Influência dos textos de notícias na queda de preços no mercado de ações brasileiro

Abstract

Forecasting financial losses and making decisions to avoid or reduce them has been a challenge for every investor. On one hand, due to the availability of data and its simple implementation, technical analysis methods have been quickly gaining supporters. On the other hand, modern computers processing power together with advances in text mining provides the opportunity to explore the investor’s behaviors in new data types: textual. This research evaluates the relationship between the Brazilian stock market and news published on national midia, focusing on automatic search for patterns related to down movements using machine learning algorithms. Six experiments were performed to analyze the possibility of predicting price falls automatically, followed by case studies in the search of explanations from the classifiers that justify the predictions.The results show that text mining based approaches overcome traditional strategies when forecasting losses, but the underlying patterns understanding is limited due to the complexity of the classifiers and high dimensional vocabulary.