Abstract:
We analyze the ability of economic and financial uncertainties in predicting movements in commodity futures markets. Using daily data over the period of 8th May 1992 to 31st August 2016 on 21 commodity futures covering agriculture, energy, metals and livestock, we find that: (a) Linear predictive tests provide virtually no evidence of predictability; (b) Linear models are misspecified due to nonlinearity and hence, results from the framework cannot be relied upon, and; (c) Using a k-th order nonparametric causality-in-quantiles test, which is robust to misspecification in the presence of nonlinearities, we find evidence that measures of uncertainty can predict returns and/or volatility of as many as 20 of the commodities considered at least at one point of their respective conditional distributions for returns and variance. In general, we highlight the importance of modeling nonlinearity, higher order moments, and quantiles of returns and volatility when carrying out predictability analysis involving commodity futures and uncertainty.
