Desenvolvimento de um simulador de mercado artificial de ações para ainteração de agentes externos

• Ano de defesa: 2014
• Autor(a) principal: Álvaro Ledo Ferreira
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: Simulador de negociação de ações; Mercado artificial por agentes; Ações (Finanças); Engenharia de produção
• Link de acesso: http://hdl.handle.net/1843/BUBD-AJVQ96

Over the last years the real interest rate of the Brazilian economy has been systematically reduced, which stimulates the migration of investments to equities, particularly the stock market. In order to ease the entry of new investors in this market, the use of simulators has shown to be an important tool to demystify the process of negotiation, without involving real money. The online simulators currently available, like FolhaInvest and UolInvest, use the quotations from BM&FBOVESPA with a 15 minutes delay in the simulator, what leads to the main limitations ofthose simulators: (i) the buy/sell orders of the players dont affect the market price; (ii) the use of the simulator is linked to the trading time of BM&FBOVESPA. An alternative would be to use Agent based Artificial Markets, but despite the extensive literature dealing with these markets,none of these models enables the interaction of external agents. They are not intended to be used as a learning tool, but to verify if the price series obtained at the end of the simulations, resulting from the proposed microstructure, present characteristics similar to real market price series. In this paper we propose the development of a simulator consisting of: i) agent based artificial market to negotiate multiple actions and that allows external user(s) to also act as agent(s) in this market, with simulations occurring at any time; ii) access structure that allows the externaluser to interact with the artificial market. The proposed simulator is based on the use of at least two instances of the R software, the first one is responsible for implementing the artificial market, and the second one is responsible for the interaction with external agents, connected bya common database. The proposals include versions of the simulator both for local use (only by one user) and online use via internet (for multiple users). The results include the development of a methodology to compare the behavior of the proposed artificial market with that observedin a real market (in this case stocks from the New York Stock Exchange - NYSE), through the frequency of occurrence of stylized facts. In the various proposed configurations, the Artificial Market did not show all the stylized facts considered simultaneously at the same frequency observed on the NYSE, but this pattern is much more demanding than simple comparison with a significance level of 5%. The proposed interface, similar to a Home Broker, was developed in HTML/CSS/JavaScript and is common for both of the proposed versions (local and online). At this point of the work, the local version of the simulator was not feasible because it presentedan installation complexity greater than initially proposed, with the change of the SQLite local database for a MySQL database server. The proposed online version is functional, but to be made available to the end user it still depends on the implementation of authentication systems that are beyond the scope and the time available for the completion of this work. Finally, we simulated the impact of the participation of external agents in Artificial Market, and the results indicate that the proposed artificial market is able to keep functioning even with the entry and exit of capital on the system.