On 10 March, the US authorities closed Silicon Valley Bank (SVB) following massive deposit outflows from the bank the preceding day. The days and weeks that followed saw further bank failures, of Signature Bank (12 March) and First Republic (1 May). The proximate factors that led to these bank collapses have been discussed at length, and include material unhedged exposures to interest rate risk and rapid growth in uninsured deposits.

Authors: David Aikman, Coskun Tarkocin

Published: 2 June 2023

On 10 March, the US authorities closed Silicon Valley Bank (SVB) following massive deposit outflows from the bank the preceding day.  The days and weeks that followed saw further bank failures, of Signature Bank (12 March) and First Republic (1 May).  The proximate factors that led to these bank collapses have been discussed at length, and include material unhedged exposures to interest rate risk and rapid growth in uninsured deposits.[1]

"Silicon Valley Bank" by Alpha Photo is licensed under CC BY-NC 2.0

“Silicon Valley Bank” by Alpha Photo is licensed under CC BY-NC 2.0

In this blog, we discuss the implications of these events for the design of the Liquidity Coverage Ratio (henceforth, the “LCR”).  The LCR is a regulatory standard introduced via Basel III, under which banks are required to maintain a buffer of liquid assets that can be turned into cash to pay depositors and debt-holders in the event of a liquidity shortfall.  This standard was not applied by the US authorities to SVB – under the Fed’s so-called “tailoring” approach, SVB’s balance sheet size was below the threshold at which the LCR takes effect.[2]  But we can nevertheless ask:  had it been in effect, would the LCR have flagged the liquidity risks being run by SVB?  If so, how might SVB have adjusted its balance sheet?   And are there aspects of the LCR rule that require a rethink given these events?

Our answers to these questions are as follows.  Yes, the LCR would have been a binding constraint for SVB bank from early 2022 at the latest.  SVB would most likely have responded by replacing its agency mortgage-backed securities with Treasuries of similar maturity, reducing its profitability somewhat but making no difference to the duration risk it was running.  And we think there is a case for revising aspects of the LCR in the light of this experience.  Specifically, we propose that regulators consider requiring (a) greater transparency over concentration risks in the liquid asset buffer, (b) that liquid assets held to meet the LCR be carried at market value on the bank’s balance sheet for regulatory capital purposes, and (c) increased run-off rate assumptions for less stable deposits.

Would the LCR have flagged the liquidity risks being run by SVB?

Let’s start with a quick recap/primer on the LCR.  A bank’s LCR is calculated by dividing its amount of high-quality liquid assets (HQLA) by its predicted net cash outflows under stress.  The most liquid assets – central bank reserves and government bonds – are treated as level 1 HQLA.  Agency mortgage-backed securities (MBS) qualify as level 2A and are subject to a 15% haircut.  Level 2B comprise certain corporate bonds, commercial paper, equities and other assets are subject to a 50% haircut.[3]

Importantly for our purposes here, there is a requirement that level 1 assets comprise at least 60% of the total HQLA; this has the effect of capping the amount of level 2 assets that can qualify in a bank’s HQLA buffer.

Net cash flows under stress are the difference between stressed outflows and inflows.  Outflows are calculated by applying pre-defined “run-off” rates to various balance sheet items, with insured retail deposits being the most stable (and hence receiving the lowest run-off rates) and committed credit lines and uninsured wholesale deposits being the least (and hence receiving the highest run-off rates).  Inflows are essentially the bank’s deposits with other financial institutions and maturing loans.

Let’s now ask the question:  had SVB been required to meet the LCR, would it have identified the liquidity risks being run by the bank or not.[4]  To assess this question, we’ll focus on the “full-fat” version of the LCR agreed in Basel III – not the weaker standard applied by the US authorities to certain banks under its 2019 tailoring approach under which these banks were required to meet only a portion of their stressed outflows.[5]  The Basel III version is the relevant focal point when it comes to drawing out lessons for global financial regulation.

According to the Fed, SVB’s LCR would have just about been in breach of the 100% standard from early 2022, with the bank’s metric declining sharply thereafter.  SVB’s full LCR was 99.3% in March 2022, a touch below the regulatory standard; by December last year, it had fallen to 90.8%; and by February 2023 it had dropped to 82.6%.[6]

Why was SVB’s LCR so weak?  This reflects the combination of three factors:

  • very rapid deposit growth in the years preceding its collapse – between 2020 and 2022, SVB’s deposits grew at an average annualised rate of 85%;
  • a very large proportion of uninsured deposit balances (94%) – which receive materially higher “run off” rates in the LCR calculation; and
  • the funds being invested in level 2A HQLA – specifically, in agency-issued mortgage-backed securities. Given the requirement that 60% of HQLA must be met with level 1 assets (see above), there was an effective cap on the amount of SVB’s agency MBS permitted to count in its HQLA, ie these assets would not have been able to exceed 40% of the bank’s total HQLA.

