Forecasting of Interval-valued Crude Oil Prices with Autoregressive Conditional Interval Models

Crude oil is a highly strategic commodity. This paper investigates the necessity of using interval data and interval econometric models for crude oil price forecasting. Compared to the traditional point-valued data, interval-valued data in a time period contain much more valuable information which is useful for market participant to make decisions. We develop three autoregressive conditional interval forecasting models, and propose three evaluating indices to compare out-of-sample forecasting results. We find that interval information can provide better forecasts than point-valued information, not only for interval prices but also for point-valued ranges of price changes. We also find that economic variables, which are widely used in point-valued forecasting models, such as past crude oil prices, OECD relative industrial petroleum inventories, speculation and OPEC surplus production capacity, still have predictive contents in interval models.