The Second Squeeze

The effective federal funds rate was 3.88% on 2026-09-30.

On 2026-09-30 the 2-year Treasury stood at 4.88% and the 10-year at 5.29%, a slope of 41 basis points.

The closest precedent is 2022–23, when the effective funds rate rose 5.21 points (quarterly averages, 2022Q1 to 2023Q4) and the 2s10s spread went from 49 basis points to −77 at its low in 2023Q3.

On the same banks at both ends of each phase, margins changed by under $1B +0.56, $1–10B +0.59, $10–100B +0.77 and over $100B +1.15 points in 2022 (2022Q1 to 2022Q4), and by under $1B −0.23, $1–10B −0.40, $10–100B −0.28 and over $100B −0.34 points in 2023 (2022Q4 to 2023Q4), when the curve was inverted. Ordered from the larger decline in 2023: $1–10B, over $100B, $10–100B and under $1B. These are point estimates for each size group; the report does not test the differences between groups.

The size groups also differ in balance-sheet shape. In 2026Q2 the median bank's non-owner-occupied CRE, relative to its total capital, was 224% for $1–10B banks, 219% for $10–100B banks and 64% for banks over $100B; cash and Treasuries were 2.6%, 3.8% and 8.6% of assets. The report does not measure loan repricing directly.

From 2024Q3 to 2026Q2, as Fed funds fell 1.63 points, the same banks' margins changed by under $1B +0.50, $1–10B +0.41, $10–100B +0.29 and over $100B +0.04 points. Single-quarter down-cycle deposit betas were under $1B 0.25, $1–10B 0.38, $10–100B 0.47 and over $100B 0.40 (Exhibit 4).

On the Federal Reserve's national index, commercial property prices dropped 13.4% between 2023Q2 and 2025Q2, then regained 8.8% by 2026Q2.

Cap rates and the 10-year Treasury

A cap rate is a property's net operating income divided by its value, so at constant income a higher cap rate means a lower value. National cap rates have risen from their 2021–22 lows in each of the five property types in the series: apartments from 4.40% (2022Q2) to 5.86% in August 2026, office from 6.06% (2022Q1) to 7.55%. Holding income constant, those moves alone correspond to value changes of −25% and −20%. The price index in Exhibit 5 and this cap-rate arithmetic differ in construction and property mix, so the two are not directly comparable.

Against the 10-year Treasury's August 2026 average of 4.68%, apartment cap rates were 1.18 points higher and office 2.87 points, compared with 2015–19 averages of 3.25 and 4.28 points. Both margins were below every 2015–19 reading for their type; on this basis only September 2022 and September 2023 were narrower. At constant income, a further 50 basis points of cap-rate expansion would lower values by another 6% to 8% across the five property types (Exhibit 5b).

Noncurrent non-owner-occupied CRE loans, on the same banks since 2022Q1: over $100B: peaked in 2024Q3 at 2.46%, 1.86% in 2026Q2; $10–100B: peaked in 2025Q3 at 1.23%, 1.13% in 2026Q2; $1–10B: peaked in 2026Q1 at 0.90%, 0.85% in 2026Q2; under $1B: at its peak for the window in 2026Q2 (0.88%).

By category, among banks holding each type of loan in both 2022Q4 and 2026Q2, noncurrent rates went: multifamily from 0.18% to 1.09%, construction from 0.45% to 1.09%, other investor property from 0.57% to 1.40% and owner-occupied CRE from 0.54% to 0.91%. Other investor property peaked in 2024Q3 at 1.86%. For owner-occupied CRE, on banks holding it in both 2019Q4 and 2026Q2, the rate was 0.85% and 0.91%.

The supervisory screen

The 2006 interagency guidance on CRE concentrations expects stronger risk management at banks whose construction and land loans reach 100% of total capital, or whose CRE loans reach 300% of total capital after growing 50% or more over the prior three years. For that test, CRE means construction, multifamily, non-owner-occupied nonresidential property and unsecured loans to finance CRE; owner-occupied CRE and farmland are excluded. The thresholds identify banks for closer risk-management attention; they are not limits.

In 2026Q2 the count of banks whose non-owner-occupied CRE was 300% of total capital or more was 191 of 3,242 banks under $1B, 178 of 895 banks $1–10B, 20 of 127 banks $10–100B and 0 of 32 banks over $100B. Of the $1–10B banks at that level, 48 had also grown that book by 50% or more over three years, satisfying the guidance's growth condition as well. The counts cover all banks in each size group, specialty business models included, against an absolute threshold; banks without the ratio count in the total.

Among $1–10B banks below 1% noncurrent in 2022Q4, 31.4% of those in the 300%-or-more tier had reached 1% by 2026Q2, against 20.3% of other banks (two-proportion test, adjusted p 0.041). The same comparison showed no statistically significant difference at under $1B and $10–100B banks.

