Trading Plan Backtesting Basics

What a bar that hits both your stop and your target does to the result

This is the quiet one. On any timeframe above a minute, some bars have a range that covers both your stop and your target, and the historical data does not record which came first. The backtest has to assume something, and the assumption moves the answer more than most people expect.

In the 120-trade sample, nine trades had both levels inside a single four-hour bar. Resolve all nine as targets and the record shows 55 winners out of 120, a 45.8% win rate and an expectancy of 9.5 pips a trade. Resolve them as stops and it shows 46 winners, 38.3% and 5.0 pips. Same rule, same data, nearly double the result, decided by a default nobody wrote down. The honest convention is to resolve every ambiguous bar as a loss, or to drop those specific bars to a one-minute chart and read the sequence.

Four ways a backtest flatters itself

• Overfitting. Test 40 parameter combinations on the same three years and the best of the 40 will look good even if all 40 are noise, because you selected on the noise. The fix is fewer parameters and a held-back period, not a better optimiser.
• A chosen date range. Starting the test in the month your idea would have worked is selection by another name. Fix the dates before you write the rule.
• Look-ahead. Using a session high inside the session, a daily close before the day ended, or a revised economic print that was not published at the time. Each one leaks the future into the entry.
• Memory. Scrolling a chart and counting the setups you notice is a survivorship test. Losing instances of the same setup are genuinely harder to see, which is why the rules have to be applied mechanically rather than spotted.

Forward-testing is the part that cannot be fitted

Hold back the most recent twelve months before you start tuning, and run the finished rule on that period once. If the out-of-sample expectancy lands within roughly 40% of the in-sample figure, that is ordinary sampling variation. If it halves, or changes sign, you fitted the rule to the first period. Paper trading alongside the backtest does the same job forwards, and adds the two things historical data cannot show: the fills you actually get, and whether you follow the rules when money is moving.

Then keep the frame honest. A favourable backtest is evidence about the past behaviour of a market, not a forecast, and price does not repeat because a spreadsheet says it used to. What a backtest genuinely earns you is a realistic distribution of outcomes, which is the input every other part of the plan needs. Feed the win rate and reward ratio into the risk of ruin calculator and vary only the risk percentage to see which sizing the record can carry, and read the risk management guide for the ceilings that sit above it.

A first pass you can run this weekend

Open 40 consecutive weeks of one instrument on one timeframe, hide the right-hand side of the chart, and step forward bar by bar. Log every trade the written rules trigger, with seven columns: date, direction, entry, stop, target, exit and the reason it closed. Do not skip one because it looks wrong. Forty trades will not settle whether the strategy works, but it settles something more useful first, which is whether a stranger could execute your rules without asking you a question.