Overfitting
Overfitting is the practice of optimising strategy parameters until the backtest looks great, only for the strategy to fail on out-of-sample or live data. The strategy has memorised the past instead of learning a generalisable edge.
Signs your strategy is overfit
- Parameters are oddly specific (EMA period 17 and 43, not 14 and 50)
- Win rate is unusually high (>75%)
- Profit factor > 3.0 on a small trade count
- Strategy uses many filters that "fix" small periods of underperformance
- Backtest looks perfect but live results don't match
How to avoid it
- Walk-forward analysis — optimise on one period, test on the next, advance, repeat.
- Out-of-sample reserve — set aside the most recent 20% of data and never look at it during optimisation.
- Simplicity bias — fewer parameters always beats more.
- Cross-asset robustness — if it only works on one symbol, it's probably curve-fit.
- Live forward test — paper-trade for 30+ days before risking real capital.
The bitter truth
Most strategies that look amazing in backtest are overfit. A strategy with profit factor 1.4 across multiple symbols and timeframes is more trustworthy than one with profit factor 3.0 on a single optimised configuration.
Related Terms
Backtesting
Backtesting runs a strategy on historical data to estimate how it would have performed before risking real money.
Walk-Forward Analysis
Walk-forward analysis tests a strategy by optimising on one window and testing on the next — repeated rolling forward through history. The most honest backtest you can run.
Sharpe Ratio
The Sharpe ratio measures excess return per unit of volatility — the most-cited risk-adjusted performance metric in finance.
Stop Reading. Start Trading.
PineForge backtests every concept on this site against real market data.
Try It Free