THE RENEWAL RATCHET

Lens As the house characterizes it, consensus reads liquidity as a stock — cash, unpledged securities, borrowing capacity — and funding cost as an average: blended deposit cost, deposit beta, NIM. Where transfer pricing is keyed to the marginal curve (EP2), a loan would be priced on a number with no published bank-level series.

Channel On the contract the house assumes (EP4), an index-priced revolver’s coupon moves with the index between renewals, not with one lender’s premium. At renewal, besides the spread, six levers can carry a premium: commitment size, collateral, covenants, deposit conditions, conversion into a government-guaranteed structure, non-renewal. Among the public surfaces this report maps, only commitment size is observed bank by bank — through unused C&I commitments, an all-C&I proxy; the others appear, if at all, in aggregate surveys or in program data that does not flag a renewal.

Switches On the house’s model of the decision sequence, which cohort banks do not disclose, the decisions that would set the SME outcome sit where SME credit is not the subject: the buffer floor is set in a forum whose test is survival; any charge on undrawn lines would inherit an outflow assumption written for the examiner; the lever would be chosen by the officer paid to keep the client.

Perimeter Regulation WW lists global systemically important BHCs, G-SIB depository institutions and Category II, III and qualifying Category IV institutions as subject to its liquidity standards (12 CFR 249.1, eCFR current to 2026-09-18) [Confirmed]. Whether the house cohort — US$5–100bn in T0 total assets (house cohort definition, as of 2026-06-30) [Range] [HOUSE-ANCHOR] — sits wholly outside that perimeter turns on each bank’s holding company, not on the cohort bound: its position against the Category IV asset line held open — dollar threshold omitted pending resolution of its basis, its weighted short-term wholesale funding, and any Board determination.

Applicant side 2025 Small Business Credit Survey (Federal Reserve Banks, report published March 2026), employer firms, prior 12 months (approximately Sep–Nov 2024 to Sep–Nov 2025), pending applications excluded: business line of credit applicants were 45% fully approved, 31% partially approved and 24% denied (N=977) [Confirmed].

Blind cut The same survey splits banks at US$10B in total assets [Confirmed] — a definitional line, applied by respondents classifying their lender with the help of a list, that falls inside the house cohort’s range. In the report pages read, lender-type outcomes appear only pooled across products, and no line-of-credit-by-lender-type table appears there.

House test First-pass separating test on 116 banks × 10 quarters (T0 2023-03-31; evaluation window 2024-03-31 to 2026-06-30) [Range] [HOUSE-ANCHOR]: funding-cost slope (year-on-year Δlog per 1 percentage point of annualized interest-bearing deposit cost) on unused C&I commitments -0.00591 (SE 0.02357, p 0.802); on the real-estate-secured CRE placebo +0.01402 (SE 0.03504, p 0.689) — house computation, as of 2026-06-30 [Range] [HOUSE-ANCHOR]. Verdict: UNIDENTIFIED, on the tercile cell table [Range] [HOUSE-ANCHOR]. Every first-pass result is carried subject to an open check of its inputs’ reporting basis (Risk Box 3). The commitment-size surface is quiet; the funding channel stands as a specified test, not a finding.

Pricing object At the cohort: the commitment run-off and loan-growth line, and the new-production spread over marginal cost. On the borrower side: the refinancing-risk component of SME credit spreads in guaranteed-pool pricing, SME securitizations and private-credit SME books. No instrument, no direction.

The Cost-Blind Buffer (M3): the house’s model of a measurement convention that counts liquidity at availability and leaves its price outside the ratio.

Liquidity carries two meanings that, on the house’s reading, the consensus folds into one. Availability is a stock: cash, unpledged securities and the contingent capacity of collateral pre-positioned at the Federal Home Loan Banks and the discount window, set against uninsured and short-tenor liabilities. Cost is the price of the next dollar of that stock. The house names the convention that scores the first, and on its reading carries no term for the second, the Cost-Blind Buffer — a reading held as an escaped predicate (EP8) until the applicable framework is classified metric by metric, availability term against cost term.

Who is measured by a regulatory ratio at all

The perimeter comes first, because it decides whether “liquidity” for this cohort is a rule or a construction. Under 12 CFR 249.1(b)(1)(i), the institutions listed as subject to Regulation WW’s minimum liquidity and minimum stable funding standards are global systemically important BHCs, G-SIB depository institutions, Category II and Category III institutions, and Category IV institutions at or above a weighted short-term wholesale funding threshold; under (b)(1)(iii) the Board may also apply the part by determination (Board of Governors, eCFR current to 2026-09-18) [Confirmed]. For an institution the part does reach, the liquidity coverage ratio is its HQLA amount divided by its total net cash outflow amount (12 CFR 249.10(c), eCFR current to 2026-10-13) [Confirmed] — a construction that, on its own, says nothing about whether supervisory review or internal assessment weighs funding cost. For a U.S. bank holding company or intermediate holding company, Category IV begins at an average total consolidated assets line held open — dollar threshold omitted pending resolution of its basis under 12 CFR 252.5(e)(1), and such an organization leaves the category downward only under a separate test — total consolidated assets below the §252.5(e)(2) threshold for each of the four most recent calendar quarters (Board of Governors, eCFR edition current to 2026-10-09) [Confirmed]. The house cohort is bounded at US$5–100bn in total assets at T0 (house cohort definition, as of 2026-06-30) [Range] [HOUSE-ANCHOR]. That bound is a bank-charter measure of Call Report total assets, not the regulation’s holding-company average, and does not stand in for the Category IV line. Whether the holding company above a cohort bank falls under Part 249 depends on more than that line: below Category IV, it is outside the part absent a Board determination; inside Category IV, it is outside unless its weighted short-term wholesale funding meets the threshold or the Board so determines. Where neither brings it in, and setting aside the parallel OCC and FDIC rules this report does not exhibit, liquidity for that bank is an analyst construction from Call Report lines plus an examiner’s rating and contingency-funding review (EP1).

