Can investor attention predict oil prices?

This paper sets out to investigate the predictive power of investor attention onto oil prices. We firstly construct investor attention index by using the Google search volume index (SVI) based on a broad set of words related to oil-related variables and terms that are directly linked to real economy to measure investor attention. Then the empirical work is performed via a novel hybrid approach and WN model (Westerlund and Narayan, 2012, 2014) that account for characteristics of persistency, endogeneity, and heteroskedasticity. The empirical results show that investor attention does exhibit statistically and economically significant in-sample and out-of-sample forecasting power to directly forecast oil prices for both daily data and weekly data. In addition, the results exhibit the term structure character, which are helpful for understanding the financial phenomena that irrational attentions have more effect in short-term decision-making.