Lender of Last Resort

Why banks need a lender of last resort

A bank's business model is a maturity transformation: deposits that can be withdrawn on demand are used to fund loans that mature over years or decades. In normal times, only a small fraction of depositors withdraw on any given day, and the bank holds enough cash and short-term assets to cover that flow. In a panic, the fraction spikes. If depositors line up to withdraw faster than the bank can convert long-dated loans to cash, the bank fails — not because its assets are worth less than its liabilities, but because the assets cannot be liquidated quickly enough at fair value.

This is the distinction the doctrine of lender of last resort is built on: a liquidity crisis (the bank is solvent but cannot meet payments on demand) versus a solvency crisis (the bank's assets are genuinely worth less than its liabilities). A central bank that lends against the bank's good assets at their non-panic value can break the panic without socialising any losses — the bank pays back the loan once the run ends, and the central bank earns interest in the meantime. A central bank that lends to an actually insolvent bank takes a real loss and effectively bails out the bank's creditors.

The trouble is that under panic conditions the distinction is hard to draw in real time. Asset prices have collapsed, mark-to-market would crystallise losses, and judging which institutions are temporarily illiquid versus terminally insolvent is the central bank's hardest call. Bagehot's response was not a formula but a policy stance — lend, and lend on terms strict enough that the question takes care of itself in expectation.

Walter Bagehot and Lombard Street

The doctrine takes its modern form from Walter Bagehot's 1873 book Lombard Street: A Description of the Money Market, written in response to the Overend, Gurney & Co. collapse of 1866. Overend Gurney was the second-largest discount house in London — a "bankers' bank" that funded itself with short-term borrowing and lent to commercial banks. When it failed on 10 May 1866 ("Black Friday"), the panic spread through the City and over 200 firms went under within a few months. The Bank of England's belated lending stopped the wider panic, but Bagehot's argument was that more aggressive action — earlier, and on Bank-of-England terms — could have prevented the contagion entirely.

Bagehot's prescription, paraphrased from Lombard Street:

"That these advances should be made on all good banking securities, and as largely as the public ask for them. … The end is to stay the panic; and the advances should, if possible, stay the panic. And for this purpose there are two rules: First. That these loans should only be made at a very high rate of interest. … Secondly. That at this rate these advances should be made on all good banking securities, and as largely as the public ask for them."

Compressed into the modern slogan, the doctrine is three rules:

• Lend freely. Do not ration. The public must see that cash is available without limit, because the run is driven by the belief that it is not.
• At a penalty rate. Charge above the normal market rate, so only the genuinely distressed borrow and so the facility does not become a permanent subsidy.
• Against good collateral. Accept any sound security at its pre-panic value, but refuse loans to firms whose collateral is itself unsound. This is the line between liquidity support and solvency bailout.

The three rules form a system. Without "freely," depositors are not reassured. Without "penalty," banks come to the central bank as a matter of course rather than only in genuine distress. Without "good collateral," the central bank becomes the underwriter of bad lending decisions.

The Fed's failure in 1929-32

The clearest historical demonstration of what happens when the rule is ignored is the United States in the early 1930s. Between 1929 and 1933 roughly 9,000 American banks failed — about a third of the country's banks. The Federal Reserve, established only in 1913 in part to perform exactly the lender-of-last-resort function, did not stop the cascade. The money supply contracted by about a third. Industrial production fell by 47 percent. Unemployment rose to 25 percent.

Friedman and Schwartz's 1963 A Monetary History of the United States argued that this was substantially a Fed failure. The "liquidationist" faction — articulated by Treasury Secretary Andrew Mellon's "liquidate labor, liquidate stocks, liquidate the farmers, liquidate real estate" — held that bad banks should fail to purge the system of unsound credit. That position is the direct opposite of Bagehot. The Fed could have lent to the surviving banks against their loan books at non-panic values, broken the run, and preserved the credit system. Instead it watched the system implode.

Ben Bernanke's later research (the 1983 "Nonmonetary Effects of the Financial Crisis" paper, citing this episode) added the credit-channel argument: bank failures destroy lending relationships and the borrower-specific information embedded in them. Those relationships take a decade to rebuild. The 1930s Fed's failure to act was not just a monetary error; it was a destruction of productive economic infrastructure.

When Bernanke became Fed chairman, this history was an explicit operating doctrine. The lesson he carried into 2008 was, in his own paraphrase: "We did not allow that to happen this time."

The Fed's alphabet soup, 2008-09

By September 2008 the US financial system faced a panic that started in mortgage-backed securities, ran through the investment banks, and threatened to seize up money markets used by every American corporation for short-term funding. The Fed's response was a coordinated rollout of emergency lending facilities, each targeting a specific category of stressed counterparty.

