Quick answer: derivatives portfolio compression is a post-trade optimisation process that tries to remove economically redundant contracts without materially changing the market risk that participants intend to keep. In the simplest case, equal-and-opposite trades can be torn up. In more complex multilateral exercises, many firms submit portfolios and risk tolerances, and an algorithm finds a smaller set of surviving or replacement trades that preserves agreed risk characteristics while reducing contract count, gross notional, counterparty links or other non-market-risk burdens.

Compression asks a surprisingly mathematical question: how much of a derivatives network is real economic risk, and how much is merely redundant plumbing created by the history of trading?

Page role: what this article owns

Bukit Timah Tutor already explains transaction reconciliation, which asks whether records agree; counterparty credit risk, which asks how much exposure exists; and collateral optimisation, which allocates assets against requirements.

This article owns a different job: given a valid population of derivatives, how can an algorithm remove redundant contractual structure while keeping intended market risk within pre-agreed tolerances?

1. A three-trade example shows why gross notional can be misleading

Start with a deliberately simple network:

Bank A pays Bank B S$100 of the same future cash flow.
Bank B pays Bank C S$100 of the same future cash flow.
Bank C pays Bank A S$100 of the same future cash flow.

The network has S$300 of gross contractual flow, but every participant’s net position is zero. If the obligations are genuinely identical in currency, date, index, legal conditions and economic effect, the cycle contains no net market exposure.

Graphically, the trades form a directed cycle:

A → B → C → A

A compression algorithm tries to detect structures like this and remove them, subject to legal, operational and risk constraints.

2. The graph representation

A derivatives portfolio can be represented as a network. Participants are nodes. Contracts or exposure components are edges. A directed edge from i to j can represent a payment obligation, a risk transfer or an economically signed exposure.

For a single homogeneous instrument, let x_ij denote gross notional from participant i to participant j. The net position of participant i can be written schematically as:

n_i = Σ_j x_ji − Σ_j x_ij

If compression preserves every n_i while reducing Σ|x_ij|, then gross contractual structure has fallen without changing the simple net exposure vector.

Real derivatives are far more complex because one trade can carry many risk dimensions. The vector that must be preserved may include rate delta by tenor, basis risk, FX delta, vega, cash-flow timing, collateral effects and counterparty constraints.

3. Exact compression versus risk-tolerant compression

There are two useful conceptual cases.

Exact compression: redundant trades can be terminated without changing the chosen risk measures at all.
Risk-tolerant compression: the algorithm may create replacement trades or slightly alter the portfolio, but only within participant-specified tolerances.

The second case is an optimisation problem. A participant may allow, for example, a tiny change in five-year rate delta but no change beyond a tighter tolerance in ten-year delta. Another may forbid new exposure to a particular counterparty. The feasible solution must satisfy all such constraints simultaneously.

4. A generic optimisation formulation

Let y denote the post-compression trade vector. A simplified optimisation can be written as:

minimise G(y)

subject to:

|R(y) − R(x)| ≤ ε for agreed market-risk measures;
legal and counterparty eligibility constraints;
product, maturity and clearing constraints;
participant-specific limits;
valuation or cash-flow tolerances.

Here x is the original portfolio, R(.) is a vector of risk measures, ε is the allowed tolerance and G(y) is an objective such as gross notional, number of trades, capital usage or a weighted combination of non-market-risk costs.

The algorithm is not asking “what is the best trade?” It is asking “what is the simplest equivalent network that stays inside the agreed feasible region?”

5. Why zero net notional is not enough

Consider two interest-rate swaps with equal notional and opposite direction. One matures in two years and the other in twenty years. Their notionals cancel arithmetically, but their interest-rate risk does not.

This falsifies a naive compression rule:

equal positive notional + equal negative notional ≠ zero economic risk.

Compression must therefore preserve the relevant risk representation, not merely a scalar total.

6. Risk vectors can be high-dimensional

A swap portfolio can be mapped to a vector such as:

R = (DV01_2y, DV01_5y, DV01_10y, basis, vega, FX, …)

A compression service may preserve each component exactly or within defined tolerances. Increasing the number of risk dimensions makes the feasible optimisation harder, but also makes the notion of “equivalence” more faithful to the real portfolio.

