Simulating Stops and Trailing Stops

A stop-loss order says "sell if the price reaches this level." A backtest usually implements that as "if the bar's low is at or below the stop, fill at the stop price" — which is exactly correct only when trading is continuous through that level. Real markets aren't continuous overnight, around earnings, or during a halt. When the next print after a stop level is already through it, the fill happens at whatever price the market reopens at, not at the level you set.

Worked example: the stop that fills fifteen points below where it was drawn

A stock is held with an 8% stop below the entry, standard practice for the strategy. The company reports earnings after the close and misses badly. The stock closes at $100 before the report and opens the next session at $85 — a 15% gap, with no trades printed anywhere between $100 and $85.

Model	Assumed fill	Realized loss
Naive stop simulation	Exactly at the 8% stop, $92	−8%
Realistic gap-aware simulation	At the next tradeable price after the gap, $85	−15%

The naive backtest simply cannot represent this trade correctly — there is no way to fill at $92 when the stock never traded there. Over many trades this doesn't show up as a small, evenly distributed error; it shows up as understated tail risk concentrated in the rare, large moves that a stop is specifically meant to protect against. A strategy's backtested max drawdown and its real max drawdown can diverge sharply on exactly the handful of gap events in the sample, precisely because those are the events the stop-fill assumption handles worst.

Stress the stop-fill assumption specifically: pull out every historical instance where the prior close was already through the stop level, refill those trades at the actual open, and compare the resulting tail of the return distribution to the naive version — the two should look meaningfully different if stops are doing real work in the strategy.