How we test a strategy

The short version

1. 01
   One rule, written down

   A strategy is code with no discretion. It either signals on a candle or it does not, and the entry, stop and targets are fixed by the rule at that moment.

2. 02
   Real price history

   The rule is run over the candles of one market on one timeframe, from a price feed we operate ourselves. Every subscriber sees results from the same data.

3. 03
   One trade at a time

   A new signal is only counted once the trade before it has reached its first target or its stop. This is a book a person could actually have held.

4. 04
   Counted, then costed

   For each trade we record the furthest target it reached. Win rate, expectancy and profit factor are worked out from those counts, before or after the spread.

The targets and the stop

Risk is the distance from the entry to the stop, written 1R. Every trade has four targets at 0.5R, 1R, 1.5R, 2R from the entry. A trade that reaches a target before its stop is a win at that target, and a trade that is stopped first is a loss of 1R. So one trade can be a win at the first target and a loss at the fourth, and the win rate always names the target it belongs to.

A trade still open when the run is made is not scored. It is counted once it has closed.

What trading costs

Price history is recorded at the bid. A buy enters at the ask and exits at the bid, and a sell does the reverse, so every trade gives up one spread. Commission is added where the account charges it. We express that cost as a share of the trade's own risk: a cost of 0.10R means the spread and commission together are a tenth of the distance to the stop.

The spread does not move the stop or the targets, so it never changes which target a trade reached. It changes what the outcome pays: a win pays the target less the cost, and a loss costs 1R plus the cost. Spreads are measured from live market data for each market, not assumed.

How much history each timeframe has

A rolling window matters more than it sounds. When a setup has 50 trades in its window, losing two old ones and gaining two new ones can move its win rate by several points and change which target looks best. That is a property of small samples, not a fault in the rule, and it is why the trade count is printed beside every percentage.

The rules that stop a number claiming too much

• A record with fewer than 30 trades is marked as thin wherever it appears. The figure is shown, because hiding it would be worse, and it is never graded as a strong edge.
• A target must be worth at least 4 times the round-trip cost to be offered as an exit. Below that, the distance is too close to the spread for the price history to say reliably whether it was touched.
• A win rate is never rounded up to 100% or down to 0%. 373 wins from 374 trades prints as 99.7%.
• A profit factor with no losing trades prints as "no losses", not as a large number.
• Expectancy is reported per 100 trades in R, so a result from a small account and a large one read the same.

What a backtest leaves out

• Slippage. Orders are assumed to fill at the modelled price. In fast markets they will not.
• Swap and overnight financing on positions held across days.
• Your broker's own spread, which may be wider or tighter than the one we measured.
• Whether you would have taken every signal. The record assumes the rule was followed each time.
• The future. A rule that worked on past data has no obligation to keep working.

One source for every figure

Each nightly run stamps its results with the time it was made and the date its market data runs to. A screen in the terminal shows figures from one run, and says so when two sources disagree instead of printing both. The pages on this site follow the same rule: a page that quotes a figure prints the run it came from.

Read next

• Strategies

  What each strategy waits for, and its record across every market.

• Tools

  Free calculators for position size, challenge rules, ruin and sample size.

• Research

  What we measured, with the run it came from and the results that went against us.

• About

  Who is behind the terminal and what it does not do.