Forecasting Crude Oil Prices Based on An Internet Search Driven Model

Internet search data, which is mainly treated as a representative index of investor attention, has been widely used in forecasting commodity markets. However, few papers discuss the role of search activity data as a real time instrument to quantify other complex market factors, such as weather, geopolitical events and macroeconomic conditions. In this paper, we use an internet search driven model to forecast crude oil prices. With a start of fifty keywords relating to crude oil markets, factors search volume indices (FSVI) are generated by Google Trends and then a small set of FSVI with best predictability are selected by filtering algorithms. Following the filtering, initial forecasts for crude oil prices are generated by estimating ARMAX model with each selected FSVI. Finally, a time varying parameter combined model is used to combine these initial forecasts to final forecasts by the accuracy. Experimental results show that the forecasting model with search data has better accuracy.