Uso de LSTMS para previsão de tendências no mercado de fundos imobiliários

Abstract

Resumo: O mercado de renda vari'avel est'a em forte expans˜ao no Brasil. Neste cen'ario o estudo de s'eries temporais financeiras 'e importante para compreens˜ao, previs˜ao de valores futuros e tomada de decis˜oes eficientes nesse mercado. Este trabalho tem por objetivo identificar o potencial de se utilizar redes neurais recorrentes do tipo Long Short-Term Memory (LSTM), para tomada de decis˜ao di'aria e de forma a obter ganhos consistentes no mercado de fundos imobili'arios brasileiro. A partir do uso de s'eries temporais da variac¸ ˜ao dos prec¸os desses fundos, 'indices gerais de mercado, cˆambio entre outros. Abstract—The variable income market is booming in Brazil. In this scenario, the study of financial time series is important for understanding, forecasting future values and making efficient decision in this market. This work aims to identify the potential of using some type of recurrent neural networks called Long Short- Term Memory (LSTM), for daily decision making and in order to obtain consistent gains in the Brazilian real estate market. Using time series of price variation for these funds, general market indices, exchange rates, among others.