Short answer: At June 30, 2026, the median bank carried an allowance for credit losses equal to 1.19% of total loans and leases, and the middle half of banks ran from 0.98% to 1.44%. That is a tight distribution: the interquartile spread is 46 basis points, and the medians of all seven asset bands fall between 1.14% and 1.44%. Almost all of the variation is among banks of the same size, and most of it is loan mix rather than credit quality. The bands with the most nonperforming loans do not carry the most reserve.
What the ratio measures, and what it leaves out
The ratio on this page is the allowance for credit losses on loans and leases divided by total loans and leases, both from the balance sheet at quarter end. The numerator is the contra-asset that sits against the loan book on Schedule RC; the denominator is total loans and leases from Schedule RC-C Part I. The quarter's movement in the allowance, provision expense, gross charge-offs and recoveries, is reported separately on Schedule RI-B Part II.
Two other allowances live outside this ratio and get mixed into it when bankers compare notes. The reserve for unfunded lending commitments is a liability, not a contra-asset, so it is outside the numerator even though it is measured under the same standard. The allowance on held-to-maturity debt securities is a third, separate balance. Investor presentations often combine the loan allowance with the unfunded commitment reserve and call the total coverage, so confirm the peer is quoting the same numerator before you tell a director yours is thinner.
The population here is the 4,205 banks that report both an allowance and a loan balance. Start from the 4,296 banks that filed for the quarter, set aside the 58 non-insured non-deposit trust companies that take no deposits and make no loans, and 4,238 remain in the benchmark population. The table below covers 99.2% of them. The 33 that drop out cannot produce the ratio at all, because the calculation requires both figures and a loan book to divide by. They are not banks with a low allowance being excluded from the low end of the distribution.
What CECL changed
The old allowance was an incurred-loss measure. A bank reserved when a loss was probable and already incurred at the measurement date, so the allowance trailed the cycle: smallest at the top of the market, when the loans being originated were the ones that would eventually default, and largest well after the losses arrived.
The current expected credit loss model removed the probable-and-incurred threshold. The allowance now estimates all credit losses expected over the remaining contractual life of the exposure, adjusted for prepayments, and records them at origination or purchase. A loan funded on the last day of the quarter carries a reserve that day, with a perfect borrower and no delinquency anywhere in the book. That day-one reserve is the largest behavioral change for a CFO, because it turns loan growth itself into provision expense.
Three consequences follow. The estimate is forward-looking, resting on a reasonable and supportable forecast over a period the bank chooses and documents, after which the model reverts to historical loss experience. Measurement is pooled by shared risk characteristics, with individually evaluated loans carved out when they no longer share them, typically collateral-dependent credits. And the standard reaches beyond loans to held-to-maturity securities and off-balance-sheet exposures, which is why the unfunded commitment reserve moves when the forecast moves.
None of that makes the resulting percentage comparable across banks. It makes it a function of what the bank holds and how long it holds it.
What coverage looks like today, by bank size
| Bank size (total assets) | Banks | Lower quartile | Median | Upper quartile |
|---|---|---|---|---|
| All banks | 4,205 | 0.98% | 1.19% | 1.44% |
| Under $100M | 524 | 0.95% | 1.24% | 1.66% |
| $100M to $300M | 1,208 | 0.98% | 1.18% | 1.48% |
| $300M to $1B | 1,429 | 1.00% | 1.20% | 1.43% |
| $1B to $3B | 624 | 0.96% | 1.15% | 1.35% |
| $3B to $10B | 263 | 0.95% | 1.14% | 1.34% |
| $10B to $100B | 125 | 0.97% | 1.14% | 1.34% |
| Over $100B | 32 | 0.62% | 1.44% | 1.74% |
Source: BankingLens, computed from FFIEC call reports for the quarter ending June 30, 2026. Quartiles are calculated across the full population of filing banks in each size band, not sampled. Non-insured non-deposit trust companies are left out of every statistic: they take no deposits and make no loans, so a margin or a funding cost computed for them has no meaning.
