NON-EXTENSIVE AND EXTENSIVE STATISTICS IN THE AGENT-BASED HERDING MODEL

Aleksejus Kononovicius1, Julius Ruseckas1

1 Institute of Theoretical Physics and Astronomy, Vilnius University

Power-law statistical properties are observed in systems with long-range memory and long-range interactions. Systems possessing these anomalous properties can not be understood in a traditional framework of Boltzmann-Gibbs statistical mechanics. Thus they are studied by using a generalized framework, known as non-extensive statistical mechanics [1]. In the core of this framework lies a distribution which is a generalization of the traditional Gaussian distribution, known as a q-Gaussian distribution,

$$p(x) \sim \exp_q\left(-\frac{x^2}{(3-q)\sigma_q^2}\right) = \left[1 + \frac{q-1}{(3-q)\sigma_q^2}x^2\right]^{-\frac{1}{q-1}} \sim x^{-\frac{2}{q-1}}.$$

In the above $[z]_+ = z$ if $z > 0$, and $[z]_+ = 0$ if $z \le 0$. It is straightforward to notice that in the limit of large values of $x$, if $q \ne 1$, the q-Gaussian distribution behaves as a power-law distribution. While if $q \to 1$ the q-Gaussian distribution reduces to the Gaussian distribution. The q-Gaussian distribution and related concepts from the non-extensive statistical mechanics have found their applications in broad range of fields, such as physics, chemistry, biology, computer science and even social sciences [2].

Contemporary modeling of the complex socio-economic systems frequently uses agent-based modeling framework [3]. In this framework distinct entities acting in the modeled system are replaced by a generalized objects, known as agents. These agents are assumed to act based on a very simple rules. For example, there are agent-based models for the financial markets in which the internal intricacy of the trading mechanics and strategies is successfully replaced by a simple herding interactions. Interestingly enough despite the simplistic behavioral rules in certain cases the ensembles of agents are able to reproduce the complex collective behavior observed in socio-economic systems [4, 5, 6]. In order for agent-based model to be useful it should also be analytically tractable. Analytical tractability introduces certain level of transparency into the model, which may be useful in understanding and further developing the model. Currently there are some approaches starting from the agent-based models, which analytically obtain non-linear stochastic differential equations as a macroscopic model for the agent-based dynamics [3, 4, 5, 7].

In this contribution we consider agent-based herding model proposed by A. Kirman [8]. This agent-based model proves to be very interesting to us as it is able to reproduce both the power-law and Gaussian-like distribution [4, 5]. In [5] Kirman model was tested on random graph, small world and scale-free networks. It was shown that Kirman model produces power-law distribution if average number of contacts of each agent scales linearly with the system size as average degree of nodes in the random graph network, $\langle d \rangle \sim N$. While if the average number of contacts is fixed, as in small world or scale-free network, $\langle d \rangle \sim$ const, the Gaussian-like distribution is observed. In this contribution we discuss how Kirman model works on a densifying network, $\langle d \rangle \sim N^{\alpha}$ ($0 \le \alpha \le 1$), obtaining continuous transition between the non-extensive and extensive statistics.


[1] C. Tsallis, Introduction to Nonextensive Statistical Mechanics (Springer, New York, 2009).

[2] Cf. M. Gell-Mann, C. Tsallis, Nonextensive Entropy - Interdisciplinary Applications (Oxford University Press, New York, 2004).

[3] M. Cristelli, L. Pietronero, A. Zaccaria, Critical Overview of Agent-Based Models for Economics, Proceedings of the School of Physics "E. Fermi" CLXXVI, 235 - 282 (2012).

[4] A. Kononovicius, V. Daniunas, Agent-based and macroscopic modeling of the complex socio-economic systems, Social Technologies 3, 85-103 (2013).

[5] S. Alfarano, M. Milakovic, Network structure and N-dependence in agent-based herding models, Journal of Economic Dynamics and Control 33, 78-92 (2009).

[6] R. Frederick, Agents of influence, PNAS 110, 3703-3705 (2013).

[7] A. Kononovicius, V. Gontis, Agent based reasoning for the non-linear stochastic models of long-range memory, Physica A 391, 1309-1314 (2012).

[8] A. P. Kirman. Ants, rationality and recruitment, Quarterly Journal of Economics 108, 137-156 (1993).