The Dynamics of Capital Flows and Property Returns: A Disaggregated Analysis of Metropolitan Areas and Property Types
by
Jeffrey Fisher, David C. Ling, and Andy Naranjo*
February 2005
*Department of Finance, Insurance and Real Estate
Graduate School of Business Administration
WarringtonCollege of Business
University of Florida
Gainesville, FL32611-7168
Phone: (352) 392-0153
Fax: (352) 392-0301
E-mail:
1
The Dynamics of Capital Flows and Property Returns: ADisaggregated Analysis of Metropolitan Areas and Property Types
by Jeffrey Fisher, David C. Ling, and Andy Naranjo
Paper Summary
We propose to examine the effects of institutional capital flows into privatecommercial real estate markets on property returns disaggregated by metropolitan area and property type. Simultaneously, we will examine the effects of property returns on subsequent investor capital flows into these metropolitan areas and property types. The dynamic relation between disaggregated, private market capital flows and property returns will be estimated using state-of-the-art regression techniques. The influence of interest rates and other national and MSA specific economic variables, such as population growth and employment, will also be incorporated into the dynamic analysis. Unlike static regression techniques, our dynamic model will produce estimates of both the short- and long-run relationships among investor capital flows and returns. For example, if capital flows do impact property returns, we will be able to determine if the effects are concentrated in the subsequent quarter or, instead, incorporated into returns over multiple quarters. In addition, our disaggregated analysis will permit us to examine the extent to which cross-sectional differences among MSAs in, for example, space market supply elasticities affect the responsiveness of returns to increased capital flows.
Research Objective
Despite an emerging general finance literature on capital flows, asset values, and the role of investor sentiment,[1] the role of capital flows in real estate pricing dynamics has received relatively little attention. Nevertheless, the seemingly pervasive view among real estate practitioners is that capital flows, at least in part, predict returns. For example, in March 2004, the Morgan Stanley research team stated that “fund flows are the only thing driving (REIT) returns higher.”[2] Motivated by such practitioner claims and a lack of rigorous research to support them, Ling and Naranjo (2003, 2005) produced the first two rigorous analyses of the dynamics of real estate capital flows and returns. In their first paper (Ling and Naranjo, 2003), the authors examine whether total equity flows into the REIT sector affect aggregate REIT prices and returns and, simultaneously, whether past industry-level returns influence subsequent capital flows into the REIT sector. In their second paper (Ling and Naranjo, 2005), the authors examine the interrelationships and short- and long-run dynamics between capital flows to dedicated REIT mutual funds and industry-level REIT returns. In summary, their work to date has produced strong evidence that capital flows into public (i.e., securitized) real estate markets do not predict subsequent returns, but that returns do affect subsequent capital flows into these securitized real estate markets.
The work of Ling and Naranjo has contributed to our understanding of the dynamics of capital flows and returns in public real estate markets. However, a rigorous analysis of the dynamics of capital flows and returns in private real estate markets has not yet been undertaken. This is a serious void in the literature given that approximately 90 percent of investible commercial real estate in the U.S. is owned and traded in private markets.[3]
The objective of the proposed research is to examine the short- and long-run dynamics among institutional capital flows and property returns in the largest U.S. metropolitan areas. In particular, we will examine whether capital invested in major metropolitan areas bymembers of the National Council of Real Estate Investment Fiduciaries (NCREIF) is associated with higher, or lower, property returns in subsequent periods. Simultaneously, we will examine whether quarterly MSA level returns, as measured by NCREIF, are predictive of future capital investments by NCREIF members in these MSAs. For each metropolitan area, this analysis will first be performed by aggregating MSA-level capital flows and returns across all commercial property types. We will then examine the short- and long-run dynamics among MSA level capital flows and returns disaggregated by the four major property types: office, industrial, retail, and apartment.
The main tool of analysis we propose to employ is a vector autoregressive (VAR) regression model. This regression technique allowsestimation of standard regression coefficients as well as impulse response functions for the variables of interest. These impulse response functions trace out the effects over time that result from changes in capital flows and returns. The application of VAR regression techniques will permit us to consider a number of interesting questions:
1.What is the effect of a change in private capital flows on currentand future property returns,and what are the time series dynamics of this effect? That is, is the effect of a shock to capital flows on returns temporary or is it permanent? If temporary, how quickly does the return effect dissipate?
2.How does the supply elasticity of the MSA affect both the short- and long-run effects of capital flows on returns? Some recent evidence documents that private real estate returns have been significantly higher in supply constrained markets than in markets with more elastic supply responses (i.e., “commodity” markets). However, despite the belief among many practitioners that investments in “constrained” markets are preferred to investments in commodity markets, the proposed research will produce the first careful analysis of this issue.
3.Do past MSA-level returns to institutional investors influence their subsequent capital flows into these markets after controlling for the potential effects of other variables? Institutional real estate investors have been labeled by some observers as “herd” investors. The proposed research will provide extensive documentation on the extent to which institutional capital flows in quarter t are predictive of flows in subsequent quarters.
