A Integração de Séries Temporais e Dados de Textos para a Previsão de Preços Futuros de Milho e Soja

Abstract

Agricultural commodities' prices perform an important role in the global market. Hence to the non-linear and non-stationary temporal series data nature, the price prediction has become a challenge. Many existing forecasting models do not take market sentiment, political events, and economic crises into account. To overcome the limitations described and motivated by the fact that agribusiness related news may have useful forecast information, text mining techniques were applied to add extracted text data and incorporate these data into two agricultural commodities temporal series. Machine learning algorithms with different arrangements were used in soybean and corn price forecasting. Four statistics evaluation techniques were applied to verify the proposed approach effectiveness. Results presented that the implemented model enhances future price forecasts. Thus, data text information offers an alternative for better and enhanced accuracy for price prediction.