What moves the measured buffer

Inside an availability convention, four ordinary treasury actions can raise measured liquidity, depending on how the measure treats each funding source’s tenor and outflow. A bank can borrow term advances to hold cash (a gross-up); replace uninsured deposits with brokered or reciprocal deposits; pledge loans as discount-window or FHLB collateral; or let undrawn commitments run off. The first three raise the marginal cost of funds or encumber the balance sheet; the fourth shrinks the credit the bank stands ready to extend. None of the four is charged for its cost inside an availability measure. To the extent the buffer is bought at the wholesale margin, it is therefore both a source of the premium — optics-driven demand for liquidity priced on the marginal curve — and the screen that keeps the premium out of the liquidity line.

The incentive map follows from the same arithmetic. On the house’s assumption (EP8), the examination lens rewards buffer size and contingent capacity and is availability-weighted. Where the buffer carries negatively, the carry has to be recovered somewhere, and this map locates it in the loan book. The FHLB earns on the advance. The deposit insurer can price non-core funding in its assessment — on the house’s assumption (EP8), a cost booked outside the liquidity line. If, as the house reads it, analysts reward the buffer ratio and penalize NIM without connecting the two, any mispricing lives in that pair: the liquidity-premium component of a cohort multiple and the NIM drag created by the same buffer, priced as if unrelated. The observable surfaces are FHLB advances (Call Report Schedule RC-M), brokered and reciprocal deposits (RC-E memoranda, read separately), pledged securities and loans (RC-B and RC-C memoranda), and the lendable-value gap between agency securities and C&I loans at the discount window open — collateral categories, valuation basis and margin-table vintage not yet fixed. On the house’s working assumption (EP9), availability is rarely the constraint on an advance against eligible collateral; cost is.

Where the buffer decision is taken

On the house’s model of the sequence, the buffer floor — a minimum liquidity ratio, a survival horizon — is set in the board’s liquidity risk-appetite statement and re-set when the bank remediates an examination finding: forums whose test is survival. Whether, and how, the cost of meeting that floor is weighed there is not disclosed for the cohort. How the buffer is funded is decided afterwards, by Treasury, at the moment of the draw — on this model, in reaction to an uninsured-deposit outflow or an exam finding. The cost appears later still, in ALCO and in earnings, as a given the loan book must recover. If optics and marginal economics diverge, the model places the divergence in the sequence of these forums rather than in anyone’s misjudgment; the sequence is the house’s model, not observed cohort practice.

Marginal-curve transfer pricing (M1): the internal device that would turn a premium visible only to Treasury into a hurdle on one product.

In a bank that runs funds transfer pricing, every new or renewed loan is charged a funding cost read off an internal curve. Keyed to the blended cost of the deposit book, that curve moves slowly and singles out nothing. Keyed to the marginal curve — the term-CD rate, the brokered-CD rate, the FHLB advance curve, plus a bank-specific liquidity add-on — it converts a premium seen only by Treasury into a hurdle on the product that consumes contingent liquidity: the revolving line, which under such a scheme also carries a liquidity charge on its undrawn balance for the buffer it requires. Whether cohort banks run marginal-cost transfer pricing with such a charge is not disclosed (EP2), and no supervisory prescription stands in for that disclosure: interagency guidance on transfer pricing for funding and contingent-liquidity risk addresses how those risks are allocated, but presents its methodologies — a weighted-average cost of debt among them — as illustrations, not as directives for any particular methodology, and carries its own applicability conditions. The marginal-versus-blended choice, and with it this chapter’s mechanism, rests on cohort practice the house has not observed. Whether supervisory guidance expects liquidity costs to be priced into the products and activities that create them, undrawn commitments included, remains an assumption; no verified clause stating that expectation is cited. Cross-border and wholesale funding conditions enter here only as inputs to the curve’s level, never as separate nodes.

The marginal curve, not the blended one, prices the next loan when the alternative source of funds — the securities book — is locked (EP3; inset below). On the house’s model, averages lag the margin by the repricing tenor of the deposit book and smooth the cross-bank dispersion a marginal curve would show; that ordering has not been measured for the cohort. Neither the cohort’s marginal-over-blended spread open nor the new-production spread over the margin open has a published series; the second surfaces, if at all, in earnings-call commentary. What can be computed from public lines is an average — blended interest-bearing deposit cost, whose cross-bank dispersion change in cohort IQR — open, cohort computation basis not yet fixed is the nearest public proxy for the premium the curve would transmit; blended-cost dispersion and marginal-premium dispersion are distinct objects.

