Running Out of Earnings Room: 36 Banks With Compressed and Compressing NIM

The 36-bank cohort distributes across the full size spectrum: 1 institution above $50B, 2 in the $10B–$50B band, 4 in $2B–$10B, 10 in $500M–$2B, 9 in $200M–$500M, and 10 below $200M. The geographic top five — Minnesota (5), Texas (3), New York (3), California (2), and Tennessee (2) — account for 41.7% of cohort membership, with Minnesota's overrepresentation notable given its modest share of the national charter count.

The NIM distribution within the cohort is severe by design: median 1.91%, p25 at 1.00%, and a floor at zero. By construction, no member exceeds 2.50%. Top-of-cohort names by asset size include Beacon Bank and Trust, Massachusetts (CERT 17798) at $23.2B with NIM of 2.21% and a 76bps YoY decline; Texas Exchange Bank, Texas (CERT 20099) at $4.5B with NIM of 2.33% and a 116bps decline; and Sumitomo Mitsui Trust Bank (U.S.A.) Limited, New Jersey (CERT 27054) at $3.8B with NIM of 2.41%. The community-scale tail includes Freedom Bank, Inc., West Virginia (CERT 15664), whose NIM of 2.37% reflects a 127bps YoY collapse — the steepest decline among named members. The cohort is not regulatory-flag-positive at the individual bank level; few of these institutions would surface on a conventional capital, asset quality, or concentration screen.

Precedent support for this cohort is deliberately narrow. Silicon Valley Bank (CERT 24735, failed March 2023) is the canonical case of NIM compression converging with rate-driven asset stress, but it is a partial fit here: the full SVB signature requires HTM portfolio composition and uninsured-deposit data that the v0.1 substrate does not yet load. SVB stands as the structural reference for what compressed earnings can produce when paired with an unhedged duration mismatch, with explicit acknowledgment that this cohort captures only the earnings-side component of that signature.

Signature Bank (CERT 57053, failed March 2023) is similarly a partial fit — earnings compression was observed in the quarters preceding failure, but the proximate failure mechanism was a deposit run rather than spread collapse. It is useful as a 'compression present but not causal' reference point.

Pattern matching at this corpus depth is illustrative, not statistically rigorous. The mechanism analysis below carries more analytical weight than the precedent reference for this cohort, and that imbalance is honest rather than incidental. When HTM and uninsured-deposit data land in the Phase 1.5 substrate, the IRRBB-overlay screen will reactivate as the more selective successor to this cohort definition.

The structural exposure works as follows. A net interest margin below 2.5% places an institution in the bottom quartile of the U.S. banking universe on the single most important measure of intermediation profitability. A 30bps-plus year-over-year decline indicates that the compression is active rather than steady-state — funding costs are repricing faster than asset yields, or asset yields are falling faster than funding costs can follow, or both. Either trajectory consumes the earnings buffer that absorbs credit losses, funds capital accretion, and supports balance-sheet flexibility under stress.

The cohort is defined entirely on earnings-statement metrics, which is both its analytical strength and its limitation. Strength: NIM compression is observable from publicly filed Call Report data with a roughly 45-day lag and requires no inference about hedge positions, deposit beta assumptions, or unrealized securities marks. Limitation: it does not isolate the cause. A bank with thin and thinning NIM may be a healthy floating-rate commercial lender riding through a curve inversion, a fixed-rate residential portfolio absorbing deposit cost catch-up, or a duration-mismatched balance sheet of the SVB archetype. The cohort is mechanism-agnostic at the screen level.

For supervisory framing, the relevant CAMELS engagement is the Earnings component directly — the cohort definition is, in effect, an earnings-trajectory screen. Liquidity becomes engaged only where compressed earnings interact with funding posture, which this screen does not measure. The 2006 Interagency Guidance on Concentrations in Commercial Real Estate Lending and similar concentration frameworks are not directly engaged at the cohort level. What this cohort surfaces is institutions with diminished capacity to absorb any subsequent shock — credit, funding, or operational — through retained earnings.

Engagement positioning. Klaros should treat this cohort as a prospecting tier rather than a triage tier. These are not institutions in acute distress; they are institutions whose strategic optionality is narrowing. The advisory conversation is about asset-mix repositioning, deposit-pricing discipline, hedge-program review, and capital-planning realism under continued spread compression. Community-scale members — particularly the 19 institutions below $500M in assets — are the natural fit for Klaros's strategic-advisory motion, where the value proposition is helping management and boards confront earnings trajectory before it becomes a CAMELS Earnings downgrade.

Examination preparation. For cohort members already in active examination cycles, Klaros's role is helping management articulate to examiners both the source of the compression and the management response. NIM trajectory is a question every examiner will ask; a board that has it framed will fare materially better than one that does not.

Cohort tracking. Klaros should monitor cohort membership quarter-over-quarter. Banks that exit this cohort by stabilizing NIM are interesting; banks that persist for three or more consecutive quarters merit closer attention; banks that persist while also acquiring funding-side or asset-quality flags become candidates for the IRRBB-overlay successor cohort once HTM data lands.

Substrate caveat. Klaros should treat this cohort as the v0.1 substitute for a more selective screen. The descriptive value is real; the precision will improve materially when HTM and uninsured-deposit data activate the IRRBB overlay.