Objective

• The scope of the engagement was to validate the statistical model used for HPI scenarios expansion
• The model used simple linear regression model to expand national level (US) HPI to (51) state level HPI in 5 different CCAR scenarios – the model used different equation for stress and normal period

CRISIL's solution

• Data Preparation: Quarterly HPI forecast was converted to monthly to predict monthly state-wise HPI
• Validation Approach: (i) review of model development document, implementation file, and data integrity; (ii) review of conceptual soundness; and (iii) review of model governance
• Goodness of Fit and Outcome Analysis: Serial correlation, heteroscedasticity, causality, structural break, etc., were examined for all the models. Out-of-sample statistics, out-of-time validations, backtesting, sensitivity analysis, and stress testing methods were used to ensure the model is appropriate. This included a few graphs and error tests/tables
  

Client impact

• Validated the entire model within stringent timelines, and created extensive and thorough documentation for regulatory submission
  

Questions

Looking for high-end research and risk services? Reach out to us at: