Using Online Economic News to Predict Trends in Brazilian Stock Market Sectors

Resumo

In this paper, we investigate the impact of online economic news on BM&FBOVESPA, the leading stock exchange in Brazil. We collected economic news published in high circulation online Brazilian newspapers between 2000 and 2015 and extracted, from each news, the reader comments, the amount of shares in social media sites and its associated sentiment. From this, we built several features that were used as input for machine learning models with the aim of predicting the trends on different sectors of BM&FBOVESPA concerning their average price, amount of trades, amount of contracts traded and financial volume. Our assumption is that the more predictable a sector is, based on these features, the more sensitive to news it is. Besides showing that sectors are indeed affected differently by news, we also show that the machine learning models employed performed better than random and other simpler baselines for all sectors, thus providing evidence that news data carry indeed an important signal for understanding BM&FBOVESPA dynamics