Your Win Rate Is Probably Noise
Thirty trades in, the dashboard says 60%. It feels like proof. Run the arithmetic and a strategy with no edge at all — a literal coin — produces that same 60% about one time in four. Here is how many trades it actually takes before the number means anything, and why win rate was the wrong number to be watching.
The short version
- At 20 trades, a coinflip shows a 60% win rate 25% of the time. Your sample is not evidence yet.
- An observed 60% over 50 trades means the true rate is somewhere between 46% and 72%. That range still contains a losing strategy.
- A realistic edge — a true 55% — takes around 370 trades to separate from luck.
- None of which matters much, because 30% at 3R beats 70% at 0.5R by four to one. Win rate was never the number.
1A coin that looks like a strategy
Flip a coin twenty times and call heads a win. No edge, no skill, nothing but chance. Do that over and over and watch what the "win rate" does.
One trader in four, running a strategy worth exactly nothing, finishes twenty trades looking like they found something. They did not do anything wrong. They ran a small sample, and small samples do this.
| Sample | Shows ≥60% | Shows ≥65% | Shows ≥70% |
|---|---|---|---|
| 10 trades | 37.7% | 17.2% | 17.2% |
| 20 trades | 25.2% | 13.2% | 5.8% |
| 30 trades | 18.1% | 4.9% | 2.1% |
| 50 trades | 10.1% | 1.6% | 0.3% |
| 100 trades | 2.8% | 0.2% | <0.1% |
| 200 trades | 0.3% | <0.1% | <0.1% |
Binomial probability, p = 0.5. The 10‑trade row repeats because with ten flips you cannot land between 6.5 and 7 wins — the discreteness of small samples showing up in the arithmetic itself.
A 60% win rate over twenty trades is not a finding. It is the most ordinary thing a coin does.
2What your 60% actually means
Turn it around. You have taken the trades and the dashboard says 60%. What is your real win rate — the one you would converge on after a thousand more?
Statistics answers that with an interval rather than a number, and the interval is wider than almost anyone expects.
| Trades taken | True win rate is somewhere in | Verdict |
|---|---|---|
| 10 | 31.3% – 83.2% | Could be a losing strategy |
| 20 | 38.7% – 78.1% | Could be a losing strategy |
| 30 | 42.3% – 75.4% | Could be a losing strategy |
| 50 | 46.2% – 72.4% | Could be a losing strategy |
| 100 | 50.2% – 69.1% | Finally clear of a coinflip |
| 200 | 53.1% – 66.5% | A real edge, size still fuzzy |
| 500 | 55.6% – 64.2% | Now you know roughly what you have |
Read the 50‑trade row again. Fifty trades is a serious amount of work — weeks of screen time, real money at risk, a spreadsheet you actually kept. And the honest reading is still "somewhere between losing slowly and doing very well."
This is where accounts die. A trader sees 60% over fifty trades, concludes the edge is proven, and doubles size. If the true rate was 47% — entirely consistent with what they observed — the larger size does not amplify an edge. It amplifies a leak, and the drawdown that follows arrives with twice the weight.
Run it on your own numbers
Your last stretch of trades, and how many of them won. Nothing is sent anywhere — this runs in your browser.
Wilson score interval, 95% confidence — the method statisticians use for exactly this, because the naive one misbehaves badly at small samples and near the edges.
3So how many trades do you actually need?
It depends entirely on how big your edge is, and the relationship is brutal: the smaller the edge, the more evidence it takes to see it at all.
Most traders assume they are in the first column. Almost everyone who is profitable at all is in the second. And a 52% edge — genuinely positive, genuinely worth having — cannot be distinguished from noise inside a normal trading lifetime without thousands of trades.
If you cannot prove your edge in fifty trades, that does not mean you do not have one. It means fifty trades was never going to be the test.
Which raises the obvious question. If the number that everyone watches needs hundreds of trades before it says anything, what are you supposed to look at in the meantime?
4The turn: win rate was the wrong number
Here is the part that reframes everything above. Win rate tells you how often you are right. It says nothing about how much you make when you are right, or lose when you are not — and that second half is where the money lives.
| Win rate | Average win | Expectancy per trade | |
|---|---|---|---|
| 70% | 0.5R | +0.05R | Wins constantly. Barely profitable. |
| 50% | 1.0R | 0.00R | A coin with commissions. Slowly bleeds. |
| 40% | 2.0R | +0.20R | Wrong most of the time. Four times better. |
| 30% | 3.0R | +0.20R | Wrong seven times in ten. Same result. |
Expectancy = (win rate × average win in R) − (loss rate × 1R), assuming losses run to the full stop.
The 70% trader is the one who feels best. They are right about everything, all the time, and their dashboard is a wall of green. They are also making a quarter of what the trader who is wrong seven times out of ten makes.
