Bounded rationality via recursion

Current trends for model construction in the field of Agentbased Computational Economics base behavior of agents on either game theoretic procedures (e.g. belief learning, fictitious play, Bayesian learning) or are inspired by theArtificial Intelligence (e.g. reinforcement learning). Evidence from experiments with human subjects puts the first approach in doubt, whereas the second one imposes significant computational and memory requirements on agents embedded in models used by social scientists. In this paper, we introduce an efficient computational implementation of n-th order rationality using recursive simulation. An agent is n-th order rational if it determines its best response assuming that other agents are (n − 1)th order rational and zero-order agents behave randomly. In recursive simulations, the simulated decision makers use simulation to inform their own decision making. We combine concepts of the n-th order rationality and recursive simulation to create an iterated construction of strategic thinking. Our goal is to provide model builders with an off-the-shelf representation of decision making processes that leads to model consistent behaviors of agents, without requirement for learning phase. Different aspects of n-th order rationality concept are illustrated on two classic games. The paper includes a discussion of experimental support for n-th order rationality as well as its applications within the domain of computational modeling.