Skip to content
Corporate & Investment Banking My progress Français

Home / Capital Markets & Securities Services / Capital Markets / Module 15

15Treasury & Funding

How a bank funds itself, prices its own balance sheet, and survives the day it runs short

How a bank funds itself and survives the day it runs short.

Reading time
6 h 20
Sections
100
Recall questions
91
Quiz questions
22
Simulators
2

Why this module exists

A bank is a maturity transformation engine, and Treasury is its control room. Depositors want their money back instantly. Borrowers want twenty-five-year mortgages. The bank stands between them and earns a margin for accepting the mismatch. That mismatch is not an accident to be minimised, it is the product. Treasury is the function that decides how much of it the bank carries, what it costs, and what happens when the assumption underneath it stops holding.

Nothing exposes a bank faster than its funding. Credit losses arrive over quarters and can be provisioned, negotiated and disclosed in an orderly way. A funding failure arrives in an afternoon. Three of the most consequential bank failures of the last two decades were liquidity events at institutions that were, on paper, solvent when they failed. Understanding Treasury is therefore not a specialist interest. It is the difference between reading a bank's balance sheet as a list of numbers and reading it as a set of promises with dates attached.

This module is also, unavoidably, a technology module. Liquidity coverage is computed daily from contract-level data across every system in the bank. Interest rate risk in the banking book depends on behavioural models whose assumptions must be documented, versioned and reproducible eighteen months later. Intraday liquidity monitoring is impossible without structured payment messages. A Treasury function is a data pipeline that happens to make decisions at the end of it, and the quality of the decisions is capped by the quality of the pipeline.

Two threads run through every page

The business thread asks what is being decided and by whom. How much wholesale funding to raise this year and in what instruments. What to charge the mortgage desk for a twenty-year fixed-rate loan. Whether to accept a deposit at a rate that looks expensive but improves the structural ratios. When to go to market and when to wait. Who bears the cost when a decision turns out badly. These are commercial decisions with owners, committees and consequences, not technical outputs.

The technology thread asks what has to be true for the decision to be possible. Which system holds the contract. How the cashflow is generated and under which behavioural assumption. Which engine applies the regulatory weights. How the number reaches the supervisor with its lineage intact. Which message turns a decision into an irreversible payment, and what happens if that message fails eleven minutes before the settlement cut-off. Neither thread makes sense alone. A funding plan that the systems cannot report on is not a plan, and a real-time dashboard that nobody is empowered to act on is not a control.

How to read this module

The module builds in five parts, each assuming only what came before. Part 1 constructs the bank balance sheet from first principles and places Treasury inside it. Part 2 walks the three cycles Treasury actually runs — the funding cycle, the transfer pricing cycle and the liquidity day — end to end. Part 3 maps every actor, internal and external. Part 4 opens the platform: architecture, data, messages, resilience and the regulatory regimes that specify them. Part 5 covers risk, real failures, and what technology is genuinely changing. A cheat sheet and a glossary close the module.

A worked fictional bank, Meridian Bank, is carried through every part so that the numbers connect. Every Meridian figure is illustrative and is flagged as such. Real market facts and real case studies are identified as real.

Seven kinds of box interrupt the prose, each with a fixed meaning. Key insight states something the rest of the section was building towards. Going deeper adds precision a specialist would want and a first-time reader may skip. IT & data lens explains what the topic demands of the systems. Watch out flags a trap, a common error, or a place where the standard answer is incomplete. Worked example carries a number through, usually with Meridian. In a nutshell closes a part. Check your understanding tests whether the part landed, with the answers given.

The module at a glance

Part What it covers The question it answers
1 — Context Maturity transformation, the funding instrument menu, LCR / NSFR / IRRBB / MREL, the 2021–2024 cycle What is a bank, and what can Treasury actually change?
1B — Capital CET1 / AT1 / Tier 2, RWA, the requirement stack and MDA, leverage ratio, ICAAP, RAROC What constrains the balance sheet besides liquidity?
1C — Lenses IFRS 9 classification, hedge accounting, rating agencies, ESG funding What decides whether a decision shows up, and what it costs?
2 Track A — Funding The annual plan, programmes, market windows, pricing, settlement, lifecycle How does the bank raise money?
2 Track B — FTP Matched maturity, the curve, replicating portfolios, governance What does internal money cost, and who pays it?
2 Track C — Liquidity Cash ladder, repo, intraday, cut-off, stress, ILAAP, contingency plan Can the bank pay today, and for the next thirty days?
2 Track D — Collateral CSAs, variation and initial margin, rating triggers, the 2022 gilt episode What does a hedge demand in cash?
2 — Worked calculations LCR, NSFR and a cash ladder built line by line, then broken on purpose Can you compute it yourself?
3 — Roles Four camps, the buy-side for bank paper, the depositor as investor, ALCO Who decides, who buys, who is affected?
3B — Group Solo versus consolidated, trapped liquidity, intragroup MREL, cross-currency, structural FX Why is a group not one balance sheet?
4 — Systems Architecture, vendors, data lineage, identifiers, messaging, resilience, five regimes What has to be true for the decision to be possible?
4B — Delivery The legacy estate, sequencing, data governance, model lifecycle, cloud, ISO 20022 How does a bank get from here to there?
5 — Risk Six risks, IRRBB in depth, KPIs, three real failures, technology and GenAI What goes wrong, and what is genuinely changing?
Applied Eight misconceptions and four worked scenarios Can you run the bank?
Reference Cheat sheet, key numbers, formulas, glossary Where do you look it up?

What you will be able to do

You will understand how a bank funds itself, instrument by instrument

You will know what a covered bond is and why it prices tighter than senior debt, what makes senior non-preferred a regulatory creation rather than a market one, why a certificate of deposit and a five-year bond solve different problems, and how the choice between them is made in a funding plan rather than opportunistically.

You will be able to map every actor and every flow

You will be able to name who inside the bank demands balance sheet and who supplies it, who outside the bank buys which instrument and why, which committee decides what, and where the line between first-line execution and second-line control actually runs.

You will understand where the price of internal money comes from

You will be able to construct a funds transfer pricing curve from a base rate and a term liquidity premium, apply it to a loan and a deposit, explain the resulting margin split to the desks it affects, and identify what a bank will over-produce if it prices that curve wrong.

You will be able to trace liquidity data front to back

You will be able to follow a single mortgage contract from the origination system through the cashflow layer, the liquidity engine, the transfer pricing engine and into a supervisory return, and say precisely which downstream number breaks when one field upstream is wrong.

You will recognise the risk and transformation landscape

You will know the six risks a Treasury function carries and who owns each, why two banks that met their regulatory ratios still failed in March 2023, which technology changes move cost or risk and which only move the dashboard, and how generative AI is being governed in a function where every number is eventually filed with a supervisor.

1A bank is a set of promises with dates attached, and Treasury manages the dates

Before any instrument, ratio or system makes sense, one thing has to be built from zero: what a bank actually is, economically. Everything Treasury does follows from a single structural fact about the institution it serves, and that fact is visible on one page of the balance sheet.

A bank exists because nobody else will hold the mismatch

Start with two people who cannot transact. A saver has ten thousand euros and wants it available tomorrow in case something happens. A homebuyer needs three hundred thousand euros for twenty-five years. No direct trade is possible between them. The saver will not lend for twenty-five years, the borrower cannot repay tomorrow, and neither has any way to assess the other's creditworthiness. The gap between them is not a market failure to be regretted. It is the space a bank occupies.

A bank performs three transformations at once, and each one earns part of the margin. Maturity transformation takes short liabilities and funds long assets. Liquidity transformation takes claims that are instantly redeemable and turns them into claims that are not. Credit transformation takes many risky individual loans and, by pooling them and absorbing the first losses with its own capital, turns them into a claim safe enough for an ordinary depositor to hold without analysis. Strip any one of these away and the institution stops being a bank. Perform all three at scale and you have created something that is profitable, socially necessary, and structurally fragile in exactly one way.

The fragility is arithmetic, not misconduct. If a bank funds twenty-five-year mortgages with deposits repayable on demand, and every depositor asks at once, no amount of asset quality helps. The mortgages are good. They are simply not due. A solvent bank with excellent assets can fail on a Tuesday afternoon because its promises had different dates from its claims, and the promises came due first. This is the single sentence from which the whole of liquidity regulation, the whole of Treasury, and every diagram in this module can be derived.

The balance sheet, read as a maturity ladder rather than a list

Read a bank balance sheet from right to left, not left to right. The conventional reading starts with assets, as though a bank first decides what to lend and then finds the money. The Treasury reading starts with liabilities, because the shape and stability of the funding determines what the asset side is allowed to look like. A bank funded overwhelmingly by insured retail deposits can hold long, illiquid assets. A bank funded by three-month wholesale paper cannot, whatever its credit quality. Funding structure is a constraint on strategy, and treating it as a residual is how banks get into trouble.

Figure 1 The bank balance sheet as Treasury sees it: stability increases up the liability stack

Every liability has two properties that matter more than its cost. The first is its contractual maturity, the date on which the bank is legally obliged to repay. The second is its behavioural maturity, the date on which the money will realistically leave. For a five-year bond these are the same. For a current account they differ by years: contractually overnight, behaviourally a multi-year liability that has sat in the same account through three rate cycles. Almost every serious Treasury question is a question about the gap between those two numbers, and almost every serious Treasury failure is a case where the behavioural assumption held until precisely the moment it was needed.

Treasury is the internal bank, and it has three mandates

Group Treasury is the counterparty of every other business line. When the retail network gathers a deposit, it does not lend that specific euro to a specific borrower. It sells the deposit to Treasury at an internal rate. When the corporate lending desk writes a five-year loan, it does not fund that loan from any identifiable liability. It buys the funding from Treasury at an internal rate. Treasury sits in the middle, absorbs the mismatch, manages it centrally, and goes to external markets only for the residual. This arrangement is not an accounting convenience. It is what allows a bank to manage interest rate and liquidity risk in one place rather than in forty.

The mandate divides into three cycles that run on different clocks, and the rest of this module follows all three. The funding cycle is annual in planning and episodic in execution: a plan agreed once a year, executed through a handful of market transactions that each take a single day. The transfer pricing cycle is continuous in application and periodic in calibration: a curve applied to every contract the bank writes, recalibrated monthly or annually. The liquidity cycle is daily and unforgiving: a cash position reconciled each morning, managed through the day, and settled against a payment-system deadline that does not move.

Cycle Clock speed Core question Primary output Who leads
Funding Annual plan, single-day execution How do we raise what we need, in the right shape? Funding plan and executed issuance Funding and Long-Term Issuance desk
Transfer pricing (FTP) Continuous application, periodic recalibration What does internal money cost, and who pays it? The FTP curve and monthly attribution ALM, with Treasury Finance
Liquidity Daily, intraday Can we meet every obligation today and for the next thirty days? Cash ladder, LCR, buffer composition Money Markets and Liquidity Management
Structural ALM Quarterly What does a rate shock do to value and to earnings? EVE and NII sensitivities, hedge programme ALM
Capital and resolution Annual, plan-driven Do we have enough loss-absorbing debt in the right entities? MREL plan, capital issuance Treasury with Group Finance

A fourth and a fifth mandate sit adjacent and cannot be cleanly separated. Structural interest rate risk in the banking book, usually shortened to IRRBB, is managed by the same people with the same data, because the cashflow model that answers a liquidity question also answers a repricing question. Capital and resolution planning overlaps because the instruments that satisfy loss-absorption requirements are debt instruments that Treasury issues, and because the amount of loss-absorbing debt a bank must hold changes its funding plan directly. This module treats funding, liquidity and FTP as the core and covers IRRBB and the capital dimension where they bind.

The funding menu runs from free and fragile to expensive and permanent

A bank does not choose a funding instrument, it chooses a position on a trade-off curve. At one end sit instruments that are cheap, short and available in unlimited quantity until the day they are not. At the other sit instruments that are expensive, long and contractually incapable of running. Every instrument in between trades cost against stability, and the funding plan is the statement of where the bank wants to sit on that curve given its asset book, its ratios and its resolution requirements.

Instrument Typical tenor Relative cost Stability under stress Key regulatory role
Retail current and savings deposits Contractually on demand Lowest High if insured, lower if not 3–10% LCR outflow; strong NSFR credit
Corporate operational deposits On demand Low Moderate — tied to the operating relationship 25% LCR outflow
Corporate non-operational deposits On demand to 1 year Moderate Low — rate-driven and mobile 40% LCR outflow
Secured funding (repo) Overnight to 1 year Low, collateral-dependent Moderate — survives longer than unsecured, but encumbers assets Encumbrance and NSFR consequences
Certificates of deposit and commercial paper 1 day to 12 months Low Low — the first market to close No NSFR credit under 6 months
Covered bonds 3 to 10 years Low for the rating achieved High — dual recourse, ring-fenced pool Encumbers the cover pool; HQLA-eligible for buyers
Senior preferred 2 to 10 years Moderate High — contractual, cannot run Full NSFR credit above 1 year
Senior non-preferred 4 to 10 years Higher High Counts towards MREL and TLAC
Tier 2 and Additional Tier 1 10 years or perpetual Highest Highest — loss-absorbing by design Regulatory capital
Securitisation Matched to the asset pool Varies Moderate — depends on market access Risk transfer and funding combined

The short end is cheap because it is a rolling promise

Deposits and money-market instruments cost little precisely because they can leave. A depositor accepts a low rate in exchange for immediate access. A money-market fund buys three-month paper at a fine spread because three months is a short exposure to a name it can stop rolling. The bank enjoys the cheap rate today and carries the risk that the roll fails tomorrow. Nothing about this is improper — it is the core trade of banking — but it is a rolling promise, and a rolling promise is only as good as the counterparty's continuing willingness to renew it.

Repo: borrowing against collateral rather than against your name

A repurchase agreement is a secured loan dressed as a sale. The bank sells a security and simultaneously agrees to buy it back at a fixed price on a fixed date; the difference between the two prices is the interest. Because the lender holds a security worth more than the cash — the excess is the haircut — the transaction survives a deterioration in the borrower's credit that would close the unsecured market entirely. Repo is therefore the funding channel that stays open longest in a stress, which is exactly why supervisors watch how much of a bank's asset base is already pledged. A security that has been repo'd out is encumbered: it is no longer available to be sold or pledged again, and it can no longer count in the liquidity buffer.

The long end is expensive because it cannot run

Bond funding is the purchase of certainty. When a bank issues a five-year senior bond, the investor cannot demand the money back. It can sell the bond to someone else, which moves the price but not the bank's cash. Every euro of term debt therefore converts a behavioural assumption into a contractual fact, and the premium over deposit funding is the price of that conversion. This is the single most useful way to read a bank's funding cost: it is not a penalty, it is the market price of removing a run from the range of possible outcomes for that portion of the balance sheet.

Covered bonds: the instrument that survives its issuer

A covered bond gives the investor two claims instead of one. It is a bond issued by the bank, so the investor has a claim on the bank. It is also secured on a ring-fenced pool of mortgages or public-sector loans that remains legally segregated if the bank fails, so the investor has a claim on that pool. This dual recourse is why covered bonds achieve ratings above the issuer's own and price materially tighter than senior debt. The cost is asset encumbrance: the pledged mortgages are no longer available to other creditors, which is why supervisors and rating agencies both cap how far a bank can go down this road.

Senior non-preferred: an instrument invented by regulation

Some instruments exist because a market wanted them; senior non-preferred exists because resolution authorities needed them. After 2008, the policy objective was to make it possible to recapitalise a failing bank by writing down its creditors rather than using public money. That requires a layer of debt that is clearly junior to deposits and operating liabilities but senior to regulatory capital, so that it can absorb losses without disrupting payments and derivatives. Senior non-preferred is that layer, and the requirement to hold enough of it is called MREL in Europe and TLAC for global systemically important banks. It prices wider than senior preferred because it is genuinely more junior, and a bank issues it because a rule says it must, not because it is the cheapest funding available.

The MREL requirement changes the funding plan in a way that is easy to miss. A bank does not merely need funding; it needs a specific quantity of bail-inable funding, issued by a specific legal entity — usually the resolution entity at the top of the group — and with a remaining maturity above one year to keep counting. That last condition creates a mechanical refinancing calendar: an MREL bond stops counting the day it drops inside twelve months to maturity, regardless of whether the cash is still there. Treasury therefore issues to a compliance deadline as well as a cash need, and that is one of the few situations in which a bank will raise money it does not economically require.

The reference case: a benchmark bond, priced in a single afternoon

Throughout this module the reference transaction is a €1bn five-year senior preferred benchmark issued by Meridian Bank. A benchmark is a bond large enough and liquid enough to be included in the main bond indices, which in the euro market means a minimum size of €500m and in practice usually €750m or €1bn. Index inclusion matters commercially: a large family of investors buys only what is in its index, so a deal that clears the size threshold reaches a materially deeper pool of demand than one that does not.

The benchmark is the reference case because it compresses the whole discipline into one day. Months of preparation — programme documentation, ratings, approvals, investor work — resolve into roughly six hours between announcement and pricing, at the end of which the bank has committed to a coupon for five years and the cash is contractually promised. There is no partial execution and no second attempt at a good price. Every other Treasury activity is either preparation for a day like that one or protection against needing one at a bad moment.

Treasury has no revenue line, and that changes everything

Every other CIB business is measured by what it earns; Treasury is measured by what it costs and what it prevents. There is no fee on a funding trade. There is no client to invoice. The output of a well-executed year is a slightly lower average cost of funds, a slightly more resilient maturity profile, and the complete absence of an incident. This asymmetry has three consequences worth stating plainly, because they explain behaviour that otherwise looks irrational from the outside.

The first consequence is that Treasury's performance is measured against counterfactuals. Did the bank fund cheaply relative to comparable issuers and to its own secondary curve? Did it avoid a maturity wall? Would it have survived the stress it did not experience? These are genuine questions with defensible answers, but they are harder to defend in a budget meeting than a revenue number, and Treasury investment cases suffer accordingly until an incident makes the argument for them.

The second is that the cost of funds is a P&L item of enormous size. A bank with €85bn of wholesale funding that improves its average spread by five basis points saves €42.5m a year for no additional risk. Very few client-facing initiatives in a bank generate €42.5m of recurring pre-tax income, and almost none do so without consuming capital. Funding cost optimisation is quietly one of the highest-return activities in the institution.

The third is that the incentive to optimise the reported ratio, rather than the underlying resilience, is permanent and must be managed. Because Treasury is judged on ratios, and ratios are reported at points in time, the temptation to arrange the balance sheet around reporting dates is structural. Supervisors know this. Reporting has moved towards daily LCR calculation and average-based metrics precisely to blunt it. The equivalent of the league-table distortion in capital markets is the quarter-end distortion in Treasury: repo books that shrink in the last week of December and reappear in January, wholesale positions that mature conveniently, and buffers that look their best on the last business day.

What a bank does when the bond market is not the answer

Term issuance is one option among several, and the alternatives have their own logic. A bank short of stable funding can issue a bond, but it can also gather deposits, pledge assets, or sell risk. Each alternative changes a different part of the balance sheet, and the choice among them is one of the more genuinely strategic decisions a Treasury function makes.

Deposit campaigns: buying stability at retail prices

Raising the rate on savings accounts is a funding decision made through the retail network. It is often the cheapest way to add stable funding, because a deposit that attracts a strong NSFR and LCR treatment can be worth more in ratio terms than its headline cost suggests. The complication is that a deposit rate cannot usually be raised for new money alone. Repricing the back book to attract new balances can cost far more than the new balances are worth, and getting that arithmetic wrong is a common and expensive error. The FTP framework exists in part to make this trade-off visible before the campaign is launched rather than after.

Central bank facilities: available, but never free of signalling

A central bank's standing facilities and targeted operations are a real funding channel with a reputational dimension. Ordinary open-market operations are routine and unremarkable. Emergency facilities are not: drawing on them, if disclosed or inferred, can be read as evidence of distress and accelerate the very outflow it was meant to cover. The targeted longer-term refinancing operations conducted by the ECB in the years to 2021 were a different animal again — cheap, long, and taken up in enormous size by banks across the euro area — and their repayment from 2022 onwards removed hundreds of billions of euros of term funding from the system on a known schedule, forcing banks back to public markets and deposit competition simultaneously.

Securitisation: funding and risk transfer in one instrument

Securitisation packages a pool of loans into securities sold to investors. Done for funding, it converts illiquid assets into cash. Done for capital, it transfers credit risk and reduces risk-weighted assets. Done for both, it does each less efficiently than a dedicated instrument would. The technique made the originate-to-distribute model possible before 2007 and its abrupt closure destroyed the banks that depended on it, which is why the modern European framework requires the originator to retain a slice of the risk and imposes transparency requirements designed to make the underlying pool analysable.

Four rules define the shape of the problem

Liquidity regulation after 2008 replaced supervisory judgement with two ratios, a sensitivity test and a loss-absorption requirement. These are covered as system specifications in Part 4 and as risk in Part 5, but they need naming here because they are the constraints inside which every decision in Part 2 is made.

Figure 2 Two ratios, two horizons: survive the month, and fund the book

The liquidity coverage ratio asks whether the bank survives thirty days. It divides the stock of high-quality liquid assets by the net cash outflows expected over a thirty-day stress period, with outflow rates prescribed by regulation rather than modelled by the bank, and requires the result to be at least 100%. Its genius is that it is comparable across institutions. Its limitation is that the prescribed outflow rates are assumptions, and a bank can satisfy the ratio completely while facing an outflow that the assumptions never contemplated.

The net stable funding ratio asks whether the balance sheet is structurally fundable over a year. It weights liabilities by how stable they are, weights assets by how much stable funding they require, and demands that the first exceed the second. Where the LCR is a stress test, the NSFR is a structural constraint: it makes funding a twenty-year mortgage book with three-month paper mathematically impossible rather than merely unwise.

Interest rate risk in the banking book asks what a rate shock does to value and to earnings. European guidelines require banks to measure the effect of six prescribed shocks on both the economic value of equity and on projected net interest income, and to explain themselves if the value impact exceeds a defined share of Tier 1 capital. Part 5 treats this in depth, because the divergence between the value lens and the earnings lens is where most ALM argument actually happens.

MREL asks whether enough of the funding stack can absorb losses. It sets a minimum quantity of own funds and eligible liabilities, issued by the right entity with enough remaining maturity, so that a failing bank can be recapitalised by writing down its investors. It is a funding requirement expressed as a capital rule, and it is the reason a bank may issue debt it does not need.

The cycle that reset the entire business between 2021 and 2024

For roughly a decade, funding was close to free, and a generation of bankers learned the wrong lesson from it. Policy rates in the euro area were at or below zero from 2014. Central bank operations supplied term funding in size at negative rates. Deposits arrived in volumes banks did not want and could not profitably deploy. In that environment the marginal value of a deposit was near zero, funding plans were modest, and the practical importance of the Treasury function inside many banks quietly declined. Deposit betas — the share of a rate rise passed on to depositors — had barely been tested in a rising cycle for fifteen years.

Figure 3 The cycle that reset the business: policy rates, spreads and the end of free funding

Three things then happened within about eighteen months. Policy rates rose at a pace without precedent in the euro area's history, taking the deposit facility rate from minus 0.5% to 4% between mid-2022 and late 2023. The targeted longer-term operations began to be repaid, withdrawing a large, cheap, term-funding stock on a published timetable. And in March 2023 two well-known banks failed within eleven days of each other, in both cases through the liability side, demonstrating that a deposit run could now happen at the speed of a mobile application rather than the speed of a queue outside a branch.

The consequences were immediate and are still working through. Deposits became genuinely competitive for the first time in a decade, and deposit beta assumptions that had been stable model parameters became live commercial decisions. Wholesale spreads widened, then partially retraced. Supervisors sharpened their attention on interest rate risk in the banking book, on the concentration and insurance status of deposits, and on the realism of behavioural models. Funding plans grew. And technology investment followed, because the March 2023 events made an argument that had been made unsuccessfully for years: that a liquidity position known at the end of the day is not the same thing as a liquidity position known now.

  1. Why can a bank with excellent assets fail on a Tuesday afternoon?

  2. Why does a covered bond price tighter than the same bank's senior preferred bond?

  3. What does a bank give up when it funds through repo rather than unsecured issuance?

  4. Why might a bank issue senior non-preferred debt it does not economically need?

  5. What is the difference between contractual and behavioural maturity, and why is it the central Treasury question?

  6. Why is a very high LCR not automatically evidence of good management?

  7. What made the 2021–2024 period a structural break rather than an ordinary cycle?

1

  1. Why can a bank with excellent assets fail on a Tuesday afternoon?

  2. What does a bank give up when it funds through repo?

  3. Why issue senior non-preferred debt the bank does not economically need?

  4. Why does a covered bond price tighter than senior preferred?

Part 1B — Capital constrains the balance sheet as tightly as liquidity does

Liquidity answers whether the bank can pay tomorrow. Capital answers whether it can absorb a loss and still be allowed to operate. Treasury sits at the intersection of the two, because the instruments that satisfy capital rules are instruments Treasury issues into the market, and the assets that consume capital are the same assets Treasury must fund. A Treasury function that understands liquidity but not capital will build a funding plan that the capital plan cannot support — and will discover the conflict in the worst possible week.

Capital is a funding structure, not a pile of money set aside

Start with the wrong mental model, because almost everyone begins there. Ask a non-specialist what bank capital is and the answer is usually some version of "money the bank keeps in reserve, in case things go wrong." That picture is intuitive, memorable, and wrong in a way that makes every subsequent regulatory concept incomprehensible. There is no vault. There is no ring-fenced account labelled "capital" that the bank draws down when a borrower defaults. If you looked at Meridian Bank's balance sheet hoping to find its €19.5bn of Common Equity Tier 1 sitting somewhere on the asset side, you would look forever.

Capital lives on the right-hand side of the balance sheet, not the left. A balance sheet has two sides. On the left sit the assets — the loans, the bonds, the cash at the central bank. On the right sit the claims on those assets — deposits, bonds the bank has issued, and, at the bottom of the queue, the shareholders. Every euro of assets is funded by exactly one euro of claims. Capital is simply the portion of that funding which comes from claims that can absorb a loss without the bank defaulting. That is the whole definition, and everything else in the prudential rulebook elaborates it.

Why the distinction matters operationally. If a depositor is owed €100 and the bank pays €98, the bank has defaulted. Deposits are a hard promise; failure to honour them in full and on time is an event of default with legal consequences that cascade. If a shareholder's stake is worth €100 and asset values fall so that it is now worth €98, nothing has happened legally at all. The shareholder simply owns something worth less. That asymmetry — the ability to lose value without triggering default — is what makes an instrument capital. It is a property of the claim, not of any asset.

So the question "how much capital does the bank have?" is really "how much of the funding stack is loss-absorbing?" Meridian funds €420bn of assets. Of that, €34bn comes from equity and subordinated instruments. If Meridian's loan book deteriorated and it had to write down €5bn of assets, the €5bn would be borne by shareholders first: assets fall to €415bn, equity falls, depositors and senior bondholders are untouched, and the bank continues to operate. That capacity to shrink the equity claim without disturbing the senior claims is what supervisors are measuring. Capital adequacy is a statement about the structure of funding, which is precisely why it belongs in a Treasury course rather than an accounting one.

Loss absorption comes in grades, and the grades are the capital stack. Not every subordinated claim absorbs loss equally well. An ordinary share absorbs loss continuously and automatically, the moment the asset value falls. A dated subordinated bond absorbs loss only when the bank has actually failed and a court or resolution authority allocates losses. Between those extremes sits a hybrid designed to absorb loss while the bank is still a going concern. European rules — the Capital Requirements Regulation, CRR, together with the Capital Requirements Directive, CRD, which member states transpose into national law — codify this into three tiers with precise eligibility criteria.

Layer What it is How it absorbs loss Typical cost (illustrative)
Common Equity Tier 1 (CET1) Ordinary shares, share premium, retained earnings, other reserves, less regulatory deductions (goodwill, most intangibles, certain deferred tax assets) First and continuously. Losses reduce retained earnings directly; no trigger, no event, no discretion required Cost of equity, typically 9–12% for a European bank
Additional Tier 1 (AT1) Perpetual instruments, no maturity, coupons fully discretionary and non-cumulative, contractual write-down or conversion to equity at a CET1 trigger of at least 5.125% While the bank is still a going concern — coupons can be cancelled at any time, and principal is converted or written down if the trigger is hit 6–9% coupon; deeply subordinated, no maturity, high volatility
Tier 2 Dated subordinated debt, original maturity of at least five years, amortised for regulatory purposes over the final five years Only in resolution or insolvency, when the authority bails in subordinated creditors. No going-concern absorption 3–5% coupon; cheapest form of regulatory capital

Read the table as a ladder of reliability, not just a ladder of cost. CET1 is expensive because it takes the first loss with no protection and no promise of return; investors price that. Tier 2 is cheap because it only bites in a scenario that, from the holder's perspective, is remote. AT1 sits in between and is genuinely unusual: a perpetual instrument whose coupon the issuer may simply decline to pay, without that constituting default. Investors accept this because they are paid for it, and because in practice coupon cancellation is a signal so damaging that boards avoid it. That practical reluctance is itself a risk — it means AT1 is loss-absorbing in theory more readily than in management's preference.

A practical Treasury consequence follows immediately. Because the tiers differ in cost by several hundred basis points, and because the regulatory requirement is expressed partly in CET1 terms and partly in total capital terms, a bank has a genuine optimisation problem: fill the requirement with the cheapest instrument that qualifies for each layer. Meridian's €19.5bn CET1, €22.5bn Tier 1 and €26.5bn total capital imply €3bn of AT1 and €4bn of Tier 2. Those are not accidents. They are the output of a deliberate calculation about the cheapest legal way to reach each threshold, and Treasury executes it through the issuance calendar.

Risk-weighted assets exist because €1bn of lending is not €1bn of risk

Suppose the rule were simply "hold 8% capital against assets." A bank holding €1bn of German government bonds would need the same capital as a bank holding €1bn of unsecured loans to speculative-grade start-ups. The second bank might lose a quarter of its book in a bad year; the first would lose approximately nothing. A size-based rule would push every bank towards the riskiest assets, since the capital cost is identical but the yield is not. Capital rules would actively subsidise recklessness.

The fix is to scale the exposure by its riskiness before applying the ratio. A risk weight is a percentage applied to an exposure to produce a risk-weighted amount. A €1bn exposure at a 25% risk weight becomes €250m of risk-weighted assets, or RWA. The capital requirement then applies to RWA rather than to the raw exposure. At an 8% total capital requirement, that €1bn exposure consumes €20m of capital rather than €80m. Same size, quarter of the capital, because the risk is a quarter as large in the regulator's calibration.

Two ways to set the weights, and the difference is worth billions. Under the standardised approach, the weights are prescribed in the regulation itself — a table lookup driven by exposure class, external rating where relevant, collateral, and loan-to-value. It is simple, auditable and conservative. Under the internal ratings-based approach (IRB), the bank estimates the risk parameters itself — probability of default, loss given default, exposure at default — from its own historical data, feeds them into a supervisory formula, and derives its own weights. IRB requires explicit supervisory permission, extensive model validation, and years of clean data history.

IRB generally produces lower weights for well-underwritten portfolios, which is the point and also the problem. A bank with genuinely good mortgage underwriting and a decade of loss data showing minimal defaults will compute a lower risk weight than the standardised table assumes, and will hold less capital. That is economically correct. It is also an invitation to optimise: the same model freedom that rewards good risk management rewards aggressive assumptions. After the 2008 crisis, supervisors found unexplained dispersion in risk weights across banks holding materially similar portfolios, which undermined confidence in reported ratios entirely.

The output floor is the answer, and CRR3 is how the EU implemented it. The final Basel III reforms — often called Basel IV in the market, though the Basel Committee does not use that name — introduced a floor on how far internal models may reduce capital below the standardised outcome. Total RWA computed under internal models cannot fall below 72.5% of total RWA computed under the standardised approach for the same portfolio. In the EU, this arrives through CRR3, applicable from 1 January 2025 with a phase-in of the floor running to 2030. For banks with large, low-risk-weight mortgage books — which describes much of European retail banking — the floor is a material constraint that binds gradually and predictably.

Why any of this belongs in a Treasury course. Consider Meridian's two lending businesses. Mortgages of €140bn at, say, an average 25% risk weight generate €35bn of RWA. Corporate loans of €90bn at an average 100% weight generate €90bn. The corporate book is smaller by a third but consumes nearly three times the capital. Funding, however, does not care: €1bn of mortgages and €1bn of corporate loans both need €1bn of funding, and both sit on the same balance sheet consuming the same leverage capacity.

So the funding decision and the capital decision cannot be taken separately. If Meridian's ALCO approves €5bn of new corporate lending, Treasury must find €5bn of funding and the bank must find roughly €685m of CET1 to support it at a 13.7% allocation. If ALCO instead approves €5bn of new mortgages, the funding requirement is identical but the capital requirement is around €171m. The same balance sheet growth carries a fourfold difference in capital consumption. Any planning process that treats funding capacity and capital capacity as separate conversations will approve business the bank cannot actually support — and will typically discover it at the point where the only remaining options are expensive.

Figure 4 The capital stack and the requirement it must clear

The requirement is a stack of stacks, and the buffers are where the pain is

Pillar 1 sets the floor and it is the least interesting part. Under CRR, every bank must hold at least 4.5% of RWA in CET1, 6% in Tier 1, and 8% in total capital. These are the hard minimums. Breaching them is a licence-threatening event: the supervisor's toolkit at that point includes withdrawal of authorisation and, if the bank is judged failing or likely to fail, referral to the resolution authority. No European bank of any size runs near these numbers, which is why they rarely feature in management discussion.

Pillar 2 is bank-specific and set by the supervisor through the SREP. The Supervisory Review and Evaluation Process is the annual assessment through which the ECB's Single Supervisory Mechanism — for significant institutions in the banking union — or the national competent authority forms a view on a bank's business model, governance, capital adequacy and liquidity adequacy. It produces a Pillar 2 Requirement (P2R), a binding add-on covering risks that Pillar 1 does not capture or does not capture adequately: interest rate risk in the banking book, concentration risk, model deficiencies, and so on. P2R plus Pillar 1 gives the Total SREP Capital Requirement, or TSCR. P2R must now be met with at least 56.25% CET1 and 75% Tier 1 under CRD.

On top of TSCR sits the combined buffer requirement, which must be met with CET1 only. Three components matter. The capital conservation buffer is a flat 2.5% of RWA for every bank. The countercyclical capital buffer is set by national authorities and varies by the jurisdiction of the exposures — it rises when credit growth is judged excessive and is released in downturns, and several European authorities have moved to a positive neutral rate, meaning a non-zero buffer even in normal conditions. Systemic buffers — the G-SII buffer for global systemically important institutions, the O-SII buffer for other domestically important ones, and the systemic risk buffer — add further CET1 for banks whose failure would do disproportionate damage.

Pillar 2 Guidance sits above all of it and is deliberately soft. P2G is the supervisor's expectation of additional CET1 to withstand stress, informed by stress test results. It is not a legal requirement, and breaching it does not trigger automatic consequences. What it does trigger is supervisory attention, and a bank operating below its P2G will find the subsequent dialogue uncomfortable and the constraints on its plans real. The ordering matters: P2G sits above the buffers, so a bank dipping into P2G has not yet breached anything automatic.

Meridian's stack, illustrative throughout. Pillar 1 CET1 of 4.5%, a P2R CET1 component of roughly 1.1%, a capital conservation buffer of 2.5%, an O-SII buffer of 1.5%, and a countercyclical buffer of around 0.8% given its exposure mix, sum to an overall CET1 requirement of 10.4%. Meridian reports 13.7%. The difference — 330 basis points, or roughly €4.7bn of CET1 at current RWA — is management headroom. That number, not the ratio itself, is what the board tracks.

Now the mechanism that makes the buffers bite: the maximum distributable amount. Breaching the combined buffer requirement does not close the bank. Nothing dramatic happens on the day. What happens is automatic and, for Treasury, decisive: the bank enters the MDA framework, which caps the total amount it may distribute as a proportion of its distributable profits, on a sliding scale by quartile of the buffer shortfall. In the lowest quartile the payout factor is zero. And "distributions" under CRD is a broad term covering dividends on ordinary shares, variable remuneration and discretionary pension benefits, and — the one that matters here — coupon payments on Additional Tier 1 instruments.

That last item is why AT1 investors watch the distance to MDA obsessively. An AT1 holder's coupon is not protected by a covenant. It is protected only by the issuer having enough headroom above its combined buffer requirement and enough distributable items under national company law to pay it. A bank that slips 50 basis points into its buffer may find its AT1 coupon restricted by regulation regardless of what management wants. The instrument's price therefore reflects, continuously, the market's estimate of the issuer's distance to MDA. Meridian's 330 basis points of headroom is comfortable; a bank running 60 basis points would see its AT1 trade materially wider for that reason alone.

And this constrains the funding plan directly. If Meridian's headroom to MDA compressed to, say, 100 basis points, three things would follow within weeks. New AT1 issuance would become expensive or impossible, because investors would demand a large premium for coupon uncertainty. Existing AT1 with call dates approaching would face a difficult decision — calling requires supervisory permission and sufficient capital, but not calling damages the bank's standing in a market that expects calls. And equity issuance, the obvious remedy, is cheapest precisely when it is least needed. The Treasury lesson is that capital headroom is an option that must be bought in good conditions, because it cannot be bought in bad ones.

The leverage ratio ignores risk on purpose, and that is why it binds on Treasury

A second constraint runs alongside the risk-based one and deliberately refuses to look at risk. The leverage ratio is Tier 1 capital divided by total exposure — on-balance-sheet assets plus derivative exposures plus securities financing transactions plus off-balance-sheet commitments after conversion factors — with no risk weighting whatsoever. Under CRR, the binding minimum is 3%, with an additional leverage ratio buffer for global systemically important institutions set at 50% of the G-SII risk-based buffer.

It exists because the risk-based framework depends on models, and models can be wrong. If a bank's internal models systematically understate risk — through bad data, optimistic assumptions, or a structural break the history does not contain — its risk-based ratio will look healthy right up to the point of failure. The leverage ratio provides a backstop that no amount of modelling can improve: it counts euros of exposure, full stop. It is crude by design, and its crudeness is the feature.

The consequence is that it binds precisely where risk weights are lowest. For a portfolio of unsecured corporate loans at 100% risk weight, an 8% total capital requirement implies capital of 8% of exposure — far above the 3% leverage minimum, so the risk-based rule binds. For a reverse repo against government collateral at a 2% risk weight, the risk-based requirement is 0.16% of exposure while the leverage requirement remains 3%. The leverage ratio is nearly twenty times more demanding. The rule that ignores risk becomes the binding constraint exactly on the low-risk, high-volume activity.

And that activity is Treasury's activity. Repo and reverse repo books, central bank reserves, high-quality liquid securities, short-dated interbank placements, and the collateral flows around cleared derivatives — these are the instruments through which a Treasury manages liquidity and short-term funding. All of them are large in notional and small in risk weight. All of them consume leverage capacity at full face value while consuming almost no risk-weighted capacity. A Treasury that expands its liquidity buffer by €10bn of central bank reserves has added €10bn to leverage exposure and roughly nothing to RWA.

Which produces a genuine institutional contradiction. Suppose Meridian's board, reacting to a stressed peer, instructs Treasury to increase the liquidity buffer from €62bn to €72bn, and separately instructs the CFO to improve the leverage ratio from 4.8% towards 5.2%. Holding Tier 1 at €22.5bn, the additional €10bn of exposure moves the leverage ratio from 4.84% to 4.74%. The two instructions are not merely in tension; they are arithmetically opposed. Executing both requires either €2.5bn of new Tier 1 capital or a €10bn reduction of exposure somewhere else on the balance sheet.

Resolving that is an ALCO decision, not a desk decision. The Treasury desk cannot choose between a liquidity target set by the board's risk appetite and a leverage target set by the same board's capital plan. What the desk can and must do is make the conflict visible, quantified, and dated — showing precisely what each instruction costs in the currency of the other, and what combinations are actually feasible. Asset-Liability Committee exists for exactly this class of decision: allocating a scarce balance sheet across competing binding constraints. A Treasury function that silently absorbs contradictory mandates ends up failing one of them at a moment it does not control.

ICAAP mirrors ILAAP: same governance, same data, different question

Supervisors do not rely solely on their own calculations. They require each bank to run its own assessment of how much capital it needs given its specific business model, risk profile and strategy, and to be able to defend that assessment. The Internal Capital Adequacy Assessment Process, ICAAP, is that exercise. It is mandated by CRD, elaborated by EBA guidelines and, for banks under the SSM, by the ECB's published expectations. It is a process with an annual documentary output, not a document with an annual process.

ICAAP has two perspectives and both must hold. The normative perspective asks whether the bank will meet all its regulatory capital requirements — Pillar 1, P2R, buffers, leverage, MREL — over a multi-year horizon, under a baseline plan and under adverse scenarios. The economic perspective asks a different question: whether the bank's internally measured economic value of capital covers all material risks measured on the bank's own methodology, including risks the regulatory framework does not quantify. A bank can be comfortable on one and tight on the other, and the supervisor will ask about the gap.

Stress testing is where ICAAP becomes concrete. The bank projects its capital ratios under scenarios of varying severity: a domestic recession, a property correction, a sharp rate move, an idiosyncratic reputational event, and combinations. The projection must run through the actual mechanics — credit losses raising impairments and reducing retained earnings, rating migration inflating RWA even where exposures have not defaulted, revenue compression, and any management actions the bank claims it would take. That last item receives scrutiny, because management actions assumed in a stress test have a way of being unavailable in an actual stress.

Alongside the internal exercise sits the supervisory one. The EBA coordinates an EU-wide stress test on a biennial cycle, with the ECB running a parallel exercise for the significant institutions it supervises directly that are not in the EBA sample. These use a common adverse scenario developed with the ESRB and a largely constrained-balance-sheet methodology, so results are comparable across banks. Outcomes feed the SREP and, in particular, inform the calibration of P2G. A weak stress test result does not produce a fine; it produces a higher capital expectation, which produces a smaller balance sheet or a larger issuance programme.

The capital plan is where all of this becomes a Treasury instruction. It is the multi-year projection reconciling three things that compete for the same resource: business growth, which consumes capital through RWA; distributions to shareholders, which consume capital directly; and issuance, which supplies it. The plan states, quarter by quarter, what RWA is expected to be, what earnings will retain, what dividends and buybacks will pay out, and therefore what AT1 and Tier 2 must be issued and when. Treasury owns the last column and must deliver it into whatever market exists on the date.

