Nonlinear multivariate modelling and forecasting of commodity prices
Time series analysis of commodity prices is one of the ongoing developments in relevant empirical studies. The usual research questions are what causes a certain price behavior and what are the consequences. As causes and consequences are sequential events, time is the natural domain of analysis. As such, time series analysis is applied to data collected over extended period of time to investigate the behavior of commodity prices. This study relies on the assumption that the current realization of a commodity price is a function of its own past realizations, as well as past realizations of related commodity prices and other exogenous variables. Additionally, this research assumes that co-evolution of a group of commodity prices may have a nonlinear pattern. That is, the behavior of commodity prices may change conditional on their own past state of nature or on the state of nature of other related variables. Hence, the main premise of this research is a nonlinear multivariate modeling of commodity prices. While a number of different methods can be used to investigate a nonlinear nature of commodity price behavior, this study adopts a smooth transition vector error correction modeling approach to examine nonlinear adjustments of commodity prices to the exogenous shocks. Specifically, Chapter 3 looks at nonlinear dynamics of major vegetable oil prices conditional on the El Ni no Southern Oscillation (ENSO) anomaly. This study is the first attempt to incorporate the effects of ENSO nonlinearities in the world major vegetable oil price dynamics. The findings of the study confirm asymmetries in vegetable price responses to shocks in different ENSO regimes. Chapter 4 examines nonlinearities in a system of exchange rates, crude oil prices and corn prices. The topic is particularly urgent due to recent high fluctuations of commodity prices and increased demand for biofuel intensive agricultural commodities. The findings of this study reveal regime-specific asymmetric responses of crude oil and corn prices to exchange rate shocks. Finally, Chapter 5 applies out-of-sample Granger causality test in linear and nonlinear models of crude oil, gasoline, ethanol and corn futures prices. The results reveal improvement of forecast accuracies from multivariate models for corn and ethanol, but not for crude oil prices. Additionally, while ethanol may have strengthened links between energy and agricultural commodity prices, they are not strong enough in terms of statistical significance.