Banks play a vital role in the economy by issuing demandable deposits and extending long-term credit to households and businesses. This liquidity transformation supports economic growth, but may also be a source of fragility. If a large fraction of depositors run, a bank may not be able to meet the redemptions, as most assets consist of long-term loans that cannot be quickly turned into cash. As famously determined by Walter Bagehot, the solution to a panic is for the central bank to lend early and freely to banks against assets that are good in normal times. Current regulatory policy would surprise Bagehot. Contrary to his recommendation, regulators and bank examiners penalize borrowing from the central bank and instead insist that banks hold large amounts of reserves and Treasuries to prevent them from having to do so.
In particular, the liquidity coverage ratio (LCR) requires banks to hold enough high-quality liquid assets (HQLA) to withstand a hypothetical 30-day run. Under the spirit of the rule, the liquidity buffer stored in the form of HQLA should be used to meet redemptions. However, the liquidity buffer must be maintained at all times, effectively prohibiting the bank from using the buffer for its intended purpose (Nelson, 2019). Recognizing that the LCR is an unusable buffer, banks must build an additional, usable liquidity buffer on top of the HQLA needed to meet the LCR’s required 1:1 ratio of liquid assets to runnable liabilities. We show a numerical example where a bank with relatively stable liabilities must maintain an LCR of 175% in normal times to barely satisfy the LCR in a stress event. Requiring banks to hoard excess liquidity reduces credit provision to the real economy and acts as a tax on banks.
Furthermore, the LCR introduces several distortions, as we document in this note. First, we show that the LCR tax is higher for banks with safer liabilities, which is at odds with the spirit of the rule. Second, we show that to meet LCR requirements, banks may be forced to fire-sell illiquid assets in a stress event, which increases bank fragility. Additionally, banks with more stable liabilities may have to fire-sell more illiquid assets. The same LCR that was meant to make banks self-sufficient and not reliant on central bank interventions ends up increasing fragility and hence the likelihood of central bank interventions.
Regulators acknowledged that liquidity rules could produce financial stability gains at the expense of long-term costs, such as a lower supply of credit to the real economy. But we show that the LCR is doubly inefficient: (i) it forces banks to create less credit to the real economy in the long run, as expected, but also (ii) it creates perverse incentives to hoard liquidity and fire-sell illiquid assets during a stress event, which increases fragility. In other words, the LCR, as designed and implemented, is all cost, no benefit.
A primary reason for implementing liquidity regulations is the presence of excessive fire sales during stress events. Thus, we should not have the LCR exacerbating the very same problem it aims to fix. When a prudential regulation taxes the safest banks most and provides perverse incentives to hoard liquidity and fire-sell illiquid assets, that rule should at least be adjusted. In this regard, we delineate an adjustment to the minimum LCR requirement that would allow banks to use the stored liquidity while still satisfying the LCR.
Furthermore, the LCR approach of requiring banks to rely solely on stored liquidity is counter to proper liquidity risk management, which instead stresses the ability of strong banks to purchase liquidity in a crisis (Nelson, 2023). Sound banks can raise cash against the pledge of good collateral. There are several avenues to do that, including from repo markets, Federal Home Loan Banks (FHLBs) and the discount window. But bank examiners have traditionally discouraged banks’ reliance on the discount window for funding. In contrast, we show that a destigmatized discount window eliminates the distortions imposed by the LCR.
1. The Liquidity Coverage Ratio
The LCR rule requires banks to hold enough HQLA to withstand a hypothetical 30-day run. Specifically, banks must maintain their LCR, defined below, above 100%:
HQLA can be of three types. Level 1 assets include central bank reserves, U.S. Treasury securities and federal agency obligations backed by the full-faith and credit of the U.S. government. No haircut is applied to Level 1 assets, so that $100 worth of Treasuries count as $100 of HQLA. Level 2A assets, which include debt and mortgage-backed securities (MBS) issued by government-sponsored enterprises, are instead subject to a 15% haircut. Lastly, a 50% haircut is applied to Level 2B assets which include investment-grade nonfinancial corporate bonds, investment-grade municipal debt and equities listed on major indexes.[1] Other assets, such as consumer loans, do not count as HQLA. Importantly, assets count as HQLA if they can be immediately turned into cash and are not encumbered.
