Omelianchyk D. Agent-based models of computational economics

The goal of this thesis is to develop and analyze of agent-based models (ABM) and computational methods, which allows to evaluate and compare, how efficient various approaches to modeling agent behavior are. Original special ABM of economics is designed as a constructor of models with homogenous structure, which allow different implementations of independent components. We define a corresponding structure for all constituent agents, including control parameters, attributes, external components and methods. Methods of production decision making are divided in groups, according to the key approach, used for their creation. In total, we distinguish a group of intuitive learning methods, two groups of specific and non-specific machine learning methods and a group of methods, using rules for learning. Comparative analysis of different agent learning methods is performed via developed technique of composite index and rating transformation. Empirical validity of the developed ABM is verified with stylized facts and statistical approaches. Therefore, in our thesis we suggest a mathematical apparatus, which includes methods of modeling agents, their behavior and learning, as well as evaluation criteria for their activity. It could be used in the other models of agent-based computational economics, aimed to reproduce various aspects of economic agent activities. In addition, developed special agent-based model of economics could be used as an experimental platform for testing of various agent learning methods, as well as a lab for checking economic assumptions and scenario analysis in real economic systems.