Backtest far enough back to collect a large sample that spans trends, ranges, and volatile events - usually one to three years, and more for infrequent strategies. The goal is condition variety and trade count, not a calendar milestone. A single quiet year proves less than the same number of trades pulled from three different market moods.
Two questions decide the window
- How often does the strategy trade? Scalpers reach 100+ trades in weeks of history; swing traders may need years to gather the same count.
- Does the window contain variety? The history must include at least one strong trend, one range, and one high-volatility stretch, or you have not tested the strategy's weak spots.
Answer both and the amount of history follows naturally - it is really a restatement of how many trades you need, expressed in time.
Why more is not always better
It is tempting to test a decade and feel thorough, but very old data can mislead. Spreads were different, volatility regimes shift, and market structure evolves. Price behavior from many years ago may reflect a market that no longer exists, so a strategy tuned to it can look better or worse than it would today. Recent history that still contains variety is usually more informative than stale history that merely adds length.
Important: do not choose your window to flatter the strategy. Picking exactly the years where a setup thrived is a form of curve fitting. Choose the window for variety and recency, then accept whatever the data says.
A practical default
For most discretionary forex strategies, one to three recent years that include a couple of trends, at least one prolonged range, and a volatile news-driven period is a sensible target. If that window does not yet give you 100 trades, extend further back until it does - keeping an eye on whether the older data still resembles current conditions.
Jump across history quickly
You do not have to replay years in sequence. In a simulator you can jump to different start dates and sample a trend from one year, a range from another, and a volatile stretch from a third, assembling a varied sample without slogging through every bar in between. That lets you cover the right conditions efficiently rather than being forced to test one continuous block of time.
Backtest data range FAQ
How far back should you backtest?
Far enough for a large sample across several regimes - usually one to three years, more for low-frequency strategies. Trade count and variety define it, not a fixed number of years.
Is more history always better?
No. Very old data may reflect spreads, volatility, and structure that no longer apply. A few recent varied years often beats a decade of stale data.
Should I backtest recent or old data?
Favor recent history for relevance, but ensure the window still contains a trend, a range, and a volatile period. Recency plus variety beats raw age.