The parallel to ILAAP is exact and worth making explicit. The Internal Liquidity Adequacy Assessment Process asks whether the bank has enough liquidity for its risk profile; ICAAP asks whether it has enough capital. Same board approval, same ALCO ownership, same three-lines-of-defence governance, largely the same underlying data — the exposure records, the maturity profiles, the counterparty attributes. Increasingly, supervisors expect the two to be internally consistent: a stress scenario that assumes a deposit run in the ILAAP and stable funding in the ICAAP is not a coherent view of the world. Banks that maintain separate teams, separate models and separate scenario libraries for the two processes reliably produce inconsistencies that the SREP dialogue finds.

ICAAP and ILAAP share a data problem. Both require exposure-level data with consistent risk attributes, both require scenario engines that can revalue the balance sheet, and both require a reproducible audit trail from source system to reported number. Banks that built these as separate programmes typically ended up with two data pipelines, two versions of the exposure master, and two answers to questions like "what is our exposure to this counterparty." The architectural conclusion is that the risk and finance data layer should be one asset serving multiple assessments, with scenario definition and aggregation logic sitting above it rather than being rebuilt inside each process.

Capital is a price, not only a constraint, and that price must reach the front office

A constraint that is not priced will be consumed until it binds. If capital is treated purely as a regulatory ceiling, every business line will book profitable-looking business until the group runs out of headroom, at which point capital is rationed administratively — by whoever argues most effectively in the planning meeting. The alternative is to attach a price to capital consumption at the point of decision, so that a transaction which does not cover its capital cost looks unattractive to the person originating it.

Two mechanisms exist for this and they must be reconciled. Some banks embed a capital charge inside the funds transfer pricing framework, so the FTP rate quoted to the business already reflects both liquidity and capital cost. Others keep FTP purely about funding and liquidity, and apply a separate capital charge in the profitability calculation. Both work. What fails is having both without a clear boundary — a Treasury that includes a capital component in FTP while Finance also deducts a capital charge in the RAROC calculation will double-charge the business, making sound transactions look uneconomic. The mirror failure, where each function assumes the other is charging, means the cost reaches nobody.

The reconciliation must be documented and owned. The cleanest arrangement states explicitly that FTP prices the funding and liquidity dimension — the term liquidity premium, the contingent liquidity cost of committed lines, the cost of the buffer the transaction obliges the bank to hold — and that a separate, clearly identified capital charge prices the regulatory capital the transaction consumes. Anyone able to see a business line P&L should be able to point to exactly one line for each cost. Ambiguity here is not a presentational issue; it changes which loans get written.

RAROC, built from zero. Risk-adjusted return on capital compares what a transaction earns, after all its costs including the expected credit loss, against the capital it consumes. Return on capital alone would flatter risky lending, since a high-margin loan looks excellent until its losses arrive. RAROC handles that by deducting expected loss — the statistically anticipated annual loss, probability of default multiplied by loss given default multiplied by exposure — as a cost rather than treating it as a surprise. The result is compared to a hurdle rate, normally the bank's cost of equity plus a margin, and transactions below the hurdle destroy shareholder value regardless of how positive their accounting profit looks.

Note what the example demonstrates about sequencing. The loan looked profitable at every intermediate step. Positive spread over FTP, comfortable coverage of expected loss, positive net income. Only when capital was priced did it become clear that the transaction destroys value. Banks that stop the calculation one line early — and many did, for years — build large portfolios of transactions that are individually defensible and collectively unable to earn the cost of the equity supporting them.

One methodological caution. The choice of capital allocation basis matters. Allocating at the actual CET1 ratio of 13.7% charges the business for the bank's full capitalisation including management headroom; allocating at the regulatory requirement of 10.4% charges only the mandated amount and leaves headroom unpriced at the transaction level. Neither is wrong, but the choice must be consistent across business lines and stable over time, or the framework becomes a device for arguing rather than deciding.

Capital and MREL overlap without being the same thing

MREL is a resolution concept, not a going-concern one. The Minimum Requirement for own funds and Eligible Liabilities, established under the Bank Recovery and Resolution Directive and set for each bank by its resolution authority — the Single Resolution Board for banks in the banking union — answers a different question from capital adequacy. Capital asks whether the bank can absorb losses and keep operating. MREL asks whether, if the bank fails, there is enough loss-absorbing and recapitalising liability to bail in so that the institution can be restructured without public money.

The requirement therefore covers a wider set of instruments. Everything that counts as own funds counts for MREL. But so do certain unsecured liabilities that are not capital at all, provided they meet eligibility conditions: remaining maturity of at least one year, not derived from derivatives, not a covered deposit, contractually or structurally subordinated where the authority requires subordination. Senior non-preferred debt — a class created specifically by EU law to sit below ordinary senior unsecured but above Tier 2 — is the workhorse instrument here. It counts fully for MREL and not at all for capital.

Meridian's position, illustrative. The requirement is 26.5% of RWA, so €37.6bn against RWA of €142bn. Current eligible stock is 28.1%, or €39.9bn, giving headroom of roughly €2.3bn. Note that the requirement is also expressed as a percentage of the leverage exposure measure, and a bank must satisfy both; for many European banks the RWA-based measure binds, but a bank with low average risk weights can find the leverage-based MREL requirement is the tighter of the two.

Instrument Counts as capital? Counts for MREL? Relative cost (illustrative) Typical use
CET1 Yes, highest quality Yes Highest — cost of equity ~10% Core solvency; not issued to meet MREL
Additional Tier 1 Yes, Tier 1 Yes High — 6–9% coupon, MDA-linked coupon risk Fill Tier 1 gap between CET1 and 6% requirement cheaply relative to equity
Tier 2 Yes, total capital Yes Moderate — MS+250 to MS+400 typical Fill total capital requirement; expensive way to meet MREL alone
Senior non-preferred No Yes Lower — MS+130 to MS+180 typical Primary MREL-filling instrument once capital layers are complete
Senior preferred Generally no Only where subordination is not required and conditions are met Low — Meridian trades ~MS+82 General funding; limited MREL role for most resolution strategies
Covered bonds No No — excluded as secured Lowest — often below senior preferred Secured term funding; consumes encumbrance capacity

The optimisation is real and annual. Once Meridian has met its capital requirements, the remaining MREL gap can be filled with either senior non-preferred or additional Tier 2. Tier 2 counts for both, which sounds efficient, but if the capital requirement is already met the extra Tier 2 buys no capital benefit while costing perhaps 100 basis points more than senior non-preferred for the same MREL credit. On a €3bn tranche that is roughly €30m a year of avoidable cost, illustrative — a number large enough to justify a dedicated analysis in the annual funding plan.

Pulling in the other direction are three considerations. Tier 2 has a five-year minimum original maturity and amortises for capital purposes in its final five years, so a bank wanting stable capital must issue long and refinance early. Senior non-preferred requires only one year of remaining maturity for MREL eligibility, which makes it cheaper but creates a rolling refinancing need that concentrates in the year before maturity. And investor bases differ: the pool that buys Tier 2 is not the pool that buys senior non-preferred, and issuing only into one keeps the other unfamiliar with the credit. Maintaining presence across the stack has an option value that pure cost minimisation misses.

A final structural point. Where the resolution strategy is single point of entry — the whole group resolved at the parent — MREL-eligible instruments are typically issued by the holding company and downstreamed to operating subsidiaries as internal MREL. Where it is multiple point of entry, material subsidiaries issue externally themselves. This choice determines who issues, in what currency, under which law, and to which investor base, and it is set by the resolution authority in the resolution plan rather than chosen by the bank. Treasury's issuance calendar is downstream of a decision it does not make.

  1. Why is "money set aside in reserve" the wrong mental model for bank capital, and what is the right one?

  2. Meridian breaches its combined buffer requirement by 40 basis points. Describe precisely what happens.

  3. Why does the leverage ratio bind specifically on Treasury activities rather than on lending?

  4. How can FTP and a capital charge double-count, and what prevents it?

  5. Meridian has met all capital requirements and has a €3bn MREL gap. Why is senior non-preferred usually the better instrument than additional Tier 2?

  6. A €1bn mortgage book at 25% risk weight and a €1bn corporate book at 100% both need €1bn of funding. What follows for planning?

  7. What does the output floor do, and why was it introduced?

Part 1C — Accounting, ratings and ESG decide what Treasury is allowed to do

Three lenses sit between a Treasury decision and its consequences. Accounting determines whether a decision shows up in earnings, in equity, or nowhere at all. Ratings determine what the decision costs, converting the entire internal state of the balance sheet into a single letter that the market prices. Sustainability frameworks determine, increasingly, whether a given pool of investors will participate in a transaction in the first place. None of the three changes the economics of a trade. All three change whether the trade can be done, at what price, and how it will be perceived.

IFRS 9 classification decides whether a loss is visible at all

Two banks can hold the identical bond and report completely different numbers. Bank A buys a €1bn ten-year government bond at par. So does Bank B. Rates rise 200 basis points; the bond's market value falls to roughly €860m, illustrative. Bank A reports no loss whatsoever — its income statement and its equity are untouched. Bank B reports a €140m reduction in equity and in regulatory capital. Same asset, same market, same date. The difference is a classification decision made at the moment of purchase, and it is one of the most consequential choices in bank Treasury.

The framework is IFRS 9, as endorsed for use in the EU. International Financial Reporting Standard 9, Financial Instruments, replaced the earlier IAS 39 for annual periods beginning on or after 1 January 2018. EU-listed banks apply it in the form endorsed by the European Commission, which is identical to the IASB text with one significant exception discussed later in this section. IFRS 9 governs how financial assets are classified, how they are measured, and how impairment is recognised.

For debt instruments, there are exactly three outcomes. Amortised cost: the asset is carried at the price paid, adjusted for the amortisation of any premium or discount and for expected credit losses. Market price movements are invisible. Fair value through other comprehensive income, FVOCI: the asset is carried at market value, but the gain or loss goes to a reserve within equity called other comprehensive income rather than through the income statement. Fair value through profit or loss, FVPL: the asset is carried at market value and every movement flows straight through earnings.

Two tests decide which one applies, and neither is optional. The first is the business model test, which asks what the entity actually intends to do with the portfolio the asset belongs to. Holding to collect contractual cash flows points to amortised cost. Holding both to collect cash flows and to sell points to FVOCI. Holding for trading or any other purpose points to FVPL. The assessment is made at portfolio level, based on observable behaviour and documented objectives, not on management's stated aspiration for an individual instrument.

The second is the contractual cash flow characteristics test, universally called SPPI. It asks whether the contractual terms give rise, on specified dates, to cash flows that are solely payments of principal and interest on the principal amount outstanding. Interest here means consideration for the time value of money, credit risk, other basic lending risks and costs, and a profit margin. A plain fixed-rate bond passes. A floating-rate note passes. A bond whose coupon is linked to equity index performance fails, because the return is no longer compensation for lending. A structure with leverage fails. Anything failing SPPI goes to FVPL regardless of business model — the standard permits no escape.

The interaction of the two tests produces the classification. SPPI-passing assets in a hold-to-collect model go to amortised cost. SPPI-passing assets in a hold-to-collect-and-sell model go to FVOCI. Everything else goes to FVPL. Additionally, an entity may irrevocably designate an asset at FVPL at initial recognition if doing so eliminates an accounting mismatch, and equity instruments not held for trading may be irrevocably designated at FVOCI, where gains and losses are never recycled to profit or loss.

Now the concrete consequence when rates rise, which is the version to remember. An amortised-cost bond shows nothing. Its carrying value is unchanged; the economic loss exists and is real but is nowhere in the financial statements, disclosed only in the fair value note. An FVOCI bond shows the full loss in the OCI reserve, reducing equity, but reports no impact in profit or loss. An FVPL bond shows the loss immediately in earnings, hitting reported profit and, through it, retained earnings and any performance metric built on net income.

For European banks, the FVOCI case carries a regulatory sting that surprises newcomers. For a period after the introduction of the current framework, a prudential filter allowed banks to neutralise unrealised gains and losses on certain FVOCI sovereign exposures for regulatory capital purposes. That filter was removed for most purposes as CRR was phased in, so unrealised FVOCI losses on debt instruments now generally flow through to CET1. An FVOCI liquidity buffer therefore transmits interest rate volatility directly into the regulatory capital ratio and, by extension, into the distance to MDA. This is the mechanism by which a rate move with no credit content and no cash consequence compresses a bank's capital headroom.

Classifying the liquidity buffer is one of the most consequential accounting choices a bank makes

The buffer sits precisely where the trade-off is sharpest. Meridian holds €62bn of high-quality liquid assets. That portfolio exists for one purpose: to be convertible into cash quickly, at predictable value, in a stress. But the same portfolio is large enough — nearly 15% of total assets — that its accounting treatment materially drives reported equity and capital. The classification decision therefore sits at the intersection of a liquidity objective and an earnings-stability objective, and the two want opposite answers.

Take amortised cost first and follow it to its conclusion. Classify the buffer as hold-to-collect at amortised cost and reported earnings and capital become immune to rate movements. In a period of rising rates this is enormously attractive: peers reporting FVOCI buffers show capital ratios sagging on unrealised losses while yours does not move. The problem appears only when the buffer is used. Selling an amortised-cost asset crystallises the loss into profit or loss immediately, converting an invisible economic loss into a visible accounting one at the worst moment.

Worse, the sale attacks the classification itself. The hold-to-collect business model assessment rests on actual behaviour. IFRS 9 permits some sales without disturbing it — sales close to maturity, sales due to credit deterioration, infrequent or individually insignificant sales. But sales that are more than infrequent and more than insignificant call the business model into question for the entire portfolio, potentially forcing reclassification. A bank that sells a meaningful slice of its amortised-cost buffer risks not only the realised loss on what it sold but a measurement change across what remains. That is a powerful disincentive to sell, embedded in the accounting, operating precisely when selling is what liquidity risk management requires.

FVOCI produces the mirror position. Market movements pass into OCI and, through the removal of the prudential filter, into CET1. The bank's capital ratio becomes rate-sensitive in a way that requires explanation to investors and supervisors every quarter. In exchange, the asset is genuinely saleable: it is already carried at fair value, so a sale realises no new accounting surprise. On disposal the accumulated OCI is recycled to profit or loss for debt instruments, which does affect earnings, but the balance sheet and capital have already absorbed the mark. The liquidity buffer behaves the way a liquidity buffer is supposed to behave.

The March 2023 episode made the trade-off unmistakable. A US regional bank, Silicon Valley Bank, had accumulated a very large held-to-maturity securities portfolio funded substantially by uninsured deposits. Under US GAAP, held-to-maturity securities are carried at amortised cost and, under the tailored regime that then applied to banks of that size, the unrealised losses were largely excluded from its regulatory capital. As rates rose sharply through 2022, the portfolio developed a substantial unrealised loss that the headline capital ratio did not reflect. When deposit outflows forced asset sales, the loss became real and public simultaneously, and the resulting confidence shock triggered a deposit run that ended the bank within days.

The European framework starts from a different place, and it is worth being precise about how. IFRS 9's hold-to-collect model is not identical to the US held-to-maturity category — it is assessed at portfolio level on the basis of actual behaviour rather than through an instrument-by-instrument intent election, and it carries the tainting consequences described above. EU prudential rules apply capital requirements and the leverage ratio to all significant institutions without the size-based tailoring that applied in the US case, and, critically, most European banks do not benefit from a filter excluding FVOCI unrealised losses from CET1. Supervisory attention to interest rate risk in the banking book and to the economic value of equity is also more explicit. None of this makes the failure mode impossible in Europe. It means the specific combination of concealed loss, excluded capital impact and concentrated uninsured funding that occurred there is harder to assemble here.

The realistic answer is a deliberate split, sized and governed. Most European banks run a tranched buffer: a liquid, FVOCI-classified core sized to cover the severe short-horizon outflow assumptions and intended to be genuinely monetisable, and an amortised-cost tranche of longer-dated, higher-yielding paper held for structural reasons and earnings stability. The split is a board-level risk appetite decision, not a Treasury preference, because it trades reported capital volatility against real crisis flexibility. What is indefensible is arriving at the split by accident — letting accounting convenience determine buffer composition and then discovering the constraint in a stress.

Hedge accounting exists because economically sensible hedges create accounting noise

Begin with the problem rather than the solution. Meridian has issued €1bn of five-year fixed-rate senior preferred debt. The board's view is that fixed-rate funding creates unwanted exposure to rate movements relative to a largely floating asset base, so Treasury enters an interest rate swap: receive fixed, pay floating. Economically the position is now clean — the bond's fixed coupon is offset, and Meridian effectively pays floating. The risk has been removed.

The accounting, left alone, does the opposite of reflecting this. The bond is a liability measured at amortised cost, so its carrying value does not move with rates. The swap is a derivative, and under IFRS 9 all derivatives are measured at fair value through profit or loss without exception. Rates fall 100 basis points: the swap gains, say, €45m, illustrative, and that gain flows straight through earnings. The bond's economic value has risen by an offsetting amount, but its carrying value is unchanged and nothing appears. Meridian reports a €45m windfall it did not earn, and will report the reversal later. A hedge that eliminated economic risk has created reported earnings volatility from nothing.

Hedge accounting is the mechanism that repairs the mismatch. It permits an entity, subject to conditions, to change the normal measurement of either the hedged item or the hedging instrument so that the offsetting economics appear together in the same period. It is elective — a bank may hedge economically without applying it, accepting the P&L noise — and it is conditional, because the alternative would be a licence to smooth earnings arbitrarily.

Fair value hedges adjust the hedged item. Where the exposure is to changes in the fair value of a recognised asset or liability — the fixed-rate bond above — the hedged item's carrying value is adjusted for the hedged risk, with that adjustment recognised in profit or loss. The bond's carrying value now moves with rates for the interest rate component, offsetting the swap's fair value movement in the same line and the same period. Reported earnings show only the small residual: hedge ineffectiveness. Fair value hedges stabilise reported earnings against balance sheet revaluation.

Cash flow hedges defer the hedging instrument's gain. Where the exposure is to variability in future cash flows — floating-rate funding whose coupon resets with the index, or a highly probable forecast transaction — the effective portion of the derivative's fair value change is recognised in OCI and held in a cash flow hedge reserve, then recycled to profit or loss in the periods when the hedged cash flows affect earnings. The derivative's mark stops distorting the current period. Cash flow hedges stabilise reported future interest flows and remove the derivative's mark-to-market from present earnings.

What both demand is documentation and evidence, and the bar is not trivial. At inception the entity must formally designate and document the hedging relationship: the risk management objective and strategy, the hedged item, the hedging instrument, the nature of the risk being hedged, and how effectiveness will be assessed. IFRS 9 replaced IAS 39's mechanical 80–125% effectiveness corridor with a principles-based test requiring an economic relationship between item and instrument, that credit risk does not dominate the value changes, and that the hedge ratio reflects actual quantities used. This is more flexible in principle and more demanding in practice, because judgement has to be evidenced rather than a ratio computed.

Macro hedging is where the framework strains. A bank does not hedge one bond. It hedges the net interest rate position of a banking book containing millions of instruments whose composition changes every day as loans are drawn and repaid, deposits arrive and leave, and prepayments run. The standard's model designates specific items in specific relationships. A portfolio whose contents change daily fits that model badly: hedged items disappear through prepayment, new items arrive, and the designations must be continuously torn down and rebuilt. The administrative burden is severe and the risk of accidental de-designation — with the resulting reclassification of accumulated amounts — is real.

Hence the EU carve-out, which is the endorsement difference mentioned earlier. When the EU endorsed IAS 39, it carved out certain provisions relating to portfolio fair value hedges of interest rate risk, permitting an approach that better fits how European banks — particularly those with large fixed-rate mortgage books and core deposit bases — actually manage banking book rate risk. Notably, it accommodates the designation of core demand deposits in a portfolio hedge, which the unmodified standard does not permit because such deposits are repayable on demand and their fair value is deemed not to fall below the amount payable. Banks may continue to apply the IAS 39 hedge accounting requirements, including the EU carve-out, rather than IFRS 9's hedge accounting chapter — an accounting policy choice many European banks have made and retained. The IASB's long-running Dynamic Risk Management project seeks a permanent model; until it lands and is endorsed, the carve-out remains structurally important to European ALM.

Which makes hedge accounting a live constraint on ALM strategy. If the economically correct hedge cannot achieve hedge accounting, the choice is between accepting the economic risk and accepting the earnings volatility. That choice is made in ALCO, and it is made frequently. Structures are sometimes selected for their accounting tractability rather than their economic precision — a slightly imperfect hedge that qualifies may be preferred to a perfect one that does not. Anyone who dismisses this as a back-office matter has not sat in the meeting where a hedge is rejected because the documentation cannot be sustained.

Hedge accounting is a data and lineage problem before it is an accounting one.

Every hedge relationship is a linked object: a designation document, an identified hedged item or portfolio, a specific derivative with a specific notional and start date, an effectiveness assessment methodology, a series of periodic effectiveness results, and the accounting entries that follow from them. Each link must be reproducible on demand, years later, for an auditor or a supervisor who will ask why a particular relationship was designated as it was and what the effectiveness test showed in a given quarter.

This makes hedge relationship data one of the harder objects in a Treasury platform. It spans systems — the trading system holds the derivative, the core banking system holds the hedged item, the accounting engine holds the entries, and the documentation frequently lives in a document store outside all three. It is temporal, since designations start, change and end, and the historical state must be recoverable rather than overwritten. And it is unforgiving: a broken link between the derivative and its designated item is not a reporting inconvenience but a potential loss of hedge accounting, with retrospective P&L consequences.

The practical implications are consistent across implementations. Hedge relationships need a first-class data model with their own identifiers and full version history, not a spreadsheet and a shared drive. Effectiveness testing should run on the same data as the accounting entries, not on a parallel extract. Designation and de-designation need workflow with approval and audit trail. And the whole object needs to survive system migration, which is where a surprising number of banks have discovered their historical hedge documentation was never actually reproducible.

A rating is a funding cost expressed as a letter

What an agency is actually doing. A credit rating is an opinion on the likelihood that a specific obligation will be repaid in full and on time. For a bank, the agencies — principally S&P Global Ratings, Moody's and Fitch, all registered and supervised in the EU under the CRA Regulation and overseen by ESMA — build that opinion in layers rather than as a single judgement, and understanding the layers explains almost everything about how ratings behave.

Layer one is the standalone credit profile. This is what the bank would be rated on its own resources, before any external support. It is built from an assessment of the operating environment — the economic and industry risk of the markets the bank operates in, which effectively anchors the starting point by geography — then adjusted for bank-specific factors. Capitalisation, measured on the agency's own metric rather than the regulatory ratio. Asset quality, meaning the level and trend of problem exposures and concentrations. Earnings, with weight on recurring rather than one-off profitability. Funding and liquidity, meaning deposit stability, wholesale reliance, maturity profile and buffer adequacy. Governance and risk management sit across all of it.

Layer two is external support. Historically, large European banks received substantial rating uplift from the assumption that governments would rescue them. BRRD and the bail-in framework were designed explicitly to remove that assumption, and the agencies reduced or eliminated sovereign support uplift for most EU banks accordingly. What partly replaced it is a different form of uplift: where a bank has issued a large stock of MREL-eligible subordinated liabilities, agencies may rate senior preferred debt above the standalone profile, on the reasoning that the subordinated layer would absorb losses first in resolution and senior preferred would be protected. Meridian's €39.9bn of MREL stock is not only a regulatory compliance item; it is a rating input.

Layer three is notching by instrument. This is where multiple ratings for one bank come from, and it is largely mechanical. From the anchor, deposits and senior preferred obligations sit at or near the top. Senior non-preferred is notched down, typically one notch, reflecting contractual subordination. Tier 2 is notched further, typically two notches or more below the anchor. AT1 is notched furthest, typically four or more notches, reflecting deep subordination, the write-down or conversion trigger, and coupon discretion. A bank with an A senior preferred rating might carry BBB+ senior non-preferred, BBB Tier 2 and BB+ AT1 — illustrative, but the shape is standard.

The practical reading is important. When someone says a bank's AT1 is rated several notches below its senior debt, this is not the agency making a separate negative judgement about the AT1. It is the same judgement about the same bank, mechanically adjusted for where the instrument sits in the loss hierarchy. Which means the notches move together: an upgrade or downgrade of the anchor moves the whole ladder in parallel.

What agencies look at Why it matters What Treasury can influence
Capitalisation on the agency's own metric Drives the standalone profile; agency measures differ from CET1 and can move independently of the regulatory ratio Indirectly — through capital instrument issuance and by ensuring the agency's methodology is correctly applied to the bank's actual position
Asset quality and concentration Non-performing exposure levels and single-name or sector concentration are core downgrade triggers Little directly; Treasury's role is accurate presentation of trends and management actions
Earnings quality and stability Recurring earnings support internal capital generation; volatile or one-off-dependent earnings are discounted Limited; net interest income sensitivity is partly an ALM outcome
Funding profile and wholesale reliance Deposit stability, wholesale share, maturity concentration and currency mismatch are directly assessed Substantially — funding plan design, maturity laddering, investor and instrument diversification are direct Treasury decisions
Liquidity buffer size and quality LCR and NSFR are inputs, but agencies apply their own stress assumptions and look at buffer composition and encumbrance Substantially — buffer sizing, composition, encumbrance discipline and the credibility of the contingency funding plan
MREL stock and subordination Determines whether senior preferred receives uplift above the standalone profile Substantially — MREL instrument mix and issuance pace are Treasury-owned
Governance and risk management A qualitative modifier that can cost a notch on its own Indirectly — the quality and consistency of what is presented is itself evidence

Now quantify the stakes, illustratively. Suppose Meridian is downgraded one notch, from A/A2 to A−/A3. Assume senior preferred spreads widen from mid-swaps plus 82 basis points to mid-swaps plus 107 — 25 basis points, a plausible magnitude for a single notch in the A category, illustrative. On the €14bn annual funding plan, the incremental cost is roughly €35m per year, building as the stock refinances. On the full €85bn wholesale book, once entirely refinanced at the wider level, the run-rate cost is roughly €212m per year. Against a bank of Meridian's size, that is a material and permanent earnings reduction from a rating action that changed no cash flow on the day it occurred.

The cost does not stop at spread, and this is the part that surprises people. Some investors and mandates have rating floors — a fund permitted to hold only single-A or better paper becomes a forced seller or a non-participant on a downgrade below the threshold. Some corporate treasury policies limit deposit placement by counterparty rating, so a downgrade can trigger deposit withdrawal that has nothing to do with the depositor's view of the bank and everything to do with an internal policy document. Losing a segment of the investor base narrows the order book on the next transaction, which widens pricing beyond what the credit deterioration alone would justify.

Rating triggers in contracts turn a downgrade into a cash event. Derivative master agreements — typically ISDA agreements with a Credit Support Annex — frequently contain provisions linked to the rating of one or both parties. A downgrade below a specified level can require additional collateral to be posted, reduce or eliminate the threshold below which no collateral is required, or in some cases give the counterparty the right to terminate outstanding transactions at mid-market plus costs. Similar provisions appear in securitisation documentation, where the issuer's downgrade can require replacement of the account bank or the swap counterparty, and in certain commercial contracts and clearing arrangements.

The liquidity consequence is immediate and must be modelled. LCR includes a specific outflow category for additional collateral requirements arising from a three-notch downgrade, and the ILAAP should quantify contractual trigger exposure at one, two and three notches. For a bank of Meridian's derivative footprint, a three-notch downgrade might require several hundred million euros of additional collateral within days, illustrative. That is not a reputational matter. It is a cash outflow with a contractual deadline, arriving at exactly the moment the bank's access to funding has deteriorated — which is why rating triggers are catalogued, monitored and, where possible, negotiated out of new documentation.

Rating agency dialogue is a managed process, not an annual formality

The cycle is regular and the content is substantial. Each agency conducts at least an annual review, typically involving a full-day management meeting supported by a detailed information pack, followed by rating committee deliberation and publication of a rating action or affirmation with a research report. Between annual reviews there is continuing dialogue: notification of material transactions, discussion of strategic changes, and engagement around results. Treasury does not own the relationship alone — it is normally shared with the CFO's office and investor relations — but Treasury owns a large share of the content.

What the bank presents is more than the published accounts. The pack normally covers strategy and business model evolution, financial performance and outlook, asset quality by portfolio with forward-looking commentary, capital trajectory including the capital plan and any planned issuance, and — squarely Treasury's territory — the funding plan, the liquidity position, the buffer composition, the encumbrance level, the maturity profile and the contingency funding plan. Agencies test whether the funding plan is credible against the bank's demonstrated market access and against the maturity wall it must refinance.

Why the funding plan is central to a credit opinion. A bank's failure mode is a liquidity failure, even when the underlying cause is solvency. An agency assessing default likelihood must therefore form a view on whether the bank can continue to fund itself under stress. A funding plan that depends on issuing more senior non-preferred in a single quarter than the bank has ever issued in a year is a plan the agency will discount, and the discount shows up in the rating. Conversely, a plan demonstrating diversification across instruments, currencies, investor types and maturities, with a track record of execution across different market conditions, supports the rating directly.

The confidentiality dimension is legally significant. Agencies routinely receive material non-public information — draft results, planned transactions, supervisory findings, internal stress results — under confidentiality arrangements, and the EU Market Abuse Regulation and the CRA Regulation both bear on the handling of that information. Selective disclosure is permitted in this context precisely because it is subject to confidentiality obligations and because the agency's function requires it. What follows practically is that agency meetings are prepared with the same discipline as regulatory interactions: controlled materials, a defined attendee list, a record of what was disclosed, and consistency with what has been said to supervisors and to the market. An inconsistency discovered by an agency is expensive.

Sustainable funding moved from marketing to plumbing

Two instrument families exist and they are routinely confused. The confusion matters because they impose entirely different obligations, carry different risks and suit different issuers. Getting the distinction right is the entry condition for any serious conversation about sustainable funding.

Use-of-proceeds instruments tie the money to specific assets. A green bond, a social bond or a sustainability bond raises funds that the issuer commits to allocate to a defined pool of eligible assets or projects — renewable energy generation, energy-efficient buildings, clean transport, affordable housing, access to essential services. The bond's financial terms are ordinary: same coupon structure, same seniority, same credit risk as the issuer's conventional debt of equal rank. What differs is the commitment about where the proceeds go and the reporting obligation that follows. The commitment is typically contractual only in a limited sense, which is why the reporting and verification architecture matters so much.

Sustainability-linked instruments tie the coupon to the issuer's own performance. No restriction on use of proceeds at all — the money funds general corporate purposes. Instead, the issuer defines one or more key performance indicators with sustainability performance targets at specified observation dates, and the coupon steps up if a target is missed. A bank might commit to reducing financed emissions in a specified portfolio by a stated percentage by a stated year, with a 25 basis point coupon step-up on failure, illustrative. The structure suits issuers whose transition is about changing what they do rather than about having a stock of identifiable green assets. It has attracted scrutiny over target ambition and calibration, and it is used less by banks than by corporates.

A green bond framework is the document that makes use-of-proceeds issuance credible. Published by the issuer before its first labelled transaction and updated periodically, it sets out the eligible asset categories with definitions and technical criteria, the process for evaluating and selecting projects, the management of proceeds including how unallocated funds are held, and the reporting commitment. Most frameworks align with the ICMA Green Bond Principles, a voluntary market standard structured around exactly these four components. The framework governs all issuance under it, which is why drafting it too narrowly constrains future capacity and drafting it too loosely invites challenge.

The second-party opinion is the external check. An independent provider — Sustainalytics, ISS ESG, Moody's and S&P's ESG arms, and others — reviews the framework against market standards and any relevant taxonomy and publishes an opinion on its credibility and alignment. It is not an audit of allocation, which comes later, and it is not a rating. Investors treat its absence as disqualifying; a labelled bond without an SPO is not a serious instrument in the European market.

The EU Green Bond Standard raises the bar for those who opt into it. Regulation (EU) 2023/2631, applicable from December 2024, creates a voluntary "European Green Bond" or "EuGB" designation. Its central requirement is that proceeds be allocated to activities aligned with the EU Taxonomy, subject to a flexibility pocket of up to 15% for activities where taxonomy criteria do not yet exist. It mandates a standardised factsheet before issuance, allocation reports, and at least one impact report, with pre- and post-issuance review by an external reviewer registered with and supervised by ESMA. The regime is voluntary — an issuer may continue with an ICMA-aligned framework — but the designation is protected, and only bonds meeting the regulation may use it.

The EU Taxonomy is what "green" means in this system. Regulation (EU) 2020/852 establishes a classification of environmentally sustainable economic activities across six environmental objectives, with activity-level technical screening criteria, a requirement to do no significant harm to the other objectives, and minimum social safeguards. It converts a contested adjective into a testable specification. Its practical difficulty for banks is that assessing taxonomy alignment demands asset-level data that lending systems were not built to hold — a point returned to below.

Now the greenium, honestly. The green premium is the pricing advantage a labelled bond enjoys over the issuer's equivalent conventional debt: the new issue prices a few basis points tighter, or trades tighter in secondary. It exists in the data, but it is small, inconsistent across issuers and currencies, and it has compressed as supply has grown. A few basis points on a €500m five-year transaction is a few hundred thousand euros a year — real, but not transformative, and easily swamped by execution timing.

The more reliable benefits are structural. A labelled transaction typically draws a broader and differently composed order book, because a segment of investors operates under mandates that can only buy labelled paper, and another segment has allocation targets that make them price-insensitive buyers of the label. That diversification is valuable in itself and it is worth most when it is scarcest — in a difficult market, an issuer with an established labelled programme and a known framework has an additional demand pool available that a first-time issuer does not. The strategic case for sustainable funding rests on durable access, not on the greenium.

ESG issuance is an asset-side data problem, not a Treasury marketing problem

The binding constraint is the eligible asset pool, and it sits outside Treasury entirely. Meridian cannot issue €3bn of green bonds because Treasury decides to. It can issue green bonds only up to the size of its identified, documented pool of eligible assets, less what previous labelled issuance has already allocated, with an appropriate margin so the pool does not fall below outstanding labelled debt as loans amortise or repay. If Meridian's identified pool of taxonomy-aligned energy-efficient mortgages and renewable project finance is €4.2bn and €2.8bn is already allocated to outstanding green bonds, the maximum new issuance is €1.4bn regardless of what investors would buy.

Identification is the hard part and it happens at loan level. A green mortgage pool is not a category in a general ledger. It is a set of individual loans, each of which must be shown to meet the framework's criteria — for buildings, typically an energy performance certificate of a specified class, or evidence of primary energy demand below a threshold, or documented renovation achieving a specified improvement. Each loan needs the attribute recorded, sourced, dated and auditable. Multiply by hundreds of thousands of loans and the scale of the problem becomes clear.

And the obligations continue for the life of every bond. Allocation reporting, normally annual, states how much of each bond's proceeds is allocated, to which categories, with what remains unallocated and how it is held. Impact reporting states the environmental outcome — tonnes of CO2 equivalent avoided, megawatt-hours of renewable capacity financed, square metres of building stock improved — with the methodology disclosed. Under the EU Green Bond Standard these are prescribed in form and externally reviewed. Under ICMA-aligned frameworks they are market expectations, but expectations investors enforce by declining to participate again.

Separately, prudential disclosure has made the same data mandatory. Pillar 3 ESG disclosures under CRR, with implementing technical standards developed by the EBA, require large institutions to publish quantitative and qualitative information on climate-related and environmental risks. Among these is the green asset ratio, the share of a bank's covered assets that finances taxonomy-aligned activities. The GAR is a disclosure about the entire banking book, not about labelled issuance. A bank could decide never to issue a green bond and would still need loan-level taxonomy attributes to compute it. The data problem is unavoidable; labelled issuance simply gives the same data a revenue use.

The origination systems were never designed for this, and that is the root of the difficulty.

A mortgage system built in the 1990s or 2000s captures what the credit decision needed: borrower identity and income, property valuation and address, loan amount, rate, term, and the collateral charge. It has no field for an energy performance certificate rating, no place for the certificate's issue date or expiry, no link to the document itself, and no way to record that the property was renovated in 2021 to a standard that changed its classification. A corporate lending system records purpose in a free-text field for credit assessment, not in a structured taxonomy-mappable code with evidence attached.

The consequences follow predictably. Banks reconstruct eligibility retrospectively through data purchases, national EPC register matching and proxy estimation, which is expensive, incomplete and weakly auditable. Coverage gaps force conservative treatment, so a loan that is probably eligible but cannot be evidenced is excluded — shrinking the pool and depressing the green asset ratio for reasons that have nothing to do with the underlying assets. And retrospective reconstruction does not scale to the ongoing obligation, because every new loan reopens the same gap.

The durable fix is at origination. Capture the attribute when the loan is written, with the evidence document linked and the source and date recorded. Make the taxonomy assessment a structured field with a defined vocabulary rather than free text. Version it, because criteria change and a loan assessed under one vintage of technical screening criteria must remain traceable to that vintage. Then eligible-pool sizing, allocation reporting, impact reporting and the green asset ratio all draw on one governed dataset rather than four reconstructions that will not reconcile to each other — and the reconciliation failure between them is exactly what an external reviewer finds first.

  1. Why is a liquidity buffer held at amortised cost not fully liquid?

  2. What is the difference between a use-of-proceeds instrument and a sustainability-linked one?

  3. Why can a rating downgrade cause an immediate cash outflow?

  4. Two banks hold the identical bond and rates rise. One reports nothing, one reports a capital hit. Explain.

  5. Why does hedge accounting exist, and what does a fair value hedge actually change?

  6. Why is the size of the eligible asset pool, not investor demand, the constraint on green bond issuance?

2Track A: The funding cycle turns a year of planning into six hours of execution

Treasury runs three processes in parallel, and they are the substance of the job. Track A raises the money. Track B prices it internally. Track C makes sure it is in the right account before the payment system closes. This part walks all three end to end, starting with the one that looks most like a capital markets transaction and ending with the one that has no equivalent anywhere else in the bank.

Figure 5 The three tracks of Part 2, and the data spine they share

The funding cycle has seven stages and a single point of no return

The annual funding plan is a budget, and everything downstream is its execution. Once a year, Treasury projects how the balance sheet will grow, what matures and must be replaced, what the regulatory ratios require, and what the resolution authority demands in loss-absorbing debt. The output is a number, a currency mix, an instrument mix and a rough calendar. The rest of the year is spent executing that plan into markets that will not always cooperate, which is why the plan is a direction rather than a schedule.

Stage Typical duration What it produces Who leads
1. Funding plan and budget Annual, reviewed quarterly Volume, currency and instrument mix, indicative calendar Treasury with ALM, approved by ALCO and board
2. Instrument selection and structuring 4–8 weeks per instrument The specific trade: format, tenor, entity, currency Funding desk with legal and Group Finance
3. Documentation and approvals 6–12 weeks, then annual updates EMTN or covered bond programme, base prospectus, ratings Legal, Treasury, external counsel, agencies
4. Market window monitoring Continuous The decision to go, or to wait Funding desk with syndicate banks
5. Execution and pricing One day, roughly six hours Announcement, IPT, guidance, book, final spread and size Funding desk with the syndicate
6. Settlement and booking T+5 for a new euro benchmark Cash received, bond issued, position booked and hedged Treasury operations, back office, ALM
7. Post-issuance management To maturity or first call Coupon payments, MREL eligibility tracking, buybacks, calls Treasury with Group Finance
Figure 6 The funding cycle: months of preparation resolve into one afternoon

The effort curve is almost flat and then vertical. Stages one through four consume most of the calendar and very little adrenaline: they are planning, drafting, negotiating and watching. Stage five compresses the entire exercise into a single session in which a spread is discovered, a book is built and a price is fixed, and during which the cost of a mistake is measured in basis points multiplied by a billion euros multiplied by five years. Stage six is mechanical but unforgiving, because a settlement failure on a benchmark deal is public. Stage seven is quiet and long, and it is where most of the operational risk actually accumulates, precisely because nobody is watching.

The point of no return is the moment the deal is launched. Before launch, the bank can withdraw at any time at the cost of some embarrassment. After launch — when the final spread and size are announced and orders are confirmed — the bank has entered into an obligation to a few hundred investors simultaneously, and pulling the trade would do more damage to its future funding cost than any single deal is worth. Everything in stages one to four exists so that the decision at that moment is a formality rather than a gamble.

Step 1 — The funding plan is a constrained optimisation, not a wish list

Four inputs determine how much a bank must raise, and only one of them is discretionary. The first is the maturity profile: whatever matures this year must be replaced unless the balance sheet is shrinking. The second is asset growth: new lending has to be funded. The third is the ratio requirement: if the NSFR or the LCR is close to its internal floor, the plan must include instruments that improve them, and those are rarely the cheapest ones. The fourth is MREL: the resolution requirement sets a floor on eligible debt outstanding, and any bond dropping inside one year to maturity stops counting.

The optimisation runs across four dimensions simultaneously. Volume answers how much. Instrument answers what kind, and therefore what it costs and which ratio it helps. Currency answers where demand is deepest and what the cross-currency basis costs to swap back. Tenor answers how long, trading a higher coupon against a longer period of certainty and a better NSFR treatment. A plan that optimises any one dimension in isolation will produce a bad answer in the others, which is why the funding plan is built and challenged at ALCO rather than assembled on a desk.

The funding plan is a modelling exercise before it is a decision, and it needs three data feeds that many banks struggle to produce cleanly. The first is a complete, entity-level maturity ladder of every wholesale liability, which is harder than it sounds in a group with dozens of issuing entities and several legacy programmes. The second is a forward projection of the balance sheet by product, which comes from the planning process rather than from Treasury and arrives in a different granularity than the ratio engines need. The third is a forward projection of the ratios themselves under the planned issuance, which requires running the LCR and NSFR engines on hypothetical future positions rather than actual current ones. The last of these is the one most often missing: many banks can calculate today's ratio precisely and next year's only approximately, which makes the plan less a projection than an educated guess.

Figure 7 The maturity profile the plan is built around

Step 2 — Instrument selection is where funding strategy meets the ratio engine

Choosing the instrument is choosing which constraint to relieve. A covered bond is cheap and improves the funding profile but encumbers mortgages and does nothing for MREL. Senior non-preferred is expensive and satisfies MREL but consumes the same investor appetite that senior preferred would have used. A three-year senior bond is cheaper than a seven-year one and gives back the refinancing problem three years sooner. Certificates of deposit are cheapest of all and improve nothing, because funding under six months receives no credit in the NSFR and is assumed to run entirely in a stress.

