Essays in agent-based computational finance

This dissertation provides theoretical economic foundations to explain key empirical properties in stock markets. The first chapter explains two features on stock return volatility, i.e., volatility clustering and asymmetric volatility. The chapter explains these features based on a close examination of how investors learn. I found that volatility clusters as investors learn through imitating the prediction methods of others. Asymmetric volatility is observed as investors learn from their own experience. Such learning implies no imitation across investors, so predictions across investors become more diversified than those with imitation learning. Heterogeneous predictions generate asymmetric volatility. I also show that both volatility phenomena can simultaneously characterize a market with a simple extension of an imitative learning economy which also allows heterogeneous predictions.;The second chapter introduces a new learning mechanism where investors can update their trading rules from their own experience and imitating others. Since this extension gives investors more ideas to update their strategies, investors become more sophisticated. I first ask if such an economy with sophisticated investors can converge to a homogeneous rational expectation equilibrium (HREE). However the result says that investors' intelligence does not contribute to the HREE. Second, I ask which learning investors often prefer. The result shows that most investors often imitate others while only rich investors use their own ideas. So, imitation dominates the market, indicating that investors often herd on others. I show that income differences are crucial to generate herding behavior.;The third chapter introduces an order-driven market with heterogeneous investors, who submit limit or market orders according to their own trading rules. The trading rules are repeatedly updated via simple learning and adaptation of the investors. I analyze markets with and without learning and adaptation, and show that a model with learning and adaptation successfully replicates long-memories of return volatility, trading volume, and signs of market orders. I also argue what drives these long-memories, and conclude that evolution on trading strategy is crucial to understand those features.