How to backtest a strategy.

Before you put money behind a set of rules, you can ask a cheaper question: what would these rules have done on prices that already happened? Working that out honestly, trade by trade, is a backtest.

A backtest cannot see the future. What it can show early is that an idea lost money in the past, or that its losing runs were longer than you could sit through.

What do you need before you start?

You need rules so exact that two people testing them on the same chart would record the same trades. The entry and exit rules lesson covers how to write them.

This lesson uses one example system. It illustrates the method and is not a recommendation:

• Market: EUR/USD on the daily chart, in a US dollar account.
• Entry: when a candle closes above the 50-day simple moving average and the candle before it closed below, buy at the open of the next candle.
• Exit: a stop loss 30 pips below entry and a take profit 60 pips above.
• Size: 0.10 lot, with one trade open at a time.

Words like "strong trend" or "clean bounce" have no place in a testable rule. Replace them with something you can measure, or hindsight will quietly pick your trades for you.

You also need price history that covers rising, falling and sideways stretches. Choose the period before you see any results, so it cannot be bent to suit them.

How do you run a backtest step by step?

You can test by hand on a chart or with code, for example an Expert Advisor in MetaTrader's Strategy Tester. Code is faster and repeats exactly. Testing by hand is slower, and you see every setup with your own eyes.

The steps are the same either way:

1. Go to the start of your chosen period and hide everything to the right of the current candle, by scrolling or with a bar replay mode.
2. On each closed candle, check the entry rule. If it is met, write down the date, the entry price at the next candle's open, the stop and the target.
3. Move forward one candle at a time until the stop or the target is hit, then record the exit and the result in pips.
4. Subtract costs from every trade: the spread, any commission, swap for nights held and an allowance for slippage.
5. Repeat until the end of the period without changing a single rule.

Daily candles hide the order of events inside a day. If one candle touches both your stop and your target, you cannot tell which came first, so count it as a loss or check a lower timeframe.

How do you read the results of a backtest?

Suppose the example rules took 40 trades over the test period, with 16 reaching the target and 24 hitting the stop. These numbers are illustrative, chosen to show the arithmetic.

On 0.10 lot of EUR/USD in a US dollar account, one pip is worth $1, as the pip value calculator confirms. Assume an illustrative spread of 1.0 pip, so each round trip costs $1.

To keep the sums simple, this example counts only the spread. A real test would also add commission if your account charges it, and swap for each night a trade stays open.

1. Wins: 16 × 60 pips × $1 = $960.
2. Losses: 24 × 30 pips × $1 = $720.
3. Result before costs: $960 − $720 = $240.
4. Costs: 40 trades × $1 = $40.
5. Result after costs: $240 − $40 = $200.

After its $1 cost, each win nets $59 and each loss costs $31. Three measures summarise the log.

Expectancy is the average result per trade. Profit factor is the total won divided by the total lost. A drawdown is a fall in the account from a previous high, covered in the drawdown lesson.

Suppose the deepest drawdown in this log came from six losses in a row.

The breakeven win rate is the share of winners at which the rules make nothing. At 40% against 34.4%, this test clears it by 5.6 percentage points.

That margin is thin for 40 trades. A standard error measures how far a percentage could move by chance, and a rough range for the true win rate is the result plus or minus two of them.

Here that is √(0.40 × 0.60 ÷ 40) ≈ 0.077, so the range runs from roughly 25% to 55%.

The bottom of that range sits below breakeven, so this sample cannot rule out a system with no edge at all. Testing four times as many trades would halve the width of that range.

Mistakes that make a backtest look better than it is

Each error below pushes results upward, which is why they are easy to miss: a test that looks good rarely gets checked as hard as one that looks bad.

Look-ahead bias

Look-ahead bias means using information the rules could not have had at the moment of the decision. Entering at a candle's open because of how that same candle closed is the classic case.

It also creeps in through indicators that redraw past values, such as many zigzag tools, and through levels you drew using candles that sit to their right on the chart.

Leaving out costs

In the worked example, costs took $40 of the $240, one sixth of the result. A system with a 10 pip target pays the same 1 pip spread on a much smaller win, so costs eat a tenth of every winner.

Add commission, swap and slippage as well. The cost of one trade lesson shows how to total them per round trip.

Overfitting

Overfitting is tuning settings until they match the past too closely. Try every moving average from 40 to 60 days with stops from 20 to 40 pips, and some combination is likely to look excellent by luck alone.

The defence is to split your history before you start. Build and adjust the rules on the older part, for example the first 70%, called the in-sample data. Then run them once on the newer 30%, the out-of-sample data.

If the out-of-sample run falls apart, the in-sample result was probably fitted to noise. Going back to adjust the rules after seeing it turns that section into in-sample data too.

Cherry-picking

Starting the test just after a bad stretch, skipping trades that felt wrong or changing a rule halfway makes the log flattering and useless. If you change a rule, restart the test from the beginning.

What a good backtest does and does not tell you

Suppose your rules still show a profit after costs, on out-of-sample data and over enough trades. That tells you they had an edge in the past, on that data. Markets change, so it says nothing certain about next year.

The next check is a forward test on a demo account, where the rules meet prices nobody has seen yet, along with live spreads and your own reactions.

Keep the backtest log. Comparing later trades with it shows whether the rules are behaving as they did in the test or drifting away from those results.