Where the curve is set

On the house’s model, two discrete choices, not the standing methodology, decide the hurdle. The first is the curve re-mark at the ALCO cycle after a funding-replacement event — the first brokered deposit or FHLB draw taken to replace outflowing deposits, the moment “marginal” acquires a value different from “blended.” On this model the re-mark follows the event rather than anticipating it. The second is the size of any charge on undrawn SME lines, which would not be decided in the pricing tool at all: it would be inherited from the outflow assumption for undrawn C&I commitments in the bank’s internal liquidity stress test and contingency funding plan open — cohort computation basis not yet fixed — a risk document written for the examiner. The nearest codified number for that assumption sits in the liquidity coverage ratio’s commitment-outflow schedule: 10% of the undrawn amount of committed credit facilities extended to wholesale customers that are not financial-sector entities, liquidity facilities excluded (12 CFR 249.32(e)(1)(iii), eCFR current to 2026-10-13) [Confirmed]. It is a regulatory outflow assumption, not a measured internal charge, in a rule that, on Chapter I’s perimeter reading, may not reach this cohort at all; it is not treated here as binding on any cohort bank. The hurdle an SME revolver must clear could therefore be fixed in an examiner-facing assumption before any lender sees the loan: the decision that would govern SME credit is taken where SME credit is not the subject.

On the house’s incentive map — a model of mandates, not disclosed positions — the conflict around the curve is four-sided. Treasury would want the true marginal cost charged, to protect spread. The commercial lines would want volume, relationship revenue and deposit-linked compensation. The CFO and investor relations, holding to stable NIM and loan-growth guidance, would have reason to favour a quiet hurdle change over visible repricing. The examiner’s interest is assumed to run to conservative liquidity charges on contingent commitments — charges that, once applied, would speed line rationing, whose shrinkage an availability metric would then score as improved liquidity.

No slope distinguishable from zero on the outcome; none on the placebo; the larger-bank subsample carries the thesis’s sign without significance. Under the disposition above, the funding channel is not identified in the commitment-size surface over the evaluation window; the credit-risk and demand reading stands as the default, not displaced and not confirmed — the surface is quiet on both models, and the non-significant coefficient is not evidence that the channel is absent. The regressor is the annualized cost of interest-bearing deposits — computed from public bank-level Call Report lines, and an average: the kind of quantity this chapter says lags the margin. That limit cuts both ways. A flat result on the average does not test the marginal curve; but over ten quarters an average built on repricing deposits should converge toward the margin, and the defence weakens with every quarter the panel adds. These coefficients, like the gradients in Chapter IV, are carried subject to the open reporting-basis check in Risk Box 3. The identical specification is re-run on the extended panel as Marker 2.

The non-price term reset at annual review (M2): why a premium that has little room in the rate would travel through the terms.

Take SME credit at the cohort to be what the house assumes it is (EP4): revolving operating lines renewed annually at the lender’s discretion, multi-year term loans with balloons, coupons indexed to Prime with floors, and the account bundled with the borrower’s operating deposits. On those terms the existing book is largely sealed against a lender-specific premium between resets: the coupon follows the index, and the spread over it — the term through which a premium could still enter at origination or reset — is the most visible and most competed term in the file. A premium set by the marginal curve then has three exits: be absorbed in NIM, be quoted in new spreads, or be moved into terms at renewal.

The third exit has six doors: (i) the committed line size and borrowing base; (ii) collateral coverage and advance rates; (iii) covenants; (iv) deposit-relationship conditions — operating-account and compensating-balance requirements, a pass-through paid as forgone deposit yield and recorded in no price series mapped here; (v) conversion into a government-guaranteed structure, which sheds risk to the guarantor and charges the borrower a fee; and (vi) non-renewal. The outward signature is the one in the house name: rates unchanged, fewer dollars.

Who turns the ratchet

On the house’s model of the decision — not disclosed by cohort banks — credit policy sets the permitted range through the chief credit officer’s overlays. The economic choice — which lever, how far — is made by the relationship manager structuring the annual-review package in the pre-expiry window, after the pricing screen returns a hurdle shortfall; the credit approver or committee ratifies a proposed structure rather than originating one. Two levers carry their own decision points. Exit is decided at the periodic portfolio exit and watch-list review by credit risk, not at the individual renewal. Guaranteed conversion is decided jointly with the borrower at application, so the bank’s choice is to offer, not to impose. On this model the ratchet would be executed by the actor whose compensation rewards retention — which is why, if it turns, “renew smaller” would dominate “exit,” and the tightening would be gradual, one-directional and absent from rates.

On the same map the incentives around that officer pull apart. Credit policy would want risk-weighted assets and contingent liquidity down. The CFO would have reason to prefer a smaller renewal to an exit, because non-renewal of a stressed borrower can manufacture a default. The guarantee program absorbs the risk the bank sheds, which would make migration into guaranteed structures individually rational and hard to see in the public record mapped here.

