Home › Guides › Myths

Do backtests predict results? What past tests can't promise

A strategy that did beautifully on 5 years of past data feels like a sure thing. A backtest is a useful tool. But a test on past charts can only show how rules behaved, and there are a few quiet ways it can mislead.

Beginner guide · 6-minute read · Updated 2026-10-21

The myth

A Backtesting report shows a smooth rising line, a win rate and an average gain. Numbers feel objective. If it worked for 5 years, why wouldn't it work next year?

It's a reasonable question. The report just doesn't show the choices that went into making it.

What's actually true

A test describes the past, nothing more

Markets change. A rule that suited a calm, trending year may struggle in a choppy one. A backtest shows how a rule behaved in conditions that already happened. That's useful evidence. It's not a forecast.

Overfitting: tuning to old noise

If you keep adjusting settings until the past looks perfect, you end up fitting the rule to random wiggles that won't repeat. That's called overfitting.

Researchers have studied this closely. Bailey, Borwein, López de Prado and Zhu wrote a paper called "The Probability of Backtest Overfitting" showing how trying many versions raises the odds that the best one is a fluke.

Look-ahead bias: peeking without meaning to

A test that uses information you wouldn't have had at the time will look better than reality. Common examples:

• Using a candle's close to enter during that same candle.
• Marking a level that was only clear after price turned.
• Scrolling forward on a chart "just to check."

On manual replay, the fix is simple: never peek. See TradingView bar replay.

Missing costs

Many tests ignore spreads, commissions and slippage. On a rule with small targets, costs can flip a positive test negative.

6 checks for a more honest test

1. Write the rules before testing, in plain words, and don't change them mid-test.
2. Use at least 100 trades. Fewer is mostly luck.
3. Include costs on every trade.
4. Split the data. Build on one period, then test once on a later period you haven't looked at.
5. Never peek on replay. Decide, then press play.
6. Check the worst stretch, not just the total. How many losses in a row? How deep did it dip? See Drawdown.

Then paper trade it live for a few weeks. If live results look very different, the test was probably too kind.

What to do instead

• Start by hand. Replay 100 past trades with our method and log each one. See how to backtest a trading strategy.
• Use code tests as a second check. Pine Script's strategy tester is handy, but its results are hypothetical. See Pine Script strategy backtests.
• Work out the expectancy after costs.
• Keep risk fixed at about 1% a trade, so a test's worst stretch is survivable live.

Common follow-up mistakes

• Trusting a seller's backtest. You can't see how many versions they tried. See the trading bot myth.
• Expecting live results to match. A good result is live trades that look roughly like the test.
• Giving up after 10 live losses on a rule tested over 300 trades. Check whether that streak happened in the test too.
• Re-tuning after every bad week. That's overfitting in slow motion.

Questions