Integrating Time Series and Cross-Sectional Signals for Optimal Commodity Portfolios

Abstract

We study the optimal combination of different commodity signals in a dynamic portfolio theoretic framework. Following Brandt et al. (2006, 2009) we parameterize the portfolio weights of a risk-averse mean-variance investor to integrate information from time series predictors and cross-sectional characteristics of commodity assets. We show that the long end of the futures curve as well as open interest have significant timing power, whereas the short end of the futures curve and past returns are relevant characteristics for tilting commodities. Combining timing and tilting strategies outperforms all factor benchmarks out-of-sample and after transaction costs. Moreover, the optimal commodity allocation is not priced by common risk factors, such as commodity carry or momentum.