How Many Trades Do You Need? Sample Size and Forward Testing

Reviewed by Leon Moyer - Published October 9, 2026 - Last reviewed October 9, 2026

Sample size is the number of trades a result is based on, and it decides how much that result can be trusted. A journal's expectancy, profit factor and win rate are estimates of how a way of trading performs, and with few trades those estimates can be far from the truth in either direction.

This is the last step of the module. How to Find Your Trading Edge set up what to record, trading expectancy and profit factor measured groups of trades, and analyzing performance by strategy and setup split the journal. This lesson asks how big a group must be before you act on it.

Why can a small sample mislead you?

Every trade has some luck in it. Over a handful of trades, luck can dominate: a mediocre strategy can string together wins, and a good one can open with a losing streak. As the count grows, luck tends to average out, and the result settles closer to what the strategy really does.

Win rate gives the clearest picture of this, because statistics has a standard rough formula for it. For a win rate p measured over n independent trades, an approximate 95% range is p plus or minus 1.96 x the square root of (p x (1 - p) / n).

The real uncertainty is larger than this table, for two reasons. The average win and average loss are estimates too, and trades are not always independent: several positions opened in the same week, on related stocks, can win or lose together. Treat the table as the most optimistic case.

The practical lesson: the closer a strategy is to break-even, the more trades it takes to know which side it is on. A strategy with a large edge shows itself sooner. Most real edges are small.

What does that mean for the small groups in your journal?

Splitting a journal makes this worse, because every split shrinks the groups. In How to Analyze Your Trading Performance, the five put spreads held through earnings won 2 of 5. The same formula puts a range of about -3% to 83% on that 40% (a negative bound means the approximation has broken down at this size), which is to say almost anything.

That does not make the earnings finding useless. It makes it a hypothesis with a sensible reason behind it, which is exactly what should be tested next.

What is overfitting, and how does it fool traders?

Overfitting means building rules that fit the trades you studied so closely that they describe chance, not a real pattern. Slice the same 40 trades enough ways, by weekday, ticker or delta, and some group will look good by luck alone; a rule such as "only trade on Thursdays" built from that group is fitted to noise. Course 4's Backtesting Options Strategies lesson covers overfitting in depth. For a journal, the defense is a forward test.

How do you test a rule without fooling yourself?

The fix is to judge a rule only on trades it was not built from. For a trader, the simplest way is forward testing.

1. Spot a pattern in past trades, ideally one with a reason behind it.
2. Write it down before testing: the exact rule, the measure you will judge it by (such as expectancy in R), and how many new trades you will wait for.
3. Trade it forward and record the new trades with the same fields as before.
4. Keep or drop the rule based on the new trades only, not the ones that suggested it.

Change one rule at a time, so you can tell which change made the difference. And keep position sizes small while a rule is being tested; a rule you are still checking has not earned extra capital.