Computational modeling of artificial economic agents' trading behaviors in a synthetic stock market
Abstract
This purpose of this study is to investigate trading behaviors in the synthetic stock market. An agent-based computational economic simulation is used to examine the impact of changes in different parameters on the trading behavior and welfare of market participants. This approach allows for the consideration of the evolutionary characteristics of the agents, as the framework is driven by feedback from evolving market environments. There are three chapters in this thesis. First, we look at the overconfident behavior. Next, we study the disposition effects in agents and lastly we study the memory effects on agent behavior. Overconfident behaviors are characterized by people's tendency to overestimate their ability to make correct decisions because they overestimate the significance of their private information. The disposition effect, a phenomenon first described by Shefrin and Statman (1985), involves investors “selling winners too soon and holding losers for too long.” In essence, an investor (agent) exhibiting the disposition effect, referred to as a disposition agent in this study, will be risk-averse when achieving gains and act in a risk-seeking manner when undergoing losses. We also analyze two cases of non-rational agents with no memory and with memory about market information in a market structure such that the true signal of the asset is revealed every few periods. Our study found that the more of the non-rational adaptive agents in the market and the more the positive private signals for non-rational adaptive agents, the greater the incidence of overconfident behaviors. In addition, market trends significantly affect the welfare of non-rational agents when they have negative market expectations. We also found the support for selling winners too soon in the market, but holding losers for too long was found to be undesirable for the disposition agents. Lastly, we learned that more memory does not necessarily translate into better performance. This is because non-rational agents have to adjust their strategies given additional information. Moreover, non-rational agents do not necessarily perform worse when true information is not revealed for longer periods.
Degree
Ph.D.
Advisors
Chaturvedi, Purdue University.
Subject Area
Management
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.