Lessons from Applying the Liquidity Coverage Ratio to Silicon Valley Bank
The failure of Silicon Valley Bank (SVB) after a run by uninsured depositors has focused attention on bank liquidity regulation in dramatic fashion.
Less than four years ago, the US bank regulators, following an Act of Congress, ruled that most banks with between $50 billion and $250 billion in assets would no longer be subject to the liquidity coverage ratio, or LCR, or other enhanced prudential standards that they apply to the most systemically important banks. Their 2019 “tailoring” rule is here, a simple chart is here.
The LCR rule requires banks to hold sufficient high-quality liquid assets (HQLA) to manage expected net cash outflows in a 30-day stress scenario. Under the original 2014 version of the rule, banks with $250 billion in assets or $10 billion in foreign exposures had to maintain their LCR ratios above 100%. SVB would have been subject to that ratio because its foreign exposures met the threshold.
We reviewed SVB’s public financials and concluded that its LCR would have been 75% at the end of 2022, substantially below the threshold. This result suggests that the 2019 tailoring rule was complicit in the run and failure at SVB.
Of course, if the bank were subject to the rule, its supervisors would not have allowed its LCR to fall so far. Supervisors shouldn’t just react when a bank breaches a limit—they should act when limit breaches become foreseeable. Even under its existing regulatory framework, SVB’s supervisors should have identified the liquidity risks the company faced due to its high concentration to, and run-inducing dependence on, a specific type of corporate depositor.
Moreover, if the bank had been subject to the rule, it would have been required to publicize more data about its liquidity risks. The market, not just bank supervisors, may have been more focused on the risks it faced. The bank’s managers would have had to manage liquidity risk very differently. To get to compliance, the bank would have needed far more high-quality liquid assets—$18 billion more to get to a 100% LCR, and $36 billion more to get to the 125% LCR that US global systemically important banks maintain on average.
To be sure, compliance with the LCR alone would not have saved SVB’s management from its mistakes managing interest-rate risk in the bank’s massive, long-dated portfolio of agency mortgage-backed securities (MBS). If faced with the LCR rule two or three years ago, they could have simply shifted from long-dated MBS to long-dated Treasuries—bringing their LCR in compliance while staying extraordinarily exposed to rising rates. The bank might have faced the same run after depositors became concerned about the realized losses on Treasuries and unrealized losses on MBS.
To manage the run, the bank would have needed more of the most liquid HQLAs—reserve balances or short-term Treasuries—not more of the long-dated stuff. The LCR as currently written does not distinguish securities by their maturity or recognize that a bank may be reluctant to take losses by selling them.
This brief note describes our calculations (Figure 2) and provides initial conclusions. The caveat is that we were limited to public reports—the company’s 10-K; Y-15 Systemic Risk Report; call reports; and investor presentations. The Fed did require the company to submit the FR-2052A liquidity monitoring report, but that is not publicly available. We believe we have made the most of the information that is publicly available.
Numerator: High-Quality Liquid Assets
HQLAs are relatively easy to calculate. Level 1 assets are considered the most liquid assets and enter the numerator with no haircut. They include reserve balances with the central bank, government debt, and government-guaranteed securities. At the end of 2022, SVB had $31.7 billion in Level 1 assets: $7.8 billion in reserve balances, $16.2 billion in U.S. Treasuries, and $7.7 billion in mortgage securities issued by Ginnie Mae, which the government fully guarantees. Level 2a assets are considered somewhat more risky and receive a 15% haircut. They include government agency debt and agency mortgage-backed securities. SVB had more than $60 billion in agency mortgage-backed securities. However, under the LCR rules, Level 2a assets can’t be more than 40% of total HQLA, so SVB’s Level 2a assets would have been capped at $21.2 billion.
Total HQLA: $52.9 billion.
Denominator: Cash Outflows
The denominator is harder to calculate and open to some interpretation. If SVB had been subject to the rule, its management would have had to hammer out all of the issues in the following paragraphs with its Fed supervisors, using far more detailed data disclosures than the company was actually required to make to the public.
To compute outflows, the LCR rules assume stressed runoff rates for each type of bank liability based on experience in past crises and on management and supervisory judgment. SVB had $173.1 billion in deposits at the end of 2022. Of those, $165.4 billion were uninsured: $151.5 billion were uninsured deposits in U.S. offices that exceeded the FDIC insurance limit; another $13.9 billion were foreign deposits, which the FDIC doesn’t insure (see the company’s last 10-K, page 80). We assume the remaining $7.7 billion was insured. Under the US rules, stressed outflows for insured deposits range from 3% to 40%: 3% for retail and small business customers, and 5% to 40% for wholesale customers that are not financial companies. We’ll assume an average 5% runoff of insured deposits. That yields a small outflow of $0.4 billion.
Calculating outflows on the $165.4 billion in uninsured deposits is the largest part of SVB’s LCR and requires the most judgment. These were also the depositors who ran the bank this month, so it is important to understand how the rule would have treated them ex ante.
