Assessing agent-based models in the social sciences

Agent-based models appear strange to many social scientists, and especially to economists. In general, they lack explicit analytical solutions and rely instead on simulation to discover their properties. Individual agents are usually heterogeneous, defying simple, easily generalizable descriptions. Probabilistic behaviour of such agents is normally incorporated in the models. It is not surprising that there is suspicion of this modelling approach, since it looks, indeed is, so different from the deterministic models in which most economists are trained. We are writing this paper from our perspective as economists, although the points we make are relevant more generally to the validation of agent-based simulation models.