Latency Arbitrage
Also known as: latency arb, stale quote arbitrage, fast-feed arbitrage, toxic flow
What is it?
Latency arbitrage is trading against a price quote that has not caught up with the real market yet. The arbitrageur watches the true price on a faster feed and hits the broker's stale one in the moments before it refreshes. The window is small and entirely mechanical.
- 1t = 0 ms - the real price moves EUR/USD ticks up three pips on the primary venue. This is now the true market price, and everything after this is a race to notice it.
- 2t = 2 ms - a fast feed carries the move A colocated direct data feed delivers the new price to anyone paying for it, about two milliseconds after it happened.
- 3t = 45 ms - the broker's quote is still stale The retail broker's aggregated quote has not refreshed yet, so it is still offering to sell at the old price the market has already left behind.
- 4t = 46 ms - the stale quote gets hit The arbitrageur buys at the outdated price and is immediately in profit. The broker absorbs that loss, then tightens its defences for every fast account on the platform.
Suppose EUR/USD ticks up three pips on the primary venue at 09:30:00.000. A colocated direct feed carries that move in roughly two milliseconds, while a retail broker's aggregated quote may not refresh until 09:30:00.045. For about 43 milliseconds that broker is still offering to sell at a price the market has already left behind, and anyone watching both feeds can buy there and be in profit the instant the quote updates.
The reason this term belongs in an automation glossary is not that you will do it, but that brokers defend against it and their defences land on you. Because the loss falls on the broker, it responds with last-look rejection windows, deliberate execution delay, wider quotes for flagged accounts, and outright restrictions. A well-tuned fast bot that never intended to arbitrage anything can trip the same detection and start collecting requotes and rejections it did nothing to earn.
Why it matters: Brokers defend against latency arbitrage with last-look delays and rejections, and those defences fall on ordinary fast bots that were never arbitraging anything.
The defences it provokes, added delay and last-look rejection and widened quotes, degrade execution for every fast automated account on the platform.
Real-world example
After a payrolls release, one broker's EUR/USD quote lagged the interbank price by roughly 40 milliseconds, long enough for a stale 1.0940 offer to sit two pips below where the pair was already trading.
How SignalBots handles it
SignalBots timestamps every signal at generation and at delivery, so you can measure your own end-to-end delay and tell a genuinely late fill apart from a broker-side rejection. See /risk-warning.
Pro tip
If your bot starts collecting rejections shortly after you reduced its latency, suspect the broker's arbitrage defences before you go looking for a bug in your own code.
Common pitfalls
Assuming a faster connection always improves fills, when past a certain point it moves your account into the broker's toxic-flow bucket and makes execution measurably worse.
Frequently asked questions
Is latency arbitrage illegal?
It is not illegal, but nearly every retail broker prohibits it in its client terms and will void trades or close accounts that practise it. Treat the contract, not the law, as the binding constraint.
Can a retail trader realistically do it?
In practice no. It requires a faster market data feed than the broker's own, which means paid direct feeds and colocated infrastructure costing far more than the edge returns at retail position sizes.
What is last look?
A brief window in which a liquidity provider or broker may reject an order after seeing it, rather than filling it automatically. It exists largely to avoid being picked off on a stale quote.
How do I tell if my broker adds artificial delay?
Compare your order-to-fill time in quiet markets against fast ones. A delay that appears only when price is moving quickly, and mostly on entries that would have been profitable, points to a defensive mechanism.
Does low latency still help a normal strategy?
Yes, up to a point. Cutting delay reduces slippage on entries, but chasing microseconds brings diminishing returns and can attract broker restrictions, and your capital is at risk either way.
Trading involves substantial risk of loss. Historical and backtested results do not guarantee future performance. Read the full risk warning.
