Why every liquidity rule exists: a practitioner history of LCR, NSFR and ILAAP

Every liquidity rule you report against was written after a specific bank ran out of cash. This post maps each rule, the LCR, the NSFR, run off rates, HQLA definitions, the ILAAP narrative, back to the failure that caused it, so the frameworks stop feeling arbitrary and start reading as sensible answers to real problems. The video below walks the same history if you prefer it in that form.

Why history matters for a liquidity team

Every liquidity rule you work with was written after something went wrong. The LCR, the NSFR, run off rates, HQLA definitions, the ILAAP stress narrative. None of it was designed in the abstract. Each piece is a response to a specific failure where a bank ran out of cash.

If you learn the rules without the failures, they feel like a compliance burden handed down from the PRA and Basel. If you learn the failures first, the rules read like sensible answers to real problems. That is the difference between filling in a PRA110 template and understanding what it is trying to protect you from.

So let us trace the failures.

The pattern that never changes: solvent but illiquid

Here is the one lesson that repeats across 150 years. A bank can be solvent on paper and still fail.

Solvency is about the balance sheet over time. Assets worth more than liabilities. Liquidity is about timing. Can you meet cash demands today, this week, this month, as they fall due.

The two come apart because bank assets are illiquid by design. You take deposits that can leave on demand and you lend them out for years. Mortgages, corporate loans, project finance. That maturity transformation is the business model. It is also the vulnerability. When depositors and counterparties want their cash faster than you can turn assets into cash, you fail, even if every loan on your book is money good.

Keep that distinction in mind. Every episode below is a variation on it.

1866 and the birth of the lender of last resort

Overend, Gurney and Company was one of the largest discount houses in London. When it failed in May 1866, it triggered a panic that spread across the City. Banks that were perfectly solvent could not raise cash because everyone was hoarding it at once.

Out of that came the doctrine we still use. Walter Bagehot set it out in his 1873 book Lombard Street. Three parts.

• Lend freely to stop the panic spreading.
• Against good collateral, so the central bank is not taking bad credit risk.
• At a penalty rate, so banks only come when they genuinely need to, not to fund normal business cheaply.

That is still, in essence, how the Bank of England and other central banks run their liquidity facilities. When you see a discount window or a term facility priced above the market rate and secured against eligible collateral, you are looking at Bagehot. The idea that there is a backstop, but a deliberately expensive one, shapes how treasury teams think about their own liquidity buffers. You hold HQLA precisely so you do not have to turn up at the penalty rate.

The Panic of 1907 and the case for a central bank

The United States spent the 19th century without a central bank. The Panic of 1907 showed what that looked like. A failed attempt to corner the stock of United Copper spread into a run on the trust companies of New York, and there was no institution able to act as lender of last resort.

The system was rescued, more or less, by J. P. Morgan personally organising other bankers to pool cash and support the market. Relying on one wealthy individual to stop a national panic is not a control framework. Everyone knew it.

The direct consequence was the creation of the Federal Reserve in 1913. The lesson for us is structural. A liquidity system needs an institution with the standing and the balance sheet to provide cash when private markets freeze. Once you have that backstop, the question becomes who gets access, on what terms, and against what collateral. Those are still live questions in every liquidity framework.

The Great Depression and deposit insurance

The bank runs of the early 1930s were retail runs. Ordinary depositors queuing to withdraw cash, banks failing in waves, and each failure feeding the fear that drove the next one.

The answer was deposit insurance. In the United States, the FDIC. In the UK today, the Financial Services Compensation Scheme, which protects eligible deposits up to £85,000 per depositor per firm. The mechanism is simple. If depositors know their money is protected up to a limit, they have no reason to run. The run stops before it starts.

This matters for how you set run off rates. Insured retail deposits are stickier than uninsured ones, and the LCR reflects that directly. Stable retail deposits attract a low run off rate. Less stable retail deposits attract a higher one. The whole idea that different deposit types leave at different speeds under stress traces back to the observation that insurance changes depositor behaviour.

Those run off assumptions are what turn each stress scenario into the numbers the ratio ultimately reports.

