Spatial Correlation in Expected Returns in Commercial Real Estate Markets and the Role of Core Markets

Abstract

This study is based on survey data on investor expectations for 40 metropolitan statistical areas (MSAs) in the US over the period from 2003:Q2 to 2014:Q2. The paper has two main objectives. These are firstly to identify whether expected rates of return across different commercial real estate markets are positively spatially correlated and secondly to analyze the role of core markets like New York, Washington, DC, Los Angeles, San Francisco, and Chicago in the information-spreading process. All the tests conducted are conditional on the maintained hypothesis that expected returns across different markets are spatially uncorrelated. To carry out these tests, we regress a measure of excess return in each of the 40 markets on (contemporaneous) measures of expected returns in other markets, controlling for the interaction between (neighboring) markets. We find very large spatial interaction coefficients across different markets. Our basic estimates yield a spatial correlation coefficient of near one and statistically significant at standard levels. Chow tests allow rejection of the hypothesis that core markets like New York, Washington, DC, Los Angeles, San Francisco, and Chicago are no different than markets like Cleveland, Columbus, Orlando, Pittsburgh, Salt Lake City, St. Louis, and Tampa in explaining the covariances of the rates of return among different commercial real estate markets. Tests of structural change also imply that core markets currently play a more prominent role that in the past in the transfer of information from one market to another.

Appendix

This appendix describes the method used to identify structural breaks in the period under investigation. Since we do not explicitly know the dates of possible disturbances, we follow the same methodology as in Ferreira and Gyourko (2011) to identify the existence of these structural breaks without imposing economic priors. In the case of Ferreira and Gyourko (2011), they use the methodology expressed concisely in equation (A1) to gain insight into the timeline of the U.S. housing boom over the 2000s. The relevant computations in Ferreira and Gyourko (2011) are carried out using house price growth rate data both at the MSA level and at the neighborhood level from 1993 to 2009. Their results show considerable heterogeneity in terms of when the housing boom began. Using aggregated MSA-level data, Ferreira and Gyourko (2011) find major jumps in price appreciation rates as early as 1997 in some MSAs and as late as 2005 in others, with only a modestly higher concentration of breakpoints around 2004–2005. Using activity in groups of 2 or 3 contiguous census tracts to define a neighborhood, their neighborhood analysis also shows similar heterogeneity in timing, with booms beginning as early as 1996 and as late as 2006.

Our approach differs from that in Ferreira and Gyourko (2011) in the following important respect. We use the methodology to identify structural breaks in the U.S. commercial real estate market. All of our data are at the MSA level. Our sample includes expected rent price growth rates for apartment buildings, CBD office buildings, industrial warehouse properties, and regional malls over the period 2003 Q2 to 2014 Q2. The empirical strategy to identify these structural breakpoints can be described with the following regression model:

$$ \left({R}_{n,c,p,t}/{R}_{0,c,p,t}\right)={\alpha}_{c,p}+{\delta}_{c,p}1\left[{q}_{c,p,t}\le {q}_{c,p,t}^{*}\right]+{\epsilon}_{c,p,t} $$

(A1)

where (R _{n , c , p , t} /R _{0 , c , p , t} ) is the quarterly expected rent price growth rate computed from the equation (1) in the text, δ _c , p estimates the importance of the potential breakpoint, q _{c , p , t} is a quarter (with τ _{c , p , 2004} < q _{c , p , t}  < τ _{c , p , 2011}), ${q}_{c,p,t}^{*}$ is the location of the potential structural break, τ _{c , p , 2004} is the first quarter of 2004, and τ _{c , p , 2011} is the fourth quarter of 2011. The subcripts are city (c), property type (p), and quarter (t).

Equation (A1) is essentially a steady state equilibrium model. The model assumes a stable market structure in which rent growth (which may be positive, negative, or zero) is constant over time. A statistically significant structural break occurs in this growth process at that quarter which maximizes the explained variance of the dependent variable. The identification of this breakpoint is achieved recursively. Equation (A1) is estimated repeatedly over the peroid τ _{c , p , 2004} to τ _{c , p , 2011}, each time by adding one observation at a time. The resulting 32 estimates of the intercept term are then investigated to find if there is a quarter in which the change in the rent price growth series has its greatest impact in explaining the rent price growth series itself.

To get a sense of whether return expectations in different markets are more spatially correlated in crisis periods than in boom periods, the idea is to use these estimated structural break dates to work out if the spatial correlation coefficients in equation (8) increase or decrease during crisis periods. Some caution is warranted in interpreting the estimates in the text, however, since the parameters of the model are estimated over a period when rents were essentailly falling as a result of the great financial crisis and so we do not have data over a full cycle.