The highly intelligent virtual agents for modeling financial markets

Authored by G Yang, J P Huang, Y Chen

Date Published: 2016

DOI: 10.1016/j.physa.2015.09.071

Sponsors: Chinese National Natural Science Foundation Shanghai Key Laboratory of Financial Information Technology Fok Ying Tung Education Foundation Program for New Century Excellent Talents in University

Platforms: No platforms listed

Model Documentation: Other Narrative Mathematical description

Model Code URLs: Model code not found

Abstract

Researchers have borrowed many theories from statistical physics, like ensemble, Ising model, etc., to study complex adaptive systems through agent-based modeling. However, one fundamental difference between entities (such as spins) in physics and micro-units in complex adaptive systems is that the latter are usually with high intelligence, such as investors in financial markets. Although highly intelligent virtual agents are essential for agent-based modeling to play a full role in the study of complex adaptive systems, how to create such agents is still an open question. Hence, we propose three principles for designing high artificial intelligence in financial markets and then build a specific class of agents called iAgents based on these three principles. Finally, we evaluate the intelligence of iAgents through virtual index trading in two different stock markets. For comparison, we also include three other types of agents in this contest, namely, random traders, agents from the wealth game (modified on the famous minority game), and agents from an upgraded wealth game. As a result, iAgents perform the best, which gives a well support for the three principles. This work offers a general framework for the further development of agent-based modeling for various kinds of complex adaptive systems. (C) 2015 Elsevier B.V. All rights reserved.
Tags
herd behavior bubbles complex adaptive system investors Stock-market Phase-transitions Stylized facts