The Multiple Dimensions of Asset Allocation: Countries, Sectors or Factors?
Anne-Sophie E. Vanroyen
Columbia Management Group
State Street Associates
Asset allocation has been performed traditionally along country lines. If the balance is shifting towards sectors, then skill in forecasting relative country returns may not be sufficient to ensure investment success. It may lead to suboptimal portfolios. There is growing evidence supporting the emergence of global sectors. We examine this claim, focusing on 19 developed equity markets between 1994 and 2000. We first identify clusters of sectors across countries, using a methodology based on neural networks. Although country stratification remains important, our results suggest a more complex and dual structure across both dimensions. Some clusters correspond to global sectors across regions. In contrast with Heston and Rouwenhorst (1994) seminal model and subsequent research, we develop a framework that allows a large degree of interaction between countries and sectors. We also relax the assumption that country and sector exposures are fixed, thus enabling us to analyze how the two dimensions evolved over time. We perform principal components analysis to identify factors driving returns. By construction, the factors combine both dimensions, geographic and industrial, and are allowed to vary over time. We measure the relative importance of country and sector effects in these factors, and find that sectors have become as important as countries since October 2000. We discuss the implications of these findings for asset allocation. Factors are interpreted as combinations of a limited number of short and long positions. We find that diversification across factors leads to lower risk than diversification across countries or sectors.
Number of Pages in PDF File: 36
Keywords: asset allocation, sector diversification
JEL Classification: G11, G15, C45
Date posted: January 16, 2002
© 2016 Social Science Electronic Publishing, Inc. All Rights Reserved.
This page was processed by apollobot1 in 2.485 seconds