Modelling slippage in a backtest without making it up
Posted: Sat Oct 31, 2026 7:54 pm
Backtests usually assume every order fills at the requested price. For scalping, this is the most optimistic assumption of all.
Realistic slippage modelling starts with live data. Collect the actual difference between requested and filled price from a few hundred real trades, split by order type, time of day and market condition. The distribution is usually skewed, with small slippage most of the time and a long tail.
In the test, draw random values from that distribution and apply them to each fill. Make it worse during news and rollover. Run the test many times, since results vary with the random draws.
The final answer is a range, not a single number. If the strategy's profitability depends on the lucky end of the range, it's fragile.
Without live data, a conservative placeholder like an extra 0.2 to 0.5 pips per trade gives a rough idea.
Even a crude version beats ignoring the issue. A one-line rule such as adding half the average spread to each stop fill, and a quarter of it to each market entry, already moves the results much closer to what live accounts record.
What slippage assumption do you build into your tests, and how did you choose it?
Realistic slippage modelling starts with live data. Collect the actual difference between requested and filled price from a few hundred real trades, split by order type, time of day and market condition. The distribution is usually skewed, with small slippage most of the time and a long tail.
In the test, draw random values from that distribution and apply them to each fill. Make it worse during news and rollover. Run the test many times, since results vary with the random draws.
The final answer is a range, not a single number. If the strategy's profitability depends on the lucky end of the range, it's fragile.
Without live data, a conservative placeholder like an extra 0.2 to 0.5 pips per trade gives a rough idea.
Even a crude version beats ignoring the issue. A one-line rule such as adding half the average spread to each stop fill, and a quarter of it to each market entry, already moves the results much closer to what live accounts record.
What slippage assumption do you build into your tests, and how did you choose it?