Why Exchanges Deliberately Slow Some Orders Down

The Race You Never See

Picture two chai stalls on the same street. One raises its price the moment the milk delivery arrives. The other is still scribbling the old price on its board. If you could somehow know the new price before the second stall catches up, you could buy cheap there and sell at the new rate a moment later. You didn't predict anything. You just saw the news first.

That, in essence, is latency arbitrage. "Latency" simply means delay — the time it takes information to travel from one place to another. In markets, the same stock often trades on several exchanges at once. When the price moves on one venue, the others don't update in the same instant. For a few microseconds — millionths of a second — the quotes sitting on the slower venues are stale. They show a price that no longer exists.

A firm with faster technology can spot the change at the source, send an order to the lagging venue, and trade at the outdated price before anyone there updates the board. The profit is the gap between the old price and the new one. As one specialist source puts it, this isn't about predicting where prices will go — it's about seeing where they've already gone somewhere else.

Why the Gap Exists at All

Prices can't be everywhere at once. Even light through fibre optic cable moves at roughly 200,000 kilometres per second — fast, but finite. Add the time data feeds need to process information, the hops between networks, and the queue inside each exchange's matching engine, and you get windows measured in microseconds to milliseconds.

These windows matter most in markets where liquidity is scattered across many venues rather than centralised in one. Forex, equities, and cryptocurrency all fit that description. A study using participant-level regulatory data found that at least 4% of dark trading — trading on venues with no pre-trade transparency — happened at stale reference prices. Dark pools are especially exposed because they often price trades by pegging them to an external reference market. If that reference moves and the dark pool hasn't caught up, the peg is pointing at a price that's gone.

Firms that specialise in speed almost always end up on the profitable side of those stale trades — between 96% and 99% of the time in one study's estimates. The pattern isn't random, either. It's driven by the behaviour of the fastest participants.

What It Takes to Win the Race

Competing at this level is expensive, and that's part of the story. Firms rent rack space inside the same data centre as an exchange's matching engine — a practice called co-location — so their servers sit metres from the machinery doing the matching. They build private microwave and millimetre-wave networks between cities, because radio waves through air travel in a straighter line than light bouncing along glass. They use FPGAs — field-programmable gate arrays, chips whose logic can be rewired for one specific job — to parse market data and fire orders in under a microsecond.

None of this produces a product you can hold. It's an arms race over time itself, and the returns on each additional nanosecond shrink as the physics gets harder. Going from ten microseconds to five is manageable. Going from one microsecond to five hundred nanoseconds demands custom silicon and a very patient chief financial officer.

The Case for Slowing Things Down

If speed creates the problem, one response is to deliberately remove the reward for it. That's the thinking behind a speed bump: an intentional delay applied to incoming orders at an exchange. The Investors Exchange, known as IEX, built one from a coil of optical fibre — tens of miles of it, boxed up — that adds 350 microseconds before an order reaches the matching engine. That's roughly a thousandth of the time it takes you to blink.

During that sliver of time, the exchange can absorb current market data from other venues and refresh its own prices. By the time the would-be arbitrageur's order arrives, the stale quote it was aiming at has been corrected. Ordinary investors never notice. The firm that spent a fortune shaving microseconds off its pipeline very much does.

IEX also runs something called the Signal, nicknamed the Crumbling Quote Indicator. Rather than treating a price change as one clean event, it treats it as a sequence — a row of dominoes starting to topple. When the Signal judges that the best bid or offer is likely about to move, certain order types automatically become less aggressive for about two milliseconds, so they aren't picked off mid-tumble.

Other Ways Venues Even the Field

Speed bumps aren't the only tool. Some venues use periodic batch auctions: instead of matching orders continuously, they collect them briefly and match everyone at a single price. If arrival time no longer decides who wins, there's nothing to gain from being microseconds faster. Others add small random delays to order processing, so being fastest no longer guarantees being first. On the broker side, some liquidity providers use "last-look" windows of roughly 10 to 200 milliseconds, during which they can check whether a requested price still reflects current conditions before agreeing to a trade.

Each of these is a market-design choice, not a moral verdict. And they're contested. Critics argue that cancelling or adjusting quotes can leave ordinary orders unfilled or filled at worse prices. Supporters counter that latency arbitrage amounts to a tax on liquidity — a cost passed along to everyone slower, including pension funds and retirement accounts. Regulators have studied the question and generally leaned toward market-structure fixes rather than outright bans, though findings vary by issue and jurisdiction.

Where It Gets Murky

Latency arbitrage as usually described is not the same as front-running. Classic front-running means trading ahead of a client's order using knowledge you were trusted with — prohibited in many jurisdictions. Latency arbitrage reacts to publicly available prices. But the line isn't perfectly crisp, and how any specific firm's behaviour is judged depends on the facts, the venue's rules, and where you are. Regulators including the SEC, the FCA, and ESMA have looked at speed-sensitive trading; outcomes have ranged from guidance to enforcement in particular conduct cases to structural reform — not one blanket rule covering every situation.

One study estimated latency arbitrage revenue in UK equities alone at hundreds of millions of pounds a year, with global figures reaching into the billions. Those estimates are debated, and some researchers argue the costs are overstated because the races might simply be won by a different fast firm. Reasonable people disagree here, which is worth remembering when someone states a number with total confidence.

Why You'd Care

You will probably never place a microsecond-sensitive order, and that's fine. But the design of these mechanisms quietly shapes the spreads you pay, the prices you get, and how much confidence people have in markets generally. The debate isn't really about whether 350 microseconds matters to you personally. It's about whether a market should reward whoever owns the fastest computer, or whether it should be arranged so that speed confers no special advantage at all.

That question has no settled answer. What's clear is that exchanges are now experimenting with the answer — some by adding delays, some by batching, some by doing nothing and letting the race continue. Watching which approach holds up is one of the more interesting unfinished stories in market structure.