The Reserve

On March 3, 2026, the Committee on Capital Markets Regulation convened a roundtable in Washington on bank liquidity and the lender of last resort. The prepared remarks were attributed to Scott Bessent, Secretary of the Treasury. They were delivered by Under Secretary for Domestic Finance Jonathan McKernan. That arrangement signals institutional weight: the attribution to the Secretary and the delivery by his Under Secretary for Domestic Finance together place the remarks within the Treasury's active reform agenda on bank liquidity regulation — one that the official positioned to coordinate across the Federal Reserve, the FDIC, the OCC, and the banking agencies is publicly advancing.

The substance of that reform agenda is ambitious in scope. Bessent's text describes the post-crisis liquidity framework — principally the Liquidity Coverage Ratio — as a "fundamental mistake" born of regulatory trauma rather than sound design. Banks, the argument runs, were forced to self-insure against liquidity stress by holding vast quantities of high-quality liquid assets: government securities and reserves that sit idle rather than circulating into productive use. That self-insurance requirement, Bessent contends, entrenched the very discount window stigma it was meant to address, because holding the mandated buffer made banks unwilling to deploy it when stress arrived. The prescription is proportionate to the diagnosis: dismantle the mandate, restore meaningful access to the Federal Reserve's lending facilities, and release the capital locked inside those buffers into AI infrastructure, defense supply chains, and domestic manufacturing. Trillions, not billions.

The argument has a coherent internal logic. It names a real problem — stigma is real, and the LCR buffer has not functioned as a deployable reserve. It promises a concrete economic payoff at a moment when industrial lending is a stated national priority.

Each element of that argument rests on specific historical and institutional claims. The Federal Reserve's own internal record — transcripts, staff briefings, and policy memoranda spanning the design and implementation of the post-crisis framework — can evaluate them.

We went into the archives.

The Insurance They Built on Purpose

The keystone question the archive must answer is whether the self-insurance mandate was a reflexive overcorrection born of post-crisis trauma, as Bessent argues, or whether it was the institutional conclusion drawn deliberately from the Fed's own documented crisis experience — the corrective to a demonstrated failure rooted in pre-crisis LOLR dependence.

Bessent frames the LCR and its underlying philosophy as a categorical error: "deputizing large banks as liquidity insurers for the financial system writ large." The logic is that only the central bank can create aggregate reserves during a flight to safety, so requiring banks to self-insure misidentifies the insurer. The claim has a surface coherence. It also inverts the institutional record.

The Fed's own analysis located the problem precisely where the LCR would later aim its remedy — and did so while the crisis was still unfolding. In June 2008, months before Lehman Brothers failed, Jeffrey Lacker, the President of the Federal Reserve Bank of Richmond and one of the committee's most consistent voices against regulatory expansion, had already identified the moral hazard at the system's core:

Lacker was not a regulatory expansionist. His concern was the opposite: that the implicit backstop had become a subsidy enabling precisely the funding fragility it was supposed to address. At the same meeting, Donald Kohn, then Vice Chairman, stated the institutional conclusion plainly — that the regulatory structure and the liquidity provision structure had simply not been sufficient to protect the economy from the new style of financial system. Both men were describing a system built on central bank access. Both were watching it fail because of that dependence.

The post-crisis framework codified what those internal diagnoses required. Governor Daniel Tarullo, who oversaw the regulatory architecture through its implementation, described what the self-insurance requirement was designed to prevent:

Runway. The LCR was not designed to replace central bank liquidity provision but to buy time — to create the interval in which intervention could be measured rather than panicked. By 2014 Tarullo could state the framework's purpose without ambiguity: liquidity regulation had a better-appreciated role to play in tandem with capital regulation and resolution mechanisms, and as a means for both complementing and limiting the LOLR function of central banks.

The internal record adds a dimension the public framing did not foreground. Staff analysis conducted before implementation showed that the LCR would permanently anchor bank demand for high-quality liquid assets and reserves — not as a transitional effect but as a structural consequence the Fed had modeled and accepted. Publicly, the regulation was described through the language of systemic resilience. The permanence of the resulting reserve-demand shift was not presented as the architectural intention it represented. That gap matters less for whether the mandate was a mistake than for what it implies about reversal: if the Fed built the structure knowingly, proposals to unwind it are redesigns of a deliberate regime, not corrections of an overlooked glitch.

