Latency Arbitrage

The same stock trades on multiple exchanges at once, and prices should stay in lockstep — if they diverge for more than an instant, someone will buy on the cheap venue and sell on the expensive one. In practice they do diverge, constantly, for tiny fractions of a second, because news doesn't reach every venue's matching engine at exactly the same moment. Latency arbitrage is trading that gap: seeing a price move on venue A before venue B's quotes have caught up, and trading against B's now-stale price before it updates.

The setup

Suppose a stock trades on both NYSE and a regional exchange, currently quoted 100.00/100.02 on both. A large sell order sweeps NYSE's bid down to 99.90 — public information the instant it happens, broadcast by NYSE's data feed. The regional exchange's quote doesn't move on its own; it only updates when someone sends it a new order or cancellation, or when its own systems process the same public news. For a window of time — often just a few hundred microseconds to a few milliseconds — the regional exchange is still showing a 100.00 bid that no longer reflects reality.

A firm watching NYSE's feed, faster than the regional exchange's other participants at reacting, sells into the regional exchange's stale 100.00 bid the instant it sees NYSE trade down, locking in the difference before the regional book catches up and someone posts a market maker's own cancel.

Worked example

• $t_0$: NYSE prints a sweep, new NYSE mid is 99.91.
• $t_0 + 200\mu s$: A latency-arbitrage firm's system has seen the NYSE print, confirmed the regional exchange's bid is still resting at 100.00 for 1,000 shares, and sends a marketable sell for 1,000 shares to the regional exchange.
• $t_0 + 350\mu s$: the sell order arrives and fills at 100.00 — the regional exchange's own market maker hasn't cancelled yet.
• $t_0 + 900\mu s$: the regional exchange's quote finally updates to 99.91-ish, roughly matching NYSE.

The arb firm sold 1,000 shares at 100.00 into a market that, 550 microseconds later, agreed the fair price was closer to 99.91 — a gain of roughly $0.09/share, $90 on the trade, extracted purely from being faster to react to public information than the regional market maker was to defend its own quote.

Who pays for it

The regional exchange's own market maker is the direct loser on each individual trade — they got picked off on a quote they didn't cancel fast enough. But the cost doesn't stop there: if this happens often enough, that market maker rationally widens spreads or quotes smaller size on the regional venue to compensate for the expected losses, which makes the regional exchange less liquid and more expensive for everyone trading there, including ordinary investors with no speed advantage at all.

The proposed fix

Because the whole strategy depends on the race being winnable in continuous time, the most discussed structural remedy is frequent batch auctions — instead of matching orders continuously, a venue would collect orders over a short window (say, every 10-100 milliseconds) and clear them all at one price, which makes being a few hundred microseconds faster than a rival worthless, since both arrive within the same batch. This remains mostly a research and policy proposal rather than a widely adopted market structure.

In interviews

The key distinction to draw is stale-quote arbitrage (this concept) versus general speed advantage in reacting to new information — interviewers often use "latency arbitrage" loosely and it's worth clarifying which one you mean before answering. Tie the mechanism back to Market Fragmentation: latency arbitrage is only possible because the same asset trades in multiple, imperfectly synchronized places.