The failure of Silicon Valley Bank, and the ensuring contagion to other small US regional banks, have served as a forceful reminder of some of the fundamental risks that make banking a heavily regulated industry. There are many lessons to be learnt for bank executives, risk managers, regulatory designers and bank supervisors – as well as a reminder to macroprudential authorities that individually non-systemic institutions can become systemic collectively if they have concentrated and correlated business models and risk. The episode and its implications will no doubt be dissected for years to come.

Author: Ronnie Driver

Published: 17 July 2023

Introduction

The failure of Silicon Valley Bank, and the ensuring contagion to other small US regional banks, have served as a forceful reminder of some of the fundamental risks that make banking a heavily regulated industry. There are many lessons to be learnt for bank executives, risk managers, regulatory designers and bank supervisors – as well as a reminder to macroprudential authorities that individually non-systemic institutions can become systemic collectively if they have concentrated and correlated business models and risk. The episode and its implications will no doubt be dissected for years to come.

Vintage Neon Sign in Window of Pawn Shop by Marti157900, Getty Images

Vintage Neon Sign in Window of Pawn Shop by Marti157900, Getty Images

One area of focus has been on liquidity risks, where the scale and pace of deposit withdrawals from some banks highlighted the fragility of the fractional reserve banking model which exists in all major economies today. This has re-opened a debate on the how to assure customer deposits, whether through commercial bank self-insurance via liquidity buffers, the calibration of government deposit guarantee schemes, or the availability of Lender of Last Resort (LOLR) lending facilities from central banks.

Former Bank of England governor, Mervyn King has recently set out his proposed policy response whereby the central bank takes on the role of a ‘Pawnbroker for all Seasons’ (PFAS).[1] The proposal involves a new bank regulation which requires the sum of banks’ liquid assets buffers and the drawable capacity of collateral they have pre-positioned at the central bank to exceed their stock of ‘runnable’ deposits. In what follows, I will refer to the sum of actual and contingent liquidity as ‘effective liquidity insurance’ (ELI) and the overall requirement as the ‘King test’.

This blog starts with a recap on the fractional system and the PFAS proposal, before exploring the feasibility of meeting the King test, using a stylised model.

Fractional reserve banking

Banks create new deposits when they make new loans.[2] Because there are few ways in which these deposits can ‘leak’ out of the banking system,[3] the aggregate banking system is self-funding. The limits on credit creation come from the constraints put on banks by the prudential authorities – mainly capital and leverage ratio requirements.[4]

The assets banks create are long-term illiquid loan assets; but the liabilities they create are short-term liquid deposits. This results in the fractional banking system – the system as a whole cannot cope with the simultaneous withdrawal of all deposits because it does not have sufficient liquid assets to do so.

This inherent fragility does, however, come with a benefit. The ability to create loan assets financed by a pool of (at least partially) behaviourally stable deposits contributes to the economy’s productive capacity, boosting employment and growth. It also keeps the cost of credit low, because the remuneration rate on demand deposits is lower (in some cases significantly so) than the rate of return required on other bank liabilities.

As a result, there is a trade-off between the benefits of fractional banking in terms of credit creation and growth, and the real economy costs of disorderly runs and bank failures that come with the fractional reserve banking model.

At one of the spectrum sit fully narrow banks, where customers deposits are backed 1:1 with reserves held at the central bank.[5] Because demand deposits and bank loans are separated, loans to the private sector must be financed by equity or term debt/deposits. Such an arrangement provides maximal safety for demand deposits, but the cost of credit will be significantly higher due to the funding costs implications of the requirement to finance loans solely with capital and term debt. There would also be likely operational diseconomies from separation of the deposit and lending function (e.g. increasing the informational asymmetry for the lender, because they no longer have any insight into the potential borrowers’ deposit behaviour).

As we move from this corner solution, and embrace some degree of ‘width’ in the system, we introduce liquidity risk whereby demand deposits can be withdrawn immediately but not all assets can be liquidated immediately. The resulting maturity mismatch is the driver of potential bank runs, as per the classic paper from Diamond and Dybvig.[6]

In that context, governing authorities should take a proactive approach in deciding how ‘fractional’ they want their banking systems to be. This involves choices around minimum levels of self-insurance by commercial banks;  the role of government deposit guarantee schemes; the role of the central bank as a lender of last resort.

The central bank as the Pawnbroker for All Seasons (PFAS)

As noted above, King’s PFAS proposal requires that banks’ ELI exceeds their ‘runnable deposits’. This ensures that short term runnable deposit outflows can always be met, irrespective of the pace or the quantum.

