CECL validations and stress-testing - your questions answered

Key takeaway:

• Best practice is a start to finish validation, not just model testing
• Start with model governance, polices and procedures surrounding the model and process
• Review and test the data inputs (loan level data) as well as the assumptions being applied
• Full replication provides value in that the entire portfolio is working accurately to produce final CECL estimate

Key takeaways:

• Data errors from extract and transfer process 
• Data assumption variances  
• User errors 

Key methodology features: 

• How is the discount rate being applied 
• Are my loss factors being vectored based on the economic environment 
• Are my prepayments shortening the life of the loan or re-amortizing the balance each month? 

Key takeaways:

• A CECL vendor’s own validation is typically a validation of their mathematical formulas and regression models within the model itself
• An independent validation focus on YOUR data within that model
• Value is here is does the software combined with your data calculate a reasonable and supportable CECL estimate

Key takeaways:

• Internally built model validations will focus on the data inputs and assumptions versus “black box” methodology
• Key here is does my assumptions applied within the calculation make sense for our institution
• Will also focus on the incorporate of a forecast since it is not as explicit in the basic methods

Key takeaways:

• Key assumptions for any CECL model whether complex or not are sensitive to changes in the economic landscape
• It is important to know the volatility of your CECL estimate so there are no surprises
• Start with your previous volatile time periods (great recession, COVID) to use for scenarios going forward