While SVB’s full LCR was very close to the 100% regulatory standard in March 2022, the relevant constraint on its actions is arguably more likely to have been market pressure.  To assess how this might have played out, we can ask: how would SVB’s LCR have compared with its peers?  Was it an outlier?  The table below presents LCRs for a selection of large and smaller US banks.  SVB’s LCR stands out as one of the weakest in the cohort in early 2022 – albeit not the weakest, which was PNC Bank with US Bankcorp close behind.

Table 1: US Banks’ LCRs

Bank Q1 2022Q4 2022
JP Morgan110%112%
Bank of America116%120%
Citi116%118%
Wells Fargo119%122%
Goldman126%129%
Morgan Stanley130%132%
US Bankcorp^96%104%
PNC Bank^93%91%
Truist^111%112%
Charles Schwab111%123%
Capital One^119%122%
Bank of New York Mellon109%118%
State Street106%106%
Silicon Valley Bank99%91%
The median for banks with full LCR116%120%
The range for banks with full LCR106% – 140%106% – 143%

Notes:

The table presents full LCRs for each bank, ie removing any scalar adjustment to net stressed outflows to make the figures comparable across banks.

^ Denotes banks with an 85% reduced LCR requirement, ie where the net stressed outflow figure is multiplied by 85%.  In Q4 2022, the reported LCRs for these banks were:  PNC 107%; US Bankcorp 122%; Truist 112%.  SVB would have been subject to a 70% reduced LCR from 2023 Q4.  Its reduced LCR in Q4 2022 was 103%.

Source: S&P Global Market Intelligence, Federal Reserve Board, and banks’ LCR disclosures.

Banks generally set internal targets for their LCRs above 100% to take account of business-as-usual volatility, seasonalities and mild market stresses.  Given its highly concentrated funding structure and its high percentage of uninsured deposits, prudent risk management considerations would suggest SVB should have targeted an LCR above its peers – eg above the 116% median LCR in Q1 2022 for banks required to meet the full LCR requirement.[7]

What impact would a binding LCR have made?

We posit that had the full Basel III LCR been in effect, SVB would have come under market pressure to adjust its balance sheet.  It then would have had various options available for increasing its LCR, which include:

  • Changing the composition of its HQLA. SVB could have reduced its holdings of agency debt/MBS and correspondingly increased its holdings of US Treasuries of similar duration.  The mathematics of the level 2A cap imply a HQLA multiplier from such a move:  shifting $6bn assets from agency debt/MBS into US Treasuries would have improved SVB’s HQLA by $10bn, pushing its LCR well above 100%.[8]  This we feel would have been the most likely option taken – an action per se that would have made no difference to the duration risk the bank was running.
  • Liability management. Alternatively, SVB could have expanded its balance sheet by raising additional short-term or medium-term debt to hold as a cash or to purchase additional level 1 assets.  This option seems less likely to us as it would have been expensive and SVB may not have had sufficient access to wholesale debt markets to improve its LCR materially.
  • Building/enhancing secured funding capacity. Finally, SVB could have raised cash by repo-ing out its level 2A securities, or alternatively it could have used collateral swaps to exchange level 2A for level 1 securities, with tenors longer than a month.  A well-established repo capability may have offered the bank valuable time by raising liquidity rather than selling and realising losses on the hold-to-maturity (HtM) portfolio.  This could have provided SVB and regulators with time to address the underlying vulnerabilities.  However, these operations would have been subject to refinancing risks due to the short-term nature of the repo market, suggesting that this might only have been a temporary solution.

Overall, therefore, SVB would have had options for adjusting its balance sheet to meet a binding LCR.  While some of these options may have helped reduce the underlying liquidity risk to a degree, it is clear they would have been insufficient to address the underlying problems the bank faced in managing its interest rate risk.

What implications are there for the design of the LCR?

We think recent events highlight three areas of the LCR’s design that should be revisited by regulators.  The first relates to the concentration of particular assets in the HQLA buffer;  the second to the inclusion of HtM assets in HQLA;  the third to the “run-off” assumptions used for calculating stressed outflows.

