The Current Expected Credit Losses (CECL) standard has been fully in effect across the banking industry for a number of years. Large SEC filers adopted it in 2020; community banks and most credit unions followed for fiscal years beginning after December 15, 2022.
FASB is in the midst of its formal post-implementation review, with a public roundtable held in May 2026 and a summary of findings expected later this year. For many community institutions, however, the core experience remains one of high ongoing cost with limited change in allowance outcomes for the portfolios that dominate their balance sheets.
Community banks consistently report that qualitative (“Q”) factor overlays equal or exceed base model output—and frequently represent the majority of the final allowance. Yet allowance coverage ratios for core portfolios showed little or no material movement at the 2023 adoption date.
The frequent observation is that CECL has delivered “a lot more cost for no change in the end result.” This is the conundrum: a standard intended to produce more forward-looking, decision-useful information has, as many smaller institutions contend, produced more processes without meaningfully different numbers. While others are simply doing much the same as they were before but face more intense scrutiny.
Three interrelated issues drive much of the friction.
First is a persistent misunderstanding of what “expected” means. In the accounting and risk-modeling sense, expected loss is a statistical concept—the mean or average outcome across a large number of similar exposures given historical experience, current conditions, and reasonable-and-supportable forecasts.
It is not a prediction that any particular loan will default. Bankers underwrite and price individual credits with the expectation of repayment. A core principle of the CECL standard is the recognition of the total risk in the loan portfolio. Requiring institutions to reserve for an “expected loss” as opposed to a statistical average loss rate creates a nomenclature disconnect that sits uncomfortably against day-to-day credit judgment. The language of the standard collides with the mindset of the banker, in that “we don’t make loans we expect will default”, where most would recognize that lending has inherent risk.
Second is the definition of “loss” for modeling purposes versus the practical risks bankers manage. Consider a $1 million real-estate loan with a 70% loan-to-value ratio. If the loan defaults and the property is taken into Other Real Estate (ORE), and the bank ultimately recovers principal after foreclosure costs, the net charge-off can be zero.
Historical loss data for that loan type therefore show low realized credit losses even though defaults occurred. When a model applies an expected default rate and then overlays recovery assumptions drawn from that same history, the quantitative output is often small. The model can look accurate against past charge-offs and defaults, but the accounting and bookkeeping practices fail to capture the consequences of the default itself: ORE holding costs, valuation volatility, liquidity drag, management distraction, and the risk that collateral values fall further before sale.
CECL measures credit losses against amortized cost after expected recoveries. Where collateral recoveries have historically been strong, the quantitative baseline stays low. That is why overlays frequently become large relative to the modeled number and why documentation around them draws intense audit and examination attention.
Third, many material losses arise from punctuated events that historical averages and reasonably supportable forecasts do not reliably capture. Sudden economic shocks, localized real-estate downturns, or idiosyncratic personal events such as illness, job loss, or divorce drive outcomes that aggregate loss rates smooth over – forces that are magnified for community banks. A model can produce a sensible mean loss rate under “normal” conditions and still miss the events that produce the largest charge-offs.
Because those outcomes are not reliably predictable from past data, management judgment and environmental considerations remain necessary. The standard anticipates this; it explicitly permits qualitative adjustments for conditions not reflected in the quantitative estimate. It does not, however, prescribe how to quantify or document those adjustments in a way that satisfies auditors. The result is a heavy process burden without a corresponding change in the final allowance for core portfolios.
Taken together, the framework asks institutions to reserve for a statistical expected loss that is net of recoveries, while the risks that matter most to bankers—the ones that turn a performing loan into a prolonged problem asset or that arrive from hard-to-forecast shocks—often sit outside the quantitative model and must be addressed judgmentally.
That gap between what models can reliably measure and what the institution needs to manage is a central reason implementation has felt more costly than beneficial for many community banks.
CECL’s intent was to recognize lifetime expected credit losses earlier and reduce reliance on subjectivity. In practice, the combination of a statistical definition of “expected” with an accounting definition of “loss” has muddied the waters. This disconnect is further amplified by the difficulty of quantifying idiosyncratic events and has left many community banks with elaborate processes, large qualitative overlays, and little net change in allowance levels for their core portfolios. The conundrum is real and the industry continues to engage FASB’s post-implementation review.