Author:

Abstract:

This paper revisits the asymmetric effect of the basis on commodity spot and futures price volatilities documented by Kogan, Livdan and Yaron (2008) and Lien and Yang (2008). Kogan et al. (2008) show both theoretically and empirically that, for a non-storable consumption good, the relationship between commodity price volatility and the basis exhibits a V-shape. Lien and Yang (2008) illustrate the existence of an asymmetric effect of the basis on commodity price volatilities for storable commodities. Their results seem to imply that both scarcity and abundance increase spot and futures price volatility, a counter-intuitive result. The aim of this article is twofold: (i) test the presence and the robustness of the asymmetric effect for storable agricultural commodities by analyzing different sample periods, longer horizons and alternative utility functions; and - given that this asymmetric effect turns out not to be robust - (ii) explore new variables besides the basis to proxy for scarcity, analyze whether they exhibit an asymmetric effect and test their performance in modeling storable commodity price volatility and in hedging futures positions. Our results provide little support for a V-shaped relationship between the basis and storable agricultural commodity price volatilities. Though an asymmetric effect is present in that the size of the coe±cient for a positive basis is much larger than for a negative basis, a negative basis does not lead to higher volatilities. Moreover, we find that the strong hedging performance documented by Lien and Yang (2008) when including the asymmetric basis in the volatility specification is not robust across sample periods, for longer hedging horizons and for alternative utility functions. More positively, though, our results indicate that alternative scarcity specifications do have the expected positive link with volatility and often outperform more simple models in terms of hedging performance. Unfortunately, no single variable consistently leads to better results out-of-sample and there is often no correspondence between the best performing model in- and out-of-sample.