What the public record can see of each lever

The first pass’s unidentified result speaks to lever (i), at all-C&I resolution, and to nothing else in this table. One further surface may bear on lever (i): the Call Report’s loans-by-original-amount schedule (RC-C Part II), whose bands sort loans by original amount, not by borrower size; its reporting rules are not verified here, and band shifts are not read as a change in small-credit supply. The first pass did not use that schedule; Marker 2 reports it beside the unchanged specification as a declared extension, not a substitute.

Where the three mechanisms meet, why the meeting can make a squeezed bank look stronger, and the test that separates supply from demand.

The “next marginal dollar” decision is not a fourth device. The marginal curve prices the dollar; the buffer convention sets how much liquidity the dollar must fund; the ratchet executes the refusal at renewal. One allocation rule carries the composite: cash and high-quality securities typically outrank SME loans on liquidity value and risk weight, so at the margin the liquidity optic and the capital ratio can tilt the same dollar the same way — toward securities — subject to each asset’s eligibility, haircut, yield and funding cost.

Loop one: the buffer’s cost becomes the loan’s hurdle

Under marginal transfer pricing (EP2), the FHLB advance and term-deposit curve is at once the price of an optics-driven buffer and the reference for the transfer-pricing curve. A buffer built with wholesale liabilities would then lift the marginal curve and the hurdle with it, handing the buffer’s negative carry to SME credit. The loop is self-limiting. A buffer built from loan run-off does not lift the curve; run-off also lowers the funding need, wholesale reliance falls, and the premium compresses. On this model rationing would cure the premium that caused it: a one-shot balance-sheet adjustment, not a spiral. The switch is the sign of buffer carry open — no published bank-level series for the marginal funding rate. Once holding the buffer pays, hurdle pressure ends — but if the ratchet is one-directional, as modelled, the commitments it cut need not be restored on the same timetable.

Loop two: the squeeze improves the ratio

When an undrawn SME commitment is cut, any liquidity measure that books undrawn lines as contingent outflows moves at both ends: the contingent-outflow requirement falls, and cash that would have funded future draws stays on the balance sheet. Where Chapter I’s composition test shows the buffer improving through run-off rather than wholesale funding, and the supply signature below also holds, a stable or improving optic is not evidence against rationing — it is partly its by-product, and the consensus variable becomes in part an output of the variable it is used to dismiss. Where the composition is run-off but the supply signature fails, the run-off is not attributed to supply — demand, credit risk and portfolio choice remain open — and the optic is not explained by this loop; where the composition is wholesale-funded, Chapter I’s reading applies instead. The offsetting loop caps the effect: capacity freed by cuts would be redeployed into securities carry, NIM would improve, the hurdle would ease and the ratchet would stop — provided securities yield more than the marginal liability open — no published bank-level series for the marginal funding rate. When they yield less, the same redeployment loses money and the pressure intensifies.

Loop three: the deposit condition funds the ratchet

Deposit conditions extracted at renewal bring operating balances onto the balance sheet and lower the marginal cost that motivated them; the ratchet partly finances itself and is self-dampening. That holds only while SME operating deposits are rate-insensitive enough for the condition to be enforced (EP5). The nearest reported proxy is the cohort’s noninterest-bearing deposit share change over 2023-03-31 to 2026-06-30 — open, cohort computation basis not yet fixed. If balances sweep to money-market alternatives, the bank can lose the deposit and the loan together, and the pass-through would switch from “renew smaller” to exit — sharper, not gentler.

The separating test, as designed and as first run

The rival model — a supervisory CRE-concentration reset, with funding cost a co-symptom of the same franchise weakness — explains every aggregate observable in Chapters I to IV. It parts from the house model on one prediction. Within a stratum of CRE concentration, the house model predicts that non-real-estate revolving commitments contract more at high funding-cost banks than at low; the rival predicts no such gradient, with contraction sitting in real-estate-secured commitments and scaling with CRE across strata. The test is a two-way sort — funding-cost tier by CRE-concentration tier — on unused C&I commitments, with real-estate-secured commitments as the placebo column and the utilization co-condition attached. Timing corroborates but does not separate: funding-driven cuts would cluster at renewals after a funding-replacement event, CRE-driven cuts after exam cycles.

A quartile sort, which the claim’s scope follows, left 8 of its 16 cells at or below the cell-size floor and was set aside as a basis for the verdict [Range] [HOUSE-ANCHOR]; the tercile sort carries the verdict.

