Institutional Investors' Preferences for REIT Stocks

2002 V30 4: pp. 567–593

REAL ESTATE

ECONOMICS

Institutional Investors’ Preferencesfor REIT StocksBrian A. Ciochetti,∗ Timothy M. Craft∗∗ and James D. Shilling∗∗∗

This article investigates the determinants of real estate investment trusts (REIT)portfolio investment and institutional REIT ownership using multivariate Tobitregressions. We contend that many institutional investors take larger positionsin more liquid assets like REIT stocks, as compared with private real estateequities, because of liquidity considerations. Consistent with this contention, wefind that liquidity constraints are significantly related to REIT portfolio investmentby institutional investors. We also find that institutional investors have differentpreferences for REIT stocks than do other investors; they generally prefer larger,more liquid REIT stocks.

Our primary interest in this article is in exploring what determines why someinstitutional investors invest in real estate investment trusts (REIT) stocks andothers invest directly in private real estate equities. Is it because institutionsdiffer in their ability or desire to invest in illiquid assets as well as in theirtolerance for overall risk? Or is it because some institutions are simply toosmall to devote significant resources to investing directly in private real estateequities? Our secondary purpose is to provide evidence on whether institutionsprefer liquid REIT stocks more than other investors do.

To test these hypotheses, this article draws on data from the Securities andExchange Commission (SEC) filings required of all institutions with $100 mil-lion under discretionary management that provide detailed information on in-stitutional ownership of REIT shares by manager type and allow matchingwith investor-investment manager data from Money Market Directories, Inc.(MMD). We then use total dollars under management to prorate the ownershipof REIT shares by manager type to the large sample of pension plans reportedin MMD. Fortunately, we generally find close correspondence between actualand prorated REIT ownership levels for the small subset of pension plans thatreport their actual REIT holdings to MMD. This close correspondence gives usencouragement that our matching strategy is reasonable.

∗University of North Carolina, Chapel Hill, NC 27599 or [email protected].∗∗Wichita State University, Wichita, KS 67260-0077 or [email protected].

∗∗∗University of Wisconsin, Madison, WI 53705 or [email protected].

568 Ciochetti, Craft and Shilling

Next, to determine whether liquidity or size considerations contribute to thecross-sectional differences in institutional investors’ propensities to invest inREIT securities, we regress actual REIT holdings (as a share of total realestate investments) on liquidity factors, size factors and institutional factors.We observe the holdings of REIT stocks for each institutional investor. Sincethese holdings are censored at zero for about one-twelfth of the observations,a censored regression is necessary. The model we estimate is a Tobit model inlogarithms (corrected for heteroskedasticity).

The model assumes that institutional investors endeavor to minimize risk fortheir plan sponsors by attempting to match asset and liability structures, subjectto a liquidity constraint. Moreover, the model assumes that there is heterogene-ity among institutional investors regarding liquidity needs. Some institutionalinvestors are assumed to be liquidity constrained, while others are not. Themodel also assumes that there is a ready (and highly organized) market forsome assets, but not for others. As a consequence, liquidity-constrained insti-tutions prefer to hold liquid assets (like REIT stocks) at low levels of wealthand illiquid assets (like private real estate equities) at high levels of wealth.

Our empirical results show that liquidity-constrained institutional investors havea strong preference for liquid REIT shares, as compared with private real estateequities.1 Our findings also show that institutional investors have different pref-erences for REIT stocks than do other investors; they generally prefer larger,more liquid REIT stocks. We interpret these results as evidence in favor of ourliquidity-constrained asset–liability portfolio choice model.

The rest of the article is organized as follows. The next section discusses thetheoretical model. The third section examines which institutions own REITshares. In this section, we compile the statistics of relative REIT stock holdingsby institutional investors, relative REIT stock holdings meaning institutionalinvestments in REIT shares expressed as a percentage of total assets (and/orexpressed as a percentage of total real estate assets). The first impression givenby these data is that relative REIT stocks holdings are quite small, generallyless than 1% of total assets, on average, and between 10 and 20% of total realestate assets for those institutions that invest in both REIT shares and real estateassets. The fourth section examines whether institutions invest in the same REITstocks as everyone else. A summary concludes the article.

1 Until recently, very few studies have dealt explicitly with institutional investors’ pref-erences for REIT stocks. The notable exceptions are Chan, Leung and Wang (1998),Graff and Young (1997) and Downs (1998), and none has considered the investmentdemand for REIT stocks by individual pension funds.