This produces a classic modelling trade-off: a coarse risk vector creates more compression opportunities but may hide important differences; a very fine risk vector preserves more detail but may leave fewer trades removable.

7. Bilateral and multilateral compression are different problems

In bilateral compression, two counterparties compare their mutual portfolio and remove offsetting or redundant trades. The opportunity set is limited to that pair.

Multilateral compression allows many participants to submit portfolios into one exercise. This can reveal cycles and replacement structures invisible to any single pair. The Bank of England’s December 2025 post-trade risk-reduction consultation notes that the effectiveness of compression depends on the number of participants, the size of submitted portfolios and the tolerances chosen.

Network size therefore has two opposing effects: larger systems create harder optimisation problems, but they also create more possible routes to remove redundant structure.

8. Compression, rebalancing and basis-risk optimisation are not the same thing

Public regulatory material distinguishes several post-trade risk-reduction services.

Portfolio compression primarily reduces contract count or notional by eliminating or replacing redundant trades without materially changing market risk.
Portfolio rebalancing can insert new non-price-forming trades to redistribute non-market risks across the network while maintaining market neutrality.
Basis-risk optimisation targets imperfect hedges and seeks mutually beneficial trades that reduce basis mismatches.

These activities are related but should not be collapsed into one algorithmic label. Their objectives and constraints differ.

9. Why compression can reduce more than trade count

A smaller derivatives network can reduce several burdens:

fewer records to reconcile;
fewer lifecycle events to process;
fewer payment and settlement instructions;
lower gross notional in some structures;
less operational complexity;
potentially lower counterparty or margin burdens, depending on the structure;
simpler default-management and valuation populations.

But none of these benefits is automatic. A replacement trade can reduce one metric while worsening another. Good optimisation therefore needs a clearly defined objective function rather than a vague instruction to “compress as much as possible.”

10. A counterexample: lower notional can create a worse counterparty shape

Suppose an algorithm reduces total gross notional by 30% but concentrates the surviving exposure with one counterparty. Operational complexity falls, yet counterparty concentration rises.

This falsifies the claim that maximum notional reduction is always the best compression outcome. The objective may need penalties or hard limits for counterparty concentration, liquidity, margin, clearing status and other constraints.

11. Valuation agreement is a hidden dependency

Two parties may agree that two trades look economically similar but disagree on value because they use different curves, market data, models or conventions. A compression exercise that creates replacement trades must therefore control valuation differences and define tolerances clearly.

If the algorithm assumes one common valuation world but participants settle in different valuation worlds, the apparent risk-neutral solution may produce transfers one party considers unacceptable.

12. Legal and operational feasibility constrain the mathematical optimum

The unconstrained mathematical optimum might connect every participant to every other participant. The legal and operational optimum cannot.

Some counterparties may not have legal documentation with one another.
Some products may require central clearing.
Some books may be segregated by entity or jurisdiction.
Some participants may reject certain maturities or product forms.
Some trades may be operationally locked or outside the submitted population.

These constraints define the feasible set. An optimisation algorithm that ignores them can produce a mathematically elegant but unexecutable solution.

13. Inputs and outputs

Typical inputs: validated trade populations, participant identities, legal relationships, notionals, cash-flow dates, product terms, market values, risk sensitivities, clearing status, collateral/netting information and participant risk tolerances.

Optimisation inputs: objective weights, forbidden edges, permitted replacement instruments, maximum risk deviations and execution constraints.

Outputs: trades to terminate, surviving trades, possible replacement trades, before/after gross notional, trade-count reduction, risk-vector changes, counterparty-network changes and a reconciliation/audit record.

14. Assumptions and weak links

Population completeness: missing trades create false compression opportunities or missed offsets.
Economic equivalence: a risk vector may omit a factor that matters under stress.
Valuation consistency: replacement trades depend on agreed pricing conventions.
Tolerance design: wide tolerances increase compression but can permit unwanted risk drift.
Legal feasibility: a proposed counterparty edge may not be executable.
Operational readiness: terminated and new trades must flow correctly to booking, collateral, settlement and reporting systems.
Objective choice: minimising notional alone can worsen concentration or liquidity.
Data timing: a stale portfolio snapshot can be obsolete by the time the exercise is executed.