Facility	Acronym	Counterparties	Collateral	Peak outstanding
Term Auction Facility	TAF	Depository banks	Discount window collateral	$493B (Mar 2009)
Primary Dealer Credit Facility	PDCF	Investment banks	Investment-grade securities	$148B (Oct 2008)
Term Securities Lending Facility	TSLF	Primary dealers	Treasuries against MBS	$236B (Oct 2008)
Asset-Backed Commercial Paper MMMF Liquidity Facility	AMLF	Banks lending to MMFs	ABCP held by money funds	$152B (Oct 2008)
Commercial Paper Funding Facility	CPFF	SPV; direct CP issuers	3-month CP	$351B (Jan 2009)
Money Market Investor Funding Facility	MMIFF	SPV; money funds	Bank short-term debt	(never drawn at scale)
Term Asset-Backed Securities Loan Facility	TALF	Investors in ABS	AAA consumer/business ABS	$48B (Mar 2010)
Central-bank liquidity swaps	—	Foreign central banks	(reciprocal swap line)	$586B (Dec 2008)

At peak, emergency lending plus liquidity swaps was above $1.5 trillion outstanding. The Fed's balance sheet expanded from about $0.9 trillion before the crisis to $2.3 trillion within a few months. Every facility was structured as a Bagehot-style lending program — collateralised, at penalty rates above market, with haircuts that protected the Fed against collateral price falls.

The outcomes vindicated the approach. Nearly all of the emergency loans were repaid with interest. The combined facilities generated a profit for the Treasury. None of the major facilities took meaningful losses on principal. The criticism that came later was not that the Fed lost money but that the rescue was so successful that the moral-hazard tax it created was understated.

"Whatever it takes" — the ECB in 2012

Two years later the euro area faced its own variant of the problem. By mid-2012 Italian 10-year yields had crossed 7 percent and Spanish yields were not far behind — levels at which the long-term solvency of those sovereigns was in doubt and at which the survival of the euro as a currency was being openly discussed in markets.

The euro-area version of the LOLR problem is more constrained than the Fed's. The ECB cannot directly fund a member-state government (Article 123 of the TFEU prohibits direct monetary financing of governments) and has historically been reluctant even to provide indirect liquidity to sovereigns. But the line between sovereign and bank liquidity is thin: peripheral banks held enormous quantities of peripheral sovereign bonds, and a sovereign spiral would have crushed bank balance sheets.

On 26 July 2012, at a conference in London, ECB President Mario Draghi delivered the now-famous lines:

"Within our mandate, the ECB is ready to do whatever it takes to preserve the euro. And believe me, it will be enough."

Two months later, in September 2012, the ECB unveiled Outright Monetary Transactions (OMT): an open-ended commitment to buy unlimited quantities of distressed peripheral sovereign bonds in the 1-to-3-year maturity range, conditional on the country signing an ESM bailout program. The structural feature was the open-endedness — there was no cap. The conditionality preserved fiscal discipline.

OMT has never been used. The announcement alone collapsed peripheral spreads — Italian 10-year yields fell from 6.6 percent at the time of the speech to 4.4 percent by the end of 2012, and to 1.6 percent by 2014. The crisis ended without a single bond being bought. This is the cleanest historical demonstration that a credible LOLR commitment can be self-fulfilling: the asset never has to be exercised because the option's existence stabilises expectations. The earlier Long-Term Refinancing Operations (LTRO) of December 2011 and February 2012, which extended €1 trillion of three-year loans to euro-area banks at 1 percent against expanded collateral lists, played the parallel role at the bank level.

SVB and the BTFP, 2023

On 8-10 March 2023, Silicon Valley Bank — California's 16th-largest US bank by assets — suffered the fastest deposit run in modern history. $42 billion was withdrawn in a single day, with another $100 billion queued for the next morning. The bank's specific weakness was a securities portfolio of long-duration Treasuries and agency MBS purchased during the zero-rate era, whose mark-to-market value had collapsed as the Fed raised rates from 0 to 4.5 percent in 2022. The bank was technically still solvent on a hold-to-maturity basis but would be insolvent if forced to mark to market.

The Fed's response was the Bank Term Funding Program (BTFP), announced on 12 March 2023. Its central design feature was the par-value rule: BTFP let banks pledge eligible Treasury and agency-MBS collateral at par (face value), not at depressed market value. Banks could borrow up to the full face amount of their securities at the one-year overnight index swap rate plus 10 basis points, for up to one year. The economic effect was to give banks an immediate cash advance against their underwater bond portfolios without forcing the recognition of mark-to-market losses.

BTFP peaked at about $165 billion outstanding in early 2024 and was closed to new lending in March 2024. It worked. No other regional bank suffered a comparable run after BTFP launched, though three did fail in the immediately following weeks (Signature Bank, Silvergate, First Republic) on more bank-specific issues. Critics noted that BTFP's par-value design quietly stretched Bagehot's "good collateral at its non-panic value" — accepting Treasuries with $20 billion of unrealised losses at face value is, in some sense, exactly what Bagehot warned against. Defenders argued that Treasuries are by definition the highest-quality collateral and that the mark-to-market losses were a pure interest-rate-risk artifact, not a credit-quality issue.