And expectancy converges faster than it looks. A trader with 40% wins at 2R is not waiting on the win rate to stabilise — they are watching the average size of the winner against the average size of the loser, which is a ratio, and ratios of magnitudes settle down sooner than proportions of counts. It is the more honest number and the quicker one.
So the answer to "how many trades do I need" is not a single number. It is: fewer than you think for expectancy, far more than you think for win rate, and you should have been watching expectancy the whole time.
5What to do with a sample that is still too small
You cannot make forty trades into four hundred. What you can do is stop treating forty as though it were four hundred. Four habits, in the order they matter.
Put the sample size next to every number
"58% win rate" and "58% win rate over 31 trades" are different claims, and only one of them is honest. A figure without its n is not a statistic, it is a rumour. If your journal shows the first, write the second beside it by hand until it shows the second.
Read expectancy in R, not P&L in currency
Currency mixes your edge with your position sizing, so a good month on big size and a bad month on small size look identical. R strips the sizing out and leaves the thing you are actually trying to measure.
Split by setup before you split by anything else
Sixty trades across four setups is not a 60‑trade sample. It is four 15‑trade samples wearing a trenchcoat, and each of them is pure noise. Tag the setup at entry so the split is possible later — you cannot reconstruct it from memory.
Grade the decision, not the outcome
While the statistics are still too thin to say anything, the one thing you can measure honestly is whether you followed your own plan. Rule adherence needs no sample size at all: either the criteria were met when you entered, or they were not.
A small sample is not useless. It is just answering a different question — not "does my edge work" but "am I trading the way I said I would."
6Why most journals make this worse
Open almost any trading journal and the first screen is a wall of large, confident percentages. Win rate. Profit factor. Average win. Every one of them rendered at the same size and in the same tone whether it was computed from eight trades or eight hundred.
That is not a small presentation flaw. It is the entire failure mode described above, built into the product — the tool that was supposed to keep you honest is the thing telling you that eight trades proved something.
What we did instead. TradeAssay states the sample size beside every figure, and says "not enough data yet" rather than printing a number it cannot stand behind. Findings get re‑tested on trades they were not discovered on. Everything is measured in R by default, so a big win and a small win are never treated as the same result — and you can switch the whole product between R, dollars and percent when you want the other view.
None of that is clever. It is the ordinary discipline any statistician would apply, applied to a trading journal — which is unusual mostly because journals are built to feel good rather than to be right.
See what your own numbers look like with their sample sizes attached
Import your history from MT4, MT5 or a CSV, or hand a screenshot to the agent. The free plan does not expire and does not ask for a card.
Frequently asked questions
How many trades before my trading statistics are reliable?
It depends on the size of the edge, not on a round number. Proving a true 60% win rate takes about 100 trades; a true 55% takes roughly 370; a true 52% needs thousands. The commonly quoted "100 trades" is the point at which an observed 60% finally stops being consistent with a coinflip — it is a floor for a strong edge, not a general answer. Expectancy in R stabilises sooner than win rate, which is one reason to watch it instead.
Is a 60% win rate good?
On its own it says almost nothing. A 70% win rate at 0.5R average win returns +0.05R per trade; a 30% win rate at 3R returns +0.20R — four times more, while being wrong seven times out of ten. Win rate only becomes meaningful once you pair it with the average size of the win against the average size of the loss.
Why is my win rate so different from month to month?
Because a month is a small sample. With no change in your trading at all, twenty trades from a genuine 55% strategy will regularly print anything from 35% to 75%. Month-to-month swings of that size are the expected behaviour of the arithmetic, not evidence that something changed.
What is expectancy in trading?
The average result of one trade, expressed in R — multiples of the risk you took at entry. Calculated as (win rate × average win in R) minus (loss rate × average loss in R). Positive expectancy means the method makes money over enough repetitions; anything at or below zero means no amount of discipline or position sizing will save it.
Does splitting trades by setup make the sample-size problem worse?
Yes, and it is the most common way traders fool themselves twice over. Sixty trades split across four setups gives four samples of fifteen, and a fifteen-trade sample is pure noise — so the "best setup" you find that way is usually just the one that got lucky. Split by setup for the qualitative reading, and wait for real depth before trusting the ranking.
Read next
- Moving your stop to breakeven breaks your R‑multiple — the other way journals produce numbers that are technically correct and completely wrong.
- Why traders fail prop firm challenges — what a normal losing streak does to a daily loss limit, in arithmetic.
- How TradeAssay reports performance — in percent, dollars or R, with the sample size attached to every figure.
Trading foreign exchange, futures and derivatives carries substantial risk and is not suitable for every investor. Only risk capital should be used. TradeAssay is a journaling and analytics tool: it does not provide investment advice, does not place orders, and past performance recorded in it is not a guarantee of future results. The figures and calculator on this page are illustrative arithmetic, not a prediction of results.