A trader finishes their first month, opens their stats, and sees a 62% win rate across 34 trades. The natural conclusion is that the strategy works. The honest conclusion is that 34 trades cannot distinguish a 62% system from a 45% system having a good month.

This matters because the wrong conclusion is expensive in both directions. Traders scale up strategies that were never profitable, and abandon workable ones after three bad weeks. Both mistakes come from asking a small sample a question it cannot answer.

The width of the error bar

Win rate is a proportion, and proportions estimated from small samples carry wide margins. The rough 95% margin around an observed win rate near 50% works out as:

Read the first line again. At 30 trades, an observed 62% is consistent with a true rate anywhere from roughly 44% to 80%. The bottom of that range is a losing system at most risk-reward ratios. Your data cannot rule it out.

At 100 trades the picture sharpens but stays blunt: 62% observed means somewhere around 52% to 72%. Only past a few hundred trades does win rate become a number you can plan around. For someone taking five trades a week, 400 trades is more than eighteen months away.

This is not a reason to ignore your statistics. It is a reason to know which ones have arrived.

It also explains something that confuses newer traders: why two people running the same strategy report opposite results after a month. Neither is lying and neither has found anything. They are both reading noise, from opposite ends of the same distribution.

Some numbers converge much faster

Here is the part that gets missed. The slow convergence applies to outcomes - win rate, expectancy, profit factor - because outcomes are produced by the market, and the market adds noise to everything you do.

Your own conduct carries no such noise. How often you opened a position within ten minutes of a loss is a fact about your behaviour, not a sample from a distribution. How often you traded without a stop is a count, not an estimate. These stabilise in weeks rather than years, because you are measuring the frequency of your own actions.

Practical consequence: judge your process on counts, and your edge on statistics. In month one you cannot know whether your strategy is profitable. You can know, with certainty, that eleven of your thirty-four trades came within ten minutes of a losing trade. Nothing about that requires a bigger sample. It is already true.

That distinction tells you what to work on first. Behaviour is measurable, fixable, and visible immediately. Edge takes a year to establish and cannot be assessed while the behaviour underneath it keeps changing.

What to do at each stage

Under 50 trades. Do not evaluate the strategy at all. Track behavioural counts: trades taken after a loss, trades without a stop, sessions where you exceeded your normal trade count. Fix the largest one. The goal for this period is a stable process, so that later data describes one method rather than five.

50 to 200 trades. Outcome numbers begin carrying signal, but treat them as direction rather than measurement. Look at average R-multiple alongside win rate, since the two together are harder to fool than either alone. A 40% win rate at 2.5R and a 65% win rate at 0.7R are similar businesses.

Past 200 trades. Now the comparisons start working: this setup against that one, this hour against that one. Segment carefully - splitting 200 trades into six buckets returns you to 33 per bucket and the ±18 point problem you started with.

Why more trades doesn’t automatically mean more information

An obvious response to all this is to trade more and reach the sample faster. It backfires for two reasons.

The first is that a sample only counts if it describes one method. Take 300 trades while adjusting your approach every fortnight and you don’t have 300 observations of anything. You have fifteen samples of twenty, each too small to read. Consistency of process is what converts trade count into evidence, which is why the under-50 stage matters more than it feels like it should.

The second is that trading more to gather data changes what you’re trading. Forcing volume means taking setups you’d normally pass, so the extra trades are drawn from a worse population than the ones you were trying to measure. The sample grows and gets less representative at the same time.

The honest position is that a real edge takes a long time to confirm and there’s no shortcut. What you can do meanwhile is make sure the trades you do take all belong to the same experiment.

Beware of splitting the sample after the fact

One more trap, and it catches careful people specifically. After a flat month, it’s natural to go looking: which setups worked, which hours, which coins, which days of the week.

Slice 150 trades enough ways and something will look excellent. With a dozen possible splits, a standout result is close to guaranteed by chance alone - that’s what randomness does when you search it hard enough. Traders then build rules around the winning slice and watch it revert immediately, which it was always going to do.

The defence is to decide what you’re testing before you look. Write down the one split you care about this quarter, check that, and let the rest wait. A hypothesis chosen in advance is evidence; a pattern found by scanning is a hypothesis for next quarter, and it should be treated as nothing more.

The rule that saves the most money

Never change strategy on a losing streak within the noise band. At a genuine 55% win rate, a run of five consecutive losses is unremarkable - it happens roughly once every 55 sequences of five. Traders live through that run, conclude the edge is gone, and rebuild from scratch. Then they do it again with the next method, and the sample never gets large enough to tell them anything.

The alternative is to decide in advance what evidence would count. Write it down: I will reassess this approach after 100 trades, or if drawdown passes X%, whichever comes first. Setting the threshold before the emotion arrives is what makes it usable. Reviewing on a schedule beats reviewing on a feeling, because a schedule does not move when you are losing.

Try this tonight

Open Trady and go to the Insight tab. Find your total trade count first, before you look at any percentage. If it’s under 100, write that number at the top of your notes and treat every performance figure below it as provisional.

Then look at the behavioural counts instead: trades opened shortly after a loss, sessions over your usual trade count, entries with no stop in place. Pick the single largest one. That number is real today, and it is the only one you can act on this week.

Trading carries risk. This article is about analytics, not financial advice.