The archival record shows that liquidity self-insurance was a deliberate, widely shared institutional conclusion — drawn by officials across the ideological spectrum, from Lacker's pre-crisis moral hazard warnings to Tarullo's post-crisis codification — rather than an accidental post-crisis reflex. That finding materially damages Bessent's "fundamental mistake" framing. It does not, however, settle every dimension of the question. Policymakers openly documented calibration uncertainty as the framework took shape: Dudley described stress-scenario standards as "relatively untested" and subject to an observation period to identify unintended consequences; Tarullo acknowledged a need for further conceptual work on several parameters; and staff briefings noted that general-equilibrium effects could not be fully modeled in advance. The 30-day stress horizon and specific runoff rates were empirically grounded but acknowledged as partly novel. A policy can be deliberately designed and still carry acknowledged uncertainty in its calibration. The investigation therefore moves to narrower ground: even granting that the mandate was purposeful, was the calibration careful or arbitrary, and did the framework itself create the very discount window stigma Bessent attributes to it?

The Stigma That Was Already There

With the self-insurance mandate documented in the institutional record as consensus rather than regulatory overreach, Bessent's causal claim moves to narrower ground: not that the framework created stigma, but that it entrenched a pre-existing condition — locking in a signaling dynamic that might otherwise have been resolved. "By driving banks to exhaust regulatory buffers before accessing the discount window, we have entrenched discount window stigma. If you only go to the window when things are really bad, then going to the window signals that things are really bad." That is a different and more careful claim than origin, and the archive must be read against the version Bessent actually makes.

Staff documented stigma as a structural equilibrium as early as 2002, years before the LCR framework existed. The 2003 Primary Credit redesign was the Fed's explicit attempt to solve it — a new facility with cleaner pricing, reduced bureaucratic friction, and an active communications effort to signal that window use carried no supervisory judgment. Brian Madigan, Director of the Division of Monetary Affairs, told the Committee what that effort produced:

The redesign failed because the problem was informational. The signal of window access — what it reveals to counterparties about a borrowing institution's condition — cannot be removed through administrative redesign. By September 2007, when the Term Auction Facility was under discussion as an alternative, the Committee already understood this. Madigan told them directly:

Governor Randall Kroszner framed the TAF's appeal in terms that made the underlying dysfunction explicit: the new facility had "a chance, by no means a certainty but a chance, to overcome that stigma" precisely because it carried none of "the same baggage of the discount window." Vice Chairman Donald Kohn observed that the Bank of England was experiencing the same problem — establishing stigma as a feature of central banking architecture, not a consequence of any particular national regulation.

The TAF was publicly described at launch as an innovative auction mechanism for distributing liquidity more efficiently. The structural failure of the 2003 redesign that had necessitated it was not part of that public account. Internally, the Committee understood it had constructed an emergency bypass around a broken tool. Publicly, the window remained available.

What the archive establishes with precision is that stigma predated the LCR by over a decade and was structural before the framework existed. What the archive does not cleanly settle is how much the LCR specifically deepened that condition once it was finalized in 2014, as against the contribution of other post-crisis changes operating in parallel. Dodd-Frank introduced disclosure requirements and reporting mandates that increased the observability of window access; supervisory examination conventions created independent disincentives to buffer drawdown; and market conventions around minimum operating buffers hardened informally alongside the regulatory ones. Internal deliberations in the 2011–2016 period do not clearly isolate the LCR's marginal contribution to post-crisis stigma from these co-equal channels. That indeterminacy does not rehabilitate the entrenchment claim — stigma's resistance to every institutional remedy suggests the informational mechanism is robust to changes in any single regulatory parameter — but it cautions against reading the archive as a precise accounting of the LCR's specific share of responsibility.

The Liquidity Coverage Ratio was finalized in 2014. Stigma was structural by 2002. It had resisted a full institutional redesign by 2007. By December of that year the institution had routed around it rather than through it — all before the framework Bessent names existed. The archive therefore refutes any origin claim with force. On entrenchment, the finding is narrower: the evidence is consistent with the LCR having reinforced a signaling equilibrium that was already self-sustaining, but the archive does not isolate the LCR's contribution from the disclosure and supervisory channels operating alongside it. The prescription Bessent offers — giving discount window access formal credit within the LCR — must in either case explain what mechanism of regulatory credit resolves something that a decade of institutional effort, international recognition, and purpose-built facility redesign could not reach, and that multiple post-crisis regulatory changes together appear not to have resolved.

The Circularity They Already Named

Stigma, as the archive has documented, operates as a structural signaling problem — highly persistent and not eliminated by the 2003 primary-credit redesign, TAF-era workarounds, or subsequent facility innovations. Bessent's two remaining claims must carry the weight that argument leaves behind. The first — that LCR calibration was "muddled guesswork" and "two parts data, one part gut" — is not a standalone critique; it is load-bearing infrastructure for the prescriptive proposal that follows. If the calibration was careful and analytically grounded, then giving the discount window formal credit within the LCR was not an option carelessly overlooked. It was an option considered and rejected. The archive contains the rejection, with the reasoning attached.

