t statistic

A t statistic compares an estimate with the value it would have if there were nothing there, in units of the estimate's own uncertainty. For an average monthly return, it is the average divided by its standard error: how much the average itself would move from one sample to another, which shrinks with more months and grows with wilder swings. A big average with a small standard error gives a high t; a big average with a huge one gives a low t.

For returns that lean on each other from month to month, the standard error has to account for that dependence, or the t comes out too confident. In the logbook the t is computed with that adjustment where it matters.

The usual bar is two. It is a convention, calibrated for particular assumptions about the data, not a universal proof of an edge: a t of two on a hundred months of a fat-tailed series is weaker evidence than the same t on a thousand months of a well-behaved one. What matters in the logbook is that the bar is fixed before the test runs and does not move afterwards: a t of zero point six nine with a positive average is still empty waters.