Basis Assets

Abstract

This paper proposes a new method to form basis assets with which to represent investors' opportunity sets and evaluate the goodness-of-fit of asset pricing models. We use return correlations to form a measure of distance between securities, as in Ormerod and Mounfield (2000), and use this distance measure to sort securities into portfolios. We compare the inferences drawn from this set of basis assets with those drawn from other benchmark portfolios commonly used in the literature such as beta and size-sorted portfolios and the Fama and French (1993) size and book-to-market sorted portfolios. Specifically, we compare the results of spanning tests, tests of asset pricing models and inferences about optimal portfolios. The results suggest that the set of basis assets formed by clustering on historical correlations in returns does at least as well as characteristic-based portfolios at generating a proxy for an investor's opportunity set. The proposed set of portfolios also appears capable of generating measures of risk-return trade-off that are estimated with lower error. Finally, we compare the cluster portfolios to principal components extracted from the same sample period, and find that the cluster portfolios generate slightly higher out-of-sample dispersion in mean returns, a slightly lower Sharpe ratio and more stable weights on individual securities.

JEL Classification: G10,G11,G12