King sets out a number of advantages of the PFAS model. It removes ‘alchemy’ from banking by significantly narrowing the system – with the central bank collateral haircut imposing an effective ‘social tax’ on alchemy which can be internalised by banks in their lending decisions.  Higher central bank haircuts increase the tax on alchemy by requiring a greater share of term debt/equity financing against those assets. King also argues PFAS recognises that insurance has to be put in place up front (rather than at the point of failure) and removes the central bank moral hazard problem associated with LOLR activities (because the central bank commits upfront, at a pre-agreed cost and haircut, to provide liquidity against assets). Finally, he argues this could pave the way for a significant simplification of the regulatory regime – as multiple complex regulations aimed at enhancing depositor protection can be removed/replaced; and as failing banks can be more easily resolved by lifting the short term runnable deposits and the associated liquid assets out of the failing bank.

However, many of these arguments are double edged. The central bank precommitment needs to be credible. And the schedule of collateral haircuts needs to be credibly fixed in advance, to avoid the central bank reneging on its commitment and pro-cyclically increasing its haircuts during an economic downturn. To do this, it needs to confront the associated ‘radical uncertainty’ of how asset values behave under stress. This is particularly challenging as it amounts to a judgment trading off the amount of liquidity insurance offered against the risk of taxpayer losses in the event the collateral is insufficient in the event of the default of a borrowing bank. This is a quasi-fiscal choice.

Moreover, despite significant and expansive reform in some jurisdiction, a range of assets remain ineligible for pre-positioning at central banks. For example, in the UK there are a range of detailed restrictions on collateral eligibility based on asset class (e.g. credit card assets, loans to banks and loans under partially drawn term facilities are not currently eligible). There are further restrictions on borrower domicile, loan currency and the legal framework governing the loan. [7]  There are also some operational restrictions (e.g. limiting the number of collateral pools each bank can have) which have scope to introduce friction into the PFAS model.

Even with fixed haircuts and asset eligibility, there is a second source of procyclicality in the PFAS model. In the event of a generalised market stress, asset prices (including loan assets that have been prepositioned at central bank) would be expected to fall, which would in turn reduce banks’ drawable capacity at the central bank.[8] In an attempt to restore the ELI and meet the King test, banks may take a range of actions which are individually sensible but collectively damaging – the same macroprudential concern that exists today. In the US, the Fed’s Bank Term Funding Programme chose to lend against US government debt at par, even though the rise in interest rates has depressed the market value of these securities to below this level. It is hard to see how the authorities would extend this type of unsecured lending against private sector debt, which would have to comprise the vast majority of the pre-positioned assets in the PFAS proposal.

Finally, definition of ‘runnable deposits’ in the PFAS proposals is subjective. This definition is crucial as it pins down the extent of the run a bank can suffer before it fails.  King focusses on liabilities with a contractual maturity of less than 1 year. But this is also a choice parameter, with an important trade-off.  Too short a horizon, and the policymaker leaves open the risk that the bank runs out of liquidity before an orderly recovery or resolution can occur.  But as the horizon is extended, the banking system is narrowed, with the associated costs.

Assessing PFAS – a stylised model

The PFAS proposal appears novel but fundamentally represents a different calibration of the existing Liquidity Coverage Ratio (LCR) regulation introduced in the aftermath of the Great Financial Crisis (GFC). The LCR requires banks to hold enough liquid assets to cover their cumulative deposit outflows over the next 30-days, with each deposit being weighted by an ‘outflow factor’ depending on the counterparty and hence the deposits’ assessed ‘stickiness’. King’s PFAS idea can be thought of as reforming / recalibrating the LCR in a few ways – it changes the horizon from 30-days to 1-year;  it increases the outflow rates on all short term deposits to 100%; and it expands the definition of liquid assets to include contingent liquidity guaranteed by the central bank by giving credit for pre-positioned collateral.

To assess theoretical feasibility, I built a stylised model of the banking system which lets us asses the scale of the transition that would be required to pass the King test.   In this model, commercial banks have:

  • Two forms of assets: liquid assets (central bank reserves) and illiquid assets (mortgage and/or corporate loans). Liquid assets are credit risk free.  All illiquid assets are denominated in the domestic currency, so there is no FX risk.
  • Three forms of liabilities: capital, callable deposits (which are ‘runnable’), and term deposits/debt (which are not). As per King’s proposal, the PFAS solution assumes no government deposit insurance is provided.

