How Long Do Stale Quotes Last? 20 Hours of Gold Ticks at Three Brokers
By Sergey Berezhnov, founder of Westernpips. Westernpips develops latency arbitrage software, so please keep that interest in mind. Every figure below comes from an open study with a published method, not from marketing.
People argue about latency arbitrage in words: some say "it stopped working years ago", others say "it is easy money". It is more useful to look at what actually happens, in milliseconds.
Our Head of Research, Alexander Vinsky, recorded gold (XAUUSD) quotes at three retail brokers and compared them with a fast reference feed. Three sessions on different days of 2026, about 20 hours in total, more than a million reference ticks. Broker names are hidden; we call them A, B and C.
1. What was measured
A "stale-quote window" is the moment when the fast feed has already moved away from the broker's price by a set amount while the broker still shows the old price. The threshold was 30 points ($0.30 per ounce). For brokers A and B that is about three median spreads; for C it is less than its spread.
Both feeds were recorded on one clock, so times compare directly. As a control, the same calculation was run on random entries.
| Broker A | Broker B | Broker C | |
|---|---|---|---|
| Fast-feed updates per second | 12.4 | 28.1 | 16.0 |
| Broker updates per second | 2.7 | 9.2 | 6.2 |
| Median gap between broker updates | 272 ms | 38 ms | 112 ms |
| Median broker spread | 10 points | 12 points | 46 points |
The broker updates its price 3–5 times less often than the fast feed. Broker C also had five feed pauses of 30 to 130 seconds.
2. Windows exist, and there are many of them
| Broker A | Broker B | Broker C | |
|---|---|---|---|
| Windows per hour | 260 | 824 | 240 |
| Median window length | 50 ms | 34 ms | 56 ms |
| 90th percentile | 217 ms | 200 ms | 201 ms |
| Closed within 20 ms | 22% | 35% | 24% |
Hundreds of windows per hour is not luck. But half of them live less than 35–56 milliseconds, and about 70% close because the broker updated its own quote, not because the market came back.
3. The number that matters: your order delay
The key question is what share of windows is still open when your order reaches the broker.
| Order delay | 10 ms | 30 ms | 50 ms | 100 ms | 150 ms |
|---|---|---|---|---|---|
| Broker A | 91% | 67% | 50% | 31% | 20% |
| Broker B | 79% | 54% | 37% | 19% | 14% |
| Broker C | 86% | 66% | 53% | 25% | 15% |
And the mean gross result per signal, in points:
| Delay | Broker A | Broker B | Broker C |
|---|---|---|---|
| 10 ms | +28 | +17 | +7 |
| 30 ms | +22 | +12 | +2 |
| 100 ms | +10 | −1 | −17 |
| 150 ms | +6 | −3 | −28 |
| Random entries | about −9 | about −12 | about −49 |
The result falls to zero at about 40 ms for broker C, about 90 ms for B and beyond 150 ms for A. The same strategy still works at 100 ms with one broker and stops at 40 ms with another.
So "does latency arbitrage work?" has no general answer. The right question is "what is my order delay, and where is the zero point at my broker?"
4. A less obvious finding: the price source matters more than the number of signals
For broker A a second fast feed was tried. It produced 6,800 signals instead of 1,469 — more than four times as many. But the mean result at 30 ms fell from +22 to +3 points. A noisy feed creates "windows" that are not really broker lag.
5. Why this is not proof of profit
- These are gross figures at displayed quotes: commission, slippage, rejected orders and last look are not included, so real results will be worse.
- Three separate sessions on different days are not a broker ranking.
- News events were not covered.
- The author states plainly that this does not prove the strategy works.
Broker terms are another reason for caution: many client agreements forbid latency arbitrage, and profits from such trades can be cancelled.
6. How to apply this yourself
- Measure your order delay, not just ping. Ping is only part of an order's path.
- Record ticks from your broker and a fast source for at least one trading session.
- Count windows at a threshold of 2–3 spreads and see what share is still open at your delay.
- Compare with random entries. If there is no difference, there is no edge.
- Only then think about live trading, with all costs included.
Economists who studied races for stale prices on the London Stock Exchange used a similar approach: they found such races about once a minute per FTSE 100 stock, with a typical race lasting 5–10 millionths of a second (NBER Working Paper 29011). In CFD markets the windows are thousands of times longer, but the measurement rules are the same.
Westernpips develops latency arbitrage and quote analysis software: westernpips.com. This post is educational. Trading Forex and CFDs carries a high risk of losing money.

Comments