# Sample Size

> Sample size in trading is how many trades sit behind a statistic. Learn why small samples mislead, how error shrinks with more trades, and how many you need.

Canonical: https://www.journalx.io/glossary/sample-size

**Sample size** is the number of trades behind a number. A 60% [win rate](/glossary/win-rate) from 15 trades and a 60% win rate from 400 trades look identical on a dashboard, but they are very different pieces of evidence. The first could plausibly come from a strategy that really wins 40% of the time. The second almost certainly could not. Sample size is how you tell which statistics are ready to act on and which are still mostly noise.

## How Sample Size Works

Any trading statistic is an estimate. Your strategy has some true win rate and true [expectancy](/glossary/expectancy) that you never see directly. What you see is the result of the trades you happened to take, and a different run of the same trades would have given a different number.

The spread of those possible results is measured by the [standard error](https://en.wikipedia.org/wiki/Standard_error):

> Standard Error = Standard Deviation ÷ √(Number of Trades)

For a win rate, the standard deviation of a single trade is √(p × (1 − p)), where p is the win rate. About 95% of the time, the true value sits within 1.96 standard errors of what you measured. That is the 95% margin of error.

![The 95% margin of error on a measured 50% win rate falls from ±19.6 points at 25 trades to ±9.8 points at 100 trades and ±4.9 points at 400 trades, so four times the trades halves the margin](https://assets.journalx.io/marketing/glossary/sample-size/win-rate-margin-by-trades.avif)

The square root is the part that catches people out. Going from 25 trades to 100 cuts the margin on a 50% win rate from ±19.6 points to ±9.8. Cutting it in half again takes 400 trades. Early trades shrink the uncertainty quickly, and every later halving costs four times as many trades as the last.

## Sample Size for Win Rate vs Expectancy

Win rate only depends on whether each trade won or lost. Expectancy also depends on how large each win and loss was, so it carries more noise and needs a bigger sample to pin down.

A common way to estimate the sample you need, with both numbers measured in [R-multiples](/glossary/r-multiple), is:

> Trades Needed ≈ (2 × Standard Deviation ÷ Expectancy)²

This gives roughly the trade count at which a positive expectancy sits two standard errors above zero. Some examples, each with 1R losses:

| Setup               | Win rate | Average win | Expectancy | Std. deviation | Trades needed |
| ------------------- | -------- | ----------- | ---------- | -------------- | ------------- |
| Frequent small wins | 60%      | 1R          | +0.20R     | 0.98R          | \~100         |
| Balanced            | 50%      | 1.5R        | +0.25R     | 1.25R          | \~100         |
| Thin edge           | 40%      | 2R          | +0.20R     | 1.47R          | \~215         |
| Very thin edge      | 55%      | 1R          | +0.10R     | 0.99R          | \~400         |

Two things drive the number. A smaller edge needs more trades, and halving the expectancy quadruples the sample. A larger spread of results needs more trades too, which is why setups with a high [payoff ratio](/glossary/payoff-ratio) and a low win rate usually take longer to confirm than steadier ones at the same expectancy. Our post on [how many trades you need to judge a strategy](/blog/how-many-trades-to-evaluate-a-strategy) works through these numbers in detail.

The same formula is behind Van Tharp's [System Quality Number](/glossary/system-quality-number), which scales expectancy by its standard deviation and the square root of the trade count. Without Tharp's 100-trade cap, an SQN of 2 is exactly the point the formula above targets.

## Where Your Sample Shrinks Without You Noticing

A journal with 500 trades can still be judging a setup on 30.

**Filtering.** Every slice divides the sample. Three hundred trades across four setups is 75 each. Split those by morning and afternoon and you are near 37. Split again by long and short and you are under 20, with a margin of error to match.

**Changing the rules.** Moving the stop, adding a filter or changing the target after 50 trades starts a new strategy. The old trades no longer describe what you are trading now.

**Testing many versions.** Try 20 variations of a setup and keep the best, and the winner looks good partly by chance. If none of the 20 has a real edge, the chance that at least one still passes a 95% test is about 64%. Statisticians call this the [multiple comparisons problem](https://en.wikipedia.org/wiki/Multiple_comparisons_problem).

**Trades that move together.** The formulas assume each trade is independent. Twenty trades taken in one week of one market, or several positions opened on the same news, carry less information than the count suggests.

## Key Takeaways

- Sample size is the number of trades behind a statistic, and it decides how far that statistic can be trusted.
- The margin of error shrinks with the square root of the trade count. Halving it takes four times the trades.
- Expectancy needs a bigger sample than win rate, and thin edges or high-variance setups need the most. A few hundred trades is common.
- Filters, rule changes and testing many variations all shrink the real sample behind a number.
- Decide the sample size before you start counting, and judge execution rather than results until you reach it.

## Common Mistakes

**Judging a setup after a few weeks.** Twenty trades can make a real edge look like a losing strategy, or a losing one look like a winner. Noise works in both directions with equal ease.

**Treating 30 as a magic number.** Thirty is a textbook rule of thumb for when an average starts to behave like a normal distribution. It is not a test of whether a trading edge exists, and trade results are often skewed enough to need more.

**Reading every slice as if it had the full sample.** A dashboard that shows expectancy by setup, by session and by weekday is showing much smaller samples than the headline. Check the trade count next to each figure before acting on it.

**Restarting the clock without noticing.** Each rule change resets the sample. A strategy that has been tweaked every month may never have had a fair test.

**Trusting a small sample because it agrees with you.** Twenty winning trades are no more conclusive than twenty losing ones. Apply the same standard to good news as to bad.

## How JournalX Tracks Sample Size

JournalX shows the trade count next to every statistic, and the count updates as you stack filters by setup, session, symbol or tag. That makes it easy to see when a striking expectancy or win rate rests on a dozen trades rather than a few hundred. Tag trades with the version of the rules they were taken under, and a rule change starts a fresh sample instead of blending into the old one. In the weekly or monthly [trade review](/glossary/trade-review), that split lets you check execution on every trade while saving strategy decisions for when the sample is large enough.