This paper applies standard maximum likelihood (ML) techniques to find an optimal agent-based model (ABM), where optimal could refer to replicating a pattern or matching observed data. Because ML techniques produce a covariance matrix for the parameter estimates, the method here provides a means of determining to which parameters and conditions the ABM is sensitive, and which have limited effect on the outcome. Because the search method and the space of models searched is explicitly specified, the derivation of the final ABM is transparent and replicable. Hypotheses regarding parameters can be tested using standard likelihood ratio methods.
More computational firepower has allowed two types of research to flourish. The first is the agent-based model (ABM), which keeps track of the decision making behavior of thousands or millions of agents. The second is optimization as a statistical method, typically involving resampling a likelihood function hundreds or thousands of times.