Read down the median column and almost nothing happens. From under $100M to $100B the median moves from 1.24% to 1.14%, a ten basis point drift across four orders of magnitude of asset size. Compare that with the efficiency ratio, which falls by roughly fifteen points across the same range. Scale buys operating leverage. It does not buy a lower reserve.
Read across instead and the picture changes. In the middle bands the interquartile spread is 37 to 43 basis points, wider than the entire 30 basis point range of medians across all seven bands. Knowing a bank's asset size tells you almost nothing about its allowance; the useful comparison is against banks holding the same kind of paper.
The two ends of the table are the widest and the most revealing. Over $100B, the lower quartile is 0.62% and the upper quartile 1.74%, a 112 basis point spread inside a band of just 32 banks. That band holds card issuers and consumer lenders alongside institutions whose balance sheets are mostly securities, custody and wholesale exposure. They do not differ in credit discipline by a factor of nearly three; they are in different businesses. Under $100M the spread is 71 basis points, the second widest, for a different reason: with few loans, one individually evaluated credit moves the whole ratio.
The ratio tracks mix, not credit quality
This is hard to say in a board meeting and easy to show in the data. If the allowance were a thermometer for credit quality, the bands with the most nonperforming loans would carry the most reserve. They do not.
| Bank size (total assets) | Median allowance to loans | Median nonperforming to loans | Ratio of the two medians |
|---|---|---|---|
| All banks | 1.19% | 0.47% | 253% |
| Under $100M | 1.24% | 0.46% | 270% |
| $100M to $300M | 1.18% | 0.38% | 311% |
| $300M to $1B | 1.20% | 0.43% | 279% |
| $1B to $3B | 1.15% | 0.48% | 240% |
| $3B to $10B | 1.14% | 0.60% | 190% |
| $10B to $100B | 1.14% | 0.66% | 173% |
| Over $100B | 1.44% | 0.85% | 169% |
Source: BankingLens, computed from FFIEC call reports for the quarter ending June 30, 2026. The final column divides one published median by the other. It is a ratio of medians, not the median of banks' own coverage ratios.
Banks with $10B to $100B in assets ran a median nonperforming ratio of 0.66% of loans, nearly double the 0.38% median for banks with $100M to $300M, yet reserved less: 1.14% against 1.18%. Across all seven bands the nonperforming median more than doubles, from 0.38% to 0.85%, while the allowance median moves within 30 basis points and not monotonically. Whatever the allowance responds to, it is not the nonaccrual loans already on the books.
Under a lifetime measure it should not be. Expected loss is loss content multiplied by how long the exposure survives, and loan mix drives both terms. An unsecured consumer or card book carries high lifetime loss content and reserves accordingly. A construction and land development book concentrates severity in a short window around completion and lease-up, which is why it reserves well above a stabilized commercial property. A first-lien residential book reserves lowest of the major categories at equivalent underwriting, because collateral coverage is high and defaults resolve with recovery rather than total loss.
Remaining life cuts partly the other way, and that subtlety separates a good CECL discussion from a bad one. The same residential book has a weighted average remaining life measured in years rather than quarters, so a low annual loss rate compounds over a long horizon, while a short commercial line matures before much of its lifetime risk accumulates. Two banks with identical historical loss rates can land 40 basis points apart on duration and prepayment assumptions alone, roughly the width of an entire interquartile range in the middle bands.
So the comparison an examiner or a director reaches for, banks of similar asset size, is the weakest one available. Build the peer set on concentration instead: if construction is a meaningful share of capital, the CRE concentration screens identify the banks whose allowance should look like yours, and building the peer group deliberately beats accepting the default.
Reading the allowance against nonperforming loans
The coverage ratio sets the allowance against nonperforming loans, : nonaccrual balances plus loans 90 days or more past due and still accruing. Both inputs here share a denominator, total loans, so coverage is one divided by the other. The median allowance of 1.19% against the median nonperforming ratio of 0.47% gives 253%.