The entity question is as consequential as the instrument question and is frequently underestimated. A banking group has a resolution entity, operating subsidiaries, and sometimes intermediate holding companies, each with its own creditors, ratings and local supervisor. MREL-eligible debt generally has to be issued by the resolution entity and then down-streamed internally to the subsidiaries that need it, in a form that will convert to equity if the group is resolved. Getting the entity wrong does not merely make a bond suboptimal, it makes it ineligible, and the error is not always visible until a regulatory return is produced months later.

The tenor decision, and why banks do not simply issue as long as possible

Longer funding is more stable and more expensive, and the trade-off is not linear. The term liquidity premium — the extra spread a bank pays purely for the length of the commitment — rises steeply from overnight to about three years and then flattens. Beyond seven or eight years, each additional year of certainty costs relatively little, but the investor base thins to insurers and pension funds whose appetite is finite and whose demand is lumpy. The practical result is that most bank senior funding clusters at three, five and seven years, not because those tenors are optimal in any deep sense, but because that is where the depth of demand and the shape of the premium intersect.

Step 3 — The programme is the real product, and the bond is a drawdown from it

Banks do not document each bond individually; they maintain standing programmes. A Euro Medium Term Note programme, or EMTN programme, is a master framework of legal documentation, updated annually, under which the issuer can bring individual bonds to market in days rather than months. The base prospectus describes the issuer, the risk factors and the possible note structures once. Each individual issue is then documented by a short set of final terms specifying size, coupon, maturity and a handful of other variables. Covered bonds run on an equivalent programme with an additional layer of cover-pool documentation and a designated cover pool monitor.

This is why a bank can issue a €1bn bond on eight hours' notice and a corporate cannot. The corporate needs a prospectus, an auditor comfort letter, a due diligence process and a roadshow. The bank has all of that standing and current, refreshed each year alongside its annual report, so the marginal cost of an individual transaction is small. That standing readiness is itself a Treasury asset. A programme that has lapsed, or whose base prospectus has gone stale, removes the bank's ability to act inside a narrow market window, and market windows do not wait for documentation.

The internal gauntlet before the bank commits

A benchmark issue passes through a defined sequence of internal approvals, and each gate exists because of a specific historical failure. The funding plan itself is approved by ALCO and, in most institutions, noted or approved by the board risk committee as part of the risk appetite framework. The individual transaction is approved against that plan by a treasury or issuance committee, which checks that the trade is within plan, within limits, in an approved format from an approved entity, and that the pricing is consistent with the mandate. Legal and compliance confirm the disclosure position, since the issuer is a listed entity and its funding plan is price-sensitive information. Finance confirms the accounting classification and the capital or MREL treatment, which for hybrid instruments is not always obvious. Only then does the desk have authority to announce.

The approval chain: delegation flows down, escalation flows back up

Step 4 — The market window is a decision to act, and waiting is also a decision

Bank issuers do not issue continuously; they issue into windows. A window is a period when the market is receptive: no major central bank meeting imminent, no significant economic data due that morning, no crowded issuance calendar competing for the same investors, no idiosyncratic news about the issuer, and secondary spreads stable rather than gapping. European bank issuance clusters heavily in January, after results announcements, and in early September, and it thins to nothing in August, over year-end, and during blackout periods before results.

The syndicate relationship is the information channel that makes window judgement possible. A bank appoints several other banks as bookrunners for a transaction, and those syndicate desks are in continuous conversation with investors about what they would buy and at what level. The issuer therefore receives, informally and continuously, a picture of demand and of the level at which a deal would clear. This is soft information — it is not a commitment and investors are not bound by it — but a funding desk that has stopped listening to it is flying on instruments alone.

Reverse enquiry and private placements: the deals nobody announces

A significant share of a bank's term funding never goes through a public book at all. An investor with a specific need — a particular maturity, a particular structure, a particular currency — can approach the bank through a dealer and ask for it. If the level is attractive relative to the bank's public curve, the bank prints a private placement under the same EMTN programme, often within a day, with no announcement and no book. These trades are typically smaller, they are always more bespoke, and cumulatively they can fund a meaningful part of the annual plan at levels inside the public curve. They also carry an operational tail: every bespoke structure has to be modelled, hedged, booked and reported, and a portfolio of two hundred idiosyncratic private placements is materially harder to manage than four benchmark bonds.

Step 5 — Pricing is a six-hour discovery process with a single output

The bank does not set the price of its own debt; it discovers it. The desk begins with an estimate of where its paper should trade — its fair value — and then runs a controlled process designed to find out whether the market agrees, and how much size the market will absorb at that level. The mechanism is a descending price auction conducted through the syndicate, and it works in three moves.

Fair value is triangulated from three independent readings. The first is the issuer's own secondary curve: where its existing bonds of similar maturity are actually trading. The second is comparables: where similar banks with similar ratings and similar structures have recently issued and where their paper trades now. The third is the derivative and capital-structure reading: the credit default swap level, and the observed basis between senior and subordinated instruments. When the three readings agree, the desk has a confident number. When they diverge, the divergence itself is information about which market is stale.

Figure 8 Triangulating fair value before the market is asked

The process then moves from initial price thoughts to guidance to launch. Initial price thoughts, or IPT, are announced deliberately wide of fair value — often 25 to 30 basis points wide for a euro senior benchmark — for a specific reason: an investor who places an order at IPT is committing to a level, and a book that builds quickly at a wide level gives the issuer the leverage to tighten. Guidance is released once the book is large enough to justify it, typically a few hours later, at a level closer to fair value, and orders that are unwilling to follow drop out. Launch fixes the final spread and the deal size. Pricing confirms the coupon.

Reading a book properly means reading its composition, not its size

A three-times-covered book is not automatically a good book. Order size can be inflated by investors who inflate deliberately, expecting to be scaled back, and by fast money that intends to sell the bond within days. What the desk actually wants to know is how much of the book is real, long-term demand that will hold the paper and buy the next transaction. That means looking at the mix of investor types, the number of orders with price limits — a limit is evidence of genuine price discipline rather than an option to participate — and the concentration of the book in a handful of large tickets.

Bank funding books are read differently from corporate bond books for one specific reason: the issuer will be back. A corporate may issue once every few years. A bank of any size issues repeatedly, every year, in several instruments. Its relationship with the investor base is a repeated game, and squeezing the last two basis points out of a transaction at the cost of leaving investors with a bond that trades badly is a poor trade when the next transaction is eight weeks away. This is the practical reason banks tolerate a new issue premium at all: it is the price of a good aftermarket, and a good aftermarket is the price of the next deal.

Allocation for a bank issuer answers a different question than for an IPO

Allocation decides who receives how much of an oversubscribed book, and the bank's own interest is unambiguous here in a way it is not in equity capital markets. In an IPO, the underwriting bank allocates its client's shares and faces a genuine conflict between the issuer's interest and its investors'. When a bank issues its own debt, the issuer and the allocator are the same institution, so the conflict disappears and the objective is simple: build a register of investors that will hold the bond, support the next transaction, and not immediately sell.

Three practical rules follow. Investors who placed limited orders and stayed through the tightening receive proportionally better allocations than those who dropped out. Long-term holders — insurers, pension funds, asset managers running buy-and-hold mandates — are favoured over accounts known to sell into the first day's strength. And investors who bought the bank's paper in a difficult market are remembered when a book is oversubscribed in an easy one. None of this is written down as policy, and all of it is tracked in the investor relationship database.

Step 6 — Settlement is mechanical, public, and the one place where operational failure is visible

A euro benchmark bond typically settles five business days after pricing. Between pricing and settlement the legal documentation is completed, the final terms are signed, the ISIN is issued, the bond is admitted to listing, and the securities are created in the international central securities depositories — Euroclear and Clearstream for euro international issues. On the settlement date the bond is delivered against payment, meaning the securities move to investors at the same moment the cash moves to the issuer, so neither side is ever exposed to the other's failure.

The payment leg and the securities leg run over different rails, and both have deadlines. Cash settlement of euro payments runs through the TARGET2 / T2 system operated by the Eurosystem, which has hard closing times. Securities settlement runs through the depositories and, for euro-denominated securities settled in central bank money, through TARGET2-Securities. The messages that carry these instructions — MT540 to MT543 for securities, MT202 or its ISO 20022 successor for the cash leg — are covered in detail in Part 4.

The settlement window on a new issue is short and the moving parts are external, which makes it one of the few Treasury processes where straight-through processing genuinely pays for itself. Manual re-keying of a final-terms document into a securities master file is how an incorrect coupon date reaches a bond that will pay coupons for ten years, and a wrong maturity date in the securities master flows straight into the maturity ladder, the NSFR calculation and the MREL eligibility test. The correction is trivial on day one, embarrassing at the first coupon date, and a regulatory finding when a supervisor asks why the reported maturity profile does not tie to the terms of the issued instruments.

Booking the trade is not the end; hedging it is. A fixed-rate bond issued by a bank that manages its banking book on a floating-rate basis creates an interest rate exposure the moment it settles. The standard response is an interest rate swap that converts the fixed coupon into a floating obligation, executed at or near pricing, together with a cross-currency swap if the bond was issued in a currency other than the bank's funding currency. That hedge is itself a transaction with a counterparty, a margin requirement and a lifecycle, and it must be linked to the bond in the systems so that the two are reported, valued and unwound together.

Step 7 — The bond's life after issuance is where the quiet risk accumulates

A bond that has settled is not finished; it has entered a lifecycle with dated obligations. Coupons must be calculated and paid on schedule through the paying agent. The instrument's regulatory status must be tracked continuously — a senior non-preferred bond counts towards MREL until its remaining maturity falls below one year, at which point it stops, and the funding plan must anticipate that date rather than discover it. Instruments with call dates require a decision, and that decision has a market convention attached to it that is not contractual.

Calls, and the expectation that is not written in the contract

Callable instruments create an obligation the documents do not contain. A Tier 2 bond issued with a ten-year maturity and a call at year five is priced by investors on the assumption that it will be called at year five, even though the issuer has no obligation to do so. An issuer that declines to call — "extending" the instrument — will save money on that specific bond and pay for it repeatedly, because every subsequent callable instrument it issues will be priced to maturity rather than to call. The economics of a single call decision are almost never worth the repricing of the entire future callable curve, which is why non-calls are rare and are read by the market as a signal of distress rather than of arithmetic.

Liability management: buying back what you issued

A bank can repurchase its own debt, and does so for three distinct reasons. The first is profile management: tendering for bonds maturing in a crowded year to smooth a maturity wall. The second is regulatory efficiency: buying back instruments that have stopped counting towards MREL or capital and replacing them with instruments that do, since paying a coupon on debt that earns no regulatory credit is pure cost. The third is opportunistic: when a bank's bonds trade well below par, repurchasing them at a discount generates an accounting gain — although supervisors scrutinise capital instrument buybacks closely and require prior permission, precisely because a bank under stress has an incentive to book gains by retiring the loss-absorbing debt it may shortly need.

A liability management exercise is a transaction in its own right, with its own approvals and its own execution risk. The bank announces a tender at a stated price or spread, investors elect whether to participate, and the results determine how much is retired. Participation is uncertain, and a tender that attracts almost no interest is a public non-event that tells the market something about where the bank's paper is really valued.

  1. Why does a bank maintain a standing EMTN programme rather than documenting each bond?

  2. What are the four inputs that determine the size of the annual funding plan, and which is discretionary?

  3. Why is a three-times-covered order book not automatically a good book?

  4. Why do banks rarely decline to call a callable instrument?

  5. What makes stage seven, the post-issuance lifecycle, the place where errors survive longest?

2Track B: Funds transfer pricing is where the bank's strategy is actually priced

Every loan a bank makes is funded by money it does not own, and every deposit it takes is money it must do something with. Funds transfer pricing is the internal mechanism that puts a price on both, and it is the single most powerful lever a bank has over what its business lines choose to do. Whatever FTP under-charges, the bank will over-produce.

Without an internal price, a bank cannot tell profit from mismatch

Consider a bank with no transfer pricing at all. The mortgage desk writes a twenty-year fixed-rate loan at 4.9%. The retail network holds current accounts paying 0.2%. The bank's management accounts show a mortgage business earning 4.9%, a deposit business earning nothing, and a comfortable-looking spread of 4.7% somewhere in the middle. Three things are wrong with this picture, and each of them has bankrupted institutions.

The first error is that the mortgage business is credited with a return it did not generate. Most of that 4.7% is not lending skill. It is compensation for taking short money and lending it long — the maturity mismatch — and the risk of that mismatch sits with the institution as a whole, not with the desk that wrote the loan. A business unit that keeps the reward for a risk it does not manage will produce more of that risk than the institution wants.

The second error is that the deposit business appears worthless. A stable, low-cost current account balance is one of the most valuable things a bank owns, because it is what makes long-dated lending possible at all. A management reporting system that shows the deposit franchise contributing nothing will underinvest in it, close branches that gather it, and reward the lending desks that consume it. Several banks discovered the cost of this when rates rose and the cheap deposits they had stopped valuing became expensive.

The third error is that nobody prices liquidity. A twenty-year loan and a two-year loan at the same margin look equally attractive without transfer pricing, even though the first consumes twenty years of the bank's scarce long-term funding capacity and the second consumes two. Fixing this is the entire purpose of the mechanism.

The mechanism: Treasury becomes an internal bank with two prices

Funds transfer pricing works by inserting Treasury as the counterparty to every business unit transaction. A desk that takes a deposit sells the money to Treasury and receives an FTP credit. A desk that makes a loan buys the money from Treasury and pays an FTP charge. The desks keep the difference between their client rate and the internal rate. Treasury keeps the spread between the two internal rates, and with it the mismatch risk that spread compensates.

Figure 9 FTP splits one client spread into three accountable pieces

The arithmetic is straightforward and the consequences are not. In the illustration above, a client spread of 3.10% between a deposit paying 1.80% and a loan charging 4.90% is decomposed into a deposit margin of 0.75%, a lending margin of 1.23%, and a term and liquidity spread of 1.12% retained by Treasury. Each of the three numbers is now attributable to a decision made by an identifiable owner. The deposit desk is rewarded for gathering cheap stable money. The lending desk is rewarded for pricing credit well. Treasury is rewarded, or penalised, for managing the mismatch — which is the risk it actually controls.

Matched maturity is the founding principle, and most FTP errors are departures from it

The transfer rate for any contract is the rate the bank would pay to raise funding with exactly that contract's cashflow profile. A five-year bullet loan is charged the five-year rate. A twenty-five-year amortising mortgage is charged a blend across the tenors at which its principal actually repays — its weighted-average life, not its final maturity. An overnight deposit receives an overnight credit unless behavioural analysis justifies otherwise. This is called matched-maturity funds transfer pricing, and it is the only method that produces the right incentives, because it charges each contract for the funding capacity it genuinely consumes.

The alternative, still found in older frameworks, is a single pooled rate. Under pooled FTP every asset is charged the same rate and every liability credited the same rate, usually an average cost of funds. The result is systematic mispricing in a predictable direction: short-dated business is over-charged and cross-subsidises long-dated business, which is under-charged. A bank running pooled FTP will therefore accumulate long-dated, fixed-rate, low-margin assets, and will do so believing them to be profitable. The pattern is visible in more than one post-mortem.

Building the curve: four components, three of which are contested

The FTP curve starts from a market base curve and adds the specific costs of being this bank. The base is the risk-free or interbank curve, in the euro area typically the euro short-term rate compounded or the mid-swap curve, which represents the time value of money with no credit or liquidity content. Everything added to it represents a cost the bank actually incurs.

Figure 10 The FTP curve: base rate plus the price of term

The term liquidity premium is the largest and most consequential addition. It is the spread the bank pays, above the base curve, to borrow at each tenor — in practice, its own senior unsecured secondary curve, adjusted for the fact that the internal curve should reflect the marginal cost of funding rather than the average of historical issuance. A bank whose five-year paper trades at mid-swaps plus 62 charges its five-year lending 62 basis points above the five-year base rate. The premium rises sharply from overnight to about three years and flattens beyond seven, which is why the incremental cost of lending long is much lower between year seven and year ten than between year one and year three.

A contingent liquidity charge covers commitments that are not yet drawn. An undrawn revolving credit facility costs the bank nothing today and may cost it everything on the day the client draws it, which will be the day the client is in trouble and the market is closed. Because the bank must hold liquid assets against that possibility under the LCR, the facility consumes buffer capacity from the moment it is granted. The charge makes that cost visible to the desk that granted it, and a bank without one will grant undrawn facilities almost free.

Additional components cover basis, optionality and, in some frameworks, capital. Basis risk arises when a loan reprices off one index and the funding off another. Optionality charges price embedded rights the client holds, such as the right to prepay a fixed-rate mortgage without penalty, which is an option the bank has sold and must hedge. Some banks also allocate a capital charge through the FTP framework rather than separately, so that the transfer rate delivered to a business line is genuinely all-in.

Component What it compensates Typical driver Contested?
Base curve Time value of money Mid-swaps or overnight index curve No
Term liquidity premium Cost of borrowing at that tenor The bank's own secondary senior curve Marginal vs average cost
Contingent liquidity charge Buffer held against undrawn commitments LCR outflow weight on the facility type Yes, heavily
Basis and optionality Index mismatch, prepayment and cap/floor options Hedge cost of the embedded option Yes
Behavioural adjustment Difference between contractual and expected life Replicating portfolio or decay model Yes, most of all

Non-maturing deposits are where FTP stops being arithmetic and starts being judgement

A current account has no maturity, so matched-maturity pricing has nothing to match. The balance is repayable on demand, and yet in aggregate it behaves like a long-dated liability: the total sits stable across years, individual accounts churn, and the aggregate barely moves. Crediting the deposit desk an overnight rate would be absurd, because the money is not behaving overnight. Crediting it a ten-year rate would be reckless, because it contractually could be.

The standard answer is a replicating portfolio. The bank decomposes the deposit balance into a stable core and a volatile non-core portion, then assigns the core a portfolio of notional maturities — for example, a share rolling overnight, a share at one year, a share at three years, a share at five — chosen so that the resulting blended rate is as stable through rate cycles as the deposit rate the bank actually pays. The deposit desk is credited the blended rate of that portfolio. European guidelines cap the assumed average repricing maturity of retail non-maturing deposits, precisely because the temptation to assume a longer life than experience supports is strong and its effect on reported earnings sensitivity is large.

Two assumptions inside this are the most consequential numbers in the bank. The first is the core-versus-volatile split, which determines how much of the deposit base is treated as long-term funding. The second is deposit beta, the share of a policy rate increase that the bank expects to pass through to depositors, which determines both the projected earnings and the assumed stability. Both are calibrated on history, both were calibrated on a history containing no fast digital run, and both are simultaneously a modelling exercise, a commercial forecast and, in effect, a strategic bet.

Figure 11 Deposit beta is not a constant, and the gap is unbudgeted expense

The FTP cycle: continuous application, periodic argument

FTP runs on two clocks. The transfer rate is applied at origination to every contract the bank writes, and for fixed-rate contracts it is generally locked for the life of the contract — the rate agreed on the day the loan was written stays with that loan, which is what allows the desk's margin to be measured cleanly over time. Separately, the curve itself is recalibrated on a defined cycle: daily or weekly for the base curve, monthly or quarterly for the liquidity premium, and annually for the policy framework and the behavioural assumptions.

Step Cadence What happens Owner
Curve construction Daily to monthly Base curve refreshed from market data; liquidity premium from own secondary levels and recent issuance Treasury / ALM
Policy approval Annual, with ad hoc changes Methodology, components, behavioural assumptions, exception rules ALCO
Application Every contract, at origination Transfer rate attached and locked to the contract record FTP engine, automatically
Attribution Monthly Business line P&L produced with FTP charges and credits Treasury Finance
Dispute and exception Continuous Challenges from business lines, escalation to ALCO Treasury Finance, then ALCO
Recalibration and validation Annual Behavioural models re-estimated and independently validated ALM with model validation

The lag between market reality and the applied curve is a real exposure, not an administrative detail. If the liquidity premium is refreshed quarterly and the bank's spreads widen 40 basis points in February, then every loan written between February and the next recalibration is charged a funding cost the bank is no longer paying, and the difference accumulates in Treasury as a loss. Through 2022 and 2023, when spreads moved quickly and repeatedly, banks with slow recalibration cycles wrote substantial volumes of business at transfer rates that had stopped being true. The fix is not complicated — it is a faster refresh — but it requires the curve to be produced by a system rather than assembled in a spreadsheet.

The FTP engine has an unusual technical profile: the computation is trivial and the data problem is severe. Attaching a rate to a contract is a lookup. Doing it correctly for every contract in a bank requires that every contract carry a clean product code, a correct origination date, a correct rate type, an accurate amortisation profile and a valid behavioural classification — the same fields the liquidity engine depends on, sourced from the same fragmented origination systems. This is why banks buy the calculation engine and build the policy layer around it: the vendor solves a problem that was never hard, while the bank solves the mapping of forty thousand product codes to a coherent set of behavioural assumptions, which is.

The second technical requirement is reproducibility. When a business line disputes its monthly attribution — and it will — Treasury Finance must be able to show which curve version, which policy version and which behavioural model version produced a specific rate on a specific contract on a specific date. An FTP engine without full version lineage cannot settle an argument, and an argument that cannot be settled becomes a governance problem.

Disputes are a feature of the system, and their volume is a health metric

Business lines argue about FTP because FTP determines their bonus. A lending desk charged 4.20% instead of 3.67% loses most of its margin on a transaction and will contest the classification, the tenor assumption, the behavioural treatment or the timing of the curve refresh. These disputes are legitimate and useful — they surface genuine errors, and a framework nobody challenges is probably one nobody understands.

What matters is that the dispute process is defined and that the outcome is not negotiable at the desk. A well-run framework has a written policy, a documented exception process, an escalation route to ALCO, and a rule that no transfer rate is adjusted bilaterally between Treasury and a business line. A poorly run framework produces a proliferation of negotiated exceptions, at which point the transfer rate stops being a price and becomes an allocation, and the incentive signal it was built to send is gone. Rising dispute volume with falling resolution consistency is one of the earlier observable signs that a bank's internal pricing discipline is deteriorating.

There is a genuine and unresolved argument about whether the term liquidity premium should reflect the bank's marginal cost of funding — what it would pay to issue today — or its average historical cost. Marginal pricing sends the economically correct signal, because the next loan is funded at today's price, not at the price of a bond issued four years ago. Average pricing produces less volatile business-line results and avoids the perverse outcome in which a bank whose spreads have widened suddenly finds most of its lending uneconomic and stops lending precisely when it is least helpful to do so.

Most frameworks land on marginal cost with smoothing, and defend the smoothing explicitly rather than pretending it is not there. The position to avoid is the unexamined one: an average-cost curve that nobody has revisited since it was built, quietly making long-dated lending look cheaper than it is.

  1. What happens inside a bank that runs no transfer pricing at all?

  2. Why does pooled FTP systematically distort the balance sheet?

  3. What is the contingent liquidity charge compensating for?

  4. Why is the replicating portfolio the point where FTP and IRRBB become the same problem?

  5. Why is a slow FTP recalibration cycle a real exposure rather than an administrative detail?

  6. What does rising dispute volume with falling resolution consistency indicate?

2Track C: The liquidity day ends at a deadline that does not move

The funding cycle happens a dozen times a year and the transfer pricing cycle happens monthly. The liquidity cycle happens every day, and it is the only Treasury process with a hard external deadline attached to it. Whatever else is unfinished at six o'clock, the bank's payments must have settled.

Everything starts with the cash ladder

A cash ladder is a projection of every expected inflow and outflow, by date, by currency, from today outwards. Today's column is the most detailed: known maturities, scheduled settlements, expected client payments, coupon dates, margin calls. Tomorrow's is slightly less certain. The thirty-day column is largely modelled. The one-year column is almost entirely behavioural. The ladder is the single artefact from which the daily position, the LCR, the survival horizon and the funding plan are all derived, and every one of those numbers is only as good as the ladder underneath it.

The ladder answers two different questions and the distinction matters enormously. The contractual ladder shows what the bank is legally owed and legally owes, and is a fact. The behavioural ladder shows what it expects to happen, applying assumptions about deposit stability, facility drawdown, prepayment and roll rates, and is a model. Regulators require both. Supervisory attention concentrates on the second, because that is where a bank's optimism can hide.

Figure 12 The liquidity day: seven checkpoints and one immovable deadline

Step 1 — The day begins with reconciliation, because yesterday's picture is never quite right

By seven in the morning the desk must know what actually happened overnight. Statements arrive from correspondent banks and from the central bank showing actual closing balances by account and currency. These are compared against what the bank's own systems expected. Differences are investigated: a payment that did not arrive, a settlement that failed, an amount that differs from the confirmation, a currency booked to the wrong account.

Breaks are not administrative noise; they are the first signal of every kind of problem. A payment that failed to arrive from a counterparty may be an operational error, or it may be the first sign that the counterparty has a problem. A settlement failure in a repo trade means collateral did not move, which means an assumed source of cash is not there. The reconciliation is where the day's real starting position is established, and a desk that begins trading from an unreconciled position is trading on an estimate.

Step 2 — The projection turns the ladder into a decision about today

With the opening position established, the desk projects the day. Known items are added: maturing interbank placements and borrowings, repo legs settling today, securities settlements from the trading businesses, scheduled corporate payments, central bank operations, coupon and redemption flows on the bank's own issuance and holdings. Expected items are estimated: client payment flows, drawdowns on committed facilities, card and retail settlement volumes.

The output is a projected end-of-day balance by currency, and it is either comfortable, short, or long. Short means the bank must raise cash today. Long means it has surplus cash to place, and placing it badly is a real cost when the alternative is a central bank deposit rate that may be well below what a secured placement would earn. Comfortable means the desk still needs to decide whether the buffer sits in the right currencies and the right accounts, because cash in the wrong entity or the wrong currency on a day when it is needed elsewhere is not liquidity.

Step 3 — Money market and repo execution closes the gap

The desk covers a shortfall or places a surplus through a small set of instruments with very different characteristics. Unsecured interbank borrowing is fastest and simplest and has become a much smaller market than it once was, because banks are reluctant to take unsecured exposure to each other and the regulatory treatment penalises it on both sides. Secured borrowing through repo is now the dominant channel: cheaper, larger, and available in a stress, at the cost of pledging assets. Certificates of deposit and commercial paper reach non-bank investors for slightly longer tenors. Central bank operations are available at published terms, and foreign exchange swaps move liquidity between currencies without altering the overall position.

Instrument Tenor Secured? Speed Main constraint
Unsecured interbank Overnight to 3 months No Immediate Counterparty limits, thin market
Repo (bilateral or triparty) Overnight to 1 year Yes Fast Available unencumbered collateral
Certificates of deposit / CP 1 week to 12 months No Same or next day Investor appetite, name concentration
FX swap Overnight to 1 year Effectively yes Immediate Cross-currency basis, counterparty limits
Central bank operations Fixed schedule Yes Scheduled Eligible collateral in the pool
Buffer monetisation (asset sale) Immediate n/a Fast for level 1 Realises accounting loss; signals stress

Repo and collateral: the plumbing that decides whether liquidity is real

A liquidity buffer is only worth what it can be converted into cash for, on the day it is needed. That conversion happens either by selling the asset or, far more commonly, by repo-ing it — pledging it as collateral for cash. The amount of cash received is the market value less a haircut, the lender's protection against the collateral falling in value before it can be sold. A German government bond might attract a haircut of half a percent; a lower-rated corporate bond might attract fifteen percent or more; an illiquid asset might attract no bid at all in the conditions where the bank needs one.

Haircuts widen in exactly the conditions where the buffer is being used, and this is the mechanism that makes liquidity crises non-linear. A bank that models its buffer using normal-market haircuts will overestimate how much cash it can raise in a stress, and the shortfall arrives at the worst possible moment. Central bank collateral frameworks apply their own haircut schedules, generally more stable than market haircuts, which is one of the reasons pre-positioning collateral with the central bank is a core part of contingency planning rather than an optional extra.

Triparty repo, collateral optimisation, and the cheapest-to-deliver problem

Triparty repo introduces an agent that manages collateral on behalf of both sides. Rather than each party selecting and delivering specific securities, the cash lender specifies eligibility criteria — ratings, asset classes, concentration limits — and a triparty agent such as Euroclear or Clearstream allocates eligible securities from the borrower's account, marks them daily, and substitutes them when needed. The operational saving is large, and it is what makes large-scale secured funding practical.

It also creates an optimisation problem that is genuinely hard. At any moment a bank holds thousands of securities with different eligibility across different destinations: the LCR buffer, several central bank pools, triparty baskets, bilateral repo counterparties, and cleared derivative margin. Each destination has its own eligibility rules and haircuts. The objective is to deliver the cheapest acceptable asset to each destination while keeping the most valuable assets — the level 1 government bonds that count fully in the buffer — unencumbered. Doing this well across a group is worth a meaningful amount of money; doing it badly means pledging a high-quality bond where a lower-quality one would have been accepted, and discovering the shortage later.

Step 4 — Intraday liquidity is a separate discipline, and the LCR does not measure it

A bank can be comfortably liquid over thirty days and fail to make a payment at eleven in the morning. The LCR is a stock measure over a horizon. Intraday liquidity is a flow problem within a single day: payments leave and arrive in an order the bank does not fully control, and at any moment the cumulative net position must remain above zero, or above whatever collateralised intraday credit the central bank has extended. A large payment that was expected at ten and arrives at three does not change the day's closing balance at all, and can still cause a failure.

The Basel monitoring framework for intraday liquidity, BCBS 248, made this measurable. It requires banks to report daily maximum intraday liquidity usage, available intraday liquidity at the start of the day, total payments made and received, and the timing of time-specific obligations. The reporting requirement was the easy part. The difficult part was that most banks' cash management systems recorded end-of-day balances and daily aggregate flows, not time-stamped individual payment events, and producing the metrics required instrumenting the payment infrastructure to a granularity it had never needed.

Real intraday liquidity management depends on one specific capability: receiving credit and debit notifications as they happen rather than in an end-of-day statement. In the ISO 20022 world this is the camt.052 intraday report and the camt.054 debit or credit notification, replacing the once-a-day MT940 statement. With time-stamped intraday messages the desk can see a live position and act on it. Without them it can only reconstruct the day afterwards, and every intraday decision is made against an estimate.

The second capability is a payment throttle: the ability to hold, prioritise and release outgoing payments according to a defined policy rather than sending everything as it arrives. Banks that can sequence their outgoing payments consume materially less intraday liquidity than banks that cannot, because the bank effectively self-funds from incoming flows. This is a queue management problem in the payment hub, and it is one of the cleanest examples in banking of a technical capability producing a direct, measurable balance sheet saving.

Figure 13 The intraday profile: flat at the close, deeply negative in the middle

Step 5 — The afternoon is about deadlines, and deadlines are absolute

As the day progresses the flexibility shrinks. Money market counterparties stop quoting for same-day value at a certain hour. The central bank's operations have fixed timings. Foreign exchange for same-day settlement becomes progressively harder and then impossible as time zones close. The euro payment system operates to a published schedule, with customer payments cutting off before interbank payments, and interbank payments before the final closing of the day. A shortfall identified at four in the afternoon has a much smaller set of solutions than the same shortfall identified at nine in the morning.

This is why the discipline is front-loaded. The desk aims to be flat, or comfortably long, well before the afternoon, and treats late-day activity as exception management rather than routine. The escalation path — who is called, at what threshold, and who has authority to draw on the central bank facility — is written down and rehearsed, because the moment it is needed is precisely the moment nobody has time to work out who decides.

Step 6 — Central bank settlement is where the day becomes final

Settlement in central bank money is final in a way that commercial bank money is not. When a payment settles across accounts at the central bank, it cannot be reversed, and legal settlement finality protects it even if one party enters insolvency minutes later. This is the property that makes the system safe, and it is also what makes the deadline absolute: a payment that misses the window is not late, it has not happened, and the obligation it was meant to discharge remains outstanding overnight with whatever consequences the underlying contract specifies.

Minimum reserve requirements are settled in the same place. Euro area banks must hold average balances at the central bank over a maintenance period, which gives some flexibility day to day but none across the period as a whole. Managing the reserve position is a routine part of the money market desk's job and interacts directly with the daily cash decision.

Step 7 — The close produces the numbers everyone else uses

At the end of the day the position is struck and the reporting chain runs. Closing balances are captured by account, currency and entity. Positions feed the overnight batch, which generates the cashflow vectors, runs the liquidity engine and produces the LCR, the maturity ladder, the intraday metrics and the internal management reports that will be on a screen the next morning. Breaks identified in the run are queued for the seven o'clock reconciliation, and the cycle repeats.

The overnight batch is the architectural centre of gravity of most Treasury platforms, and also their principal limitation. It produces a clean, dated, auditable snapshot — genuinely valuable, and not something to give up carelessly. It also means the bank's official view of its own liquidity is a photograph of last night. Through a quiet week that is entirely adequate. Through the week of March 2023 when uninsured deposits moved in hours, it was not, and the gap between the batch cycle and the speed of events is the strongest argument for the architectural changes described in Part 4.

Beyond the day: stress testing, the survival horizon, and the plan for when it goes wrong

Regulatory ratios are floors, and no serious Treasury function manages to them alone. The LCR applies one prescribed scenario with one set of assumptions over thirty days. A bank's own stress framework applies several scenarios, over several horizons, with assumptions calibrated to its own business — and it is the internal framework, documented in the ILAAP, that supervisors actually interrogate.

Three scenario families are standard. An idiosyncratic stress assumes something happens to this bank alone: a ratings downgrade, an adverse news story, a loss announcement, with wholesale funding closing and deposits leaving while markets function normally. A market-wide stress assumes the opposite: the bank is unaffected but funding markets close, asset haircuts widen and secured funding tenors shorten. A combined stress assumes both at once, which is what actually happens, and is the scenario the board should be shown.

Survival horizon under three scenarios

The output that matters is the survival horizon: the number of days the bank can meet its obligations under stress without new funding. It is a more honest measure than a ratio because it has an intuitive meaning that a non-specialist board member can interrogate. It is also more revealing, because it exposes the shape of the problem rather than a single point: a bank with ninety days of survival that falls off a cliff on day thirty-one has a different problem from one that declines smoothly.

The Internal Liquidity Adequacy Assessment Process, or ILAAP, is the annual document in which the bank sets all of this out. It describes the risk appetite, the measurement framework, the models and their validation, the stress scenarios and their results, the buffer composition, the contingency plan, and the governance around all of it. Supervisors assess it as part of their annual review, and its quality shapes the bank's supervisory outcome — including, in the euro area, any institution-specific liquidity requirement imposed above the regulatory minimum.

The contingency funding plan is only a plan if it has been rehearsed

Every bank has a contingency funding plan, and their quality varies enormously. A serious plan names quantitative triggers rather than describing conditions qualitatively: a defined fall in the LCR, a defined level of deposit outflow over a defined period, a defined widening in the bank's own spreads, a rating action. It assigns an owner to each trigger and specifies who is convened when it fires. It lists the available actions in the order they would be taken, with an estimated amount and an estimated time to execute for each: monetising level 1 assets, drawing pre-positioned central bank collateral, curtailing new lending, suspending buybacks, accelerating asset sales. And it specifies the communication protocol, internal and external, because in a liquidity event what the bank says is itself a driver of what happens next.

The rehearsal is what converts the document into a capability. A plan that has never been tested contains assumptions nobody has checked: that collateral can be mobilised in the time assumed, that the people named are contactable, that a system can produce an intraday position outside its normal schedule, that legal agreements permit an action the plan assumes. European operational resilience requirements have made this testing explicit rather than optional, which is a change from a decade ago when a well-drafted plan was frequently the whole of the deliverable.

The regulatory calendar runs alongside all three tracks

Deliverable Frequency What it reports
LCR Daily internally, monthly to the supervisor Buffer against 30-day stressed net outflows
Additional liquidity monitoring metrics (ALMM) Monthly Funding concentration, contractual maturity ladder, unencumbered assets, roll-over of funding
Intraday liquidity metrics Monthly Peak usage, available intraday liquidity, time-specific obligations
NSFR Quarterly Structural funding over one year
IRRBB Quarterly EVE and NII under six prescribed shocks, outlier test
Asset encumbrance Quarterly and annually Share of assets pledged and their sources
ILAAP Annual Full internal liquidity adequacy assessment
Resolution and MREL reporting Annual and quarterly Eligible liabilities by entity and maturity

  1. Why are initial price thoughts announced deliberately wide of fair value?

  2. What is the new issue premium actually buying?

  3. Why might a bank issue a bond it has no cash need for?

  4. Under matched-maturity FTP, why is a ten-year loan at the same client rate as a five-year loan potentially value-destroying?

  5. What does a replicating portfolio do, and why is it contentious?

  6. Why can a bank with an LCR of 143% still fail to make a payment at eleven in the morning?

  7. Why do haircuts make liquidity crises non-linear?

  8. What distinguishes a contingency funding plan that works from one that does not?

  9. Why is the overnight batch both the strength and the weakness of a Treasury platform?

2Track D: The collateral a bank posts is a funding cost nobody budgeted for

A hedge is supposed to remove risk. It also creates an obligation to deliver cash or securities to a counterparty, at short notice, in amounts the bank does not control, on precisely the days when markets move most. Derivative collateral is the mechanism by which a well-hedged bank can run out of liquidity, and it has produced more sudden funding stress in the last decade than the unsecured market has.

Hedging converts market risk into liquidity risk, and the conversion is not free

Start with the problem the collateral system was built to solve. Two banks enter a ten-year interest rate swap. Over the following year rates move, and the swap acquires value: one side is now owed a substantial sum that will only be paid over the remaining nine years. That side is exposed to the other's failure. Before 2008 this exposure was managed largely by choosing good counterparties and hoping. Afterwards, it is managed by requiring the losing side to post collateral continuously, so that the exposure never accumulates.

The consequence is that a paper loss becomes a cash payment. A swap that has moved against the bank by €400m does not merely show a negative mark in the valuation system. It generates a demand for €400m of cash or eligible securities, usually the next morning, delivered to the counterparty or the clearing house. The bank has not lost €400m — the swap is hedging something whose value moved the other way — but the hedged item, typically a loan book or a bond portfolio, generates no offsetting cash at all. The hedge is symmetric in value and asymmetric in cash. That sentence is the whole of Track D.

The contractual machinery: master agreement, CSA, and what "eligible" means

Two documents govern almost every bilateral derivative in the market. The ISDA Master Agreement sets the legal relationship: the events of default, the right to terminate all transactions and settle a single net amount rather than each trade separately, and the governing law. The Credit Support Annex, or CSA, sits underneath it and governs collateral: what may be posted, how positions are valued, how often, and how disputes are handled. A bank's liquidity exposure to its own derivative book is determined almost entirely by the terms negotiated in these annexes, often years earlier, frequently by people who were optimising for something else.

Four CSA parameters decide how much cash a given move demands. The threshold is the amount of exposure a counterparty will tolerate before requiring any collateral at all; a zero threshold means every euro of movement is collateralised. The minimum transfer amount prevents trivial calls by suppressing anything below a floor. Eligible collateral defines what may be delivered — cash in specified currencies, government bonds within a rating and maturity band, sometimes nothing but cash — and each type carries a valuation percentage, the CSA's word for a haircut. Rehypothecation rights determine whether collateral received may be reused, which decides whether an incoming margin flow is usable liquidity or an inert holding.

Variation margin and initial margin answer different questions

Variation margin settles today's mark-to-market and moves in both directions. Each day the position is revalued, the change is calculated, and the party whose position has deteriorated transfers the difference. Over the life of a trade the net variation margin flows equal the net value change, so variation margin is economically a settlement mechanism rather than a security deposit. It is also, for accounting and clearing purposes in most cleared markets, treated as settlement of the exposure rather than collateral against it.

Initial margin is a buffer against what might happen next, and it only moves one way at inception. It covers the potential change in value between a counterparty's default and the point at which the surviving party has closed out the position, conventionally assumed to take several days. It is posted by both parties, held in segregated accounts so that it is not available to the collecting party's creditors, and — decisively for Treasury — it is dead weight. The bank cannot use it, cannot earn a normal return on it, and does not get it back until the trade matures or is terminated. The phased introduction of margin requirements for uncleared derivatives extended this to a large population of banks and buy-side firms that had never posted initial margin before.

Feature Variation margin Initial margin
Covers Today's realised change in value Potential future change during a close-out period
Direction Two-way, changes daily Posted by both parties, held throughout
Reusable Usually, where rehypothecation is permitted No — segregated
Liquidity effect A daily cash flow, sometimes very large A permanent encumbrance of assets
Sized by The mark-to-market move A model or a regulatory schedule

Central clearing changes the counterparty but not the arithmetic. A cleared trade faces a central counterparty rather than a bank, which removes bilateral credit risk and imposes a stricter, less negotiable margin regime: initial margin set by the clearing house model, variation margin in cash and typically settled intraday, plus a contribution to the default fund and an exposure to further calls if another member fails. A clearing house can and does raise margin requirements when volatility rises, which means the liquidity demand from a cleared book increases exactly when markets are disorderly.

Rating triggers turn a credit event into an immediate cash outflow

Many collateral agreements contain terms that activate on a downgrade. The mildest require additional collateral once the bank's rating falls below a specified level, by lowering the threshold to zero or increasing the valuation haircuts. Harsher versions give the counterparty the right to terminate the agreement and demand settlement of the net amount. Similar provisions appear well beyond derivatives: in securitisation documentation requiring a replacement account bank or swap provider, in clearing membership terms, and in some corporate deposit agreements.

The liquidity regulation treats this explicitly, and the reason is instructive. The LCR requires a bank to hold liquid assets against the additional collateral it would have to post following a defined downgrade of its own credit quality, calculated as though the downgrade had occurred. It also requires an add-on for the potential increase in collateral demanded by adverse market movements, computed by looking back over a historical window and taking the largest observed net collateral outflow over a thirty-day period. Both are attempts to make a bank hold liquidity against events driven by its own deterioration — the moment when everything else is also going wrong.

The September 2022 gilt episode showed the mechanism at full scale

In late September 2022 UK government bond yields rose extremely rapidly following a fiscal announcement, and the resulting collateral spiral nearly broke a market. The parties under pressure were liability-driven investment funds used by UK defined-benefit pension schemes. These funds held leveraged, collateralised positions in gilts and gilt derivatives, designed to hedge the schemes' long-dated liabilities against falls in interest rates. The hedges were economically sensible and, in a slow-moving market, well constructed.