2008: the wholesale funding run

Here is where the story turns modern. The 2008 crisis was not a retail run. Depositors were not queuing outside branches, at least not at first. It was a wholesale funding run.

Banks had come to rely on short term market funding. Repo, commercial paper, interbank lending, all rolling over constantly. When confidence went, that funding did not get withdrawn slowly. It simply stopped rolling. Counterparties refused to renew overnight and short dated positions, and banks that depended on them to fund longer dated assets were left with a hole they could not fill.

Northern Rock is the retail image everyone remembers, but the deeper problem across the system was this reliance on markets that could close in days. A bank could look well capitalised and still be unable to fund itself by the end of the week.

That exposed two gaps. First, banks did not hold enough genuinely liquid assets to survive a short sharp outflow. Second, they were funding illiquid long term assets with unstable short term money, and nobody was measuring that mismatch consistently.

Basel III: turning lessons into rules

Basel III answered those two gaps with two ratios. This is the part where the history maps directly onto your day job.

The Liquidity Coverage Ratio (LCR) addresses the short sharp shock. It requires a bank to hold enough HQLA to cover net cash outflows over a 30 day stress. Net outflows here means expected outflows minus expected inflows, where inflows are capped at 75 per cent of outflows so you cannot rely fully on money coming in during a crisis. HQLA is the modern version of Bagehot's good collateral, assets you can actually turn into cash quickly. The outflow side applies run off rates to your funding, and those run off rates encode the lesson about which money leaves fastest under stress. Wholesale funding runs quicker than insured retail. That is 2008 written into a formula.

The Net Stable Funding Ratio (NSFR) addresses the structural mismatch. Over a one year horizon it compares your Available Stable Funding to your Required Stable Funding, and requires the first to be at least as large as the second. It is not an asset by asset test. It weights every funding source by how stable it is and every asset by how much stable funding it needs, then checks the totals. That is the direct answer to funding illiquid loans with overnight repo.

Alongside the ratios sits the ILAAP, where a firm sets out its own liquidity risk profile and the stresses it thinks it should survive. The ILAAP is where the history becomes a narrative. You are effectively telling the PRA which of these historical failures could happen to you, and showing you would survive them. The aim is to build that story without it becoming a panic, keeping the assessment focused on the stresses you can genuinely justify.

2023: SVB, Credit Suisse and the speed problem

Then came 2023, and it changed one assumption badly.

On 9 March 2023, Silicon Valley Bank faced reported withdrawal requests of around 42 billion dollars in a single day, with a further 100 billion dollars reportedly queued for the next day. Credit Suisse bled funding at a pace nobody had modelled. Both runs were faster than anything the standard assumptions were built for.

Two things drove the speed. Digital banking, where a withdrawal is a few taps and settles instantly, no queue outside a branch. And social media, where the fear that used to spread over weeks now spreads in hours. SVB's depositor base was also concentrated and largely uninsured, which is exactly the profile the run off rates flag as unstable.

The uncomfortable point for practitioners is that a 30 day horizon assumes outflows spread over days. In 2023 the damage was done in hours. The ratio can still pass while the timing assumption underneath it no longer holds.

What this means for your run off assumptions today

The framework is not finished, because the failures are not finished. Here is what to take back to your desk.

• Treat run off rates as a starting point, not gospel. They were calibrated on historical runs that were slower than what we now see. If your deposit base is concentrated, uninsured, or digitally mobile, your own stress should assume it leaves faster than the standard tables suggest.
• Look at concentration, not just totals. SVB's problem was not the size of its deposits but who held them and how alike they were.
• Test an intraday and multi day stress, not only the 30 day picture. The regulatory ratio is a floor. The real question is whether you survive the first 24 hours.
• Write the history into your ILAAP narrative. Naming the specific failure each stress represents makes the document defensible and makes the numbers mean something.

If you want to get closer to your own position rather than the templated view, read it directly from the data in Python rather than waiting for the monthly report. And if you are building the numbers that feed a PRA110 return, the same failures explain why each row is broken down by maturity and counterparty type.

The rules are old lessons in current language. Know the lesson and you will know when the rule is protecting you and when it is quietly out of date.