Forecasting monthly LME Copper returns

Abstract

We evaluate if monthly LOCADY returns on the London Metal Exchange can be accurately predicted one, two and three months ahead. In total ten models are constructed using time-varying parameters and bandwidth optimization. The models are evaluated against one another using the following pseudo-out-of sample test statistics: Diebold and Mariano (1995), Clark and West (2006), Giacomini and White (2006) and the Campbell and Thompson (2008) out-of-sample R2. The test statistics generated are inconsistent. A few models are able to generate positive out-of-sample R2 values for one and two month predictions. No model significantly outperforms a random walk for the three step ahead prediction.