This paper studies portfolio decisions of boundedly rational agents in a financial market. Learning is modeled via a genetic algorithm. Learning as modeled in this paper leads agents to hold too much risk as compared to the optimal portfolio of rational investors. Moreover, adaptive agents exhibit an asymmetric response after positive and negative returns where the portfolio adjustment is more pronounced after negative returns. It is demonstrated that investors in mutual funds show the same investment patterns as the adaptive agents in the model. A model with entry and exit of agents is able to match the mutual fund data closely.