Verdict: UNIDENTIFIED, on the tercile cell table [Range] [HOUSE-ANCHOR]. Confirmation required a negative gradient in a majority of strata, a significantly negative pooled coefficient and the utilization co-condition; the disconfirmation set — the CRE coefficient significant on its own, a funding-cost slope in the placebo, utilization falling — was not met either. On the house computation’s record, core fields were reported at 100% and the tercile cells met the cell-size floor [Range] [HOUSE-ANCHOR]; on that record the tercile verdict is a result, not a data shortfall, while at the quartile scope the claim follows, the cells are too thin for a verdict. The first pass tested two legs of the AQ-01 signature — the within-stratum funding-cost gradient and the utilization co-condition — against a real-estate-secured placebo; the term-balance leg was not tested. One stratum of three (C2) carries the house’s sign. One check on the first-pass inputs is open: whether numerator and denominator items were taken on a single reporting basis (RCON or RCFD) for FFIEC 031 filers has not been verified from the available execution records. Until it is, the utilization co-condition is reported as computed, not as a verified leg of the signature, and the gradient results and the funding-cost coefficients in Chapter II are carried under the same reservation. The commitment-size surface is quiet on both models; by the disposition in Chapter II, the default reading stays with credit risk and demand — as the undisplaced default, not as a result of this test.

What the survey shows, and where it stops. Business line of credit applicants were 45% fully approved, 31% partially approved and 24% denied (N=977) — by-product chart, all sources, 2025 SBCS (Federal Reserve Banks, published March 2026), employer firms, prior 12 months [Confirmed]. Among applicants that did not receive all of the financing they sought, 37% did not accept some or all of what was approved (N=1,936; multiple responses; prior 12 months) [Confirmed]: an offer made and not taken up, one place a terms-based pass-through, if it operates, could register, among other reasons for declining. The survey then stops where the thesis needs it. Lender-type outcomes are pooled across loans, lines and cash advances; banks are split at US$10B in total assets by respondents classifying their lender with the help of a list, a definitional line inside the house cohort’s range; and no table in the report pages read crosses product with lender type. The cross-section the thesis needs — line-of-credit outcomes at high-funding-cost lenders against low — is not in the report pages read.

The borrower cohort, the containment channels, and the three regimes in which the mechanism would operate, reverse or migrate.

The thesis is not about SMEs as a class. Its borrower is narrower: a firm that needs a revolving operating line above business-card size, carries thin collateral, and holds a deposit relationship the lender can condition — the bank-dependent revolving borrower. For that borrower, a revolving line underwritten on soft information is a bank-specific product if substitutes supply mainly term and receivables credit (EP6); for other small firms, the house does not claim a bank-specific product.

The survey’s lender-type outcomes give a first look at where refused borrowers can go — pooled across loans, lines of credit and cash advances, not specific to lines; 2025 SBCS (Federal Reserve Banks, published March 2026), employer firms, prior 12 months, pending applications excluded [Confirmed]. Fully approved / partially approved / denied: small banks (less than US$10B) 57/23/20 (N=597); finance companies 50/31/19 (N=305); credit unions 44/26/29 (N=150); large banks (at least US$10B) 43/26/31 (N=919); online lenders 38/39/23 (N=665); CDFIs 27/39/34 (N=116). No non-bank source fully approves a larger share of its applicants than small banks do, and online lenders and CDFIs carry the highest partial-approval shares. The table records who applied where, not how one borrower would fare elsewhere; selection runs through every cell. The share of applicants at each lender type that received less than the amount requested is open — computation scope, channel definition, lender grouping or table locator not yet fixed.

Regime map

The September SEP year-end medians alone do not identify the switch into the Sign-flip regime. The federal funds target range stood at 3.75–4.00% (as of 2026-09-16) [Confirmed] [HOUSE-ANCHOR], and the September 2026 SEP median — participants’ projections of the midpoint of the appropriate target range, not a commitment — is 4.1% for end-2026 and 4.1% for end-2027 (as of 2026-09-16) [Confirmed] [HOUSE-ANCHOR], above the current range on both dates. These medians are levels at end-2026 and end-2027, not the path between them, so they do not settle whether policy easing could be a route; at a flat-to-higher policy level, the switch would have to come from the curve’s shape or from dispersion compressing. Excluded by scope: an acute liquidity event moves the availability measure itself and validates the variable through it rather than testing it.

Sensitivity of the report’s asset-pricing objects under each regime, plus the identification-null state. Directional and qualitative by design; no cell is a trade.

Out of scope: an acute liquidity event moves the availability measure itself; it is not a regime of this map.

Reader-side parameters, not house advice. Every boundary is analytical and carries the [Scenario] label; the house sets no exposure limits.

On the house’s model, the structure the consensus does not map is a premium that would be generated in one forum, priced in a second and executed in a third — none of which has small-business credit as its subject. On that model the board and the examiner set the buffer floor; ALCO and the stress test set the curve and any charge on undrawn lines; a relationship manager in a pre-expiry review chooses the lever. What binds the three, on the house’s reading of the cohort’s book (EP4), is a contract: a revolving line indexed to a national rate and renewed, at the lender’s discretion, once a year. That contract is the sector-level distortion. It leaves the premium a narrow, heavily competed route through spreads at origination and reset and a wider route through terms — and the measurement convention around the bank, on the house’s assumption (EP8), scores the terms route as strength.

