Time-Varying Correlations of REITs and Implications for Portfolio Management

Abstract

This study uses bivariate dynamic conditional correlations (DCC) to analyze REITs’ relation with stock and bond markets from 1999 to 2018. The results show that the daily DCCs of both Equity REIT and Mortgage REIT returns experienced several structural changes attributed to the state of the economy, levels of leverage, inclusion or exclusion of REITs from the major S&P indices, and REITs getting their own Global Industry Classification Standard (GICS) category. To account for the structural changes, we allow the impact of the macroeconomic driving forces of the DCCs to vary over time. First, we formulate an OLS model using dummy variables regression (DV) to indicate regime membership, using endogenous break-dates. Then, we estimate a Markov regime-switching model (MRS) that allows the impacts of macroeconomic variables to differ during high and low variance regimes. Both complementary regime-sensitive models (DV and MRG) exhibit significant improvement relative to a traditional OLS model. The findings have significant implications for portfolio and risk management. For example, we find that with the new GICS sector, Equity REIT returns decoupled from the Financial Sector and the overall market as measured by the SP 500. These types of correlation shifts can significantly alter optimal portfolio weights whether trying to maximize returns, minimize risk, or achieving the highest risk-adjusted returns.

Keywords:

• Structural changes
• REITs
• portfolio rebalancing
• Markov-regime switching
• dynamic conditional correlations

Notes

1 Graphs of each data series are available from the authors upon request.

2 A table of the DCC-GARCH results are available from the authors upon request but have been omitted here for the sake of brevity. The results confirm the strong persistence of the dynamic conditional correlations between the Equity REIT, Mortgage REIT, Bond, SP500, and SP500 Financials markets. Moreover, the long-run persistence of the conditional correlations is stronger than the short-term persistence.

3 In other models, we use the predetermined breaks identified in Yang et al. (2012). The results are qualitatively similar and are available from the authors upon request.

4 See, for example, Beracha et al. (2019), Freybote and Seagraves (2018), Liow and Ye (2017), Evans and Mueller (2016), Tsolacos et al. (2014), and Anderson et al. (2012).

5 A graph that plots the smoothed probabilities of the DCCs being in the low-variance regime conditioned on the available information at time t-1 is available from the authors upon request.