How Many Trades Before Your Statistics Mean Anything

Twelve trades, eight winners, a 67% win rate — and a strong temptation to conclude the strategy works. It might. It also might be a coin that came up heads eight times in a row, which happens more often than it feels like it should. The number itself does not tell you which one you are looking at. The sample size is what tells you how much to trust the number.

Why small samples lie in a specific direction

The problem is not that small samples are wrong on average — it is that they are wrong with confidence. A strategy with a true 50% win rate will, over any given run of 20 trades, regularly print a run of 65% or 35% just from variance. Over 20 trades that swing is normal. Over 500 trades, a strategy that is actually 50-50 will almost never sit at 65% for long — the noise gets diluted by volume. The win rate does not become more true as the sample grows; the noise around it shrinks, and what is left is closer to the real number.

This is why two traders can run the same setup and one calls it a winning strategy after a good week while the other calls it broken after a bad one. Neither has enough trades to know yet. Both are reading variance and calling it signal.

A rough number to work with

There is no single trade count where a strategy suddenly becomes provably good — statistical confidence is a curve, not a switch. But as a working floor: below 30 trades, treat any win rate or expectancy figure as directional at best, not a verdict. Between 30 and 100, the number is worth watching but still wide enough that a handful of trades can swing it several points. Past 100, particularly if the results hold up across different market conditions, the number starts to describe the strategy rather than the last few weeks of it.

The reason 30 shows up so often as a rough floor is that it is where the underlying distribution of outcomes starts behaving predictably enough for the average to stop jumping around on every new data point — not because trade 31 is magically different from trade 29. Treat it as a floor to clear, not a target to stop at.

Reward-to-risk needs a larger sample than win rate does

Win rate is a proportion — it converges reasonably fast because every trade is either a win or a loss, a simple binary. Average win size and average loss size are not binary; they are continuous, and they are much more sensitive to outliers. One trade that runs unusually far before getting stopped, or one loss that gaps past a stop-loss, can move an average reward-to-risk figure more than a dozen ordinary trades combined. Expectancy — which depends on both — inherits that sensitivity, which is why it takes noticeably more trades to trust than win rate alone does.

In practice this means a strategy can have a stable, believable win rate at 40 trades while its expectancy is still bouncing around because one or two outsized trades are doing most of the work. Both numbers deserve a sample check before either gets treated as settled.

Splitting the sample makes the problem worse, not better

The instinct to slice results — this setup on this pair, this setup in this session, this setup after 2pm — is reasonable on its face, but every split divides the sample that was already thin. A strategy with 80 trades total might have 12 of them in any one slice, which puts every subgroup conclusion back below the floor that made the overall number worth trusting in the first place. Segment for ideas worth testing forward, not for conclusions to act on immediately.

What to actually do with a small sample

• Track the trade count next to every statistic, not just the statistic itself — a win rate without an n attached is not really a number.
• Widen the time window before narrowing the trade type — more history on the same setup beats a smaller number of setups.
• Treat early results as a hypothesis, not a track record — a good first 20 trades is a reason to keep going, not a reason to size up.
• Re-check the same statistic after every additional batch of tradesand watch whether it is stabilizing or still swinging — a number still moving significantly past 100 trades is telling you something about the strategy's consistency, not just its average.

None of this replaces having an edge — no amount of sample size turns a losing strategy into a winning one. What it does is stop a few lucky or unlucky trades from being mistaken for proof either way, which is a mistake that costs real money in both directions: sizing up too early on a false positive, or abandoning a sound strategy on a false negative.

This is also why a journal needs enough history before its own numbers are worth acting on. getALPHA's journal keeps the full trade count next to every expectancy and win-rate figure it shows, and process review is built to read the decisions behind the trades rather than lean on a small-sample outcome number — so the read on a strategy does not get decided by whichever dozen trades happened to come first.