Paris Wholesale Market

Simulation of the Rungis Wholesale Market: lessons on the calibration, validation and usage of a Cognitive Agent-based Simulation

Philippe Caillou∗ , Corentin Curchod† and Tiago Baptista‡ ∗ LRI, Universite Paris Sud, France, Email: caillou@lri.fr † Audencia Nantes School of Management, France, Email: ccurchod@audencia.com ‡ CISUC, Department of Informatics Engineering, University of Coimbra, Portugal, Email: baptista@dei.uc.pt

Abstract

In this paper, we present some methodological lessons and thoughts inferred from a research we are making on a simulation of the Rungis Wholesale Market (in France) using cognitive agents. The implication of using cognitive agents with an objective of realism at the individual level contradicts some of the classical methodological assertions about simulations. Three such lessons are of particular interest: the calibration and validation focus on individuals rather than global values (1); the definition of the simulation model is made independently from the research objectives (2), and without targeting the usual objective of hypothesis simplicity (3). Our goal here is to briefly present the simulation and to discuss more in-depth the main methodological lessons learned from this work.

1. Introduction

Multi-Agent simulations (MAS) are increasingly being considered as flexible and versatile modeling frameworks, enabling positive and normative investigations of phenomena out of reach when one uses analytical studies[1], [2]. Investigated domains usually suppose a large number of interacting agents who, at an aggregated level of the simulation, must act in coherence with chosen stylized facts derived from empirical compilation of data. In other words, in Agentbased Computational Economics (ACE), the global dynamics of the system is supposed to be realistic, but not the individual behavior. ACE constitutes a powerful tool to test the impact of clearly delineated variables on outputs at an aggregated level, without going through...