Identifying and managing volatility in your CECL model 

Assumptions are typically the main driver of volatility in CECL models. CECL models may use different assumptions depending on the selected model methodology. Common methods involve projecting cash flows and estimating future losses using a cash flow model. This normally involves developing a prepayment assumption, or a payment made beyond what is contractually obligated. To run an accurate cash flow projection, the model requires an assumption regarding expected prepayments. For institutions using the WARM (weighted average remaining maturity) method, prepayments are also commonly used to calculate the weighted average remaining life. Regardless, including prepayments in the calculation will shorten the effective life of the portfolio and thus decrease the exposure for future losses. Higher prepayment rates reduce projected losses, while lower rates increase them. In cash flow calculations, the expected loss calculation can be extremely sensitive to changes in prepayment rates, especially long-term loans such as mortgages. It's crucial to understand and document how prepayment rates in your model are calculated. Many models use a constant prepayment rate (CPR) that remains the same throughout the entire life of the portfolio. In these cases, more recent look-back periods are recommended because they better align with the expectations of prepayment activity over the near term. Other models use regression analysis to predict prepayment rates over a forecast period (typically one or two years) and then revert to a long-term historical average. This is often recommended because the long-term average will remain more consistent and thus will reduce volatility related to prepayment rates.

Another source of volatility, regardless of methodology, is the loss rate assumption. This can either take the form of probability of default (PD) / loss given default (LGD), net charge-off rates, or gross charge-off rates with a recovery rate assumption. Changes to the loss rates will affect the outcome of the projected loss model regardless of the loss rate methodology. Generally, these changes have less effect on cash flow compared to similar prepayment rate changes, unless the portfolio is short-term. Like prepayment rates, reverting to a longer-term historical loss rate after a forecast period will help reduce volatility.

CECL now requires the consideration of reasonable and supportable forecasts. Depending on your forecasting methodology, they may also be a significant source of volatility. Models often use economic factors to predict future assumptions over a forecast period. If the forecast period is only one to two years before reverting to a historical calculation, it may not result in high volatility. The initial years of cash flow calculations have the highest exposure since the cash flows have not yet amortized for the payments and prepayments. Therefore, changes to a year one or two loss rate will significantly affect any changes compared to changes in subsequent years. Assuming the economic forecast utilizes regression analysis to predict assumptions, these forecasted assumptions may be sensitive to variations in the forecasted economic metrics, particularly if the slope of the regression line is high. 

Another source of potential volatility is qualitative adjustments.  A significant qualitative adjustment indicates a major change in current conditions that is not captured within the actual modeling using historical data, which management could document and understand why a change to the allowance was made. With the introduction of CECL, there has been a shift towards quantifying qualitative factors and establishing a more supportable and measurable qualitative factor framework. (For more information on developing a qualitative factor framework, follow this link). This involves creating a framework that is consistently applied, reducing the subjectivity of making adjustments and ensuring consistency in the future. Depending on how the framework is set up, small changes to current conditions may lead to large swings in the applied qualitative adjustment. It’s important to understand how your qualitative framework operates so you can identify any areas of potential volatility.  

For example, factor scales are a common way for institutions to quantify the application of risk ratings within the framework. For changes in delinquency, institutions may develop a sliding scale to determine which risk rating to apply to which levels. It’s important to develop scales that are not overly restrictive, ensuring that minor variations in delinquency trends do not result in significant risk designations when no risk was apparent in the last period. To avoid this unnecessary volatility, it’s important to keep the sensitivity of the framework in mind when developing it.  

Many CECL models use complex methodologies, including discounted cash flow models that integrate industry data, economic factors and contractual data and model loans on an individual basis. This means assumptions are applied on an individual loan level and not at a segment level. These models are often seen as “black box” as the calculation of the assumptions is done behind the scenes using proprietary formulas and calculations. It can be difficult for users of these models to understand how the model calculates assumptions; therefore, it can be difficult to understand where volatility in the model originates from. For example, does volatility originate from changes to economic forecasts? Which underlying assumptions are driving the volatility; did the economic forecasts change the predicted prepayment rates or was the volatility related to changes in predicted loss rates, default rates, loss given defaults or recovery rates? All these questions can be difficult to answer if the users do not understand how their model works. Another solution is to have a model validation completed on your CECL models. Model validations will ensure the model is working properly and can provide valuable insight into how the calculations “behind the scenes” are working together to produce your institution’s CECL estimate. These value-added insights tend to provide further insight to management and users of the model to help better understand where the volatility is originating from  For more information on model validations read more here.

The best way to understand potential sources of volatility in your CECL model is with sensitivity testing. Sensitivity testing is meant to quantify, (1) which inputs and assumptions the model is most sensitive to and (2) how sensitive the model is to changes in these inputs and assumptions. It is important to isolate assumptions so users can understand how each assumption impacts the calculation. For example, how do changes in prepayment rates affect the model compared to similar changes in loss rates? By performing the sensitivity tests, users can create scenarios with varying changes to key assumptions. Comparing the results of the different scenarios will help the user understand the overall sensitivity of the model.