This paper develops and calibrates an agent-based model of boundedly rational, adaptive agents in a two-good production and exchange economy to replicate human-subject outcomes in the same eight-person experimental economy. To test agents’ ability to capture human behavior, it extends the model and uses its output to make predictions about a second experimental environment in which the group of eight agents is slowly constructed by merging smaller groups. This environment improves human-subject performance in the specialization and exchange task, and commensurate improvement emerges for some parameterizations of the agent-based model. This iterative process yields incremental improvement of decision-level theories about economic discovery.
Find article at Science Direct.