Learn / Risk & Position Sizing
Expectancy vs win rate
A win rate on its own is uninterpretable. A strategy winning 80% of its trades can lose money and one winning 30% can make it — what decides is expectancy, the average result per trade measured in units of risk.
The reason win rate persists as a headline number is that it feels like a quality score. It is not. It is one of two terms, and the two trade against each other by construction.
The formula
Using R multiples, where 1R is the amount risked:
expectancy = (win rate × average win in R)
− (loss rate × average loss in R)Positive means the approach makes money over enough trades. Negative means it does not, regardless of how the last few felt.
Three strategies, same question
| A | B | C | |
|---|---|---|---|
| Win rate | 80% | 80% | 30% |
| Average win | +0.2R | +0.5R | +4R |
| Average loss | −1R | −1R | −1R |
| Expectancy | −0.04R | +0.20R | +0.50R |
A: (0.8 × 0.2) − (0.2 × 1.0) = 0.16 − 0.20 = −0.04R
B: (0.8 × 0.5) − (0.2 × 1.0) = 0.40 − 0.20 = +0.20R
C: (0.3 × 4.0) − (0.7 × 1.0) = 1.20 − 0.70 = +0.50RA and B have identical win rates and opposite signs. A wins four times out of five and loses money, because the wins are too small to cover the losses.
C loses seven times out of ten and is the best of the three, by a factor of two and a half over B.
So the win rate column tells you nothing on its own — it is identical for the worst strategy and a good one, and lowest for the best. It is not that high win rates are bad. It is that the number is not an answer to any question you have.
Why the two terms move against each other
This is the part that makes win rate seductive and misleading.
A closer target is hit more often and pays less. A distant target is hit less often and pays more. So improving win rate by bringing targets in mechanically reduces average win, and improving average win by pushing targets out mechanically reduces win rate.
The break-even line makes the trade explicit. For a strategy with planned R, the
win rate required to break even is 1 ÷ (1 + R):
| Planned R | Break-even win rate |
|---|---|
| 0.5R | 67% |
| 1R | 50% |
| 2R | 33% |
| 3R | 25% |
| 5R | 17% |
A 0.5R approach needs to be right two times in three. A 3R approach needs one in four. Neither is better until you know which win rate you can actually achieve at that R — which is an empirical question about your setups, not a preference.
The risk/reward calculator computes both from your own inputs.
The psychological trap
Win rate is emotionally salient in a way expectancy is not. Being right feels good, being wrong feels bad, and a high win rate delivers more of the first.
Which produces a specific and very common failure: cutting winners early and letting losers run. Taking profit at +0.3R raises the win rate visibly. Giving a loser room converts a −1R into a −1.5R invisibly. Both feel like management. Together they hollow out expectancy while the number you were watching improves.
The diagnostic is straightforward if you keep records: if your win rate is rising while your average win in R is falling, you are trading the wrong metric.
Why sample size matters more than it seems
Expectancy is an average, and averages over small samples are noise. Twenty trades tells you almost nothing — a run of good luck at a 40% win rate is entirely ordinary, and so is a run of bad luck at 60%.
Worse, the variance is driven by the winners. A strategy with a 30% win rate and +4R wins has most of its return concentrated in a minority of trades, so a sample that happens to miss them understates the edge badly. High-R, low-win-rate approaches need substantially more trades before their statistics mean anything.
This connects directly to losing streak probability: at a 30% win rate, a run of seven losses is unremarkable, and it will arrive while the sample is still too small to interpret.
What to record
Per trade, the minimum that makes expectancy computable:
- Planned R at entry
- Realised R at exit
- Whether it was a win or a loss
- The setup type, so expectancy can be computed per setup
That last one is where the useful findings are. Aggregate expectancy can be positive while one setup is quietly negative and being subsidised by the others — and you cannot see that without the breakdown.
FAQ
What is a good expectancy?
Positive, sustained over a sample large enough to mean something, is the bar that matters. The magnitude is less comparable across approaches than it looks, because expectancy per trade interacts with how many trades you take — a small edge at high frequency and a large edge at low frequency can be equivalent. What is not ambiguous is the sign.
Is a high win rate bad?
No — it is uninformative. A high win rate paired with an average win larger than the average loss is excellent. Paired with small wins and full-size losses it can be negative. The reason it gets criticised is that it can be manufactured by cutting winners, which improves the visible number while damaging the one that matters.
How many trades before expectancy means anything?
More than people assume, and more for high-R approaches than low-R ones, because their returns are concentrated in fewer trades. A sample of twenty is noise under any configuration. The practical check is whether your estimate is stable as new trades arrive — if it moves materially with each one, the sample is still too small.
Can I improve win rate and expectancy at the same time?
Sometimes, by better selection rather than by moving targets. Filtering out setups that are individually negative raises both, since you are removing trades rather than changing their shape. What does not work is bringing targets closer: that trades average win for win rate at roughly break-even, and often worse once costs are included.