Tese Mestrado
Predicting Commodity Turning Points with Mixed Causal-Noncausal Models and Exogenous Variables
Inês Pires e Castro
Commodity prices are often subject to substantial fluctuations and periods of explosive price dynamics, potentially associated with speculative bubbles and subsequent sharp turning points. Mixed causal-noncausal autoregressive (MAR) models provide a framework for capturing such dynamics through forward- and backward-looking components, while their extension with exogenous variables (MARX) allows additional economic information to be incorporated.
This thesis investigates whether the shape of predictive densities generated by MAR and MARX models can provide information about bubble-related dynamics in commodity markets. A predictive density bimodality measure is developed to quantify the extent to which one-stepahead predictive densities exhibit two distinct modes. The proposed measure is examined through simulation experiments under different MAR and MARX specifications and innovation distributions.
The empirical analysis focuses on three commodities: West Texas Intermediate (WTI) crude oil, Henry Hub natural gas, and nickel. MAR(1,1) and MARX(1,1,q) models are compared to assess how the inclusion of exogenous variables affects parameter estimation, predictive densities, and bimodality. The bimodality measure is further compared with the Phillips-Shi-Yu (PSY) procedure for detecting and dating explosive episodes. The results show that exogenous regressors improve model fit for WTI and nickel, but not natural gas.
Positive bimodality coincides with several PSY-detected explosive episodes, although correspondence varies across commodities and bimodality also occurs outside these periods. Including exogenous information generally maintains or improves this correspondence. Thus, bimodality offers a complementary perspective on explosive dynamics, but its ability to anticipate turning points in real time remains unverified.