The mechanism was the one described above, running in a feedback loop. As yields rose, the value of the hedging positions fell and margin calls were issued. To meet them, funds sold gilts. Gilt selling pushed yields higher still, which generated further margin calls, which forced further selling. The speed was the binding constraint: schemes had assets, but converting pension scheme assets into eligible collateral takes days, and the calls came in hours. The Bank of England intervened with temporary purchases explicitly framed as a financial stability measure to restore orderly conditions rather than as monetary policy.

Three lessons transfer directly to a bank treasury. First, a hedge sized correctly for the risk can be sized wrongly for the liquidity it demands, and the two calculations are different. Second, collateral eligibility and the operational time required to mobilise it matter as much as the quantity of assets held; assets that cannot be delivered by the deadline are not collateral. Third, the correlation that matters is not between the hedge and the hedged item — that one works — but between the margin call and every other demand on the bank's liquidity, and that correlation is close to one.

Collateral management is an optimisation problem with a hard constraint

Figure 14 One inventory, several destinations, and a rule that permits only one

The bank holds one pool of securities and faces several competing demands on it. The liquidity buffer requires unencumbered assets. The central bank pool requires pre-positioning, which takes time and encumbers what is placed there. Triparty baskets and bilateral repo consume assets to raise cash. Clearing houses and bilateral counterparties demand initial margin in segregated form. Each destination has its own eligibility criteria and its own haircut schedule, and a security delivered to one is unavailable to all the others.

Optimisation means delivering the cheapest acceptable asset to each destination. If a counterparty accepts single-A corporate bonds, delivering a German government bond is a real and avoidable cost, because that bond would have counted fully in the liquidity buffer while the corporate bond would not. Doing this well requires a single, live inventory of every security with its eligibility across every destination and its current encumbrance status — which is exactly the system component most banks discovered they did not have when they first tried to answer the question.

Collateral management is one of the clearest cases in Treasury where the data model determines whether the business capability exists at all. Three requirements are non-negotiable. The inventory must be maintained at the level of the individual lot rather than the position, because half a holding can be pledged and half free. Encumbrance must be a live attribute updated on settlement rather than a nightly reconciliation, since a decision taken at ten in the morning against last night's picture can double-pledge an asset. And eligibility must be a rule set evaluated against current attributes rather than a static flag, because a security's eligibility changes when its rating changes, when it approaches maturity, or when a clearing house amends its schedule.

The optimisation itself is a constrained allocation problem that is well understood mathematically and hard operationally. The difficulty is almost never the algorithm; it is that the inputs live in five systems that disagree about what the bank owns.

  1. Why is a hedge symmetric in value but asymmetric in cash?

  2. What is the difference between variation margin and initial margin, from a liquidity standpoint?

  3. Why does the LCR require an add-on for rating triggers?

  4. What broke in the September 2022 gilt episode, given that the funds held sufficient assets?

  5. Why is negotiating securities eligibility into a CSA worth doing?

  6. Why must collateral inventory be tracked at lot level rather than position level?

2Worked calculations: build the ratios by hand once, and they stop being magic

A ratio you have never computed is a number you can only repeat. This part takes Meridian Bank's balance sheet and constructs the LCR, the NSFR and a cash ladder line by line, then breaks each of them on purpose to show what actually moves them. Every figure is illustrative. The arithmetic is the point.

Building the LCR: the numerator is easier than it looks and the denominator is where the judgement lives

The formula is a fraction, and both halves are constructed rather than observed. The numerator is the stock of high-quality liquid assets, taken at market value, reduced by prescribed haircuts, and subject to composition caps. The denominator is total expected cash outflows over thirty days of prescribed stress, less expected inflows, with the inflows capped. Neither half is a number that exists anywhere in the bank's accounting records. Both are assembled.

Step 1 — The buffer, after haircuts and after caps

HQLA component Market value (€bn) Level Haircut Weighted (€bn)
Central bank reserves 18.0 1 0% 18.00
EEA sovereign bonds 32.0 1 0% 32.00
Covered bonds, highest quality 8.0 1 7% 7.44
Corporate and covered, high grade 5.0 2A 15% 4.25
Lower-rated eligible securities 0.6 2B 50% 0.30
Total 63.6 61.99

Two caps then have to be tested, and both pass here. Level 2 assets in total may not exceed 40% of the buffer: €4.55bn of weighted Level 2 against a buffer of €61.99bn is 7.3%, comfortably inside. Level 2B may not exceed 15%: €0.30bn is 0.5%. Meridian's buffer is dominated by Level 1, which is expensive in foregone yield and is the reason the constraint does not bite. A bank running closer to the caps would find that adding Level 2A assets increases the buffer by less than their weighted value, because the excess is simply disallowed.

The buffer is therefore €62.0bn, rounded, and this is the numerator.

Step 2 — Gross outflows, category by category

Outflow category Balance (€bn) Run-off rate Outflow (€bn)
Stable insured retail deposits 130.0 5% 6.50
Less stable retail deposits 50.0 10% 5.00
Corporate operational deposits 55.0 25% 13.75
Corporate non-operational deposits 40.0 40% 16.00
Unsecured wholesale maturing within 30 days 12.0 100% 12.00
Secured funding backed by Level 1 collateral 22.0 0% 0.00
Secured funding backed by non-HQLA collateral 2.0 100% 2.00
Committed credit facilities to corporates 40.0 10% 4.00
Committed liquidity facilities 5.0 30% 1.50
Derivative collateral and rating-trigger add-ons — — 2.00
Other contractual outflows — — 1.00
Gross outflows 63.75

Three lines in that table carry most of the information. The two corporate deposit lines differ by fifteen percentage points for the same money in the same currency at the same bank, purely because one relationship is operational — tied to payments, payroll or custody services the client cannot easily move — and the other is a rate-seeking balance. The secured funding lines differ by a hundred points depending only on what was pledged against them. And the committed facilities generate €5.5bn of outflow from balances that have not been drawn and may never be, which is exactly the cost the contingent liquidity charge in the FTP framework is designed to push back to the desk that granted them.

Step 3 — Inflows, and the cap that limits them

Inflow category Balance (€bn) Inflow rate Inflow (€bn)
Interbank placements maturing within 30 days 11.4 100% 11.40
Performing corporate loan repayments due 10.0 50% 5.00
Reverse repo secured by non-HQLA collateral 3.0 100% 3.00
Reverse repo secured by Level 1 collateral 7.0 0% 0.00
Other contractual inflows — — 1.00
Gross inflows 20.40

The inflow cap prevents a bank from relying on its own book to save it. Recognised inflows may not exceed 75% of gross outflows, here 75% of €63.75bn, or €47.81bn. Meridian's €20.40bn sits far below the cap, so the cap does not bind. It exists because without it a bank could report a comfortable ratio purely by holding a large book of short-dated loans, which would be a claim that its borrowers will repay punctually during precisely the stress in which they are least able to.

Step 4 — The ratio

Net outflows are €63.75bn less €20.40bn, or €43.35bn. The buffer is €62.0bn. The LCR is 62.0 divided by 43.35, which is 143%.

Figure 15 Building the LCR denominator line by line

Step 5 — Break it three ways and watch what moves

Reclassify €10bn of corporate deposits from operational to non-operational. Nothing about the bank changes: the same clients hold the same balances. But the run-off rate on that €10bn rises from 25% to 40%, adding €1.5bn to outflows. Net outflows become €44.85bn and the ratio falls to 138%. A five-point move in the headline ratio, caused entirely by a classification judgement about whether a deposit is tied to a service.

Encumber €5bn of Level 1 sovereign bonds in a repo maturing beyond thirty days. The bank has raised cash and improved nothing: the buffer falls to €57.0bn while net outflows are unchanged, and the ratio falls to 131%. This is the arithmetic behind the rule that secured funding is cheap in spread and expensive in ratio terms.

Grant €10bn of new committed credit facilities to corporate clients. Not a euro is drawn. At the 10% rate the facilities add €1.0bn to outflows, net outflows become €44.35bn, and the ratio falls to 140%. The business line that signed the facilities booked a commitment fee and consumed roughly €1.0bn of the group's liquidity capacity, which is the entire argument for charging it internally.

Building the NSFR: the same balance sheet, weighted for stability rather than speed

The NSFR asks a structural question, so it weights every liability by how long it will stay and every asset by how long it must be funded. Available stable funding is the numerator, required stable funding the denominator, and the requirement is that the first covers the second.

Available stable funding Balance (€bn) ASF factor ASF (€bn)
Equity and subordinated capital 34.0 100% 34.00
Wholesale debt, residual maturity over 1 year 47.0 100% 47.00
Wholesale debt, 6 months to 1 year 12.0 50% 6.00
Wholesale funding under 6 months, incl. repo from financials 26.0 0% 0.00
Stable retail deposits 130.0 95% 123.50
Less stable retail deposits 50.0 90% 45.00
Non-financial corporate deposits under 1 year 95.0 50% 47.50
Total ASF 303.00
Required stable funding Balance (€bn) RSF factor RSF (€bn)
Cash and central bank reserves 18.0 0% 0.00
Level 1 securities, unencumbered 32.0 5% 1.60
Level 2A securities, unencumbered 5.0 15% 0.75
Level 2B and other securities 0.6 50% 0.30
Securities encumbered over 1 year 8.0 100% 8.00
Residential mortgages over 1 year, low risk weight 120.0 65% 78.00
Other residential mortgages over 1 year 20.0 85% 17.00
Corporate loans over 1 year 70.0 85% 59.50
Corporate loans under 1 year 20.0 50% 10.00
Other loans (SME and consumer) 30.0 85% 25.50
Trading and other financial assets 45.0 50% 22.50
Net derivative assets 12.0 100% 12.00
Other assets 19.4 100% 19.40
Committed facilities, off balance sheet 45.0 5% 2.25
Total RSF 256.80

The NSFR is €303.00bn divided by €256.80bn, or 118%. Two features of the calculation are worth extracting. The first is that retail deposits, which the LCR treats as an outflow to be covered, are treated by the NSFR as the most valuable funding a bank has, contributing €168.5bn of stable funding from €180bn of balances. The two ratios read the same liability in opposite directions, and both are right, because they are asking different questions. The second is that the mortgage book alone requires €95bn of stable funding, which is why a bank cannot grow a mortgage book faster than its deposit franchise without issuing term debt — the constraint is mechanical, not prudential judgement.

Building a cash ladder: the one calculation that is done every morning

A five-day euro ladder for Meridian, constructed from the opening position forward. Known items are contractual; expected items are modelled from history. The projection tells the desk not just whether it is short, but on which day.

Item (€bn) Day 1 Day 2 Day 3 Day 4 Day 5
Opening balance 4.10 2.90 3.20 1.55 2.35
Interbank placements maturing +3.40 +1.10 +0.60 +2.20 +0.40
Other receipts +0.10 +0.90 +0.35 +0.20 +0.60
Repo maturing, to repay or roll −2.90 −0.40 −1.80 −0.35 −1.20
Expected net client flow −1.20 −0.90 −0.30 −1.25 −0.55
Coupon and redemption payments −0.60 −0.40 −0.50 0.00 −0.90
Projected close 2.90 3.20 1.55 2.35 0.70
Internal floor 2.50 2.50 2.50 2.50 2.50
Headroom +0.40 +0.70 −0.95 −0.15 −1.80

The ladder says something the daily balance never would. Today is comfortable. Day 3 breaches the internal floor by €0.95bn and Day 5 by €1.80bn, driven by a repo concentration and a coupon date. The desk therefore acts today, not on Day 3, because a five-day-forward shortfall can be covered with a one-week placement or a term repo arranged at leisure, whereas the same shortfall discovered on the morning it occurs must be covered in whatever market is open at the time. The value of the ladder is entirely in the lead time it creates.

Figure 16 The same logic extended: the full contractual maturity ladder

Extend the ladder from days to years and it becomes the maturity gap. The short buckets are heavily negative because deposits are contractually immediate and loans are not. That negative short gap is not a fault to be corrected — it is the business, quantified. The questions it answers are how large the gap is, whether it is growing, whether behavioural analysis justifies treating part of it as long-dated, and how much of it the bank is willing to carry. Those four questions are the entire agenda of an ALCO meeting.

  1. Meridian reclassifies €10bn of corporate deposits from operational to non-operational. Compute the new LCR.

  2. Why does encumbering €5bn of sovereign bonds hurt the LCR twice as much as granting €10bn of undrawn facilities?

  3. Meridian's Level 2 assets are 7.3% of the buffer against a 40% cap. What would change if they were at 45%?

  4. Why do the LCR and the NSFR treat a €180bn retail deposit base in opposite directions?

  5. A five-year bond issued four years and seven months ago still holds its cash. Why has the NSFR deteriorated?

  6. The cash ladder is comfortable today and breaches the floor on Day 3. Why does the desk act today?

  7. What does a persistently negative short-bucket gap in the maturity ladder indicate?

2

  1. Which input to the annual funding plan is genuinely discretionary?

  2. Why does a bank rarely decline to call a callable instrument?

  3. What happens in a bank with no transfer pricing?

  4. Under matched-maturity FTP, why can a 10-year loan at the 5-year client rate destroy value?

  5. Why is a hedge symmetric in value but asymmetric in cash?

  6. Meridian moves €10bn of corporate deposits from operational to non-operational. What happens to the LCR?

3Treasury sits in the middle of every flow, internal and external

Every other desk in a bank faces a client. Treasury faces the bank itself. That single structural fact — that its principal counterparty is its own institution — explains almost everything about how Treasury behaves, why its authority is delegated rather than commercial, and why the people who work there are permanent insiders. This part maps the four camps Treasury deals with, the flows that move between them, and then goes deep on the two constituencies that matter most and are least understood: the investors who buy the bank's debt, and the depositors who are also investors whether or not they know it.

Treasury has no clients, no revenue, and no choice about who it deals with

A trading desk can walk away from a bad counterparty. Treasury cannot walk away from its own bank. If the mortgage business originates €12bn of 25-year fixed-rate loans this year, Treasury must fund them, hedge them, and account for them in the liquidity ratios — there is no version of the conversation where Treasury declines. It can price the activity punitively, escalate to the Asset and Liability Committee, or ask for a limit, but the balance sheet arrives regardless. This is the deepest asymmetry between Treasury and every client-facing function. Understand it and the rest of the governance architecture stops looking bureaucratic and starts looking necessary.

Treasury also earns no external revenue of its own, and any apparent profit it books is a warning sign rather than an achievement. A well-run Funds Transfer Pricing system — the internal mechanism by which Treasury charges business lines for the funding they consume and pays them for the funding they raise, covered in detail in Part 5 — should run close to flat over a cycle. A large positive Treasury P&L usually means the FTP curve is mispriced and one business line is subsidising another. A large negative one usually means the same thing in reverse. Treasury is a redistribution engine, not a profit centre, and the moment it starts being managed as a profit centre the incentives inside the bank quietly invert.

Because Treasury cannot select its counterparties internally, its influence has to be exercised through price and process rather than through refusal. The three levers are the FTP curve, which makes expensive balance sheet expensive to the business that consumes it; the limit framework, which caps how much of a scarce resource any one unit may take; and the ALCO mandate, which converts board-level risk appetite into operating constraints. None of these is a veto. All of them are slow. That is why a Treasury function that is respected in the annual planning cycle is worth more than one that is merely obeyed in a crisis.

The four camps: who Treasury actually deals with

Treasury's counterparty universe divides cleanly into four camps, and confusing them is the most common analytical error newcomers make. The internal business lines are borrowers and depositors — they consume and supply balance sheet. Group Treasury and ALM is the internal bank that intermediates between them. External investors and counterparties supply the funding the internal bank cannot source internally. And market infrastructure plus the official sector — central securities depositories, central counterparties, TARGET2, the central bank, the supervisor, the rating agencies — provide the plumbing and the permission without which none of the rest functions.

Camp Who they are What they want from Treasury What they give Treasury
Internal business lines Retail and mortgage banking, corporate lending, transaction banking, markets/trading, wealth Cheap, certain, unlimited balance sheet; stable FTP rates they can quote clients against; fast answers on new product Balance-sheet demand and supply; client behaviour data; deposit gathering; early warning on pipeline
Group Treasury and ALM Funding, money markets, ALM, collateral, treasury operations, FTP administration A coherent mandate, credible limits, and data it can trust Term funding, intraday liquidity, hedges, internal pricing, ratio management
External investors and counterparties Bank treasuries, money market funds, insurers, pension funds, asset managers, hedge funds, central banks as reserve managers, retail bondholders, repo counterparties, derivative counterparties, depositors Compensation for credit, liquidity and structural subordination risk; disclosure; predictability of issuance Cash, term, collateral capacity, and a secondary market that sets the bank's funding cost
Market infrastructure and official sector CSDs, CCPs, TARGET2 and T2S, correspondent banks, the ECB and national central banks, the supervisor (SSM/national competent authority), EBA, rating agencies, resolution authority Settlement finality, correct and timely regulatory returns, collateral eligibility compliance, credible resolvability Settlement, collateral eligibility, standing facilities, licence to operate, external ratings, resolution planning certainty

The camps are not equally negotiable, and the gradient runs in a specific direction. Business lines negotiate constantly and loudly. Investors negotiate on price at the moment of a transaction and not otherwise. Infrastructure and the official sector do not negotiate at all — a failed settlement, a breached LCR, or a rejected COREP return is a fact, not a discussion. A Treasury that spends its political capital arguing with the last camp instead of the first has misread where its leverage lies.

The flows that move between the camps define the working day

A Treasury team's calendar is almost entirely determined by the flows below, and each has a rhythm that rarely changes. Some are intraday and continuous — payments, cash positioning, collateral substitution. Some are daily — cash forecasting, repo rollover, LCR estimation. Some are monthly or quarterly — regulatory returns, ALCO. Some are annual — the funding plan, the ICAAP and ILAAP submissions, the rating review. Reading the table below as a clock rather than as a list is the fastest way to understand what Treasury does with its time.

From To What flows When
Business line Treasury Balance-sheet demand: loan pipeline, deposit forecast, undrawn commitment growth, new product requests Continuous; formalised in annual plan and monthly forecast
Treasury Business line FTP charge on assets, FTP credit on liabilities, liquidity and balance-sheet limits, product sign-off Rates published daily or weekly; limits set annually, reviewed quarterly
Treasury Investors Benchmark and private-placement issuance, roadshows, investor presentations, funding plan guidance Issuance windows around results; typically 4–8 public transactions a year for a bank of this size
Investors Treasury Orders into a bookbuild, cash at settlement, secondary-market price signals, reverse enquiry Order books build over 2–4 hours on launch day; secondary signals daily
Treasury Repo counterparties and CCPs Collateral delivery, margin, substitutions, haircut negotiation, GC and specials borrowing Intraday and overnight, continuously
Central bank Treasury Standing facilities, open market operations, collateral eligibility decisions, reserve remuneration Weekly tenders; facilities available daily; eligibility rules updated periodically
Treasury Supervisor LCR and NSFR returns, ALMM (additional liquidity monitoring metrics), COREP and FINREP submissions, ILAAP, funding plan templates Monthly for LCR and ALMM, quarterly for NSFR and COREP, annually for ILAAP
Treasury Rating agencies Funding and liquidity data packs, management meetings, pre-issuance briefings, resolution structure explanation Annual review cycle plus event-driven contact
Treasury CSD / TARGET2 / correspondents Payment instructions, settlement instructions, intraday liquidity positioning, reserve account management Continuous through the settlement day; cut-offs drive the intraday clock
ALCO Treasury Mandate, risk appetite translation, limit approvals, hedging strategy decisions, FTP methodology approval Monthly, with ad-hoc sessions in stress
Risk (second line) Treasury Limit breach notifications, early warning indicator triggers, stress test results, model validation findings Daily monitoring, monthly reporting, immediate on breach
Treasury Board / board risk committee Survival horizon under stress, contingency funding plan status, funding plan execution against target, ratio headroom Quarterly in normal conditions; daily or more in a liquidity event
Treasury Group Finance Interest income attribution, FTP outturn, hedge accounting documentation, balance-sheet forecasts Monthly close cycle
Resolution authority Treasury MREL targets, subordination requirements, resolution entity structure constraints Annual determination, with transitional paths
Figure 17 Treasury as the hub: the only desk whose main counterparty is its own bank

The investors who buy bank paper are not the investors who buy corporate paper

A corporate issuer sells a claim on a business. A bank sells a claim on a balance sheet that is itself made of claims — and the buyer base reflects that. The people who buy Meridian Bank's senior preferred paper are, to a large degree, other financial institutions managing their own regulatory ratios, insurers matching long-dated liabilities, and index funds mechanically replicating a benchmark. Very few of them are making an unconstrained judgement about whether the bank is a good investment. Most are making a constrained judgement about whether this instrument, at this tenor, in this currency, fits a rule they must obey. Understanding which rule each investor obeys is the single most useful piece of buy-side knowledge in Treasury.

This matters commercially because the shape of the funding plan is chosen to reach specific pockets. Meridian's €14bn annual wholesale programme (illustrative only) is not €14bn of one thing. It is covered bonds to reach bank treasuries and central banks, senior preferred to reach asset managers and insurers, senior non-preferred and Tier 2 to satisfy MREL while reaching credit funds, commercial paper and certificates of deposit to reach money market funds, and structured private placements to reach specific insurance mandates. Each instrument is a key cut for a particular lock.

Bank treasuries buy other banks' paper, but only the parts that count as liquidity

A bank treasury investing its own liquidity buffer is not looking for yield — it is looking for assets that count as High Quality Liquid Assets under the LCR delegated regulation. The Liquidity Coverage Ratio requires a bank to hold enough HQLA to survive thirty days of a defined stress. HQLA is tiered: Level 1 (central bank reserves, most EEA sovereign debt, and — critically — certain very high quality covered bonds), Level 2A (some covered bonds, high-rated corporate and public sector paper, subject to a 15% haircut and a cap), and Level 2B (a narrow tail, more heavily haircut). What is not HQLA in any tier, as a general matter, is another bank's senior unsecured debt.

The exclusion is deliberate and it is the reason the interbank market shrank so dramatically after 2010. If Bank A's liquidity buffer were allowed to consist of Bank B's unsecured paper, and vice versa, the system's aggregate liquidity would be an accounting illusion — in a stress, both banks would need to sell exactly when the other could not buy. Regulators call this wrong-way correlation, and the LCR framework removes it by refusing HQLA status to unsecured financial-sector claims. The practical consequence: a bank treasury will happily buy another bank's covered bond, because a covered bond is secured on a ring-fenced pool of mortgages or public-sector loans and can qualify as Level 1 or Level 2A depending on rating, issue size and structural features. It will buy short-dated CDs for operational cash management. It will be much more reluctant to hold five-year senior unsecured, because that instrument consumes balance sheet, attracts capital, and delivers no liquidity value.

Add to this the outflow treatment on the funding side and the interbank market's decline becomes overdetermined. Under the LCR, unsecured wholesale funding from a financial institution maturing within thirty days attracts a 100% outflow assumption — the regulation assumes it will not be rolled. So short interbank borrowing is expensive from the borrower's perspective too. Both sides of the trade are penalised. What replaced it is secured financing: repo, where the lender takes collateral and the borrower gets a far better regulatory outcome. Meridian's €38bn of secured wholesale funding out of €85bn total (illustrative only) is a fairly typical European profile and reflects exactly this migration.

Money market funds are the fastest money in the book and the easiest to lose

Money market funds are collective investment vehicles that invest in very short-dated, high-quality instruments and offer investors near-daily liquidity — which makes them structurally the most fragile buyer a bank has. They buy commercial paper and certificates of deposit, typically under twelve months and heavily concentrated inside three months. In Europe they operate under the Money Market Funds Regulation, which constrains their weighted average maturity, weighted average life, and the minimum proportion of assets maturing daily and weekly. Those constraints are not soft preferences. A fund approaching its weekly liquidity floor will stop rolling paper immediately, regardless of its credit view on the issuer.

The consequence is that MMF funding disappears in exactly the conditions where a bank needs it most. When the fund's own investors redeem, the fund must generate cash, and the fastest way to generate cash is to let maturing paper run off rather than reinvest. There is no negotiation, no relationship call, no roadshow that fixes this. It is mechanical. Supervisors read heavy reliance on short-dated MMF funding as a vulnerability indicator, and it shows up directly in the ALMM concentration-of-funding-by-counterparty template and in the funding-maturity profile template.

MMF funding is nonetheless useful and no large bank does without it. It is cheap, it is deep, it absorbs day-to-day cash-flow mismatch, and it supports the operational plumbing of a transaction bank. The discipline is proportion, not abstinence: keep the reliance small relative to the buffer, avoid concentration in a handful of fund complexes, ladder the maturities so no single week carries a cliff, and never let the treasurer's convenience in a benign market become the structure the bank has to defend in a stressed one.

Insurers and pension funds are the only natural buyers of the long end

An insurer selling a forty-year annuity has a forty-year liability, and its overriding investment problem is finding assets that behave like that liability. This is asset-liability matching, and it produces a buyer that genuinely wants ten-, fifteen- and twenty-year paper — a preference almost no other investor class shares. For a bank issuer, insurers and pension funds are therefore the constituency that makes long-dated funding possible at all. Without them, the term structure of a bank's liabilities would be far shorter and its structural liquidity position correspondingly worse.

Solvency II shapes what they can buy, and any Treasury team pitching them needs to know how. Under the standard formula, the capital charge for spread risk on a bond rises with both duration and credit quality deterioration, which pushes insurers toward higher-rated long paper and makes them relatively less willing to hold long-dated, low-rated bank debt. The matching adjustment and volatility adjustment mechanisms, which allow insurers to discount liabilities using a spread-enhanced rate when assets are held to maturity against matched liabilities, further reward predictable, non-callable cash flows. Callable structures, extension risk, and anything whose cash flows are uncertain are penalised. This is why a bank chasing insurance demand issues bullet paper in size, in euro, with clean documentation — and why a structure that looks clever to the issuer often kills the bid.

Insurers are slow money in the best sense: they are price-sensitive but not headline-sensitive. They will decline a deal because the spread is ten basis points too tight for their internal hurdle. They will not dump a position because of a negative news cycle, because doing so would break the matching that justified the purchase. For a Treasury building resilience, a bond held by an insurer to maturity is worth more than the same bond held by a fund that will sell it on the first bad print.

Asset managers buy the index before they buy the credit

A very large share of the demand for a benchmark bank bond comes from funds whose mandate is to track or beat a credit index, and index membership rules therefore price the deal. The major euro credit indices apply eligibility criteria on minimum issue size, currency, remaining maturity, rating, and instrument type. A €300m tap of an existing line will not enter the index. A €1bn new benchmark with five years to maturity in euro almost certainly will. That difference is worth real basis points, because index inclusion converts a discretionary bid into a mechanical one: funds tracking the index must buy some quantity of the bond simply because it exists.

This is why Meridian's reference trade is structured the way it is. A €1bn five-year senior preferred benchmark (illustrative only) clears the size threshold comfortably, sits in the most liquid part of the euro curve, and is the instrument type with the broadest index representation. The €3.2bn book and the three basis point new issue premium — the concession over where the issuer's existing curve trades, paid to get the deal done — reflect a deal built to catch index demand rather than one that had to be sold house by house. Had the same funding been raised as €250m of four separate private placements, the all-in cost would likely have been higher and the secondary curve would have gained nothing.

The corollary is that index-driven demand is not loyalty. Index funds do not care about the issuer's story. When a bond approaches one year to maturity and falls out of the index, those same funds become mechanical sellers. A Treasury that reads a huge index-driven order book as a vote of confidence has misread the arithmetic. It is a vote of construction.

Hedge funds in the book are information, not a problem

Fast money — hedge funds, relative value desks, credit arbitrage funds — participates in bank issuance for reasons that have little to do with holding the bond. Typical strategies include the asset swap basis trade, buying the new bond and swapping it to floating to capture a spread against a reference rate; capital structure arbitrage, going long one layer of an issuer's stack and short another when the relative pricing between senior preferred, senior non-preferred and Tier 2 looks inconsistent; and simple new-issue flipping, buying at the reoffer price and selling into the immediate aftermarket if the concession was generous.

Syndicate desks watch fast-money participation carefully during a bookbuild, because it changes what the order book means. A €3.2bn book made up of long-only asset managers and insurers is a deal that will trade stably in the secondary market. The same €3.2bn book with a large fast-money component may see substantial selling in the first days, pushing the bond wider than reoffer and damaging the issuer's curve — which then raises the cost of the next transaction. Books are therefore analysed by investor type, not just by size, and allocations are skewed toward accounts expected to hold.

None of this makes fast money unwelcome. It provides the marginal bid that makes tight pricing possible, it makes the secondary market liquid, and its willingness to arbitrage the capital structure is part of what makes the structural subordination of senior non-preferred paper properly priced rather than notional. It is a component to be measured and allocated deliberately, not a contaminant to be excluded.

Central banks and official institutions are the anchor bid, especially in covered bonds

Central banks and sovereign wealth funds manage reserves under mandates that prioritise capital preservation and liquidity over return, which makes them natural buyers of the safest end of bank issuance. In practice this means covered bonds, supranational and agency paper, and short-dated high-grade instruments. A reserve manager buying a euro-denominated covered bond is diversifying currency exposure and picking up a modest spread over the sovereign while retaining a claim that is secured, ring-fenced, and legally insulated from the issuer's insolvency under the covered bond directive framework.

The Eurosystem's purchase programmes have also functioned as a structural bid at various points, and their presence or absence changes the market's clearing level. When a central bank is buying covered bonds in the primary market, the bid is deep, price-insensitive, and reliable, and issuers can print size at tight spreads. When purchases are tapered or reversed, that bid is withdrawn and private demand has to clear the market at a wider level. Any Treasury reading historical spread series has to adjust for this — the covered bond spread of a period with heavy official purchases is not comparable to one without.

Official-sector participation also carries a signalling weight disproportionate to its size. A named central bank in a covered bond order book is read by other investors as an independent quality endorsement, and it draws real-money accounts in behind it. That is why syndicate desks work official-sector orders early and disclose their presence (in aggregate, anonymously) during the bookbuild.

Retail investors buy bank paper through the branch, and that is where the politics live

Banks with large branch networks can and do place their own debt securities with their own retail customers, and this is simultaneously the cheapest distribution available and the most dangerous. The economics are attractive: a retail buyer typically accepts a lower spread than an institutional one, the distribution cost is a network already paid for, and retail holders are sticky. The instruments range from plain senior preferred notes to structured notes with equity- or rate-linked payoffs sold as savings products.

The supervisory concern, expressed repeatedly across European authorities, is about loss-absorbing instruments in particular. Under the Bank Recovery and Resolution Directive, subordinated instruments and senior non-preferred debt are designed to absorb losses in resolution — that is their regulatory purpose and the reason they count toward MREL, the minimum requirement for own funds and eligible liabilities. If those instruments are held by retail savers who understood them as deposits with a better rate, then imposing losses becomes politically explosive precisely when the resolution authority most needs it to be routine. The authorities' response has included minimum denomination requirements, tightened suitability and appropriateness rules under MiFID II, and explicit supervisory attention to the retail holder base of MREL instruments.

For Treasury the operational lesson is that the distribution channel is part of the instrument's design, not an afterthought. A senior non-preferred deal placed with institutional investors at a €100,000 minimum denomination behaves in resolution exactly as intended. The same economic exposure placed in €1,000 clips through a branch network creates a resolvability question that the supervisor will ask about — and that a resolution authority may treat as a constraint on the credibility of the entire resolution strategy.

Investor type Typical tenor Price sensitivity What makes them buy What makes them disappear
Bank treasuries Under 3 years unsecured; 3–10 years covered Moderate; ratio value dominates spread HQLA eligibility, low risk weight, operational relationship Loss of HQLA eligibility, issuer downgrade, own liquidity stress
Money market funds Under 12 months, concentrated under 3 months High; a few basis points move allocation Yield pickup within MMFR constraints, short WAM, name diversification Own redemptions, weekly liquidity floor pressure, any credit headline
Insurers and pension funds 7–20+ years High on absolute level, low on news flow ALM matching, Solvency II treatment, predictable bullet cash flows Spread below internal hurdle, callable or uncertain cash flows, downgrade below mandate floor
Asset managers (index and active) 3–10 years, benchmark maturities Moderate; relative value against curve and peers Index inclusion, benchmark size and currency, relative value versus comparables Index exclusion, deal too small, rating migration out of index band
Hedge funds and fast money Any; often short holding period Very high; trade the concession Asset swap basis, capital structure dislocations, new issue concession Concession too thin, funding costs rise, volatility spike
Central banks and official institutions 3–10 years, covered and secured Low; mandate-driven Reserve diversification, secured structure, programme eligibility Programme end, currency reallocation, structure falls outside eligibility
Retail investors 2–7 years; also structured notes Low on spread, high on headline rate Branch relationship, headline coupon versus deposit rate Adverse press about the issuer, regulatory restriction on distribution

The depositor is an investor too — and the most important one

Meridian's €275bn of retail and corporate deposits (illustrative only) dwarf its €85bn of wholesale funding, and yet depositors never see a term sheet, never negotiate a spread, and never appear in an order book. They are still investors. They have supplied unsecured, unrated, callable-at-will funding to the bank at a rate the bank largely sets unilaterally. The entire discipline of Asset and Liability Management rests on the fact that this funding behaves nothing like its contract says it does — and on models that attempt to predict how it will behave when it matters.

Deposit beta is the most consequential assumption in the entire bank

Deposit beta is the fraction of a change in the policy rate that a bank passes through to its depositors, and small differences in it move net interest income by hundreds of millions. If the ECB raises rates by 200 basis points and Meridian passes 30% through to its retail savings book, the retail deposit rate rises 60 basis points while the assets funded by those deposits reprice by considerably more. That gap is deposit margin, and in a rising-rate environment it is where a large share of a European retail bank's earnings improvement comes from.

The difficulty is that beta is not a constant, it is a function of level, speed, competition and channel. Beta is typically low for the first increments of a hiking cycle and rises as rates go higher, because depositors ignore the difference between 0% and 0.3% but not the difference between 2% and 3.5%. It is higher for large balances than small ones, higher in digital-only channels than branch-based ones, higher for corporate than retail, and it rises sharply once a visible competitor starts advertising. It is also asymmetric: banks pass rate increases through slowly and rate cuts through quickly, which is why deposit margin compresses more slowly than it expands.

Every beta assumption is a forecast dressed as a parameter, and the EBA IRRBB guidelines require banks to treat it as such. The guidelines expect institutions to document their behavioural assumptions, justify them with historical evidence, test their sensitivity, and demonstrate that the resulting interest rate risk in the banking book measures — both the economic value of equity perspective and the net interest income perspective — remain within the supervisory outlier thresholds. Beta assumptions are also a favourite target of supervisory review, precisely because a bank that assumes low beta reports lower interest rate risk and better projected earnings.

Non-maturing deposits are contractually overnight and behaviourally multi-year

A current account has no maturity date: the customer can withdraw the full balance today. Contractually, it is an overnight liability. Behaviourally, aggregate current account balances at a large retail bank have been remarkably stable for decades, with a substantial core that has never left. If Treasury modelled these balances as overnight, it would conclude the bank has a catastrophic maturity mismatch and hedge accordingly — buying enormous quantities of short-dated protection against an interest rate exposure that does not economically exist.

So banks model non-maturing deposits by splitting the balance into a volatile portion and a stable core, then assigning the core a behavioural maturity profile. A typical approach — illustrative only — might treat 20% of a retail current account balance as volatile and repriceable immediately, and distribute the remaining 80% across a replicating portfolio with an average life of three to five years. That replicating portfolio is then the basis for both the FTP credit paid to the retail business and the hedge Treasury actually executes. The EBA IRRBB guidelines constrain this: they impose a cap on the average repricing maturity that may be assigned to non-maturing deposits, with tighter caps for less stable categories such as wholesale non-operational balances, precisely to prevent banks from assuming away their exposure.

The modelling choice has a direct commercial consequence that business heads feel immediately. A longer assumed behavioural life means a higher FTP credit for the deposit-gathering business, because the deposit is being treated as term funding rather than overnight money. Retail always argues for longer. Second-line risk always argues for shorter. The ALCO adjudicates, and the number it lands on quietly determines whether retail banking looks profitable.

Insured and uninsured deposits run at completely different speeds

Under the Deposit Guarantee Schemes Directive, deposits are protected up to €100,000 per depositor per institution across the EU — and that threshold divides the deposit book into two populations with entirely different behaviour under stress. A depositor whose full balance is covered has little economic reason to run, because the guarantee pays out regardless of what happens to the bank. A depositor holding €4m has every reason to move at the first hint of trouble, because the loss is real and the cost of moving is a few clicks.

The LCR framework encodes this asymmetry directly in its outflow weights. Stable retail deposits — covered by a guarantee scheme and held in an established relationship or a transactional account — attract a 5% thirty-day outflow assumption. Less stable retail deposits attract 10% or more depending on category. Corporate deposits that are operational, meaning they arise from clearing, custody or cash management services the depositor genuinely depends on, attract 25%. Non-operational corporate deposits attract 40%, or 100% where the counterparty is a financial institution. The ratio between the best and worst treatment is twenty to one, and that ratio is the entire reason transaction banking is strategically valuable to a bank's liquidity position.

Uninsured concentration is therefore a first-order risk metric, not a disclosure detail. A bank with €95bn of corporate deposits (illustrative only) needs to know what fraction is above the guarantee threshold, what fraction is operational versus non-operational, and how concentrated it is by single name — because the ALMM concentration-of-funding template will ask, and because in a stress the answer determines how many hours the bank has.

The digital channel compressed the run clock from days to hours

Classical bank run models were calibrated on a world where withdrawing money meant physically going to a branch during opening hours and standing in a queue. That physical friction was a genuine circuit breaker: it capped the hourly rate of outflow, it gave management a night to organise a response, and the visible queue itself was the mechanism by which panic spread — which meant panic spread at walking speed. Every deposit behaviour model built before 2020 embeds this friction, whether or not its documentation says so.

Events in March 2023 removed any remaining doubt that the friction is gone. Deposit outflows at affected institutions ran at a pace that would have been physically impossible through a branch network, driven by mobile banking transfers initiated by customers who learned of the problem through social media rather than by seeing a queue. The relevant change is not just speed but coordination: information now propagates faster than the institution's ability to respond to it, and a concentrated depositor base connected through the same professional networks can act effectively simultaneously.

The supervisory and industry response is still working through the implications, and Treasury teams should assume the assumptions will tighten. The thirty-day LCR horizon was always a convention rather than a physical constant, and a run that empties a material share of the book within a single business day tests the intraday and multi-day survival horizon far more severely than the LCR's monthly arithmetic suggests. Practical mitigations include far more granular real-time deposit monitoring, early warning indicators built on transaction-level flow rather than end-of-day balances, and contingency funding plans whose first actions are executable within hours rather than days.

Deposit type Contractual maturity Behavioural maturity (modelled) LCR outflow weight Run sensitivity
Insured retail current account (stable, transactional) Overnight Core portion often modelled 3–5 years (illustrative only) 5% Low, but no longer negligible in a digital panic
Insured retail savings, rate-shopping segment Overnight or notice 1–2 years (illustrative only) 10% Moderate; rate-driven attrition rather than panic
Uninsured retail (balance above €100,000) Overnight Materially shorter than insured core 10% or higher category High; economically motivated to move first
Corporate operational (clearing, custody, cash management) Overnight Multi-year while the service relationship persists 25% Low-to-moderate; sticky because the service is embedded
Corporate non-operational Overnight or short term Close to contractual 40% (100% if financial counterparty) High; treasurer-managed and actively monitored
Financial institution deposits Overnight to short term Contractual 100% Very high; first to leave on any signal
Retail term deposit within 30 days Fixed, short Fixed, with early-withdrawal option Treated per residual maturity and penalty structure Low within term; renewal risk at maturity

Treasury sits permanently inside the information wall, not beside it

An equity capital markets banker becomes an insider when a client shares a transaction and ceases to be one when the deal is announced. A Treasury officer is an insider on the first day and stays one until they leave. The material non-public information Treasury holds is not about a client — it is about the employer, a listed issuer with publicly traded shares and bonds. The true intraday liquidity position, the internal stress test results, the survival horizon under the severe scenario, the funding plan for the next twelve months, the content of the supervisory dialogue, and any early warning indicator that has triggered are all price-sensitive facts about that issuer that are not in the public domain.

This produces a regime of permanent restrictions rather than deal-by-deal ones. Treasury staff are typically placed on a permanent insider list under the Market Abuse Regulation, which requires issuers to maintain lists of persons with access to inside information. They are subject to closed periods around results announcements during which dealing in the bank's own shares is prohibited, and to standing restrictions on dealing in the bank's own debt instruments. Personal account dealing pre-clearance applies not only to the employer's securities but frequently to peer bank securities as well, because knowledge of one's own funding conditions carries inferential information about the sector.

The most delicate structural conflict is that the same institution both issues bonds and makes markets in them. Meridian's own markets division may quote two-way prices in Meridian's senior preferred paper. That desk must not know what Treasury knows about the upcoming funding plan, the buyback intentions, or the liquidity position — otherwise the bank would be trading its own paper on inside information about itself. The separation is enforced by physical and logical segregation, distinct reporting lines, restricted-list controls that block the trading system, and clear rules on when Treasury may cross the wall to brief the desk (typically only at the point where information becomes public, or under a formally logged wall-crossing with a compliance officer present).

The funding plan itself deserves specific attention because people underestimate it. A decision to issue €2bn of senior non-preferred in September rather than €1bn tells the market something about MREL headroom, about the maturity profile, and about how the bank views its own cost of funds. A decision to pull a planned transaction tells the market something worse. Both are price-sensitive information about a listed issuer, and both are known inside Treasury weeks before they are known outside. Handling the funding plan as an ordinary internal planning document rather than as inside information is a common and serious control failure.

The enforcement stack for Treasury information barriers is technical, and it should be auditable end to end.

Restricted and insider lists. A permanent insider list maintained under MAR, with timestamped entries for every person granted access to a defined category of inside information, plus a restricted list that feeds directly into the order management system to block trading in named securities. The block must be enforced at the system level, not by policy alone.