4.Are thereother variables that affect the dynamic relation between institutional capital flows and property returns at the MSA level? In addition to past returns, the general literature on capital flows suggests other variables such as interest rates and capitalization rates may affect capital flows. We will also experiment with other macroeconomic variables such as MSA level changes in population and employment.
Knowledge of the factors that influence private capital flows into MSAs and various property types,and the effects these flows have on both short-and long-run returns, should prove to be extremely valuable to commercial real estate practitioners in making investment and portfolio allocation decisions.
Research Methodology
The more recent research which investigates the dynamics of capital flows and returns employs structural vector autoregressive (VAR) and vector error-correction (VEC) frameworks that also include additional exogenous variables. In our analysis, we propose to use VAR and VEC models to examine the relationships among institutional investor real estate capital flows and property returns at the aggregate, MSA, and property type levels. The models will also include various national and MSA level control variables such as equity market returns, interest rates, unemployment, and other macroeconomic state variables. By separately examining the aggregate, MSA, and property types, we will be able to disentangle the influences that location and property type have in the conditional dynamics of institutional capital flows and returns.
Several of the variables that we propose to investigate (e.g., institutional real estate capital flows) are likely to be non-stationary.[4] If so, single-equation time-series techniques may result in misleading (i.e., spurious) values of inferential statistics (i.e., t-statistics, R2, and DW) that may erroneously cause one to conclude that a meaningful relation exists among the regression variables. To address these potential non-stationarity problems, we will also examine vector error-correction (VEC) models in our investigation, which are designed for use with non-stationary time series (see, for example, Hamilton, 1994).
To better understand the VEC framework, it is useful to first briefly describe a vector autoregression (VAR) model. In its simplest form, a VAR model is composed of a system of regressions where a set of dependent variables are expressed as linear functions of their own and each other’s lagged values, and possibly some other independent variables. For example, consider the following two-variable, one-period lag VAR model:
Yt = a1 + b1Yt-1 + c1Zt-1 + e1t
Zt = a2 + b2Zt-1 + c2Yt-1 + e2t.
Simultaneous VAR models have proven successful for forecasting systems of interrelated time series variables (see Sims, 1980). Using more technical terminology, a vector autoregression model is the unconstrained reduced form of a dynamic simultaneous equations model. A problem, however, arises with the VAR framework if the variables in the system are non-stationary.
The VEC model is a more appropriate model than the VAR if the variables in the VAR system are integrated of order one or more (i.e., are non-stationary). A group of non-stationary time series is cointegrated if there is a linear combination of them that is stationary. These cointegrating relations are incorporated into the VEC. For example, consider the following two-variable VEC with non-stationary time series:
Yt = a1 + b1Yt-1 + c1Zt-1 + 1(Yt-1 - Zt-1) + e1t
Zt = a2 + b2Zt-1 + c2Yt-1 + 2(Yt-1 - Zt-1) + e2t,
where all terms involving (i.e., first differences) are stationary. This two-variable error correction model is a bivariate VAR in first differences augmented by the error-correction terms 1(Yt-1 - Zt-1) and 2(Yt-1 - Zt-1) from the cointegrating relation. In general, the kth order vector error-correction model can be represented by the following system:
where
Xt = vector of pI(1) variables,
= p x 1 vector of intercepts,
1, 2, k, = p x p matrices of parameters,
et = error term [NID(0,)],
= difference operator,
I(1) = integrated of order one (i.e., first-difference stationary).
In the above system, the coefficient matrix provides information about the long-run relations among the variables, while the ’s provide information on short-run relations. Using Johansen’s (1988) method, we can obtain estimates of the long-run relationships, short-run relationships, impulse response functions, and forecast variance decompositions. The impulse response functions provide the time path of the short-run dynamic relationships from a shock to the variables in the system, while the forecast variance decompositions provide the forecast error variance explained by variations in the variables.
In our analysis, we will estimate the above constrained system or an unconstrained VAR version if the variables in the system are non-stationary and cointegrated. We will estimate the models over the 1978-2004 sample period using quarterly data at the aggregate level (pooled cross-sectional times series) and at the disaggregate level by MSA and property type.
Data
Our sample period will begin in the first quarter of 1978 and end in the fourth quarter of 2004. We are currently gathering aggregate, MSA, and property type (industrial, office, retail, and multi-family) quarterly data from NCREIF. For each of the data types, we are collecting the total return, appreciation return, market value, and implied capital flow. The implied capital flow is based on the extent to which NCREIF members have shifted the relative amount of capital they have invested in different MSAs and property types over time. Each quarter the aggregate market value of properties in the NCREIF Property Index (NPI) in a particular property type and MSA changes for several reasons including: 1) capital appreciation return in the property sector and MSA, 2) net flow of funds into the MSA from acquisitions less dispositions, and 3) properties added to the database from new members. We remove the effect of capital appreciation from the net change in market value so that we only capture the change due to a flow of funds into the property sector and MSA from net acquisitions or properties added by new members. We than calculate the net capital flow on a relative basis by dividing by the total amount of capital in the NPI during the same quarter to adjust for the change in market value over time due to adding new members.
