Mulit-agent based simulation is a method for modelling social phenomena bottom-up. Different from the usual statistical methods it aims to generate data and results instead of deriving them from existing data. Hence also non-linear results are possible.
The basic idea of agent-based modelling is the represenation of actors, e.g. humans or groups, as computer entities. To get a executable model, often many assumptions about motiviations, plans etc. have to be made. For these micro-level information usually no information is available or has to be acquired first.
A main field of application is therefore the use as “foresight tool” that allow to simulate possible trend, developments and different scenarios. Validation is usually done on an abstract level, e.g. by experts, the comparison of stylized facts or, if data are available, the comparison with empirical distributions.
A possible area of application is marketing research. In this area a rich academic research body about consumer choice, innovation diffusion, the simulation of business strategies and similar questions already exists.
In Application in Agent-based Economics an approach for simulating bounded rational agents is proposed. This approach forms the bases for the simulation framework gsim. gsim provides an API that eases the development of simulations. The thesis contains three applications in Game Theory and Health Economics. The re-implementation of the statistical discrimination model can be accessed on the software page.