Backtesting
Backtesting is testing a trading strategy against historical data before risking money on it. Done honestly, it is the cheapest way to find out a strategy loses. Done carelessly, it manufactures confidence in strategies that were never going to work. This page covers both: what the numbers say when famous strategies are tested exactly as taught, how to run a backtest that does not fool you, and free tools to do it yourself.
What 53,856 backtested trades taught us
June 1, 2025 – June 1, 2026 (12 months) · BTC & ETH · rules pinned exactly as taught · commission 0.05% per side · same deterministic engine as the in-app AI backtester
We took 21 well-known strategies, from the ICT Silver Bullet to opening range breakouts to RSI reversion, wrote each one down as mechanical rules with no discretion, and ran them over twelve months of real 1-minute data on BTC and ETH. Every combination lost money after costs. The median profit factor was 0.62, meaning the typical strategy gave back $1 and change for every dollar it made.
That is not an argument that trading is impossible. It is a measurement of what the popular rule-sets are actually worth without a human filtering the setups, and it is exactly the kind of result a careless backtest hides. The per-strategy pages below publish the losers as losers, with the exact rules, monthly breakdowns, and drawdowns, so you can see what honest results look like before trusting anyone's screenshot of an equity curve.
Bar-by-bar replay on real BTC data in your browser. No account.
Describe a strategy in plain English; a deterministic engine runs it on real data.
How to backtest without fooling yourself
The seven rules our own engine enforces, in the order they matter.
- 1Write the rules so a machine could follow them
If two people reading your strategy could take different trades, it is not testable yet. Pin down the entry trigger, the stop, the target, the position size, and the session, with numbers.
- 2Pick one market and one timeframe first
Prove the idea somewhere specific before generalising. An edge that only exists when you average eight markets together is usually noise.
- 3Include costs from the first run
Commission, spread, and slippage decide marginal strategies. Our study paid $225,949 in modeled fees across 53,856 trades, and costs alone flipped most of the field from looking viable to losing.
- 4Resolve each bar pessimistically
When a candle touches both your stop and your target, count the stop first unless you have finer data that proves otherwise. Optimistic tie-breaks quietly inflate every stat.
- 5Collect enough trades before believing anything
A hundred trades is a floor, not a target. Under that, win-rate differences are luck. The fastest way to get there honestly is more data, not looser rules.
- 6Hold out data the strategy never saw
Tune on one stretch of history, then run the frozen rules on a later stretch once. If performance collapses out of sample, you fitted the past instead of finding an edge.
- 7Forward test the survivors
Paper trade or replay-trade the strategy in real time briefly before funding it. It catches execution problems a historical run cannot, like fills you would never actually get.
Step-by-step walkthrough with examples: how to backtest a trading strategy.
The five ways backtests lie
- Testing without costs, then discovering the edge was smaller than the fees.
- Tuning parameters until the past looks perfect (overfitting), then wondering why the future disagrees.
- Cherry-picking the window: every breakout strategy looks brilliant if you only test a trending year.
- Lookahead bias: using information (like the close of the current bar) that was not available at the moment of entry.
- Stopping at 20 trades because the equity curve already looks good.
Strategy backtest results
Original research: exact pinned rules, real data, fees on, losers published as losers.
Backtesting guides
Frequently asked questions
What is backtesting in trading?
Backtesting is testing a trading strategy against historical price data to see how it would have performed before risking real money. A proper backtest pins the entry, exit, and risk rules down mechanically, runs them bar by bar over past data with realistic costs, and measures the outcome with statistics like win rate, profit factor, expectancy, and maximum drawdown.
Is backtesting accurate?
Only as accurate as its assumptions. The common failure modes are ignoring fees and slippage, testing on a cherry-picked period, letting the strategy peek at future data (lookahead bias), and tuning parameters until the past looks perfect (overfitting). In our own study of 21 famous strategies, most looked viable before costs and none were profitable after a realistic 0.05% commission per side.
How many trades do you need for a reliable backtest?
As a rule of thumb, at least 100 trades before drawing any conclusion, and several hundred before trusting fine differences. Below that, the difference between a 40% and a 55% win rate is well within luck. Sample size matters more than the length of the calendar window.
What is the difference between backtesting and paper trading?
Backtesting replays historical data, so you can evaluate years of trades in hours; paper trading executes the strategy forward in real time on a simulated account, so it takes as long as real trading but includes live conditions. The strongest workflow is both: backtest to filter out losers cheaply, then paper trade the survivors briefly before going live.
Can I backtest for free?
Yes. The replay simulator on this site runs bar-by-bar on real BTC data in the browser with no account. Beyond that, the AI backtester takes a strategy described in plain English and runs it through a deterministic engine on real data, so you can test an idea without writing code.
Key terms
Precise definitions for the numbers a backtest produces: