Backtests assume every order fills. In live trading, some are rejected, and the rejections aren't random. They happen more when price is moving in the direction of the order, which is exactly when the trade would have been profitable.
Modeling this requires a rule: if price moves in the order's favor by more than a threshold during the delay, the order is rejected with some probability. A simple version rejects orders when the market has moved 0.3 pips in the trade's favor during the round trip.
The effect on results is a reduction in the best trades. The winners that depend on speed disappear. The losers, which often come from entries into a market that is moving against the order, still fill.
That skews the distribution toward worse outcomes. A strategy with a profit factor of 1.6 in an idealized test might fall to 1.2 under realistic rejection.
Calibrating the rule requires live data: the fraction of orders rejected and the conditions under which it happens. Logs from a few weeks of live or small-size trading provide enough.
An edge that survives this test is far more likely to survive the real thing.
What fraction of orders gets rejected on the live account, and does it correlate with price movement?
Simulating last-look rejections
-
FXS HongKong Scalper
- Posts: 543
- Joined: Sat Oct 24, 2026 10:15 pm
Re: Simulating last-look rejections
Good rule. One refinement: the threshold should depend on the instrument and the hour. Gold during the US data window moves far more within a 50 ms delay than EURUSD at Asia midday, so a single pip threshold across everything understates rejections in exactly the conditions where they bite. Calibrating it from your own rejected-order log, if the platform keeps one, gives a more realistic figure.
FXS Official Team | Hong Kong Session Desk
-
Amanhasnoname
- Posts: 2060
- Joined: Fri Oct 30, 2026 9:45 am
Re: Simulating last-look rejections
I only started counting rejections properly after a frustrating month where my live results trailed the demo by far more than spreads could explain. Once I pulled the execution report into a spreadsheet, roughly one order in twelve had been rejected or requoted, and when I lined those up against the tick data almost all of them came when price had already ticked in my favour before the fill. The quiet periods hardly had any.
So yes, in my experience it correlates strongly with movement, and it clusters around the open of London and the first minutes after scheduled data. I ended up adding a rule to my tester that rejects a share of entries whenever price moves my way during the assumed delay, using the rate from my own log for each hour of the day.
It knocked a strategy I was excited about from looking great to barely worth running, which was painful but saved me money. The ones that held up were the slower entries that didn't need a perfect fill to work.
So yes, in my experience it correlates strongly with movement, and it clusters around the open of London and the first minutes after scheduled data. I ended up adding a rule to my tester that rejects a share of entries whenever price moves my way during the assumed delay, using the rate from my own log for each hour of the day.
It knocked a strategy I was excited about from looking great to barely worth running, which was painful but saved me money. The ones that held up were the slower entries that didn't need a perfect fill to work.