Deadlocks, gaps and bottlenecks

Regulatory deadlock, on the house’s assumption (EP8): the examination metric rewards buffer size and contingent capacity and is availability-weighted; any negative carry has to be recovered, and this map locates it in the loan book; the deposit insurer can price non-core funding in an assessment the liquidity ratio does not register. Three cost signals, no shared ledger — and on this map each actor behaves correctly by its own mandate. Funding-cost gap: where an acquirer funds below the cohort’s marginal curve, consolidation can capture the gap by re-funding the SME book on the cheaper curve; the guarantee program offers the borrower a different gap — public-balance-sheet risk at a borrower-paid fee. Indirect bank dependence: where the non-bank lenders that absorb displaced borrowers fund themselves with bank credit, migration relocates the bank link rather than removing it. Certification bottleneck: contingent capacity counts only for collateral pre-positioned and accepted at a margin open — collateral categories, valuation basis and margin-table vintage not yet fixed; eligibility is not unused capacity, since haircuts and collateral rules set how much can be drawn; pledging C&I loans to manufacture capacity encumbers the very book the ratchet would shrink. External channel: cross-border and wholesale funding conditions reach this map only through the level of the marginal curve; they are inputs, not nodes, and the chain terminates domestically.

Cohort banks are not named in this register: the house computation is published as cohort statistics, and no bank-level value has been verified for this report. Bank-level classification follows the composition test (Marker 1).

What survives, and where. The house claims a structure, not an outcome. Cohort: Tier-2 banks in the top marginal-funding-cost quartile with high uninsured or wholesale reliance — not the cohort as a whole. Borrower: bank-dependent SMEs on revolving operating lines above card size with thin collateral — not SMEs as a class. Regime: negative buffer carry with persistent dispersion and no acute event; the claim lapses in the Sign-flip regime and migrates in Containment. Horizon: one renewal cycle from premium to commitment reset for revolvers, the balloon wall for term loans, and hysteresis after the premium compresses if the ratchet is one-directional. Identification: the funding channel is asserted only where the revolver-and-utilization signature separates it from credit risk and demand — for revolving commitments at the top funding-cost quartile over post-2023 renewal cycles — and the house’s first pass on 2024–26 data, run on terciles because quartile cells were too thin, did not identify that signature; its inputs carry an open reporting-basis check.

Consensus under interrogation

The reading under interrogation — the house’s characterization of it, not a measured survey of positions — treats regional-bank liquidity through aggregate cash buffers, deposit stability and regulatory liquidity compliance, and infers from stable metrics that the system retains the capacity to keep lending to smaller businesses. As the house reads them, the two live camps — “liquidity is fine” and “margins are under pressure” — share one premise: funding cost is a bank-margin variable, absorbed in NIM and competed away before it reaches a borrower. This report interrogates that premise by separating liquidity availability from the price of replacing liquidity, and by asking where that price could travel if not into the margin.

For a global or emerging-market portfolio manager, the transferable object is not U.S. regional banks but a pairing: a liquidity metric that counts availability, and a credit contract renewed at the lender’s discretion. Wherever both exist, a funding premium can be paid by borrowers in terms rather than rates — and the local liquidity ratio can register the payment as strength.

What this report is — and is not

It is a structural map of the gap between headline liquidity optics and the marginal funding cost that would govern SME credit supply: the mechanisms, the control and decision points, the incentive conflicts, and a published separating test with its first-pass result. It is not a set of single-name calls, price targets, long/short recommendations or P&L-optimized trade lists. Readers translate the mechanism and its thresholds into their own strategies.

Reader application

Long-only

Read a cohort bank’s loan-growth guidance against its unused-commitment run-off and utilization, not against deposit cost alone.

Treat a liquidity ratio that rises in the same quarter unused C&I commitments fall as a question for management, not as comfort.

Ask how the undrawn-commitment outflow assumption in the contingency funding plan is set; under marginal transfer pricing, that is where the SME hurdle would be fixed.

Event-hedge

The ratchet produces no event; the transitions to watch are the buffer-carry sign and funding-replacement quarters, not a run.

The dated checkpoints are Call Report releases and the annual SBCS employer-firm report (Markers 1–3), with SLOOS rounds as the quarterly aggregate check.

An acute liquidity event sits outside this framework; it moves the availability measure itself.

Macro-allocator

SME credit can tighten through terms while rate-based credit indicators stay calm; read non-price survey terms alongside spreads, not spreads alone.

The Sign-flip regime needs compressed dispersion or a steeper curve; the September 2026 medians are year-end levels, not the path between them, so they neither confirm nor rule out a lower policy rate as a route.

Containment moves SME risk to guarantee programs and private-credit books; follow the carrier, not the bank.

Falsifiable, dated and checkable against public sources. Each marker states in advance what confirms and what invalidates.

Methodology & House Rules

Four layers

Layer 1 Surface Narrative — the headline liquidity optics and the reading under interrogation (Key Takeaways; Consensus Under Interrogation).

Layer 2 Organ / Macro Vectors — the buffer and the funding curve: the Cost-Blind Buffer (M3) and marginal-curve transfer pricing (M1) (Chapters I and II).

Layer 3 Nerves / Transmission — the Renewal Ratchet (M2), the three loops and the separating test (Chapters III and IV).

Layer 4 Scars / Regime Shifts — the borrower cohort, containment, the regime map, the matrix, the blotter and the markers (Chapter V and the pillars that follow it).

