CECL allowance model support for community lenders, credit unions and fintech lenders

The revised Interagency Policy Statement on Allowances for Credit Losses was published in the Federal Register on April 27, 2023, and takes effect for each institution when it adopts ASC 326. The NCUA joined the OCC, the Federal Reserve and the FDIC in issuing it, so credit unions are inside its scope.

It covers CECL measurement for loans held for investment, net investments in leases, held-to-maturity debt securities and off-balance-sheet credit exposures, plus the allowance for available-for-sale debt securities under ASC 326-30. On method, it states that ASC 326 does not require a specific loss estimation method and lists loss-rate, probability of default and loss given default, roll-rate, discounted cash flow and aging methods.

On governance, it expects written policies, defined roles, reliable data, internal controls and prompt write-offs. The board oversees significant judgments and reviews the estimate. Management compares estimates with actual charge-offs and validates the estimation process. The statement scales those expectations to an institution's size, complexity and risk, which is why a small lender and a large bank both answer to it with documentation of very different weight.

Footnote 12 of the policy statement groups the weighted-average remaining maturity method, vintage analysis and the snapshot or open pool method under loss-rate approaches. The table sets out what each one needs and where it tends to strain.

ASC 326-20-30-3 lets an entity use any of these without reconciling to a discounted cash flow method. The model documentation therefore has to show why the chosen method fits the pool, since "we have always done it this way" is not support for a method choice.

Historical loss information is the starting point, and ASC 326-20-30-8 and 30-9 require adjustments where the past does not describe the present. Differences in underwriting standards, portfolio mix or asset term count, and so do changes in conditions and forecasts, such as unemployment rates, property values, commodity values and delinquency.

Each adjustment gets one row in a register: the factor, the pool it applies to, the direction and size of the adjustment, the evidence behind it and the date it was last reviewed. An adjustment without evidence is the first thing a reviewer pulls out. Where the lender cannot support a reasonable forecast over the full contractual term, the model reverts to historical loss information beyond the forecast period, immediately, on a straight-line basis or using another rational and systematic basis, as paragraph 30-9 allows.

Loans that stop sharing a pool's risk characteristics move to individual evaluation. Collateral-dependent loans, purchased credit-deteriorated assets and unfunded commitments each have their own treatment in the model, so the pools tie back to the general ledger without leftovers.

The board does not need a model walkthrough. It needs enough to challenge the number. The quarterly package Finbryn prepares for management contains:

1. The allowance by pool against the prior quarter, with the movement split into volume, loss rates and qualitative changes.
2. A roll-forward from the opening allowance through provision, charge-offs and recoveries to the closing balance.
3. A back-test comparing what the model predicted with what was actually charged off.
4. Sensitivity to the two or three assumptions that move the number most.
5. A list of open items: data exceptions, model limitations and changes made since the last meeting.

Management reads it first, edits it, and presents it. The package carries no recommendation about capital or dividends, and it does not substitute for the institution's own credit committee minutes.

The policy statement expects management to validate the loss estimation process and any changes to it, and describes review by a party with appropriate knowledge, technical expertise and experience who is independent of the credit approval and allowance estimation processes. It also says management should evaluate models before they are used.

That language is why Finbryn does not validate the models it builds. The builder cannot be the independent reviewer. We write the documentation, keep the inputs and calculations traceable and answer the reviewer's questions, and the institution engages the reviewer. Depending on size, that may be an internal reviewer outside the credit function or an external model validation firm.

The same logic applies to lenders that no agency supervises. A credit facility agreement or an investor may ask for a review by someone other than the builder, and the documentation is written to the same standard.