How Many Trades Do You Need to Validate a Trading Strategy?

Twenty profitable trades feels like proof. Statistically, it is closer to a rumour.

The problem is variance. A strategy with genuinely zero edge, a coin flip after costs, will still produce runs of 15 winners in 20 trades often enough that thousands of traders are living inside one right now and calling it validation. The smaller the sample, the more the result reflects luck rather than the strategy, and small samples are precisely where confidence feels strongest, because nothing bad has happened yet.

The honest thresholds

There is no magic number, but there are defensible bands.

Below roughly 5 trades, a result contains no usable information about the future. Any scoring or judgement built on it should treat it as zero evidence, not weak evidence.

Around 30 trades, a consistent result starts to be distinguishable from noise for a strategy with a meaningful edge. This is a minimum for a first verdict, not a finish line. A small edge, the kind most real strategies actually have, needs hundreds of trades to separate from zero with any confidence.

Per condition is where most traders go wrong. If you slice your trades by session, volatility and setup, each slice needs its own sample. A 200-trade book split eight ways is eight 25-trade verdicts, and every one of them is provisional.

The trap that makes small samples dangerous

Slicing multiplies the problem. Cut any losing book enough ways and some slice will look brilliant, because with enough small samples, luck produces perfection somewhere. This is the multiple-comparisons problem, and it is why a tool that rewards filtered results without sample floors is a machine for manufacturing false edges.

Our own rule, stated publicly: conditions with fewer than 5 trades score zero, and full weight requires 30. You cannot filter your way to a good score, because the scoring will not let you.

What validation actually requires

Sample size is necessary but not sufficient. The order of operations that holds up:

First, verify the arithmetic. Reconcile your total P&L against an independent calculation of exit minus entry times size, outside whatever produced the results. Every downstream test inherits an upstream accounting bug.

Second, reach the sample. Trade or backtest until each condition you intend to trust has crossed its floor.

Third, test forward. Freeze your conclusions on the data you have, then check them against trades that did not exist when you drew them. A verdict that only describes the past is a description, not a validation. We published a full worked example of this method, including the conditions that failed it, in our case study.

The uncomfortable summary: most traders have never validated anything. They have observed a streak, sliced until something shone, and named it an edge. The sample-size discipline is boring, and it is the whole difference.