Multifamily

201 banks hold multifamily loans at or above 100% of total capital. By concentration tier, the share of multifamily lenders reaching 1% noncurrent showed no statistically significant difference in any size group.

Does the definition matter? The non-owner-occupied and total-CRE concentration measures are correlated at 0.94 across banks in 2026Q2, and aggregate noncurrent rates were 1.18% for total CRE and 1.26% for non-owner-occupied CRE. On the total-CRE measure (total CRE over total capital as reported in the UBPR; absolute tiers below 200%, 200–400% and 400% or more, fixed at 2022Q4, among banks with total CRE of at least 10% of loans) the tier comparison showed no statistically significant difference in any size group. The supervisory definition remains the primary measure.

CRE concentration and the 2022–23 pass-through gap (asset-yield minus funding pass-through) have Spearman rank correlations of $1–10B −0.21 (n 768, adjusted p below 0.001) and $10–100B −0.41 (n 98, adjusted p below 0.001): banks with more CRE relative to capital tended to have a lower gap. This is an association; the report does not measure the mechanism.

Unrealized gains and losses on securities, as a share of amortized cost, on the same banks since 2021Q4: under $1B −12.1% in 2023Q3 and −5.7% in 2026Q2, $1–10B −12.6% in 2023Q3 and −5.7% in 2026Q2, $10–100B −11.8% in 2023Q3 and −4.9% in 2026Q2 and over $100B −12.2% in 2023Q3 and −5.6% in 2026Q2. Per point on the end-of-quarter 10-year yield, the quarterly change in that share was under $1B −4.25 (correlation −0.92), $1–10B −4.34 (correlation −0.95), $10–100B −4.41 (correlation −0.96) and over $100B −4.24 (correlation −0.96), over 2021Q4–2026Q2.

The securities estimate assumes that the historical relationship between yields and portfolio marks remains representative; changes in portfolio maturity composition could alter realized sensitivity.

On the same banks at both dates, non-interest-bearing deposits went from under $1B 22.7% to 18.0%, $1–10B 23.1% to 17.9%, $10–100B 21.7% to 17.7% and over $100B 19.9% to 15.0% of assets. Median Tier 1 leverage ratios, 2022Q4 and 2026Q2, were under $1B 10.48% and 11.15%, $1–10B 9.81% and 10.63%, $10–100B 9.76% and 10.20% and over $100B 8.73% and 9.39%.

Rising rates move deposits in two ways. The gradual one: as Treasury bills and money-market funds pay more, rate-sensitive depositors move unless their bank matches the yield. The sudden one: uninsured depositors who doubt a bank's solvency leave at once, as they did in March 2023.

Exhibit 10a covers only banks that came through the period, so it leaves out the sudden kind. Uninsured deposits were 73% of total assets at Silicon Valley Bank at the end of 2022, 72% at Signature Bank (New York) and 56% at First Republic Bank. All three failed in the spring of 2023.

Among banks that came through the period, from 2022Q1 to 2023Q4: $1–10B banks in the top quarter for uninsured deposits had a median deposit change of +0.6%, against +3.3% in the bottom quarter (adjusted p 0.020); $10–100B banks in the top quarter for uninsured deposits had a median deposit change of −4.1%, against +4.4% in the bottom quarter (adjusted p 0.020). Two-sided Mann–Whitney tests.

What the Run Vulnerability Index showed. BankAtlas's Run Vulnerability Index (RVI) ranks each bank's funding structure and liquidity against banks of its size in the same quarter; a higher percentile means more structurally exposed to a run. Being peer-relative, it ranks banks within a quarter rather than measuring risk through time. In point-in-time historical testing that applies the current RVI method to end-2022 reports, among the 34 banks with assets over $100B (every bank in that group had an RVI), Silicon Valley Bank ranked at the 76.5th percentile and First Republic Bank at the 82.4th, in RVI's amber zone (above the 75th percentile); Signature Bank (New York) ranked at the 67.6th, below the amber zone. RVI measures funding structure; it does not measure concentrations of uninsured deposits, the funding that ran in 2023, which is why this report tracks uninsured deposits separately.

The gap between Fed funds and banks' average deposit cost (interest-bearing deposits) was 2.43 points in 2022Q4 and 1.39 points in 2026Q2. Uninsured deposits funded between 28% and 34% of assets across the size groups over $1B in 2026Q2.

Does today's system carry the 2023 pattern?

