After 20 trades with 13 wins, is the win rate 65%? A Bayesian view says: probably not exactly, and here's how uncertain it is.
Start with a prior belief, such as "the win rate is somewhere between 40% and 60%, most likely near 50%". As trades accumulate, update the belief. With a simple Beta distribution, the update is easy: add the wins and losses to the prior's parameters. Starting from a flat prior, after 40 wins in 70 trades the posterior mean is about 57%, and the range of plausible values runs from roughly 45% to 68%.
That wide range is the lesson. Even 70 trades leave huge uncertainty about the true win rate. A strategy with a 57% observed rate could easily have a true rate of 48%, which would be a losing system after costs.
The approach also encourages sensible behaviour: more data narrows the range, and decisions can be tied to the probability that the true edge exceeds some threshold, such as the break-even win rate.
Spreadsheets handle the Beta distribution through built-in functions, so the calculation takes minutes.
A tip for spreadsheet users: the BETA.INV function returns the percentile of a Beta distribution, so the 2.5% and 97.5% points give a credible interval directly. With wins and losses in two cells, the whole update takes a single formula, and it can be recalculated after every session.
It also supports a practical decision rule. Instead of asking "is the win rate above 50%?", ask "what's the probability that the true win rate exceeds the break-even level after costs?" If that probability is only 60%, the evidence is thin, whatever the point estimate says.
What is the plausible range for the current strategy's true win rate, given the number of trades logged?
Bayesian thinking: updating a win rate as trades accumulate
-
Shadow Trader
- Posts: 5594
- Joined: Wed Sep 09, 2026 5:19 pm
-
FXS London Scalper
- Posts: 976
- Joined: Sun Oct 18, 2026 12:25 pm
Re: Bayesian thinking: updating a win rate as trades accumulate
Thirteen wins out of twenty is consistent with a true win rate anywhere from about 45% to 80%. That range is the honest answer, and it's why changing a system after twenty trades usually means reacting to noise. A Beta prior updated trade by trade makes the narrowing visible over time.
FXS Official Team | London Session Desk
-
Amanhasnoname
- Posts: 2060
- Joined: Fri Oct 30, 2026 9:45 am
Re: Bayesian thinking: updating a win rate as trades accumulate
This was a real eye opener for me when I first tried it. I had a breakout system that won fifteen of its first twenty two trades and I was ready to double my size. A friend who works with statistics sat me down, put the numbers into a Beta distribution in a spreadsheet, and showed me that the plausible range ran from somewhere in the low fifties up to the high seventies. Given my average loss was a bit bigger than my average win, the bottom of that range was not a winning system at all.
So I kept the size where it was and carried on logging. By around a hundred and twenty trades the observed rate had drifted down to just under sixty percent, and the interval had tightened enough that I could see it was probably above my break even, but only modestly.
The habit I kept from that is a single cell that shows the probability the true rate beats break even after costs. I only allow myself to change size when that number has stayed above ninety percent for a few weeks, and it stops me reacting to every good or bad patch.
So I kept the size where it was and carried on logging. By around a hundred and twenty trades the observed rate had drifted down to just under sixty percent, and the interval had tightened enough that I could see it was probably above my break even, but only modestly.
The habit I kept from that is a single cell that shows the probability the true rate beats break even after costs. I only allow myself to change size when that number has stayed above ninety percent for a few weeks, and it stops me reacting to every good or bad patch.