Análise do modelo de Kiyotaki-Wright em simulações multiagentes que utilizam a plataforma Swarm

• Ano de defesa: 2011
• Autor(a) principal: RODOVALHO, Wildener Monteiro
• Orientador(a): CRUZ JÚNIOR, Gélson da
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Goiás
• Programa de Pós-Graduação: Mestrado em Engenharia Elétrica e de Computação
• Departamento: Engenharia
• País: BR
• Palavras-chave em Português: Sistema multi-agentes; Plataforma Swarm; Modelo de Kiyotaki-Wright; Aprendizado por Reforço
• Palavras-chave em Inglês: Multi-Agent System; Swarm Platform; Kiyotaki-Wright Model; Reinforcement Learning
• Área do conhecimento CNPq: CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::METODOLOGIA E TECNICAS DA COMPUTACAO::ENGENHARIA DE SOFTWARE
• Link de acesso: http://repositorio.bc.ufg.br/tede/handle/tde/965

The goal of this dissertation is to make a methodological repetition of Duffy and Ochs's and Rouchier's works. Nonetheless, it tries to have a different focus. Among many theoreticaleconomical models which explain the emergence of money, there is one that distinguishes itself due to its simplicity, efficiency, and robustness: Kiyotaki-Wright's model. Therefore, some economists devoted themselves to apply this model in experiments with human beings and in computer simulations. Among those economists, Duffy and Ochs applied it later, correcting some flaws of the application process used by their predecessors. Duffy and Ochs's studies showed that Kiyotaki-Wright's model is actually valid for its purpose. However, besides all the efforts made, their agents were not able to achieve the foreseen equilibrium. In the same way, the French researcher Rouchier repeated methodologically the works of Duffy and Ochs while criticizing their implementation details. From this research, Rouchier creates other four sub-models in order to take her agents to the proposed equilibrium, obtaining partial success. In this dissertation, the computer system development platform is Swarm, especially created to multi-agent simulations and used, most of the time, in social and biological scope models. The agents' rationalization process is undertaken in trial and error way in accordance with the reinforcement learning technique, analyzing the answer given by the environment in each period. Finally, a new entity called Mediator is created to intermediate the relations between agents and mainly to pair them randomly. The obtained results are, in general, closer to the equilibrium when compared to the results of previous research. From them, some studies are made about the relation between the number of agents and the number of periods of simulation as values approximates the equilibrium.