Applications in Agent-Based Computational Economics

A constituent feature of adaptive complex system are non-linear feedback mechanisms between actors. This makes it often difficult to model and analyse them. Agent-based Computational Economics (ACE) uses computer simulation methods to represent such systems and analyse non-linear processes. The aim of this thesis is to explore ways of modelling adaptive agents in ACE models. Its major contribution is of a methodological nature. Artificial intelligence and machine learning methods are used to represent agents and learning processes in ACE models. In this work, a general reinforcement learning framework is developed and realised in a simulation system. This system is used to implement three models of increasing complexity in two different economic domains. One of these domains are iterative games in which agents meet repeatedly and interact. In an experimental labour market, it is shown how statistical discrimination can be generated simply by means of the learning algorithm used. The aim of this model is mainly to illustrate the features of the learning framework. The results resemble actual patterns of observed human behaviour in laboratory settings. The second model treats strategic network formation. The main contribution here is to show how agent-based modelling helps to analyse non-linearity that is introduced when assumptions of perfect information and full rationality are relaxed. The other domain has a Health Economics background. The aim here is to provide insights of how the approach might be useful in real-world applications. For this, a general model of primary care is developed, and the implications of different consumer behaviour (based on the learning features introduced before) analysed.