Moral hazard and constructive ambiguity

The standing critique of the LOLR is that the existence of the rescue option creates moral hazard. If a bank's management and creditors expect that a panic will trigger central-bank support, they will run thinner capital, take on more interest-rate risk, hold less liquid asset portfolios, and fund themselves more cheaply with short-term wholesale debt. The ex-post benefit of stopping the panic is bought with an ex-ante distortion that makes panics more likely. This is the too-big-to-fail problem in its purest form: the threat of a systemic event becomes an implicit guarantee, and the guarantee becomes a competitive subsidy.

Bagehot's penalty rate is the intended antidote. If borrowing from the central bank is genuinely expensive, only banks in real distress will take the option; in equilibrium, the rescue is not free, and the moral-hazard tax is correspondingly smaller. The empirical record on this is mixed. In 2008-09 banks borrowed at the Fed's facilities in enormous volumes despite penalty pricing, because the alternative was collapse — at that margin, the penalty does not deter. In quieter periods, however, banks famously refuse to borrow at the discount window precisely because of the stigma and price, and the Fed has spent decades trying to reduce that stigma so the window actually functions.

Mervyn King's 2009 doctrine of constructive ambiguity is the institutional answer: never commit in advance to rescue any particular institution or any particular event. By leaving the existence and scale of support deliberately uncertain, the central bank preserves discretion and forces banks to behave as if rescue might not arrive. The cost is that the very uncertainty can deepen a panic in real time; the benefit is that the implicit guarantee cannot be capitalised in advance into bank funding costs.

Modern policy combines all three responses: a more transparent rule-based system for routine liquidity (discount window, standing repo facility) that should be used unstigmatised; constructive ambiguity preserved for genuinely systemic interventions; and a separate regulatory regime — capital requirements, stress tests, resolution authority, living wills, Total Loss-Absorbing Capacity — that attempts to suppress the underlying risk-taking incentive directly rather than through pricing the option.

Variants and adjacent doctrines

• Domestic LOLR. The traditional Bagehot role — central bank lends to domestic banks against domestic collateral. Most discount-window operations fall here.
• International LOLR. A central bank lends in its own currency to foreign banks (typically via swap lines with foreign central banks). The Fed extended $586 billion of swap lines in late 2008 to fix a global shortage of dollar funding among foreign banks; that line, not the domestic facilities, was arguably the single largest 2008 intervention.
• Market-maker of last resort. Rather than lending to a specific bank, the central bank buys or sells a market-wide asset to stabilise its price. The Bank of England's October 2022 gilt-market intervention during the LDI pension-fund crisis is the canonical recent example: £19 billion of long-dated gilt purchases over 13 days, sterilised afterward.
• Investor of last resort. The central bank purchases assets outright, not as collateralised loans. This is closer to quantitative easing than to Bagehot lending; the Fed's mortgage-backed-securities purchases in 2008 sat on this line.
• Fiscal LOLR. Direct government recapitalisation of failing banks (TARP in the US, ESM bank recapitalisations in the euro area). When losses are real rather than mark-to-market, the central bank cannot absorb them without monetisation; the Treasury must.

Common confusions

• LOLR is not quantitative easing. QE is balance-sheet expansion to ease monetary policy after the policy rate hits zero — a deliberate, long-run, system-wide tool. LOLR is short-run emergency lending to specific institutions under stress. The two can coexist (the Fed did both in 2009) but they are doing different jobs.
• "Bailout" is not the same as LOLR. A bailout, strictly, is a transfer that absorbs losses on bad assets. LOLR is a collateralised loan against good assets that is expected to be repaid with interest. The distinction matters: most LOLR loans in 2008 were repaid; the bailouts (TARP recapitalisations, Fannie/Freddie, AIG) cost the Treasury real money.
• Penalty rate vs. punitive rate. Bagehot's "high rate" was the contemporary high-quality market rate, not a punitive rate. Modern penalty rates are typically 50-100bp above the policy rate — meaningful, but not so high as to deter use in a real panic.
• "Good collateral" is endogenous. A bond that is good collateral in normal times may not be in a panic, and vice versa. The central bank's willingness to lend against an asset class at non-panic values is itself the policy lever that determines whether the collateral remains "good."
• Solvent-but-illiquid is hard to verify. In a panic, every borrower claims to be solvent-but-illiquid; some are insolvent and unwilling to admit it. The central bank's haircuts, collateral discounts, and inspection rights are the institutional machinery for forcing the question. Failures of the LOLR doctrine almost always come from getting this distinction wrong, in either direction.