Data source hierarchy

Tier 1 — Sovereign / Regulator: Board of Governors of the Federal Reserve System (12 CFR Parts 249 and 252 via eCFR; Senior Loan Officer Opinion Survey; FOMC statement and Summary of Economic Projections); interagency liquidity and funds-transfer-pricing guidance; Federal Financial Institutions Examination Council (Call Reports, CDR bulk data); Federal Reserve Banks (Small Business Credit Survey; discount window); Federal Reserve Bank of Kansas City (Small Business Lending Survey); FDIC; U.S. Small Business Administration; National Credit Union Administration.

Tier 2 — Listed IR / Consensus: holding-company earnings disclosures and calls (new-production yields), sell-side consensus revisions, market valuation data.

Tier 3 — Macro proxies: Federal Reserve H.8 aggregates for small domestically chartered banks; NFIB small-business credit-conditions series.

Analytical labels

[Confirmed] — a value or clause printed by a Tier-1 source and exhibited or audited.

[Range] — a house computation or candidate band; house-computed anchors carry this label.

[Market Estimate] — a consensus or market-derived estimate; none is used in this report.

[Scenario] — a conditional analytical state; every matrix cell, regime row and blotter boundary.

[N/A] — no admissible value. In this report an open quantity is not reduced to a bare label: it is carried as an open quantity — a numeric slot whose ID and definition are held fixed until a value meeting that definition is retrieved, listed under Open quantities.

House conventions alongside the five labels: [Assumed] marks an unverified predicate (anchor predicates and EP-coded escaped predicates), never stated as fact. Provenance tag: [HOUSE-ANCHOR] marks values from the house anchor library, carried verbatim with unit and as-of date; on a house-computed anchor it records the computation as registered, not a reproduction of it from raw data. Observability tags: [Observable], [Inferable], [Structurally Unobservable].

Open quantities

22 distinct quantities remain open, each carried as an open quantity with its ID and definition held fixed, listed below: A1–A11 (cohort funding, buffer, valuation and stress-test quantities), B1–B4 and B6–B9 (cohort commitments and utilization, SLOOS non-price net shares, SBA guaranteed approvals, substitute channels, rate dispersion, partial funding by lender type) and C1, C3 and C4 (the Category IV asset line, the credit-union business-lending cap, the discount-window margin gap). A1, A3 and A11 turn on the marginal funding rate, which has no published bank-level series and no fixed measurement protocol here; A2 and A4–A10 wait on a fixed cohort computation basis — cohort membership, the reporting basis of each Call Report input, a valuation source and a disclosure standard for internal stress tests; B1, B2 and B7–B9 wait on a fixed computation scope, substitute-channel definition, lender grouping and survey table locator; C4 waits on fixed collateral categories, valuation basis and margin-table vintage; C1 is held open with the dollar thresholds omitted from Exhibits 1 and 2. For B3, B4, B6 and C3, source material has been read but does not meet the slot’s definition as printed; no derived value, distribution or subtraction of rounded percentages is substituted. Two slots are filled in this edition: the LCR commitment-outflow rate (Chapter II) [Confirmed] and the SLOOS small-firm spread reading (Chapter III and the blotter) [Range].

House computation: the first-pass separating test

Cohort: 116 banks × 10 quarters, T0 2023-03-31, evaluation window 2024-03-31 to 2026-06-30 [Range] [HOUSE-ANCHOR]. Tier-2 definition: T0 total assets US$5–100bn; C&I loans at least 10% of total loans; panel-consistent; M&A-flagged banks excluded [Range] [HOUSE-ANCHOR]. Exclusions: 124 (panel exit 27; business model 68; CBLR filers 4; zero commitments 4; M&A 21) [Range] [HOUSE-ANCHOR]. Source: FFIEC CDR Call Reports bulk data, 2022Q4–2026Q2, all schedules, with the FFIEC MDRM dictionary. Outcome: year-on-year Δlog of unused C&I commitments (Schedule RC-L), used as an all-C&I proxy rather than as SME revolving lines; placebo: real-estate-secured CRE commitments. Regressor: annualized cost of interest-bearing deposits. Estimation: pooled regression with quarter fixed effects and bank-clustered standard errors, plus a subsample of banks with T0 assets of at least US$10bn [Range] [HOUSE-ANCHOR]. Separating design: two-way sort of funding-cost tier by CRE-concentration tier, tercile cut. Decision rules: CONFIRM requires a negative within-stratum gradient in a majority of strata, a significantly negative pooled coefficient and the utilization co-condition; the DISCONFIRM set is the CRE coefficient significant on its own, a funding-cost slope in the placebo and falling utilization; a result meeting neither is UNIDENTIFIED. Input reproducibility from public FFIEC Call Report data is not yet confirmed: the item-code mapping used in the actual run — including the RCON or RCFD basis of each input for FFIEC 031 filers — has not been obtained from the available execution records, and it is carried as an open check; the funding-cost coefficients, the gradients and the utilization co-condition are all carried subject to it, and no calculation error has been found.

Literature: the public studies opened for this report do not establish the post-2023 funding-dispersion channel for this population and product separately from CRE concentration, credit risk and demand; studies that could not be read are not counted as evidence either way.