What we are interested in is the changes in the relative amount of capital in each property sector and MSA. For example, properties may be sold in Dallas and properties purchased in San Francisco during a particular quarter. This changes the relative amount of capital in San Francisco versus Dallas. It is this relative change in the propensity to invest in different property sectors and MSAs that we use as a proxy for capital flows from institutional investors. To address the concern that NCREIF’s NPI return data suffer from appraisal smoothing and lagging, we will test the robustness of our results by using NCREIF’s Current Value Index (CVI) in place of the NPI and by experimenting with several unsmoothing techniques.
At the MSA level, we are also gathering available economic variables, including unemployment and other variables. These additional MSA level state variables will allow us to also control for relative regional effects. National level data including stock returns, interest rates, and other fundamental macroeconomic data are available from CRSP and the DRI/Citibank databank, both of which have been purchased by the University of Florida.
References
Cha, H.S. and B. Lee, 2001, “The Market Demand Curve for Common Stocks: Evidence from
Equity Mutual Fund Flows,”Journal of Financial and Quantitative Analysis 36(2): 195-220.
Choe, H., B.C. Kho, and R. Stulz, 1999, “Do Foreign Investors Destabilize Stock Markets? The
Korean Experience in 1997,” Journal of Financial Economics 54(2): 227-264.
Dickey, D. and W. Fuller, 1979, “Distribution of the Estimates for Autoregressive Time
Series with a Unit Root,” Journal of the American Statistical Association, pp. 427-431.
Edelen, R. M. and J. B. Warner, 2001, “Aggregate Price Effects of Institutional Trading: A Study
of Mutual Fund Flow Data and Market Returns,”Journal of Financial Economics 59(2):
195-220.
Edwards, F. R., and X. Zhang, 1998,“Mutual Fund Flows and Stock and Bond Market Stability,”
Journal of Financial Services Research 13(3): 257-282.
Fortune, P, 1998,“Mutual Funds, Part II: Fund Flows and Security Returns,”New England
Economic Review, Federal Reserve Bank of Boston, Jan/Feb: 3-22.
Froot, K.A., P.G.J. O’Connell, and M.S Seasholes, 2001, “The Portfolio Flows of International
Investors,” Journal of Financial Economics 59(2): 151-194.
Hamilton, J. D., 1994, Time-series Analysis, Princeton Press.
Johansen, S., 1988, “Statistical Analysis of Cointegrating Vectors,” Journal of Economic
Dynamics and Control, pp. 231-254.
Karceski, J., 2002,“Returns-Chasing Behavior, Mutual Funds, and Beta’s Death,”Journal ofFinancial
and Quantitative Analysis 37(3): 559-599.
Ling, D. C. and W. Archer, Real Estate Principles: A Value Approach, McGraw-Hill Irwin, Burr Ridge
Illinois, 2005.
Ling, D.C. and A. Naranjo, 2003,“The Dynamics of REIT Capital Flows and Returns,”Real Estate
Economics 31(3): 405-436.
Ling, D.C. and A. Naranjo, 2005,“Dedicated REIT Mutual Fund Flows and REIT Performance,”
Journal of Real Estate Finance and Economics, forthcoming.
Remolona, E.M., P. Kleiman, and D. Gruenstein, 1997,“Market Returns and Mutual Fund Flows,”
Federal Reserve Bank of New York Policy Review 3(2): 33-52.
Sims, C., 1980, “Macroeconomics and Reality,” Econometrica, pp. 1-49.
Warther, V.A., 1995, “Aggregate Mutual Fund Flows and Security Returns,” Journal of Financial Economics 39, pp. 209-235.
Warther, V.A., 1998,“Has the Rise of Mutual Funds Increased Market Instability?” Brookings Papers on
Financial Services, The Brookings Institution.
1
[1]See for example, Cha and Lee (2001), Choe, Kho and Stulz (1999), Edelen and Warner (1999), Edwards and Zhang (1998), Fortune (1998), Froot, O’Connell and Seasholes (2001), Karceski (2003), Remolona et al (1997), and Warther (1995, 1998).
[2]“Will Fund Flows Buoy Valuations,” Real Estate Investment Trusts Industry Report, Morgan Stanley REIT Team, March 2004.
[3]See Ling and Archer (2005), Chapter 15, page 387.
[4] In general, a series is non-stationary if its mean, autocovariances, or other higher moments are time dependent. For example, if the mean of a series varies with respect to time, it is likely to be non-stationary. Simply stated, the test for a unit root (i.e., non-stationarity) in a time-series is the test that a regression of a series on itself lagged one period yields a coefficient of one. This test is complicated by several features arising from the non-stationarity of the series under the null hypothesis. In a well-known paper, Dickey and Fuller (1979) suggest a method for computing a test for a unit root in a time-series.