LEARN · EN · easyquanttrading.com
What a backtest really is
The word does a lot of quiet work. A backtest sounds like a test — something you either pass or fail — and that framing leads directly to the mistakes most traders make with it. It is not a test. It is a measurement, and measurements only mean something when you know exactly what was measured.
- A backtest measures fixed rules against fixed history
- It cannot forecast, because the market is not obliged to repeat
- Every result is conditional on cost and fill assumptions
- It is very good at rejecting ideas, and poor at endorsing them
The precise definition
A backtest takes a set of rules, applies them to a historical price series in order, and records what would have happened under a stated set of execution assumptions. Every part of that sentence is load-bearing.
The rules must be complete, because anything left to judgement was not measured. The series must be fed in order, because using later data at an earlier time invalidates everything. And the execution assumptions must be stated, because a result at zero cost and a result at your broker's real cost are two different numbers that can point in opposite directions.
Why it cannot forecast
A forecast requires that the future resembles the sample in the ways that matter. Sometimes it does, for a while. Nothing obliges it to, and there is no way to verify that it will before the fact.
This is not pessimism, it is the structure of the problem. The same reasoning applies to every statistical model of markets: the estimate is about the sample, and the transfer to a new sample is an assumption layered on top.
The one question it answers well
A backtest is genuinely good at rejecting ideas. If a strategy loses money across a long history with realistic costs, it is almost certainly not worth more of your time. That is a real and valuable service, and it is where most of the benefit lies.
It is much weaker at endorsing ideas. Passing a backtest means 'this was profitable in the past under these assumptions', which is a much smaller claim than people hear. A trader who internalises only this asymmetry will do better than one who memorises every performance metric.
What the assumptions cost you
The gap between the backtest and reality is produced by assumptions, and they are worth listing explicitly because each one has a direction.
| Assumption | Default in most tools | Direction of the error |
|---|---|---|
| Spread | Fixed, often a default pip value | Understates cost when it widens |
| Commission | Zero unless entered | Understates cost |
| Fill price | Signal bar's close | Unobtainable; flatters entries |
| Slippage | Zero unless modelled | Understates cost, worst on breakouts |
| Data quality | Whatever the broker supplies | Unknown direction, sometimes large |
Five assumptions, four of which make the result look better than reality by default.
What to do with a result
Use it to filter, not to decide. Run the strategy with costs you would actually pay, on data you did not use to design it, and with an execution delay. If it survives those, you have an idea worth observing on live prices. If it does not, you have saved yourself an account.
And keep the number in proportion. A backtest is a description of one path through history. There were many other paths that the same rules would have taken under slightly different conditions, and you have measured exactly one of them.
FAQ
- How long should a backtest be?
- Long enough to include a genuinely difficult period for that market. That is usually several years for daily data and requires careful thought for intraday data, where long histories are often unavailable or unreliable.
- Can a backtest be trusted at all?
- Yes, for the narrow claim it supports: these rules, over this history, under these assumptions, produced this result. Distrust begins when that claim is silently upgraded into a statement about the future.
- Is a backtest the same as an optimisation?
- No, and the distinction matters. A backtest evaluates one set of rules. An optimisation searches many sets and keeps the best, which introduces a selection effect that has to be corrected for before the winner's result means anything.
More guides
- How EasyQuant validates strategies — evidence you can filter
- Honest backtesting, not pretty curves
- Gold strategy research that stays honest
- Overfitting detection: catch it before you deploy
- System Forge: design, then prove
- Walk-forward analysis: the only backtest that fights overfitting
- MT5 export without custody
- Glass box, not black box AI signals
Not investment advice. Historical results do not guarantee future performance. EasyQuant is a research factory — you execute on accounts you control.