Be careful with that number. It is a ratio of two medians computed over slightly different populations, 4,205 banks and 4,203 banks, and no individual bank has to sit at both. The median of banks' own coverage ratios is a different statistic from a far wider and more skewed distribution, because the denominator approaches zero at the clean end of the book: a bank with almost no nonaccrual loans posts coverage in the thousands of percent, which says nothing about the adequacy of its reserve. Treat 253% as an order-of-magnitude reference, not a percentile you can rank against.
Read directionally, the last column is still informative. Coverage falls steadily with size, from 270% under $100M to 169% over $100B, because the nonperforming median rises across the bands while the allowance median does not. Coverage below 100% is not automatically a shortfall: a nonaccrual loan secured by real estate with a current appraisal needs little specific reserve. It is a signal to test the appraisals behind the largest nonaccrual relationships, not to add a qualitative overlay. The Texas ratio answers the adjacent question, whether capital and reserves together cover the problem assets, and adds the foreclosed real estate that coverage ignores.
What a build costs in earnings and capital
Every change in the allowance runs through provision expense, then net income, then retained earnings and regulatory capital. Sizing it against the earnings and capital of a typical bank is what turns a modeling debate into a decision.
| All banks, June 30, 2026 | Banks | Lower quartile | Median | Upper quartile |
|---|---|---|---|---|
| Return on assets | 4,230 | 0.83% | 1.23% | 1.65% |
| Tier 1 leverage ratio | 4,238 | 9.70% | 10.96% | 13.00% |
Source: BankingLens, computed from FFIEC call reports for the quarter ending June 30, 2026. Quartiles are calculated across the full population of filing banks in each size band, not sampled. Non-insured non-deposit trust companies are left out of every statistic: they take no deposits and make no loans, so a margin or a funding cost computed for them has no meaning.
Take a bank at the lower quartile, 0.98% of loans, that concludes it belongs at the median of 1.19%. The build is 21 basis points of the loan book, recognized as provision in the quarter it is taken. Convert it with your own loans to assets before you present it, because the earnings and capital ratios are struck on assets: the median bank earns 1.23% of assets in a year, with the lower quartile at 0.83%, and carries a tier 1 leverage ratio of 10.96%, with the lower quartile at 9.70%.
For most balance sheets that build is a meaningful bite out of one quarter's earnings and a rounding error against a 10.96% leverage ratio. The asymmetry is why allowance debates get heated in the wrong room: the number that moves is the quarterly earnings line the board watches, and the number that matters for solvency barely registers. A bank at the lower quartile for capital should model the build against its own buffer rather than the industry median.
The reverse deserves the same discipline. Releasing reserve flatters ROA in the quarter it happens, and a bank whose ratio drifted from 1.19% toward 0.98% while the book grew into higher loss-content categories has booked earnings it will give back. Ask whether the ratio moved because the model moved or because the mix did.
What an examiner will ask
The interagency policy statement on allowances for credit losses expects the estimate to be documented, governed and supportable. In practice the questions are narrower than the guidance, and consistent.
- Why this method for this pool? Weighted average remaining maturity, vintage, discounted cash flow and probability-of-default approaches each imply different assumptions. Expect to explain why the method fits the pool's behavior, not why it was easiest to implement.
- How long is the reasonable and supportable forecast period, and what happens after it? The reversion technique and the point at which the model returns to historical experience are judgment calls. Both need to be written down and applied consistently across pools.
- What supports each qualitative adjustment? Every Q factor needs an evidence trail and a direction that matches it. An overlay that only ever increases, or that quietly absorbs the gap between the model output and the number management wanted, is the most common finding.
- How has the estimate back-tested? Prior-period allowances against realized charge-offs. Persistent one-directional error is a model problem, not a conservatism argument.
