He will then run another couple hundred tests and randomly buy each of the candlestick patterns tested if they are also at a 10 day low. Again he will average out the 5 and 10 day % returns of the closed positions.

If the average 5 and 10 day returns of the patterns are better than the average 5 and 10 day returns of randomly buying...is that adequate to suggest that the patterns have a 'predictive' quality?