Reinforcement Learning in Repeated Portfolio Decisions