On calibration, the record is specific. Governor Jeremy Stein noted in April 2013 that quantitative impact studies estimated worldwide incremental demand for high-quality liquid assets at roughly $3 trillion, derived from scenario modeling anchored in actual crisis outflow data. More important is what happened after further analysis: the recalibration in the January 2013 final version reduced the HQLA demand estimate for U.S. banks by roughly one-third, with the global reduction falling in the range of $800 billion to $1.6 trillion. A standard that revises its own requirements downward by a third, in response to updated quantitative evidence, is not operating arbitrarily. The framework embedded an explicit observation period precisely because regulators acknowledged they could not fully model all general-equilibrium effects in advance. Policymakers documented those limits candidly: the 30-day stress horizon was acknowledged as a novel parameter without a long track record; runoff assumptions required structural judgment under a regulatory regime with no historical precedent; and Stein's own framing of the QIS estimates centered on the scale of market impact that regulators needed to monitor rather than on the precision of the estimates themselves. The calibration was empirically anchored and iteratively revised — but it incorporated substantial judgment on parameters that policymakers themselves described as untested. What matters for the prescriptive question Bessent raises is not whether that calibration was perfect but whether the discount-window-recognition option was overlooked or deliberately rejected.

The prescriptive claim collapses on different grounds. Stein addressed the discount window credit proposal directly in the same April 2013 speech:

The circularity is precise: a buffer that counts the backstop as part of the buffer eliminates the buffer's purpose. Stein acknowledged a narrow exception — a committed liquidity facility with an up-front fee, for jurisdictions with structurally insufficient HQLA supply — and was explicit that this priced structure was fundamentally different from simply allowing banks to count central-bank-eligible collateral at no charge. The United States was not the target jurisdiction. The concession was constructed to preserve the self-insurance principle, not to erode it.

The operational history then supplies the empirical anchor. The Term Auction Facility was designed specifically to route around discount window stigma through anonymous auction mechanics. Dudley assessed the result in April 2008:

The workaround intensified the signal it was designed to suppress. (The same recursion — each operational redesign looping back to the signaling equilibrium the institution identified in 2002 — was documented in the Standing Repo Facility's inheritance of identical structural constraints: stigma, counterparty limits, and stress-period failure.) Public communications, meanwhile, framed the discount window's exclusion from the LCR as a principled self-insurance choice without foregrounding the deeper institutional assessment — that the window was operationally unreliable precisely when it would be needed most. The distinction between choosing not to count something and knowing it cannot function when counted was collapsed into clean prudential language.

Bessent presents the DW-in-LCR idea as a structural insight the original architects overlooked. The archive shows it is the position that lost the original argument — examined, named as circular, and set aside for reasons now on the public record. What the archive settles precisely is that unpriced, unconditioned access to the discount window was rejected on circularity grounds: a buffer that treats the backstop as part of the buffer eliminates the buffer's purpose. What the archive does not settle is whether Bessent's specific proposal — capped recognition, conditioned on collateral prepositioning and operational readiness, subject to haircuts and supervisory discretion — changes the circularity mechanism or merely relabels it. Stein's own CLF carveout, reserved for jurisdictions with structurally scarce HQLA and explicitly priced to preserve the self-insurance principle, confirms that the design space between full recognition and none is not empty. The burden falls on the proposal to demonstrate that its conditioning features actually interrupt the signaling recursion Stein identified — rather than reproducing it at a different scale. That question the archive cannot answer in advance; it can only establish where the argument must begin. What remains alongside it is the economic promise: whether capital freed by dismantling these requirements would actually flow toward AI infrastructure and defense supply chains, or whether the binding constraint lies somewhere else entirely.

The Trillions That Never Arrive

The circularity between the LCR and discount window stigma was named inside the institution and left structurally unresolved — a finding that bears on the regulatory architecture but does not answer the economic claim beneath the reform proposal: whether dismantling liquidity requirements would direct freed capital toward the investments Bessent names.

The need he identifies is real. America's AI buildout requires capital at a scale that private credit markets have historically struggled to mobilize alone. The defense industrial base needs long-duration manufacturing investment against timelines that do not fit conventional credit cycles. Domestic supply chains, reshaped by a decade of geopolitical realignment, require lenders willing to commit against multi-year industrial assets. When Bessent observes that "25 percent of large banks' balance sheets are allocated to safe assets — up from roughly 10 percent before the crisis — that necessarily means less lending for mortgages, small businesses, AI hyperscalers, and critical infrastructure", the arithmetic is not wrong. Capital held in Treasuries is capital not held in loans. The aspiration behind "unlocking hundreds of billions — potentially trillions — in new lending capacity" is the aspiration most worth being achievable.