The bank sets itself 2 constraints:

  • Meeting its minimum capital requirement, with a management buffer on top. For simplicity, I ignore leverage ratio rules and simply assume that the bank must meet its risk-based Minimum Requirement for Eligible Liabilities (MREL).  In my baseline calibration, I assume that the target MREL ratio is 30% of RWA.[9]
  • Meeting its minimum Liquidity Coverage Ratio (LCR) requirement, with a management buffer on top. In the baseline calibration, I assume the target LCR ratio is 150%.[10]

The model then requires user assumptions on the banks’ initial funding model (which determine the bank’s liquidity requirement under the LCR) and on the haircuts set by the central bank on pre-positioned assets. It also requires assumptions about the bank’s asset Risk-Weighted Asset (RWA) density. Given these initial conditions, the model solves for the term deposits/debt required (as a % of the total balance sheet) to pass the King test, holding all other things equal.[11]  Different parameter calibrations generate sensitivity analysis and a quantitative assessment of the main drivers of the results.

Assessing PFAS – results

Table 1 summarises some initial results for two different business models.

Bank A is calibrated to represent a household-deposit funded mortgage monoline – this could be a building society or a narrow ring-fenced bank within part of a wider group. I have assumed it has a RWA density of 15%.  Its non-capital funding is assumed to comprise solely of household demand deposits with an assumed 10% LCR outflow factor.[12]  Assuming a 30% haircut on pre-positioned assets at the central bank,[13] Bank A’s ELI would cover 77% of its demand deposits. It would need to restructure its balance sheet so that ~23% of it was term (non-runnable) deposits to meet the King test.

Bank B resembles a corporate-focussed lender (or a non-ring fenced bank subsidiary of a wider banking group). It has corporate credit exposures which deliver a higher RWA densities – I have assumed 30% – increasing the capital requirement and reducing the amount of non-capital funding on the balance sheet. And it has a mix of corporate operational, non-operational and financial institution deposits. These deposits have materially different LCR outflow rates (ranging from 20% to 100%) but are materially higher than for household deposits. For illustration purposes I have assumed the blended outflow rate is 40%. Even with no term funding, this bank can now meet 91% of its demand deposits using its liquid assets and drawing capacity at the central bank. It only needs 10% of its balance sheet funded by term deposits to meet the King test.

Table 1 – Model results for two stylised banks

The analysis reveals and quantifies three intuitive results:

  • Banks who are funded exclusively through household demand deposits will face the biggest challenge. This is because PFAS requires these deposits to be 100% backed by liquid assets, (compared to as little as 5% for insured deposits under today’s regulatory requirements). By the same intuition, the greater the share of non-household deposits, the higher the starting King test position, and the fewer term deposits required. That is because these non-household deposits already attract higher outflow factors against the LCR, which means they are closer to the ‘narrow’ bank solution. As expected, changing the bank’s assumed target LCR (for a given deposit mix) has a much more limited impact relative to changing the underlying deposit mix itself.
  • Higher central bank haircuts makes meeting the King test more challenging, for any given asset mix and set of other parameters. The second calibration in the table above shows the impact of increasing the CB haircut from 30% to 60% – to cater for the fact not all assets are eligible as collateral at the central bank.  This significantly increases the share of term debt required to pass the King test, to over 50% in the case of Bank A and 32% for Bank B. This is a very material narrowing of the system.
  • Intuitively, changing the capital requirement and/or risk density doesn’t materially change the outcome. This is because, for all plausible calibrations, the nominal capital required is relatively small relative to the overall balance sheet. As such, increasing capital requirements doesn’t materially reduce the amount of non-capital deposit liabilities required to fund the balance sheet.

NIM implications

The model can then be extended to assess the NIM implications of the PFAS proposal, by making assumptions about the spread on loan assets, demand deposits and term deposits.

The Bank of England publishes a range of loan and deposits interest rates data, which can be used for calibration purposes. Choosing the appropriate rates to use for calibration is difficult, however. There has been much structural change since the GFC, which means that pre-2008 spreads may no longer be representative of today’s banking sector. At the same time, post-GFC spreads are also unlikely to be representative given official rates were at the effective lower bound, and Quantitative Easing created significant additional deposits and excess liquidity in the banking system.  More recently, spread are likely to have been distorted by behaviour during Covid, and the speed of the recent increases in official interest rates.

Given this is just an illustrative model, and for simplicity, I present the results of three different calibration choices – as set out in the table below. The illiquid asset spreads are based on the average spread of the 2Y 75% LTV quoted mortgage rate; the demand deposit and term deposit spreads are based on the effective interest bearing sight deposit rates and effective term deposits rates respectively. I have assumed there is no spread earned on liquid assets (which in the model comprise entirely of central bank reserves), given they are remunerated at Bank Rate.

