Abstract

During the 1970's, the increase in the price of oil and many other commodities dominated macroeconomic discussions. In the late 1980's and early 1990's, commodity prices generally declined and not much attention was given to the topic. Currently, the surge in the price of oil and many other commodities, both in nominal and real terms, drew back attention to the issue.
This thesis follows the empirical approach in modelling nonlinear behavior of commodity prices. The approach was motivated by the observation that commodity prices tend to move together in groups in response to a common macroeconomic variable or group of variables. The thesis attempts to explain this phenomenon by, first, classifying commodity prices according to their recorded border prices (an issue that has been ignored in previous studies), and then by trying to find the best transition (threshold) variable that can explain this common dynamic in each group.
The observed nonlinearity in commodity prices is modelled using the smooth transition regression (STR) model with external threshold variables. The use of external threshold variables, in addition to the commonly used autoregressive lags of the dependent variable, is a theme that distinguishes this thesis from the majority of the studies in the regime switching literature. The STR model, which technically models regime switching in the conditional mean equation of the data generating process, is extended to model regime switching in the conditional variance. Both models (the STR in mean and the STR in variance) were fitted to the Grilli & Yang's (1988) commodity price index and to the individual price series forming the index.
Two external transition variables were found successful in capturing the regime switching dynamics of the commodity price index: inflation rate and the price of oil. Using both transition variables in the STR in mean and the STR in variance models, both models displayed the same dynamics in the limiting processes of the commodity price index. This result suggests that both models can be seen as substitutes when modelling nonlinearity in the commodity price index. As for the two transition variables, inflation was capable of capturing the early dynamics (between 1900 and 1950) of the commodity index whereas oil price captured the late ones (between 1980 and 2007). This result motivates the use of external threshold variables in regime switching models in general and, in particular, the use of inflation and oil price in the STR model when applied to an index of commodity prices.
More insight about the co-movement of commodity prices is gained by studying the individual price processes forming the index. However, it is worth noting that there is no single variable that is capable of explaining the behavior of all commodity prices. The way each price series is recorded and the history of the commodity's major exporter and importer is crucial in styding and modelling its dynamics.