How many trades does a backtest need? The luck in small samples

A backtest needs enough trades for its average result to stand out from luck, and for most strategies that means somewhere between 100 and 300 trades. The exact number depends on how big the edge is compared with how much a single trade’s result varies.

Twenty trades, a steadily rising equity curve, an expectancy of +0.2R a trade. It looks like a strategy. It’s also a result that a strategy with no edge at all produces about one time in four over twenty trades, and from the report alone you can’t tell the two apart.

This guide explains why the number of trades matters so much, with simulations of how far luck can move a small backtest in either direction. Then it gives a rule for the number of trades you need, explains why some strategies need far more than others, and covers what to do when a strategy trades rarely. It ends with how EdgeLuma shows the size of the sample in every report.

Why does the number of trades matter?

Because the numbers in a backtest report are averages, and an average of a few results swings a lot. Expectancy, the average result per trade in R (R is the amount risked on one trade), is the clearest case. Each trade is a draw: a win of some size or a loss. Over a handful of draws, one lucky winner or one bad streak moves the average far from the strategy’s real edge.

As trades add up, the luck starts to cancel out. That’s the law of large numbers: the average of many independent draws settles close to the true value. How close depends on the standard error, the spread of one trade’s result divided by the square root of the number of trades. Four times the trades halves the luck that’s left.

How much can luck move a small backtest?

More than most traders expect. We simulated 40,000 backtests of each length for a strategy that wins four trades in ten at +2R and loses 1R on the rest, which is worth +0.2R a trade, and for a strategy with no edge at all, which wins four in ten at +1.5R.

Over 20 trades, the strategy with a real edge measured anywhere from about −0.4R to +0.8R a trade in 95 runs out of 100. So a single short backtest can bury a good idea, or crown a useless one, and the two mistakes happen about equally often.

How many trades do you need?

Enough that luck alone would rarely produce your result. A rule of thumb follows from the standard error: the number of trades at which an edge sits about two standard errors away from zero.

Trades needed ≈ (2 × spread of one trade ÷ expectancy)²

The spread is the standard deviation of a single trade’s result in R. Here’s what the rule gives for five common shapes of strategy, all before costs:

Treat the result as a minimum. The rule assumes every trade is independent of the others and that the edge stays the same throughout, and real markets break both assumptions. Bailey and López de Prado work out a more careful version of the question for a track record, which they call the minimum track record length, in The Sharpe Ratio Efficient Frontier. Their version also accounts for skewed results and fat tails, and it asks for more data, not less.

Why do low win rate strategies need more trades?

Because their results vary more. A strategy that wins three trades in ten loses 1R most of the time and makes 3R now and then, so one extra winner or one missing winner moves its average a lot. A strategy that wins seven in ten with small winners produces results that look alike, and its average settles sooner.

That’s why the strategy that wins three in ten needs more than twice as many trades as the one that wins seven in ten in the table above. It also has the longer losing streaks, which our guide to win rate and expectancy measures, and the deeper drawdowns that come with them, simulated in our guide to max drawdown.

Trades or years: which counts?

Trades carry the information, so count them first. A backtest that covers five years and produces 15 trades is still a sample of 15.

Years still matter for a different reason. Trades taken in the same kind of market tend to win and lose together, so 300 trades from one long rally tell you how the strategy did in that rally and not much more. Look for a sample that spans rising, falling and sideways markets, and check the result in each of them separately.

What if your strategy trades rarely?

You have a few honest options:

• Use more history. On a daily chart, an extra year adds only a handful of trades, so start the backtest as early as the data allows.
• Test the same rules on more markets. Run the unchanged rules on several pairs and look at each result and at all of them together.
• Treat a faster version as a new strategy. The same idea on a lower timeframe produces more trades, but it’s a different System with different costs, and it needs its own test.

Avoid loosening the rules just to get more trades. That changes the strategy you’re testing and adds another round of tuning, which is how backtests get overfit. Our guide to backtest overfitting covers the checks.

How EdgeLuma shows the size of the sample

EdgeLuma puts the number of closed trades at the top of every backtest report, right under the total result and next to the expectancy, so you see how much evidence sits behind each number before you read it. Results are counted in R after fees and slippage, the same unit as the rule above.

You can run the same System on other markets and other periods to grow the sample without changing the rules. To check a result, attach the run in the Chat and ask the EdgeLuma AI assistant, for example, whether its number of trades is enough to trust the expectancy. It answers from the run’s own trades. A backtest replays the past: it doesn’t tell you what to trade, and nothing in a report is financial advice.