Agent Based Simulation Design Principles – Applications to Stock Market Lev Muchnik 1 , Yoram Louzoun 2 , Sorin Solomon 3,4 1 Department of Physics, Bar Ilan University, Ramat-Gan, Israel 2 Department of Mathematics, Bar Ilan University, Ramat-Gan, Israel 3 Racah Institute of Physics, The Hebrew University, Jerusalem 91904, Israel. 4 Complex Multi-Agent Systems Division, ISI Torino Summary. We present a novel agent based simulation platform designed for general-purpose modeling in social sciences. Beyond providing convenient environment for modeling, debugging, simulation and analysis, the platform automatically enforces many of the properties inherent to the reality (such as causality and precise timing of events). A unique formalism grants agents with an unprecedented flexibility of actions simultaneously isolating researchers from most of the overhead of the virtual environment maintenance. Key words. Agent-Based Simulation; Experimental Markets; Artificial Financial Markets; Market Microstructure. Introduction The classic analysis of financial phenomena is usually based on simple (often linear) macroscopic models, which preferably can be solved analytically. Such models can reproduce basic market macroscopic features. This type of models fails to reproduce emergent features of markets that cannot be directly deduced from the microscopic interaction producing them. Emergent phenomena, were studied over the last couple of decades in a wide range of systems. A general approach is to model the system in question as a set of microscopic elements and define microscopic interactions between them so that the desired macroscopic phenomenon emerges. Being frequently and successfully exploited in physics, this method is now being applied in social sciences as well. In the specific context of the stock market, a variety of simplified microscopic models have been introduced over the last decade, (Bak et al. 1997, Stauffer 2000, Levy et al. 1994, Mantegna R., Stanley 1999, Maslov 2000, Solomon 2000 and many others). Most of these models focus on specific aspects of the problem: basic features of the agent’s behavior or of the stock exchange procedures. They show that even a small set of simple assumptions can explain the set of ‘‘stylized’’ experimental facts characterizing generically the market (Mantegna, Stanley 1999, Cont 2001, Lux, Heitger, Takayasu): power (Pareto–Zipf) laws, fat tails (and/or