also: execution slippage, price slippage, deviation
The difference between the price an order asked for and the price it was filled at. It can fall either way, but for a strategy that trades on speed it is a cost that backtests rarely model.
You asked for one price and got another. Between the moment an order leaves the terminal and the moment it is filled, the market moves — sometimes in your favour, more often not, and most of all when you least want it to.
A backtest fills at the price it decides to fill at. A live account fills at whatever the broker had available when the order arrived. That difference is small per trade and compounds with trade count, which is why it decides whether a fast strategy survives contact with a real account.
| Setting | Typical value | Note |
|---|---|---|
| Deviation, swing strategies | 10–30 points | Generous. A few points of slippage is irrelevant against a target measured in hundreds. |
| Deviation, scalping strategies | 1–5 points | Tight enough that a bad fill is refused rather than accepted — at the cost of more rejections. |
| Average slippage worth accepting | under 0.5 pips | Measured over 100+ live trades on your own account, not quoted from a broker's marketing. |
| Average slippage worth investigating | over 1.5 pips | For a strategy with small targets this is a material share of every win. |
| Filling mode | as the symbol allows | Set by the broker per symbol; sending an unsupported mode gets the order rejected, not adjusted. |
The same 0.4 pips of average slippage, applied to two strategies with different targets. The per-trade cost is identical; the share of the edge it consumes is not.
Nothing about the broker changed between these two rows. The difference is how much of the strategy's edge sits inside the execution window.
Calculation 0.8 ÷ 6 = 13% · 0.8 ÷ 120 = 0.7%
Result One execution quality, two entirely different verdicts
Slippage is measured, not assumed. Every conclusion below comes from comparing requested prices with fill prices on your own account and broker.
| Range | What it means |
|---|---|
| Average under 0.5 pips over 100+ trades | Good execution. Backtest results transfer with a modest haircut. |
| 0.5–1.5 pips | Acceptable for strategies with targets of tens of pips; a real drag on anything smaller. |
| Over 1.5 pips average | Investigate before scaling up. Either the strategy is trading at the wrong moments or the execution is not competitive. |
| Consistently one-directional | Positive slippage should occur too. If fills are always worse than requested, that is an execution characteristic rather than market movement. |
AIStrategyMiner listings publish which broker and account conditions each backtest was produced on, so the execution assumptions behind a result are stated rather than implied.
Positive slippage — a fill better than requested — happens under normal conditions with genuine market execution. A record showing only negative slippage over hundreds of trades is describing the execution, not the market.
The Strategy Tester models spread but not the delay between decision and fill. Every tester result is therefore optimistic by an amount that grows with trade frequency, and the gap is largest for exactly the strategies that look best in testing.
That does not remove slippage; it removes fills. The EA ends up trading a fraction of its signals, and which signals it misses is decided by market speed rather than by the strategy.
A tight advertised spread with poor execution can cost more than a wider spread filled cleanly. The comparable number is the total cost per round trip — spread plus commission plus measured slippage.
Every listing here ships a fill tolerance, and the shelf disagrees on its meaning. Deviation decides which fills get refused, not how bad the accepted ones are — and the same numeral here means two sizes.
| Across the 14 EAs published here | Figure |
|---|---|
| Listings with a fill-tolerance input | 14 |
| Parameter entries describing it | 40 |
| Stated tolerance of 0.3 pips | 3 |
| Stated tolerance of 3 pips | 11 |
| Pages saying it does not affect the backtest | 8 |
| Pages saying the record was measured at it | 2 |
| Runs with a measured slippage figure | 0 |
Rows three and four are the trap. Cairn and Thunderhead print a bare 3 and count it in points; nine others print a bare 3 and count it in pips. Same numeral, ten times apart. Only Iridescence, Lattice Weave and Tidewell Slack print the unit — the three you cannot misread.
Rows five and six describe one setting in two registers, and row seven says which to trust: none of our backtests measures a fill against a request. That makes it a live-trading instruction, not a backtest assumption — the boundary our methodology draws around each record.
It bites hardest where it is tightest: at 0.3 pips a fill that moves further is refused, so Cairn’s 4,725 trades become fewer than the record shows, not worse.
