Can you trust your backtest? Learn how to find out.

You can’t tell from the backtest alone. It needs to hold up on data it was never tuned on, after you account for how many versions you tried, and with realistic costs included.

Decide what you’ll test before you see results, and stop when a change has no reason behind it other than a better backtest. Once you’ve opened your final test, you’re done tuning.

Set a stretch of history aside before you start, build and tune on the rest, and run the finished strategy on it once. Every extra look turns it into tuning data.

Usually one of four reasons: it was overfit, the backtest left out real costs and fills, a bug or look-ahead mistake flattered it, or the market it was built on has changed. Sometimes it’s simply a normal losing stretch that the backtest had too.

Charge every trade the commission and fees you’d really pay, add slippage on entries and exits, and only assume fills that could have happened. Then re-run with higher costs to see if the edge survives.

It happens more than people think. The classic mistake is look-ahead bias: deciding a trade with information you couldn’t have had yet, like a bar’s close. Results that look too good deserve a line-by-line check.

No one can promise that. What you can do is test how it did across different conditions, calm and volatile, trending and choppy, so you know where it struggles before the market shows you.

There’s no magic number. Rules of thumb like 30 or 100 trades aren’t proof. It depends on how big the edge is compared with the noise, how many versions you tried, and whether the trades span different market conditions.

Long enough to see the strategy behave the way the backtest said it would, across more than one kind of market. Forward testing is slow at proving an edge, so an honest backtest matters even more.

No single number. Read the win rate together with the average win and loss, look at the worst drawdown and the number of trades, and treat profit factor and Sharpe as summaries of the same trades. Every one of them is easy to inflate by overfitting.

Yes. No-code backtesters and AI tools can turn written rules into a test. Whatever you use, ask the same three things: can it show you the actual trades, the costs it assumed, and the checks it ran?

AI is good at turning a description into rules and suggesting ideas. Trust the numbers only if they come from a real backtest you can inspect, not from text the AI wrote. Ask to see the trades and the checks.

Yes, mostly for what it rules out. A backtest can’t promise a strategy will work, but it’s the cheapest way to find out that one doesn’t, before real money does.

Turn it into rules first: exactly what has to be true to enter, where the stop goes and when you exit. The parts that stay “it depends” can’t be tested honestly, so make them as specific as you can.

A challenge asks whether your strategy can reach a profit target without breaking rules like a daily loss limit or a maximum loss. A strategy that’s profitable overall can still fail on one bad day, so test it against the rules, not just for profit.