Quick answer: banks optimise collateral by deciding which assets to pledge to which secured-funding, central-bank or margin obligation while respecting eligibility rules, haircuts, currencies, maturities, concentration limits, legal-entity constraints and liquidity requirements. The cheapest asset to pledge is not always the lowest-quality asset, and the highest-value asset is not always the best collateral. A useful optimiser minimises the total economic cost of meeting all obligations, including haircut-adjusted funding needs and the opportunity cost of making a liquid asset unavailable for another purpose.

Collateral has at least three values at once: market value, borrowing value after haircut, and opportunity value in the next place it might be needed.

Why this belongs in mathematics

Collateral optimisation is a constrained-allocation problem. It uses matrices, eligibility filters, haircut transformations, cost functions, linear or mixed-integer programming, liquidity buffers and scenario analysis. It also shows why a balance-sheet asset cannot be valued only by its price: the same government bond may be valuable as HQLA, central-bank collateral, repo collateral, CCP margin or an unencumbered emergency reserve.

The wider eduKate FinanceOS estate already explains collateral, haircuts and forced-sale spirals at system level. This Bukit Timah Tutor article claims a narrower job: how a bank allocates a finite collateral inventory across competing uses.

1. Repo turns securities into secured funding

In a repurchase agreement, one party receives cash and provides securities with an agreement to reverse the transaction later. Economically, the securities protect the cash lender if the borrower fails.

If collateral has market value V and haircut h, a simplified borrowing value is:

Cash capacity ≈ V × (1 − h).

A S$100 million security with a 5% haircut supports roughly S$95 million of cash in this simplified representation. With a 20% haircut, it supports only S$80 million.

The ECB uses the same intuition in its public explainer: a haircut is a valuation discount applied when collateral is accepted, with larger haircuts generally applied to assets with greater price, liquidity or credit risk. See What are haircuts?.

2. Eligibility comes before optimisation

An asset cannot be assigned to a funding route merely because the optimiser likes its cost. The asset must be legally and operationally eligible.

Eligibility can depend on:

asset class;
credit quality;
issuer;
currency;
maturity;
settlement location;
legal owner and entity;
central-bank or CCP rules;
whether the asset is already pledged;
concentration or wrong-way-risk restrictions.

The ECB’s collateral-management framework, for example, distinguishes eligible marketable and non-marketable assets and applies detailed mobilisation, credit-quality and haircut rules. See ECB Collateral Management.

3. Build the eligibility matrix

Suppose the bank has n collateral assets and m obligations. Define:

E_ij = 1 if asset i is eligible for obligation j, and 0 otherwise.

Now define x_ij as the amount of asset i assigned to obligation j.

The optimiser must obey:

x_ij = 0 whenever E_ij = 0.

This simple binary matrix can eliminate thousands of impossible assignments before cost optimisation begins.

4. Haircuts create different effective values at different venues

The same S$100 million bond might receive:

a 2% haircut in one repo market;
a 5% haircut at a central bank;
an 8% haircut for a different counterparty;
no eligibility at all for a particular CCP.

Its effective collateral value is therefore venue-dependent:

Effective value_ij = market value_i × (1 − haircut_ij).

This transforms collateral allocation into a matrix of route-specific capacities rather than one inventory value.

5. A simple allocation example

Suppose a bank needs S$90 million of secured funding and has:

Asset	Market value	Haircut	Effective value
Government bonds	S$60m	2%	S$58.8m
Covered bonds	S$50m	8%	S$46m
Corporate bonds	S$40m	15%	S$34m

If the objective were only “use the lowest haircut first,” the bank might pledge all government bonds and enough covered bonds to meet the remaining requirement.

But suppose the government bonds are the bank’s most valuable unencumbered HQLA and might be needed for a severe liquidity stress. A more intelligent optimiser may preserve some government bonds and use more covered bonds, accepting a larger haircut today to keep emergency liquidity optionality tomorrow.

6. Opportunity cost turns “cheapest collateral” into a harder question

For each asset-route pair, define a total cost:

c_ij = funding cost + haircut cost + liquidity opportunity cost + balance-sheet cost + operational cost.

