Signs an EA has been curve-fitted to tick data
Posted: Tue Oct 20, 2026 9:25 am
Overfitting is the most common reason backtests fail live. Some signs to look for:
- Too many parameters. An EA with 15 optimised inputs can fit almost any data.
- Parameters with very specific values (e.g. RSI period 17, stop 23 points, take profit 41 points).
- Performance collapses when parameters change slightly. A solid EA should work across a range of values.
- Great results in-sample, poor out-of-sample.
- Most profit comes from a small number of trades.
- Performance varies hugely by year.
Tests for robustness:
- Walk-forward analysis: optimise on one period, test on the next, repeat.
- Parameter sensitivity: plot results across a range of each parameter. Look for plateaus, not peaks.
- Different data: test on another broker's tick data.
- Monte Carlo: shuffle trade order to estimate drawdown distribution.
- Different pairs: does the logic work on similar instruments?
An honest backtest often looks less impressive but is more likely to survive live trading.
What's your main test for overfitting?
- Too many parameters. An EA with 15 optimised inputs can fit almost any data.
- Parameters with very specific values (e.g. RSI period 17, stop 23 points, take profit 41 points).
- Performance collapses when parameters change slightly. A solid EA should work across a range of values.
- Great results in-sample, poor out-of-sample.
- Most profit comes from a small number of trades.
- Performance varies hugely by year.
Tests for robustness:
- Walk-forward analysis: optimise on one period, test on the next, repeat.
- Parameter sensitivity: plot results across a range of each parameter. Look for plateaus, not peaks.
- Different data: test on another broker's tick data.
- Monte Carlo: shuffle trade order to estimate drawdown distribution.
- Different pairs: does the logic work on similar instruments?
An honest backtest often looks less impressive but is more likely to survive live trading.
What's your main test for overfitting?