Exhibit conventions

Primary source exhibits render verified figures with the [Confirmed] label and carry each figure’s period and scope qualifier wherever it is cited. Dollar thresholds in the Regulation WW and Regulation YY exhibits are omitted pending resolution of their basis; the Category IV asset line is carried as an open slot. Money figures are stated in U.S. dollars. As-of conventions: Call Report data to 2026-06-30; regulation text current to 2026-09-18 (Part 249, §249.1), 2026-10-13 (Part 249, §§249.10 and 249.32) and 2026-10-09 (Part 252); SLOOS as of the July 2026 round; survey data as published March 2026; policy-rate anchors as of 2026-09-16.

Assumptions this report rests on (EP1–EP9)

EP1 [Assumed] — The cohort sits below the Category IV threshold and outside LCR/NSFR, so its liquidity measures are analyst Call Report ratios and supervisory expectations. Conditional on the Category IV asset line — held open, dollar threshold omitted pending resolution of its basis and on reconciling bank-charter cohort measurement with holding-company categorization; entity-level applicability has not been mapped.

EP2 [Assumed] — The cohort runs marginal-cost transfer pricing with liquidity charges on undrawn commitments. Unverified; not disclosed. Interagency transfer-pricing guidance on funding and contingent-liquidity risk presents weighted-average as well as marginal methodologies as illustrations, not directives, and carries its own applicability conditions; the marginal-versus-blended choice therefore rests on cohort practice, not on a supervisory prescription. Actual cohort practice is neither established nor disproved.

EP3 [Assumed] — Unrealized securities losses make monetization capital- or earnings-prohibitive, so the marginal loan is funded at the marginal liability cost. Regime-dependent; see cohort unrealized securities losses as a share of tangible equity — open, cohort computation basis not yet fixed. An unrealized loss does not by itself show that a sale is infeasible or uneconomic.

EP4 [Assumed] — Cohort SME credit is predominantly annually renewable, index-priced, relationship-bundled revolving credit plus balloon term loans. Unverified. Indexation alone does not establish a fixed spread or exclude bank-specific pricing at origination or reset; the decision model in Chapter III — who chooses the lever, what the committee does — is part of this predicate.

EP5 [Assumed] — SME operating deposits are rate-insensitive enough for deposit conditions to be enforced. Unverified; proxy change in cohort noninterest-bearing deposit share — open, cohort computation basis not yet fixed.

EP6 [Assumed] — Substitute channels supply mainly term and receivables products, not revolving operating lines. Unverified; revolving alternatives exist in at least some channels, and the matched comparison that would test “mainly” has not been made; fintech and private-credit SME revolving volumes are not observed in the sources this report maps.

EP7 [Assumed] — Commitment and loan-count data are frequent and granular enough to observe renewal-cycle changes at cohort level. Under strain: the first pass returned UNIDENTIFIED at all-C&I quarterly resolution, and aggregate commitment movements do not identify renewal events, contractual resets or borrower-level utilization.

EP8 [Assumed] — The supervisory and analytic liquidity metric set applied to the cohort carries no marginal-cost term. Unverified; metric-by-metric classification pending. A ratio without an explicit funding-price input does not by itself establish that supervision or internal assessment ignores cost.

EP9 [Assumed] — FHLB advance availability is not collateral-constrained for the cohort; cost binds. Unverified; collateral eligibility does not establish unused capacity.

Adjacent structures outside the core mechanisms

SAT2 — Deposit-beta dispersion: the raw cross-bank difference in marginal deposit cost that the marginal curve transmits; enters this report through its dispersion proxy.

SAT7 — Regulatory buffer optimization versus economic liquidity cost: absorbed into the Cost-Blind Buffer as its definition.

SAT9 — Reciprocal-deposit programs and deposit-insurance assessment adjustments for non-core funding: costs of retaining uninsured deposits, or of relying on wholesale funding, that, on the house’s assumption (EP8), sit outside the liquidity ratio.

SAT11 — The contingent-liquidity charge on undrawn commitments as an assumed supervisory expectation (unverified): if guidance installs the transmission device on the cost side while the metrics score the availability side. Guidance on allocating funding and contingent-liquidity risk can inform internal pricing; method, institution scope and the actual charge decide whether the hurdle exists.

Rendered elsewhere: the accounting lock (Chapter II), the maturity wall under the buffer (Chapter I), how the ratchet would choose its borrowers (Chapter III) and the substitute with a statutory ceiling (Chapter V) as insets; the balloon wall (Chapter III) and the containment carriers (Chapter V) as risk boxes; the allocation rule that carries the composite in the body of Chapter IV.

Claim perimeter: Tier-2 banks in the top marginal-funding-cost quartile with high uninsured or wholesale reliance; bank-dependent SMEs on revolving lines above card size with thin collateral; negative buffer carry with persistent dispersion and no acute event, lapsing in Sign-flip and migrating in Containment; one renewal cycle for revolvers, the balloon wall for term loans, hysteresis after compression if the ratchet is one-directional; asserted only where the revolver-and-utilization signature separates funding from credit risk and demand, for revolving commitments over post-2023 renewal cycles.