How to backtest a forex strategy

01Write the rules before you open a chart

A strategy you can test is one a stranger could execute without asking you a question. That means five things in writing: what has to be true to enter, where the stop goes, how the trade is exited when it works, how position size is decided, and which sessions or days the strategy trades at all.

Anything you leave vague will quietly be decided in your favour once the bars are moving, and the result will be a record of your hindsight rather than a test of the rules. If a rule cannot be written down, it is a discretionary judgement, which is fine, but then what you are testing is your judgement and the sample needs to be bigger.

• Entry conditions, stop placement, exit rules, sizing, and the sessions it trades
• Name the strategy, so every trade you log can point at this version of it
• Write the conditions you will not trade in, such as the minutes around a release

02Pick a period that contains more than one market

The point of choosing a period is to include conditions you will actually meet: a trend, a range, a quiet holiday week and at least a few high impact releases. A test that covers one clean trending quarter tells you how the strategy behaves in a trend and nothing else.

Sample size matters more than calendar length. A hundred trades is a reasonable floor for an intraday strategy, and getting there on a 5 minute chart usually takes several months of data. For a swing strategy on 4 hour or daily bars, the same hundred trades may take a few years.

Know what your data can and cannot support. In the DT Terminal backtester the smallest bar is one minute, with forex, metals, crypto and CME futures including the US index contracts. There are no stocks, and futures history starts in August 2024. A strategy whose edge lives inside the spread, or inside a single minute, cannot be settled on minute bars in any tool, and the spread and slippage you assume are yours to justify.

• Cover a trend, a range and a news week, not one flattering quarter
• Aim for a hundred or more trades before you believe a win rate
• Check the data actually exists for the symbol and timeframe you trade

03Replay bar by bar, never scrolled

The mechanical requirement of an honest manual backtest is that you cannot see the candle you are deciding on. Scrolling a chart backwards and marking entries is not a backtest, because the outcome is already on the screen.

So step the replay forward one bar at a time and decide with only what is left of the chart. In DT Terminal the space bar plays and pauses, the arrow keys step a candle, and L and S open a long or a short at the cursor price. Set the risk percentage and a commission per lot when you create the session, so the simulated cost resembles what you actually pay.

• One bar at a time, deciding before the bar closes where the rules say so
• Set risk and commission at the start, so the result is net of costs
• Keep the session to one strategy, so the sample means something

04Log every trade while you are in it

Record the trade as you take it, not from the summary afterwards: the setup, the reason, the stop and target at entry, and any rule you broke. A backtest with no reasons attached can tell you the strategy lost money, but not whether the strategy lost money or you did.

It helps to name what the session is for before it starts. Practising a setup, validating a written strategy, and drilling one weakness such as moving stops are three different sessions, and mixing them produces a sample you cannot read.

• Reason and mistakes on the trade itself, while the bar is still on screen
• One session, one purpose, so the numbers answer one question
• Screenshot the chart at entry when the setup is visual

05Read expectancy, not the equity curve

An equity curve is a picture, and over a small sample it is mostly a picture of luck. The number that answers whether the strategy is worth trading is expectancy: the average result of a trade, expressed in R, where one R is the amount you risked.

Expectancy of plus 0.2R means each trade returned a fifth of what it risked on average, so a hundred trades returned about 20R before you touch position sizing. It combines win rate and average win against average loss, which is why a 40% win rate can beat a 70% one. Read it beside the sample size, the largest losing run and the maximum drawdown, because those decide whether you could have sat through it.

Then look for where the result comes from. If the whole edge sits in one instrument, one session or three outlier trades, you have found a coincidence rather than a strategy.

• Expectancy in R, with the trade count beside it
• Longest losing run and maximum drawdown, because those are what you have to endure
• Break the result down by instrument, session and day, and see if it survives

06Forward test on live prices before any money

A backtest is a test against known history, and you were the one choosing which bars to trade. A forward test removes both problems: the same written rules, on prices nobody has seen yet, at the times you would really be at the screen.

Log forward tests as trades in their own right, in a separate account so they never touch your live numbers. In DT Terminal a forward test account is created the first time you flip the pill on the trade form, forward tests still count toward your logging streak, and a finished backtest session can be exported into the trade log as forward test trades on Pro.

Run it long enough to hit a real sample and to include a week you would rather forget. If the forward numbers look nothing like the backtest, the difference is usually spread, slippage, or the trades you took in the test that you would never take at 3am.

• Same rules, unseen prices, real session times
• A separate account, so practice never flatters your live record
• Compare expectancy against the backtest, not the profit

07Go live small, and keep comparing

Live trading adds the two things neither test can simulate: your money and your nerves. Start at a size where a losing run is boring, and keep logging the same fields with the same tags, so the live sample can be read against the tested one.

The comparison is the whole point of having journalled the backtest. Same strategy tag, same fields, three columns: tested, forward tested, live. Where live expectancy falls short, the gap is either execution, which your mistake tags will show, or costs, which your commission and swap fields will show.

• Size so that the drawdown you measured is survivable
• Same tags across backtest, forward test and live, or the comparison is meaningless
• Review by strategy, not by day, once there is a live sample to read