We briefly discuss each in turn.

Regulators should consider revisiting the potential for concentration risk in the HQLA buffer

To cover deposit outflows, a bank needs to have confidence that it can sell securities without resulting in material losses on capital or raising concerns about its solvency.  A well-diversified liquid asset buffer that avoids excessive concentration of particular asset classes – in terms of issuer, maturity, currency, asset type – is therefore important.

While Basel III is very clear on the need for a diversified liquid asset buffer,[9] it includes no specific policies to achieve such an outcome (beyond the cap on level 2 assets in HQLA).  We feel there is a case for revisiting this issue.  As a starting point, there should be a clear supervisory expectation that diversification of the liquid asset buffer should be monitored regularly by the bank’s internal liquidity management function.  Another option for regulators to consider would be requiring greater disclosure over composition of HQLA, and the concentration of particular assets within the buffer.  A third would be introducing specific quantitative limits on the quantum of particular assets that can be included.

Regulators should consider requiring that HQLA be carried at market values not amortised cost

The current design of the LCR has a schizophrenic quality in that banks are permitted to include hold to maturity assets in their liquid asset buffers.  When calculating the amount of HQLA these assets provide, the bank must use their market values.  But for the purposes of calculating its capital ratio, the bank can use the assets’ amortised cost value.  Aside from the logical inconsistency here, this treatment potentially inhibits the usability of HQLA as when there are unrealised losses, any attempt to sell the assets in question to raise liquidity will trigger a capital loss.

The simplest way to resolve this inconsistency is to require all assets that are included in HQLA to be held at market value on the bank’s balance sheet.  If this approach is unappealing, the logical alternative would be a limit on the duration of securities (allowing for any hedges in places) that can be held in the liquidity asset buffer portfolio, whose sale would crystalise a regulatory capital hit.

One contentious aspect of the Fed’s “Bank Term Funding Program”, which it set up to provide additional emergency funding to banks after the SVB collapse, was to value banks’ Treasury and agency MBS collateral at par.  We note that this had the effect of squaring the circle and resolving the inconsistency we describe here.  It alleviated the need for banks (other than SVB) to sell their liquid asset portfolios, avoiding crystalising losses on HtM portfolios that had been included in HQLA.  While that is an understandable response in a crisis, it raises the question of what the steady state regime should be – we pick up this question at the end of the article.

Regulators should consider revisiting the deposit run-off assumptions in the LCR

It is clear from recent events that deposits can leave a distressed bank significantly more rapidly than is assumed in the LCR’s outflow assumptions.

SVB bank experienced a total deposit outflow of over $40bn on March 9th (c. 25% of its deposit base).  That evening, the bank communicated to supervisors that it expected an additional $100 billion in outflows on March 10 (80%+ of its remaining deposits).  These outflow rates are far more severe than those anticipated in the LCR calibration, where, roughly speaking, over a 30-day period the bank is assumed to lose 10-20% of its uninsured deposits.[10]  See the table below for other prominent examples, several of which exceed run-offs assumed in the LCR.

The experience, coupled with the perception that social media might have the effect of speeding up future deposit flights, raises the question of whether the run-off rate assumptions for uninsured, non-operational deposits should be revised upwards.

We think regulators should look again at this assumption.  This could be done either across the board as a recalibration of the LCR parameter, or via Pillar 2, which would allow for a more tailored treatment depending on the nature of the bank’s depositor profile.  Tailoring provides regulators with more flexibility to capture bank-specific vulnerabilities.  While in principle this can be a good thing, as we have seen it adds complexity to the regime and opens up avenues for regulatory arbitrage and lobbying pressure over time.

Table 2: Major deposit outflows in recent banking stresses

Bank NameDuration Size (billions)Run off rateMonthly run off rate
Northern RockA few weeks13 (GBP)56%56%
Washington Mutual16 days18.7 (USD)10.1%18.60%
Dexia1 month7 (EUR)8.75%8.75
Wachovia2 weeks15 (USD)3.6%7.80%
Banco Popular2 months18(EUR)24%12%
Silicon Valley Bank9 March42 (USD)24%> 80%
10 March (expected)c.100 (USD)75%
First Republic3 months101 (USD)58%>20%
Credit Suisse3 months67 (CHF)29%>10%
Signature Bank *10 March, following SVB collapse18 (USD) estimated20%>80%*

Source:  Northern Rock-Banco Popular: https://www.ecb.europa.eu/pub/pdf/scpops/ecb.op250~c7a2d3cc7e.en.pdf.  SVB-Signature: authors’ calculations.

