Last look and no-last-look: the mechanism, not the villain

Last-Look is probably the most poorly explained mechanism in the FX market. It's presented as a trap or defended as a harmless standard, and both positions dodge the interesting part: it's a tool that can be used well or badly, and there are specific questions that tell one case from the other.

What it is, mechanically

When a provider quotes a price and you accept it, Last-Look is a time window — typically a few milliseconds — during which that provider can confirm or reject the transaction after you've already committed.

That is: you accept its price, and it takes one last look before deciding whether to honor it.

Put that way it sounds like the deal is skewed, and in a sense it is. The question is why it exists and what's done with that window.

Why it exists

The technical justification is real: protection against latency.

A provider quotes to thousands of counterparties simultaneously. Between publishing a price and your acceptance reaching it, milliseconds pass, and in that gap the market may have moved. Without Last-Look, anyone with a faster connection could systematically accept prices that are already stale, at the provider's expense.

That behavior — accepting stale prices on purpose — is called latency arbitrage, and with no defense at all it would make quoting aggressively unviable. The practical consequence would be wider spreads for everyone.

Last-Look is the defense. The problem isn't its existence: it's what's done with it.

Where the abuse starts

There are three uses that turn a defensive mechanism into value extraction:

Asymmetry. The provider rejects when the price moved against it but executes when it moved in its favor. The window should apply equally in both directions; when it doesn't, it's a free option at the client's expense.

Long windows. A few milliseconds is defense. Tens of milliseconds is something else: enough time to observe where the market is heading before deciding.

Pre-hedging during the window. The provider uses knowledge of your order to position itself before confirming it. You revealed your intention and it used it before committing.

These three practices are what gave the mechanism its bad name, and rightly so. But none is intrinsic to Last-Look: they're implementation decisions.

The question that settles it

The entire evaluation comes down to one thing:

Do they publish the rejection rate, is it symmetric, and how long is the window?

A provider that uses Last-Look and publishes its rejection rate by upstream counterparty is operating to institutional standards. One that uses it without publishing that isn't giving you enough information to evaluate whether it suits you.

Transparency is the criterion, not the mechanism.

Specific questions worth asking:

What's the duration of the window, in milliseconds?
Is the rejection policy symmetric with respect to the direction of movement?
Do you pre-hedge during the window?
Do you publish the rejection rate broken down by upstream and by instrument?
Can I opt for No-Last-Look, and what does it cost me?

What No-Last-Look really costs

No-Last-Look — where the provider commits firm with no review window — exists and is usually available. It isn't free, and the cost has logic.

Without Last-Look, the provider takes on the risk of being handed stale prices. It compensates in the only way possible: by quoting a bit wider. It's insurance, and like all insurance it has a premium.

The right decision depends on your profile:

No-Last-Look suits you if: you trade large sizes where execution certainty is worth more than a few tenths of a pip, or if your strategy breaks with unpredictable rejections.

Last-Look suits you if: you trade high volume in moderate sizes and the tighter spread offsets the occasional rejections — provided the rate is known and symmetric.

Mixed suits you: which is what most serious operations do. No-Last-Look in the instruments and sizes where certainty matters, Last-Look where the spread rules.

There's no universal answer, but there is a correct answer for your specific book, and it's calculated with your own data: expected cost of rejections against expected saving on spread.

How to measure it before deciding

With a production-parity sandbox you can run the calculation instead of estimating it:

Send representative test flow under both configurations
Measure the average effective spread in each
Measure the rejection rate and the retry slippage under Last-Look
Compare it during a news window, not just in a flat market

The result usually surprises: for many books, No-Last-Look comes out cheaper than the nominal spread difference suggests, because the cost of retries in a moving market weighs more than expected.

A note on asymmetry

If I had to keep a single red flag, it would be this: a rejection rate that rises exactly when the price moves in your favor.

It's measurable with your own logs. Record, for each rejection, the direction the market moved during the window. If rejections concentrate in the movements favorable to you, the window isn't being used as defense: it's being used as an option.

That analysis requires no one's permission and is the most direct way to know who you're dealing with.

Frequently asked questions

Is Last-Look legal? Yes, and it's a widespread practice in wholesale FX. What's under regulatory and industry code-of-conduct scrutiny is its asymmetric use and pre-hedging during the window, not the mechanism itself.

Do Tier-1 banks use Last-Look? Many do, in their electronic flows. It's one of the reasons an aggregator connected to banks has some rejection rate: it inherits it from its sources.

How do I know the duration of the window if they don't tell me? It can be estimated by measuring the time between the send and the rejection response in your own logs. It's not exact — it includes network latency — but a consistent pattern of tens of milliseconds is informative in itself.

Can I negotiate the policy? At the institutional level, often yes: the choice between Last-Look and No-Last-Look, even by instrument group, is usually configurable. If they tell you it isn't, either the infrastructure is rigid or they don't want you to choose.

Related: fill ratio and rejection rate for the detail on the metrics, and how a broker chooses liquidity for the full framework.

At Exura Prime the Last-Look and No-Last-Look options are configured by instrument class, with the rejection policy agreed in writing before production rather than discovered in it. Clients onboard and execute through us; the technical detail is in FIX API.