The archive does not dispute the need. It disputes the mechanism.

The archive does not contain a direct test of LCR relief as a transmission channel — the HQLA framework was finalized in 2014, and no subsequent episode produced a regulatory relaxation large enough to trace its lending effects in isolation. What the archive does contain is the closest available historical analogy: the large-scale asset purchase programs between 2009 and 2021, which examined repeatedly whether abundant bank liquidity translated proportionally into real-economy credit extension. QE and LCR relief operate through different channels — reserve injection versus portfolio-composition rebalancing — and that distinction matters for any precise claim about mechanism. What the QE-era record nonetheless establishes is the directional behavior of freed balance-sheet capacity in the absence of sector-specific facility design, and on that question the evidence pointed in a predominant direction while including notable exceptions.

In October 2012, as QE3 was expanding the balance sheet toward its eventual peak, Governor Jeremy Stein documented where the issuance capacity enabled by low borrowing costs was flowing among speculative-grade corporate borrowers:

Two-thirds. Not to AI data centers or defense manufacturing facilities. To retiring existing debt and returning cash to shareholders. Stein was describing how nonfinancial firms used bond proceeds once borrowing costs fell — the demand-side response to easier financial conditions rather than the bank-side decision about how to deploy released reserves. Board staff had identified the complementary supply-side finding. During the September 2011 FOMC meeting, staff economist Mr. Carpenter stated it directly:

The constraint was not that banks lacked liquidity. It was that creditworthy borrowers with bankable projects were not presenting themselves in sufficient volume to absorb expanded supply. Taken together, the QE-era episodes constitute the closest available analogy to the "abundant liquidity → industrial lending" transmission Bessent projects — and in those episodes transmission flowed to financial asset prices, debt refinancing, and shareholder distributions rather than to the industrial credit channels he names. That evidence does not directly test the LCR-specific rebalancing mechanism, but it leaves Bessent's sector-specific forecast historically undemonstrated by the most relevant analogue the record contains. The archive does record a partial exception: targeted facilities with specific collateral conditions — notably the PPP Liquidity Facility, which accepted SBA-guaranteed loans as collateral and supported a defined category of small business lending — did transmit credit to particular borrowers under particular terms. That exception, however, cuts against rather than for Bessent's general claim: it was the specificity of the facility's design, not the abundance of reserves, that directed credit toward the named borrowers. A general relaxation of HQLA requirements operates through a different channel entirely, removing a portfolio-composition constraint without specifying where the resulting balance-sheet capacity flows. (The identical finding — valid diagnosis, broken transmission pipe — appeared in the rate-cut record, where the binding constraints on small business credit were demand-side and structural, not supply-side.) The Tealbook prepared by Board staff for the January 2011 FOMC meeting recorded that "demand for most types of loans appeared to remain subdued, and lending standards likely remained tight for many households and firms." That observation reached back further still: the Greenbook prepared for January 2009 had already found that "demand for loans from both businesses and households at domestic institutions reportedly continued to weaken, on balance, over the survey period." The pattern persisted across the cycle because the constraint was structural, not transient.

A separate question attaches to the specific sectors Bessent names. The archive does not address sectoral financing structures directly, but the structural features of these sectors suggest a poor fit with the proposed mechanism. AI hyperscalers appear to be predominantly financed through capital markets and internal cash flow rather than bank credit; the binding constraint on their investment programs may be equity market access, risk-weighted capital treatment of long-duration assets, or procurement structure rather than HQLA mandates. Defense supply chain financing faces analogous issues of contract certainty, credit tenor, and leverage-ratio treatment that a reduction in liquid-asset requirements does not directly address. The claim that relaxing the LCR would unlock trillions in credit directed toward those sectors asserts a sectoral mapping that the archive does not support and that the structural features of those sectors do not obviously sustain.

Publicly, the mechanism was framed as operative. Chairman Bernanke described Fed operations as "easing conditions in interbank funding markets, thus increasing the willingness of banks to extend loans" — the aspiration stated as the outcome. Internally, the picture the institution documented was different: two-thirds of expanded issuance capacity flowing to financial engineering, loan demand described as subdued across consecutive preparation cycles, and staff naming borrower risk profiles and risk appetite — not liquidity availability — as the binding constraints.

Bessent's diagnosis survives: industrial investment needs capital, and the displacement created by HQLA mandates is real. His prescription does not survive on the terms he states. The historical record shows that abundant liquidity is not sufficient for broad-based industrial lending; the specific mechanism by which relaxing the LCR would yield trillions in targeted credit to AI hyperscalers, defense manufacturers, and domestic supply chains is not established by the archive and is not supported by the transmission evidence the institution accumulated across the episodes it documented most carefully.