Notes:  *Assuming all uninsured deposits would leave over the course of the month.

Macroprudential considerations

In making these suggestions, we are taking the current regulatory regime as given and assuming the current calibration of the LCR is in the vicinity of being optimal.  One reasonable challenge is that there is little basis for making such an assumption as it is unclear how the benefits of the current calibration of the LCR compare to its costs.  There are two particular difficulties in making such an assessment.  First, liquidity risks are by their nature all-or-nothing phenomena, which makes it unclear how to value any marginal change in liquid assets in terms of lower run-risk.  Second, it is unclear from a social welfare perspective whether the LCR is providing the right amount of self-insurance for the liquidity risks being run by banks vis-à-vis what should be provided publicly via central bank liquidity support policy.

This second point especially is a very difficult question to answer.  It links to suggestions by Lord King, former Governor of the Bank of England, that central banks should act as “pawnbroker for all seasons” (see FT, May 12).  Under this proposal, banks would pre-position sufficient collateral at the central bank to provide them with sufficient contingent liquidity to cover all runnable liabilities.  There wouldn’t be a need for an LCR in this situation, and the amount of liquidity risk being publicly insured would be determined by the haircuts and valuation approach set by the central bank.  We hope to explore this interesting proposal more fully in future blogs on this site.

Disclaimer: The views and opinions expressed in this blog article are those of the authors, and they do not represent the views of the HSBC Group and do not constitute advice.

Other resources:

https://ir.svb.com/financials/quarterly-results/default.aspx

https://www.fdic.gov/news/press-releases/2023/pr23033a.pdf

https://www.fdic.gov/resources/resolutions/bank-failures/failed-bank-list/

Data Source: S&P Global Market Intelligence

Endnotes

[1] A detailed report prepared by the Federal Reserve Board provides a comprehensive discussion of the reasons for the failure of the SVB. See https://www.federalreserve.gov/publications/files/svb-review-20230428.pdf.

[2] According to the Fed, SVB crossed the threshold of having $50 billion or more in average weighted short-term wholesale funding in December 2022, and would have been required to comply with the “reduced” LCR at a 70% calibration from the start of Q4 2023.

[3] See https://www.bis.org/publ/bcbs238.pdf for details.

[4] There have been other studies of SVB’s LCR.  See Feldberg (2023) who provides an earlier estimate of the bank’s LCR:  https://som.yale.edu/story/2023/lessons-applying-liquidity-coverage-ratio-silicon-valley-bank.  And see Nelson (2023) who focuses on whether the tailored LCR would have caught the risks being run by the bank:  https://bpi.com/silicon-valley-bank-would-have-passed-the-liquidity-coverage-ratio-requirement/.

[5] Specifically, “category III” banks with under $75bn in short-term wholesale funding face an LCR requirement calculated as HQLA divided by 85% of net stressed outflows; “category IV” banks with more than $50bn of short-term wholesale funding instead apply a 70% scaling factor to their modelled outflows.  See https://www.federalreserve.gov/aboutthefed/boardmeetings/files/tailoring-rule-visual-20191010.pdf.

[6] See Table 13 of the Fed’s SVB report, https://www.federalreserve.gov/publications/files/svb-review-20230428.pdf.

[7] An interesting question is whether, even in the absence of regulatory disclosure, bank analysts might have produced reasonable estimates of SVB’s LCR.  And if so, why this alone was not sufficient to lead the bank to adjust its risk profile.

[8] As discussed in the text, SVB’s level 2A assets would have been subject to 40% cap in its liquid asset buffer.  It is straightforward to show that the effect of such a cap is to limit the post-haircut value of the bank’s level 2A assets that can be included in its HQLA to 2/3 of the bank’s level 1 assets.  Given this, shifting $6bn “excess” level 2A assets into level 1 would have increased SVB’s HQLA by the $6bn + 2/3*$6bn = $10bn.  Intuitively, this reflects the fact that, with more level 1 assets, the cap loosens allowing the bank to include more of its excess level 2A in its HQLA.

[9] The Basel LCR standards require the stock of HQLA to be well diversified within the asset classes themselves, except for sovereign debt of the bank’s home jurisdiction or from the jurisdiction in which the bank operates, central bank reserves, central bank debt securities, and cash.

[10] For other funding items, the LCR assumes run-off rates of 3% of its insured retail deposits, 25% of its operational deposits, 40% of its non-operational corporate deposits, and 100% of its deposits from other financial firms.