Validation and Functional Complexity

T his paper provides a frame work for discussing the validity of computer simulation models of market phenomena. It defines functional complexity and derive sm easures of this for a well known agent-based simulation model and suggests methods to overcome the obstacle of complexity in validating such models.. Acknowledgments: It hank Yaneer Bar-Yam, Ian Wilkinson, and three referees for their assistance. 1. Intr oduction One possible reason for the relative s lowness of the economics discipline to embrace computer simulation models of economic phenomena might be lack of confidence in the behaviour and results exhibited by such models. Even if there are other reasons, better validation of such models would reduce the skepticism about their results. Leombruni et al. (2006) go further ,a rguing that a common protocol for conducting and reporting agent-based social simulations is necessary ,i ncluding fiv et ypes of validation (see below). Midgle ye ta l. (2007) also argue for a common approach for establishing the "assurance" (verification, or program validity ,a nd validation) of agent-based models, introducing afi ve-step process for modellers. This paper discusses the general issue of validation (for whom? with regard to what?) and its relationship to the use of computer models for explanation in Section 2. Section 3 delineates agent- based computer models from other simulation techniques. Section 4 discusses notions of complexity, and argues that functional complexity is the appropriate measure for the complexity of computer simulation models. We c alculate the functional complexity of a well known simulation model, and conclude that such models are too comple xt ob ep roperly tested. Section 5 discusses possibilities for overcoming this obstacle and concludes. 2. V alidation