To illustrate the potential NIM impact, I consider Bank A again (which resembles a mortgage focussed, retail deposit funded bank).  In the baseline calibration, the bank has no term deposits and so the King test is not met. Using the whole sample spreads (see first column of Table 2), Bank A has a starting NIMs of 2.06% – this is close to the recent simple average NIMs across mortgage focussed UK banks. If the bank restructures its balance sheet to meet the King test – with 23% term deposits as per Table 1 above – it would see its NIM fall to 1.79%.  Note that this is a very simplistic calculation, and that it (unrealistically) assumes no change in the market spreads on term deposits relative to the baseline, despite a significant increase in banks’ demand for them.[14]

This fall in NIM ultimately has to be borne by someone. One option is that it is borne by bank shareholders, in the form of lower profits. Assuming no non-interest income, an initial 50% cost-income ratio (with cost held fixed as the bank rebalances to meet the King Test), a fixed 25% effective tax rate, full pay out of all profits and a 10% cost of equity discount factor, a perpetual loss in income of this scale would reduce the bank’s share price by around 25%.

Table 2 – impact of different spread assumptions on Bank A’s NIM and equity value, based on meeting the King test

If the central bank haircut was 60% instead of 30%, NIM would drop to 1.44%, leading to a ~60% drop in share prices based on similar assumptions. The other columns in Table 2 show that the magnitude of the drop in equity value doesn’t move materially across these specific alternative calibrations.

Bank shareholders are unlikely to be happy with such a drop in the value of their investment, and will expect bank management to react. So an alternative, and more plausible outcome, would be that banks would try to rebuild their margin to ensure that they can continue to meet investor Return on Equity (ROE) expectations. They can do this through a combination of increasing the rate they charge on loans (i.e. the asset spread) or reducing the rate they pay on deposits (i.e. the deposit spread).

In that case, the ‘cost’ of PFAS would be borne by borrowers and savers in the economy – with the burden split across the two groups based on how much of the margin of adjustment comes through asset prices relative to liability prices. Note that, in the current financial system, the cost of assuring deposits is borne by different actors.  The cost of assuring insured deposits is (contingently) borne by bank shareholders via the industry funded deposit guarantee scheme.[15] The cost of assuring uninsured deposits, in the event they are protected from loss in the event of a bank failure, is borne by taxpayers.

To illustrate the potential impact of moving the burden of PFAS onto borrowers, the model can run another sensitivity. Taking Bank A again, and assuming a starting asset spread of 1.4% as per the whole sample spread calibration, the chart below shows how much the loan spread would need to increase (Y axis) to restore the bank’s starting NIM position, whilst meeting the King test, for a variety of term deposit spreads (X axis).

Chart 1 – increase in loan spread required (Y axis), for a given term deposit spread (X axis), to restore Bank A’s initial NIM.

The chart intentionally includes a long-tail of assumed term spreads, because the very significant increase in term funding needs would require banks to pay up considerably more than today to attract the necessary deposits. In reality, there would be a limit to how much money households and businesses were willing to lock up for >1 year term (given a minimum underlying demand for liquidity), which would make relationship between the supply of term funding and the term spread offered non-linear at some point. In the face of limits on the supply of term deposits, some banks (those who are large enough to have a market presence) could instead raise more wholesale long term unsecured debt.

This source of funds is more expensive still, and would also reprice higher given increased encumbrance levels associated with PFAS levels of pre-positioning at the central bank, which would increase the risk of losses for bond holders in resolution. Finally, note that the increases in loan spreads would also reduce demand for credit, and shrink banks’ balance sheets and reduce their scope for pre-positioning collateral. A further feedback loop would ensue. Modelling these factors is beyond the scope of this stylised model, but their impact is likely to be material. As a result, it seems likely that the PFAS would have a very significant impact on the spreads required to stabilise margins, and hence the cost of credit for the economy.

Conclusion

The modern banking system is inherently exposed to liquidity risk. To date this has been managed through a combination of liquidity regulation, deposit insurance, and central bank LOLR. The recent failure of some regional US banks has re-opened discussions on the appropriate mix of these factors in providing depositor protection and defining the narrowness of the fractional reserve banking system.

King’s PFAS proposal narrows the banking system and hence reduces its inherent liquidity risk. Alternative options, such as recalibrating LCR outflow factor requirements, could deliver a similar outcome. All such proposals have both social benefits in terms of increasing the assurance around demand deposits, but also potentially significant costs in the form of a higher equilibrium cost of credit in the economy. This same trade-off is repeated in the discussions around CBDC and fully reserved stablecoin.

