[gwern]

I would suggest a Bayesian decision-theory test of mean profitability: http://www.gwern.net/Candy%20Japan (Negative, so the decision is to use the old box.)