Modeling Nonlinear Price Relationships in Commodity Markets

Three essays investigate the nonlinear price relationships in commodity markets. The unifying theme of all three essays is that time series econometric models, especially nonlinear time series models, are used.;The first paper presents an empirical analysis of the effect of exchange rate shocks on import and export prices in the forest industry. In particular, exchange rate pass-through for four important tropical timber commodities are considered: sawnwood (hard and soft), plywood, spruce lumber, and logs (hard and soft) for Africa, Southeast Asia and Japan to United States using some linear and non-linear regression approaches taking into account the structural changes.;A nonlinear smooth transition regression (STR) (Terasvirta 1994, 1998) model, which can be viewed as a generalization of threshold models with a continuous transition function that allows for smooth changes during the transition period rather than discrete changes, is considered. Results suggest evidence for the convenience of the STAR type models (SETAR and LSTAR) to model deviations from LOP in a nonlinear fashion for tropical forest product markets. Reasonable estimates of the threshold values that may be a representation of transaction costs in line with the theoretical arguments in international trade were found. It was also observed that the values of threshold variables greatly vary across different countries.;The second essay provides the empirical results for spatial price dynamics in soybean and corn markets in accordance with the theory of semi-parametric Vector Error Correction Generalized Autoregressive (VECGAM) models and standard VAR models for three North Carolina terminal markets taking into account the nonlinearity of the time trend component. Overall results indicate that the non-parametric drifts coincide with the general price movements, and when compared with the standard VAR results, the addition of nonparametric mean shift affects the overall implication of impulse-responses in a way that the VECGAM model impulse responses tend to imply a smaller degree of reaction towards the shocks and exhibit shorter time of adjustment for the convergence into a stable equilibrium level. Also, the number of significant impulse response coefficients under VECGAM models is larger compared to the standard VAR impulse responses. Responses confirm integration of markets in both VAR and VECGAM models.;The third paper's objective is to investigate the potential of a time series analysis technique, namely the Time Varying Parameter Vector Autoregressive Model (TVPVAR) technique, in the development of daily forecasting models for cattle prices in the presence of structural changes by using a Bayesian approach. More specific objectives include integrating smoothing techniques and stochastic volatility into TVPVAR modeling framework based exclusively on time series for cash-cattle prices and comparing the accuracy of the forecasting performance of this model with the standard VAR model that ignores time variation in parameters and possible time variation in the variance-covariance matrix of disturbance terms to determine if the inclusion of the time varying component into the conventional VAR structure. Another purpose was to extend the TVPVAR to include stochastic volatility improves the forecast results.;One of the main conclusions from this paper is that letting time variation in VAR models improves the forecast performance. Overall, taking into account expectations about the behavior of cross-the-impulse responses, the process of reaching pre-shock levels and the MNG test results suggest that the nonlinear TVPVAR model is an improvement over the standard VAR forecast for 1 month, 3 months and 9 months ahead horizons for most of the time. Also, the main results are partially consistent with the literature on cattle prices.