Personal account dealing surveillance. Pre-clearance workflow for every employee trade, automated matching of employee accounts against the restricted list, and post-trade surveillance that flags trades in the employer's securities or peer names ahead of price-sensitive events.

Entitlement-based access to ALM and stress outputs. Stress test results, survival horizon calculations, ILAAP working papers and the funding plan should sit in access-controlled repositories with role-based entitlements, not in shared drives or email attachments. Access reviews should be periodic and evidenced. Under BCBS 239 — the principles for effective risk data aggregation and risk reporting — the same systems must also demonstrate lineage and accuracy, which conveniently means the access controls and the data controls can be built once.

Wall-crossing logs. Every instance of Treasury briefing a public-side function, including the bank's own trading desk or investor relations, recorded with the person, the information disclosed, the timestamp, and the compliance approver. The log is the evidence base if a regulator asks.

Communication surveillance. Lexicon and behaviour-based monitoring across email, chat and recorded voice, tuned to Treasury-specific risk phrases around liquidity, ratios and issuance timing. Off-channel communication controls matter more here than almost anywhere else in the bank, because the information at stake is about the firm itself.

The internal teams divide along a first-line / second-line fault line

Treasury is not one team, and the boundary that matters most is not between desks but between the people who take risk and the people who independently constrain it. The first line executes: it issues bonds, borrows and lends cash, hedges the banking book, moves collateral, and settles payments. The second line — Liquidity Risk within the risk function — sets limits, validates models, monitors compliance, and reports breaches to a governance body that does not report to the Treasurer. Collapsing these two into a single reporting line is a governance failure that supervisors specifically look for, because it means the person managing the position is also the person deciding whether the position is acceptable.

Team Division Core role Primary technology Stage of the flow owned
Funding & Long-Term Issuance Group Treasury (first line) Build and execute the term funding plan; run benchmark and private placement issuance; manage investor relations for debt Issuance and deal management platforms, DCM syndicate interfaces, funding plan models Origination and execution of term liabilities
Money Markets & Short-Term Cash Group Treasury (first line) Manage the daily cash position, CP/CD programmes, unsecured short-term borrowing and placement, central bank reserve balances Treasury management system, cash forecasting engine, dealing platforms Daily and intraday cash and short-term funding
ALM / Balance Sheet Management Group Treasury (first line) Model behavioural assumptions, manage IRRBB, run the replicating portfolios, execute the banking book hedges, own the FTP curve construction ALM system (cash-flow engine, EVE and NII simulation), behavioural models, hedge accounting tools Structural interest rate and liquidity position
Liquidity Risk Risk function (second line, independent) Set and monitor liquidity limits, own the stress testing framework and survival horizon, validate ALM and deposit models, report breaches Risk reporting and limit monitoring platforms, independent stress engines, model validation tooling Independent oversight across the whole flow
Collateral & Repo Management Group Treasury / Markets (first line) Manage the collateral pool, execute repo and securities financing, optimise collateral allocation, manage central bank eligibility and haircuts Collateral management system, triparty agent interfaces, CCP connectivity Secured funding and collateral supply
Treasury Operations & Payments Operations (support) Settle payments and securities, manage nostro and correspondent accounts, monitor intraday liquidity under BCBS 248, confirm and reconcile trades Payments engine, TARGET2/T2S connectivity, SWIFT infrastructure, intraday liquidity monitoring Settlement, payment and intraday liquidity execution
Treasury Finance / FTP Administration Finance (support / control) Run the FTP calculation and allocation, attribute net interest income to business lines, produce regulatory liquidity returns, maintain hedge accounting documentation FTP engine, general ledger, regulatory reporting platform (COREP/FINREP/ALMM) Measurement, attribution and regulatory reporting

Treasury Operations deserves more respect than it usually gets, because BCBS 248 made intraday liquidity a supervised discipline rather than a plumbing concern. The Basel Committee's monitoring tools for intraday liquidity management require banks to measure daily maximum intraday liquidity usage, available intraday liquidity at the start of the day, total payments, and time-specific obligations. A bank can be comfortably compliant with a 143% LCR and still fail to meet a payment at eleven in the morning because collateral was in the wrong place. The LCR is a thirty-day measure. The settlement day is measured in minutes, and the people who manage it sit in Operations.

The FTP administration function sits in Finance rather than Treasury for a reason that is easy to miss. Whoever calculates the FTP charges and credits decides, in effect, how net interest income is split between business lines. If that calculation sat inside a Treasury that also negotiated with those business lines, the numbers would be contested every month. Placing the calculation in Finance, applying a methodology approved by ALCO, makes the outcome a measurement rather than a negotiation. Treasury builds the curve. Finance applies it. ALCO owns the method.

ALCO is where the balance sheet is actually decided

The Asset and Liability Committee is the body that converts board risk appetite into operating decisions about the balance sheet, and it is the single most important governance forum in Treasury. A typical composition includes the Chief Financial Officer or Chief Executive as chair, the Group Treasurer, the Chief Risk Officer, the heads of the major business lines, the head of Liquidity Risk, and a senior Finance representative, with the head of ALM presenting. It meets monthly in normal conditions, and moves to weekly or daily during stress. Attendance by business heads is not a courtesy — the decisions ALCO takes bind their P&L.

What ALCO actually decides is narrower and more specific than its name suggests. It approves the FTP methodology and any significant change to the curve. It sets and reviews internal liquidity and interest rate risk limits within the board-approved appetite. It approves the annual funding plan and material deviations from it. It approves behavioural assumptions for non-maturing deposits, prepayment models, and the replicating portfolio design. It decides the strategic hedging position for the banking book. It reviews stress test results and the survival horizon, and it decides whether to invoke the contingency funding plan. It does not run the trading book, it does not set credit policy, and it does not approve individual transactions.

Escalation to the board follows a defined path and it should be short. Breaches of board-approved risk appetite, any invocation of the contingency funding plan, material changes to the funding plan, and results of the ILAAP go to the board risk committee and then the board. The board sets appetite; ALCO operates within it; Treasury executes; second-line risk verifies and escalates. Delegation flows down that chain. Escalation flows up it. When either direction is broken — a board that sets appetite without understanding the survival horizon, or a Treasury that manages a breach without escalating — the failure mode is the same: the institution discovers its actual risk position at the moment it can no longer do anything about it.

The governance chain: delegation flows down, escalation flows up

  1. Why can a bank not count another bank's senior unsecured bond toward its High Quality Liquid Assets?

  2. How does investor type determine the shape of a funding plan, not just its cost?

  3. What is deposit beta and why is it the most consequential ALM assumption?

  4. Why is Treasury staff insider status permanent rather than transaction-specific?

  5. A corporate client keeps €200m with the bank purely to earn a rate, having no cash management relationship. Another keeps €200m as clearing balances for its payroll and supplier payments. Why does Treasury value these differently?

  6. Why does modelling non-maturing deposits at their contractual maturity produce a worse outcome than modelling them behaviourally?

  7. Why should the head of Liquidity Risk not report to the Group Treasurer?

  8. Meridian's LCR is 143% and yet it could still fail to make a payment at 11:00. Explain.

3

  1. Why can't a bank count another bank's senior bond as HQLA?

  2. What is deposit beta?

  3. Why is Treasury staff insider status permanent?

  4. Which investor pocket supplies the long end of a funding plan?

Part 3B — A banking group is not one balance sheet, and liquidity does not flow freely inside it

Everything described so far assumed a single balance sheet in a single currency: one pool of assets, one pool of funding, one ratio to manage. Real groups are nothing like that. They are dozens of legal entities spread across jurisdictions, each with its own regulator, its own capital and liquidity ratios, its own board, and its own firmly held view about whether money may leave. The group balance sheet you see in the annual report is an accounting construct. The constraints that bind on a bad Tuesday morning are entity-level, and they do not net.

The requirement applies at several levels at once, and the group number is the least binding of them

Start with the legal architecture, because the regulation follows it. A banking group is a set of separately incorporated companies linked by ownership. Meridian Bank — our running example, a European universal bank with €420bn of assets — is headquartered in the euro area. It owns a UK subsidiary, a Polish subsidiary, a US broker-dealer subsidiary, and it operates a branch network in the United States. A subsidiary is a separate legal person: it has its own balance sheet, its own creditors, its own insolvency estate, and it can fail without its parent failing. A branch is not a separate legal person: it is an extension of the parent, and its assets and liabilities are the parent's. That distinction, which sounds like a lawyer's detail, determines almost everything about how liquidity behaves inside the group.

Prudential requirements under the EU framework apply at three levels. The Capital Requirements Regulation (CRR) and Capital Requirements Directive (CRD) impose requirements on a consolidated basis — the group as a whole, as supervised by the consolidating supervisor, which for a significant euro-area institution is the ECB under the Single Supervisory Mechanism. They also impose requirements on an individual (or "solo") basis — each credit institution in the group must meet capital and liquidity requirements on its own books. And where a subsidiary sits outside the consolidating supervisor's jurisdiction, the local host supervisor imposes its own regime: the UK subsidiary answers to the PRA, the US broker-dealer to the SEC and the Federal Reserve, the Polish subsidiary to the KNF and, for euro-area purposes, sits outside the SSM perimeter entirely.

Now the uncomfortable arithmetic. Meridian reports a consolidated LCR of 143%. That single number is a weighted composite. It is entirely consistent with a UK subsidiary running at 105% — comfortably above the 100% minimum but with almost no headroom — and a Polish subsidiary running at 260% because local deposit funding vastly exceeds local lending opportunities. The consolidated figure is dominated by the parent, which holds the largest share of HQLA. Nothing in the consolidated ratio tells you whether the liquidity is sitting where the obligations are. If the group's stress bites in sterling, the €62bn HQLA total is irrelevant; what matters is what the UK entity can reach without asking anyone's permission.

This is the single most common analytical error made about banking groups. The group ratio is reassuring precisely because it aggregates. Aggregation is what hides the problem. A treasurer who manages to the consolidated LCR and a supervisor who reads only the consolidated LCR will both be surprised at the same moment, and it will be a moment when surprise is expensive. The discipline is to read the entity-level ratios first and the group ratio last, and to treat the difference between them as the real subject of the conversation.

Waivers exist, and they are the exception that proves the rule. CRR permits a competent authority to waive the application of liquidity requirements at the individual level, so that a parent and its subsidiaries are treated as a single liquidity sub-group. The conditions are demanding: there must be no current or foreseen material practical or legal impediment to the transfer of funds; the entities must be subject to common risk management, measurement and control; there must be a legally binding commitment from the parent to support the subsidiary; and the supervisors involved must agree. Within a single member state, such waivers are granted with some regularity. Across borders, they are granted sparingly, and the reason is structural rather than technical: a host supervisor who waives the local requirement is accepting that, in a crisis, the depositors and creditors in its jurisdiction depend on a decision taken by a parent board in another country, under the supervision of another authority, with another electorate. Few host supervisors are willing to be in that position, and their reluctance has not diminished since 2008. The EU has a cross-border liquidity waiver regime on paper; its take-up has been modest, and the Banking Union's incomplete state — no fully mutualised deposit insurance, no unified resolution funding at the required scale — is the reason usually given.

Trapped liquidity is the defining constraint of a cross-border group

Define it precisely, because loose use of the term causes bad decisions. Trapped liquidity is liquidity held in a legal entity within the group that cannot be upstreamed to the parent or moved to another entity, because a local prudential requirement, a local ring-fencing or subsidiarisation regime, a contractual restriction, or the exercise of supervisory discretion prevents the transfer. It is not the same as liquidity that is merely inconvenient to move, and it is not the same as liquidity that is committed to a business use. It is liquidity the group owns on its balance sheet and cannot deploy where it is needed.

Several distinct mechanisms create it, and they compound. The first is the local liquidity requirement itself. If the UK subsidiary must run an LCR of at least 100% on its own books, every pound of HQLA needed to reach 100% is unavailable to the group, because withdrawing it breaches the requirement. In practice the constraint binds well above 100%, because no subsidiary board will operate at the regulatory floor; a management buffer of 10 to 20 percentage points is normal, and that buffer is trapped too. The second is large exposure limits on intragroup exposures. Under CRR, exposures to a single counterparty — including a group affiliate — are capped as a percentage of Tier 1 capital, with the possibility of exemption for certain intragroup exposures at national discretion. Where the exemption is not granted, a subsidiary simply cannot lend the parent more than the limit allows, regardless of how much surplus it has.

The third is structural reform legislation. The UK's ring-fencing regime requires large banks to separate retail deposit-taking activities into a legally distinct ring-fenced body, with restrictions on its exposures to the rest of the group and on the dividends and loans it may pass out. The purpose is to insulate retail depositors from investment banking losses; the effect on Treasury is that retail deposit surpluses inside the ring-fence are the most trapped liquidity in the entire group. Comparable structural constraints apply in other jurisdictions in different forms, and the US intermediate holding company requirements for foreign banking organisations create an analogous pool.

The fourth is resolution planning, and the fifth is discretion. Resolution authorities require that operating subsidiaries be capable of being recapitalised without disrupting critical functions, which in practice means pre-positioned resources that must stay put. And when stress arrives, host supervisors have the power — sometimes formal, sometimes exercised through a phone call that is never written down — to instruct a local entity not to remit funds. Every experienced treasurer plans on the assumption that discretionary restrictions appear exactly when the group most wants the money, because that is when the host supervisor is most worried about its own depositors.

The practical consequence is a permanent cost, not a contingent one. A group with trapped liquidity must hold more liquidity in aggregate than a single-entity bank running an identical business, because the surpluses in one entity cannot offset the deficits in another for regulatory purposes. Buffers do not net. The excess sits in HQLA — central bank reserves and high-quality government bonds — earning less than the bank pays to fund it, and that negative carry is a recurring charge against group earnings for as long as the structure exists. It is one of the largest hidden costs of operating a cross-border group, and it is almost never disclosed as a line item.

Figure 18 Consolidated comfort, entity-level constraint

Intragroup funding is a transaction with a price, a limit and a resolution consequence

Money does move inside a group, but it moves as a contract. Parents and subsidiaries fund each other through recognisable instruments: intragroup deposits placed by a surplus entity with a deficit entity, term loans, committed and uncommitted liquidity facilities, and subordinated instruments that count towards the recipient's capital. Every one of these is a legal obligation between two separate companies. It has a maturity, a price, a governing law, and a place in the recipient's insolvency waterfall. Treating intragroup positions as internal bookkeeping is the mistake that resolution planning was designed to eliminate.

Pricing is not optional, and it is not free of consequence. Intragroup transactions must be priced at arm's length — on terms that unconnected parties would have agreed — because tax authorities in both jurisdictions will otherwise reallocate profit and levy penalties. That means an intragroup loan carries a spread reflecting the borrower's standalone credit, the tenor, and the currency, and the transfer pricing documentation must support it. It also means the group's internal FTP framework and its transfer pricing policy have to be reconciled: FTP is a management tool designed to drive behaviour, transfer pricing is a tax obligation designed to allocate profit, and where they diverge somebody has to be able to explain why in writing.

Internal MREL is the mechanism that makes single point of entry resolution work. MREL — the minimum requirement for own funds and eligible liabilities, set under BRRD — obliges a bank to hold a minimum amount of instruments that can absorb losses and recapitalise the firm in resolution. In a group, the resolution authority designates a resolution entity, typically the group holding company or parent bank, which issues external MREL to third-party investors in the market. The operating subsidiaries then issue internal MREL — bail-inable instruments subscribed by the resolution entity rather than by external investors. When a subsidiary makes losses, the resolution authority writes down or converts the internal MREL, the loss travels up to the resolution entity, and the subsidiary is recapitalised without entering local insolvency and without any external creditor of the subsidiary being touched. Critical functions keep running. Depositors never notice.

That is single point of entry, and its alternative is meaningfully different. Under single point of entry (SPE), resolution powers are applied at one entity — the top of the group — and losses are channelled there from below. Under multiple point of entry (MPE), the group is divided into resolution groups that can each be resolved separately by their own authorities, with each maintaining its own external MREL issued into the market. SPE suits groups managed as an integrated whole with centralised funding; MPE suits groups whose regional operations are self-funded and could plausibly stand alone. The choice is made by the resolution authorities in a resolution college, and it has direct funding consequences: an MPE group issues term debt from several entities into several investor bases, pays for the fragmentation in spread, and loses the cheapest funding structure in exchange for a more credible cross-border resolution.

Intragroup mechanism What it moves Constraint on it Who approves
Intragroup deposit / short-term loan Cash and liquidity, usually short tenor Local LCR outflow treatment; large exposure limit; host supervisor discretion in stress Group Treasury within pre-agreed limits; subsidiary ALCO
Term intragroup funding Stable funding for local asset growth NSFR at both entities; transfer pricing documentation; withholding tax Group ALCO; local board; tax function
Committed liquidity facility (parent to subsidiary) Contingent liquidity, drawn only on trigger Counts as outflow at the provider under LCR; may not be recognised as inflow by host Group ALCO; host supervisor acceptance
Internal MREL (down-streamed) Loss-absorbing capacity, not usable liquidity Set by resolution authority as a percentage of the subsidiary's requirement; cannot be reduced unilaterally Resolution authority; resolution college
Capital injection (equity or AT1) Permanent capital, effectively irreversible Local capital requirement; repatriation requires supervisory consent Host supervisor; group board
Dividend upstreaming Retained earnings to the parent Local distributable reserves; supervisory distribution restrictions; MDA constraints Host supervisor; subsidiary board

Read the last column, because it is the one that matters in a crisis. Almost every mechanism that moves meaningful resources upward requires the consent of somebody outside group management. That is the design intent of the post-crisis framework, and a group funding plan that assumes consent will be forthcoming under stress is not a plan.

Currency is a second dimension of the same problem, and swapping does not solve it

Liquidity has to be adequate in each material currency, not merely in total. The EU framework requires institutions to monitor and report liquidity positions by significant currency — broadly, any currency accounting for more than 5% of total liabilities — and supervisors may impose currency-specific expectations where a mismatch is material. The logic is simple: a euro is not a dollar on the day you owe dollars. Meridian may hold €62bn of HQLA and still be unable to meet a $3bn outflow if the HQLA is euro-denominated government paper and the dollar market has stopped functioning.

The instinctive response — "we will swap" — deserves to be taken apart. An FX swap is a contract to exchange one currency for another today and reverse the exchange at a fixed rate on a future date. It is economically a secured loan in one currency collateralised by the other, and it is one of the largest funding instruments in global banking. Covered interest parity is the textbook proposition that the forward exchange rate should exactly offset the interest rate differential between the two currencies, so that borrowing euro and swapping into dollars costs the same as borrowing dollars directly. If it did not hold, the argument goes, arbitrage would close the gap.

The cross-currency basis is the amount by which it does not hold. Since 2008 there has been a persistent, measurable deviation from covered interest parity, expressed as a spread — the basis — added to one leg of the swap. For a bank swapping euro into dollars, the basis has typically been negative, meaning the euro-based borrower pays more than the interest differential alone would imply. The reasons are structural rather than mysterious: post-crisis balance sheet constraints, leverage ratio costs and counterparty limits mean the dealers who once arbitraged the gap now charge for the balance sheet they commit, and a persistent structural demand for dollar funding from non-US banks and investors meets a finite supply.

The basis widens exactly when it hurts. Dollar funding stress is the classic pattern: at quarter-ends, when balance sheet is scarce, and in genuine crises — March 2020 being the most recent large-scale example — the cost of swapping into dollars spikes, sometimes violently. This is not a coincidence but the mechanism working as expected: the basis is the price of scarce intermediation, and intermediation is scarcest when everyone wants it. The consequence for Treasury is direct. A funding plan that says "we are short dollars but we will swap" is making an assumption about the price and availability of a market on the worst day of the year, and it is making that assumption silently unless somebody writes it down.

FX swaps used as funding carry rollover risk that the balance sheet does not display. A three-month FX swap funding a five-year dollar asset is a maturity mismatch, but it appears in the accounts as a derivative rather than as short-term wholesale funding. Twenty times over five years, the bank has to return to the market and roll. If on one of those occasions the market is closed, or the basis has moved 80 basis points against the bank, the position has to be unwound at whatever price is available. Supervisors have become considerably more attentive to this in recent years, and the sensible internal control is a limit on the proportion of foreign-currency assets funded by swaps shorter than a defined tenor, monitored and escalated like any other limit.

Structural FX risk is the exposure nobody trades into deliberately

Define it from zero. When Meridian owns a Polish subsidiary, that subsidiary's assets, liabilities and earnings are denominated in złoty. To produce consolidated euro accounts, the subsidiary's net assets are translated into euro at the closing spot rate, and its earnings at an average rate. If the złoty depreciates against the euro, the euro value of Meridian's net investment falls, and the difference flows to a translation reserve within consolidated equity. Nobody chose this exposure. It arises automatically from owning a foreign operation, which is why it is called structural FX risk — it is a property of the structure, not a position in a trading book.

The naive hedge is to sell the foreign currency forward, and it creates a subtler problem. If the exposure is a PLN 8bn net investment, selling PLN 8bn forward against euro will stabilise the euro-translated value of that investment: when the złoty falls, the loss on the net investment is offset by a gain on the hedge. Under IFRS, a net investment hedge allows that offsetting gain to be recognised in the same translation reserve rather than in profit or loss, so reported equity stops moving with the currency. So far the hedge does exactly what was intended.

Now look at the capital ratio, and the result reverses. The CET1 ratio is capital divided by risk-weighted assets. The Polish subsidiary contributes to both. Its capital contribution translates at spot; its RWA, being złoty-denominated assets, also translates at spot. When the złoty falls, both numerator and denominator shrink together — and because the subsidiary's RWA is generally larger relative to group RWA than its capital is relative to group capital, the ratio moves in a way that partly self-stabilises. A hedge that fixes the euro value of the numerator removes that natural offset: the capital stops falling, but the RWA still does, and the ratio moves in the opposite direction to what the unhedged position would have produced. Hedging the amount destabilises the ratio.

Two practical notes. Hedging costs money: the forward points on a high-yielding currency represent a real, recurring carry cost, and hedging a złoty or forint net investment is materially more expensive than hedging a Swiss franc one. And a net investment hedge only qualifies for hedge accounting if it is documented as such at inception, with effectiveness tested; a hedge put on without the documentation delivers the economics and pushes the accounting volatility straight through profit and loss, which is the worst of both worlds.

Booking model and entity architecture are a technology decision as much as a legal one

Every position lives somewhere, and where it lives determines who regulates it. A trade, a deposit, a loan, a security — each is booked in a specific legal entity, and often in a specific branch of that entity. The booking determines which balance sheet it sits on, which LCR and NSFR it affects, which capital requirement it consumes, which supervisor sees it, which tax authority taxes the income, and which insolvency estate the counterparty joins if things go wrong. The booking model is the set of rules that decides, for each product and client relationship, which entity books what. It is a legal and regulatory construct that has to be implemented in software.

The systems usually do not treat entity as a first-class concept. Most core banking and trading platforms were designed when the bank was one bank. Legal entity arrived later, as an attribute bolted onto an account or a book — a code in a field, sometimes derived from the branch, sometimes defaulted, sometimes populated by a downstream mapping table maintained outside the system of record. The symptom is familiar: the group can produce a consolidated position quickly and an entity-level position slowly, with a reconciliation step and a set of manual adjustments that somebody explains verbally. That is exactly backwards relative to which number binds.

Structural reform and resolution planning forced expensive corrections, and the UK's departure from the EU forced another. Ring-fencing required UK banks to move retail portfolios into a new legal entity — re-papering contracts, migrating accounts, splitting systems and rebuilding reporting. Resolution planning required groups to demonstrate that critical functions could survive the failure of an affiliate, which meant unpicking shared service arrangements and intragroup dependencies. And the loss of EU passporting rights meant UK-booked business with EU clients had to be re-papered and re-booked into EU entities, at a scale involving hundreds of thousands of contracts across the industry. These programmes ran for years, consumed large fractions of change budgets, and produced no new revenue. They are the reason entity architecture is now treated as strategic rather than administrative.

One consequence for anyone evaluating a Treasury platform. Ask to see the entity-level LCR for a single subsidiary, produced from the same data as the group number, with no manual step in between. If the answer involves an export, a mapping file and an analyst, the platform has an entity problem, and every group-level number it produces should be read with that in mind.

  1. Meridian reports a consolidated LCR of 143%. Why might a supervisor still be concerned?

  2. Define trapped liquidity and name three distinct mechanisms that create it.

  3. Why does trapped liquidity impose a cost even when nothing goes wrong?

  4. What is the cross-currency basis, and why is it a funding planning problem rather than a pricing curiosity?

  5. Why can hedging structural FX risk make the capital ratio more volatile?

  6. What does internal MREL do, and why is it central to a single point of entry strategy?

  7. A trading system stores legal entity as a code derived from a downstream mapping table. Name two concrete risks.

4The Treasury platform is a data problem wearing a risk-management costume

Every liquidity ratio, every transfer price, every intraday alert is the output of a calculation performed on a contract record that some system created months or years earlier. The sophistication of the calculation is rarely the constraint. The constraint is whether the contract record is complete, correctly attributed, versioned, and reconcilable against the accounting ledger that describes the same asset. Part 4 walks the machine from the contract to the supervisory return and back, and treats the plumbing — identifiers, message formats, cut-offs, failover — with the seriousness it earns on the day something breaks.

1. Ask what functions a Treasury needs before you ask who sells them

Start with the clean question: strip away every vendor name and ask what a bank Treasury must be able to do. The answer is stable across institutions, jurisdictions and decades, which is why it is the right starting point. A Treasury must know what it owns and owes and when those positions turn into cash; it must move cash and settle payments within hard deadlines; it must measure liquidity risk under stress; it must measure interest-rate risk in the banking book; it must price internal funding; it must raise money in markets; and it must prove all of the above to a supervisor in a prescribed format. Seven functions. Every product brochure you will ever read is a claim to cover some subset of them.

Position and cashflow capture is the foundation, and it is the block most often underestimated. Its job is to take every contract the bank has entered — a mortgage, a corporate revolver, a term deposit, a repo, an interest-rate swap, a bond in the buffer — and turn it into a dated vector of expected cash movements. This is not a copy of the general ledger. The ledger tells you the balance today; the cashflow vector tells you what that balance becomes over the next ten years under contractual terms and behavioural assumptions. A ledger says a mortgage book is worth €38bn. The cashflow layer says how much of it amortises in month seven and how much prepays if rates fall 100bp.

Cash and payment management is the only block that operates against a deadline measured in minutes. It maintains the intraday view of nostro accounts, central bank reserve balances, expected receipts and outgoing payments, and it drives the decision to fund or place at each point in the day. Its inputs arrive as payment messages and account statements; its outputs are payment instructions and an intraday liquidity usage profile. If it is wrong at 15:00 and the T2 cut-off is at 18:00, the bank has three hours to find out and fix it.

The liquidity risk engine converts the cashflow universe into ratios and ladders. It applies the LCR outflow and inflow weights defined in the Commission Delegated Regulation, classifies assets into HQLA levels with their haircuts and caps, builds the contractual and behavioural maturity ladders, computes NSFR available and required stable funding factors, and runs the internal stress scenarios that sit alongside the regulatory ones. Its output is a set of numbers that the board sees monthly and the supervisor sees at prescribed frequencies.

The ALM/IRRBB engine answers a different question with the same raw material. It measures how the economic value of equity and projected net interest income move when the yield curve moves, under the six prescribed shock scenarios in the EBA guidelines plus internal scenarios. It needs repricing dates rather than cashflow dates — the distinction matters, because a floating-rate loan with a twenty-year maturity reprices every three months and therefore carries almost no economic-value interest-rate risk while carrying twenty years of liquidity risk. Same contract, two engines, two entirely different characterisations.

The FTP engine is where risk measurement becomes commercial behaviour. It attaches an internal transfer rate to every balance-sheet item so that business units earn or pay a rate that reflects the funding and liquidity value of what they originate. Meridian Bank's five-year FTP rate of 3.67% — 3.05% base plus a 62bp term liquidity premium, all figures illustrative — is an output of this engine, and it determines whether a relationship manager fights for a five-year loan or walks away from it.

Funding execution and dealing is the block that touches the outside world. Money-market deals, repo, commercial paper and certificate-of-deposit issuance, benchmark bond execution, FX swaps for currency funding, and the hedges that accompany them. It captures trades, generates confirmations, feeds settlement, and updates positions.

Regulatory reporting is the block that costs the most and is loved the least. COREP liquidity templates, ALMM additional monitoring metrics, FINREP reconciliation, IRRBB supervisory outlier reporting, MREL, and the intraday metrics derived from BCBS 248. Its input is everything; its output is a validated file transmitted on a calendar.

Five cross-cutting capabilities sit underneath all seven blocks and belong to none of them. Reference and market data — curves, indices, security master, counterparty master — because every engine consumes them and inconsistency between them produces reconciliation breaks that look like risk. The limit and escalation framework, which turns a number into an action by defining who is told what at which threshold. Collateral inventory, which tracks where every security actually is. Immutable audit, which makes past results reproducible. And the scenario library, which stores the definition, calibration and approval history of every stress scenario the bank runs.

Functional block What it must do Core input Core output Latency requirement
Position and cashflow capture Explode contracts into dated cashflow and repricing vectors Contract records from source systems Cashflow vectors, repricing schedules Daily batch; intraday for traded positions
Cash and payment management Manage nostro and reserve balances, sequence payments Payment messages, account statements Payment instructions, intraday usage profile Near real time, minutes
Liquidity risk engine Compute LCR, NSFR, ladders, internal stress Cashflow vectors, HQLA inventory, weights Ratios, ladders, survival horizon Daily; monthly for full stress suite
ALM/IRRBB engine Compute EVE and NII sensitivity under shocks Repricing vectors, behavioural models, curves EVE and NII deltas, duration gap Overnight batch; monthly full run
FTP engine Attach transfer rates to every balance-sheet item Contract terms, FTP curve, policy rules Transfer rates, unit P&L attribution Monthly standard; daily for new business
Funding execution and dealing Capture and settle funding trades and hedges Market prices, funding plan, limits Trades, confirmations, settlement instructions Real time during dealing hours
Regulatory reporting Produce validated supervisory returns All of the above plus accounting data COREP, ALMM, FINREP, IRRBB submissions Calendar-driven, monthly and quarterly
Figure 19 Functional architecture: contract to cashflow to risk to price to report

The blocks run on different clocks, and that mismatch is the source of most Treasury architecture pain. Cash management runs continuously through the business day. The ALM engine runs overnight in a batch that may take four to eight hours on a large book. The FTP engine runs monthly for the standing book and daily for new business. Regulatory reporting runs on a calendar. When a trader asks at 14:00 what the LCR impact of a proposed €500m repo would be, the honest answer in a batch-only architecture is that the bank will know tomorrow morning. That answer is unacceptable, so institutions build approximation layers — spreadsheets, tactical calculators, desk-level estimators — that produce a second version of the truth. The second version then disagrees with the overnight run, and a reconciliation process is born. Almost every Treasury reconciliation you will encounter traces back to two systems computing the same quantity at different clock speeds from different snapshots.

2. Banks buy the rulebook and build the judgement

The market splits cleanly along one line: institutions buy what is standardised by regulation and build what encodes their own strategy. The LCR calculation is defined identically for every bank in the European Union, the rules change several times a decade, and maintaining them is pure cost. A vendor writing that logic once and amortising it across three hundred clients wins on economics and on regulatory currency. The FTP policy layer is the opposite: it determines which desk is rewarded for a sticky operational deposit and which is charged for a five-year drawdown facility, which is a management decision about incentives, and no vendor can supply it.

ALM and liquidity risk platforms form the most established category. QRM is the long-standing specialist in balance-sheet and behavioural modelling with deep penetration in sophisticated ALM shops. Moody's Analytics, Wolters Kluwer OneSumX, FIS and Finastra offer integrated ALM-plus-regulatory suites, with OneSumX in particular strong where the regulatory reporting and the risk calculation are wanted from the same data model. Empyrean serves the mid-market ALM segment. Nomis sits adjacent, focused on deposit and loan price optimisation rather than risk measurement — a reminder that pricing analytics and risk analytics are separate disciplines that consume overlapping data.

Dealing and trade capture is dominated by the cross-asset trading platforms. Murex and Calypso (now part of Adenza) cover money markets, repo, FX and derivatives with position keeping, valuation and confirmation generation. ION's Openlink and related products serve treasury and commodity-adjacent workflows. Bloomberg TOMS provides order and trade management with native connectivity to the data and dealing environment traders already inhabit. Finastra Kondor remains widely deployed for treasury trading and position management. The choice usually turns on the derivative complexity the bank runs rather than on the money-market functionality, which is comparatively simple.

Cash and payments is more fragmented and more likely to be partly in-house. SWIFT connectivity is typically obtained through a service bureau or a cloud connectivity provider rather than built, but the payment hub that sequences, prioritises and throttles outgoing payments is frequently proprietary, because payment sequencing logic is intimately tied to the bank's own liquidity policy and client commitments. A bank that gives correspondent-banking clients a same-day guarantee needs a hub that understands that promise.

Collateral management spans vendors and market infrastructure. Adenza and Murex provide collateral inventory, margin call and optimisation modules. Triparty agents — Euroclear and Clearstream in Europe — hold and allocate collateral on behalf of counterparties under triparty repo arrangements, and the bank's internal inventory must reconcile to the agent's records continuously. Central bank collateral pools add a third location for the same securities.

Regulatory reporting has its own specialist tier. AxiomSL, now part of Adenza, with ControllerView, alongside Wolters Kluwer, Vermeg and Regnology, supply the taxonomies, validation rules and submission channels for COREP, FINREP, ALMM and national returns. The value proposition is unambiguous: EBA taxonomies change on a schedule, validation rules run to thousands of assertions, and no bank wants to maintain that alone.

Market data is a duopoly-plus. Bloomberg, LSEG (Refinitiv) and ICE supply curves, indices, security reference data and pricing. The Treasury-specific point is that the curve used to build the FTP transfer curve, the curve used to discount in the ALM engine and the curve used to value the trading book must be the same curve, sourced once and distributed, or the bank will spend its life explaining differences that have no economic content.

Functional block Typical answer Buy or build Why
Cashflow capture and ALM data layer Vendor ALM platform, heavily configured Buy, then build the mappings Engine is generic; source-system mappings are institution-specific
Liquidity risk calculation Vendor engine with configurable weights Buy Rules are common, update burden is high, vendor amortises it
IRRBB engine Vendor ALM platform Buy Prescribed scenarios and metrics; differentiation is in the behavioural models
Behavioural models In-house calibration on vendor framework Build the calibration Deposit stickiness and prepayment are the bank's own customer base
FTP calculation kernel Vendor or in-house Either Arithmetic is simple once the curve and policy are fixed
FTP policy layer Always proprietary Build Encodes incentives and strategic allocation, not a computation
Dealing and trade capture Vendor trading platform Buy Instrument coverage and confirmation plumbing are commoditised
Payment hub and sequencing Frequently in-house Build Sequencing logic reflects the bank's own liquidity policy and client promises
Collateral inventory Vendor plus triparty agent feeds Buy, integrate hard Multi-location tracking demands reconciliation the vendor cannot own
Regulatory report production Specialist reporting vendor Buy Taxonomy and validation maintenance is pure cost
Management reporting and board packs In-house Build Content, framing and thresholds are governance choices

3. Follow one mortgage from origination to the supervisory return

Pick a single contract and trace it end to end, because the abstraction only becomes real when you watch one object move. Take a €300,000 twenty-five-year fixed-rate residential mortgage originated by Meridian Bank (illustrative). At origination the retail lending platform creates the contract record: borrower identifier, product code, notional, rate, maturity date, amortisation schedule, collateral property reference, loan-to-value. That record is the atom. Everything downstream is a transformation of it.

The ALM data layer explodes the contract into a dated cashflow vector under behavioural assumptions. Contractually the mortgage produces three hundred monthly payments of principal and interest. Behaviourally, the bank's prepayment model — calibrated on its own historical book, with a documented methodology — predicts that a portion of the balance disappears earlier, and that the prepayment rate rises when market rates fall below the contract rate. The vector that leaves this stage is not the contract; it is the contract plus a model. That distinction becomes critical two sections from now.

The liquidity engine buckets the vector and applies regulatory weights. For the LCR, a performing residential mortgage generates inflows in the thirty-day window only to the extent of contractual payments due, and those inflows are subject to the cap on total inflows at 75% of outflows. For the NSFR, the mortgage is an asset requiring stable funding, with a required stable funding factor determined by its residual maturity, its risk weight under the standardised approach, and whether it qualifies as an encumbered asset. For the internal maturity ladder, it populates the one-to-five-year and over-five-year buckets. Three different treatments, one contract.

The FTP engine attaches a transfer rate. A twenty-five-year fixed-rate mortgage with a modelled behavioural life of roughly seven years receives a transfer rate built from the base curve at the behavioural tenor plus the term liquidity premium at that tenor, plus any policy add-ons the bank applies to illiquid retail assets. The retail unit pays this rate to Treasury; the margin it keeps is the customer rate minus the transfer rate. If the behavioural life is modelled at seven years and the truth is four, the retail unit is being overcharged and will under-originate mortgages relative to what the bank actually wants.

The reporting layer aggregates it into supervisory returns. The same mortgage appears in the FINREP loans-and-advances tables at amortised cost, in the COREP liquidity templates as an inflow and an RSF item, in the ALMM contractual maturity ladder in its contractual bucket, and in the IRRBB submission as a repricing cashflow in the fixed-rate slot. Four representations, all of which must be explicable from one contract record.

The cascade principle: an error in the contract record reaches every downstream system, and the cost of correction rises with distance from the source. Suppose the maturity date was captured as 2049 instead of 2044. At origination this is a one-field fix. In the cashflow vector it has already generated sixty spurious payment rows. In the liquidity engine it has shifted RSF weighting and moved the ladder bucket. In the FTP engine it has produced a transfer rate calibrated to the wrong tenor, which has already been booked into two months of business-unit P&L. In the COREP return it has been submitted to a supervisor, and correcting it means a resubmission with an explanation. The same one-field error costs minutes at stage one and weeks at stage five.

Stage System Data object created What breaks if it is wrong
Origination Retail lending platform Contract record Everything downstream; the single point of maximum leverage
Cashflow explosion ALM data layer Dated cashflow and repricing vectors Ladder buckets, EVE and NII sensitivity, FTP tenor
Behavioural overlay Behavioural model service Adjusted vector plus model version tag Stress survival horizon, deposit and prepayment assumptions, supervisory challenge
Liquidity classification Liquidity risk engine HQLA level, inflow/outflow weight, ASF/RSF factor LCR and NSFR ratios, buffer adequacy, potential breach reporting
Transfer pricing FTP engine Transfer rate and allocation record Business-unit P&L, origination incentives, product pricing
Aggregation and submission Reporting platform COREP, FINREP, ALMM, IRRBB records Supervisory return accuracy, resubmission risk, findings
Figure 20 The cascade: one bad field corrupts the ladder, the ratio, the price and the return

4. The contract cashflow record is the atom, and its fields are not decoration

Every Treasury system is ultimately manipulating one object: the contract cashflow record. It is the Treasury equivalent of the order record in equity capital markets — the smallest unit that carries enough information for every downstream calculation. Get its field set right and the architecture has a chance. Get it wrong and no amount of engine sophistication recovers the loss.

Field What it is What it drives downstream
Contract ID Unique internal identifier for the position Lineage, audit trail, reconciliation to source and ledger
Legal entity Booking entity within the group Solo versus consolidated LCR, ring-fencing, intragroup exposure limits, entity-level reporting
Counterparty ID and LEI Identity of the other side LCR outflow category (retail, SME, operational, financial), concentration metrics, EMIR and MiFIR reporting
Product code Classified instrument type Weight lookup for LCR/NSFR, IRRBB slotting, FTP policy rule selection
Currency Denomination Currency-level LCR where a currency is material, FX gap, funding plan by currency
Notional / current balance Outstanding amount today Every ratio numerator and denominator, ladder amounts, RSF and ASF calculation
Contractual maturity date Legal final date Contractual maturity ladder (ALMM), NSFR residual maturity bands, structural gap
Behavioural maturity assumption Modelled effective life Behavioural ladder, stress survival horizon, FTP tenor selection, EVE for non-maturing items
Rate type (fixed/floating) Repricing character IRRBB repricing gap, EVE sensitivity, choice of FTP curve point
Reference index and reset frequency Which benchmark and how often it resets Basis risk measurement, NII projection, benchmark transition exposure
Spread Margin over index or over the FTP rate Business-unit margin attribution, NII forecast, product profitability
Amortisation profile Bullet, linear, annuity, custom schedule Shape of the cashflow vector, ladder distribution, weighted average life
Embedded optionality Prepayment right, call, cap, floor Automatic option risk under IRRBB, convexity, stress behaviour, FTP option cost charge
Collateral flag and encumbrance status Whether the asset secures an obligation and to whom HQLA eligibility, unencumbered asset reporting, NSFR treatment, contingent funding capacity
Deposit insurance flag Whether covered by a deposit guarantee scheme LCR outflow rate (stable versus less stable retail), stress assumption severity
Netting set Enforceable netting agreement grouping Derivative exposure netting, margin calculation under EMIR, counterparty concentration

Two of these fields carry problems large enough to sink projects, and both deserve direct treatment.

Encumbrance tracking fails because the same security can honestly appear in three places at once. A German Bund held by Meridian Bank might be counted in the liquidity buffer by the liquidity team, pre-positioned in the central bank collateral pool by the collateral team, and delivered out under a repo by the funding desk — and each of those systems may have a record that is internally consistent. Only one of these can be true for LCR purposes: an asset pledged or delivered is encumbered and cannot count as HQLA. The failure mode is not fraud, it is architecture. When collateral inventory lives in the repo system, the central bank interface and the liquidity engine as three separate stores, double-counting is the default outcome rather than an anomaly. The fix is a single collateral inventory that owns location state for every security lot, where every other system reads rather than asserts, and where the sum of allocations by lot is reconciled against custody and triparty positions daily. Assets pre-positioned at a central bank but not drawn against are a particular trap: they are typically unavailable for immediate sale, and whether they may still be counted depends on the specific arrangement and how quickly they can be recalled.

