Measuring Asymmetric Correlations for Portfolio Diversification

Summary

This review challenges the assumption that diversification works equally in rising and falling markets. It argues that the useful question is how a prospective diversifier behaves when a portfolio’s main growth asset falls. To measure downside correlation, it recommends conditioning on the return of that main asset alone, rather than requiring both assets to fall; the latter excludes cases where one asset offsets the other’s loss. It also explains that conditional correlations can change mechanically even under a bivariate normal distribution, so observed values should be compared with the values expected under that distribution.

An empirical analysis of six major asset classes finds differing asymmetric relationships, including historically favorable behavior from government bonds in several pairings. The review presents global optimization with a kinked utility function as a way to account for asymmetric returns and investor loss aversion when selecting portfolio weights. The evidence is historical, and the article cautions that bond behavior may differ when interest rates are near zero. Its conclusions depend on the sample and the specified utility preferences; they are not a guarantee that diversification will protect against future losses.

Key ideas

• Downside correlation should be estimated by conditioning on the portfolio’s main growth asset, allowing the other asset to offset its losses.
• Requiring both assets to fall omits successful diversification episodes and can distort estimates.
• Conditional correlations can shift for mathematical reasons, so empirical estimates should be compared with a distribution-based benchmark.
• The review reports asymmetric correlations across six asset classes and describes government bonds as historically favorable diversifiers in several pairings.
• Global optimization with a loss-sensitive utility function can account for return asymmetry when choosing portfolio weights.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.