Written By: John Hurlock
What does model validation mean for the allowance for loan and lease losses and what will it mean under CECL? Ever since the 2011 Interagency Guidance (2011-13) was issued and began to be applied to ALLL model, model validation has been going through a maturing process. Every model is different, and there have been some misunderstandings and misinterpretations of what validation means for ALLL. These divergent views could get amplified when the industry enacts the FASB’s current expected credit loss model, or CECL, potentially resulting in a lot of non-value added extra work and headaches. To begin this discussion, let’s start with the current ALLL model, which focuses on the incurred loss approach.
Incurred loss model and model validation
Incurred loss is exactly what its name implies, so think of the ALLL model as a real-time study of the institution’s credits (and really, principally, your loans). As FASB describes it, “Current GAAP requires an ‘Incurred Loss’ methodology for recognizing credit losses that delays recognition until it is probable a loss has been incurred. This model has been criticized for restricting an organization’s ability to record credit losses that are expected, but do not yet meet the ‘probable’ threshold.”
These are pools of loans you recognize as having correlated default characteristics but you have not individually recognized the defaulting credit because it is either immaterial or just hasn’t surfaced yet by displaying missed payments, reduced collateral values, etc. The credits you know about you have (most likely) already downgraded, and if they are material, you are determining the Loss Given Default (LGD) individually. For the pool loans, you look to the recent past along with outside factors such as unemployment, and use historical averages to determine what you have lost in order to estimate what you are currently probably going to lose in the pool.
With this type of “absolute” model, validation techniques like sensitivity analysis, backtesting or stress testing aren’t usually applicable. I won’t go into definitions of sensitivity, backtesting or stress testing in this article, but I do encourage you to look them up to get a better understanding of why they aren’t applicable to the ALLL model. The incurred loss model lives in the present and uses past data to recognize losses being incurred.
Expected loss model and model validation
The CECL model, on the other hand, creates a future view of the (static) asset side of the balance sheet at a specific point in time, much like Asset Liability Management (ALM) does, but without the rollover and growth rates. In essence, what are the losses expected to take place over the life of the earning asset (loans and investments)? Now we begin to introduce future-based concepts like Net Present Value (NPV), along with more granular views such as vintage loss analysis. These can – and most likely will – have sensitivity analysis, backtesting and stress testing techniques applied to them. While we still wind up with an absolute value to be reserved for, it is likely we will have ranges around that value identified and used in discussion as to why the amount we are reserving for is the correct one. For example, one factor of NPV is the projected interest rate, which could be incrementally adjusted up or down to see the impact on the value of future losses. Or stress testing could use different vintage periods to determine the impact of a worsening economic picture.
What about validation and maturity of the process of validation? With the current ALLL model, financial institutions wound up reacting to the 2011 validation guidance and learned how to apply it to their model. Is it an auditing function? “Wait, we already do that!” Or is it the credit quality component? “Wait, that’s what Loan Review does!” The answer to both is no. Validation, however, is becoming a standardized process with the focus on the model and how it calculates, as well as how it works in determining the reserve.
Supporting model validation during CECL implementation
CECL effective dates are rapidly approaching and validation will be an important component of each institution’s implementation. There is the opportunity to be proactive and prepare for model validation as you implement your plan for determining reserves.
Here are six suggestions to support the model validation process as you go through implementation.
- Rewrite your reserve policy and include discussion regarding your approach for the expected loss model.
- Identify the components that should be subject to sensitivity analysis and stress testing, and define how the results will be used.
- Define a backtesting strategy and how you will use it.
- Encourage your software vendor to make available a test environment using actual data. If they already have one, learn how to use it within defined parameters for sensitivity, backtesting and stress analysis.
- If you have an Excel model, make certain it is transparent and logical to follow through the calculation process.
- Document your transition plan.
With all of the newly collected data and future based analytical approaches involved under CECL, financial institutions have the opportunity to plan for validation while the model is being adopted. This makes the validation more efficient and effective, as well as supporting the decisions made throughout the ALLL determination process. In short, this makes it less costly as well.
Validation is going to become more robust and complex with the movement from incurred losses to expected losses. Being proactive makes all the difference when it comes to challenges and surprises. There are certain activities where being boring and predictable is not a bad thing. This is one of them.