Checking in on CECL: Hot topics and what matters in 2025 

• Clear oversight structure: define roles and responsibilities related to model design, monitoring, validation and board reporting. 
• Model development documentation: detailed methodology selection, segmentation logic. 
• Data governance: ensure data quality controls, assumptions and input reviews are documented. 
• Validation and monitoring: assess conceptual soundness and ensure models perform as intended on an ongoing basis. 
• Stress testing and back testing frameworks: include testing model volatility and accuracy procedures. 

Common gaps identified include a lack of detailed process documentation, missing loan reconciliation at the segment level and insufficient internal controls, especially in vendor-based models. 

• Max loss assumption: Establish a “ceiling” for reserve adjustments based on historical or modeled data. 
• Factor Scales: Quantify risk levels (minor, moderate, major) using historical benchmarks. 
• Anchoring and weighting: anchor adjustments to max loss assumptions and apply weights based on factor relevance. 
• Documentation: maintain clear records of assumptions, calculations and rationale and be able to support max loss assumptions, factor scales and factor weights. 

This framework helps institutions streamline qualitative calculations and reduce bias in reserve estimates. 

Volatility in CECL models can stem from: 

• Assumptions and inputs: prepayment rates, average life and forecast methodologies. 
• Q- factors: inconsistent frameworks lead to unpredictable reserve swings. Also, “tight” factor scales may result in large swings over time. 
• Model changes: Switching vendors or methodologies introduces variability. 

The first step is to analyze historical data to understand trends and mitigate volatility. For example, institutions using long lookbacks that include COVID-era prepayments may see declining prepayment rates as those periods roll off, and it is important to know how that may impact CECL calculations. 

Back testing is now a best practice which includes validating model accuracy by comparing forecasts to actual outcomes or comparing the reserve to historical experience. Examples include: 

• Coverage tests: Compare reserves to historical loss rates. 
• Forecast accuracy: compare predicted vs. actual losses. 
• Assumption accuracy: test prepayment rates and other input assumptions against historical experience. 

Stress testing and scenario analysis complement this by modeling stressed scenarios  to assess model sensitivity. Both techniques support regulatory expectations and internal governance. 

Beyond stress testing, institutions can use: 

• Peer benchmarking: Compare reserves and charge-off rates to industry peers. 
• Ratio analysis: Identify anomalies in reserve levels relative to performance metrics. 

These tools help institutions calibrate their models and justify assumptions during audits or validations.