Predicting energy futures prices
KeywordsMcCracken and Ng ; Crude oil ; Commodities ; Energy ; Futures ; Heating oil ; Natural gas ; Predictability ; Principal compenents analysis
Forecasting energy commodity prices is of great importance for policymakers, individuals and researchers. Using end-of-month settlement prices of the first three shortest maturity NYMEX energy futures (i.e. WTI Crude Oil, Heating Oil, and Natural Gas) over the period Jan.1990-Dec.2016, this thesis examines whether the evolution of futures log-returns can be predicted across multiple forecast horizons and, if so, by which variables. Based on three alternative linear model specifications, in-sample and out-of-sample point forecasts are generated and evaluated under different performance measures, including the modified Diebold-Mariano Test. The economic model is constructed by means of macroeconomic and financial indicators which have been found to predict the time-varying risk-premia of traditional asset classes (i.e. equities and bonds). Three joint Principal Components (PCs) are also extracted from McCracken and Ng’s (2016) large macroeconomic database and used as potential predictors in a latent factor model. The results are then compared to a univariate autoregressive AR(1) model. While the results provide evidence of significant in-sample predictability under the economic model, the benchmark AR(1) model outperforms both the economic and the PCA models out-of-sample.