15. Failure modes

False cycle: trades appear identical by notional but differ by maturity, index or legal terms.
Risk leakage: the compression preserves delta but materially changes vega or basis risk.
Concentration transfer: notional falls but exposure becomes concentrated in fewer counterparties.
Unbookable optimum: replacement trades violate documentation or product eligibility constraints.
Valuation dispute: participants disagree on replacement values after the optimisation has solved.
Stale snapshot: new trades arrive after the portfolio cut and invalidate intended offsets.
Reconciliation break: one participant terminates a trade that another participant still carries.

16. Diagnostics and falsifiers

Recalculate all preserved risk measures before and after compression.
Measure gross notional, trade count, counterparty degree and concentration before and after.
Replay the exercise with tighter tolerances. Which compression benefits disappear first?
Run an independent pricing and sensitivity check on replacement trades.
Verify that every proposed counterparty pair has the required documentation and product permissions.
Reconcile terminated and replacement trades across all participants before settlement.
Stress the post-compression portfolio under scenarios not included in the optimisation risk vector.

A useful falsifier for “market risk was preserved” is a stress scenario that produces a material pre/post P&L difference even though the optimisation reported near-zero risk change. That indicates the chosen risk representation omitted an important dimension.

17. Alternatives and limits

Compression is not the same as hedging. Hedging intentionally changes or offsets market risk using new positions. Compression primarily removes redundant contractual structure while keeping intended market risk substantially unchanged.

Nor is compression always appropriate. If a portfolio is small, highly bespoke, legally fragmented or already operationally simple, the implementation cost may exceed the benefit. A highly compressed network can also reduce gross complexity while increasing concentration in ways that require separate monitoring.

Alternatives or complements include bilateral tear-ups, novation, clearing, portfolio rebalancing, basis-risk optimisation, hedging and collateral optimisation. Each changes a different part of the system.

18. Verification and update triggers

rapid growth in trade count or gross notional;
material changes in counterparty network structure;
new product types or risk factors;
changes in clearing requirements or legal documentation;
persistent portfolio-reconciliation breaks;
new valuation-model conventions;
large changes in margin or capital after compression;
evidence that post-compression stress risk differs materially from the pre-compression portfolio.

Current public regulatory context

In the United States, CFTC rules require swap dealers and major swap participants to maintain policies and procedures for bilateral and multilateral portfolio compression when appropriate. In the United Kingdom, the Bank of England’s 11 December 2025 consultation on post-trade risk-reduction services described compression as reducing the number of contracts and total notional outstanding without materially affecting market risk, and distinguished it from portfolio rebalancing and basis-risk optimisation. The consultation closed on 11 March 2026; this article uses it as a public description of the mechanism rather than as a claim about the final legal status of every proposed exemption.

Connections across the finance-algorithms lane

Interest-rate swap valuation — the pricing and sensitivity layer needed to judge economic equivalence.
Counterparty credit risk — one burden compression may reduce or redistribute.
Transaction reconciliation — the record-integrity layer compression depends on before and after execution.
Collateral optimisation — a separate optimisation over assets rather than over the trade network itself.

Research anchors

The deeper mathematical lesson

Portfolio compression is a clean example of optimisation over a network. It separates state from representation. The market risk a participant wants to keep is the state. Thousands of bilateral contracts are one possible representation of that state. Compression searches for a smaller representation without crossing the boundaries that define equivalence.

The chain is:

validated trades → network representation → risk vector → tolerances → feasible set → optimisation → replacement/termination set → independent verification.

A smaller answer is only better if it still represents the same thing.

Educational boundary: This article explains public derivatives post-trade mathematics and operational logic. It is not trading advice, investment advice, legal advice or a recommendation to compress any specific portfolio.

Small Group Tutorials

How Derivatives Portfolio-Compression Algorithms Remove Redundant Trades: Graphs, Netting, Risk Tolerances and Optimisation