Why Win Rate Is Misleading: Expectancy, Risk-Reward and the Martingale Trap

Win rate is misleading because it counts how often trades win but ignores how much they win or lose. A signal service that wins 80% of its trades can still lose money steadily if its average loss is several times its average win, and one that wins only 35% can be profitable if its winners are much larger than its losers.

In short: what matters is expectancy, the average result per trade once wins and losses are both counted. Expectancy = (win rate x average win) minus (loss rate x average loss). A win rate means nothing until it is paired with the average win and the average loss, ideally after costs. Very high win rates are often a symptom of a risky design, such as wide or missing stops, or doubling down after losses.

The pitch you have probably seen

“92% accuracy.” “9 out of 10 signals hit take profit.” Win rate is the favourite number in signal marketing because it is easy to understand and easy to make large. It is also, on its own, close to useless as evidence of skill. The CFTC has warned that bogus forex signal schemes often play on the “hot hand” bias, the belief that recent wins mean more wins are coming.

Expectancy: the number that actually matters

Expectancy is the average profit or loss per trade:

Expectancy = (W x average win) - (L x average loss)

where W is the win rate and L = 1 - W is the loss rate. Measure it in pips or in dollars per trade; it is positive if the system made money on average and negative if it lost.

Four example systems, before costs:

System	Win rate	Average win	Average loss	Expectancy per trade	Per 100 trades
A	80%	10 pips	50 pips	-2 pips	-200 pips
B	35%	60 pips	20 pips	+8 pips	+800 pips
C	90%	5 pips	60 pips	-1.5 pips	-150 pips
D	50%	30 pips	30 pips	0 pips	0 pips

Working through A: 0.80 x 10 = 8 pips gained on average from winners, and 0.20 x 50 = 10 pips lost on average from losers, so the system loses 2 pips per trade. At $10 per pip (one standard lot on a USD-quoted pair like EUR/USD; see pips, lots and signal math), that is -$2,000 per 100 trades, despite winning four trades out of five.

System B wins barely a third of the time and makes 0.35 x 60 - 0.65 x 20 = 21 - 13 = +8 pips per trade.

Costs make it worse, and they hit high-win-rate systems hardest

Every trade pays the spread (and sometimes commission). Assume a total cost of 1 pip per trade, which shrinks each win by a pip and enlarges each loss by a pip:

System A: 0.80 x 9 - 0.20 x 51 = -3.0 pips per trade.
System B: 0.35 x 59 - 0.65 x 21 = +7.0 pips per trade.

A system that takes many small wins pays the same cost per trade as one that takes few large wins, so costs eat a far bigger share of its edge. Transaction costs in retail FX goes into this in detail.

The break-even win rate for any risk-reward ratio

The risk-reward ratio (R) is the average win divided by the average loss. Setting expectancy to zero gives the win rate you need just to break even, before costs:

Break-even win rate = 1 / (1 + R)

Average win vs average loss (R)	Break-even win rate
1 : 5 (R = 0.2)	83.3%
1 : 4 (R = 0.25)	80.0%
1 : 3 (R = 0.33)	75.0%
1 : 2 (R = 0.5)	66.7%
1 : 1	50.0%
1.5 : 1	40.0%
2 : 1	33.3%
3 : 1	25.0%

A provider claiming 80% wins is telling you something only if their average loss is smaller than four times their average win. If their typical stop is 50 pips and their typical target is 10, 80% is exactly break-even before costs, and a loser after.

How win rates get inflated

Win rate is easy to raise without improving results at all. Common ways:

Taking profit early and letting losses run. Closing winners at +5 pips and holding losers until they hit -40 can produce an 88% win rate with an expectancy of 0.88 x 5 - 0.12 x 40 = -0.4 pips per trade.
Wide or missing stop losses. A trade with no stop loss can sit in a loss until the market comes back, so it “eventually wins,” until the time it doesn’t.
Counting partial closes as wins. Taking a small profit on part of a position, then counting the trade as a win even if the rest is stopped out.
Not reporting some losers. Signals that hit their stop are deleted, edited, or described as “cancelled” or “breakeven.”
Averaging down. Adding to a losing position so a small bounce turns the whole thing into a “win.”

The last one leads to the most dangerous pattern of all.

Martingale: winning almost every time, then losing almost everything

A martingale system doubles the position after every loss, so that one win recovers all previous losses plus the original target. Grid systems, which open more positions at fixed intervals as price moves against them, follow similar logic. Both can produce win rates near 100% for long stretches. The full maths is in martingale and grid strategies explained; here is the core of it.

Setup. A $1,000 account. Each trade risks the same amount it aims to make (a 1 : 1 risk-reward), with a 50% chance of winning, ignoring costs. The base stake is $10; after each loss the stake doubles.

Losses in a row	Stake on that trade	Cumulative loss	Balance left
1	$10	$10	$990
2	$20	$30	$970
3	$40	$70	$930
4	$80	$150	$850
5	$160	$310	$690
6	$320	$630	$370
7	$640 needed		not enough money

Each “cycle” ends the first time a trade wins, and every completed cycle nets exactly +$10. The account can survive six losses in a row. The chance of six losses in a row is 0.5 raised to the 6th power, or 1.56%, so 98.4% of cycles win. That is the win rate the marketing will show.

But the one losing cycle costs $630, wiping out 63 winning cycles at once, and leaves too little to keep doubling. Over 100 cycles, the chance of hitting at least one six-loss streak is 1 minus 0.984375 raised to the 100th power, about 79%. Before costs, the expected profit per cycle is exactly zero: 0.984375 x $10 - 0.015625 x $630 = $0. With spreads included the odds tilt against you, and if losses happen 52% of the time instead of 50%, the expected result per cycle becomes about -$2.65.

A blow-up path. Suppose the system runs 30 winning cycles in a row. The account grows to $1,300 (+30%), and the record looks flawless. Now it hits seven losses in a row (which a $1,300 balance can just about fund): stakes of $10, $20, $40, $80, $160, $320 and $640 add up to $1,270. The balance falls to $30, a 97.7% loss from the peak. A seven-loss streak at 50/50 odds has a 0.78% chance on any given cycle: rare on a given day, close to inevitable over enough cycles.

What to ask for instead of a win rate

Average win and average loss, in pips and in dollars at a stated position size.
Expectancy per trade, after costs.
The largest single loss and the longest losing streak.
Maximum drawdown, measured on open (floating) positions, not just closed trades. See drawdown and risk of ruin.
Whether position size ever increases after a loss. If it does, treat any high win rate as a warning.
The number of trades. An expectancy estimated from 30 trades is mostly noise.

What can go wrong

The average loss is hidden. Many providers publish wins in pips and losses as “small” or not at all.
Rare large losses haven’t happened yet. A short record from a martingale or grid system can show no losses at all.
Costs and slippage. A thin positive expectancy before costs can easily be negative after them.
Correlated signals. Five “separate” winning trades on EUR/USD, GBP/USD and EUR/GBP can be one bet on the dollar or the euro.

The short version

Win rate answers “how often,” not “how much.” Expectancy combines both and is the only way to see whether a system makes money on average. Any win rate can be break-even or worse with the wrong risk-reward ratio, and the highest win rates often come from designs that take small profits and carry large, occasional losses. A martingale can win 98% of its cycles and still lose most of an account in one bad streak.