The bank can then minimise:

Minimise Σ c_ijx_ij

subject to eligibility, inventory, funding, HQLA, concentration and legal-entity constraints.

This is a classic linear-programming structure when costs and constraints are linear. Integer or nonlinear terms appear when collateral must move in indivisible lots, transaction fees have fixed components, or concentration and market-impact functions are nonlinear.

7. Inventory constraints stop the optimiser from pledging the same bond twice

If asset i has available quantity Q_i, then:

Σ_jx_ij ≤ Q_i.

This seems obvious, but real collateral systems must reconcile ownership, settlement, substitutions and encumbrance across many systems. An asset can appear “available” in one ledger while already pledged, pending settlement or trapped in another legal entity.

That connects directly to How Banks Reconcile Transactions: optimisation cannot be more accurate than the inventory ledger it consumes.

8. Encumbrance changes the balance sheet before default happens

An encumbered asset has been pledged or otherwise committed and is no longer freely available for another use. A bank can therefore own many high-quality securities while having only a small unencumbered buffer.

A good optimiser tracks not only total assets but:

available unencumbered collateral;
pledged collateral;
collateral received and reusable where legally permitted;
assets pre-positioned at central banks;
assets reserved for liquidity requirements;
collateral trapped by legal entity, currency or settlement location.

The result is a state-dependent inventory rather than one static balance-sheet number.

9. HQLA creates an explicit opportunity cost

Some assets are valuable because they count toward liquidity buffers such as the Liquidity Coverage Ratio. Pledging them can reduce the unencumbered stock available for liquidity protection, depending on the transaction and regulatory treatment.

ECB research on liquidity transformation shows the trade-off clearly: an asset that is both HQLA-eligible and central-bank-eligible can have different value depending on whether it remains unencumbered or is pledged to generate central-bank reserves. See Liquidity transformation and Eurosystem credit operations.

10. Margin obligations compete with repo funding

A bank may need the same collateral inventory for:

repo borrowing;
CCP initial margin;
derivative variation margin;
central-bank operations;
secured customer obligations;
contingency liquidity.

When volatility rises, margin demands can increase at the same time secured funding becomes more expensive. The optimiser must therefore solve a moving problem under time pressure rather than a once-a-day static allocation.

For CCP margin mechanics, see How Central Counterparties Calculate Margin.

11. Haircuts are safety buffers—and can amplify stress

A haircut protects the cash lender against collateral price decline during close-out. Higher volatility or weaker liquidity can justify a larger haircut. But system-wide haircut increases can force borrowers to find more collateral or repay funding, potentially causing asset sales.

The FSB’s haircut framework for non-centrally cleared securities-financing transactions was designed partly to limit excessive leverage and procyclicality. See the FSB regulatory framework.

Basel also sets minimum haircut-floor treatment for specified securities-financing transactions. See Basel CRE56.

12. The 2019 repo spike is a useful anti-normalisation case

Repo is often treated as ordinary market plumbing until the plumbing becomes scarce. In September 2019, US overnight repo rates rose sharply as Treasury settlement and corporate-tax flows reduced cash while dealers needed to finance large Treasury inventories. The Federal Reserve noted that SOFR rose above 5% and repo operations were used to restore reserves and stabilise rates.

See What Happened in Money Markets in September 2019?.

The lesson for optimisation is subtle: a collateral allocation that is cheap in normal markets may be fragile if it assumes cash, repo capacity or preferred collateral can always be sourced at yesterday’s terms.

13. Wrong-way collateral should carry an extra penalty

If the value of collateral is likely to fall when the borrower or counterparty becomes weaker, the collateral provides less protection exactly when needed. An optimiser that selects only by haircut and funding rate can therefore choose economically dangerous collateral.

A bank can represent this by adding a wrong-way-risk penalty or restricting eligibility for highly correlated issuer/counterparty combinations. The same concept appears in counterparty credit-risk models.

