Online Convex Optimization Against Adversaries with Memory and Application to Statistical Arbitrage