Adding latency to a backtest changes the story
Posted: Sat Oct 31, 2026 8:55 pm
A backtest assumes the order fills at the signal price, instantly. In reality, there is a delay between signal, order, and fill. During that time, price keeps moving.
For a strategy that enters on a breakout, the delay matters most. The signal comes as price accelerates, and by the time the fill happens, the entry might be a pip or two away.
Adding a delay to the test, say 200 milliseconds, and using the price at that later moment shows the damage. Strategies that look strong with zero latency often lose their edge once delay is included.
Some testers allow a configurable delay. Where this is not possible, a rough approach is shifting each entry price by a typical slippage amount in the unfavorable direction.
Latency affects different strategies differently. Mean-reversion systems that place limit orders suffer less, while momentum systems that chase moves suffer more.
Measuring actual latency from the platform to the broker gives the number to use.
How much delay is assumed in tests, and did including it change any decisions?
For a strategy that enters on a breakout, the delay matters most. The signal comes as price accelerates, and by the time the fill happens, the entry might be a pip or two away.
Adding a delay to the test, say 200 milliseconds, and using the price at that later moment shows the damage. Strategies that look strong with zero latency often lose their edge once delay is included.
Some testers allow a configurable delay. Where this is not possible, a rough approach is shifting each entry price by a typical slippage amount in the unfavorable direction.
Latency affects different strategies differently. Mean-reversion systems that place limit orders suffer less, while momentum systems that chase moves suffer more.
Measuring actual latency from the platform to the broker gives the number to use.
How much delay is assumed in tests, and did including it change any decisions?