JAS-mine provides a unique tool for both microsimulations and large-scale agent-based applications. Dynamic microsimulations are computer models aimed at applying specific processes (as aging, educational choices, labor market events, household formation, etc.) to a representative sample of a given population, in order to forecast its evolution or understand the distributional impact of some policy of interest. While some processes might be deterministic (as aging, or pension eligibility), other processes are stochastic (as mortality, or retirement choices) and are generally estimated in the data.
Agent-based models (ABM) also take an initial population and simulate it forward in time. The main differences between the two approaches can be traced down to the following: (i), microsimulations are more policy-oriented, while agent-based models are more theory oriented; (ii) microsimulations generally rely on a partial equilibrium approach, while agent-based models are most often closed models. The initial population in an agent-based model is typically not meant to reproduce a real population of agents: for example, in a labor market model with firm creation all individuals might be initiated as unemployed, or randomly employed. The focus is on the emergence of aggregate patterns from the interaction of the individual agents, with the aim to replicate some observed stylized fact (business cycle fluctuations, for instance). Accordingly, the value of the parameters that drive the processes are chosen ad-hoc, or only roughly calibrated with real data. However, in their struggle to replace dynamic stochastic general equilibrium (DSGE) models, agent-based models are becoming more empirically oriented. At the same time, microsimulations are becoming more complex, by including more behavioral responses and general equilibrium feedbacks.
If the two approaches retain different goals and perspectives, from a mathematical and computational perspective they are identical. Both agent-based models and microsimulations are recursive models, where the number and individual states of the agents in the system are evolved by applying a sequence of algorithms to an initial population. As computer-based simulations, they face the problem of reproducing real-life phenomena, many of which are temporally continuous processes, using discrete microprocessors.
For a more in-depth analysis, see:Richiardi M (2013). The missing link: AB models and dynamic microsimulation. In: Leitner S., Wall, F. (eds). Artificial Economics and Self Organization. Springer, Lecture Notes in Economics and Mathematical Systems, vol. 669, Berlin.