be presented to the Supervisory Authorities
• Statistical testing of the models programmed in R to determine the adequacy and reliability of the model design, selection of risk factors, and underlying data using a selection of statistical tests such as correlation tests, rank ordering properties, benchmarking, population stability testing and indicator concentration.