Computational modeling of collective human behavior: The example of financial markets

Abstract

As a result of the increased availability of higher precision spatiotemporal datasets, coupled with the realization that most real-world human systems are complex, a new field of computational modeling is emerging in which the goal is to develop minimal models of collective human behavior which are consistent with the observed real-world dynamics in a wide range of systems. For example, in the field of finance, the fluctuations across a wide range of markets are known to exhibit certain generic stylized facts such as a non-Gaussian 'fat-tailed' distribution of price returns. In this paper, we illustrate how such minimal models can be constructed by bridging the gap between two existing, but incomplete, market models: a model in which a population of virtual traders make decisions based on common global information but lack local information from their social network, and a model in which the traders form a dynamically evolving social network but lack any decision-making based on global information. We show that a combination of these two models - in other words, a population of virtual traders with access to both global and local information - produces results for the price return distribution which are closer to the reported stylized facts. Going further, we believe that this type of model can be applied across a wide range of systems in which collective human activity is observed. © 2008 Springer-Verlag Berlin Heidelberg.