The denominator of the LCR measures net cash outflows over a 30-day hypothetical stress event. Net cash outflows represent the amount of cash the bank owes its creditors in addition to the amount it is obligated to provide under pre-existing lines of credit, net of the amount of cash it receives from its performing assets. The LCR weighs each liability that becomes due within 30 days by a presumed runoff rate. For instance, unsecured wholesale funding due within 30 days has a 100% runoff rate, implying that lenders would not roll over any wholesale funding to the bank in case of a stress event. By contrast, federally insured retail deposits are subject to a 3% runoff rate. Additionally, since credit lines can be drawn down at any time, the LCR assumes that some undrawn credit lines will be drawn down in a stress event, giving rise to a cash outflow. The LCR rule attaches different runoff rates (ranging from zero to 100%) to undrawn credit lines depending on the counterparty type. The sum of bank liabilities and pre-existing lines of credit weighted by runoff rates gives rise to the 30-day Cash Outflow.
At the same time, banks have cash inflows, such as payments on performing loans that are owed to the bank. Each type of incoming payment is weighed by an inflow rate, and the weighted sum represents the 30-day Cash Inflow. Importantly, the LCR rule does not allow for a full offset of cash outflows with inflows, and it caps the 30-day Cash Inflow at 75% of the 30-day Cash Outflow. The difference between 30-day Cash Outflow and the capped 30-day Cash Inflow represents the denominator of the LCR. [2]
2. Some Unpleasant LCR Arithmetic: A Simple Example.
To understand why the LCR is internally inconsistent, let us first consider a simple example. Assume that a bank funds itself with $100 in deposits that are subject to a 25% LCR runoff rate, resulting in 30-day cash outflows of $25. The bank invests primarily in loans that do not generate any incoming payments within the next 30 days. As a result, 30-day cash inflows equal zero and net cash outflows are $25. To satisfy the LCR, the bank must hold at least $25 in HQLA, which we assume the bank satisfies only with Level 1 assets (e.g., central bank reserves). Thus, HQLA0 = $25 and D0 = $100, where the subscript denotes the time period, with zero being the initial period and one the crisis period. The bank is compliant at time zero, with an LCR0 of 100%.
If the hypothetical scenario envisioned by the LCR occurs, namely that 25% of deposits run, one would assume that the bank would still be LCR-compliant during the stress event, but that is not the case. Suppose that 25% of the deposits run, in line with the assumed runoff rate, and that the bank meets redemptions with its HQLA. At that point, the bank is out of compliance because it no longer has any HQLA left to meet any future run on its remaining liabilities. At time one, the bank has no HQLA left against the remaining $75 of deposits, resulting in an LCR1 of zero.
But banks must maintain a minimum LCR of 100%,[3] which implies that the bank must hold HQLA1 equal to $18.75, namely 25% of the remaining deposits D1. To do that, the bank can only accommodate $7.25 out of the $25 in deposit redemptions by liquidating its stock of HQLA, and the remaining $18.75 of redemptions by selling non-HQLA assets, such as loans.
To meet redemptions in the crisis period while still satisfying the LCR, the bank must either: (i) hold excess HQLA before the crisis; or (ii) fire-sell illiquid assets during the crisis period. This is clearly not the scenario envisioned by regulators, but it is nevertheless the outcome of the LCR rule.
If the bank wishes to avoid fire-selling illiquid assets, it must hold excess HQLA to satisfy the LCR while still being able to meet redemptions during the crisis. This is the reason why the LCR is internally inconsistent. It requires the bank to hold HQLA to meet a 25% depositor run. But when the 25% run occurs and HQLA is used to meet the redemptions, the LCR continues to require that HQLA is held to meet an additional 25% run from the remaining 75% of depositors. However, the HQLA has already been fully used to meet the actual run, so no HQLA is left to meet an additional hypothetical run on the remaining deposits.
To be able to accommodate redemptions with HQLA while still satisfying the LCR at time one, HQLA0 has to equal the HQLA required at time one plus the expected withdrawal of deposits at time one. This LCR buffer can be quite substantial. For instance, to accommodate a 25% redemption (the same magnitude expected by the rule) while still satisfying the 100% minimum LCR requirement during the stress event, the bank needs an LCR of 175% in normal times. See the Appendix for the full derivation.
