The objective of this work is to empirically test the EMH (efficient market hypothesis) and compare its results to those of a viable agent-based competitor using computational simulation. The models are not directly fit to the data; random walk and agent-based methods of stock price determination are statistically compared using the criteria of stationarity, randomness, and autoregressive behavior. The agent-based approach used, styled the "ant trader" model, is based on the ant model established by Kirman in his 1993 work "Ants, Rationality, and Recruitment". Daily returns of the Hang Seng and Nikkei 225 indices are used over the periods 1987-2007 and 1984-2007, respectively. Preliminary simulations run with the agent-based model indicate high sensitivity to parameter changes; parameter imbalances lead to unrealistic growth in returns. Batch stationarity tests using ADF and PP tests suggest that the two models behave similarly under the chosen parameter conditions. However, the random-walk model is found to be more consistent with the available data when using the Wald-Wolfowitz runs test and the LoMacKinlay variance ratio test. We conclude that the EMH can be theoretically challenged by the ant trader model, but not empirically. The agent-based model has more realistic assumptions and is more flexible; however, the random walk model agrees with the stationarity and randomness properties of realworld stock index return.
Swisher IV '08, Scott N., "Stock Index Pricing with Random Walk and Agent-Based Models" (2008). Honors Projects. Paper 84.