What LIBOR rigging teaches us about benchmark design

Submission-based benchmarks asked parties holding positions in the output to supply the inputs, which is a design flaw that was described in academic literature before it appeared in an indictment. The reforms that worked replaced estimates with observed transactions; the enforcement, which produced billions in penalties, addressed the conduct rather than the vulnerability.

Published 2026-06-24 · 11 min read

The interest rate benchmark scandals produced some of the largest financial penalties ever imposed. This library records the five-bank FX action at $1.4 billion, the RBS matter at $325 million, and the Société Générale action at $475 million. The full benchmark facet contains the aggregate.

The lesson usually drawn is about culture. The more useful lesson is about design, and it was available in advance.

The flaw, stated simply

A submission-based benchmark is calculated from figures panel members report rather than from observed transactions. Members submit an estimate — typically of the rate at which they could borrow — an administrator trims the extremes and averages the rest, and the result is published as an official rate. Contracts worldwide settle against it.

The vulnerability is a single sentence: the parties supplying the inputs hold positions in the output.

That is not a subtle problem, and it was not discovered by prosecutors. It had been described in the academic literature well before it appeared in an indictment. What was missing was not analysis but the assumption that the conflict would be managed by professional norms.

Why the trimming defence fails

The standard reassurance about these benchmarks was that trimming protected them. Discard the highest and lowest submissions, average the middle, and no single submitter can move the result much.

Two problems.

Trimming does not remove a submission that survives the trim. A submitter who understands the likely distribution can position their figure just inside the boundary — skewed, but not skewed enough to be discarded. It still pulls the average.

Take sixteen panel members with the top four and bottom four discarded. One bank moving its surviving submission by a basis point moves the published rate by one eighth of that. On $12 billion of notional over a three-month period:

$12,000,000,000 × 0.0000125 × (90/360) = $37,500

Thirty-seven thousand dollars, from one submission, on one day. Individually unremarkable — which is precisely the point, because nothing about it looks worth investigating.

Coordination defeats trimming entirely. If four members move together, the trim removes the honest outliers instead. And coordination makes each individual submission less conspicuous, because the submissions no longer diverge from one another.

That inversion — where colluding makes you harder rather than easier to detect — is the feature that made the conduct persist for years.

Why surveillance failed and documents succeeded

This is the family where market surveillance performed worst of anything on this site, and it is worth understanding why rather than attributing it to inattention.

The statistical case against a submission is weak by construction. A submission is an estimate. In stressed conditions with few observable transactions, honest estimates vary widely and legitimately. Comparing a bank’s submission against its own funding costs works only when those costs are observable, and during the periods that mattered most they frequently were not.

So what broke the cases open was not data. It was the participants’ own messages: requests from traders to submitters, and coordination between traders at competing institutions, in writing, on systems their employers retained.

Two consequences followed, and both are still with us.

Communications surveillance expanded enormously. The subsequent enforcement against firms for allowing business to be conducted on unmonitored messaging channels — producing very large penalties entirely unrelated to any manipulation — follows directly from this experience.

Where a technique’s statistical signature is weak, the record of people arranging it is what remains. That generalises well beyond benchmarks.

The reforms that actually worked

The response divided into two kinds, and only one of them addressed the vulnerability.

Enforcement produced billions in penalties, criminal prosecutions in several jurisdictions, and outcomes for individuals that varied considerably — a number of convictions were subsequently overturned on appeal. It punished the conduct. It did not change the design.

Structural reform changed the design, and it is the part that worked.

• Transaction-based benchmarks. Replacement rates are calculated from large volumes of observed overnight transactions rather than from estimates. There is no discretionary input left to skew, which removes the attack surface rather than policing it.
• Separation of submission from trading. In many institutions, submitters had sat on the same desk as the traders whose positions depended on the result. That was treated as a control failure independent of any specific submission.
• Regulated benchmark administration. Benchmark contribution and administration became regulated activities in their own right in several jurisdictions, with dedicated offences.
• Comprehensive communications retention and monitoring.

The first of these is the one that matters. A benchmark computed from observed transactions is not harder to manipulate; it is manipulable only by actually trading, which is expensive and visible.

What remains vulnerable

The reformed headline rates are substantially more robust. The residual risk sits where the same design features persist.

Benchmarks still involving judgement. Any rate computed from expert estimates rather than observations retains the flaw. Several exist, particularly at longer tenors and in less liquid markets.

Benchmarks computed from thin underlying markets. Transaction-based is not sufficient if the transactions are few. A rate derived from a handful of daily trades can be moved by trading, which converts the attack from submission rigging into settlement price manipulation — a different technique addressing the same objective.

