Backtesting: how to know whether your edge is real
Every trader believes their method works; backtesting is how you find out whether that belief survives contact with data. Done honestly it builds the confidence to execute through drawdowns and the humility to drop what does not work. Done carelessly, with hindsight and cherry-picking, it manufactures false confidence that the live account quickly corrects. This module is about testing in a way that tells you the truth, even when the truth is inconvenient.
Why one trade proves nothing
A single winning trade, or even five, tells you almost nothing; it is well within what random chance produces. An edge is a statistical claim, and statistical claims need samples. Aim for at least a few dozen, ideally a hundred or more, instances of one specific model before you conclude anything. The goal is not to find trades that worked, which always exist in hindsight, but to measure how a fixed rule performs across many examples.
Test one model at a time
Backtest a single, fully-defined model, not a vague feel for the method. Write the exact rules: the sweep, the structure shift, the point of interest, the stop and the target. Then apply them mechanically to historical data. If the rules are fuzzy you will unconsciously count the winners and forgive the losers. A model precise enough to backtest is a model precise enough to trade.
Avoid hindsight bias
The great danger is hindsight: on a chart you have already seen, every setup looks obvious. Fight it by testing bar by bar with the future hidden, using a replay tool, so you only decide with the information you would have had live. If you can only find your setups by looking at the completed chart, you have not found an edge; you have found a pattern your eye draws after the fact.
Include costs and be pessimistic
Record entries at the ask and exits at the bid, and include the spread, especially the wider spread around news and rollover. Where a candle could have hit either your stop or your target, assume the worse outcome unless you have tick data to prove otherwise. Honest, pessimistic accounting is what makes a backtest resemble live results instead of flattering you into a strategy that only works on paper.
Read the equity curve
Plot the running total of your results as an equity curve. A real edge shows a generally upward curve that is lumpy, with real drawdowns, not a straight diagonal line, which usually signals a bug or curve-fitting. Look at the depth and length of the worst drawdown; that is what you must be able to sit through live. A high average R with an unbearable drawdown is not a tradeable edge for most people.
Watch for curve-fitting
Beware of rules tuned until they fit the past perfectly. A model with many finicky conditions that make a specific history look flawless usually fails forward, because it learned that history's noise rather than a repeatable edge. Prefer simple rules that work acceptably across many conditions over complex ones that work beautifully on one slice of data. Robustness beats optimisation.
Forward-test before you scale
A backtest is a hypothesis; a forward test confirms it. Trade the model on current data at small or simulated size and check that live results resemble the backtest. Markets change, and a model that worked historically can decay, so forward-testing catches that before it costs real money. Only scale up a model after it has proven itself forward, not on backtest hope alone.
Make it a habit
Backtesting is not a one-time gate; it is ongoing maintenance. Periodically re-test your live models to see whether the edge is holding, and retire any that have clearly decayed. Use the replay function on the live XAUUSD chart to practise spotting your setups bar by bar. The journal in the next module is where your forward results are recorded and read.
FAQ
How many trades do I need to backtest?
At least a few dozen, and ideally a hundred or more, instances of one specific model. An edge is a statistical claim, and a handful of trades is well within what random chance produces, so small samples prove nothing either way.
What is hindsight bias in backtesting?
The tendency for setups to look obvious on a chart you have already seen. Fight it by testing bar by bar with the future hidden, so you only decide with information you would have had live. Setups you can only find on the completed chart are not an edge.
Do I need to include the spread in a backtest?
Yes. Record entries at the ask and exits at the bid, include the spread, and widen it around news and rollover. Honest, pessimistic accounting is what makes a backtest resemble live results instead of flattering a strategy that only works on paper.
What should a healthy equity curve look like?
Generally upward but lumpy, with real drawdowns. A perfectly straight diagonal line usually signals a bug or curve-fitting. Pay attention to the worst drawdown, because that is the stretch you must be able to sit through when trading live.
What is the difference between backtesting and forward-testing?
A backtest measures a model on historical data as a hypothesis; a forward test confirms it on current data at small or simulated size. Markets change, so forward-testing catches edge decay before it costs real money, and you scale up only after it proves itself forward.