3. A General Framework.
We now build a simple dynamic model to document the many distortions that result from the LCR. Consider a bank that must satisfy the LCR in two periods, time zero representing normal times and time one representing a crisis. The bank funds long-term loans with deposits, D, that have a runoff rate of r. Against them, the bank holds high-quality liquid assets, HQLA. The liquidity coverage ratio at time zero, called LCR0, is defined as the ratio of HQLA to the projected amount of deposit withdrawals:
In period one there is a run on the bank. A fraction R of deposits run, where R may be different from the hypothetical runoff rate r. In line with the spirit of the LCR, the run is accommodated by liquidating HQLA. Thus, HQLA at time one equals the previous amount of HQLA net of deposit withdrawals:
Next, the quantity of deposits left at time one equals the fraction of time zero deposits that did not run:
Finally, the bank’s LCR at time one equals the time-one HQLA divided by the fraction r of current deposits D1 that are projected to run:
Equations (1) to (4) allow us to show the dynamic inconsistency of the LCR regulation. Combining these equations, we obtain
Suppose that the actual run at time one is exactly of the magnitude hypothesized by the LCR, so that r = R. Equation (5) simplifies to the following law of motion for the LCR:
Result 1 (The LCR Tax). If the bank is expected to maintain an LCR of 100% even in a crisis period (LCR1 = 1), then it must maintain an LCR well above 100% in normal times. Indeed, requiring that LCR1 = 1, equation (6) implies that LCR0 = 2 – r, which is 1 – r more than the minimum requirement of 100%. We can call this excess liquidity (1 – r) the LCR tax.
Note that by requiring banks to hold excess liquidity, the LCR forces banks to provide less credit to businesses and households, a long-run cost.
Result 2 (The Safe Bank Distortion). The LCR tax, 1 – r, is the most severe when the bank funds itself with the most stable liabilities. Indeed, as r approaches zero, the LCR tax reaches its maximum of 100%. On the contrary, the tax goes to zero if the bank funds itself with the least stable liabilities. Indeed, as r approaches one, the tax goes to zero. Therefore, the LCR distorts stable funding sources the most and unstable funding sources the least.
For example, consider a bank that funds itself only with insured retail deposits (subject to a 3% runoff rate). This institution faces an LCR tax of 97%, meaning that it must carry an LCR of 197% in normal times to remain compliant with the LCR rule in a crisis. On the other hand, a bank that funds itself only with short-term unsecured wholesale funding (such as one-week commercial paper) which faces an assumed 100% runoff rate would have no LCR tax and can keep an LCR of 100% at all times.
Result 3 (Countercyclical Buffer). Suppose instead that banks hold an LCR of 100% in good times. Equation (6) then requires that the LCR goes to zero in a crisis. It follows that the LCR must be conceived as a countercyclical buffer to be drawn down in a crisis.
Taking the derivative of LCR1 in equation (5) with respect to the intensity of the run, R, gives a negative number under reasonable assumptions, which means that the LCR must be allowed to adjust downward after a deposit outflow. Even when a run occurs that is less intense than expected under the rule, the LCR must be allowed to dip under 100%. Indeed, consider the following two scenarios.
Scenario 1: Normal Times. With no run (R = 0), equation (5) becomes: LCR1 = LCR0. In other words, in normal times the LCR can remain constant.
Scenario 2: Mild Run. With a milder-than-expected run (R = 10%, r = 25%), equation (5) becomes: LCR1 = 1.1 × LCR0 – 0.44. In other words, in bad times the LCR must decline, even when the run is less intense than expected (R < r). Indeed, with LCR0 = 100% and a 10% run on deposits that were expected to run at 25%, the example indicates that LCR1 = 1.11 – 0.44 = 67%.
In sum, Basel III should allow the minimum LCR requirement to dynamically adjust downward according to equation (5). Requiring that the LCR stays above 100% at all times is inconsistent with the spirit of the rule. As shown above, a bank that experiences even milder-than-expected outflows should be able to use its stored liquidity as intended by the LCR rule and, in doing so, it would mechanically see its LCR go below 100%. Table 1 shows how the minimum LCR should be adjusted downward for different values of runoff rates r (on the horizonal axis) and actual run intensities R (on the vertical axis).
Table 1: Minimum LCR During a Crisis.