Short determination windows. FX fixing demonstrates this. The rate is transaction-based, and it was manipulated anyway, because compressing price determination into a short window creates a well-defined and affordable target. The response there — lengthening the window substantially — is the same kind of structural fix.

Crypto reference prices. Index prices for digital assets are frequently computed from a small number of venues whose own volume statistics are unverified. This combines a thin underlying market with an unauditable input, which is the original problem with additional steps.

The generalisable lesson

Wherever a small market determines the value of a much larger set of contracts, the arithmetic attracts someone. That ratio — cost of moving the reference against value of moving it — is the thing to look at, and it is usually calculable in advance.

It is the same ratio that drives banging the close, where trading in a five-minute settlement window determines marks across an entire market. The same ratio drives cash versus derivatives schemes, where a small spot market prices a large derivative complex. The HSBC matter in this library carries a $1.73 billion penalty for conduct in that category.

The benchmark scandals were the most expensive demonstration of a general principle: a price that matters more than it costs to move will eventually be moved. The durable defence is to make the price cost more to move — more volume, longer windows, observed transactions, multiple sources — rather than to rely on the people who could move it choosing not to.

That is a design question, and it is answerable before the enforcement action rather than after.

Who was actually harmed, and why restitution was hard

A striking feature of the benchmark resolutions is how much went to penalties and how little to compensating anyone.

The reason is not indifference. It is that identifying victims of a benchmark manipulation is genuinely close to impossible.

A submission moved a published rate by a fraction of a basis point on a particular day. Every contract resetting against that rate on that day paid or received slightly differently than it would have. Some of those counterparties were harmed and some benefited, depending on which side of the contract they were on — and the same institution could be on both sides across its book.

To compute an individual’s loss you would need to know their exposure on each affected date, the direction of the manipulation on each of those dates, and the counterfactual rate. The first is private, the second varied, and the third is contested.

The consequence is that these matters resolved as penalties paid to governments rather than as compensation paid to counterparties, with private litigation running separately and reaching mixed outcomes. That is unsatisfying, and it is a general property of manipulation harm: where the injury is a distorted market-wide price, the victims are everyone and therefore nobody in particular.

It is also why Fair Funds — the mechanism allowing penalties to be distributed to harmed investors — are used far less in manipulation cases than in ordinary fraud cases.

The individual outcomes diverged from the institutional ones

Something else in this record deserves attention, because it is unusual.

Institutions settled, in very large amounts, with detailed statements of facts. Individuals were prosecuted criminally in several jurisdictions, and the outcomes varied considerably — a number of convictions were subsequently overturned on appeal.

That divergence is worth sitting with rather than explaining away. It reflects genuine difficulty about what the submission obligation actually required. A submission was an estimate, and where the guidance about how estimates should be formed was unclear, establishing beyond reasonable doubt that a particular submitter knew their figure was dishonest proved harder than the settlements implied.

The lesson is not that the conduct was acceptable. It is that a rule everyone assumed was clear turned out, under adversarial examination, to be less clear than the settlements suggested — and that institutional settlements resolved without admissions do not establish what individuals knew.

This is a recurring feature of financial enforcement and a reason to read settled matters carefully. Our editorial policy treats a settlement as an outcome rather than a finding for exactly this reason.

A checklist for any benchmark

If you want to assess whether a reference price is vulnerable, four questions cover most of it.

Is the input observed or estimated? Estimates can be skewed at no cost. Observations require trading.

How much volume underlies it? A transaction-based rate computed from a handful of daily trades is transaction-based in name only.

How long is the determination window? Compressing price formation into minutes creates a well-defined and affordable target, which is what FX fixing demonstrated even for a transaction-based rate.

Who supplies the input, and what do they hold? The original flaw, and the one that generalises furthest. Any arrangement where the parties supplying inputs hold positions in the output will eventually be tested.

For the mechanics: benchmark submission rigging, FX fixing, and the benchmark family.

Techniques referenced

• Benchmark submission rigging — Benchmark submission rigging is skewing the figures a panel member reports into a reference rate calculation, so that the published benchmark moves in favour of positions that settle against it.
• FX fixing — FX fixing manipulation is trading during the short window used to calculate a daily currency benchmark, using advance knowledge of client orders that must be executed at that fix.
• Settlement price manipulation — Settlement price manipulation is trading to move an official settlement price, because that price determines margin, portfolio marks and the payoff on contracts many times larger than the trading itself.
• Banging the close — Banging the close is trading heavily during the short window used to calculate a settlement price, in order to move that settlement in favour of a much larger position that references it.
• Cash versus derivatives schemes — A cash versus derivatives scheme establishes a large derivative position and then trades the smaller underlying market at a deliberate loss, in order to move the reference price the derivative settles against.

Cases referenced

Reviewed June 24, 2026. Spotted an error? Tell us.