Bad ticks that make a backtest look brilliant

[Amanhasnoname]

One spike tick printed 15 pips away from the market and gone a millisecond later. A scalper with 3 pip targets and limit orders can book a fill and a profit on it in the test, while no live order ever saw that price. A handful of those in a sample and the equity curve gets a nice boost nobody could have collected.

Cheap way to find them: sort the trade list by profit and read the top 20. If the biggest winners sit on odd timestamps, at the edge of sessions or right at feed gaps, treat the results as suspect.

Scan the raw data too:

- ticks that jump more than a set number of pips from their neighbours and revert within a second
- zero or negative spreads
- duplicate timestamps and out-of-order ticks
- missing hours, where interpolation can produce phantom fills

Weekend gaps and holiday sessions deserve a separate look, since thin data there often contains the worst artefacts.

Filtering has its own risk. Cut too aggressively and real fast moves, the very ones that matter at news, get removed. Keep a log of what was dropped and rerun with and without to see how much of the edge depends on the questionable data.

What filters do you apply to raw tick data, and what was the strangest bad tick found?