Probability: Beat the Base Rate First
Probability in trading is a frequency over many repetitions, not a statement about the next trade. A signal is only worth taking if it beats the base rate — what the market does anyway — by more than the round trip costs. Win rate alone decides nothing; expectancy does.
How it works
A probability here is a frequency, not a forecast. It describes how often an outcome occurs across many repetitions. It says nothing about the next trade, and that gap is the whole difficulty.
The base rate is the number to beat, and nobody checks it. Before judging a signal, measure what the market does without it. On this site’s shared 576-bar history, 54% of ten-bar windows closed higher, across 566 observations.
An edge is a tilt in the odds, not a forecast. A method with a genuine edge is wrong constantly and still profitable, because the tilt only shows across a long run. Any single result is noise.
Four things that break the intuition
A high win rate with small wins is still a losing system. Expectancy — the average result per trade, combining how often you win with how much — decides whether a method makes money. Winning seven times in ten while losing more on each loser still loses.
Long losing runs are ordinary, not evidence. A genuine edge produces them regularly. Work out what runs yours should produce before you trade it, so you can tell normal from broken.
And twenty trades cannot tell you anything. A hundred is still thin for anything with high variance, which is why backtesting a short sample invites overfitting instead of an answer.
Trades are not independent, which breaks the arithmetic. Consecutive positions in one market share the same conditions, so correlation between them clusters losses and makes the drawdown deeper than a coin-flip model predicts.
In practice
Costs move the break-even rate before you start. The round trip on this series is 0.0098 price units, which is 2% of a median bar’s range and 45% of the smallest bar. Every calculation begins from behind.
A thin market changes the odds you calculated. Low volume widens the spread and worsens the fill, so the same rule pays a different cost depending on when it fires.
A slow system needs years to reach a usable sample. A method trading a few times a month spends years gathering results, and the market it began in is not the one it ends in.
And the rare event is the one that sets the outcome. An opening gap straight past your level is rare and decisive, and why a distribution’s shape beats its average.
The stop loss decides the shape of the distribution. Trailed by a multiple of the 14-bar average true range, the median position here survived 3 bars at one multiple, 10 at two, 22 at three and 32 at four, across 562 trials each.
Every round trip costs 2% of a bar. On 15 of the 576 bars it exceeds 10% of that bar’s range, so it is never a rounding error on the quiet bars that make the rest.
The stop and the target set the rate you need
A stop and a target define the distribution before the market does anything. The stop fixes what a loss costs, the target fixes what a win pays, and the ratio between them sets the rate you need to break even.
Neither side moves for free. Widen the target and the required rate falls, but so does the frequency of reaching it. Tighten the stop and losses get cheaper, while more trigger on noise. Judging the pair together is what risk management means.
Costs sit on top of that arithmetic. The rate the ratio implies comes before the round trip is paid, and here that is 2% of a median bar’s range. Work out the rate your ratio needs, add the cost, then ask whether your measured base rate is anywhere near it.
What probability is not
It is not a prediction about the next trade. Nothing is.
It is not a win rate. A rate without a size is half the figure.
It is not stable. The conditions that produced the number move.
And it is not a substitute for risk per trade. Sizing is what handles the tail.
When it fails
A model built in a trend fails in a trading range. In a range the odds drift and the model does not: its frequencies were measured under conditions that have since changed, and nothing announces it.
The second is reading an ordinary losing run as a broken method. Direction runs here average 2.01 bars with a longest of 11, across 286 runs — and that is a simple two-outcome process. A method’s runs go further, and quitting mid-run turns an ordinary stretch into a permanent loss.
A third is treating a measured frequency as a stable one. The 85% figure for 20-bar breakouts closing back below the level within ten bars rests on 39 events, and the 100% figure for 55-bar breakouts on 11.
A fourth is ignoring the tail. A small chance of an outcome that ends the account is not offset by a healthy average; there is no next repetition to average over.
And a fifth is the mind rather than the maths. Trading psychology is where probability really breaks; Mark Douglas wrote a book on the gap between knowing the odds and acting on them. It is most of why traders lose money.
The original data
On this site’s shared 576-bar history the base rate for a higher close ten bars later is 54% across 566
observations, and 52% over one bar across 571 observations. Those figures are in
research/series-measurements.json, produced by site/measure_series.py.
The same 31,760-video corpus in research/broker-coverage.json puts “probability” in 83 titles at a median
of 3,118 views across 57 channels. “Win rate” appears in 211 titles at a median of 11,527 views across 130
channels.
Three terms appear in none of those 31,760 titles: “expectancy”, “risk of ruin” and “monte carlo”.
The base rate is the whole argument. 54% of ten-bar windows here closed higher, so a signal claiming a 55% hit rate has added roughly one percentage point to doing nothing — and the round trip costs 2% of a median bar’s range, more than that point is worth.
The corpus says the same from the other side. 211 videos discuss win rate and none discuss expectancy or risk of ruin — the exact pair of ideas that turns a win rate into money.
Two things follow, in order. Measure the base rate for your instrument and horizon before you evaluate any signal, and work out the rate your reward-to-risk ratio needs before you take the trade.
Related
Risk per trade is where the tail gets handled, and the next thing to read. Backtesting is how a frequency gets measured without paying for it, and the six errors that flatter the result. And trading psychology is why knowing the odds and acting on them are different skills.
The run that shook me most was not the deepest one; it was the ordinary one I had never bothered to expect. I had not worked out how long a losing streak my own method should produce, so when it arrived I read it as proof the method had stopped working. I changed it, and the next stretch went the other way without me. Now I write down the run length I should expect before the first trade, and I keep the note where I can see it.
— Michael Whitman
This page is educational, not financial advice. Test every idea on your own charts before risking money.