Blog · September 30, 2026
The Average Bank Is Fine. The Typical Bank Isn't.
Both are also fine.
Noncurrent loans (90+ days past due plus nonaccrual) as a share of total loans, deposit-taking U.S. banks, 1Q2019 to 2Q2026. Tap or hover any chart for the numbers.
The short version
Add up every dollar of loans that are 90+ days late or stopped accruing interest, divide by all loans, and you get 0.93%. That is the same number it was two years ago, and it matches the FDIC's. Flat. Nothing to see.
Now line up the banks and look at the one in the middle. Its rate went from 0.28% to 0.47%. That is up about 70% since mid-2022.
Same country, same banks, same quarter. One number says calm. The other says the typical bank is getting a little more nervous. That gap is the whole story, and it is why a headline number can be true and still miss what is going on at the bank down the street.
Mean vs. median, literally.
MeanZero point nine three. Same as two years ago. Next slide.
MedianSay that at the bank down the street.
MeanI've never been down the street. I'm dollar-weighted.
Three ways to average one industry
Show the numbers
Why the two numbers disagree
The aggregate is dollar-weighted. A handful of very large banks hold most of the loans, so they set that number. The median counts every bank once, so the 2,687 banks between $100 million and $1 billion are 64% of all banks and the $100B+ giants are under 1%.
Last year the big banks got a little better, which held the total flat. Over the last four quarters the aggregate actually fell by about 3 hundredths of a percentage point, while the typical bank kept drifting up.
It isn't a few bad apples
Take the 4,173 banks that exist at both dates. 58% got worse, 36% got better, 6% didn't move. That is a broad drift, not a few blow-ups.
It is in every size group. The median bank rose in all of them.
It moved in every size group
Show the numbers
| Size (in 2022) | Banks | Median, 2Q22 to 2Q26 | Share of banks worse |
|---|---|---|---|
| Under $100M | 634 | 0.25% to 0.47% | 47% |
| $100M to $1B | 2,687 | 0.27% to 0.42% | 57% |
| $1B to $10B | 716 | 0.29% to 0.54% | 70% |
| $10B to $50B | 95 | 0.37% to 0.65% | 75% |
| $50B to $100B | 12 | 0.55% to 0.92% | 92% |
| $100B and up | 29 | 0.71% to 0.85% | 55% |
Small banks carry the count. But the pace is fastest at $1 to $50 billion: the typical bank there is up 23 to 28 hundredths of a point and 70–75% of them are worse. (The $50B+ groups are a dozen and 29 banks we could follow from 2022, so don't hang a hat on them.)
MedianEvery size group went up. Every one.
MeanThe $50–100 billion banks went up thirty-seven basis points.
MedianThe twelve that were there in 2022. There are nineteen now.
MeanNineteen very large banks.
What's driving it
Commercial real estate first, then business loans, then homes. About 40% of the rise in the typical bank comes from commercial real estate (57% if you count owner-occupied buildings), about 20% from business (C&I) loans, and about 13% from home loans. Credit cards, car loans and farm loans are not what is moving the typical bank.
Roughly: put commercial real estate back where it was in 2022 and the typical bank's number falls 8 of the 19 hundredths. Do it for business loans and it falls another 4.
What's driving the typical bank
In dollars, it is a different list. The aggregate is set by big banks: non-owner commercial real estate and credit cards at the very largest banks are the two biggest dollar increases since 2022. Credit cards barely register at the typical bank.
In dollars, it's a different list
MeanSee? In dollars it's the big banks: commercial real estate and credit cards. That's my number.
MedianThen your big banks got better and you called it flat.
MeanI call it stable. Shareholders love stable.
MedianWe don't have shareholders. I made sure of that.
Is it a crisis?
No. Look at the dates. The typical bank bottomed out at 0.24% in early 2023 and is now at 0.47%. In mid-2019 it was 0.55%. The unweighted average, 0.96%, is also still just under 2019's 1.00%.
This is normalization: a very clean stretch coming back to ordinary. It is not a break. The ones to watch are the same as the ones that moved: commercial real estate and business loans at mid-sized banks.
MeanSo it's a crisis.
MedianIt's 2019.
Mean2019 was fine.
MedianMostly.
Who is stressed: the securities-heavy or the loan-heavy?
Both, in different ways.
- Banks loaded up on securities have the weaker margins. Median net interest margin 3.68% vs. 4.14% for banks light on securities. They are the banks sitting on big unrealized bond losses, about 17% of equity at the median.
- Banks loaded up on loans have the credit problems. Aggregate noncurrent rate 1.16% for the most loan-heavy third vs. 0.83% for the least.
Since 2022 the typical securities-light bank got worse on credit by 25 hundredths of a point; the typical securities-heavy bank by 10. Margins recovered in both (bond-heavy +77 hundredths, bond-light +53) but the bond-heavy are still 46 hundredths behind. In other words: the bond-heavy banks have an earnings problem and the loan-heavy banks have a credit problem. Different diseases.
Two different diseases
New words in the Call Report
Banks have to fill in a few free-text lines on the Call Report. We read all of them, seven years' worth, and looked at what showed up for the first time in the last four quarters. We are not naming names. Trends only:
- “Stablecoin” is now in the Call Report. It first appeared in the fourth quarter of 2025 and two banks used it by mid-2026.
- “Seasoned” spread. 21 banks started using it from the fourth quarter of 2025 on; the example wording is an allowance on purchased seasoned loans. Probably a new reporting line or accounting treatment, but we have not confirmed which.
- Otherwise: about 3,600 new words and phrases, only 54 of them used by more than one bank. Most are one bank's private spelling for an ordinary line item. One bank spelled “miscellaneous” four different ways. We counted each as a new word. Mean says that's rigorous.
What we don't know
We measure late loans, not losses. Rates are on loans that already went bad, so this looks backward. A 0.47% median isn't a warning by itself; it is a level. The signal is direction and breadth, and both point the same way: up, and widely.
MeanOn average.