Big Data Meets the Turbulent Oil Market

Abstract

This paper introduces novel news-based measures for tracking energy markets. Our parsimonious yet broad set of indicators reflects the information content of millions of news articles and forecasts oil spot, futures, and energy company stock returns, and changes in oil volatility, production, and inventories. Our measures are not spanned by existing text or nontext variables suggested in the literature. We identify the specific aspects of news flow that are informative for each outcome  variable. By developing a methodology to determine, ex-ante, times when effective predictions are possible, we show our text-based measures provide robust and economically important out-of-sample forecasts.

Keywords: Asset Pricing, Commodity Markets, Energy Forecasting, Model Validation

JEL Classification: C52, G10, G14, G17, Q47