The current expected credit loss (CECL) adoption deadline of Jan. 1, 2023 has many financial institutions evaluating various models and assumptions. Many financial institutions haven’t had sufficient time to evaluate their CECL model performance under various stress scenarios that could provide a more forward-looking view, taking the model beyond just a compliance or accounting exercise.
One critical element of CECL adoption is model validation. The process of validating a model is not only an expectation of bank regulators as part of the CECL process — it can also yield advantages for institutions by providing crucial insights into how their credit risk profile would be impacted by uncertain conditions.
In the current economic environment, financial institutions need to thoroughly understand what an economic downturn, no matter how mild or severe, could do to their organization. While these outcomes really depend on what assumptions they are using, modeling out different scenarios using more severe assumptions will help these institutions see how prepared they may or may not be.
Often vendors have hundreds of clients and use general economic assumptions on them. Validation gives management a deeper dive into assumptions specific to their institution, creating an opportunity to assess their relevance to their facts and circumstances. When doing a validation, there are three main pillars: data and assumptions, modeling and stress testing.
Data and assumptions: Using your own clean and correct data is a fundamental part of CECL. Bank-specific data is key, as opposed to using industry data that might not be applicable to your bank. Validation allows for back-testing of what assumptions the bank is using for its specific data in order to confirm that those assumptions are accurate or identify other data fields or sources that may be better applied.
Modeling (black box): When you put data into a model, it does some evaluating and gives you an answer. That evaluation period is often referred to as the “black box.” Data and assumptions go into the model and returns a CECL estimate as the output. These models are becoming more sophisticated and complex, requiring many years of historical data and future economic projections to determine the CECL estimate. As a result of these complexities, we believe that financial institutions should perform a full replication of their CECL model. Leveraging this best practice when conducting a validation will assure the management team and the board that the model the bank has chosen is estimating its CECL estimate accurately and also providing further insight into its credit risk profile. By stripping the model and its assumptions down and rebuilding them, we can uncover potential risks and model limitations that may otherwise be unknown to the user.
Validations should give financial institutions confidence in how their model works and what is happening. Being familiar with the annual validation process for CECL compliance will better prepare an institution to answer all types of questions from regulators, auditors and other parties. Furthermore, it’s a valuable tool for management to be able to predict future information that will help them plan for how their institution will react to stressful situations, while also aiding them in future capital and budgeting discussions.
Stress testing: In the current climate of huge capital market swings, dislocations and interest rate increases, stress testing is vital. No one knows exactly where the economy is going. Once the model has been validated, the next step is for banks to understand how the model will behave in a worst-case scenario. It is important to run a severe stress test to uncover where the institution will be affected by those assumptions most. Management can use the information from this exercise to see the connections between changes and the expected impact to the bank, and how the bank could react. From here, management can gain a clearer picture of how changes in the major assumptions impact its CECL estimate, so there are no surprises in the future.