Why Most Backtests Lie, and How Not to Be Fooled

You've probably seen equity curves that climb in a perfectly straight line: 100% a year, no drawdowns. When the same strategy goes live, the story is usually very different. The problem isn't backtesting. It's bad backtesting.

Trap one: overfitting

Tweak the parameters until the past looks perfect and you've taught the strategy the noise of that particular period. A 17-period average instead of 20, a 1.3% stop instead of 1.5%... each one nudges the result up, and together they build something that has merely memorised history.

• Simple rules with few parameters tend to last longer.
• Test the strategy on several symbols and periods. If it only works on one, be suspicious.
• Set part of the data aside and test on it only at the very end.

Trap two: look-ahead bias

The most common technical mistake is letting a strategy use information that didn't exist yet at the moment of decision, for example deciding on a candle's close but entering at that same candle's open. The usual symptom is unrealistic profits and a suspiciously high win rate.

A sound backtest checks conditions only on closed candles and enters at the next candle's open, which is what's actually possible in a live market.

Trap three: forgetting the costs

Every trade pays a fee, and in a real market you don't always fill at the exact price you want. A strategy making dozens of small trades a day can be profitable without fees and losing with them. Always ask: are fees included?

Trap four: a short or one-sided period

A strategy tested only during a six-month rally has proved one thing: it works in a rally. A trustworthy backtest has seen uptrends, downtrends and ranges, with enough trades that the result isn't luck. Ten trades aren't enough to judge.

Which numbers to look at

1. Maximum drawdown: the worst fall from a peak to a trough. Could you live through it?
2. Profit factor: total wins divided by total losses. Above 1.3 is usually meaningful.
3. Average R: how many times its risk each trade made or lost on average.
4. Compared with buy and hold: if the strategy does worse than simply holding the asset, why complicate things?

The bottom line

A backtest isn't there to prove your strategy is great. It's there to show you its weaknesses before you lose real money. One pessimistic, precise backtest is worth more than ten optimistic ones.