Testing for Spanning with Futrures Contracts and Nontraded Assets : A General Approach

This paper generalizes the notion of mean-variance spanning as de- ned in the seminal paper of Huberman & Kandel (1987) in three di- mensions.It is shown how regression techniques can be used to test for spanning for more general classes of utility functions, in case some as- sets are nontraded, and in case some of the assets are zero-investment securities such as futures contracts.We then implement these tech- niques to test whether a basic set of three international stock indices, the S&P 500, the FAZ (Germany), and the FTSE (UK), span a set of commodity and currency futures contracts.Depending on whether mean-variance, logarithmic, or power utility functions are considered, the hypothesis of spanning can be rejected for most futures contracts considered.If an investor has a position in a nontraded commodity, then the hypothesis of spanning can almost always be rejected for fu- tures contracts on that commodity for all utility functions considered.For currency futures this is only the case for a power utility function that re ects a preference for skewness.Finally, if we explicitly take into account net futures positions of large traders that are known to have predictive power for futures returns, the hypothesis of spanning can be rejected for most futures contracts.