Backtesting and Forward-Testing a Setup

What you will learn

• How to write rules precise enough to test.
• How to record results in R and calculate expectancy and profit factor.
• Why sample size matters and how a few trades can change everything.
• How to avoid curve-fitting and look-ahead bias, and how forward-testing helps.

Explanation

Backtesting means applying fixed rules to past data and recording what would have happened. Forward-testing means applying the same rules to new data as it happens, in a demo account or with very small size. A good backtest is evidence about the past, not a promise about the future.

Testable rules

Every rule must be answerable with yes or no. "Strong trend" is not testable. "Daily close above the 50-day SMA and the last daily swing low unbroken" is testable. Write down: market, timeframes, setup conditions, entry trigger, stop, target or exit, and the maximum risk per trade.

Recording in R

Record each result as a multiple of the planned risk. From the list you calculate:

• Win rate = winning trades ÷ all trades
• Expectancy = total R ÷ number of trades
• Profit factor = sum of winning R ÷ sum of losing R (as a positive number)
• Longest losing streak and largest drawdown (biggest fall from a peak in your running total)

Include estimated costs (spreads, fees, slippage) as a deduction in R per trade.

Sample size

A small sample is noisy. The fewer trades, the more one or two lucky trades decide the result. Collect as many trades as you reasonably can, from different market conditions (trends, ranges, high and low volatility), before you trust a number.

Curve-fitting and look-ahead bias

Curve-fitting (over-optimisation) means adjusting rules and settings until the past looks perfect. The rules then describe the past noise, not a repeatable behaviour. Protection: keep rules simple, fix them before testing, and test on a separate period you did not use while building the rules.

Look-ahead bias means using information that was not available at the time of the decision. Examples: using a swing point before it was confirmed, or using the day's close to decide an entry during that day. When testing by hand, scroll the chart forward one candle at a time.

Worked example

Hypothetical example: 40 recorded backtest trades of one setup. Planned risk €100 per trade (1R).

1. Wins: 16 trades, total +33.6R (average +2.1R). Losses: 24 trades, total −24.0R.
2. Win rate: 16 ÷ 40 = 40%
3. Net result: 33.6R − 24.0R = +9.6R. Expectancy: 9.6R ÷ 40 = +0.24R per trade
4. Profit factor: 33.6 ÷ 24.0 = 1.4
5. Estimated costs 0.1R per trade: 40 × 0.1R = 4.0R. Net after costs: 9.6R − 4.0R = +5.6R, or 0.14R per trade (about €14 per trade at €100 risk). Taxes are excluded.

Sensitivity check, before costs: remove the two best trades (+4.5R and +4.0R, total +8.5R). Then 9.6R − 8.5R = +1.1R over 38 trades ≈ +0.03R per trade. The edge almost disappears. This sample is far too small to trust, so the next step is more backtest data and a forward-test.

Common mistake

Practice activity

Hypothetical results in R of 10 trades: +2, −1, −1, +3, −1, +1.5, −1, −1, +2, −1. Calculate win rate, net R, expectancy and profit factor. Then write one reason why 10 trades are not enough.

Key takeaways

• Only yes/no rules can be tested honestly.
• Record every trade in R, including costs, and calculate expectancy, profit factor and drawdown.
• Small samples mislead; check how much a few trades drive the result.
• Avoid curve-fitting and look-ahead bias; confirm with a forward-test on new data.

Next step

You have finished this course; continue with "Risk & Trade Management" to turn a tested setup into a complete plan for position sizing, drawdowns and trade handling.