In the past data we checked, the stocks this screen picked did better than the ones it left out, but a portfolio built from them did not clear our test after trading costs. We looked back over 64 months. Each month we checked whether the stocks this screen picked did better, over the following month (21 trading days), than the stocks it didn't pick. We score that from -1 (the picks always did worse) to +1 (always did better); 0 means no pattern. This screen scored 0.0492. As a measurement, the score is small, but clearly above zero. If the screen had no real edge, a score at least this far from zero would show up less than 1 in 1,000 by luck alone.

We also tested a portfolio of the screen's picks, using only what was known at the time and after estimated trading costs: it earned 0.40 for each unit of ups and downs, per year. Because this screen has been tested 2 times, we mark down results that could be flukes. After that, the chance that it truly beats simply holding the index is 0.324, about 32 in 100.

A pattern can exist and still not be worth trading after costs. This is a record of the past, not a prediction. A screen is still a way to narrow a list by your own rules, whatever the measurement says.