Risk-Adjusted Performance of Real Estate Stocks: Evidence from Developing Markets

Journal of Real Estate Research, The,  Oct-Dec 2004  by Ooi, Joseph T L,  Liow, Kim-Hiang

• E-mail
• Print
• Link

Debt ratio has also been found to significantly explain cross-sectional variations in common stock returns (Bhandari, 1988; and Fama and French, 1992). There is surprisingly scarce evidence on the direct relationship between financial leverage and risk-adjusted returns of real estate securities. Chan, Henderson and Saunders (1990) observe that highly levered equity REITs are more sensitive to macroeconomic factors than moderately levered REITs. This observation highlights the need to control for leverage when evaluating the relationship between macroeconomic factors and real estate returns.

The empirical evidence is not clear whether dividend yield plays any role in explaining REIT returns. While Sanders (1997) found the coefficient on his dividend variable to be negative and highly significant in a simplified model, its explanatory power ceased to be significant once a more well-defined asset pricing model was employed.

Another fundamental factor that may explain cross-sectional returns is the portfolio characteristics of individual REITs. Studies by Howe and Shilling (1990), Redman and Manakyan (1995), Gyourko and Nelling (1996) and Chen and Peiser (1999) conclude that property type specialization has a significant impact on individual REIT returns and risks. The impact of geographical concentration is, however, less conclusive. On the one hand, Gyourko and Nelling (1996) and Ambrose, Ehrlich, Hughes and Wachter (2000) conclude that diversification strategy by geographical regions has no significant benefit on REIT value. Chen and Peiser (1999), on the other hand, find geographical concentration has a positive impact on individual REIT returns.2

Regression Models

The panel regression employed to explain the risk-adjusted returns of real estate securities is specified as follows:

with the i and t subscripts denoting the cross-sectional and time-series dimensions respectively. The dependent variable, S^sub i^, is the Sharpe index of the individual firms. X^sub it^ is the predetermined vector of firm-specific attributes, while M^sub t^ represents the time-variant macroeconomic factors, α is a scalar, while β and γ are column matrices of the partial regression coefficients for the explanatory variables to be estimated.

The error term, u^sub it^, may be further specified as: u^sub it^ = μ^sub i^ + v^sub it^; where μ^sub i^ accounts for any unobservable firm- or country-specific effect that is not included in the regression model, and v^sub it^ represents the remaining disturbances in the regression, which varies with individual firm and time. In the estimation model, μ^sub i^ is fixed for each company over the study period.3 This represents the effects of omitted variables unique to each company that stay constant over time. An obvious way to estimate the model is to introduce dummy variables into the regression model. The least squares dummy variable model may be specified as:

where α^sub i^ is the unique intercept for the individual ith firm in the sample. The fixed-effects model provides a common set of partial regression coefficients for the explanatory variables while allowing a different intercept for each of the cross-sectional units.