Within the uninsured deposit category, the LCR rules makes key distinctions:
- Retail clients are considered more loyal (sticky) and less likely to run than wholesale clients.
- Among wholesale accounts, operational deposits—deposits that customers need to place with a bank in order to use services such as payment settlement systems, and don’t bear interest—are considered stickier than non-operational deposits—which are interest-bearing; the rules assume these depositors are more likely to switch to other banks to get better rates.
- Also among wholesale accounts, nonfinancial company deposits are considered stickier than financial company deposits.
SVB reported its end-2022 short-term wholesale funding in its Y-15 Systemic Risk Report, as the Fed requires. Uninsured, nonfinancial wholesale deposits were $109.6 billion; uninsured, financial wholesale deposits were $28.8 billion. It is typically difficult to estimate from public filings how much of a bank’s deposits are operational, but the company in a January 2023 investor presentation said that 47% of clients’ funds were in “operating cash typically held in on-balance sheet, noninterest-bearing deposits” (see the same investor presentation, p. 11). Accepting the company’s definition of “operating” and applying that proportion yields $65.0 billion in operational uninsured deposits from both financial and nonfinancial customers, which have a 25% run-off factor under the rule; $58.1 billion in non-operational uninsured deposits from nonfinancial customers (40%); and $15.3 billion in non-operational uninsured deposits from financial customers (100%). The combined outflow from uninsured wholesale clients comes to $54.7 billion.
We assume the remaining $27.1 billion of uninsured deposits are retail or small business. Applying the 10% runoff factor under the rule yields an outflow of $2.7 billion.
The LCR rules also require banks to assume that its customers draw down on a portion of their outstanding lines of credit and other commitments. The drawdown rates are 5% for commitments to retail customers, 10% or 30% for nonfinancial wholesale customers, and 40% for financial entities other than banks (12 CFR 249.32 (e)(1)). SVB had $62.2 billion in such commitments at the end of 2022; assuming a 20% average drawdown rate, the outflow would be $12.5 billion.
Finally, SVB had $13.6 billion in short-term borrowings, of which $13.0 billion were advances from the Federal Home Loan Bank (FHLB) of San Francisco. The outflow for such borrowings is 25%, or $3.4 billion.
The total cash outflow comes to $73.7 billion.
Denominator: Cash Inflows
Next, inflows. The LCR allows banks to subtract contractual inflows from their estimate of 30-day outflows, according to the LCR guidelines (para. 142).
Contractual inflows may include loans that are scheduled to mature during the 30-day period. The rules then allow banks to treat half of such repayments as inflows. However, a close look at SVB’s $73.6 billion portfolio of performing loans suggests that it should have counted on very little of those loans to be repaid over 30 days.
First, of that total, $46.0 billion were revolving loans (SVB 10-K, p. 132-33). Borrowers are expected to roll over such loans and they are not included in inflows, per the US final rule: “With respect to revolving credit facilities, already drawn amounts would not have been included in a covered company's inflow amount.” To make clear why such loans should not be considered inflows in SVB’s case, note that about $40 billion of the total represents loans to private equity or venture capital funds. According to the company: “The vast majority of this portfolio consists of capital call lines of credit, the repayment of which is dependent on the payment of capital calls by the underlying limited partner investors in the funds managed by these firms” (SVB 10-K, p.70). Surely, it makes sense that the rules would not expect SVB to be able to demand that venture capital investors repay their revolving loans in stressed market conditions.
Eliminating the revolving loans leaves $27.6 billion in term loans. These included $6.1 billion in “investor-dependent” loans. Borrowers of these loans may require additional equity financing from venture capital firms or other investors, or in some cases a successful sale to a third party or an IPO, according to the company’s 10-K (p. 70). Further, the risk of “prolonged market volatility… particularly applies to Investor Dependent loans, where repayment is dependent on the borrower’s ability to fundraise or exit” (p. 74). Without specific backing from the Basel guidelines or the agencies’ final rule, let’s agree for the purpose of today’s exercise that it wouldn’t be prudent to assume repayment of investor-dependent loans in stressed market conditions. Of the remaining $21.5 billion in term loans, less than $3 billion mature in under one year (p.73). Assuming one-twelfth of those loans pay back at maturity within 30 days and applying the 50% haircut in the rule gives a cash inflow of $0.1 billion.
SVB reported $5.3 billion in deposits at other financial institutions. Under the rules, banks can’t include operational deposits in their inflow estimates, since “these deposits are required for operational reasons, and are therefore not available to the depositing bank to repay other outflows” (LCR guidelines, January 2013, para. 98). The US regulators’ final rule in 2014 similarly excluded “any inflows derived from amounts that a covered company holds in operational deposits at other regulated financial companies… [T]he agencies reasoned that it would be unlikely that a covered company would be able to withdraw these funds in a crisis to meet other liquidity needs, and therefore excluded them.” However, it’s impossible to tell from the public filings how much of SVB’s bank deposits are operational. We assume 50% are operational. That yields an inflow of $2.6 billion, using the 100% factor in the rule.
