Community Institutions Face Unique CECL Modeling Limitations
With the effective date of Accounting Standards Update (ASU) 2016-13 Financial Instruments – Credit Losses: Measurement of Credit Losses on Financial Instruments (better known as CECL) getting closer, institutions should be prepared to comply with its terms. But the community institutions we work with have grappled with general limitations and data issues in their initial modeling attempts. You can learn from their efforts to help avoid similar challenges in your institution.
Updates amend original requirements
Before we explore the particular challenges community institutions have experienced when approaching modeling, let’s review the standard’s most recent clarifications.
ASU 2019-04 clarifications
The initial standard was written so that accrued interest would be included in the amortized cost basis of the receivable. The ASU amends this requirement and allows an entity to:
- “Measure the allowance for credit losses on accrued interest receivable balances separately from other components of the amortized cost basis of associated financial assets.”
- “Make an accounting policy election not to measure an allowance for credit losses on accrued interest receivable amounts if an entity writes off the uncollectible accrued interest receivable balance in a timely manner and makes certain disclosures.”
- “Make an accounting policy election to write off accrued interest amounts by reversing interest income or recognizing credit loss expense, or a combination of both. The entity also is required to make certain disclosures.”
- “Make an accounting policy election to present accrued interest receivable balances and the related allowance for credit losses for those accrued interest receivable balances separately from the associated financial assets on the balance sheet. If the accrued interest receivable balances and the related allowance for credit losses are not presented as a separate line item on the balance sheet, an entity should disclose the amount of accrued interest receivable balances and the related allowance for credit losses and where the balance is presented.”
- “Elect a practical expedient to disclose separately the total amount of accrued interest included in the amortized cost basis as a single balance to meet certain disclosure requirements.”
Recoveries should be included when “estimating the allowance for credit losses.” The ASU also clarifies that “expected recoveries of amounts previously written off and expected to be written off should be included in the valuation account and should not exceed the aggregate of amounts previously written off and expected to be written off by the entity. In addition, for collateral dependent financial assets, the ASU clarifies that an allowance for credit losses that is added to the amortized cost basis of the financial asset(s) should not exceed amounts previously written off.”
The ASU clarifies the guidance “by specifically requiring that an entity consider the estimated costs to sell if it intends to sell rather than operate the collateral when the entity determines that foreclosure on a financial asset is probable.” Additionally, the ASU clarifies that “when an entity adjusts the fair value of collateral for the estimated costs to sell, the estimated costs to sell should be undiscounted if the entity intends to sell rather than operate the collateral.”
The ASU clarifies “that an entity should consider extension or renewal options that are included in the original or modified contract at the reporting date and are not unconditionally cancellable by the entity.”
ASU 2019-05 clarifications
ASU 2019-05 provides entities that have certain instruments within the scope of Subtopic 326-20 with an option to irrevocably elect the fair value option in Subtopic 825-10, Financial Instruments—Overall, applied on an instrument-by-instrument basis for eligible instruments, upon adoption of Topic 326. The fair value option election does not apply to held-to-maturity debt securities. An entity that elects the fair value option should subsequently apply the guidance in Subtopics 820-10, Fair Value Measurement—Overall, and 825-10, Financial Instruments—Overall.
Community institutions must take a close look at data
Before you begin modeling, it is critical to first understand the general limitations and data issues that we have seen many community institutions discover after beginning implementation.
Renewal date data must be properly coded
The most pressing data issue we have seen among clients is how an institution documents loan renewals. Calculation of lifetime losses over a contractual term is defined by the standard as:
“An entity shall estimate expected credit losses over the contractual term of the financial asset(s) when using the methods in accordance with paragraph 326-20-30-5. An entity shall consider prepayments as a separate input in the method or prepayments may be embedded in the credit loss information in accordance with paragraph 326-20-30-5. An entity shall consider estimated prepayments in the future principal and interest cash flows when utilizing a method in accordance with paragraph 326-20-30-4. An entity shall not extend the contractual term for expected extensions, renewals, and modifications unless it has a reasonable expectation at the reporting date that it will execute a troubled debt restructuring with the borrower.”
The key statement is that extensions or renewals (other than those at the behest of the borrower as outlined in ASU 2019-04) are not to be included in the contractual term. In working with community institutions around the country, it is common practice that when a loan reaches its maturity and the institution extends it, only the maturity date is updated in the core system; no other data is maintained to flag that a renewal occurred. This becomes an issue when accumulating data for CECL, as when a renewal happens at contractual maturity, you effectively have a new loan.
Let’s look at an example: An institution originates a loan on December 31, 2019, on a 20-year amortization with a balloon payment on December 31, 2024, which is also the stated maturity date on the signed loan document. If the institution renews the loan on December 31, 2024, for another five-year term, and takes a charge-off on December 31, 2025, this charge-off relates to the loan originated on December 31, 2024, and not to the December 31, 2019 loan, as under CECL, the 2019 loan was paid off.
