Sample Size and Variance: Why 10 Trades Mean Nothing

The Trap of the Small Sample

Here's a scenario that destroys accounts every year: a trader takes 8 trades with a new strategy, wins 6 of them, and concludes they've found their edge. They size up. Then they hit a stretch of 7 losses in a row and blow up an account they'd have kept if they'd sized appropriately from the start.

The strategy might have been fine. The problem was drawing a conclusion from a sample too small to draw any conclusion from at all.

Coin Flips Explain This Better Than Charts Do

Flip a fair coin 10 times. It is genuinely common to get 7 heads and 3 tails, or worse. Nobody would conclude from that the coin is biased — we know the true probability is 50/50, and 10 flips just isn't enough to reveal it reliably. Random variance dominates small samples.

Your trading strategy is the same. Even a strategy with a real, durable 55% win rate will regularly produce stretches of 10 trades with only 3 or 4 wins, purely from variance — no different in kind from a fair coin running cold. If you judge the strategy (or yourself) off that stretch, you're reacting to noise, not signal.

How Much Data Do You Actually Need?

There's no single magic number, but useful rules of thumb:

• 30 occurrences is the rough statistical floor where a win rate starts to mean something more than noise — this is the same threshold used broadly in statistics for a sample to start approximating a normal distribution.
• 50-100 occurrences gives you a much tighter read on your true win rate and average R, and starts to reveal your worst realistic losing streak.
• 100+ occurrences, ideally spanning different market conditions (trending, choppy, high and low volatility), is what you want before trusting a strategy with meaningful size.

If your setup only triggers once or twice a week, reaching 100 occurrences legitimately takes the better part of a year. That's not a flaw in the process — that's the actual timeline for building real evidence. Trying to compress it by cherry-picking backtested trades or overtrading a rare setup just reintroduces bias.

Expect Losing Streaks — Calculate Them, Don't Guess

Given a strategy's real win rate, you can estimate the probability of a losing streak of a given length. A strategy with a 50% win rate will, over enough trades, produce a losing streak of 5+ in a row with real regularity — that's not the strategy failing, that's the strategy behaving exactly as its statistics predict.

This is why risk-of-ruin planning (covered in the risk module) has to be based on your worst realistic losing streak, not your best week. If you don't know your strategy's real numbers because your sample is too small, you can't plan for this at all — you're just hoping.

Variance Cuts Both Ways

The same logic that says "don't panic after 4 losses in 10 trades" also says "don't get overconfident after 8 wins in 10 trades." A hot streak from a genuinely mediocre strategy looks identical, in a small sample, to a real edge. The only way to tell them apart is more data.

This is uncomfortable because it means you can't get quick certainty. Nobody wants to hear "you need 100 more trades before you know if this works." But wanting a faster answer doesn't make small-sample noise into real information.

Worked Example

Two traders each run a strategy for 10 trades. Trader A wins 7. Trader B wins 4. Both conclude something about their strategy after this single sample. But if the true win rate of the strategy is 55%, then both 7/10 and 4/10 are well within the range of outcomes you'd expect from ordinary variance — neither result actually tells you much on its own.

Now extend both traders to 100 trades. Trader A's win rate settles to 54%. Trader B's settles to 53%. Same strategy, same real edge — the small-sample difference was just noise, and it evaporates as the sample grows.

The Practical Takeaway

Treat any conclusion drawn from fewer than 30 trades as provisional at best. Don't resize your risk, don't abandon a strategy, and don't declare yourself "figured out" off a small sample. Keep logging, keep executing the rules consistently, and let the sample grow before you trust the numbers.