I have built an automated system that seems to be able to correctly predict upward stock movement with about 65% precision (i.e., whenever it makes a prediction, it is correct 65% of the time). The problem is that the prediction horizon is on the scale of minutes, so we're often pursuing price...