- Who owns the number? Independent review, approval authority and validation appropriate to the bank's size. Prepared by the same person who approves it is a governance finding whether or not the estimate is right.
- Does the unfunded commitment reserve move with utilization? A largely undrawn construction pipeline carries real exposure that never appears in the loan allowance.
- Why is your ratio where it is against peers? The answer is loan mix, remaining life and concentration, with a specific comparison. At 1.44% a bank sits at the upper quartile for all banks; at 0.98% it sits at the lower quartile. Either is defensible with a portfolio explanation and neither is defensible with an assertion about underwriting quality.
Bring the distribution to that last conversation rather than a single peer average. The call report supports the portfolio explanation directly: RC-C gives your mix by category, RI-B Part II the charge-off history the model is calibrated on, RC-N the nonperforming balances behind coverage. The UBPR adds a peer percentile, but on a ratio this dependent on mix the percentile is a prompt for the explanation, not a substitute for one.
Frequently asked questions
What is a normal allowance for credit losses to total loans ratio?
At June 30, 2026, the median bank carried an allowance for credit losses equal to 1.19% of total loans and leases, with the middle half between 0.98% and 1.44%, across the 4,205 banks that report both figures. Medians by asset size run in a narrow band, from 1.14% for banks with $3B to $100B in assets to 1.44% for banks over $100B.
How did CECL change the allowance for credit losses?
CECL replaced an incurred-loss allowance with a lifetime expected-loss allowance. Under the old model a bank reserved only once a loss was probable and had already been incurred. Under CECL the bank estimates all credit losses expected over the remaining contractual life of the loan and books them at origination, supported by a reasonable and supportable forecast that reverts to historical experience. The measure also covers held-to-maturity debt securities and, through a separate liability, unfunded lending commitments.
Why is my allowance higher than a bank my size?
Almost always loan mix and remaining life rather than credit quality. CECL measures lifetime expected loss, which is loss content multiplied by how long the exposure lasts, so a consumer or construction book reserves well above a first-lien residential book at identical underwriting. In the Q2 2026 data the medians across seven asset bands span only 1.14% to 1.44%, while the middle half of banks inside a single band is 37 to 50 basis points wide in the five bands between $100M and $100B. Most of the variation is among banks of the same size, not between sizes.
What is a good allowance coverage ratio of nonperforming loans?
Coverage is the allowance divided by nonaccrual loans plus loans 90 days or more past due and still accruing. There is no supervisory floor, and the right level depends on collateral behind the nonaccrual credits. For scale, the median allowance of 1.19% of loans set against the median nonperforming ratio of 0.47% of loans is about 253%, but that is a ratio of two medians and not the median of banks' own coverage ratios, which is a different and wider distribution.
Does a higher allowance mean worse credit quality?
Not across banks. Banks with $10B to $100B in assets had the second-highest median nonperforming ratio of the seven bands at 0.66% of loans, yet carried a lower median allowance, 1.14%, than banks with $100M to $300M in assets, which reserved 1.18% against a 0.38% nonperforming median. If the allowance tracked observed problems, those two rows would be reversed.
What will an examiner ask about the allowance?
How the estimation method fits the portfolio, how long the reasonable and supportable forecast period runs and how the model reverts after it, what evidence supports each qualitative adjustment and whether the direction matches the data, how the estimate has back-tested against realized charge-offs, who reviews and approves the number, whether the unfunded commitment reserve moves with utilization, and why the ratio sits where it does relative to peers. The last question is answered with loan mix and remaining life, not with an assertion about underwriting.
Where these numbers come from
Every BankingLens figure on this page is computed from FFIEC call reports for the quarter ending June 30, 2026, across every bank in the benchmark population rather than a sample. Nothing is modeled or estimated. You can see these ratios for a real bank, ranked against its FFIEC peer group, on the sample scorecard, or walk the peer comparison on the dashboard. Plans that cover your state or every bank are on pricing.