Win rate

What it is

Win rate is the share of trades that made money.

It is the first number most people ask for, and the least informative one on its own.

Why it misleads

Win rate tells you how often you are right. It says nothing about how much you make when right or lose when wrong. Those are two different facts, and only together do they decide whether a strategy makes money.

Expectancy per trade is:

expectancy = win_rate × payoff_ratio − (1 − win_rate) × 1

Put numbers in. A strategy winning 90% of the time, making 1 on a win and losing 20 on a loss:

0.90 × 1 − 0.10 × 20 = 0.90 − 2.00 = −1.10 per trade

Nine wins out of ten, and it still bleeds. That is not a contrived example — it describes most strategies that sell options, average down, or hold losers “until they come back”.

Why it matters for AI quant

A model optimises whatever you measure. Ask it to maximise accuracy and you get a model that predicts the common case well and is silent about the tail — which is exactly the failure mode of a high win rate.

This is why overfitting to win rate is so easy to do by accident: the metric rewards being right often, not being right where it counts.

How to check

Never accept a win rate alone. Ask for three numbers together:

Number	Why it is needed
Win rate	how often
Payoff ratio	how much
Number of trades	whether the sample means anything

Two out of three is not enough. Any two can be made to look good while the third hides the problem.