Compute a t-Statistic for Your Own Results

[Fairman]

Compute a t-Statistic for Your Own Results

A trader has 64 trades with an average of plus 0.30R and a standard deviation of 1.4R. Is the result evidence of an edge?

Compute the standard error: 1.4 divided by the square root of 64, which is 1.4 divided by 8, or 0.175R. The t-statistic is the average divided by the standard error: 0.30 divided by 0.175, about 1.71.

Interpretation. A t-statistic around 2 is often taken as moderate evidence that the true mean is above zero. At 1.71, the evidence is suggestive. Under a one-sided test, the probability of seeing such a result from a zero-edge strategy is about 4 percent. Under a two-sided test, about 9 percent.

But apply caution. If the trader tried many strategies before finding this one, the result is subject to selection bias. If trades are not independent, the standard error is understated. If the sample was drawn from a single market regime, it may not generalize.

Compute how many trades would raise the t-statistic to 2.5 if the average stays at 0.30R: n equals (2.5 times 1.4 divided by 0.30) squared, which is 136 trades.

So the trader needs about 70 more trades to reach stronger evidence, if the edge persists.

Meanwhile, avoid overconfidence. The honest summary: the evidence leans positive, and it is not conclusive.

Exercise: compute the t-statistic for your last 50 trades, and note how many more trades you would need for 2.5.