Artificial stock markets with different maturity levels: simulation of information asymmetry and herd behavior using agent-based and network models

Abstract

This paper aims mainly at building artificial stock markets with different maturity levels by modeling information asymmetry and herd behavior. The developed artificial markets are multi-assets, order-driven and populated by agents having heterogeneous behaviors and information. Agents are defined by their information and their herd behavior levels. Agents trade multiple risky assets based on their wealth, their behaviors and their available information which spread among multiple behavioral networks. In a novel contribution to artificial stock markets literature, agents’ behaviors modeling is mixed with social network simulation to reproduce different degrees of information asymmetry and herd behavior based on several assortative topologies. Several simulations validated the proposed model since univariate and multivariate stylized facts were reproduced both for mature and immature stock markets. The proposed artificial stock market can be considered as a first step toward decision and simulation tools for optimal management, strategy analysis and predictions evolution of immature stock markets.

Appendices

Appendix 1: Behavioral network algorithm

For simplicity, we drop the asset subscript k from all variables. We consider an initial network, $G_0$ formed by two agents, where the fundamentalist exerts influence on the chartist. At each time step, a new agent i comes. With probability $\alpha$ it chooses to emit $m_{out} \in \{0,\dots ,m\}$ out-edges proportional to its fundamentalist weight, by selecting a set of existent agents following their in-strength $s_j^{in} + \delta _{in}$. $\delta _{in}$ reflects that even if one agent has no-influence, it has always a chance to receive in-edges from new comers. Then, agent i will select other out-neighbors from in-neighbors set of its out-neighbors, based on their out-strength depending on parameter $\theta$ (influential agents share private information between them). With probability $\gamma$, new agent i chooses to receive $m_{in} \in \{0,\dots ,m\}$ influences proportional to its chartist weight, from existent agents selected based on their out-strength $s_j^{out} + \delta _{out}$. Therefore, agent i will look for clusters by establishing other in-edges based on the out-neighbors set of its in-neighbors (influenced agents share private information between them). Finally, to enrich the behavioral specificity of the network, with probability $\beta$, existent agents will exchange connections among them. Indeed, more influential agents will have more chance to exert influence on more influenced agents. Following Bollobás et al. (2003), we define parameters as $\alpha + \beta + \gamma = 1$, and we favor the probability $\beta$ rather than $\alpha$ and $\gamma$ to promote behavioral out and in connections, instead of randomness. We choose also, $\alpha = \gamma$ and $\delta _{in} = \delta _{out}$ in order to introduce symmetry in random $out-in$ edges.

Appendix 2: Multi-Assets BI-ABM algorithm

By considering agents with their behaviors and multiple traded assets, Algorithm 2 describes the way of forming multiple artificial price series.