Cash inflows: $2.8 billion. Net outflows: $71.0 billion.
The resulting LCR is 75%: 52.9/71.0.
It is a myth that the LCR is a complex rule. It is relatively simple to compute for a relatively simple company. The calculations in this brief were not complicated. The challenge lies in working with the available public data for a company that is not required to disclose such data. The exercise also revealed some lack of clarity around definitions, which a bank would have to work out with its supervisors. For those reasons, there was much guesswork involved in our calculations, and the results may be debated.
Still, this exercise should make it clear that SVB would have managed its liquidity risks much differently if it had been subject to the LCR. It would have disclosed more information to supervisors and market participants. Supervisors and market participants would have demanded it hold more HQLAs, probably at a level of 125% or more, where other US companies sit, and potentially higher, given the bank’s high concentration in a risky industry. To get to 100%, SVB would have needed $18 billion more HQLA. Assuming the average actual LCR for US global systemically important banks of 125%, SVB would have needed $36 billion more.
This result is notably different from an analysis published by the Bank Policy Institute, an industry think-tank. We made the following different assumptions: assuming the company would be subject to the full 100% LCR; defining the bulk of SVB’s nonfinancial business customers as wholesale, rather than small business-cum-retail, per the Y-15; finding that the bulk of SVB’s loans would not provide much cash inflows over a 30-day period; and excluding half of inflows from deposits held at other banks, since it’s reasonable to assume much of those would be needed for operational purposes.
The policy community is already rethinking the regulatory tailoring that the banking agencies undertook in 2019. Most jurisdictions—the UK, Europe, most of Asia—have gone in the other direction. They have maintained full LCR disclosure and threshold requirements for all banks, not just the largest ones. At the same time, they have strengthened requirements for larger institutions. Outside the US, the global average is about 140%, according to the Basel committee’s dashboards (see Figure 1). That would have required nearly $50 billion more HQLA for SVB.
While there is broad agreement in the US that most banks are small and should not need to follow enhanced prudential standards designed for large, complex, and internationally active banks, the definition of “small” has crept up beyond any sense or reason. SVB was not small.
The ex-post policy review should also reconsider the “modified” LCR that the bank regulators implemented in 2014. Under the original 2014 rule, the full 100% LCR only applied to banks with more than $250 billion in assets or more than $10 billion in foreign exposures. Other large banks were subject to a modified version that allowed them to hold 30% less HQLAs. Banks with less than $50 billion in assets were fully exempt from the LCR. If US regulators had not changed the rule in 2019, it would have required SVB to hold the full 100% of estimated net cash flows in HQLA because its foreign exposures met the threshold. But similar-sized banks without those exposures would have been required to hold only 70% of that.
The episode suggests other flaws in the LCR rule. Most important, as noted above, LCR compliance would not have saved the bank alone. The bank’s management would likely have made the same mistake with respect to interest rate risk—it might have replaced its long-dated MBS (HQLA Level 2) with long-dated Treasuries (HQLA Level 1) to improve the LCR, and faced the same mark-to-market losses and breakdown in confidence when interest rates rose.
Indeed, the proximate cause of the depositors’ run was SVB’s sale of its portfolio of available-for-sale Treasuries, on which it realized losses, creating the need to raise new capital. It also financed its long-dated HQLAs—mostly agency mortgage-backed securities—with the FHLB; but the FHLB would stop funding the bank if realized losses turned its tangible equity negative (reportedly, there were also operational delays at the FHLB). Underwater HQLAs proved difficult to sell or raise cash against.
The lesson is that securities with mark-to-market losses are not perfectly liquid. But the LCR does not distinguish between short- and long-dated securities, or between securities with unrealized losses and securities trading at par. Perhaps the LCR needs a new approach to held-to-maturity securities, or a penalty for assets that you can't sell without a loss. Alternatively, the LCR could require banks with concentrated uninsured deposits to hold a portion of their HQLA in reserve balances and short-term Treasuries. There are many options to insert maturity and potential losses into the calculation.
The broader lesson is that regulators need to take a holistic approach to the risks banks face, rather than bucketing one set of rules for liquidity and another set of rules for solvency.
Thank you to Carey Mott for research assistance.
 Note that this piece focuses on SVB. A similar analysis of First Republic Bank suggests it also would have been out of compliance with the LCR if it had been required to meet it. Its 10-K for 2022 notes that, although the LCR didn’t apply, the bank had $26 billion in high-quality liquid assets. But that would be their HQLA ignoring the haircuts and cap on Level 2 assets that the LCR rule imposes. Based on the rule, its HQLA would have been about $10 billion. Meanwhile, First Republic had $176 billion in deposits, of which about one-third was insured; $51 billion in commitments; $10 billion in short-term FHLB commitments; and minimal inflows. Assuming a 3% drawdown for insured deposits and 10% for uninsured deposits (generously) gets an LCR of roughly 50%, before the 70% adjustment for banks with less than $250 billion in assets.