What is win rate?
How often is the direction of the signal correct?
That is, for a “buy signal” triggering at time t, how often did the price go up from t to t+1? Conversely, for a “sell signal”, how often did the price go down?
Here’s an example:
Suppose that CoinFi Trading Signals predicted at 12:00 AM that prices for ETH could drop in the next 12 hours. If the the price of ETH is lower at 12:00 PM than it was at 12:00 AM, the signal was correct and the win rate for the signal increases. If prices decrease in the next 12 hours 70% of the time, the win rate is 70%.
For our trading signals, we aim to achieve a win rate well above random performance (50%).
If you are more familiar with terminology from machine learning than trading (like me!), win rate is effectively precision for binary classification.
Note: Regarding precision vs recall: we care less about recall since we are not really trying to identify every price increase (or decrease). We’re more interested in getting it right when we do trigger a signal.