← School

How to read a backtest without fooling yourself

A well-read backtest tells you a lot; a badly-read one ruins you with confidence. Here are the numbers that matter and the traps they hide.

The metrics beginners look at (and why they mislead)

Net profit and % gain: they impress but say nothing about risk. A +500% with an 80% drawdown is a dead account that got lucky.

High win rate: 90% wins sounds great until you see the average loss is 20x the average win (classic martingale). Win rate alone means nothing.

The metrics that actually matter

The deadly trap: in-sample vs out-of-sample

If you optimise a strategy over a period and test it on THE SAME period, it will always look perfect — you've memorised the past. The honest test: optimise on one stretch (in-sample) and validate on another the system never saw (out-of-sample). If performance collapses out-of-sample, it was overfitting. And even that isn't enough: the final judge is real ticks.

The question that sums it all up

Don't ask "how much did it make?". Ask: "does it survive real ticks, its tail risk, and perturbing its parameters?". If all three answers are yes, you have something. If you haven't measured it, you just have a pretty curve.

See 17 backtests that lied → Back to School