Behavioural overlay versioning matters because the same contract produces different cashflows under different model versions. A non-maturing current account balance of €1bn has no contractual maturity at all. Under the model version approved in 2024 it might be treated as 60% stable core with a ten-year effective life; under the 2026 recalibration, reflecting observed rate sensitivity during a hiking cycle, it might be 45% core with a six-year life (both illustrative). Both are legitimate. But the LCR reported for March 2025 was computed under the first, and if a supervisor asks in late 2026 for the calculation to be reproduced, running today's model over the archived data produces a different answer and proves nothing. The model version is therefore part of the data lineage, not a property of the code repository. Every calculation record must carry the identifier of the model version and parameter set used, those versions must be immutably archived with their calibration evidence and approval date, and the platform must be able to run a historical date with its contemporaneous model. Institutions that treat models as deployable code rather than as versioned data lose the ability to explain their own history.

5. Identifiers are where systems agree to be talking about the same thing

An identifier is a promise that two systems mean the same object, and the quality of Treasury data is largely the quality of that promise. ISIN — the twelve-character International Securities Identification Number — names a security issue, and is the workhorse for the liquidity buffer, repo collateral and issuance. It names an issue, not a holding, and not a lot: two ISIN-identical bonds, one encumbered and one free, are the same ISIN and different positions, which is exactly why encumbrance must be tracked at lot level and not at ISIN level.

The LEI names the legal entity and is mandatory rather than optional. Under EMIR and MiFIR reporting obligations, a counterparty without an LEI cannot be reported, and therefore in practice cannot be traded with. For Treasury the LEI does more than satisfy reporting: it is the key that connects a deposit in one system, a derivative in another and a credit line in a third to a single counterparty for concentration monitoring under the ALMM concentration-of-funding metric. Without it, the largest single funding provider is invisible because it appears under four different name spellings.

FIGI provides vendor-neutral instrument identification, and CFI classifies instrument type. FIGI matters in architectures that consume multiple data vendors and need a mapping key that no vendor controls. CFI — the ISO 10962 classification code — encodes the instrument's character in six letters and is useful for automated product classification when the internal product taxonomy is incomplete.

The UTI is the identifier that Treasury people underestimate. The Unique Trade Identifier is generated once per derivative or securities-financing transaction and shared by both counterparties, so that regulators can pair the two sides of the same trade in reporting under EMIR and SFTR. Generation rules define which party creates it, and mismatches produce unpaired reports and regulatory breaks. A repo desk that generates UTIs inconsistently produces a reconciliation queue that no amount of downstream cleverness clears.

Internal contract IDs are where the honest pain lives. Every bank has them; almost no bank has them consistent across systems after a merger. The mortgage that is `RM-4471829` in the retail platform is `0004471829` in the ALM feed and `RM4471829-01` in the collateral system, and the mapping table that connects them is maintained by three people, two of whom have left. This is not a hypothetical: post-merger identifier reconciliation is one of the largest hidden costs in Treasury integration programmes, and it surfaces as unexplained differences in the maturity ladder that take months to attribute.

Identifier What it names Who assigns it Where it bites
ISIN A security issue National Numbering Agency Names the issue, not the lot; encumbrance must be tracked below it
LEI A legal entity Local Operating Units under GLEIF Mandatory for EMIR/MiFIR; without it, concentration monitoring is blind
FIGI An instrument, vendor-neutrally OpenFIGI / Bloomberg-originated, openly licensed Cross-vendor mapping when multiple data sources disagree
CFI Instrument classification ISO 10962 scheme, assigned with ISIN Automated product typing when internal taxonomy is thin
UTI A single trade, shared by both sides Generating party per regulatory rules Unpaired EMIR/SFTR reports; repo reconciliation queues
Internal contract ID A position in one system The originating system Never reconciles across legacy estates after a merger

6. Batch, API and streaming are three different theories of what truth is

Batch is the legacy default and it is not merely slow — its latency is a risk, not an inconvenience. In a batch architecture the ALM and liquidity engines consume an end-of-day snapshot, run overnight, and publish results in the morning. The consequence is that at 11:00 the Treasury desk is managing today's balance sheet using yesterday's picture. On a calm day this is tolerable. On the day a large corporate client withdraws €3bn, a rating action prompts wholesale funding providers to shorten tenor, and a repo counterparty declines to roll, the desk is making decisions on a picture that predates all three events. The batch cycle is precisely the interval during which a liquidity stress is invisible.

APIs make calculation on-demand, which changes what questions can be asked. An API-exposed liquidity engine allows the desk to ask what the LCR becomes if a proposed €500m three-month repo is executed, and to get the answer before deciding rather than after. That is a different management capability, not a faster version of the same one. The engineering cost is real: an engine designed for a single nightly run over the full book must be re-architected to compute marginal impact incrementally, because recomputing the entire balance sheet per query does not scale.

Event streaming changes the topology rather than the speed, and that is the deeper shift. In a publish-subscribe architecture — Kafka-style — a position change is published once as an event, and every consumer that cares subscribes to the same stream. The liquidity engine, the collateral system, the intraday monitor and the reporting layer all receive the identical event in the identical order. They do not each pull their own extract at their own time from their own view of the source.

Batch architectures require reconciliation as a structural necessity, not as a consequence of poor engineering.

When the liquidity engine takes an extract at 23:00 and the collateral system takes its own extract at 01:30, and a trade is booked at midnight, the two systems hold different positions and both are internally correct. The reconciliation process exists to discover and explain this. It is permanent, it consumes headcount, and it never finishes, because the cause is architectural rather than operational.

Event architectures dissolve the problem instead of accelerating it. One system owns the position and publishes changes; everyone else is a subscriber to that single ordered stream. Two subscribers cannot disagree about what happened, because they consumed the same events in the same sequence. There is nothing left to reconcile — only lag to monitor.

The honest counterpoint deserves equal weight. The overnight batch provides something streaming does not give away for free: a clean, dated, frozen, auditable snapshot that everyone agrees represents the close of business on a specific date. Supervisory reporting, board reporting and financial accounting all depend on that discipline. Streaming platforms must deliberately reconstruct it — through consistent point-in-time snapshots, offset-anchored replays, or event-sourced state reconstruction at a defined timestamp. Institutions that migrate to streaming without building this back discover at the next reporting cycle that they have real-time numbers and no defensible official ones.

7. Reproducibility is a regulatory requirement, and the FTP policy layer is never for sale

A supervisor may ask a bank to reproduce a liquidity ratio it reported eighteen months earlier, using the data and the model version in force at the time. This single requirement dictates a great deal of architecture. It means calculation results cannot be overwritten when a correction is made — they must be superseded by a new record that references the old one, with the reason and the approver. It means the input data for every reporting date must be archived in the form it had at the time, not reconstructed from a current system that has since been updated. It means model versions and parameter sets must be immutably stored alongside approval evidence. Append-only is not a design preference here; it is the only structure that satisfies the question.

BCBS 239 is best read as a data-architecture specification, not as a governance policy. Its principles map directly onto engineering requirements. Accuracy and integrity demands automated aggregation with minimal manual intervention, because every manual step is an unreproducible transformation. Completeness demands that risk data be capturable across all material entities, business lines and exposures, which is a scoping and coverage requirement on source-system integration. Timeliness demands that aggregated data be available fast enough to act on, which is a latency requirement measured against the decision it supports rather than against a generic standard. Adaptability demands the ability to produce ad hoc aggregations on request — which in practice means a flexible data model rather than hard-coded report extracts, because a supervisor's question during a stress episode will not match any report the bank has pre-built. Institutions that implemented BCBS 239 as a committee and a policy document rather than as a data-platform programme are the ones that still cannot answer a novel question in under a week.

The FTP calculation kernel can be bought; the FTP policy layer cannot. The arithmetic of applying a curve point plus a liquidity premium to a balance is trivial. What is not trivial, and what no vendor can supply, is the set of decisions encoded in the policy: whether operational deposits from cash-management clients receive a credit that reflects their genuine stickiness or a conservative haircut; whether committed but undrawn facilities are charged a contingent liquidity cost and at what rate; whether the term liquidity premium is charged at contractual or behavioural maturity for retail mortgages; whether new business is priced at marginal cost or at a pooled average; whether internal loss-absorbing capacity costs are allocated to business units at all. Each of these is a statement about which activities the bank wants more of. The 62bp term liquidity premium in Meridian Bank's five-year FTP rate (illustrative) is not an observation about markets — it is a decision, informed by market evidence, about how expensive five-year illiquidity should feel to a relationship manager. Buying that decision from a vendor would be outsourcing strategy.

8. A funding decision only becomes irreversible when a message leaves the building

Four distinct message layers separate a Treasury decision from an irreversible cash movement, and confusing them is a common source of operational error. Trade capture records intent. Confirmation records agreement. Payment executes value transfer. Reporting records what happened. Only the third is irreversible, and only after settlement finality.

The front-office layer uses less FIX than a trading desk, for structural reasons. FIX — the Financial Information eXchange protocol — dominates order routing in equities and increasingly in fixed income, where anonymous electronic execution against many venues is the norm. Treasury money-market and repo activity is bilateral, relationship-driven and often voice- or chat-negotiated, and trade capture happens through vendor platform APIs, dealing-platform integrations and direct feeds from the dealing system into position keeping. FIX appears where Treasury touches electronic markets — some repo platforms, FX execution, and secondary bond trading in the buffer — but a Treasury desk's dominant capture path is its trading platform's own interface rather than a FIX session.

The confirmation layer is where SWIFT message types still carry the freight. MT320 confirms a fixed-term deposit or loan between financial institutions, carrying principal, value date, maturity date, rate and interest amount. MT300 confirms an FX transaction, which matters because FX swaps are a primary tool for raising foreign-currency funding. MT540 through MT543 handle securities settlement instructions — receive free, receive against payment, deliver free, deliver against payment — which is the message family that moves repo collateral and buffer securities. A mismatch at confirmation stage is cheap to fix; the same mismatch discovered after settlement is a failed trade with a funding consequence.

The payment layer is where money actually moves and where cut-offs bind. MT202 is the general financial institution transfer — the instruction by which one bank moves funds to another for its own account — and its ISO 20022 successor is pacs.009. MT210 is a notice to receive, telling a correspondent that funds are expected, which is what allows the receiving side to plan intraday position. In the euro area, high-value payments settle in TARGET2, now operating on the T2 platform following consolidation with TARGET2-Securities, with hard daily cut-offs for customer and interbank payments and legally defined settlement finality: once settled, the payment cannot be recalled unilaterally. That finality is the reason the payment layer deserves different controls from every other layer in the stack.

The reporting layer is migrating from MT to camt, and one of those messages changes what is possible. MT940 delivers an end-of-day customer statement and MT950 a statement of an account held with another institution; both are being replaced by camt.053, the ISO 20022 bank-to-customer statement. camt.054 delivers debit and credit notifications for individual entries. The consequential one is camt.052 — the bank-to-customer account report — which delivers intraday account movement rather than an end-of-day summary. Without an intraday feed, a bank's view of its own nostro balances during the day is an estimate built from what it believes it has sent and what it expects to receive. With camt.052, the balance is observed. That is the difference between managing intraday liquidity by inference and managing it by measurement.

Layer Standard Carries Timescale
Trade capture Vendor platform APIs, FIX where electronic Intent to transact: notional, tenor, rate, counterparty Seconds, during dealing hours
Confirmation MT320 deposits, MT300 FX, MT540–MT543 securities Bilateral agreement on economic terms Same day, typically within hours
Payment MT202 / pacs.009, MT210 notice to receive Irreversible transfer of value Minutes, against a hard cut-off
Settlement infrastructure TARGET2 / T2, T2S for securities Central bank money settlement with legal finality Real time gross settlement, cut-off bound
Reporting MT940/MT950 to camt.053, camt.052 intraday, camt.054 notifications Observed account state and individual entries End of day; intraday for camt.052
Figure 21 From trade capture to irreversible payment

9. Treasury has a daily irreversible moment that most functions face once a year

The payment system cut-off is a hard, external, non-negotiable deadline that arrives every single business day. An equity capital markets team faces one irreversible moment per transaction — pricing night. A Treasury faces one every afternoon. If the bank has not funded its position by the T2 cut-off, it either borrows from the central bank's marginal lending facility at a penalty rate, fails an obligation, or both. This asymmetry is why Treasury resilience requirements are stricter than the rest of the bank's, and why the four concepts below must be distinguished rather than used interchangeably.

Redundancy means a duplicate component exists. Two data centres, two network paths, two SWIFT connections. Redundancy alone guarantees nothing about recovery, because a spare component that requires forty minutes of manual intervention to bring into service is a spare component, not a resilient system.

Automatic failover means the duplicate takes over without human action, within a defined time. This is the property that matters at cut-off. The distinction from redundancy is the whole argument: at 17:20 with a cut-off at 18:00, the question is not whether a second payment hub exists but whether it is already processing.

Disaster recovery is quantified by two numbers that people routinely confuse. RPO — recovery point objective — is how much data the institution accepts losing, measured backwards in time from the failure. RTO — recovery time objective — is how long the institution accepts being unavailable, measured forwards. For a payment hub, an RPO of anything above zero means payment instructions may be lost, which is unacceptable, so payment systems demand synchronous replication with zero RPO. For an overnight ALM batch, an RPO of several hours may be perfectly tolerable because the input data can be re-extracted. Applying payment-grade RPO to the ALM engine wastes money; applying ALM-grade RPO to payments loses instructions.

Business continuity is the organisational layer: people, locations, decision rights and communication when the technology answer is inadequate. It covers who is authorised to instruct a payment by telephone when the hub is down, which correspondent relationships have manual fallback procedures, and how the desk operates from an alternate site.

Four Treasury-specific failure scenarios illustrate why generic IT resilience planning is insufficient. Loss of SWIFT connectivity severs both confirmation and payment channels simultaneously, forcing fallback to correspondent telephone and authenticated fax procedures that must exist and be tested rather than assumed. Loss of the payment hub near cut-off is the acute scenario: the bank knows what it needs to pay, cannot instruct it, and has a deadline that no one will move. A failed overnight ALM batch leaves the desk trading blind — positions unknown, ratios unrefreshed, limits unmonitored — and the correct response is usually a defined reduced-risk operating mode rather than business as usual on stale numbers. And a central bank collateral pool that cannot be mobilised in time converts a solvency-adequate bank into a bank that cannot make a payment, which is the precise definition of a liquidity crisis.

The contingency funding plan is the business continuity plan for liquidity itself. Required under CRD and elaborated in EBA guidance, a CFP defines the early-warning indicators that trigger escalation, the actions available at each stage, the collateral pre-positioned and ready to mobilise, the decision rights at each escalation level, and the internal and external communication protocol. The indicators are institution-specific: widening of the bank's own senior spread beyond a threshold, deposit outflow rates above a level, a shortening of wholesale funding tenor, a rating outlook change, a decline in the LCR toward the internal buffer floor above the 100% regulatory minimum.

A CFP that has never been rehearsed is a document, not a plan. The rehearsal reveals what the document conceals: that the collateral pre-positioning takes three days rather than the assumed same day, that the person authorised to draw on a committed facility is on leave with no delegate, that the process for mobilising the central bank pool requires a form that only one team knows how to complete, that the board escalation path assumes a meeting that takes six hours to convene. None of these appear on paper. All of them appear in a live exercise.

DORA turned this from aspiration into obligation. The EU Digital Operational Resilience Act imposes requirements on ICT risk management, incident classification and reporting, digital operational resilience testing including threat-led penetration testing for significant institutions, and oversight of critical third-party ICT providers. The last of these is the pointed one for Treasury: when the ALM engine, the reporting platform, the SWIFT connectivity and the market data all come from external providers, the bank's resilience is a property of its supply chain, and DORA makes that supply chain the bank's explicit responsibility with contractual and exit-strategy requirements attached.

Concept What it guarantees Why it matters at cut-off
Redundancy A duplicate component exists Necessary but insufficient; a cold spare does not meet a deadline
Automatic failover The duplicate takes over without human action within a defined window The only property that survives a failure at 17:20 with an 18:00 cut-off
RPO Maximum acceptable data loss, measured backwards Must be zero for payment instructions; may be hours for ALM inputs
RTO Maximum acceptable outage, measured forwards Must be shorter than the time remaining to cut-off, not shorter than a generic SLA
Business continuity People, sites, authority and communication when technology fails Determines whether a manual payment can lawfully be instructed at all
Contingency funding plan Named triggers, pre-positioned collateral, rehearsed actions Converts a liquidity shock from improvisation into execution
DORA testing regime Documented, supervised resilience testing and third-party oversight Makes the above verifiable rather than asserted

10. Read the regulation as a specification and the system requirements write themselves

A supervisory rule is a functional requirements document written by someone who will audit the implementation. Reading each regime for what it demands of data and systems, rather than for its legal content, produces a build specification directly.

LCR — Regulation (EU) 575/2013 as amended, with the Commission Delegated Regulation specifying the calculation. The specification requires daily calculation capability, which forces a data pipeline that completes reliably every business day rather than a monthly process that is merely run more often. It requires HQLA classification with Level 1, Level 2A and Level 2B categorisation, associated haircuts and the composition caps, which forces a security master carrying eligibility attributes that must be maintained as issuers and ratings change. It requires encumbrance logic capable of excluding pledged assets, which forces the single collateral inventory discussed earlier. And it requires outflow and inflow weights that are configurable rather than hard-coded, because the weights are amended by regulation and a code release should never be the mechanism by which a parameter change is applied. Meridian Bank's 143% LCR against a €62bn buffer (illustrative) is the output of exactly this machinery running without failure every day.

NSFR — the same contract data on a one-year clock. The specification requires available stable funding factors applied to liabilities and capital by residual maturity and counterparty type, and required stable funding factors applied to assets by residual maturity, asset quality and encumbrance duration. What makes it architecturally interesting is that it consumes the same contract records as the LCR but bands them differently — under six months, six to twelve months, over twelve months — and applies encumbrance treatment based on the remaining term of the encumbrance rather than its mere existence. A system that models encumbrance as a boolean flag can compute the LCR and cannot compute the NSFR correctly.

IRRBB — EBA guidelines on the management of interest rate risk arising from non-trading book activities. The specification requires both economic value of equity and net interest income measurement under the six prescribed shock scenarios: parallel up, parallel down, steepener, flattener, short rates up, short rates down. It requires the supervisory outlier test, under which a decline in EVE beyond the prescribed threshold of Tier 1 capital triggers supervisory attention. Systemically, this demands full repricing cashflows rather than maturity cashflows, a documented behavioural model for non-maturing deposits with a cap on the assumed average repricing maturity, prepayment and early-redemption modelling, and calibration evidence that survives challenge. The behavioural model is not an enhancement here — without it, a current-account balance has no repricing date and the calculation cannot be performed at all.

BCBS 248 intraday liquidity monitoring — the requirement most legacy cash systems were never built to meet. The specification asks for daily maximum intraday liquidity usage, available intraday liquidity at the start of each business day, total payments sent and received, and time-specific and other critical obligations. Every one of these is a time-stamped, transaction-level question. A cash system that stores end-of-day balances holds none of the required information, and no amount of aggregation recovers it. This is the single clearest case in Treasury where a regulatory requirement forced a data-capture change rather than a calculation change, and it is why the camt.052 intraday feed and ISO 20022 granularity discussed above are load-bearing rather than cosmetic.

BCBS 239 with ALMM and COREP — the aggregation and lineage regime. The additional liquidity monitoring metrics extend supervisory visibility beyond the two headline ratios: concentration of funding by counterparty and by product type, the contractual maturity ladder, concentration of counterbalancing capacity by issuer or counterparty, prices for various lengths of funding, and rollover of funding. Concentration of funding by counterparty is the metric that makes LEI discipline non-negotiable, because it requires the bank to know that four separately-booked relationships are the same legal entity. The contractual maturity ladder requires contractual dates unmodified by behavioural assumptions, which means the platform must retain both views of every contract rather than overwriting one with the other. BCBS 239's aggregation, timeliness and lineage principles then apply across all of it.

Two further regimes make demands worth naming. MREL reporting under BRRD requires the bank to identify which liabilities are eligible for bail-in, which is an attribute of the instrument's legal terms, ranking and residual maturity — data that lives in the issuance system and must reach the reporting layer intact. EMIR margin requirements generate variation and initial margin flows on non-cleared derivatives, which are a contingent liquidity outflow under stressed market moves and therefore an input to the LCR's derivative outflow calculation and to the contingency funding plan. A bank that cannot project its margin calls under a market shock cannot size its buffer honestly.

  1. The seven functional blocks run at different speeds. Name the fastest, the slowest, and explain why the mismatch causes problems. — Cash and payment management runs in near real time, in minutes, against the payment cut-off. Regulatory reporting runs on a monthly or quarterly calendar, and the ALM engine runs overnight in batch. The mismatch causes problems because a desk needing an intraday answer cannot wait for the overnight run, so it builds a tactical approximation; that approximation then disagrees with the official run, and a permanent reconciliation process is created. Two systems computing the same quantity from different snapshots at different times will always differ.

  2. Why does camt.052 matter more for intraday liquidity than camt.053, despite both being ISO 20022 statements?

  3. A German Bund appears in the liquidity buffer, in the central bank collateral pool and as delivered-out repo collateral. Which is true for LCR purposes, and what architectural fault produced the situation?

  4. Why is redundancy alone insufficient for a payment hub, and what property is actually required?

  5. Why is the FTP policy layer never bought from a vendor even when the calculation engine is?

  6. A supervisor asks Meridian Bank to reproduce the LCR reported eighteen months ago. What must the architecture have preserved?

  7. Why can a system that models encumbrance as a simple yes/no flag compute the LCR but not the NSFR?

  8. What does BCBS 239's adaptability principle demand of the data model, and why do report-extract architectures fail it?

  9. What does a contingency funding plan rehearsal reveal that the written plan does not?

4

  1. Why does camt.052 matter more than camt.053 for intraday liquidity?

  2. A Bund appears in the buffer, the central bank pool and as repo collateral. What is true for LCR?

  3. Why is redundancy alone insufficient for a payment hub?

  4. Which Treasury block runs in near real time?

Part 4B — Knowing the target architecture is the easy half; getting there from the legacy estate is the job

The previous part described what a Treasury platform should look like: one data layer, entity-aware, cashflow-complete, feeding calculation engines that produce ratios, limits and prices consistently. No bank starts from there. Every institution starts from an estate assembled over thirty years by people solving the problem in front of them with the budget available that year. This part is about the distance between the two, and about the organisational machinery — governance, sequencing, sourcing decisions — that determines whether the distance is ever closed or merely surveyed annually.

The legacy estate is the rational outcome of decades of local optimisation

Describe the starting point honestly, because pretending otherwise wastes the first year of any programme. In a bank the size of Meridian — €420bn of assets, several jurisdictions, a history of acquisitions — the Treasury-relevant estate typically looks like this. A general ledger that is authoritative for accounting balances and completely unauthoritative for contractual cashflows, because it records what happened, not what is scheduled to happen. Three or four origination systems for lending, inherited from mergers, with product taxonomies that overlap but do not agree: what one calls a revolving facility another calls a committed line with a drawdown schedule, and the mapping between them lives in a translation table nobody has reviewed since it was written.

The liquidity engine is usually fed by files rather than by an interface. Extracts are produced overnight from each source system, landed on a file server, validated by a set of scripts, and loaded. When a file is late, the calculation is late. When a file is short, somebody notices during the reconciliation, or does not. The engine itself is often a competent piece of vendor software doing a competent job on data it has no way to verify, which is the specific failure pattern that makes ratio production feel simultaneously automated and unreliable.

The ALM system's behavioural assumptions live outside the ALM system. Deposit decay curves, prepayment speeds, deposit beta by product and segment, credit line drawdown factors — these are the parameters that turn contractual cashflows into economic ones, and in a large number of institutions they are maintained in spreadsheets by two or three specialists, reviewed annually, and loaded into the engine as static tables. The people who maintain them are excellent. The arrangement is nonetheless a single point of failure with no version control, no automated lineage and no way to reproduce which parameters were used on a given date twelve months ago.

Payments report end of day, and reporting reconciles by hand. The payment infrastructure produces a statement the next morning, which is why intraday liquidity monitoring — the BCBS 248 subject — is so often reconstructed retrospectively rather than observed live. And the regulatory reporting layer sits at the end, assembling outputs from the liquidity engine, the capital engine, the general ledger and several manual adjustment files, reconciling them in a controlled spreadsheet environment, and producing a submission that is correct because a team of experienced people worked late, not because the pipeline is sound.

None of this is evidence of incompetence, and treating it as such guarantees failure. Every component was a sensible decision at the time. The origination system was bought because it was the best fit for that business line in that year. The file interface was built because the source system had no API and building one would have required a change to a platform owned by a different division with a different budget. The spreadsheet exists because the model changed twice a year and the vendor's release cycle was eighteen months. Each choice optimised locally, under a real constraint, with a real deadline. The estate is the accumulated residue of thousands of such choices, and the way to change it is to understand why each one was made, not to declare the whole thing broken. Programmes that open with a condemnation of the existing estate lose the people who built it, and those people are the only ones who know where the exceptions are.

Figure 22 Current state, target state, and the sequence between them

The order of a Treasury transformation is not negotiable

There is one defensible sequence, and the attractive order is not it. Data layer first: a single, complete, entity-aware store of contracts and their cashflows, with the reference data — counterparty, product, entity, currency, rate basis — that gives them meaning. Then calculation engines on top of that layer: liquidity ratios, NSFR, ALM measures, FTP, stress scenarios, all reading the same positions. Then reporting and analytics, drawing from engine outputs that are already consistent. Then real-time and self-service capability, which is only safe once the layers beneath it are trustworthy.

Defend the first step, because it is the one that gets cut. The data layer delivers nothing visible. It produces no dashboard, no new report, no capability a business user can demonstrate to a steering committee. Its entire value is that everything built afterwards is cheaper, faster and consistent — a benefit that arrives eighteen months later and is attributed to whatever was built on top. This is why it is under-funded in almost every institution, and why the same institutions rebuild their reconciliation logic four times in a decade.

The attractive order starts with a dashboard, and the reason is political rather than technical. A dashboard is demonstrable. It can be shown to the executive committee in month four. It generates the sense of momentum that keeps a programme funded. And a real-time dashboard sitting on an unreliable data layer is worse than no dashboard, because it delivers confident wrong answers faster and to more people. The old process was slow and manual, but the slowness contained a control: an analyst looked at the numbers, recognised the ones that looked odd, and investigated. Automate the pipeline without fixing the inputs and you have removed the human check while preserving the defect, and you have simultaneously increased the audience's confidence in the output. Institutions discover this when a decision is taken on a figure that turns out to have double-counted a portfolio migrated from a decommissioned system.

Phase What it delivers What it unblocks Indicative duration Main failure mode
1. Data layer Entity-aware contract and cashflow store; reference data model; golden source designations; measured data quality controls Everything downstream; single reconciliation point; ability to answer entity-level questions 18–30 months (indicative) Under-funded because invisible; scope creep into a full enterprise data programme; no business owner for definitions
2. Calculation engines LCR, NSFR, ALM, FTP, stress and scenario engines reading common data Consistent numbers across risk, finance and Treasury; ability to explain differences 12–24 months (indicative), overlapping phase 1 Engines implemented against a data layer that is not yet complete, forcing local workarounds that become permanent
3. Reporting and analytics Regulatory submissions, ALCO packs, management reporting from engine outputs with lineage Retirement of manual reconciliation; auditable submission chain 9–18 months (indicative) Preserving legacy report formats verbatim, which re-imports the old data model into the new stack
4. Real-time and self-service Intraday liquidity view, on-demand scenario running, business self-service query Genuine intraday management; faster ALCO cycles; reduced central bottleneck 12+ months (indicative), continuous thereafter Started first; or started before data quality controls are measured, producing fast wrong answers

Overlap is allowed; inversion is not. Phases can run concurrently where dependencies permit — an engine can be built against an agreed data contract before the full data layer is populated, provided the contract is real and enforced. What cannot work is delivering phase 4 capability against a phase 0 data estate and calling it a quick win. The honest framing for a steering committee is that the early phases buy optionality, and that the demonstrable outputs come later but come reliably. Programmes that cannot make that argument successfully tend not to survive their second budget cycle, which is itself useful information about the institution's appetite.

Data governance is an organisational design problem wearing a technical name

The roles are simple to state and hard to fill. A data owner is accountable for the definitions and the quality of a data domain — deposits, derivatives, counterparties, entity hierarchy. This is always a business role, held by someone with the authority to decide what a term means and the budget to fix it when it is wrong. A data steward is the operational custodian: monitors quality metrics, investigates breaks, coordinates remediation, maintains the documentation. Stewards can sit in technology or operations. Owners cannot, because the questions they resolve are business questions — is an overdraft facility a committed line, does an internal transfer count as a deposit inflow, which entity owns a syndicated participation — and technology has no standing to answer them.

A golden source is a decision, not a discovery. For each attribute, one system is designated authoritative; every other system that holds the same attribute becomes a consumer, not a peer. Counterparty identity comes from the client reference system and nowhere else. Contractual cashflows come from the origination system. Accounting balances come from the ledger. Once designated, differences are not reconciled as equals — they are treated as defects in the consumer.

The hardest part is not the technology; it is that the losing system has users who preferred it. When two systems disagree about a counterparty's legal entity, the resolution is not a matching algorithm. It is a decision that one of them is wrong, taken by someone with the authority to impose it on a business area that has been using the other system happily for a decade and gets a better answer for its own purposes. That business area will produce reasonable objections — their version is more granular, more current, better aligned to how they actually manage the relationship — and some of those objections will be correct. The decision still has to be made and enforced, because an institution with two authoritative sources has none. Governance that cannot make this decision produces a data catalogue, a glossary and a reconciliation team, which is data governance theatre.

Data quality has to be measured, not aspired to. Four dimensions carry most of the weight in a Treasury context. Completeness: what proportion of expected records arrived, measured against a known expected count, not against yesterday's. Validity: what proportion conform to the defined domain — dates that parse, currencies that exist, entity codes in the approved hierarchy, maturities after origination dates. Timeliness: what proportion arrived before the cut-off needed for the downstream calculation. Consistency: whether the same attribute agrees across systems where it should, and whether cross-field logic holds.

Each dimension needs a metric, a threshold, and a named person who is called when it breaches. "We monitor data quality" is not a control. "Completeness of the derivative cashflow feed is measured as records received divided by records expected, threshold 99.9%, breaches page the domain steward within thirty minutes and are reported to the data owner daily" is a control. The difference is testable, and an auditor or supervisor will test exactly that difference.

BCBS 239 is at least half an organisational specification, and that is why it is still unmet. The Basel Committee's principles for effective risk data aggregation and risk reporting, issued in 2013, cover governance and infrastructure, data aggregation capability, reporting practices and supervisory review. Read the governance principles closely: they require board-level accountability for risk data, ownership of definitions, independent validation of aggregation processes, and the ability to aggregate rapidly and accurately across the group including in stress. Very little of that is a technology deliverable. It is a specification for how an institution decides things and who is answerable when a number is wrong.

Which explains the supervisory record. More than a decade after issuance, the Basel Committee and the ECB have repeatedly reported that compliance remains materially incomplete at large institutions, and the ECB has made risk data aggregation a recurring supervisory priority, with targeted reviews and, in some cases, enforcement. The technology has improved enormously over that period. The reason the gap persists is that BCBS 239 asks institutions to establish clear ownership of contested definitions and to resolve them permanently, and that is an exercise in organisational authority that no platform purchase delivers.

Behavioural models are software, and should be run with an engineer's discipline

Start with what a behavioural model is. Contractual cashflows are what the documents say: a five-year bond pays coupons and repays principal on stated dates. Behavioural cashflows are what actually happens: current account deposits with no contractual maturity stay for years, mortgages prepay when rates fall, revolving facilities get drawn when a borrower is under stress, deposit rates follow policy rates with a lag and a partial pass-through. A behavioural model is the estimated mapping from contract to expected behaviour — deposit decay, prepayment speed, deposit beta, drawdown factor — and it sits directly underneath the LCR, the NSFR, interest rate risk in the banking book, FTP and the funding plan. Change the deposit decay assumption and the reported NSFR moves. These are among the highest-leverage parameters in the institution.

Treat their lifecycle as an engineering lifecycle, because that is what it is. Development, with documented methodology, data used and assumptions made. Independent validation — review by a function separate from and not reporting to the developers, consistent with supervisory expectations on model risk management, testing conceptual soundness, data quality, implementation correctness and outcomes. Version control on code, parameters and documentation together. Controlled release, with the same change management as any production software: approval, testing in a representative environment, defined rollback. Ongoing performance monitoring against defined tolerances. Backtesting against realised behaviour — did deposits actually decay at the modelled rate. And periodic recalibration on a fixed cycle, with explicit triggers for out-of-cycle review when the environment breaks the model's assumptions.

The trigger design deserves attention, because the 2022–2023 rate cycle exposed it. Deposit models calibrated across a decade of near-zero rates encoded a world in which deposits were insensitive to rate changes, because rates did not change. When policy rates rose rapidly, realised deposit beta and outflow behaviour departed sharply from the modelled path at many institutions. An annual recalibration cycle discovers that eleven months late. A trigger framework — recalibrate out of cycle when realised behaviour breaches a defined tolerance for a defined number of consecutive periods, or when a specified macro variable moves beyond a specified range — discovers it in weeks. Write the triggers when the model is built, not when it breaks.

Model version belongs in the data lineage, and this is the point most institutions miss. Reproducing a ratio reported eighteen months ago requires two things: the data as it stood at that date, and the model as it stood at that date. Institutions invest heavily in the first and neglect the second. When a supervisor asks why the LCR reported in a given quarter differs from what the current system produces from the same data, the answer is usually "the deposit model was recalibrated in the interim" — and if the prior model version and its parameters are not retrievable, that answer cannot be demonstrated. A reported number that cannot be reproduced is a control finding, and it is one of the more common ones.

Challenger models are how you find out whether the main model is any good. A challenger is an alternative specification run in parallel against the same data, producing an alternative answer that is compared to the champion's. It might be a simpler model, a different functional form, or a machine-learning approach used specifically as a benchmark. The value is not that the challenger replaces the champion; usually it does not. The value is that persistent, systematic divergence tells you something the champion is missing, and that a model which has never been compared to an alternative has never really been tested — its errors are invisible because there is nothing to see them against. Validation functions increasingly expect a challenger for material models, and the expectation is reasonable.

Cloud in a regulated Treasury context: a genuine case and genuine constraints

The real case is elastic compute, and it is a strong one. Treasury and risk calculation demand is spiky by nature. Full stress scenario suites, Monte Carlo simulations for behavioural models, historical revaluation across long lookback windows, ad hoc reverse stress testing requested by ALCO on a Tuesday afternoon — these consume enormous compute for hours and nothing for days. On-premise capacity has to be sized for the peak, which means it is largely idle, and the idle capacity is capitalised, powered and depreciated regardless. Elastic compute matches cost to actual demand, and it changes what is feasible: a scenario suite that takes fourteen hours on fixed capacity, and is therefore run monthly, can be run in forty minutes and therefore run daily. That is a capability change, not a cost optimisation, and it is the argument worth making.

The constraints are real and should not be minimised. Data residency: some jurisdictions require certain categories of data to remain within national borders, and confidentiality obligations under banking secrecy laws restrict what may be processed by a third party. The EBA Guidelines on outsourcing arrangements require institutions to maintain a register of outsourcing arrangements, to notify supervisors of arrangements covering critical or important functions, to conduct due diligence proportionate to the risk, to secure audit and access rights for both the institution and the supervisor, and to document an exit strategy. These obligations apply whether or not the arrangement is labelled "cloud".

DORA raised the bar and shifted part of the oversight. The Digital Operational Resilience Act establishes a harmonised framework for ICT risk across the EU financial sector: ICT risk management requirements, incident classification and reporting, digital operational resilience testing including threat-led penetration testing for significant entities, and a register of information on all ICT third-party arrangements. Its most structurally significant feature is the oversight framework for critical ICT third-party providers — designated providers, which in practice includes the largest cloud infrastructure and software providers, become subject to direct oversight by a lead overseer at European level, with powers to conduct inspections and issue recommendations. That is a recognition of concentration risk: when a small number of providers underpin a large share of the sector's infrastructure, the failure or compromise of one is a systemic event, and a framework built only on individual institutions managing their own vendors does not address it.

Exit strategy is where most arrangements are weakest. DORA and the EBA guidelines require contractual arrangements for critical functions to include exit provisions and require institutions to maintain exit plans that are tested. A credible exit plan has to answer specific questions: to what — another provider, on-premise, a hybrid arrangement; over what period, with the transition milestones identified; at what cost, including data egress charges, parallel running, re-engineering of provider-specific services, and the staff effort; with what data, including how many terabytes move by what mechanism and how integrity is verified; and under what trigger, distinguishing an orderly commercial exit over eighteen months from a stressed exit following provider failure or a supervisory instruction.

Build versus buy, decided by a rule rather than by preference

Use a rule that can be applied by someone who was not in the room. Buy where the requirement is defined by an external rule that applies identically to every institution and changes on someone else's schedule. Regulatory calculation engines, reporting submission and taxonomy handling, payment message formats and standards implementations: the specification is public, identical across the industry, and revised by an authority on its own timetable. Every bank building its own is paying separately to track the same rule changes, and the vendor's entire business model is amortising that tracking cost across a client base.

Build where the requirement encodes an institution-specific choice. The FTP policy layer expresses how this institution wants to allocate funding cost and drive behaviour, and no two frameworks are identical. The limit framework encodes this board's risk appetite and escalation structure. Management reporting reflects how this executive committee thinks. Capital and liquidity allocation logic embeds negotiated internal agreements. Buying these means adopting the vendor's model of how a bank should be run, then customising it into something the vendor never intended, at a cost that recurs at every upgrade.

Integrate where the requirement is a commodity with a network effect. Market data, payment network connectivity, trading venue access, benchmark rates: the value comes from connecting to the network everyone else is on, and there is no version of building it that produces an advantage.

Requirement type Decision Reason Hidden cost
Regulatory ratio calculation (LCR, NSFR, ALMM) Buy Rules are external, identical across institutions, and revised on the regulator's schedule; vendor amortises change tracking Vendor data model may not represent your products; mapping layer becomes permanent and needs its own governance
Regulatory reporting submission and taxonomy Buy Taxonomies change frequently; submission mechanics are pure compliance with no differentiation available Version lock-step with the vendor's release cycle; late vendor releases become your late submission
FTP policy and rate assignment logic Build Encodes an institution-specific policy choice about behaviour and allocation that no vendor default matches Permanent ownership of the logic; needs documented policy governance or it drifts into undocumented local rules
Limit framework and escalation Build Expresses this board's risk appetite and this institution's governance structure Integration burden with every position source; must be maintained as the entity and product structure changes
ALM and behavioural modelling engine Buy the engine, build the models Numerical machinery is generic; behavioural assumptions are institution-specific and must be owned internally Vendor engines vary widely in whether custom models can be injected as first-class objects; check before signing, not after
Market data and reference rates Integrate Commodity with network effects; no differentiation from building Licensing terms restricting redistribution and derived use inside the group; often discovered during an audit
Payment connectivity and message standards Integrate Network effect is the entire value; standards are external Format changes propagate through every downstream consumer that parsed the old format
Management and ALCO reporting Build on bought data Reflects institution-specific decision-making; must change as the conversation changes Proliferation without governance produces four hundred reports and no agreed hierarchy of measures

Total cost of ownership, stated honestly on both sides. The buy case is usually presented as licence plus implementation. The number that matters is licence, plus implementation, plus the recurring cost of carrying every customisation across every upgrade, plus the integration and mapping layer that translates your data model into the vendor's, plus the constraint that your release calendar is now partly the vendor's. Institutions that customise heavily discover that the upgrade they deferred for three years has become a re-implementation.

The build case is equally under-stated. Build cost is not the initial development. It is development, plus the permanent obligation to track regulatory change yourself — reading the consultation papers, interpreting the final text, implementing and testing before the effective date, every time — plus the key-person risk that accumulates around a system three people understand, plus the documentation and validation the supervisor will eventually ask for. A five-person team maintaining a built regulatory engine is a recurring annual cost that persists for as long as the engine does, and it should be in the comparison from the first slide.

ISO 20022 is the worked example of a multi-year infrastructure change

Explain the standard from zero. ISO 20022 is a message standard for financial communication using structured XML with a defined data dictionary, replacing the older, terse, positional formats — MT messages in the correspondent banking world — that were designed when bandwidth was expensive and fields were scarce. The practical difference is data richness and structure. Where a legacy payment message carried a free-text remittance field of limited length into which anything might be typed, an ISO message carries structured, separately identified elements: fully specified debtor and creditor identities and addresses, structured remittance information linking to invoices, purpose codes, and unambiguous currency and amount representation.

The migration is a genuine multi-year programme with a coexistence period. Cross-border payments and reporting have been migrating across the industry, with high-value payment systems including TARGET2 and its successor T2 in the euro area, and equivalents elsewhere, moving to the standard, and a coexistence window during which both old and new formats circulate. Coexistence is the expensive part: every institution has to handle both, translate between them, and reconcile outputs across the boundary — meaning that for the duration, the estate is more complex, not less, and the benefits have not yet arrived. Programmes lose sponsorship precisely in this window.

The temptation that destroys the value is internal translation back to the legacy model. The cheapest way to become compliant is to accept ISO messages at the gateway, map them into the internal legacy format that every downstream system already consumes, and process as before. It works. Messages flow. The programme is declared complete on time. And the structured data — the very thing the standard exists to deliver — is discarded at the gateway, because the internal format has nowhere to put it. The institution has paid the full cost of the migration and retained the old data model, which means every future capability that depends on rich payment data requires the programme to be done again.

The Treasury prize is specific and worth naming. Structured, richly attributed payment data arriving intraday is what makes genuine intraday liquidity measurement possible. BCBS 248 monitoring tools require institutions to measure daily maximum intraday liquidity usage, available intraday liquidity at the start of the day, total payments, and time-specific obligations. Doing that properly requires knowing what each payment is, whose it is, when it must settle, and which entity and currency it affects — attributes that ISO carries as structured fields and legacy formats largely did not. The same data makes automated reconciliation, cash forecasting for corporate clients, and payment purpose analytics feasible. All of it depends on the data surviving past the gateway.