Therefore, institutions should begin to take steps to address how renewals are being coded in the system to address this issue, as some of the methods will rely heavily on these data points.
Limited data make a community institution’s loss curve less predictable
The other issue we have noted in modeling efforts is that community institutions do not have predictable loss curves, which stems from the fact that these institutions do not have a large enough pool of data points (i.e., both number of loans and charge-offs) to generate a mathematically/statistically relevant result.
Below is a loss curve (i.e., net charge-offs) for a large national institution:
As you can see, this is a very predictable pattern. A model can be built that is mathematically supportable when you have a predictable and consistent charge-off curve.
Now let’s look at a community institution:
As you can see, there is no pattern to these charge-offs. Community institutions typically take several large charge-offs as opposed to a large number of small charge-offs. This is particularly true for community institutions primarily lending to commercial customers.
Oftentimes, community institutions do not have enough charge-offs or loans to mathematically support a lifetime loss calculation. Thus, qualitative/forecasting adjustments are going to be a critical part of CECL analysis, as a community institution loss curve is less predictable than that of a large national institution.
Mapping charge-off data to economic data
Since the loss curve for a community institution is generally less predictable, it is challenging to quantify qualitative adjustments. For example, the chart below shows the correlation between the national unemployment rate and the net charge-off rate for all banks in the United States.
Using a regression analysis in Microsoft Excel, these data tell you that there is an 87 percent correlation between these two data points (i.e., if you know the national unemployment rate, you can say with 87 percent certainty what the nationwide net charge-off rate will be). This also tells you on a linear basis that for every 100 basis point move in the unemployment rate, the annualized net charge-off rate moves approximately 38 basis points.
This type of analysis can be done because you are dealing with two consistent curves with thousands of data points. But when you try to accomplish this analysis for an individual institution, the results may show no correlation at all. This should weigh heavily in the method an institution selects, as a more sophisticated methodology may not necessarily produce more accurate results because the results are heavily predicated on the consistency of the loss curve.
Grouping loans of similar risk characteristics
Institutions should start with their charge-offs and work backwards to determine if correlations can be established in order to justify extrapolating charge-offs to a broader portfolio. This is a challenge for community institutions, as usually charge-off rates are a result of what happens on a handful of loans, as opposed to a larger population of loans.
For example, in one institution we worked with, a loan group had three loans in it, and one loan was charged off in full, resulting in a 30 percent lifetime loss rate. This was going to be extrapolated to the current loan group in which the average loan-to-value was 50 percent. The institution justified not using the 30 percent loss rate, as it was not reflective of the current portfolio. Community institutions need to be cognizant of this issue; otherwise, you may end up extrapolating a charge-off on one loan to a large population, which is not statistically supportable.
Modeling methods break down into two categories
Since CECL was not prescriptive in regards to how the credit loss computation should be done, there are numerous modeling methods that are available for use. In general, there are two categories in which you can divide the methods:
1. Use the annual loss rate and extrapolate this loss rate over the lifetime of the loan
This method aligns with the weighted average remaining maturity method and is the closest to the current allowance methodology, with the exception of incorporating loan terms into the calculation. It extrapolates an entity’s annual loss rate over the expected remaining term.
The pros of this method include:
- The calculation is simple and easy to understand
- Limited data points are needed
- There is no need to harvest loan by loan data for multiple time periods
- It most closely resembles current calculation
The cons of this method include:
- It is heavily reliant on qualitative factors and forecast adjustments
- It can be very volatile based on annual loss rates
- It does not mathematically factor in vintage information of loan portfolio
2. Compute a lifetime loss rate and multiply this rate by the current principal
Other methods include, but are not limited to:
- Probability of default/loss given default — Computes two ratios (percent of loans in a portfolio that default, and percent of net charge-offs of the defaulted loans) and multiplies them together
- Snapshot/cohort/open pool —Takes a loan portfolio at a point in time and tracks net charge-offs from that point forward to compute a lifetime loss rate. This pools can be bifurcated by any number of different metrics.
- Migration — Tracks changes in credit criteria over time to compute a lifetime loss rate based on loan type and a credit indicator
- Vintage — Tracks losses on loans originated in the same year to compute the lifetime loss rate
The pros of these methods include, but are not limited to:
- Loan-by loan-level detail is needed, which could prove more accurate if correlations between charge-offs and loan types can be established
- Institutions can apply loss rates more granularly, which could allow for applying higher loss rates to problem loans with lower loss rates to pass-rated credits
- Institutions can factor in the vintage information of a loan portfolio
The cons of these methods include, but are not limited to:
- Institutions need to harvest loan-by-loan data for multiple periods, and more data points for each loan
- It is difficult to factor in the actual contractual terms given renewals, depending on how loan data are maintained
- Depending on the size of the portfolio, Microsoft Excel may lack the processing power necessary to successfully execute one of these methods
How we can help
At CLA, we believe in making the complex simple for community institutions. Our professionals are well-versed in the CECL standard, so we can anticipate the limitations you will face and help you select an appropriate method for the complexity of your institution.