How many trades to prove an edge?
The most expensive mistake in prediction-market trading is believing a small winning sample. A hot week feels like proof, but the math says otherwise. This guide explains how many trades it really takes to know whether you have an edge, and why the number is usually much larger than people expect.
It comes down to signal versus noise. Your edge is the signal; variance is the noise. And noise shrinks far more slowly than most traders assume.
- Variance shrinks with the square root of your sample size, not linearly.
- Contracts priced near 50 cents carry the most variance, because a coin-flip outcome is the least predictable.
- First, stop over-updating on small samples, both the good and the bad.
The square-root problem
Variance shrinks with the square root of your sample size, not linearly. Doubling your trades only cuts the noise by about 30 percent. So to detect a smaller edge, you need disproportionately more trades, halving the edge you're chasing roughly quadruples the sample you need to prove it.
That's why a 10-point edge might show up clearly in a hundred trades, while a genuine but modest 3-point edge can take well over a thousand to confirm. The edge is real in both cases; one is just far harder to see through the noise.
Price changes the answer
Contracts priced near 50 cents carry the most variance, because a coin-flip outcome is the least predictable. An edge on those markets takes more trades to prove. Lopsided prices, like 15 or 85 cents, carry less variance, so an edge there becomes visible sooner.
This is worth internalizing: the same edge is easier to prove on lopsided markets than on toss-ups, purely because of how variance works.
What to do about it
First, stop over-updating on small samples, both the good and the bad. A great month isn't proof you've cracked it, and a bad month isn't proof you're broken. Keep a clean, complete record and let the sample grow.
Second, know your number. Use the trades-to-prove-edge calculator to see how many trades your suspected edge actually requires, then check your real record against the edge significance calculator once you're in range. Until then, treat your results as provisional.