2 / 5 · 7 min
How many trades before you can trust the result
Over twenty trades a random system shows anywhere between 28% and 72% wins. Proving an edge of five percentage points takes a thousand trades — and that changes which ideas are worth looking for at all.
Why a small sample lies in both directions
Toss a coin twenty times and you get not ten heads but somewhere between six and fourteen, and that is normal. Trading is the same: a system with no edge whatsoever produces any result over a short run. The danger is not that you will see a bad result from a good system — everyone accepts that. The danger is the reverse: you will see a brilliant result from a useless rule, believe it, and start increasing size.
What a system with NO edge shows (true 50%)
| Trades | 95% of results land between |
|---|---|
| 20 | 28% and 72% |
| 50 | 36% and 64% |
| 100 | 40% and 60% |
| 400 | 45% and 55% |
The first row explains most stories about a discovered holy grail. Twenty trades is a week of day trading, and in that week a system with no edge comfortably delivers 70% wins. The person draws a conclusion, increases size, and then the sample grows and the result slides back towards the truth. Note the last row too: even at four hundred trades the spread remains ±5 percentage points.
How many trades to PROVE an edge
The risk course says fifty trades are enough for a rough estimate of expectancy, and that is true: the question there is whether there is a plus at all and of what order. But proving a specific edge is a different and far more expensive task. Computed at 95% confidence: telling 60% wins from 50% takes 258 trades. Telling 55% from 50% takes 1046. Telling 53% from 50% takes 2912.
The price of proof
| Edge | Trades needed | For a swing trader that is |
|---|---|---|
| 60% versus 50% | 258 | about 2 years |
| 55% versus 50% | 1046 | about 7 years |
| 53% versus 50% | 2912 | about 19 years |
The main conclusion: a small edge cannot be proven
The table implies something that changes the whole approach to finding ideas. An edge of three percentage points may exist, but you will not manage to verify it within a trading lifetime — which means decisions cannot be built on it even if it is real. Hence the practical rule: look not for a tiny surplus in win rate but either for a noticeable effect visible over hundreds of trades or for an edge in the SIZE of the win. The second is verified faster: the difference between an average win of 2R and 1R shows up within fifty trades, while the difference between 53% and 50% is invisible even at a thousand.
Treating a winning streak as significant
Ten winning trades in a row are impressive, but at a 50% win rate such a streak occurs roughly once in a thousand attempts — and the market offers many attempts. Measured over a hundred trades: a streak of five losses appears in 95% of cases even for a system with no edge, a streak of seven in half of cases. The conclusion cuts both ways: a winning streak does not prove the system works, and a losing streak does not prove it broke. Only a sample of the required size proves anything, and that size is known in advance from the calculation.
How to split history honestly
A sample can be not only small but dirty — if you test on the same stretch where the idea was found. The technique is simple and it works: split the available history into two unequal parts. On the first, say two thirds, you search and tune. The remaining third you DO NOT OPEN at all until the rule is finally formulated. Then you run it once and accept the result as it comes. If after a failure you went back to tuning and tested on that same third again, it has stopped being clean: you are now fitting parameters to it, and you need a fresh untouched piece.
Paper trading in the service keeps a journal: every trade with entry price, exit and result. It is the cheapest way to build a sample without risking money, and the only one where you cannot retroactively remember a trade differently from how it was. Fifty trades in the journal give a rough estimate; several hundred are already grounds for conclusions.
About paper tradingCompute when you will get your answer
Take your rule and estimate what edge you expect from it — in percentage points of wins or in average R. Find the required number of trades in the table. Divide by your trades per week from the styles course. The resulting period is when you will learn the truth. If it comes out in years, you have two ways out: look for a larger effect, or move the testing into bar replay where a week of market passes in an hour. There is no third way — believe it and increase size is not one.
A system gave 70% wins over 20 trades. Can I increase size?
No, because that result fits comfortably inside the ordinary spread of a system with NO edge: over twenty trades, 95% of outcomes lie between 28% and 72%. Seventy percent here is evidence of a small sample, not of quality. What to do: continue at the same size and accumulate trades until there are at least a hundred. Increasing size based on a short run is the most common way of turning random luck into a regular loss.
Why is an edge in win size verified faster than one in win rate?
Because win rate is a proportion, and its spread on a small sample is enormous: telling 55% from 50% needs a thousand observations. The average result in R is measured in different units, and the difference between 2R and 1R on the average win shows up within tens of trades — it is simply larger than the noise. The practical implication for idea hunting: a rule of taking few trades but with a target three times the stop can be verified in months, while a rule of guessing slightly more often than half can never be verified at all.