Resources

CECL Model Validation: Common Pitfalls for Financial Institutions

Written by: Alan D. Lloyd , CPA , Michael V. Fallon , CAP ,

As of the date of this article, many banks and credit unions in the United States have implemented the Current Expected Credit Loss (CECL) accounting standard, ASC 326 – Financial Instruments – Credit Losses. In a previous article, we outlined some of the impacts the standard had on banks adopting in 2020 through January 1, 2023.

Wolf has been heavily involved in the standard, including providing our clients with independent audits of adopters, assisting institutions with implementation and model development, and performing CECL model validations.

In our experience, there have been common implementation pitfalls across institutions we’ve evaluated. Below are a few themes that we have identified:

Change Management

Change management is a key concept outlined in the regulatory guidance (SR 11-7) over model risk management. There should be a sound framework for processing, approving, and monitoring changes to the CECL model. To start, you should ensure that the model providers’ report on the Statement of Controls (SOC) exhibit strong controls over the software within the model.

The key item that we will focus on is the model settings. Several factors must be considered as many institutions are more heavily model-dependent under CECL. Changes to key settings in the model, whether accidental or nefarious, could cause significant variations in the model results. A few of these selections include:

  • How are loans amortized in the model?
  • How long is the forecast period?
  • How long is the reversion period?
  • Are historical losses weighted or straight-line?
  • What is the time to recovery?
  • Which calculation method is being used (i.e., discounted cash flow, weighted average maturity, PD/LGD, etc.)?

Your organization should be working with your internal teams and model providers to find a suitable audit trail or reporting mechanism where changes to these settings can be quickly and efficiently analyzed on a quarterly basis. If there are no such reports, consider contacting your providers to have this capability added. The model results are generally more challenging to comprehend given the additional reliance on software and internal calculations. Ensuring that these key model settings are not causing unexpected period-over-period variations is a critical step to gaining comfort with the results.

CECL Model Assumptions

As mentioned above, the model assumptions play a key role in producing the output generated by the CECL model. Assumptions drive the model output and should be thoroughly assessed, documented, supported, and monitored for appropriateness. Key assumptions of the CECL model may include, but are not limited to the following:

  • Prepayment speeds
  • Loss rates
  • Recovery periods
  • Qualitative factor adjustments

One of first requests typically asked by an independent party is to provide the documentation to support the basis of the model assumptions selected. This documentation should allow the reviewer to understand how and why the institution selected its assumptions.

Assessing other data components of the institution’s assumptions is also worthwhile. For example, if an institution’s asset and liability management (ALM) model has a conditional prepayment rate (CPR) speed of 12 on 30-year mortgage loans, but the CECL model is using a CPR of 18, then the institution should be prepared to explain the difference.

Most CECL model vendors provide an out-of-the-box assumption parameter set that their clients can utilize. These assumptions may not be appropriate for a given institution for a variety of reasons (assumed confidence intervals, loss rates, etc.). As such, assumptions should be tailored to be institution-specific, thoroughly assessed, monitored, and approved on a periodic basis.

Regulatory Guidance

An institution should be well-versed over the inner workings of CECL. The published guidance includes provisions on what is acceptable per the standard. For example, loans held for sale do not apply to the standard. See below for a provision taken from the guidance:

The new accounting standard applies to all banks, savings associations, credit unions, and financial institution holding companies (hereafter, institutions), regardless of size, that file regulatory reports for which the reporting requirements conform to U.S. generally accepted accounting principles (GAAP).

Further, ASU 2016-13 applies to all financial instruments carried at amortized cost (including loans held for investment (HFI) and held-to-maturity (HTM) debt securities, as well as trade receivables, reinsurance recoverables, and receivables that relate to repurchase agreements and securities lending agreements), a lessor’s net investments in leases, and off-balance-sheet credit exposures not accounted for as insurance or as derivatives, including loan commitments, standby letters of credit, and financial guarantees. The new accounting standard does not apply to trading assets, loans held for sale, financial assets for which the fair value option has been elected, or loans and receivables between entities under common control.

Sensitivity Analysis

Having a strong understanding of which factors drive the model result is also going to be very useful in having a sound model. Documenting a sensitivity analysis is a great way to understand what really drives the model results. We suggest you approach it as follows:

  1. Create a sandbox with a point-in-time model to run the analysis.
  2. Stress key model settings and assumptions one-by-one. Consider some set ranges for each assumption (e.g., up 20% and down 20%) to get a feel for up and down movements in a given factor. Having some meaningful range of stressing these up and down from the current setting is the most useful. Try to hit the low end and high end of the reasonableness range, or if there are numerous options for a setting, select a few to see how much the results change.
  3. Document each assumption and setting, and the related results. Include dollar and percent change from the base case to have a good dashboard of how each item impacts the model results.
  4. Use this data to drive your focus going forward. For example, if prepayment assumptions in a discounted cash flow have a significant impact on your results, make sure to have strong data and documentation backing that assumption. Conversely, if certain settings have immaterial changes, you can possibly “set and forget” them, and revisit them periodically via the sensitivity analysis.

Off-Balance Sheet Commitments

A key difference between CECL and the previous incurred loss method is how off-balance sheet commitments are reported. Under the CECL guidance, no credit losses should be recognized for off-balance sheet credit exposures that are unconditionally cancellable by the issuer. To illustrate this, please see the below example taken from the ASU 2016-13 guidance:

326-20-55-55 Bank M has a significant credit card portfolio, including funded balances on existing cards and unfunded commitments (available credit) on credit cards. Bank M’s card holder agreements stipulate that the available credit may be unconditionally cancelled at any time.

326-20-55-56 When determining the allowance for credit losses, Bank M estimates the expected credit losses over the remaining lives of the funded credit card loans. Bank M does not record an allowance for unfunded commitments on the unfunded credit cards because it has the ability to unconditionally cancel the available lines of credit. Even though Bank M has had a past practice of extending credit on credit cards before it has detected a borrower’s default event, it does not have a present contractual obligation to extend credit. Therefore, an allowance for unfunded commitments should not be established because credit risk on commitments that are unconditionally cancellable by the issuer are not considered to be a liability.

As a result of this update in guidance, institutions have seen significant increases in required reserves for these instruments in comparison to the incurred loss methodology which had minimal reserves. Instruments to consider for whether they are unconditionally cancelable include, but are not limited to the following:

  • Written loan commitments
  • Lines of credit
  • Letters of credit
  • Unfunded construction loans

A best practice recommendation would be to assess the terminology of the agreements for these instruments to determine whether they are unconditionally cancelable. We have seen institutions make updates to their agreements to be unconditionally cancelable to reduce these required reserves.

As part of our CECL model validation offering, we have detailed procedures to assess the CECL model, including reviewing change management and other CECL related controls, assessment over the reasonableness of assumptions, and alignment of current practices with regulatory guidance.

Are you seeking a trusted partner to help you avoid the common pitfalls that come with implementing CECL models? Reach out to our team of seasoned experts today!