Six numbers a strategy report hides, and how to recover them

Take an example report that reads profit factor 1.80, win rate 45%, 180 trades. It is an illustration, not a result of mine. Every number below is computed exactly from those three inputs, so anyone can reproduce it.

1. The payoff it implies

Profit factor is gross profit over gross loss. Divide both by the number of trades and you get a relationship between the average win, the average loss and the win rate:

payoff = PF × (1 − win rate) / win rate = 1.80 × 0.55 / 0.45 = 2.20

The report is claiming that the average winner is 2.2 times the average loser. That is a statement about exits and fills, not about the win rate, and it is the part of a backtest most exposed to how the engine handles fills.

2. The breakeven win rate for that payoff

At a payoff of 2.2, the win rate that breaks even is:

breakeven = 1 / (1 + payoff) = 1 / 3.2 = 31.25%

So the headline 45% is not the edge. The edge is the gap between 45% and 31.25%, about 13.75 points. That gap is what costs, slippage and bad luck have to eat through before the system stops making money.

3. How wide the 45% really is

A win rate measured on 180 trades has a margin of error. With 81 wins in 180 trades, a Wilson 95% interval runs from 37.9% to 52.3%.

The whole interval sits above the 31.25% breakeven. For this report, sample size is not the objection, and it is worth saying so first. A review that leads with the check the system passes earns a hearing for the two it does not.

4. Expectancy, in units of one average loss

expectancy = win rate × payoff − (1 − win rate) = 0.45 × 2.20 − 0.55 = 0.44

Per trade, the system expects to make 0.44 of whatever it loses when it loses. That is a ratio, not money. Everything that follows turns on how large that average loss is.

5. The cost that zeroes it

Costs come out of that 0.44. Suppose the average loss is 0.50% of the position, an assumption for this example, not a measurement. Then the edge is gone at:

0.44 × 0.50% = 0.22% round trip, or 22 basis points

That is commission, spread and slippage together. Whether 22 bps is comfortable or fatal depends entirely on where and how the system trades. Read the report's commission and slippage settings rather than assuming them; a backtest run at zero cost will draw exactly the same kind of chart as one that was priced.

6. Where the argument actually is

Put the five together and the win rate stops being the question. It survived its interval. The result lives or dies on two things the screenshot never shows:

• the cost line, against a 22 bps budget, and
• the payoff, which rests on how the engine filled exits, including the bars where price touched both the stop and the target. An engine working from bar data has to assume which came first, and the report rarely says which.

A quick stress test shows why the payoff matters. If the live payoff came in 20% lower, at 1.76, the breakeven win rate rises to about 36.2% and expectancy falls from 0.44 to about 0.24 average losses. The lower end of the interval, 37.9%, is then less than two points clear of breakeven. Same win rate, much thinner margin.

The order matters

Six numbers, three inputs, no code:

1. payoff, 2.20
2. breakeven win rate, 31.25%
3. Wilson 95% interval on the win rate, 37.9% to 52.3%
4. expectancy, 0.44 average losses
5. the cost that zeroes it, 22 bps on a 0.50% average loss
6. the payoff under a 20% haircut, and what it does to the margin

Run them in that order and you will know which questions to ask before you ever see the code: what were the costs, and how were exits filled?

These are the same checks I run before I believe any backtest, including my own.

Get the Backtest Verification Checklist