14. Creative-work lens: Margin Call and the value of optionality before everyone sells

Margin Call is not a repo or collateral-optimisation manual, but it makes one resource-allocation truth emotionally visible: an asset can have a quoted value while its executable value changes rapidly when everyone needs the same exit. Collateral optimisation faces the quieter version of that problem every day—how much optionality should the bank preserve before stress makes optionality expensive?

The film supplies the intuition. Haircuts, eligibility, funding costs and inventory constraints supply the evidence.

15. The optimisation pipeline

Build a real-time collateral inventory.
Mark assets to current values.
Map eligibility by venue, counterparty, currency and legal entity.
Apply route-specific haircuts.
Calculate effective collateral value.
Estimate funding and opportunity costs.
Reserve required liquidity buffers and restricted assets.
Model concentration, wrong-way and settlement constraints.
Formulate the allocation problem.
Solve with linear, mixed-integer or other optimisation methods appropriate to the structure.
Generate settlement instructions.
Reconcile actual pledges and substitutions.
Re-optimise when prices, haircuts, margin calls or funding needs change.
Stress the solution for market closure and haircut increases.

16. Failure modes

Haircut-only optimisation. The bank uses the smallest haircut assets without valuing lost liquidity optionality.
Double-pledge data error. Inventory systems disagree about whether an asset is already encumbered.
Static eligibility. Rules change but the optimiser still treats an asset as eligible.
HQLA cannibalisation. Cheap secured funding consumes the assets needed for stress liquidity.
Wrong-way blindness. Collateral quality deteriorates with the counterparty.
Settlement blindness. The mathematically optimal asset cannot be mobilised before the deadline.
Concentration. Too much funding depends on one collateral class or venue.
Normal-market objective. The cost function has no value for survival optionality during stress.

17. Diagnostics and falsifiers

How much effective funding value disappears if haircuts rise across the portfolio?
Which obligations can use only one narrow collateral class?
How much unencumbered HQLA remains after the optimiser runs?
Which asset is cheapest to pledge today but most valuable in a stress scenario?
Can the inventory system prove that every assigned asset is free and deliverable?
Does the solution change materially if repo rates or CCP margin calls move together?
Which legal entity has trapped collateral another entity cannot use?
How long does it take to substitute collateral operationally?

Suppose someone claims, “Government bonds are the best collateral because they have the lowest haircut.” A falsifier is a situation where pledging them destroys more liquidity value than the haircut saving is worth, while another eligible asset could meet the obligation at lower total economic cost. Lowest haircut and optimal allocation are not the same criterion.

18. Verification and update triggers

reconcile collateral inventory across custody, treasury and risk systems;
test every eligibility rule against current venue requirements;
compare predicted and realised funding cost;
stress haircut increases and collateral downgrades;
re-run after margin or liquidity requirements change;
verify settlement and mobilisation times;
review the opportunity-cost assumptions assigned to HQLA;
keep a fallback strategy for optimiser failure or incomplete market data.

Connections across the finance-and-banking algorithms lane

CCP margin algorithms — one major source of collateral demand.
Liquidity stress testing — why collateral optionality matters beyond ordinary funding cost.
Counterparty credit risk — where collateral reduces exposure but creates wrong-way and close-out questions.
FX settlement — another area where time-to-deliver can matter as much as nominal value.

Research anchors

The deeper lesson

Collateral optimisation is the mathematics of preserving choices under constraints. Haircuts determine how much cash an asset can raise. Eligibility determines where it can go. Encumbrance determines whether it is still free. Opportunity cost determines what future option is lost when it is pledged today. A strong optimiser therefore does not merely ask, “What collateral can satisfy this obligation?” It asks, “Which assignment leaves the whole bank strongest after every obligation is satisfied?”

Educational note: This article explains public banking and collateral-management concepts. It is not funding advice, trading advice, legal advice or instructions for managing a specific institution’s collateral.

Small Group Tutorials

How Banks Optimise Collateral for Repo and Margin: Eligibility, Haircuts, Funding Value, Encumbrance and Linear Programming