Diversification When It Hurts? The Joint Distributions of Property and Other Asset Classes

Abstract

Research on the role of real estate in the mixed asset portfolio, whether in the form of direct private holdings or indirect securitised forms of property investment, has focused on the diversification potential of the asset class. In addition to the investment characteristics of real estate (stable base income but growth potential), it has been argued that portfolios containing real estate has superior risk adjusted performance compared to portfolios excluding property. Initial naOve analyses relied on unadjusted appraisal-based ex-post portfolio indices and contemporaneous covariance measures. More recently, there has been greater sophistication in analyses, with careful treatment of lag structures, data transformation, consideration of the uncertainties inherent in a forward-looking expectations framework, use of downside-risk measures and the search for unique priced real estate factors. While the more advanced models add greatly to our understanding of the nature of real estate as an asset class, further extensions are possible. In this paper, we explore whether the relationship between real estate performance and the return distributions of other investment assets is constant over time and cycle. For example, of real estate has a generally low correlation with other assets but exhibits a high correlation in poor economic conditions (that is, strong correlations in the negative tails), then diversification benefits may be illusory or, at best, muted. The investigation of relationships between variables across the whole of their distribution has been an area of considerable technical development in financial economics and econometrics. We will use some of the new techniques available, including the use of copulas and dynamic conditional correlation models, to explore the relationship between UK real estate performance and the other main asset classes. Briefly, copulas model the joint distribution of multiple variables and allow researchers to characterise common movement patterns at different points on the joint distribution surface, while DCC approaches allow for time-varying correlation structures between variables. While such models have been used primarily to analyse high frequency public market data series, they are sufficiently robust to generate results for lower frequency direct property market variables, in both raw and transformed forms. There are many practical applications of such analyses: the aim is to enhance our understanding of the nature of real estate as in asset class.