Sample Size (N)
The same '70% win rate' can carry completely different weight. Flipping a coin 10 times and getting 7 heads is ordinary — it happens roughly one time in six with a perfectly fair coin. But 700 heads out of 1,000 flips is practically impossible. The first 70% is explained by luck; the second is evidence the coin itself is bent.
Trading statistics work the same way. A 70% win rate from 10 trades is a number that pure coin-flip skill produces all the time. A 55% win rate from 200 trades can mean far more. Before the size of the number, ask how many cases produced it.
With chart signals, rarer signals mean smaller N almost by definition. The more extreme the condition (say, RSI collapsing below 20), the more likely years of data yield only a few dozen cases. Win rates on signals like that swing hard when even a few cases change — read those numbers with extra care.
Small N creates another problem: cherry-picking. Try dozens of condition combinations and, by chance alone, a few will show flattering win rates. Pull out the one good-looking result and it looks like a big discovery — but it's the same as flipping 10 coins over and over and bragging only about the best round.
So wherever you see a win rate or a return figure, ask reflexively: "What's the N?" A statistic that won't or can't answer that question can safely be filtered out.
What the data actually shows
Barobara shows the sample size (N) next to every win rate and expected value, and attaches no conclusion at all to combinations with too few cases. Signals only make it onto the statistics pages if their historical occurrence count clears a threshold. Rare signals like RSI deeply oversold (1d) have few cases, so their numbers wobble more — always check the N printed on the page. The full signal list is in the setup statistics.
Common misconceptions
"10 correct calls in a row proves skill?" If calling direction is a coin flip, 10 in a row is a 1-in-1,024 event. Sounds rare — but let 1,024 people each make calls and on average one of them nails all ten. Social media shows you that one person while the other 1,023 quietly disappear — that's survivorship bias. A short winning streak is weak evidence of skill.
FAQ
Q. How big does N have to be before I can trust it?
There's no clean cutoff, but with samples in the dozens, win rates routinely wobble by ±10 percentage points or more. Barobara chose not to attach conclusions to thin combinations. Whatever the number, the habit that matters is reading it together with its N.
Q. If N is large, can I trust the number unconditionally?
Even a large N can mislead if the data is concentrated in one kind of period (say, a bull market), or if the market's character changes in the future. A large N is a minimum requirement, not a sufficient one — past statistics are, in the end, past distributions.