How Many Trades Do You Need to Detect an Edge?

[Fairman]

How Many Trades Do You Need to Detect an Edge?

A trader believes their strategy has an expectancy of 0.2R per trade. The standard deviation of individual trade results is 1.5R. How many trades are needed before the average would be about two standard errors above zero?

The standard error of the mean is the standard deviation divided by the square root of n. Set the average divided by the standard error equal to 2: 0.2 divided by (1.5 divided by the square root of n) equals 2.

Solve: square root of n equals 2 times 1.5 divided by 0.2, which is 15. So n is 225 trades.

At 6 trades a month, that is 37.5 months, over three years. At 20 trades a month, 11 months.

Smaller edges need more. An edge of 0.1R would need 900 trades. A larger edge of 0.4R needs only 56 trades.

What does it mean for planning? A swing trader with a modest edge cannot verify it quickly with live trading alone. They need backtests, multiple pairs, or patience.

It also means that early results are mostly noise. After 30 trades, the standard error is 0.27R, larger than the edge itself, so an average of plus 0.5R or minus 0.1R would be equally unremarkable.

The calculation assumes independent trades, and results with fat tails need larger samples.

Use it to set expectations before starting, and to decide what you will and will not conclude at each stage.

Try this: compute the sample size for your own edge and trade frequency.

[FXS HongKong Scalper]

Solve: square root of n equals 2 times 1.5 divided by 0.2, which is 15.

With 0.2R expectancy and 1.5R standard deviation, two standard errors above zero needs n such that 0.2 > 2 × 1.5 / √n, so √n > 15 and n > 225 trades. That is a useful number for anyone who judges a strategy after thirty trades.

[Amanhasnoname]

Running this exact calculation on my own numbers was humbling. My scalping system showed an average of roughly a quarter of an R per trade, but the spread of outcomes was wide because I let some winners run, so the standard deviation came out near 1.8R. Plugging that in told me I needed around two hundred trades before I could say anything, and at the time I had forty and was already convinced it worked.

What I ended up doing was splitting the problem. I backtested the same rules on four pairs with realistic spreads to get a decent sample quickly, then traded it live at minimum size and treated the live results only as a check that execution matched the test, not as proof of the edge.

The other thing I learned is to write down in advance what I will conclude at each milestone. Before I started I noted that at one hundred trades I would only stop if the average was clearly negative. Having that in writing kept me from abandoning it during a bad streak around trade sixty.