Loss Forecasting Models for CCAR

bankingThe Client

Top 10 U.S. banking institution

The Challenge

A U.S. banking institution needed to validate multiple Loss Forecasting models in preparation for CCAR (Comprehensive Capital Analysis & Review).

The Approach

Cognilytics took a 3-step approach which included Conceptual Soundness, Review/Replication and Testing:

Step #1: Conceptual
  • Reviewed choices regarding model design, performance window chosen, target variable definition, and choice of predictors
  • Leveraged professional experience/reviewed relevant research
  • Interviewed model developers
Step #2: Review/Replication
  • Reviewed documentation related to
  • Data gathering
  • Model selection
  • Implementation
  • Monitoring reports
  • Selective replication of areas with highest risk of influencing model performance
Step #3: Testing
  • Backtesting – evaluated prediction accuracy of models outside of development window
  • Stress Testing – analyzed sensitivity of predictions to changes in the inputs
  • Benchmarking – constructed alternate models and compared with model in use

The Solution

Cognilytics provided a full validation, which included:

  • A review of conceptual design
  • Select replication of data gathering
  • Data exploration
  • Model selection
  • Outcome analysis
  • Stress testing

The results of validation and findings were presented to model owner and model risk management leaders. This validation report included:

  • Summary of conceptual design
  • Detailed model description for spreadsheet models
  • Review of all validation activity
  • Assessment of model adequacy
  • Risks/limitations of model usage
  • Document written to stand on its own with minimal cross-referencing