On end-2022 reports: Silicon Valley Bank was in the top fifth of its size peers for uninsured deposits and in the worst fifth for securities losses relative to equity; Signature Bank (New York) was in the top fifth of its size peers for uninsured deposits and at or below the peer median for cash and Treasuries; First Republic Bank was in the top fifth of its size peers for uninsured deposits and at or below the peer median for cash and Treasuries. The cash-and-Treasuries measure uses carrying values, so it does not reflect a loss that selling Treasuries would crystallize.

A screen requiring all three features would have matched none of the failed banks at the end of 2022; it matched 19 other banks, none of which failed. Requiring heavy uninsured funding plus either of the other two would have matched all three failed banks along with 102 others. That second rule is drawn from the 2023 failures rather than tested on them, so it describes the 2023 pattern; it is not a tested screen.

On June 2026 reports, 103 banks over $1B show heavy uninsured funding together with one or both of the other two features.

Counting the 6 rows of Exhibit 11, the number on which each group ranks first is: under $1B 1, $1–10B 4, $10–100B 0 and over $100B 1. No group ranks first on all of them, which is why the summary reports each channel separately.

Every bank is screened on six exposures as of June 2026. Size groups use total assets; peer-relative flags compare a bank with its own size group, so each flags about a fifth of a group by construction and is not used to compare groups.

• Margin: from 2022Q1 to 2023Q4, the rise in asset yield less the rise in interest expense, both relative to earning assets and divided by the rise in Fed funds (single-quarter UBPR ratios); flagged in the bottom fifth of the bank's size group (peer-relative). Scored only where both quarters are reported and total assets stayed within 0.67 to 1.5 times over 2022Q1–2023Q4 and 2023Q4–2026Q2; otherwise not scored.
• CRE concentration: flagged when the UBPR's 2026Q2 ratio of non-owner-occupied CRE to total capital is 300% or more, or its ratio of construction and land loans to total capital is 100% or more (absolute thresholds; the guidance's growth condition is not applied). Banks without the ratio are not measured.
• CRE credit: among banks with non-owner-occupied CRE of at least 10% of loans, the noncurrent rate on that book (90+ days past due or nonaccrual) in the top fifth of the size group and above zero, 2026Q2 (peer-relative). Banks below 10% are not eligible; banks without the rate are not measured.
• Securities: unrealized losses on held-to-maturity and available-for-sale securities relative to tangible equity before securities marks, in the worst fifth of the size group, 2026Q2 (peer-relative). Banks without the inputs are not measured.
• Funding: uninsured deposits relative to assets in the top fifth of the size group, together with cash and Treasuries at or below the group median, 2026Q2 (peer-relative). Banks under $1B usually do not report uninsured deposits; banks without either input are not measured.
• Capital: Tier 1 leverage ratio in the bottom fifth of the size group, 2026Q2 (peer-relative). Banks without the ratio are not measured.

Excluded as specialty business models: banks reporting no deposit cost, card banks (cards at least 50% of loans), banks with a margin of 7% or more, banks with brokered deposits at least 50% of deposits, and captive banks (no non-interest-bearing deposits and uninsured deposits under 2%).

Share of banks with three or more exposures, against a coverage-aware benchmark for six unrelated exposures: 7.8% and 7.5% at $1–10B; 7.6% and 6.4% at $10–100B. For each bank the benchmark counts only those exposures that bank can be scored on.

Tested pairings (two-sided Fisher exact tests, Benjamini–Hochberg across the 12 pair tests; pairs with fewer than 20 banks showing both flags are not reported): at $1–10B banks, a squeezed 2022–23 margin and CRE concentration appear together more often than chance (lift 1.45, adjusted p 0.005); at $1–10B banks, CRE concentration and large securities losses appear together less often than chance (lift 0.52, adjusted p below 0.001).

• Single-quarter deposit costs at $1–100B banks in 2026Q3.
• Securities marks against the model estimates and intervals in Exhibit 9.
• Noncurrent investor CRE on the same-bank paths in Exhibit 6.
• Non-interest-bearing deposits by size group.

The bank figures are charter-level Call Report and FFIEC Uniform Bank Performance Report data for 2019Q4 to 2026Q2, taken from BankAtlas's accepted 2026Q2 panel. Every comparison over time uses the same banks at each date, with the size group (and, where used, the concentration tier) fixed at the start; counts of banks are shown. Margin components use single-quarter annualized ratios. Deposit cost is interest expense on deposits over average interest-bearing deposits, aggregated across banks. Deposit betas use the same banks in every quarter from 2021Q4 to 2026Q2. Statistical tests are preregistered and two-sided, with Benjamini–Hochberg adjustment within each family; group differences without a test are shown as point estimates only. Interest rates and the CRE price index are Federal Reserve series, obtained from FRED. The Run Vulnerability Index is a BankAtlas measure. The report uses 2022–23 as its reference episode. Balance sheets have changed since then, and hedging or repositioning is captured only to the extent it shows in reported ratios.