  1. Why does starting a Treasury transformation with a real-time dashboard tend to fail?

  2. What makes the golden source decision difficult, and what happens if governance cannot take it?

  3. Give a specific data quality control that would satisfy an auditor, and one that would not.

  4. Why must the model version be part of data lineage?

  5. What does a credible cloud exit plan have to demonstrate that most do not?

  6. Apply the build-versus-buy rule to an FTP engine and to an LCR engine, and say why they differ.

  7. How should an ISO 20022 migration be judged?

5Treasury risk is the risk that the bank runs out of time

A bank does not fail because its assets are worth less than its liabilities. It fails because on a particular morning, at a particular hour, it cannot deliver a particular amount of a particular currency to a particular counterparty. Solvency is a statement about a balance sheet; liquidity is a statement about a clock. Every risk that Treasury manages — funding, market, interest rate, collateral, model, settlement — reduces to the same question: when the obligation falls due, will the cash be there? This part builds the risk taxonomy from zero, goes deep on the interest rate risk that sits at the heart of asset-liability management, sets out how to distinguish a well-run Treasury from a lucky one, dissects three real failures that each broke in a different place, and then examines what technology — including generative AI — actually changes.

1. Six risks, and not one more: a taxonomy that leaves no gaps

A risk taxonomy is only useful if it is MECE — mutually exclusive and collectively exhaustive. Mutually exclusive means no risk appears twice under different names, because duplication creates the illusion of coverage and the reality of confusion about who acts. Collectively exhaustive means nothing falls between the categories, because a risk without a category is a risk without an owner, and a risk without an owner is a control gap waiting to be discovered by a supervisor or by an event. Treasury risk decomposes cleanly into six types. Everything a Treasurer worries about belongs to one of them.

Funding liquidity risk is the inability to meet obligations as they fall due without incurring unacceptable losses. The bank has promised to return deposits on demand, repay maturing wholesale debt on its maturity date, honour drawn commitments under corporate credit lines, and post variation margin when derivatives move against it. Funding liquidity risk is the risk that the cash to honour those promises is not available at an acceptable price, or at any price. The mechanism by which it bites is a mismatch between the contractual or behavioural speed at which liabilities can leave and the speed at which assets can be turned into cash. Under the EU framework, this is the risk that the Liquidity Coverage Ratio (LCR) and the Net Stable Funding Ratio (NSFR) under CRR are designed to constrain, and it is the risk that the Internal Liquidity Adequacy Assessment Process (ILAAP) requires the bank to assess in its own terms. The Group Treasurer owns it. The main defence is a buffer of high-quality liquid assets (HQLA) large enough, and a funding profile diverse and long enough, to absorb outflows for long enough that management can act.

Market liquidity risk is the risk that an asset cannot be sold or repo'd at or near its modelled value. Funding liquidity risk assumes that the liquidity buffer converts into cash at the haircut the model assumed; market liquidity risk is the risk that this assumption fails. A covered bond that trades with a two-basis-point bid-offer in normal conditions may show a forty-basis-point spread and no depth in stress. The mechanism is that market makers withdraw exactly when everyone needs to sell the same instrument, so realised proceeds fall short of modelled proceeds precisely when the shortfall matters. This is why the LCR's HQLA definition is deliberately narrow and why Level 2B assets carry haircuts of 25 to 50 percent. The Treasurer owns it jointly with the head of the liquidity portfolio; the main defence is conservatism in what counts as buffer, genuine periodic monetisation testing rather than assumed monetisation, and central bank eligibility as a backstop route.

Interest rate risk in the banking book — IRRBB — is the risk that changes in interest rates damage either the economic value of the balance sheet or its earnings. The banking book holds loans and deposits that are not marked to market, so the risk is invisible in the profit and loss until it crystallises through the interest margin or through a forced sale. The mechanism is repricing mismatch: assets and liabilities reprice at different times and to different reference rates, so a rate move changes the spread between what the bank earns and what it pays. The EBA Guidelines on IRRBB and the associated regulatory technical standards under CRD govern it. ALCO owns it as a body; ALM within Treasury runs it day to day. The main defence is a measurement framework that uses both an economic value lens and an earnings lens, plus a hedging programme executed through the interest rate swap market. Section 2 treats this in full.

Collateral and intraday liquidity risk is the risk of failing to have the right asset, in the right place, at the right minute. A bank can hold ample collateral in aggregate and still fail a margin call because the eligible securities sit in the wrong custodian, the wrong currency, or the wrong legal entity, and the movement takes longer than the deadline allows. Intraday liquidity risk is the same failure in cash: the bank's end-of-day position is comfortable, but at 11:00 the payment system requires a settlement the bank cannot fund without central bank credit. The mechanism is operational and legal frictions — settlement cycles, cut-off times, ring-fenced entities, collateral eligibility schedules — colliding with a hard deadline. BCBS 248 sets the monitoring framework for intraday liquidity; the ECB and national supervisors examine it directly. Treasury operations and the collateral management function own it. The main defence is real-time visibility of positions, pre-positioned collateral at the central bank, and buffers held specifically for intraday use rather than counted twice against the LCR.

Model risk is the risk that a model used to measure or price Treasury exposures is wrong in a way that matters. Every liquidity and ALM number depends on behavioural assumptions: how long non-maturing deposits stay, how fast mortgages prepay, how much of an undrawn corporate line is drawn in stress, what fraction of a wholesale maturity rolls. These are estimated, not observed. FTP model risk is a specific and consequential subtype — if the funds transfer pricing curve misprices a tenor or a product, the bank systematically over-produces the business it under-charges, and the error compounds silently through origination volumes rather than announcing itself in a report. The mechanism is that a wrong assumption produces a right-looking number, so the error is invisible until reality tests it. Independent model validation, under the second line of defence, owns the challenge; the model owner in Treasury owns the model. The main defence is documented assumptions, back-testing against realised behaviour, benchmarking against alternative specifications, and sensitivity analysis that shows the board how much the answer moves when the assumption moves.

Operational and settlement risk is the risk that the process breaks even though the position was right. A payment instruction fails, a system is unavailable at cut-off, a cyber incident takes down the payments interface, a key person is unreachable, a data feed arrives corrupted and the overnight batch produces a liquidity report that nobody can rely on. The mechanism is that Treasury operates on daily, irreversible deadlines with no natural retry — a missed settlement is a public event that counterparties observe and price. The Chief Operating Officer for Treasury owns it, with the operational resilience function. Under DORA, the EU's Digital Operational Resilience Act, this now carries explicit obligations around ICT risk management, incident reporting, testing and third-party oversight, with critical payment and market infrastructure dependencies squarely in scope. The main defence is redundancy, tested manual fallback procedures, named escalation with named substitutes, and scenario testing that includes the unavailability of the systems the bank uses to detect problems.

Risk type What it is How it bites Who owns it Main defence
Funding liquidity Inability to meet obligations as they fall due at acceptable cost Liabilities leave faster than assets convert Group Treasurer HQLA buffer, diversified and termed-out funding, ILAAP
Market liquidity Buffer assets cannot be monetised at modelled value Bid-offer widens and depth vanishes in stress Treasurer / liquidity portfolio head Narrow HQLA definition, monetisation testing, central bank eligibility
IRRBB Rate moves damage economic value or earnings Repricing mismatch between assets and liabilities ALCO, run by ALM Dual EVE/NII measurement, hedging programme, limits
Collateral & intraday Right asset or cash unavailable at the required minute Operational and legal frictions meet hard deadlines Treasury operations / collateral management Real-time visibility, pre-positioned collateral, dedicated intraday buffer
Model (incl. FTP) A measurement or pricing model is wrong in a material way Wrong assumptions produce right-looking numbers Model owner, challenged by validation Documented assumptions, back-testing, benchmarking, sensitivity analysis
Operational & settlement Process fails though the position was correct Daily irreversible deadlines with no retry Treasury COO / operational resilience Redundancy, tested fallback, named escalation, DORA-aligned testing
Figure 23 Six risks, six owners — unowned risk is a control gap

2. IRRBB requires two lenses because one lens will always flatter you

Interest rate risk in the banking book is measured through two views that answer different questions and can disagree violently. The first is economic value of equity (EVE), a present-value measure. Take every future cashflow on the banking book — asset and liability — discount them at current rates, subtract liabilities from assets, and you have the economic value of equity. Now shock the rate curve and repeat. The change in that value is the EVE sensitivity, and it answers the question: if the bank had to be valued today at new rates, how much of its equity would the rate move destroy? It is a full-horizon measure, capturing everything out to the last cashflow of the longest mortgage.

The second is net interest income (NII), an earnings measure over a short horizon, normally twelve months. Project the interest the bank will earn and pay over the next year under current rates, then reproject under the shocked curve. The difference is the NII sensitivity, and it answers: what does this rate move do to next year's profit? NII is short-horizon and cash-based, ignoring anything beyond the projection window. It is the number the CFO cares about because it lands in the income statement; EVE is the number the risk function cares about because it captures the whole balance sheet's exposure.

The two can point in opposite directions, and the reason is the most instructive fact in ALM. Consider the treatment of non-maturing deposits — current accounts and savings accounts with no contractual maturity, which the customer can withdraw on demand but in practice mostly does not. The bank must assume a behavioural life. Assume that life is short, say one year, and the deposits are modelled as repricing quickly: NII becomes volatile because when rates fall the bank is assumed to pass the fall through immediately and lose margin, but EVE sensitivity is small because the liability has little duration to lose value. Assume the life is long, say five years, and the mirror image appears: NII looks stable because the deposit is treated as a fixed-rate liability that does not reprice, so falling rates do not compress margin — but EVE sensitivity rises sharply, because the bank is now carrying a long-duration liability whose modelled value moves substantially with rates.

The same modelling choice therefore reads as prudent through one lens and reckless through the other. Lengthening deposit life stabilises reported earnings and lets the bank invest longer, improving margin. It also builds a larger economic exposure that only appears in the EVE number, and only crystallises if the assumption breaks — if depositors turn out to be faster than the model assumed. This is not a technicality. It is the mechanism by which a bank can report stable earnings for years while accumulating an exposure that will be revealed only under stress. Both lenses are required precisely because each one is blind where the other sees.

The six EBA shock scenarios exist because a parallel shift is not how curves move

European supervisory expectations require IRRBB measurement under six prescribed interest rate shock scenarios, not one. A single parallel shift would miss the risks that arise from the shape of the curve rather than its level, and shape risk is where much of the banking book exposure actually lives. The six are: parallel up, parallel down, steepener (short rates down, long rates up), flattener (short rates up, long rates down), short rates up, and short rates down. Shock magnitudes are currency-specific and calibrated by the standards; for the euro the parallel shock has been calibrated at 200 basis points, with short-rate shocks larger and long-rate shocks smaller, and post-shock rates subject to a floor that rises with tenor.

Two supervisory outlier tests attach to these scenarios and function as tripwires. The EVE outlier test: a decline in economic value of equity exceeding 15 percent of Tier 1 capital under the worst of the six scenarios triggers supervisory attention. The NII outlier test, introduced through the more recent standards: a large decline in net interest income under the parallel up or parallel down scenario relative to Tier 1 capital triggers the same. Breaching an outlier test is not automatically a breach of law. It obliges the supervisor to investigate and, under the Supervisory Review and Evaluation Process, potentially to impose a capital add-on or require the bank to reduce the exposure. Banks manage deliberately inside these thresholds, and a bank sitting persistently just below one should expect to be asked why.

Figure 24 One deposit assumption, two opposite verdicts

Replicating portfolios turn a behavioural guess into a tradeable instruction

A replicating portfolio is the technique that makes non-maturing deposits manageable. The problem is concrete: the bank has €180 billion of retail deposits with no maturity date and an administered rate the bank itself sets. It cannot hedge an instrument with no defined cashflows. The replicating portfolio solves this by representing the deposit book as a portfolio of fixed-term instruments whose combined behaviour — in rate sensitivity and in balance stability — best matches the observed historical behaviour of the deposits. A typical construction might allocate the balance across overnight, one-year, three-year and five-year tranches, rolling a fixed proportion each month. The result is a synthetic maturity profile with a weighted average life, and that profile can be hedged, transfer-priced and reported.

Constructing it requires splitting the balance into a core and a volatile component. The core is the portion that historical data suggests remains regardless of rate moves or seasonality; the volatile portion is what comes and goes. The core is replicated long, the volatile portion short. Calibration uses regression of deposit balances and deposit rates against market rates over a historical window, plus stability analysis of the balance series, and is constrained by supervisory caps — the EBA standards impose a cap on the average behavioural life of non-maturing deposits, differentiated by counterparty type, with retail transactional deposits treated as the stickiest and wholesale deposits capped much shorter.

The replication tenor is the single most consequential and least verifiable assumption in the entire bank. Consequential, because it determines the transfer price paid to the retail network for deposits, the duration the bank runs, the reported EVE sensitivity, the reported NII sensitivity and the size of the hedge programme. Move the assumed average life of €180 billion of deposits from three years to four and you have changed the bank's interest rate position by an amount comparable to a large hedging decision — without executing a single trade. Least verifiable, because the assumption is a claim about how depositors would behave in conditions that have not occurred. Back-testing confirms the model against the past. It cannot confirm it against a future that differs from the past in the relevant way.

3. Six KPIs separate a well-run Treasury from a fortunate one

A Treasury that has not been tested looks identical to one that would survive testing, unless you measure the right things. Regulatory ratios are necessary but insufficient — they are standardised, backward-calibrated and gameable at the margin. The metrics that actually discriminate combine regulatory outputs with internal measures whose whole purpose is to reveal what the standardised ratios smooth over.

Survival horizon under combined stress is the number that answers the board's real question. It is the number of days the bank can meet all obligations, using its buffer and its contractual inflows, under a scenario that combines an idiosyncratic shock — a rating downgrade, a name-specific loss of confidence — with a market-wide shock, in which wholesale markets close to everyone. "Combined" is the load-bearing word: banks fail in exactly the conditions where their own problem coincides with a market that has no appetite to solve it. A well-run bank targets a survival horizon well beyond the LCR's 30-day construct — three to six months is a common internal ambition — and knows how the number decomposes by currency and by legal entity, because a group-level horizon of ninety days is meaningless if the dollar horizon is eleven.

LCR and NSFR headroom, and the deliberate management of that headroom, is the second. The requirement is 100 percent for both under CRR. What matters is not the level but the intent behind it and the stability of the gap. A bank running an LCR of 143 percent should be able to explain, in a sentence, why 143 and not 115 — pipeline of committed lending, a maturity concentration in the next quarter, currency mismatch, supervisory expectation, rating agency comfort. If the answer is "that is where it came out," the buffer is not being managed.

Funding concentration is the third, and it must be measured along four axes simultaneously. Largest single counterparty as a share of total funding; largest single day of contractual maturities; reliance on any one instrument type (covered bonds, senior preferred, commercial paper, central bank facilities); and reliance on any one currency relative to the currency of the assets it funds. Each axis has produced failures on its own. A bank with a diversified investor base but a single €4bn maturity date has a concentration problem that a Herfindahl index across investors will not show.

Weighted average maturity of wholesale funding, and the shape of the maturity wall, is the fourth. The average alone hides the risk. Two banks can both report a 3.2-year weighted average maturity while one has smooth quarterly redemptions and the other has 40 percent of its wholesale book maturing in a single eighteen-month window. The second bank is fine until that window opens in a closed market. The metric to watch is the largest rolling twelve-month redemption as a percentage of total wholesale funding, alongside the average.

Cost of funds against a defined peer group and against the bank's own secondary curve is the fifth. Peer comparison answers whether the market prices this bank's credit in line with comparable institutions. The comparison against the bank's own secondary levels answers something sharper: whether primary issuance is being executed efficiently. A new issue concession of five basis points on a €1bn deal is €500,000 a year for five years — €2.5m of value on a single transaction, which is the kind of number that justifies the entire debt capital markets execution function.

Intraday peak liquidity usage against available intraday capacity is the sixth, and the most neglected. BCBS 248 requires monitoring of daily maximum intraday liquidity usage, available intraday liquidity at the start of the day, total payments, and time-specific obligations. The discriminating metric is peak usage as a percentage of available capacity, tracked as a distribution rather than an average. A bank whose peak usage touches 85 percent of capacity on the busiest days of the month has far less room than one averaging 40 percent, and the average will not reveal it.

FTP dispute volume is a governance health metric, and it belongs on the list. Count the number of formal challenges from business lines to the transfer prices they receive, and the number that are upheld. Near-zero disputes usually means the businesses have concluded that challenging is futile, or that the prices are not visible enough to challenge — either way, the framework has stopped being contested and has therefore stopped being validated. A high and rising dispute rate with a low upheld rate means the methodology is poorly explained. A moderate dispute rate with a meaningful upheld rate means the framework is under genuine scrutiny and improving, which is the healthy state.

KPI What it measures What good looks like How it can be gamed
Survival horizon (combined stress) Days of obligations coverable under idiosyncratic + market stress 90–180 days, decomposed by currency and entity Soften the scenario; assume secured funding stays open; net across entities that cannot transfer
LCR / NSFR headroom Buffer above the 100% requirement and the intent behind it A stated target with a stated rationale, moving with the balance sheet Window-dress at month-end; classify volatile deposits as operational
Funding concentration Dependence on any counterparty, day, instrument or currency No single counterparty above a low single-digit share; smooth maturity ladder Report only the investor axis; measure at group level where entity level is the constraint
WAM and maturity wall Average tenor and the shape of redemptions Stable average with no rolling 12-month cluster above ~20% of wholesale Issue very long tail deals to lift the average while the near wall is untouched
Cost of funds vs peers and own secondary Credit pricing and execution efficiency Spread to peers explained by rating and franchise; concession within a few bp Choose a flattering peer set; quote reoffer and ignore secondary performance
Intraday peak vs capacity Headroom in the payment system on the worst day Peak usage well below capacity on the busiest days, tracked as a distribution Report the average rather than the peak; exclude time-specific obligations
FTP dispute volume and upheld rate Whether the pricing framework is genuinely contested Moderate volume, meaningful upheld rate, documented resolution Discourage escalation; resolve informally so nothing is recorded

Survival horizon: the number the board actually asks for

4. Three real failures, three different places to break

Silicon Valley Bank broke on the asset side, and concentration did the rest

Silicon Valley Bank failed in March 2023 in what became, at the time, one of the largest US bank failures on record, with approximately $209bn of assets at the point of failure according to the public record. The mechanism was a textbook combination of interest rate risk in the banking book, deposit concentration and modelling failure — a combination that regulators had described in guidance for years.

The asset side carried unhedged duration. During the deposit surge of 2020 to 2021, SVB invested a very large share of the inflow into long-dated fixed-rate securities, heavily agency mortgage-backed securities, with a substantial portion classified as held-to-maturity. Held-to-maturity accounting carries securities at amortised cost, so unrealised losses do not flow through the income statement or, for most banks under the applicable US rules at the time, through regulatory capital. As the Federal Reserve raised rates by roughly 500 basis points from March 2022, the market value of that portfolio fell sharply while the reported book value did not. Public filings and the subsequent Federal Reserve review indicated unrealised losses of roughly $15bn on the held-to-maturity portfolio by late 2022 — a figure of the same order as the bank's tangible common equity. The bank had, by the public record, substantially unwound or not maintained interest rate hedges on this exposure.

The liability side was extraordinarily concentrated. SVB's depositor base was dominated by venture-backed technology and life sciences companies — one industry, one geography, one funding cycle, and a network in which the depositors shared investors, board members and communication channels. Critically, the overwhelming majority of deposits exceeded the $250,000 FDIC insurance limit; the widely cited figure is that roughly 90 percent or more of deposits were uninsured. Uninsured depositors have every incentive to run first and ask questions later, because they bear loss directly. A deposit behaviour model calibrated on the stickiness of an insured retail base has no application to this population.

Regulatory perimeter and speed completed the failure. Following the 2018 tailoring of US prudential rules, SVB's size category was not subject to the full LCR requirement that applies to the largest institutions, and it was subject to lighter liquidity stress testing than a globally systemic bank. When the bank announced a securities sale crystallising a loss and a capital raise on 8 March 2023, the signal was interpreted immediately across a tightly connected depositor network. Withdrawal requests on 9 March were reported at approximately $42bn in a single day, with a further very large amount queued for the following morning. The bank was closed on 10 March. No liquidity buffer sized on a thirty-day standardised outflow assumption survives a quarter of the deposit base leaving in one business day.

The mapping is exact. IRRBB: long-duration assets funded by liabilities modelled as long but behaviourally overnight, with hedges removed. Deposit modelling: behavioural assumptions applied to a population for which they had no evidential basis. Concentration: single industry, single depositor type, overwhelmingly uninsured. Every one of these was measurable in advance.

Credit Suisse met its ratios and failed anyway

Credit Suisse was acquired by UBS in a state-brokered transaction announced on 19 March 2023, ending 167 years of independent existence. The instructive fact for Treasury professionals is that Credit Suisse was not, on the regulatory metrics, an illiquid bank. It reported an LCR above the requirement throughout, with the group ratio in the region of 150 percent at end-2022 by the public disclosures, and it was capitalised above its minimum requirements. It failed anyway.

The cause was a loss of confidence in the franchise rather than a shortfall against a ratio. Years of accumulated damage — the Archegos loss of approximately $5.5bn in 2021, the Greensill supply-chain fund collapse, multiple senior management changes, repeated restructurings, and a string of legal and compliance matters — had eroded belief that the institution had a stable future. In October 2022, social media speculation about the bank's health triggered client withdrawals that accelerated through the fourth quarter. The bank disclosed net asset outflows of approximately CHF 110bn in Q4 2022, with roughly CHF 138bn of deposit outflows in the quarter across the group — a scale that no LCR scenario contemplates, since the standardised outflow rates for stable retail and operational wholesale deposits assume single-digit to low-double-digit percentages over thirty days.

The final phase in March 2023 combined a specific trigger with an already-fragile base. The annual report was delayed after the auditor identified material weaknesses in internal control over financial reporting. Days later, the bank's largest shareholder stated publicly that it would not provide further capital. Outflows resumed at pace. The Swiss National Bank made emergency liquidity available, reported at up to CHF 50bn initially and subsequently extended to a facility of up to CHF 100bn with additional government-backed support, and the authorities brokered the UBS acquisition over a weekend. The transaction also produced the write-down of approximately CHF 16bn of Additional Tier 1 instruments — a decision that reordered the expected creditor hierarchy and reshaped how the AT1 market prices resolution risk.

The mapping is to the limits of ratio-based regulation. The LCR encodes assumptions about how depositors behave, derived from historical experience, and then defines compliance as holding enough HQLA to survive those assumed outflows for thirty days. It works when behaviour resembles the calibration. When clients conclude that the institution has no viable future, the outflow is not a stress scenario applied to a going concern — it is the market withdrawing the franchise. No ratio calibrated on going-concern behaviour describes that. The lesson is not that ratios are useless; it is that they are a floor, and a bank whose only liquidity narrative is its ratio has no narrative at all.

Northern Rock's funding model was viable until the market it needed closed

Northern Rock was a UK mortgage lender whose funding model depended on continuous access to wholesale and securitisation markets. By 2007 it had grown rapidly, with total assets of roughly £113bn, funding a book that was overwhelmingly UK residential mortgages of generally sound credit quality. Retail deposits funded only around a quarter of the balance sheet. The remainder came from wholesale borrowing and, crucially, from the securitisation of mortgage pools through its Granite programme — an originate-to-distribute model in which loans were written, packaged and sold to investors, releasing capacity to write more.

The model was internally coherent and, in normal markets, genuinely profitable. It carried a single, absolute dependency: the securitisation market had to remain open. In August 2007, following the deterioration of US subprime credit, the asset-backed securities market effectively closed. Northern Rock could not complete its planned securitisation, could not roll its short-term wholesale funding, and had no meaningful retail deposit base to fall back on. Its assets were not impaired — the mortgages were performing. Its funding channel had simply disappeared.

The bank approached the Bank of England for emergency liquidity support, and news of the facility became public on 13 September 2007. Queues formed outside branches the following day, producing the first run on a British bank in over 140 years. The retail run was a consequence of the wholesale failure, not its cause, but it made the position irrecoverable. The UK government guaranteed deposits, and the bank was taken into public ownership in February 2008. Among the durable consequences: the UK deposit guarantee arrangements were reformed, and the episode became a central exhibit in the case for the NSFR — a ratio explicitly designed to constrain the funding of long-dated assets with short-dated market borrowing.

The mapping is to funding concentration and market-access risk. Northern Rock was not concentrated in a single counterparty; it was concentrated in a single funding channel and a single instrument type. Market-access risk — the risk that a market on which the bank depends becomes unavailable regardless of the bank's own credit quality — is a distinct risk from counterparty concentration, and a diversification metric that counts investors rather than channels will not detect it.

Case Failure mode Which risk type Which stage of the Treasury process Lesson
Silicon Valley Bank, 2023 Asset-side duration plus uninsured deposit concentration IRRBB, model risk (deposits), funding concentration Balance sheet strategy and ALM measurement Hedge the duration you take; never apply retail behavioural assumptions to an uninsured, concentrated base
Credit Suisse, 2023 Loss of confidence in the franchise despite compliant ratios Funding liquidity, market liquidity Stress testing and contingency funding planning Ratios are a floor calibrated on history; franchise viability is the real liquidity asset
Northern Rock, 2007 Dependence on a single funding channel that closed Funding liquidity, market-access risk Funding plan design and diversification Diversify channels, not just counterparties; stable funding must match asset tenor

6. Generative AI in Treasury is safest where it drafts and most dangerous where it computes

Large language models have real uses in Treasury, and they cluster in one region of the work — language-heavy, judgement-supported, human-verified — and stay away from another. Five applications carry genuine value.

Drafting and reviewing funding documentation and investor materials. A benchmark bond issue requires a prospectus or drawdown documentation, investor presentations, roadshow scripts and internal approval papers, much of it derived from prior transactions with defined variations. Models are effective at producing first drafts from structured inputs and at reviewing documents for internal inconsistency — a maturity date stated one way on page four and another on page thirty-one, a defined term used before definition. The gain is cycle time on documentation that is genuinely repetitive.

Summarising and cross-checking supervisory correspondence and internal policy against actual practice. A large bank accumulates supervisory letters, SREP feedback, on-site inspection findings and internal policy documents in volumes no individual can hold in mind. A model with access to that corpus can answer "what have we committed to on intraday liquidity monitoring, and where" in seconds rather than in a week of searching. More valuably, it can surface tension between a policy statement and an operating procedure — the policy says collateral positions are reconciled daily, the procedure describes a weekly reconciliation. Finding those gaps before an inspection does is worth real money.

Interrogating regulatory text and mapping a change to affected systems and fields. When a technical standard is amended, someone must determine which reports, which data fields and which systems are affected. Done manually this takes weeks and is error-prone. A model working over regulatory text alongside the bank's own data dictionary and report specifications can produce a candidate impact map in hours. The candidate must be verified — but verifying a draft map is a different task from building one from nothing.

First-pass drafting of model documentation and validation reports. Model documentation is a substantial obligation under EBA and SSM expectations, and it is chronically under-resourced because writing it is tedious and the people capable of writing it are the people needed to build models. A model that produces a structured first draft from code, parameter files and test results shifts the human effort from composition to review, which is both faster and a better use of the expert.

Natural-language interrogation of liquidity and ALM data for management. A question such as "which currencies have deteriorated most in LCR terms over the last quarter, and what drove it" currently routes through an analyst and a reporting queue. A well-governed natural-language interface over a controlled data layer can answer it directly. The critical design constraint is that the model must translate the question into a query against a governed dataset, not compute the answer itself.

The risks are serious and must be treated as such rather than acknowledged and moved past.

Leakage of material non-public information about the bank's own liquidity position is the most acute. A Treasurer's questions are themselves sensitive. "Model the impact of a two-notch downgrade on our funding costs" reveals concern about a downgrade. Submitted to an external service without contractual and technical guarantees on data handling, retention and training use, that prompt is a disclosure. The controls are architectural, not procedural: private deployment or contractually ring-fenced hosting, no retention for training, prompt logging inside the bank's own perimeter, and classification rules that block the highest-sensitivity categories at the interface rather than relying on user discipline.

Hallucinated regulatory or numerical content in a document destined for a supervisor is the second. Language models generate plausible text, and plausible-looking regulatory citations and figures are among the things they generate. A fabricated article reference in an ILAAP submission, or a plausible but wrong outflow percentage in a policy document, is a serious matter — supervisors treat inaccurate submissions as a governance failure regardless of the mechanism that produced them. The control is that no model-generated regulatory citation or figure reaches an external document without verification against source, with the verification recorded.

Model risk governance obligations apply to any model that influences a regulatory number. EBA and SSM expectations on model risk management do not distinguish between a regression and a large language model — if the output influences a reported figure or a risk decision, it requires an owner, documentation, validation, monitoring and periodic review. Layered on top, the EU AI Act introduces risk-tiering, transparency and documentation obligations, with phased application, and systems used in credit and financial decisioning falling into higher-obligation categories. A bank deploying GenAI in Treasury needs an inventory of where it is used, a classification of each use against both the internal model risk policy and the AI Act tiering, and evidence of the controls applied. Building the inventory retrospectively, after tools have spread through the department informally, is considerably harder than building it from the start.

Explainability matters where the output feeds a decision the bank must justify. If a model contributes to setting a transfer price, sizing a buffer or shaping a hedge, the bank must be able to explain to ALCO, to internal audit and to a supervisor why that answer and not another. "The model produced it" is not an explanation. This is the strongest argument for keeping GenAI in the drafting and summarising layer, where the human retains and exercises the judgement, rather than in the calculation layer, where the judgement migrates to the machine and the explanation goes with it.

Use of GenAI The gain The governing risk The control
Funding documentation and investor materials Faster drafting and consistency review Hallucinated terms, figures or dates in a legal document Full legal and DCM review; models draft, they never approve
Supervisory correspondence and policy cross-check Rapid recall across a corpus no human holds Confidential supervisory information exposed to a third party Private deployment, no training retention, access controls mirroring the source systems
Regulatory change impact mapping Impact analysis in hours rather than weeks Missed or invented linkages; false confidence in coverage Output treated as a candidate map; verified by a named subject matter expert
Model documentation and validation drafting Expert time moves from writing to reviewing Documentation that describes a model the bank does not run Model owner attests to accuracy; validation independently confirms against code
Natural-language data interrogation Direct management access to liquidity and ALM data Numerical hallucination presented with confidence Model generates queries against a governed dataset; it never computes the number itself

  1. Why can EVE and NII sensitivity point in opposite directions after the same modelling change?

  2. Credit Suisse reported an LCR well above the requirement and still failed. How?

  3. What is the headroom fallacy, and what is the right diagnostic?

  4. Which model risk governance obligations attach to a GenAI tool used in Treasury?

  5. Why are Silicon Valley Bank, Credit Suisse and Northern Rock the right three cases rather than any three failures?

  6. What makes the replicating portfolio tenor the most consequential assumption in the bank?

  7. A real-time liquidity dashboard is installed and the escalation path is unchanged. What has improved?

5

  1. Credit Suisse had a healthy LCR and still failed. How?

  2. What is the headroom fallacy?

  3. Why can EVE and NII sensitivity point in opposite directions?

  4. A GenAI tool influences a regulatory number. What applies?

The six ideas that matter most

Idea 1 — The liability side is a strategy, not a residual

Most descriptions of a bank start with the assets, as though funding were something that arrives afterwards to pay for decisions already made. That sequence is backwards. The shape of the liability side — how much retail deposit, how much corporate, how much wholesale, at what tenor, in which currencies, from which investors — is chosen, and once chosen it determines what the asset side is permitted to do. A bank funded overwhelmingly by short wholesale borrowing cannot hold long fixed-rate assets safely, whatever its credit appetite. A bank with a deep, granular, insured retail deposit base can. The funding structure is the constraint, and constraints set strategy.

Meridian Bank's illustrative composition makes the point concrete. With €180bn of retail deposits against €85bn of wholesale funding, the bank has the stable base to hold long-dated mortgages and to run a term lending business through a market closure. Invert those numbers and the same asset strategy becomes Northern Rock. The assets in both cases could be identical residential mortgages of identical quality. The difference in survivability is entirely on the liability side, and it was decided years before any stress arrived.

This reframes what Treasury is for. A Treasury that is asked to fund whatever the businesses originate is an execution function operating under someone else's constraints. A Treasury that shapes the funding profile deliberately — deciding tenor, diversification, instrument mix and currency mix, and then communicating what that profile permits — is setting the boundary within which the bank's strategy is feasible. The second version is the one that prevents failures.

Idea 2 — Data cascades, so a single contract field reaches the ratio, the price and the supervisory return

One field in one contract — the maturity date, the counterparty classification, the collateral eligibility flag — does not stay in one place. It flows into the liquidity calculation and changes the LCR. It flows into the behavioural model and changes the assumed cashflow profile. It flows into the FTP engine and changes the price the originating business is charged. It flows into COREP and appears in a return filed with the supervisor. A single wrong classification propagates into a regulatory number, an internal price and an external submission, and it does so silently.

This is why data lineage in Treasury is a control rather than an IT housekeeping matter. BCBS 239 framed risk data aggregation and reporting as a governance requirement precisely because the alternative — figures that cannot be traced to source — makes it impossible to answer a supervisory question or to know whether a discovered error is isolated or systemic. Many European banks still carry open findings in this area more than a decade after the principles were issued, which is a statement about how hard the problem is rather than about how optional it is.

The practical implication for anyone working in Treasury is to trace before trusting. When a number looks wrong, the question is not "what is the correct number" but "which field, in which system, produced this, and where else does that field go." The second question finds the other places the same error is already sitting.

Idea 3 — FTP is where the bank's strategy is priced

Funds transfer pricing is the internal mechanism that charges each business for the funding it consumes and credits each business for the funding it supplies. It sounds like accounting. It is allocation. The FTP curve determines which businesses appear profitable, which products get invested in, which balance sheet gets committed where. It is the single most powerful non-executive lever over the composition of the bank, because it operates continuously on every origination decision rather than periodically on a strategy document.

Whatever FTP under-charges, the bank will over-produce. If the transfer price for a five-year fixed-rate corporate loan does not reflect the true cost of funding five-year money — Meridian's illustrative 3.67 percent five-year FTP rate, built from the bank's own term funding curve — then that product looks more profitable than it is, the business writes more of it, and the funding gap grows. Nobody made a decision to take more term funding risk. The pricing simply made it attractive, and volume followed.

The corollary is that FTP methodology changes are strategic decisions and must be governed as such. Adjusting a liquidity premium or changing a deposit crediting tenor reallocates value between businesses and changes origination incentives across the bank. That belongs at ALCO with documented rationale, not in a technical update circulated by the ALM team. And the businesses must be able to challenge it — which is why FTP dispute volume, discussed among the KPIs, is a genuine measure of whether the framework is alive.

Idea 4 — Liquidity regulation encodes assumptions about behaviour, and assumptions can be wrong

The LCR is not a measurement. It is a model with regulatory force. It specifies which assets count as liquid and at what haircut, and which liabilities are assumed to leave and at what rate over thirty days — 5 percent for stable retail deposits, higher for less stable retail, 25 percent for non-operational wholesale, 100 percent for certain categories. Every one of those numbers is a behavioural assumption derived from historical experience and negotiated through the Basel process. The NSFR does the same over a one-year horizon with available and required stable funding factors.

Credit Suisse demonstrated what happens when behaviour departs from the calibration. The bank complied with a ratio built on the assumption that depositor behaviour under stress resembles the past, and then encountered depositors who had concluded the institution had no future. Silicon Valley Bank demonstrated the adjacent failure: assumptions about deposit stickiness applied to a population — concentrated, uninsured, digitally connected, single-industry — for which no historical basis existed.

The operating conclusion is to treat ratios as a floor and to build the judgement layer above them. That layer is the ILAAP, the internal stress scenarios that are harsher and more specific than the regulatory ones, the survival horizon under combined stress, and above all the reverse stress test — the question of what would have to happen for the buffer to be exhausted, and how far that is from what the models predict. A bank that can only describe its liquidity position by quoting its LCR has outsourced its risk judgement to a committee in Basel that never saw its balance sheet.

Idea 5 — Treasury has a daily irreversible moment, so resilience is a business topic

Most functions in a bank can absorb a bad day. Treasury cannot, because Treasury settles. Payments have cut-off times. Margin calls have deadlines. Bond settlement dates do not move. A missed settlement is not an internal issue to be resolved next week — it is an event observed by counterparties, reported through market infrastructure, and priced into how the bank is treated tomorrow. The irreversibility is what makes Treasury operational risk different in kind from operational risk elsewhere.

This is why DORA lands on Treasury with particular weight. The EU's Digital Operational Resilience Act requires ICT risk management, incident classification and reporting, resilience testing including threat-led penetration testing for significant entities, and oversight of critical third-party providers. For Treasury this means the payment interfaces, market data feeds, collateral management platforms and central bank connectivity that the daily process depends on — and it means the bank must be able to operate, at least in a degraded mode, when one of them is unavailable at 11:00 on a settlement date.

The business consequence is that resilience investment cannot be delegated wholly to technology functions. Deciding which payment flows must continue in a degraded state, which can wait, who has authority to invoke manual processing, and what the manual process actually is — these are business decisions about the bank's obligations and relationships. Testing them, including at inconvenient times with the primary decision-makers deliberately absent, is the only way to know whether the plan is real.

Idea 6 — Real transformation moves cost or risk, and GenAI is governed as tightly as it is useful

The filter applied throughout this module is the one to carry out of it. Does the change move the economics — cost of funds, buffer size, collateral consumption, cost of a controlled process? Does it move the risk — earlier detection, shorter time to act, fewer places an error can enter? If neither, it has changed how the work looks, and looking better is not a return.

Applied to the five technology trends, the filter is discriminating. Collateral optimisation moves economics measurably. Lineage-as-a-product moves the cost of assurance and supervisory response. Elastic computation moves risk, conditional on the faster answer reaching someone empowered to act. Real-time intraday data moves risk, conditional on the escalation path changing with it. Machine-learning behavioural models move both economics and risk, while adding a model risk burden that must be paid rather than assumed away.

Applied to GenAI, the filter gives the clearest answer in the module. GenAI moves economics in drafting, summarising, regulatory impact mapping and documentation, where the human expert stays in the judgement seat and being wrong produces a correction. It moves risk where it surfaces gaps between stated policy and actual practice before an inspection does. It is dangerous where it computes, because a hallucinated figure entering a ratio, a price or a supervisory return is a governance failure regardless of its cause — and because EBA and SSM model risk expectations plus the EU AI Act's tiering and documentation obligations apply in full to anything influencing a regulatory number. Useful and governed are not in tension here; the governance is what makes the usefulness bankable.

Common misconceptions, and an applied exercise

Understanding a subject means being able to say what is wrong about it as precisely as what is right. Everything up to this point has built the positive account: how funding works, how it is priced, how it is measured, how it fails. This section does the negative work. It attacks the eight mental models that most reliably mislead intelligent people about Treasury — each of them plausible, each of them held by senior practitioners, each of them wrong in a way that has cost money. Then it hands you a bank and asks you to run it for a year.

Section A — Eight beliefs that sound reasonable and are wrong

"A high LCR means a bank is safe"

The Liquidity Coverage Ratio is the single number most often quoted as evidence of liquidity strength, and it is the single number least capable of carrying that weight. The belief is intuitive because the ratio is constructed to look like a safety measure: high-quality liquid assets divided by thirty days of stressed net outflows, with 100% meaning survival. A bank at 143%, like Meridian, appears to have 43% more resource than it needs. The mental image is a reservoir with comfortable freeboard.

The image is wrong because the denominator is not a forecast. It is a set of prescribed run-off assumptions written into Delegated Regulation (EU) 2015/61 — 5% on stable insured retail deposits, 10% on less stable retail, 25% on operational corporate balances, 40% on non-operational non-financial corporate deposits, 100% on financial institution funding. Those weights were calibrated against historical episodes and negotiated politically. They are a common yardstick, not a prediction of what your depositors will do. Credit Suisse reported an LCR of roughly 150% in late 2022 and around 144% at the end of that year — comfortably above requirement — and then lost deposits at a rate no prescribed assumption contemplated. Silicon Valley Bank's concentrated, uninsured, digitally connected depositor base behaved in a way that a 40% thirty-day weight does not describe; the run compressed into hours.

Three things about the buffer matter more than the headline ratio, and none of them is visible in it. The first is composition: €62bn of HQLA is not a single thing. Level 1 government bonds monetise instantly at scale in any market; Level 2A covered bonds and Level 2B corporate paper monetise more slowly, at wider bid-offer, and precisely when everyone else wants to sell them. Two banks at 143% with different Level 1 shares are not equally liquid. The second is encumbrance and availability: HQLA sitting in a subsidiary behind a local regulator's ring-fence, or already committed as collateral, counts in the consolidated ratio and is unavailable to the group in the moment it is needed — the trapped-liquidity problem. The third is the currency and entity ladder underneath the consolidated number; a group can be at 143% in aggregate and short in dollars.

And the buffer is not free. Holding €62bn of low-yielding assets funded at Meridian's marginal cost is a permanent, quantifiable drag on earnings — the difference between the HQLA yield and the blended funding cost, applied across the buffer. Excess LCR is therefore a decision, not a virtue, and it should be defended as a decision. The corollary is uncomfortable: a ratio that sits at 142–145% quarter after quarter, never moving, is more likely to indicate a bank that manages the reported number at reporting dates than a bank with a genuinely stable liquidity position. Window-dressing around period ends is a documented supervisory concern. The right question to ask of an LCR is not "how high" but "how volatile intra-period, of what composition, and available to whom".

"Deposits are free money"

Deposits carry no coupon in the way a bond does, so they get treated as costless funding. In a zero-rate world this was nearly true, and a generation of commercial bankers formed their instincts there. It is not true now, and the error has three parts.

The first part is beta. When policy rates rise 200 basis points, a bank does not pass on zero and it does not pass on 200; it passes on some fraction, the deposit beta. Meridian's €275bn of retail and corporate deposits at a blended beta of 45% cost an additional €2.5bn a year after a 200bp move — illustrative, but the order of magnitude is real. Beta is not a constant. It is low early in a hiking cycle and rises as depositors notice, as competitors advertise, and as money market funds become visible alternatives. It is higher for non-operational corporate balances than for insured retail current accounts. It is higher when switching is one screen tap. The bank that budgeted 45% and realised 62% did not misprice a product; it mismodelled a behaviour.

The second part is the back book. A mortgage written at 1.4% in 2021 does not reprice. The deposit funding it does. The entire margin compression of the 2022–2024 cycle for European retail banks is that arithmetic, and it is why the sequence of rate moves matters more than the level. The third part is that deposits are not free even before interest: branch networks, payment infrastructure, fraud and financial crime controls, and deposit guarantee scheme contributions are all real costs of the deposit franchise, and they are typically carried somewhere other than the line that shows the deposit's rate.

