Efetividade de modelos de previsão em fundos imobiliários: um estudo comparativo entre E LSTM E ARIMA

Abstract

The financial market plays a fundamental role in the global economy by facilitating the efficient allocation of resources. It serves as a mechanism to connect investors with investment opportunities, promoting economic growth and innovation. Within this context, real estate investment funds are particularly important as they allow small investors to access the real estate market with greater ease and diversification. In this scenario, predicting the returns of brazilian real estate investment funds - Fundo de Investimento Imobiliário (FII), can help investors select the assets to be included in their investment portfolios. This study compares the effectiveness of different price prediction models in real estate investment funds, evaluating the predictive capacity for the largest equity funds in each sector, as defined by ANBIMA. For this purpose, Long Short-Term Memory (LSTM) and AutoRegressive Integrated Moving Average (ARIMA) models were used to predict the next day's closing price (D+1). Models were developed for each of these funds using different sliding windows, aiming to identify which approaches offer the best predictive results. The error metrics used in the comparison of the models were the mean squared error (MSE). The analysis results indicate that the ARIMA model consistently outperforms the LSTM model in all comparisons, demonstrating greater accuracy in predicting the closing prices of the analyzed real estate investment funds.