With that trade-off in mind, the key question is on the appropriate ‘narrowness’ of the banking system and hence the appropriate level of liquidity insurance. Here, there may be case for review given developments in technology and the potential speed of deposit flows. But at the same time, the prudential authorities should be careful of lurching to a radical overhaul of the liquidity framework, given that the issues in the small US bank sector are localised to that population and jurisdiction. Significant buffers of bank capital, the development of the resolution regime, regular stress testing, effective supervision and deposit insurance schemes should all help mitigate the risk of rapid deposit flight.

If reform is deemed necessary, the PFAS appears a relatively extreme solution. The stylised model discussed in this blog demonstrates how the implications would be very significant for some business models, with significant consequences for real economy credit supply and cost. A significant transition period would be required to ensure individual banks are able to adjust (King himself proposed a 10-20 year transition period).

Moreover, practical constraints are likely to make it an unrealistic solution. For example, the PFAS is likely to require wholesale term debt issuance, as well as term deposits.  This poses a significant problem for smaller banks, who do not have market issuance capability and for whom establishing it would not be cost effective.   The solution also significantly increases asset encumbrance (actually or contingently in stress) and hence would leave greater losses to be imposed on unsecured creditors (including depositors) in the event of a capital-related bank failure.

The PFAS also leaves the central bank in a very difficult position where it has to pre-commit its balance sheet, and hence ultimately putting taxpayer funds at risk.  As such it needs to set its haircuts to limit the risk of taxpayer loss in the event of a bank failure and the associated liquidation of any collateral it has monetised, whilst also resisting the siren calls in normal times that its haircuts are too punitive and that this is constraining growth in the economy. That is an unenviable tightrope to walk, particularly for an unelected body.

However, there is a key element of the PFAS proposal – namely the recognition of the benefit of pre-positioned collateral – that is highly valuable. Although pre-positioning is now standard practice for UK banks, current liquidity regulation gives no value to the associated contingent liquidity. Setting some requirement and giving banks credit for the liquidity value of this collateral (with a suitable cap given the absence of an ex ante central bank lending commitment) would provide additional liquidity insurance without significantly affecting the cost of credit to the real economy. This is worthy of further thought in policy circles.

Endnotes

[1] King originally discussed this idea in his book “The End of Alchemy”, long before the recent US bank failures.

[2] See, for example, McLeay, Radia and Thomas – Money creation in the modern economy | Bank of England

[3] The exchange of deposits for physical cash (notes and coin) is one example of such leakage.  The transfer of money to non-bank entities that are able to deposit directly with the central bank, is another.

[4] In theory, liquidity and funding based requirements such as the Liquidity Coverage Ratio (LCR) and the Net Stable Funding Ratio (NSFR) can also constrain credit creation.  But in practice these are unlikely to be a binding constraint.

[5] This proposal, which eliminates fractional reserve banking entirely, was initially put forward in the 1933 “Chicago plan”.

[6] Bank Runs, Deposit Insurance, and Liquidity

[7] For more details on loan level collateral eligibility, see Section 4.1 of this document.  Loan Collateral: guidance for participants in the Sterling Monetary Framework (bankofengland.co.uk)

[8] There is potentially some offset here from an increase in the value of ‘safe’ assets, such as sovereign bonds, in the event of a flight to quality.  But this is unlikely to be sufficient to offset the broader fall in the market value of non-sovereign bank assets.

[9] This is calibrated just above the regulatory “MREL requirements + buffers” published by the Bank of England here, to allow for an (arbitrary) management buffer on top

[10] This is calibrated to be broadly in line with the average LCR across the European banking since 2016 (noting higher recent data reflect the extraordinary central bank support measures offered following the Covid pandemic.   See the EBA’s January 2023 Report on Liquidity for more information here.

[11] In particular, I have retained the assumption that the bank is required to meet its target LCR ratio – this pins down the share of the bank’s ELI that is met with on-balance sheet liquidity.  The alternative assumption, which holds the bank’s liquid asset buffers constant at their initial position, would reduce the term deposit requirement marginally, but at the cost of lower NIM (because the bank now holds a greater share of demand deposits as on balance sheet liquidity, at the expense of higher yielding illiquid loan assets.

[12] Blended across insured and uninsured deposits which attract different factors under the LCR.

[13] This is towards the top end of the Bank of England’s collateral haircuts, albeit those are for securitised portfolios (rather than ‘raw’ loan portfolios) and exclude a 5pp add on for own-name securitisations.  Details are available here.

[14] Modelling the price elasticity of demand and hence an endogenous response to the increase in demand for term deposits is beyond the scope of this exercise.

[15] In theory this cost should already be priced into banks’ loan and deposit pricing, and so ultimately also borne by the real economy users of financial services.  But this appears unlikely in practice.