The correct model is that a deposit's value lies in its behaviour, not its rate. A €1bn insured retail current account balance with a five-year observed half-life and a 20% beta is a fundamentally different instrument from a €1bn non-operational corporate balance with a 90% beta and a thirty-day half-life, even if both pay 1.80% today. The first is a long-dated, low-cost, LCR-favourable liability. The second is expensive short-term wholesale funding wearing a retail-sounding name. Any framework that prices them the same — and pooled-rate FTP does exactly that — will cause the bank to buy the wrong one.

"Liquidity risk is a treasury problem, not a business problem"

Treasury manages liquidity. Treasury does not create it. Liquidity risk is manufactured on the trading floor and in the branch network, by people making lending and product decisions, and Treasury's role is to hold the consequences. A relationship manager who grants a €500m committed revolving credit facility to win a mandate has written a liquidity option: the client draws when the client's own funding is impaired, which correlates with the moment the bank's funding is also impaired. A retail product designer who launches a savings account with a rate guarantee and no notice period has created a liability that behaves like overnight money in a stress. A structured products desk that embeds a client early-termination right has created a contingent outflow with no offsetting asset.

The intuition behind the misconception is organisational: liquidity limits live in a Treasury policy, liquidity reporting is produced by a Treasury team, and the ILAAP has a Treasury author. So liquidity looks like something that happens in one building. But the balance sheet is the aggregation of thousands of front-office decisions, and no amount of downstream management fixes an upstream product that should not have been sold at that price.

The funds transfer pricing curve is precisely the mechanism that converts this from a Treasury problem into a business problem, and it works only if the contingent leg is charged. Meridian's five-year FTP rate of 3.67% — 3.05% base plus a 62bp term liquidity premium — makes a five-year loan visibly more expensive to originate than a one-year loan. That part is usually built. The part that is usually missing or under-calibrated is the contingent liquidity charge on undrawn commitments. If a business line can grant an €80m undrawn revolver at a charge of zero, it will grant them, because they are free to it and they win business. The cost does not disappear; it lands as an outflow assumption in the LCR, as required stable funding in the NSFR, and as counterbalancing capacity that Treasury must hold. The whole bank pays so that one desk can hit a target. Charging 12bp on the undrawn amount is not an accounting nicety; it is the difference between a bank that knows what its contingent book costs and a bank that discovers it in March.

"FTP is an internal accounting entry with no real consequence"

The objection sounds sophisticated: transfer prices net to zero at group level, therefore they cannot change group profit, therefore they are bookkeeping. The first two clauses are true. The conclusion does not follow, because transfer prices do not change group profit directly — they change behaviour, and behaviour changes group profit.

An FTP curve is a set of instructions to a sales force. It tells them which loans clear their hurdle and which do not, which deposits are worth chasing and at what rate, and which tenors are worth writing. Set the term liquidity premium too low and the bank originates long-dated fixed-rate assets that nobody prices correctly, accumulating an interest rate and funding mismatch that only becomes visible when the curve moves. Set the deposit credit too high on non-operational corporate balances and the corporate bank will go and buy hot money at a loss the corporate bank does not see. The transfer price is where the balance sheet's shape is decided, one deal at a time, by people who never read the ALCO pack.

The most consequential and most common error is the pooled or average-cost rate. A single blended funding rate applied to every asset regardless of tenor systematically overprices short assets and underprices long ones. The mechanism is arithmetic: the pooled rate sits somewhere in the middle of the funding curve, so anything longer than the pool's average duration is being funded at less than its true marginal cost. Every rational originator responds by writing the longest-dated asset they can, because that is where the subsidy is. Over years, this produces a balance sheet stuffed with long-dated fixed-rate assets funded short — precisely the position that destroyed Silicon Valley Bank, and precisely the position no ALCO ever voted for. Nobody decided it. The transfer pricing framework decided it, quietly, thousands of times.

"If the assets are good, the bank is fine"

Solvency and liquidity are different failure modes, and the second kills faster. A bank is solvent when the present value of its assets exceeds the present value of its liabilities. A bank is liquid when it can meet obligations as they fall due. These are related but independent, and it is entirely possible — routine, in fact — to be the first and not the second.

Northern Rock is the cleanest case in the European record. In September 2007 its mortgage book was performing; arrears were below the UK market average; the assets were, in the ordinary sense, good. What failed was the funding model: heavy reliance on securitisation and wholesale markets that closed simultaneously when the asset-backed market seized. The bank could not wait for its good assets to pay. A twenty-five-year mortgage that will repay in full is worthless as a means of settling a wholesale maturity on Thursday. The distinction is between a bank being right and a bank being able to wait long enough to be proved right.

Liquidity is what buys the time in which solvency can be demonstrated. This reframes the buffer's purpose. The HQLA stock is not there because the assets are bad; it is there because the market's willingness to believe the assets are good is not under the bank's control, and belief can withdraw faster than any asset can be converted. Credit Suisse in March 2023 had a CET1 ratio around 14% — capital was not the proximate cause. Confidence was, and confidence is settled in the funding market, not the capital ratio.

"Hedging removes risk"

Hedging transforms risk. It rarely removes it, and it always creates a new one. Three transformations matter.

The first is accounting. An economically perfect hedge can produce a violently imperfect P&L if hedge accounting under IFRS 9 or the carried-forward IAS 39 macro hedge model is not achieved or is subsequently broken. The derivative is fair-valued through profit or loss; the hedged item may be at amortised cost. The mismatch is not an economic loss, but it is a reported one, and reported volatility has consequences for capital, for analysts, and for the credibility of the people who put the hedge on.

The second is basis. A hedge is a different instrument from the exposure. A swap referencing €STR against a deposit book that reprices on an administered rate leaves the spread between them unhedged. A cross-currency basis swap hedging a dollar asset leaves the bank exposed to the basis itself, which moves — often widening exactly when dollar funding is scarce. A government bond hedge on a corporate credit exposure leaves the spread. The residual is smaller than the original exposure, which is the point, but it is not zero and it is often correlated with stress.

The third transformation is the one that actually causes failures: a hedge is a collateralised contract, and collateral is a liquidity obligation. Under a standard CSA, a mark-to-market move against you produces a variation margin call in cash, next day, regardless of whether the hedged item has moved in your favour by the same amount — because the hedged item, typically, is not collateralised. The hedge converts a valuation risk into a cash flow risk. The September 2022 UK LDI episode is the canonical demonstration: pension schemes holding leveraged gilt exposure were economically hedged against rate moves, and were destroyed operationally by the speed of the margin calls that the hedge itself generated when gilt yields gapped. The positions were right. The liquidity to service them was not there in the hours available.

And there is a case where hedging the obvious quantity makes the ratio worse. Structural FX is the standard trap. A bank with a foreign subsidiary holds a net investment denominated in the foreign currency; that investment is also carrying foreign-currency risk-weighted assets. Hedging the full net investment amount stabilises the euro value of the equity — and destabilises the capital ratio, because a currency move now changes the denominator without a matching change in the numerator. The correct hedge ratio is the one that stabilises the ratio, which is generally not the full net investment. The lesson generalises: define the variable you are protecting before you size the hedge, because hedging the wrong variable perfectly is worse than hedging the right one approximately.

"The regulatory ratio is the risk measure"

LCR, NSFR, CET1, the leverage ratio and MREL are floors and comparability devices. They are not statements of risk appetite, and they were never designed to be. Their function is to make banks comparable across jurisdictions and to establish a minimum below which supervisors intervene. They achieve this by prescribing assumptions — which is the same as saying they achieve it by ignoring what is specific about your bank.

The internal framework is where the actual risk measurement lives: the ILAAP's own scenarios, calibrated to your depositor concentration and your funding channels; the survival horizon under a name-specific stress, a market-wide stress, and a combination; the cash ladder run daily at entity and currency level; the ICAAP's capital scenarios. These measure what the regulatory ratio deliberately abstracts away. Meridian's LCR of 143% is a compliance fact. Meridian's modelled survival horizon under an idiosyncratic stress with 60% non-operational corporate run-off is a risk fact, and it is the one that should set limits.

Managing to the ratio is the mechanism by which a bank acquires precisely the risk the ratio does not capture. If the ratio prices a thirty-day horizon, the optimising bank funds at thirty-one days. If the ratio weights financial-institution deposits at 100% and non-financial corporate at 40%, the optimising bank recharacterises balances. If the ratio caps Level 2B at 15%, the bank fills to 15%. Each step is legal, each improves the reported number, and the aggregate is a portfolio selected for the characteristic of being invisible to the metric. This is not a hypothesis; it is what regulatory arbitrage means. The defence is a second, internal, differently-shaped measure that the optimisation was not aimed at — and a Treasury willing to say that a compliant position is nonetheless outside appetite.

"Real-time data would have prevented it"

This is the most seductive of the eight, because it is half true, and because it is the one with a vendor behind it. After every liquidity failure, an analysis appears showing that the bank's deposit reporting was T+1 or worse, and concluding that real-time visibility would have changed the outcome. Sometimes it would have. The condition under which it does is specific, and stating it is more useful than either endorsing or dismissing the claim.

Detection speed only converts into outcome improvement if the escalation path and the decision authority operate at the same speed. A bank that observes a €400m outflow at 10:00, routes it to a liquidity risk analyst at 11:00, escalates to a Treasurer at 14:00, convenes an ALCO at 16:00 and takes a decision the following morning has bought nothing with its real-time feed. It has bought a more precisely timestamped record of a decision it took a day late. The binding constraint was never the data latency; it was the governance latency. Real-time data in front of a monthly committee is an expensive telemetry system for a slow car. The corollary is that the cheapest liquidity resilience investment in many institutions is not a data platform: it is a standing delegation that lets a named individual monetise €5bn of Level 1 collateral without a committee, pre-positioned central bank collateral that can be drawn the same day, and a rehearsed crisis team with a thirty-minute convening standard.

The counter-case is real and should not be waved away. Where an institution has genuine same-day decision capability — a pre-delegated authority, a pre-positioned collateral pool, a repo desk that can execute size — detection speed becomes the binding constraint, and hours matter enormously. Intraday liquidity monitoring under BCBS 248 exists for exactly this reason: a bank that can see its payment throughput and its collateral position at ten-minute granularity can throttle outgoing payments, prioritise time-critical settlement and avoid a TARGET2 cut-off failure that would otherwise be discovered at 17:45 with nothing to be done about it. In a 2023-style run propagating at social-media speed, the difference between hourly and daily deposit visibility is the difference between acting on day one and acting on day two, and there may not be a day two. The honest formulation: real-time data is a multiplier on decision capability, not a substitute for it. Build the decision capability first, then the data pays for itself. Build the data first, and you will have a beautiful dashboard of your own failure.

Section B — The applied exercise: run Meridian Bank for a year

What follows is the part of the module where you do the work. Four scenarios, in the order a Treasurer would meet them: build the year's funding plan, price a transaction, absorb a rate shock, survive a stress event. Each gives you a situation, the data you need, and an explicit question. Each worked answer sits in a box below the question — attempt the answer before reading the box, because the value of the exercise is entirely in the attempt. All figures are illustrative and constructed for teaching; they are internally consistent with the Meridian balance sheet used throughout the module but they are not calibrated to any real institution.

Scenario 1 — The funding plan is constrained by MREL and NSFR, not by cash

It is October. You are preparing the following year's funding plan for ALCO and board approval. The commercial plan projects €3bn of net lending growth, weighted towards mortgages and mid-market corporate lending. Deposit gathering is expected to add €1.2bn. The wholesale maturity schedule and the regulatory position are below.

Item Position (illustrative)
Wholesale maturities falling due in the plan year €9.5bn
Projected net lending growth €3.0bn
Projected net deposit growth €1.2bn
Senior non-preferred losing MREL eligibility (residual maturity falling below one year) €2.6bn
MREL stock 28.1% of RWA
MREL requirement 26.5% of RWA
MREL management buffer policy 150bp above requirement
RWA, current €142.0bn
Projected RWA growth from lending plan €2.0bn
NSFR 118% (ASF €309.2bn / RSF €262.0bn)
Internal NSFR floor 112%
Board NSFR operating target 118%
ASF lost during the year to term debt rolling inside twelve months €6.2bn
Asset encumbrance policy cap 20% of total assets

The question. How much should Meridian raise, in which instruments, and — the part that matters — which constraint is actually binding? Do not answer "cash", and be prepared to say why not.

Scenario 1

Start with the cash requirement, because it is the one everybody computes first and it is the one that turns out not to bind. Maturities of €9.5bn must be refinanced. Net lending growth of €3.0bn less deposit growth of €1.2bn leaves €1.8bn to be funded from markets. Total cash need: €11.3bn. If cash were the constraint, the plan would be €11.3bn.

Now MREL. Current stock is 28.1% of €142.0bn, which is €39.90bn. The €2.6bn of senior non-preferred crossing inside one year ceases to be eligible, taking the stock to €37.30bn. Meanwhile RWA grows by €2.0bn to €144.0bn, so the 26.5% requirement rises to €38.16bn. Meridian is therefore €0.86bn short of the requirement before any issuance — a breach, arriving mechanically, with no adverse event at all. Add the 150bp management buffer, 1.5% of €144.0bn, or €2.16bn, and the target stock becomes €40.32bn. Required senior non-preferred issuance: €40.32bn less €37.30bn, or €3.02bn. Call it €3.0bn, and note that this must be subordinated, MREL-eligible paper — covered bonds and senior preferred do not substitute, whatever their cost advantage.

Now NSFR. RSF grows with the lending: €3.0bn at a blended required stable funding factor of roughly 78% (residential mortgages at 65%, corporate loans over one year at 85%) adds €2.34bn, taking RSF to €264.3bn. On the funding side, €6.2bn of ASF is lost as term debt rolls inside twelve months, and deposit growth of €1.2bn at a 90% ASF factor adds €1.08bn. Available ASF before new issuance: €309.2bn − €6.2bn + €1.1bn = €304.1bn. To hold the board's 118% target on RSF of €264.3bn requires ASF of €311.9bn. The gap is €7.8bn — meaning at least €7.8bn of the plan must be issued with residual maturity above one year, and realistically well above, because paper issued at thirteen months provides ASF for one month and then rolls down. Note carefully what this does and does not constrain: it constrains the tenor mix, not the volume. Note also that against the 112% internal floor, required ASF is only €296.0bn, so there is €8.1bn of headroom. The regulatory-adjacent floor is not what binds. The board's own operating target is.

The plan therefore builds to approximately €14bn, not €11.3bn: €11.3bn of cash need, plus roughly €2.7bn of pre-funding carried into the following year, which is prudent practice and which also protects against the possibility that the covered bond or senior non-preferred window is shut when it is needed. A defensible composition, all illustrative: €4.5bn covered bonds (cheapest funding, longest tenor, best ASF, but encumbering roughly €5.4bn of mortgage collateral at 120% overcollateralisation — check against the 20% encumbrance cap before committing); €4.5bn senior preferred; €3.0bn senior non-preferred to satisfy MREL; €0.5bn Tier 2 to maintain the capital stack; €1.5bn of private placements, structured notes and retail-distributed paper for diversification.

The binding constraints are MREL — which dictates €3.0bn of a specific, expensive, subordinated instrument regardless of whether Meridian needs the cash — and NSFR, which dictates that the bulk of the remainder be term rather than short-dated. Cash is not binding; it is the easiest of the three to satisfy and would have been satisfied by €11.3bn of one-year paper that would have failed both of the others. This is the general lesson of funding planning: the volume falls out of the cash arithmetic, but the shape falls out of the constraints, and the shape is what the plan is for.

Scenario 2 — The undrawn facility decides a loan that fails anyway

The corporate bank brings a transaction to the pricing committee. A seven-year €120m term loan to a mid-market borrower at 5.05%, alongside an €80m undrawn committed revolving credit facility. The relationship manager argues it is a strategically important client. You have the following inputs.

Input Value (illustrative)
Term loan amount / rate / tenor €120m / 5.05% / 7 years
Seven-year FTP base curve 3.20%
Seven-year term liquidity premium 78bp
Undrawn committed revolver €80m
Contingent liquidity charge on undrawn 12bp
Borrower risk weight 100%
Credit conversion factor, committed facility 40%
Expected loss 0.55%
Operating cost allocation 0.10%
CET1 allocation rate 13.7%
Cost of equity 10%
RAROC hurdle 12%
Tax rate 25%

The question. Does this transaction clear the 12% RAROC hurdle? If not, by how much does it miss, and what would have to change for it to clear? Show every step.

Scenario 2

Revenue first. The term loan yields 5.05% on €120m, or €6.060m a year.

The transfer price is the base curve plus the term liquidity premium: 3.20% + 0.78% = 3.98%, applied to €120m, or €4.776m. Net interest income after FTP is therefore €6.060m − €4.776m = €1.284m, a margin of 107bp over the transfer price. That number alone looks respectable, and it is where an under-built framework would stop.

Now the costs the under-built framework omits. The contingent liquidity charge on the undrawn revolver is 12bp on €80m, or €0.096m. Expected loss is 0.55% on €120m, or €0.660m. Operating cost allocation is 0.10% on €120m, or €0.120m. Profit before tax: €1.284m − €0.096m − €0.120m − €0.660m = €0.408m. After tax at 25%, net profit is €0.306m.

Now the capital. The drawn loan carries €120m of RWA at a 100% risk weight. The undrawn commitment carries a 40% credit conversion factor, giving €32m of exposure at default and, at 100% risk weight, €32m of RWA. Total RWA is €152m. Allocated CET1 at 13.7% is €20.824m.

RAROC is €0.306m divided by €20.824m, or 1.47%. Against a 12% hurdle, this is not a marginal miss. It is a failure by an order of magnitude. Even on the more generous convention that credits the allocated equity with a notional return at the base rate — €20.824m at 3.20%, or €0.666m of additional pre-tax income — pre-tax profit rises to €1.074m, net profit to €0.806m, and RAROC to 3.87%. Still nowhere near.

What would have to change? To reach 12% on €20.824m of allocated equity requires net profit of €2.499m, therefore pre-tax profit of €3.332m, against the €0.408m actually generated. The gap is €2.924m, which on €120m of drawn balance is 244bp. The loan would have to be priced at approximately 7.49% rather than 5.05%. That is not a negotiation; it is a different transaction.

The revolver is where the analysis is won or lost, though not in the way the phrase usually implies. It contributes €4.384m of the €20.824m of allocated equity — 21% of the capital charge — and €0.096m of annual cost, for zero revenue. Restructuring it as a 364-day uncommitted line would cut the conversion factor to 10%, reducing RWA to €128m and allocated equity to €17.536m, and would cut the contingent charge to a few basis points; RAROC improves to roughly 2.1%. Materially better in relative terms, and still a decisive failure. The point is that a framework which charges nothing for undrawn commitments and applies no capital to them would have shown this deal at roughly 2.5% RAROC on €16.44m of equity — still failing, but by a margin small enough to argue about, and small enough for a persuasive relationship manager to win the argument. The contingent leg does not usually flip a good deal to bad. It removes the ambiguity that lets bad deals through.

Two traps worth naming. First, reducing the hold from €120m to €40m and syndicating the rest does not improve RAROC — it scales numerator and denominator together. It reduces the absolute capital consumed, which is a portfolio decision, not a pricing fix. What does help is the syndication fee: €80m distributed at 60bp is €0.48m of upfront income, which transforms year one and is worth €0.069m a year if amortised across seven. Be explicit about which convention you are using, because the same deal passes or fails depending on it. Second, the honest conclusion is that the transaction is uneconomic on lending alone and requires roughly €2.9m of annual pre-tax ancillary revenue — cash management, FX, hedging, debt capital markets — to justify. That is a large and specific number. If the relationship case is real, someone should be prepared to commit to it and be measured against it. Most corporate lending in Europe clears its hurdle this way or does not clear it at all; what distinguishes a well-run bank is that the ancillary requirement is quantified upfront rather than asserted afterwards.

Line Calculation Amount
Interest income €120m × 5.05% €6.060m
FTP charge €120m × 3.98% (€4.776m)
Net interest income €1.284m
Contingent liquidity charge €80m × 0.12% (€0.096m)
Operating cost €120m × 0.10% (€0.120m)
Expected loss €120m × 0.55% (€0.660m)
Profit before tax €0.408m
Tax at 25% (€0.102m)
Net profit €0.306m
RWA €120m + (€80m × 40%) €152.0m
Allocated CET1 €152.0m × 13.7% €20.824m
RAROC €0.306m / €20.824m 1.47%

Scenario 3 — The beta assumption breaks and the temptation is to fix the model

Policy rates have risen 200bp over nine months. Two competitors have launched aggressively priced instant-access savings accounts and a third is advertising a twelve-month term deposit at 150bp above Meridian's board rate. The ALM model assumed a blended deposit beta of 45% across the €275bn retail and corporate book. Realised beta over the last two quarters is running at 62%.

Item Modelled Realised (illustrative)
Blended deposit beta 45% 62%
Rate-sensitive deposit base €275bn €275bn
Annual pass-through cost at +200bp €2.475bn €3.410bn
Modelled replicating portfolio average life, core deposits 4.2 years Behaviour implies ~2.6 years
Five-year FTP credit paid to deposit business 3.67% —
EVE limit 15% of Tier 1 —

The question. What is the effect on net interest income, on the FTP credit paid to the deposit business, and on EVE sensitivity? What should ALCO do, in what order, and what is the trap?

Scenario 3

The NII effect is direct. At the modelled 45% beta, a 200bp move costs €275bn × 2.00% × 45% = €2.475bn a year. At the realised 62%, it costs €275bn × 2.00% × 62% = €3.410bn. The difference is €0.935bn of unbudgeted annual interest expense — against a bank of this size, roughly one seventh of net interest income. This is not a rounding error in a model; it is the year's plan.

The EVE effect runs the other way and is larger. The economic value of the deposit franchise in a rising-rate world comes from the spread between the market rate and the administered rate, capitalised over the modelled behavioural life. Modelled at a 4.2-year average life with 45% pass-through, the deposit book contributes roughly €275bn × 4.2 × 2.00% × (1 − 0.45), or about €12.7bn of value under the +200bp shock. At a 2.6-year life with 62% pass-through, it contributes roughly €275bn × 2.6 × 2.00% × (1 − 0.62), or about €5.4bn. A swing of some €7.3bn on a crude linear approximation, against an EVE limit of 15% of Tier 1 — around €2.9bn on CET1 of €19.5bn. The single most consequential assumption in Meridian's interest rate risk framework is a behavioural parameter with no contract behind it, and it has just moved. This is precisely why the standardised IRRBB framework caps the average repricing maturity that may be assigned to non-maturity deposits — five years for retail transactional, lower for non-transactional retail and for wholesale. The cap exists because the incentive to extend is overwhelming and the evidence base is thin.

The FTP effect is where the governance gets uncomfortable. The deposit business receives its credit from the replicating portfolio yield. Shorten the replicating portfolio from 4.2 to 2.6 years and, on an upward-sloping curve, the credit falls — at exactly the moment the deposit business's actual cost of funds has risen. The business is squeezed from both sides for a change it did not cause. The instinct is to raise the credit to compensate. Resist it, or at least do it consciously: raising the FTP credit rewards a deposit business for a market-wide rate move, converting a transfer price into a subsidy and destroying the information content of the number. If the deposit franchise is genuinely worth defending at a lower margin, that is a strategic decision to be taken, funded and disclosed as such — not smuggled in through the transfer pricing curve.

The order of operations matters. First, re-forecast NII and EVE on realised beta before deciding anything; do not debate remedies against a model everyone in the room knows is wrong. Second, separate the market move from the commercial decision: decide segment by segment whether Meridian is defending volume or defending margin — the answer will differ between insured retail core, less stable retail savings, operational corporate and non-operational corporate, and a single board rate for all four is the error that produced the problem. Third, route the model change through model risk governance with the same rigour as an initial build: backtesting evidence, documented rationale, independent validation, supervisory notification if material. Fourth, change the FTP with a forward effective date and grandfather the back book, so that the transfer price change does not retroactively rewrite business-line performance. Fifth, revisit the structural hedge: if the true behavioural life of the deposit base is shorter than modelled, the receive-fixed position is too long and should be shortened, deliberately and in size.

The trap has two halves. The first is the NII–EVE conflict: lengthening the replicating portfolio improves reported EVE stability in a falling-rate world and worsens it in a rising one; shortening it does the reverse. There is no setting that optimises both, and a bank that quietly re-tunes the parameter each cycle in whichever direction currently flatters the reported number has no interest rate risk framework at all — it has a reporting process. The second half is subtler and more dangerous: recalibrating a model because its output has become inconvenient rather than because its inputs have been shown wrong. The legitimate reason to move beta from 45% to 62% is that observed behaviour says so. The illegitimate reason is that 45% produces an EVE number the board does not like. These two actions look identical in the model documentation and are opposite in kind. The only defence is a governance record that shows which one happened.

Scenario 4 — In a stress, the sequence matters because early actions are reversible

It is Tuesday. On Monday evening a national newspaper published an investigation into Meridian's commercial real estate exposure, alleging under-provisioning in a specific regional portfolio. Over three days the following occurs.

Event Magnitude (illustrative)
Corporate deposit outflow, primarily non-operational €7.0bn
Rating action Outlook to negative from one agency, flagging potential two-notch downgrade
Senior preferred spread move +45bp, to MS+127bp
Collateral call on the swap book €900m cash variation margin
Starting HQLA €62.0bn
Starting LCR 143% (implied net outflows €43.4bn)
Unencumbered non-HQLA collateral pre-positioned at the central bank €18.0bn nominal, ~35% haircut
Committed inbound facilities €2.0bn
Internal survival horizon target, combined stress 90 days
Survival horizon before the event 118 days

The question. What happens to the LCR? What actions are taken, in what order? Which contingency funding plan triggers fire?

Scenario 4

Take the LCR first, because the answer is instructive precisely because it is undramatic. HQLA falls by the €7.0bn of deposit outflow and the €0.9bn of margin, to €54.1bn. The denominator also falls: the €7.0bn that has already left no longer generates a modelled outflow, and at a 40% weight on non-operational non-financial corporate deposits, modelled net outflows reduce by €2.8bn. Add roughly €0.4bn for additional collateral outflow assumptions and net outflows land near €41.0bn. The LCR is therefore approximately €54.1bn / €41.0bn = 132%. The bank has lost €7.9bn of liquid assets in three days, has been publicly accused of under-provisioning, has had its outlook cut and has seen its funding cost rise 45bp — and its headline regulatory ratio has fallen eleven points and remains 32 points above requirement. If you believed the first misconception in Section A, you would conclude that nothing much has happened.

The internal measure tells the real story. Under Meridian's idiosyncratic stress — non-operational corporate run-off at 100% over thirty days, operational corporate at 25%, less stable retail at 15%, stable insured retail at 5%, no rollover of maturing unsecured wholesale, widened repo haircuts — the remaining €33bn of non-operational corporate balances is now the dominant exposure, and the counterbalancing capacity of €54.1bn of HQLA plus roughly €11.7bn of post-haircut pre-positioned collateral plus €2.0bn of committed facilities does not comfortably cover it. The modelled survival horizon falls from 118 days to approximately 41 days, against a 90-day internal target. That is the number that should be on the first page of the crisis pack.

Triggers. On an illustrative CFP ladder, amber fires on any two of: five-day cumulative deposit outflow above €3bn (fired, at €7bn); senior five-year spread more than 25bp above its twenty-day average (fired, at 45bp); adverse rating action including outlook (fired); survival horizon below 60 days (fired, at 41); LCR below 130% (not fired, at 132%). Four of five amber conditions are met. Red requires LCR below 115%, survival horizon below 30 days, an actual downgrade, or five consecutive days without unsecured market access — none yet. Meridian declares amber, which activates the crisis team, daily supervisory contact, and the delegated authorities that make the next steps executable in hours rather than days.

Now the sequence, and the principle that governs it: take reversible actions first, because they are free to undo if the event fades, and irreversible actions last, because each one you take reduces your options and signals distress to the market you may still need. In order. One, within the first hours, freeze discretionary balance sheet consumption — suspend the buyback, halt new committed facility grants, stop non-relationship lending approvals, pause own-paper market-making, defer non-essential intraday payments. All of this is invisible, costless and instantly reversible. Two, monetise Level 1 HQLA through repo rather than outright sale: repo preserves the position, avoids crystallising fair-value losses through OCI, and does not signal. Note the subtlety — repoing government bonds for cash is LCR-neutral, since both legs are Level 1; it solves a cash problem, not a ratio problem. Three, draw on pre-positioned central bank collateral. This is the step that genuinely moves the ratio: €18bn nominal of loan collateral already pledged, at a 35% haircut, yields roughly €11.7bn of cash that was not previously in HQLA, taking the LCR to approximately €65.8bn / €41.0bn = 161% if taken at a tenor beyond thirty days. Regular Eurosystem operations carry no stigma; emergency liquidity assistance is a different instrument in a different part of the ladder. Four, mobilise unencumbered collateral that is not yet pre-positioned — and observe that loan-level pre-positioning takes weeks of legal and data work, which is why it is done in peacetime and why a bank discovering this on Tuesday has already lost. Five, curtail the asset side: reprice, decline, and let facilities run off. Six, and only if the event persists, the irreversible instruments — a retail term deposit campaign at above-market rates, which is public and hard to withdraw; covered bond issuance, which encumbers assets and structurally subordinates the senior investors you will need later; outright asset sales, which crystallise losses and, as March 2023 demonstrated, can be the event that converts a funding problem into a solvency narrative.

Two things run in parallel with all of it. Communication: the supervisory team should hear this from Meridian before it hears it from the market, and the substantive response to the article — the actual provisioning position on the actual portfolio — is a liquidity action, because the outflow is a confidence phenomenon and confidence responds to evidence. And collateral: the two-notch downgrade warning should trigger an immediate recalculation of CSA rating triggers across the derivative book, because a downgrade that crosses a threshold generates additional collateral demands mechanically, on top of anything market-driven, and that number should be known before the downgrade rather than after.

Section C — Four things to do next, all of them concrete

Read your own institution's ILAAP and funding plan, cover to cover, once. Not the executive summary — the scenario definitions, the run-off assumptions by depositor category, the counterbalancing capacity table, the survival horizon calculation and the limit framework. Two hours will teach you more about how your bank actually works than any external course, because it is the one document that connects the balance sheet, the behavioural assumptions and the governance in a single argument. Then read the funding plan alongside it and check whether the two agree about what the bank is going to do. Where they disagree, you have found something worth asking about.

Follow one European bank's issuance for a full quarter and watch the spread against its own secondary curve. Pick an issuer of Meridian's approximate size and rating. Note every deal: instrument, tenor, initial price thoughts, guidance, launch spread, book size, and where the bond trades one week and one month later. You will develop, faster than by any other method, an instinct for what a new issue premium of 10bp versus 25bp actually signals, for how a book builds and collapses, and for the difference between a market that is open and a market that is open to that issuer. Extend the exercise by watching what happens to the curve when the issuer publishes results or receives a rating action.

Read the official post-mortems of March 2023 rather than the commentary about them. The Federal Reserve's review of Silicon Valley Bank supervision, the FDIC's Signature Bank report, the Basel Committee's report on the 2023 banking turmoil, and the Swiss and European supervisory assessments of Credit Suisse. They are long, unsensational and specific, and they contain the operational detail — the hour-by-hour outflow figures, the collateral that could not be mobilised, the governance findings — that the secondary literature strips out. The Basel Committee report in particular is the most useful single document for understanding what the regulatory framework did and did not catch.

Build a cash ladder by hand, once, in a spreadsheet. Take a small illustrative balance sheet — a dozen asset lines, a dozen liability lines, a few derivatives with CSAs — and construct the daily contractual and behavioural cash flow profile out to ninety days, with a counterbalancing capacity block underneath it and haircuts applied. Then shock it: pull 40% of the non-operational corporate deposits over ten days, close the unsecured market, widen the repo haircuts by ten points, and find the day on which the ladder goes negative. Doing this manually once, without a system, is the fastest way to understand what every liquidity report you will ever read is actually doing, and why the assumptions buried in it matter more than the number on the front page.

Process Cheat Sheet — Treasury and Funding on One Page

One reference sheet for the three clocks a Treasury runs against: the funding cycle that plays out over a year, the FTP cycle that repeats every month, and the liquidity day that starts at 07:00 and must be closed by 18:00. Every figure marked indicative is a market-typical order of magnitude, not a rule; every Meridian Bank number is illustrative.

Track A — The funding cycle

Stage Timing Core activity Key output Lead and systems
1. Funding plan and budget Q4 for the following year, reviewed quarterly Project balance sheet growth, maturities rolling off, MREL and NSFR requirements; convert the gap into a euro issuance need by instrument and tenor Approved annual funding plan (Meridian: €14bn, illustrative) with tenor and format split Long-Term Funding, ALM; planning model, ALM system, finance forecast
2. Instrument selection Rolling, refreshed each quarter Match each funding need to the cheapest eligible instrument — covered bond for secured term, senior preferred for NSFR, senior non-preferred for MREL, AT1 and Tier 2 for capital stack gaps Instrument-by-instrument issuance calendar with target size and currency Treasury with Capital Management; funding plan tracker, capital model
3. Documentation and approvals 4–8 weeks before first use; annual programme update Update the EMTN and covered bond programmes, refresh the base prospectus, secure ALCO and board delegated authority, confirm rating agency and legal opinions Live programme, signed delegated authority, board resolution Treasury Legal, Company Secretary; document management, regulator and listing authority filings
4. Market window monitoring Daily, intensifying pre-deal Track index spreads, comparable secondary levels, supply pipeline, central bank and data calendars; identify a clean window with no blackout and no competing supply Go / no-go recommendation with an indicative spread range Treasury Funding with syndicate banks; market data terminals, comparables sheet
5. Execution One day, 07:30 to ~15:00 Announce with initial price thoughts, build the book, revise to guidance, declare books covered, set size, launch and price at the final spread Priced trade: Meridian €1bn 5-year senior preferred, IPT MS+115 area, guidance +90 area, landed +88, book €3.2bn, NIP 3bp (illustrative) Head of Funding with syndicate; order book platform, pricing screens
6. Settlement T+5 (European benchmark convention) Sign subscription agreement, confirm allocations, deliver signed documents and legal opinions, receive proceeds, list the bond, book into the funding ledger Cash in the nostro, bond in the ledger, cash ladder and LCR updated Treasury Operations, Legal; ICSD (Euroclear/Clearstream), settlement and accounting systems
7. Post-issuance lifecycle Continuous to maturity Monitor call dates and MREL eligibility (senior non-preferred loses eligibility inside one year to maturity), run buybacks and tender offers, manage investor relations and disclosure Refreshed maturity profile, MREL headroom, clean liability stack ALM and Treasury Funding; liability register, MREL monitoring tool

Track B — The FTP cycle

Step Cadence Core activity Key output Owner
1. Curve construction Monthly build, daily refresh of the base leg Take the base curve (swap or OIS), add the term liquidity premium at each tenor, add contingent liquidity and other charges Full FTP curve by currency and tenor — Meridian 5-year: 3.67% = 3.05% base + 62bp TLP (illustrative) Treasury ALM quantitative team
2. ALCO approval of the FTP policy Annually, with intra-year amendments Agree the methodology: which curve, how the term premium is derived, how non-maturing deposits are replicated, which charges apply to whom Signed FTP policy with methodology annexes ALCO, chaired by CFO or CEO
3. Application at origination Every contract, at deal capture Stamp each new loan and deposit with the FTP rate for its behavioural tenor at trade date; the rate then locks for the life of the contract Locked transfer rate on every contract in the book Front office booking, enforced by the pricing engine
4. Monthly attribution Monthly close Aggregate FTP charges and credits, split net interest income between the business lines and the Treasury book, isolate the interest-rate and liquidity result held centrally Business line margin reports and the Treasury residual Finance and Performance Management
5. Dispute and governance Monthly, escalating to ALCO Business challenges a rate or an allocation; Treasury evidences the methodology; unresolved cases escalate to ALCO with a documented decision Dispute log with rulings and any policy clarifications Treasury with the ALCO secretariat
6. Annual recalibration Annually, or after a market or model break Re-estimate deposit betas, behavioural maturities, replicating portfolio weights and the term premium against realised data; backtest the prior year Recalibrated parameters, validation report, updated policy ALM with independent Model Validation

Track C — The liquidity day

Time Activity Decision made System
07:00 Overnight reconciliation Confirm the true opening cash position: match expected against actual settlement, identify fails and unexpected flows Nostro reconciliation, camt.053 end-of-day statements
08:30 Cash ladder projection Build the intraday and forward ladder — known maturities, client flows, collateral moves; size today's surplus or shortfall Liquidity management system, cash ladder
10:00 Money market and repo execution Close the gap: place surplus in the deposit facility or reverse repo, cover a shortfall via repo, CP, CD issuance or the central bank Dealing platforms, triparty repo agent, e-trading
12:00 Intraday monitoring Check throughput against the intraday buffer and BCBS 248 monitoring metrics; release or hold queued payments RTGS dashboard, intraday liquidity monitor, camt.052
15:00 Payment cut-off watch Prioritise the payment queue ahead of client and correspondent cut-offs; escalate anything at risk of missing Payment engine, queue management
16:30 Central bank settlement Settle the day in central bank money, place or draw the reserve balance, confirm the minimum reserve trajectory TARGET2 / T2, central bank account
18:00 Close and reporting Lock the closing position, feed the daily LCR and internal limit reports, flag breaches and prepare the next morning's opening view Regulatory reporting engine, limit monitoring, ALM data warehouse

Key numbers

Metric Rule of thumb
LCR minimum 100% — high-quality liquid assets must cover 30 days of stressed net outflows
NSFR minimum 100% — available stable funding must cover required stable funding over a one-year horizon
LCR internal management target Typically 120–150%, held above the minimum so ordinary volatility never touches the regulatory floor (indicative; Meridian runs 143%)
NSFR internal management target Typically 110–120% (indicative; Meridian runs 118%)
Benchmark deal minimum size €500m; €750m–€1bn is the usual size for index inclusion and reliable secondary liquidity (indicative)
Typical benchmark tenors 3, 5, 7 and 10 years for senior; covered bonds stretch to 15–20 years; AT1 is perpetual with a first call at 5–7 years
Bond settlement convention T+5 for European syndicated benchmarks; T+2 for secondary trades
IPT-to-landing compression 20–30bp on a well-supported benchmark (Meridian: 115 area to 88 = 27bp, illustrative)
New issue premium 0–10bp over the issuer's own secondary curve in normal markets; 15–30bp when markets are stressed (indicative)
Oversubscription on a well-received benchmark 2.5x–4x the deal size (Meridian: €3.2bn book on €1bn = 3.2x, illustrative)
Term liquidity premium shape Rises with tenor and flattens beyond 5–7 years; steepest between overnight and 2 years (Meridian 5-year: 62bp, illustrative)
Retail deposit LCR outflow weight 5% for stable insured retail; 10%+ for less stable retail — the cheapest and stickiest funding in the LCR
Corporate non-operational outflow weight 40% where covered by deposit insurance and 100% for financial institution deposits — the reason operational and non-operational corporate balances are priced very differently
Operational deposit outflow weight 25% — balances tied to clearing, custody or cash management that the client cannot easily move
TARGET2 / T2 cut-off 18:00 CET for customer payments (pacs.008); 18:45 CET for interbank payments, with a short late window for the marginal lending facility
Survival horizon expectation Supervisors expect the bank to survive a combined idiosyncratic and market-wide stress for at least 30 days on the regulatory scenario; internal appetite is usually 3–6 months on milder scenarios (indicative)
ILAAP frequency Annual, submitted to the supervisor and assessed as part of SREP; the contingency funding plan is tested at least annually
Asset encumbrance Typically 15–30% of total assets for a European universal bank; above ~40% invites supervisory and rating agency scrutiny (indicative)
Deposit beta 20–50% for retail current accounts, 60–90% for corporate term deposits over a full rate cycle (indicative)

Key formulas

Formula Written out What drives it
LCR LCR = HQLA stock / net cash outflows over 30 calendar days of stress, where net outflows = gross outflows minus inflows capped at 75% of outflows Improves by holding more government bonds and central bank reserves, or by shifting funding from 100%-outflow wholesale deposits to 5%-outflow insured retail. The inflow cap means you cannot rely on loan repayments to fix a weak ratio
NSFR NSFR = available stable funding / required stable funding, both measured over one year with regulatory weights by instrument and residual maturity Improves by lengthening liability maturity beyond one year or by shortening and de-risking the asset side. Overnight wholesale funding contributes zero ASF; a 5-year covered bond contributes 100%
Deposit beta Beta = change in the deposit rate paid / change in the reference market rate, over the same period The share of a policy rate move that reaches the depositor. Low beta means a large, stable earnings cushion when rates rise — and a large repricing risk if competition forces the beta up faster than assumed
FTP all-in rate FTP rate = base curve rate at the contract's behavioural tenor + term liquidity premium + contingent liquidity charge + other charges (optionality, basis, regulatory). Meridian 5-year: 3.05% + 0.62% = 3.67% (illustrative) Tenor and the bank's own credit spread drive it. A longer behavioural maturity means a higher rate charged to a lending business and a higher credit paid to a deposit business
Net interest income sensitivity Delta NII = sum over repricing buckets of (repricing gap in the bucket x rate shock x fraction of the year the gap is open), with behavioural assumptions applied to non-maturing deposits The size and sign of the repricing gap, plus the assumed deposit beta. The beta assumption usually moves the answer more than the gap does
Survival horizon Survival horizon = the number of days until the cumulative counterbalancing capacity (unencumbered HQLA after haircuts, plus committed lines) is exhausted by cumulative stressed net outflows Buffer size after haircuts and the speed of the assumed outflow. Encumbering collateral shortens the horizon without changing the size of the balance sheet
Economic value of equity sensitivity Delta EVE = present value of assets after the shock minus present value of liabilities after the shock, less the same difference before the shock; measured across the six EBA shock scenarios Duration mismatch between assets and liabilities. The outlier test flags a bank whose worst-case delta EVE exceeds 15% of Tier 1 capital

Treasury exists so that every euro the bank lends is funded at a price the business can see, on a maturity the balance sheet can survive, with enough liquid collateral left over to settle every payment on the worst day of the year.