A plain-language t-test for a strategy's results

[Shadow Trader]

A t-test asks whether an average result is large enough, relative to the noise, to be unlikely under "no edge". It sounds technical and is easy to compute.

Take the net result of each trade in R. Calculate the mean, the standard deviation and the number of trades. The standard error is the standard deviation divided by the square root of the number of trades. The t-statistic is the mean divided by the standard error.

An example. Mean 0.12R, standard deviation 1.1R, 300 trades. The standard error is 1.1 divided by about 17.3, or 0.064. The t-statistic is roughly 1.9. A value around 2 is the conventional threshold for significance at about the 95% level, so this result is borderline. More trades would sharpen the picture.

Caveats apply. Trades may be correlated, results are not normally distributed, and strategy selection inflates the statistic. A t-statistic of 2 after testing 50 variations means little.

Still, computing it forces the right question: is the average big enough compared with its own uncertainty?

One more point: the same calculation shows why a longer record matters so much. Doubling the number of trades shrinks the standard error by about 30%, so the same mean produces a higher t-statistic. A borderline result at 300 trades may become convincing at 600, or evaporate, which is useful to know before putting real size behind it.

What t-statistic does the current strategy produce, and how many variations led to it?