The paper develops an oil price forecasting technique which is based on the present value model of rational commodity pricing. The approach suggests shifting the forecasting problem to the marginal convenience yield which can be derived from the cost-of-carry relationship. In a recursive out-of-sample analysis, forecast accuracy at horizons within one year is checked by the root mean squared error as well as the mean error and the frequency of a correct direction-of-change prediction. For all criteria employed, the proposed forecasting tool outperforms the approach of using futures prices as direct predictors of future spot prices. Vis-à-vis the random-walk model, it does not significantly improve forecast accuracy but provides valuable statements on the direction of change.Keywords: oil price forecasts, rational commodity pricing, convenience yield, singleequation models.JEL classification: E37; G12, G13, Q40; C22.
Non-Technical SummaryThe paper develops an oil price forecasting technique on the basis of the present value model of rational commodity pricing. The central equation of the theoretical model describes the current spot price of crude oil as the sum of all discounted expected future payoffs received by the owner of one unit of this commodity ("convenience yields"). The discount factor is the sum of the risk-free interest rate and the oil-specific risk premium. The latter compensates for the holder's nondiversifiable risk. Convenience yields are defined as differences between the cost of carry and the futures prices of the commodity. The forecasting problem is moved to the marginal convenience yield because this entity is clearly more predictable than, say, the oil price percentage change directly. The indirect method, however, requires the oil-specific risk premium to be estimated. This is done by a cointegration approach. Market expectations of the marginal convenience yield can be derived from the term structure of the oil market. Alternatively, the marginal convenience yield can be forecast on the basis of autoregressive (AR) models. Combinations between the two approaches are possible, too. Multi-step AR forecasts can be performed by either the plug-in technique or the direction estimation method. Moreover, several information criteria can be applied for model selection issues.The forecast accuracy of the proposed technique is evaluated by out-of-sample projection exercises at horizons up to eleven months. The random-walk model and the approach of using futures prices as direct predictors of future spot prices serve as benchmarks. The sample of Brent oil prices used (i.e. starting in April 1991) is split into an estimation and an evaluation period. The latter comprises the post-January 1997 data in the first and the post-July 2000 data in the second experiment. The root mean squared error is the central evaluation criterion. The mean error and the relative frequency of a correct direction-of-change prediction are also considered. Moreover, statistical hypothesis testsà la Diebold and Maria...