Forecasting realized volatility using news flow

Abstract

Economic news contain valuable information to predict future movements in financial market prices. We explore the relative importance of news flow to forecast realized volatility in the Brazilian stock market. We build news-based uncertainty indicators from articles of major newspapers in Brazil. We then incorporate these indicators into volatility models that already account for persistence, leverage effects, jumps, and market microstructure noise. We find that adding news-based indicators significantly improves the forecasting ability of volatility models, especially for the most liquid stocks and, perhaps surprisingly, for longer horizons. Because news cycles are more persistent than negative returns and jumps, they contribute more than the latter in forecasting realized volatility up to four weeks ahead.