The Most Predictive Energy Search Terms

Abstract

Internet search activity data has been widely used as an instrument to approximate trader attention in different markets. This method has proven effective in predicting market indices in the short-term. However, little attention has been paid to comparing various search keywords and finding the most effective terms representing attention in different markets. This study attempts to build the best practically possible proxy for attention in the market for major energy commodities using Google search data.

Specifically, first we confirm that Google search activity for energy-related keywords are significant predictors of energy price volatility. We show that search trends data have incremental predictive power beyond the conventional GARCH models. Next, starting with a set of ninety terms used in the energy sector, the study uses a multistage filtering process to create combinations of keywords that best predict the volatility of crude oil (Brent and West Texas Intermediate), conventional gasoline (New York Harbor and US Gulf Coast), Heating Oil (New York Harbor), and natural gas prices. For each commodity, combinations that enhance GARCH most effectively are established as proxies of attention. The results indicate investor attention is widely reflected in internet search activities. The results also demonstrate search data for what keywords best reveal the direction of concern and attention in energy markets.

Keywords: Google Search Activity, Energy Market, Volatility Prediction, Energy Price Volatility

JEL Classification: C53, Q02, Q47