Why most backtests lie
Backtesting is one of the most powerful tools in a trader's arsenal — and one of the most abused. Done correctly, a backtest tells you whether a strategy had a statistical edge in historical data. Done incorrectly, it tells you whatever you want to hear. The difference comes down to a few critical principles that most retail traders skip.
What backtesting actually is
Backtesting means applying a set of defined trading rules to historical price data and measuring the outcome. If your rules are: "Buy when the 20 EMA crosses above the 50 EMA on the 4H chart, stop below the previous swing low, target 2× the stop distance" — you go back through the last 5 years of data and log every instance of that signal and its result.
The output: win rate, average R:R, maximum drawdown, EV per trade, and total return over the period.
The 5 most common backtesting mistakes
1. Curve-fitting (overfitting)
Adjusting your rules until they perfectly fit the historical data. The result looks great on paper but falls apart on live data because you've engineered a strategy for the past, not the future. Solution: define your rules before you test them. Don't modify the rules based on what you see in the data.
2. Survivorship bias
Testing on assets that are still trading today, ignoring companies/pairs that went bankrupt or were delisted. This makes historical performance look better than it was. For Forex traders this is less of an issue, but for equity traders it's significant.
3. Ignoring trading costs
A strategy that looks profitable before spread and commission often isn't after. Always include realistic costs: at minimum 1 pip spread for majors, 0.1% per trade for crypto. High-frequency strategies are particularly vulnerable to this.
4. Testing on too-short a period
Three months of backtesting means nothing. Markets cycle through trending, ranging, and high-volatility regimes. Test across at least 3–5 years to see how your strategy behaves in different conditions.
5. Look-ahead bias
Using information that wouldn't have been available at the time of the trade. Example: placing a trade based on how a candle "closed" when using live data where that candle was still forming. Automated backtesting tools often introduce this by accident.
The correct backtesting process
- Define your rules precisely — Every condition, every parameter, in writing. No ambiguity.
- Choose your data source — Use a reliable broker's historical data or a dedicated data provider. At least 3 years.
- Test on unseen data — Split your data: develop on 70% (in-sample), validate on 30% (out-of-sample). The out-of-sample results are what matter.
- Log every signal — Not just the "clear" ones. If it met your rules, log it. Cherry-picking destroys validity.
- Calculate realistic EV — Include spread, commission, and slippage.
- Forward test before going live — Trade the strategy on a demo account for 1–3 months. Does live performance match backtest? If not, investigate why before risking real money.
Backtest results vs live results
It's normal for live performance to be 10–20% worse than backtest performance. This is due to execution differences, emotional interference, and market evolution. If live performance is more than 30% worse, your backtest had flaws — or you're not executing the strategy correctly.
Use TradeLab's journal to track live trades and compare your actual performance against your backtest projections. The gap between the two tells you whether your edge is real.