How Many Trades Do You Need to Backtest a Strategy?

Enough that your result could not plausibly be luck, which is usually far more than the “100 trades” rule of thumb. To separate a 55% win rate from a 50% coin flip at 95% confidence takes about 385 trades. A much stronger edge needs far fewer: about 43 trades at 65%. You can check any figure yourself with the win rate confidence interval calculator.

The weaker your edge, the more evidence it takes to prove it exists. That is the part traders get backwards. The strategies people are least sure about are exactly the ones they test least.

Why “100 trades” became the standard answer

It sounds reasonable: big enough to feel serious, small enough to finish in a few weekends of replay. It has no statistical basis.

The number of trades you need is not a fixed constant. It depends on the size of the effect you’re trying to detect. Proving a coin lands heads 90% of the time takes a handful of flips. Proving it lands heads 52% of the time takes thousands. Trading strategies sit at the difficult end, because real retail edges are usually small.

Other guides also quote 385, for a different reason. Theirs is the sample that gives a margin of 5 points either side of a 50% win rate. Ours is the sample at which a 55% result stops being compatible with a coin flip. The two land together because clearing 50% from 55% also takes a margin of about 5 points.

The actual numbers

Here’s what a 55% observed win rate means at different sample sizes, using two sided 95% Wilson confidence intervals (z = 1.96). Edwin B. Wilson introduced this interval in 1927. It holds up at small sample sizes, where the familiar Wald interval loses coverage. Every figure here was computed with a script.

Trades	Observed win rate	95% confidence interval	Rules out a coin flip?
20	55%	34% to 74%	No
50	55%	41% to 68%	No
100	55%	45% to 64%	No
200	55%	48% to 62%	No
500	55%	51% to 59%	Yes

Read the 100 trade row again. You have a hundred replayed trades and you win 55% of them. Your evidence says only this: the true win rate sits between 45% and 64%. That range contains comfortably profitable strategies and comfortably losing ones. You have not yet learned whether your strategy works.

At 200 trades the lower bound is 48%, still short of a coin flip. It takes about 385 before the interval clears 50%, which is break even when wins and losses are the same size.

Bigger edges need less evidence

This cuts the other way too, and it’s the useful part. These counts use the same 95% Wilson interval as the table:

• A strategy winning 65% clears a coin flip at about 43 trades.
• At 60%, it takes about 97.
• At 52%, about 2,400.
• A strategy whose edge comes from payoff rather than win rate (40% winners at 3R) needs a different calculation. The thing being measured is no longer a proportion.

So the answer to “how many trades do I need?” is this: enough that the confidence interval around your result no longer includes break even. That is a moving target, and it is the right target.

Win rate is the wrong number to act on

Win rate appears in every sample size example, this one included, because proportions give the cleanest math. It is not the number to act on.

When each loss costs 1R, a 40% win rate at 3R earns 0.6R a trade. A 60% win rate at 0.5R loses 0.1R a trade. Win rate on its own cannot tell them apart, which makes it the most overweighted number in retail trading.

Use expectancy: the average result of one trade, in R or currency. It combines how often you win with how much you win, and that decides whether the account grows. The expectancy calculator works it out from your win rate and reward to risk.

Expectancy usually varies more than a win rate, so it needs more data before you can trust it, not less. The same variance drives how often losing streaks show up at a given win rate.

The mistake that quietly destroys sample size

Everything above assumes you’re evaluating one number over your whole history. The moment you break results down, your effective sample collapses.

Split 60 trades across five trading sessions and you have twelve trades per bucket. Twelve. Seven wins out of twelve is 58%, and its 95% Wilson interval runs from 32% to 81%. The gap between your “best” and “worst” session is almost certainly noise. It will still look convincing on a bar chart.

This is how traders drop good setups over a losing streak in one arbitrary slice of their data. Every filter you apply (by pair, by session, by weekday, by setup) divides the evidence again. Your best session to trade is something you measure, once each bucket holds enough trades.

Practical rules:

• Look at the trade count in each bucket, not just the percentages
• Be suspicious of any bucket below a few dozen trades
• Ask whether the pattern survives removing your two best and two worst trades
• Prefer one large conclusion over five small ones

A challenge pass rate needs the same sample math. A pass rate from a handful of attempts is mostly noise.

Trades also need to be independent

A Wilson interval assumes each trade is a separate draw. Three hundred trades from one trending quarter tell you less than they look. They share one market regime, so they behave like fewer trades. Spread the sample across years and conditions. BacktestFlow’s forex history starts between 2010 and 2015, depending on pair. If live results later drift from the backtest, these are the usual causes.

What this means for how you backtest

Meaningful backtesting takes longer than most people plan for. Several hundred trades is a real commitment, so the process needs to be efficient and completely recorded. Five hundred trades you half remember are worth less than fifty you recorded properly. Follow a manual backtesting method from the first trade.

Two things make that sample achievable:

1. Speed. A replay with four speeds, step controls and keyboard shortcuts covers more chart time in each sitting.
2. A record you cannot skip. If saving a trade is optional, the ugly ones go missing first, and the sample flatters you.

BacktestFlow is manual backtesting software for forex traders who trade by hand, and it is built around both. Every close opens a short write up, and the replay won’t advance until it is saved. The record is complete by construction. On Pro and Max, backtest trades are unlimited. The free plan holds 50, enough to learn the method.

On Pro and Max, more than 40 metrics in five sections build themselves as you go. The free plan shows the Verdict and Edge sections. Pro and Max also run Monte Carlo: 1,000 seeded paths resampled from your own R multiples, once 30 trades have an R.

Flow Coach keeps a finding only if it passes a significance check: a 90% Wilson interval and a group of at least 8 trades. Its report may only quote numbers the engine measured. The first report per strategy unlocks at 35 backtest trades. Its 90% check is built to surface a lead early. The 95% interval in this article is the bar for a final verdict.

The short version

There is no magic number. The sample depends on the size of the edge you are proving. At 55%, about 385 trades; at 65%, about 43; at 52%, about 2,400. Track expectancy rather than win rate. Watch what happens to your sample every time you filter. And stop treating 100 trades as a finish line.

---

Nothing here is financial advice. Backtested performance does not guarantee future results.