Performance also: curve-fitting, over-optimisation, data mining bias

Overfitting

Tuning EA parameters too precisely to historical data, producing a strategy that backtests well but performs poorly on unseen data because it exploits noise rather than real market patterns.

How overfitting happens

Every EA has tunable parameters. Optimising all of them simultaneously across a long backtest produces parameters that fit the idiosyncratic features of that specific historical data, not underlying market dynamics.

Warning signs in EA marketing

  • Win rate above 90% on a 5+ year backtest with no live track record
  • Strategy parameters that look suspiciously precise (e.g., EMA period 37, stop 23.7 pips)
  • No walk-forward or out-of-sample performance shown
  • Backtest window less than 3 years
  • Sharpe ratio above 3.0 without explanation

Partial remedies

Use fewer, more interpretable parameters; require walk-forward analysis before publication; test across multiple instruments beyond the optimised pair.