How Long Do Stale Quotes Last? Measuring Latency Arbitrage Windows on XAUUSD at Three Retail Brokers

Latency arbitrage in retail foreign exchange and CFD markets exploits the interval during which a broker’s displayed quote lags a faster reference price. Practitioner sources describe this interval with loose ranges such as 50–200 ms, and the peer-reviewed evidence concerns equities at microsecond scales. This paper measures the interval directly. Using tick recordings of XAUUSD from three sessions at three anonymised retail brokers, each paired with a fast reference feed on a single recording clock, we define stale-quote windows by a cross-side gap threshold and measure their duration exactly within the recording. At a 30-point (0.30 USD) threshold the brokers showed 240 to 824 windows per hour, with median durations of 34 to 56 ms; between 22% and 35% of windows closed within 20 ms and between 19% and 31% remained open beyond 100 ms. A simulated order arriving 30 ms after the signal and closed one second later would have captured a mean gross result of +22, +12 and +2 points on displayed quotes, against −9, −12 and −49 points for random entries. The delay at which the mean result reaches zero ranges from about 40 ms to beyond 150 ms, depending mainly on quote update frequency and spread. A second reference feed produced 4.6 times as many windows but almost no edge, showing that window counts are not a measure of opportunity. Results are reproduced on the export of the recording software and compared with its own 210 simulated trades. The data contain displayed quotes only, so all results are gross upper bounds. Disclosure: the author is an employee of Westernpips Group, which develops the software used to record and analyse the data. This is a vendor case study intended to be repeated and checked, not an independent audit.