Forecasting Interval-valued Crude Oil Prices via Autoregressive Conditional Interval Models

We propose two parsimonious autoregressive conditional interval-valued (ACI) models to forecast crude oil prices. The ACI models are a new class of time series models proposed by Han et al. (2009). They can characterize the dynamics of economic variables in both level and range of variation in a unified framework and hence facilitate informative economic analysis. A minimum DK-distance estimation method can also simultaneously utilize rich information of level and range contained in interval-valued observations, thus enhancing parameter estimation efficiency and model forecasting ability. Compared to other existing methods, the ACI models deliver significantly better out-ofsample forecasts, not only for interval-valued prices but also for point-valued highs, lows, and ranges. In particular, we find that the oil price range information is more valuable than the oil price level information in forecasting crude oil prices, which is consistent with observed facts of price movements in crude oil markets. We also find that speculation has predictive power for oil prices in our interval framework..