Note: Table 1 shows the minimum LCR in a crisis period as by equation (5), assuming that the bank keeps an LCR of 100% in the pre-crisis period. We also assume that the actual run rate is never greater than the assumed runoff rate and hence some cells in the lower-left corner are left empty.
The main message is that an LCR below 100% after an outflow is not in violation of the rule; instead, it is perfectly consistent with the rule. However, the current LCR rule stigmatizes the use of HQLA to meet outflows as intended by the rule itself. As economists would say, the LCR is internally inconsistent. The LCR rule concedes that the LCR could drop below 100% for two reasons. First, in case of “unanticipated liquidity needs,” which is not correct. As shown above in Scenario 2, a bank that faces a 10% outflow of deposits that carry a 25% runoff rate, thus experiencing less-than-expected liquidity needs, can see its LCR go well below 100%.
Second, regulators state that the LCR could drop below 100% as a result of “significant deficiency in a [bank]’s management of liquidity risk”.[4] Thus, any drop below 100% triggers immediate and heightened supervisory scrutiny. Bank managers want to avoid this type of scrutiny and therefore keep the LCR above 100% at all costs. To do that, banks keep sizable liquidity buffers, with LCRs often above 120%. This is stigma all over again. Indeed, as documented in BCBS (2021) and Nelson and Waxman (2021), banks avoid using liquidity buffers in order to avoid regulatory scrutiny.
3.1 Allowing Liquidations of Illiquid Assets.
One may argue that the previous exercise is too restrictive because we require banks to meet redemptions only with HQLA (which is by the way the intent of the LCR rule). Next, we relax that assumption and allow for a more flexible model. Banks sell HQLA to accommodate only a portion p of deposit withdrawals and meet the remaining portion (1-p) of withdrawals by selling illiquid non-HQLA assets, such as loans, at a fire-sale discount. In particular, the law of motion of HQLA, equation (2), becomes
while equations (1), (3), and (4) remain unchanged. Following the same procedure as before, including the simplifying assumption that that r = R, leads to
Result 4 (The Safe Bank Shall Fire-Sell More). Suppose that the bank is satisfying the LCR requirement pre-crisis (LCR0 = 1). Equation (6b) then implies that to also satisfy the LCR requirement during the crisis (LCR1 = 1), we need p = r. Hence, a bank with stable funding (r = 3%) can only meet a very small fraction (p = 3%) of redemptions by selling HQLA and a much larger fraction, 97%, by selling illiquid assets at a fire-sale discount. Safer banks need to engage in more fire-sales of illiquid assets to remain LCR compliant. On the other hand, a bank that uses very unstable funding (r = 100%) can accommodate the full outflow with HQLA, without having to sell any illiquid assets. In other words, the LCR penalizes banks with more stable funding, another distortion.
4. Possible Solutions, Including the Discount Window
One solution to the dynamic inconsistency of the LCR would be to allow the minimum LCR to adjust downward in a stress event as HQLA are sold for their intended purpose.[5] Yet another solution could be unstigmatized access to the discount window, namely as a term loan that is renewable and prepayable.[6] Indeed, recent changes to the discount window were made specifically to make it LCR-friendly.
The U.S. LCR imposes a 25% run-off rate on overnight borrowing from the discount window against non-HQLA collateral. Thus, in a stress event, a bank would have to keep 25% of the amount borrowed from the central bank as HQLA instead of using it to meet redemptions. Put differently, the LCR assumes that the central bank may run on the bank. However, in 2020, some changes to the discount window were made to eliminate this LCR distortion. Specifically, discount window loans could be extended for as long as 90 days. Importantly, a 90-day loan does not trigger an outflow under the LCR. However, a 90-day loan eventually becomes a loan with less than 30 days of residual maturity, at which point it generates outflows under the LCR and requires an increase in HQLA. To avoid that, the discount window loan can now be prepaid or renewed before it crosses the 30-day threshold. Thus, a bank can now borrow from the discount window without ever triggering an increase in the LCR denominator.
