Artificial Stock Market Research Based On Multi-Agent System

The modern standard finance theory cannot explain many anomalies which exist in real markets. Behavioral finance theory considers investors as irrational and financial markets inefficient. Actually, it studies investors' market behavior from psychological principles of decision making, and concludes that it is the investors' irrational behavior that leads to various market anomalies. At present, the behavioral finance theory stays in its babyhood for lack of mature financial models and analyses of real market data.Facing the questions of explaining anomalies under the EMH, verifying the EMH, or finding smart tools to study behavioral finance theory, the researchers find that there are no mature models or tools in mathematics to describe a financial market when being as a complex adaptive system.Based on the previous works, the author applies multi-agent technique with reference to concepts of standard finance theory and behavior finance theory, and constructs an artificial stock market with aggressive auction regime to study financial anomalies. The investor models and market structure model are linked by time model in the market. In the investor models, the investors form expectations from views of fundamentalists, chartists, noise traders respectively, make decisions by optimizing constant absolute risk aversion (CARA) utility functions, modify their decisions with their own confidence level and the consistency of decisions from others which are gained under contacting mechanism, and evolve gradually with GA learning.Five experiments are performed and the results show that: (1) When there are only pure noise traders in the market, the GARCH property is found in time series, but there is no fat tail phenomenon. (2) When the market contains fundamentalists and noise traders, the data properties are consistent with those in standard finance in the case that there are more fundamentalist: the market is controlled by fundamentalist, the price is close to REE price, and there are small volumes with little GARCH property but some fat tails. The fact is discovered that the less fundamentalists, the weaker strength in controlling market by them, which validates the limited ambiguity theory put forward by Fama. What's more, in the case of permitting traders to switch investing strategy, the noise traders convert into fundamentalists soon. (3) When chartists and noise traders are in the market, the data exhibits the periodic property and strong GARCH which show that chartists have controlled the market. But there are no fat tails in return's distribution which is not satisfied with the normal distribution. (4) When there are three types of investors in market who don't refer to others' decisions, the GARCH and fat tails are found in data, which seems to be the fusion of experiment 2 and 3. (5)When three types trader can refer to others' decisions in the experiment which is performed under the same parameters with experiment 4, the stronger GARCH and fat tails are found. It shows that the psychological and social factors influence the market.The consistency in data properties between above experiments and real world shows that the models are basically right and can be used to describe the anomalies from the view of quantity or system science.