Borrowing from the discount window is a form of purchased liquidity that allows banks to accommodate deposit withdrawals with the loan receipts. To borrow, a bank needs to pledge collateral to the Federal Reserve. While Treasuries and other liquid securities can be pledged quickly, loans take more time to pledge (Carlson, Styczynski, 2025). Thus, pre-pledging loans allows banks to quickly borrow against them when necessary. Borrowing at the discount window against Treasury (Level 1) collateral does not improve the LCR, as this maneuver would encumber the Treasury security pledged at the window, thus reducing the amount of HQLA in the same way that selling it would. Pledging agency MBS (Level 2A) would provide only a modest benefit, as HQLA would decline by 85% of the amount pledged at the window, which then generates an inflow of cash equivalent to 94% of the pledged amount.[7] The biggest advantage occurs when the bank pledges non-HQLA to the window and borrows for 90 days – that is, if there is no supervisory stigma associated with borrowing from the discount window.
Under the assumption that the bank pledges non-HQLA assets at the discount window, the law of motion of HQLA (equation 2) becomes HQLA1 = HQLA0. The other equations (1, 3, and 4) remain the same. To pay for redemptions, R × D0, the bank borrows that exact amount from the discount window. Since the discount window loan is a 90-day loan that can be both prepaid and renewed, it never generates a 30-day outflow and thus never appears at the denominator of the LCR. Suppose that LCR0 = 100% so that HQLA0 = r × D0. Then,
This exercise shows that pledging non-HQLA to the window allows the bank to keep its LCR above 100%. Since discount window loans tend to be more expensive than deposits, the bank will only rely on them when necessary to stave off a run.
Result 5 (The Discount Window). A 90-day renewable and prepayable discount window loan against non-HQLA collateral allows a bank to manage redemptions without facing LCR distortions.
As just demonstrated, a bank that is willing to borrow from the discount window and has abundant non-HQLA collateral pre-pledged at the Fed will be well-positioned to withstand a run. While this ex-post use of the discount window eliminates LCR distortions and allows a bank to always keep an LCR of 100% (as shown in Result 5), there is an even more balance-sheet-efficient way to incorporate the borrowing capacity from the discount window ex-ante. Indeed, if the LCR is intended to evaluate the liquidity of a bank in a forward-looking manner, the capacity to borrow ex-post should be reflected in the metric. It is illogical that the discount window only improves the metric ex-post when the borrowing occurs. The LCR recognizes that undrawn lines of credit are likely to be drawn down in a stress event, leading to hypothetical outflows that increase the denominator of the LCR. In the same way, the LCR rule should recognize that banks could borrow from the discount window against their pre-pledged collateral. Along these lines, Waxman (2025) proposes a method that subtracts “proven discount window capacity” from the denominator of the LCR. Doing so would free up balance sheet space and allow banks to provide more credit to businesses and households instead of holding excessive amounts of reserves and Treasuries.
Conclusion
We have shown that the LCR provides perverse incentives to both hoard liquidity during a stress event and accommodate outflows by selling illiquid assets. Banks have three options: hold an outsized liquidity buffer well above the 100% minimum in normal times, which reduces credit provision to households and businesses; fire-sell illiquid assets in a stress scenario, which would increase insolvency risk in already challenging times; or borrow from the discount window against non-HQLA collateral. Similarly, we present two options that regulators should consider to eliminate the documented LCR distortions: adjust the minimum LCR requirement downward during a stress event, so that banks can use HQLA without triggering regulatory scrutiny; or promote use of the discount window without any supervisory stigma, including ex-ante recognition of discount window capacity in the calculation of the LCR.
[1] In practice, banks use almost no Level 2B assets to satisfy the LCR.
[2] Additionally, the U.S. implementation of the LCR rule includes a maturity mismatch add-on to the denominator of the ratio. The U.S. LCR final rule is available at https://www.ecfr.gov/current/title-12/chapter-II/subchapter-A/part-249.
[3] See https://www.ecfr.gov/current/title-12/part-249#p-249.10(a).
[4] See https://www.govinfo.gov/content/pkg/FR-2014-10-10/pdf/2014-22520.pdf.
[5] See Result 3 and Table 1. See also https://bpi.com/bpi-comments-on-bank-of-england-and-prudential-regulatory-authority-discussion-paper-on-usability-of-liquid-assets/
[6] See https://www.federalreserve.gov/monetarypolicy/discountrate.htm.
[7] For information on discount window haircuts by collateral type, see https://www.frbdiscountwindow.org/pages/collateral/collateral_valuation.