Institutional Investors’ Preferences for REIT Stocks 569

Modeling the Demand for REITs with and without Liquidity Constraints

We consider an environment in which institutional investors are called on tomanage a portfolio so as to meet a set of liability payments over time. Wefurther assume that these liability payments are stochastic. This assumption ismade to capture the fact that many things can happen to a firm’s employees. Forexample, some employees may voluntarily leave the company before normalretirement age. Others may be fired before normal retirement age. Still othersmay die prematurely. Intuitively, these assumptions mean that the pension planmanager will need to have assets of at least equal value to the present value ofexpected benefits for past services on hand at all points in time to meet potentialliquidity needs.

In this environment, the decision to invest in REIT stocks is explained byliquidity and transaction-cost motives. We assume that investors want portfolioswith greater mean surplus return (equal to assets minus liabilities) and lowersurplus return variance. We also assume that institutional investors are subjectto liquidity constraints, some more so than others. Additionally, the modelassumes a one-period decision horizon.2

The institutional investor’s portfolio problem is to

Max ψ =N∑

i=1

Ri Xi − ρF L X F L − ρC L XC L , (1)

subject to

P

{N∑

i=1

(1 + Ri − Ci )Xi − (1 + ρF L )X F L − ρC L XC L ≤ XC L

}≤ α, (2)

N∑i=1

Xi = 1, (3)

Xi ≥ 0 i = 1, . . . , N , (4)

X F L + XC L = XT L , (5)

2 The model is an extension of Chun, Ciochetti and Shilling (2000). In that work weadvance a formal model suggesting that the solution to the portfolio choice problemwhere the fund manager allocates assets to maximize the risk-adjusted future surplusvalue can be very different from the solution to the portfolio choice problem wherethe fund manager allocates assets to minimize the variance of a portfolio for a givenvalue for the portfolio mean. The main contribution of this article is to show how someinstitutional investors are compelled by their circumstances to invest in REIT stockswhereas others seek primarily to invest in direct private real estate equities.


where Xi is the portfolio weight of each asset i, Ri is the expected return on eachasset i, Ci is the liquidation cost of asset i, and P is the probability operator.To be consistent with the asset portfolio weights, XC L and X F L are definedas the ratio of current and future liabilities to assets. The sum of current andfuture liabilities is the total pension liability, XT L , which is the reciprocal ofthe funding ratio. For example, if assets were 103% of pension liabilities, thefunding ratio would be 1.03 and XT L would be 0.97. Values of ρC L and ρF L

are the expected annual change in current and future liabilities per employee,respectively. We assume no short selling or riskless borrowing or lending.

With the further assumption that asset returns and pension liabilities are dis-tributed normally, the deterministic equivalent of (2) becomes

N∑i=1

(1 + Ri − Ci )Xi − (1 + ρF L )X F L − ρC L XC L + K (α)[(X ′�X )

+ X2F Lσ 2

F L + X2C Lσ 2

C L − 2(X ′X F L ) − 2(X ′XC L )]1/2 ≥ XC L , (6)

where K(α) is the inverse of the standard normal cumulative distribution, X ′ isthe vector of asset weights, and � is the covariance matrix of the assets. Withregard to the pension liabilities, is the vector of covariances between futureliabilities and assets, while is the vector of covariances between changes incurrent liabilities and asset returns. The first term on the left-hand side (LHS)of (6),

N∑i=1

(1 + Ri − Ci )Xi − (1 + ρF L )X F L − ρC L XC L ,

represents the potential liquidation value of the portfolio (net of any liquidationcosts) minus the expected future benefits and the return on current liabilities.The second term on the right-hand side (RHS) of (6) is some number, K(α),times the standard deviation of the surplus return. For plausible values of α, (6)is strictly convex, so that the model consisting of (1), (3), (4), (5) and (6) is anonlinear program with a unique optimum.

To solve for the optimal value of X ′ = (Xi , . . . , Xn), we can use numericalmethods in the spirit of Markowitz (1959). The solution proceeds in two steps.First, we use historical values for Ri , ρF L , ρC L , �, and . Second, valuesof X ′ = (X1, . . . , Xn) are computed for alternative values of XC L and X F L ,conditional on XT L . We first select a value for XC L (which we normalize bydividing by XT L ). Initially, we assume XC L = 0. We then compute the optimalvalue of X ′ = (X1, . . . , Xn), which comes from maximizing (1) subject to (6),


(3), (4) and (5). We then increment the value of XC L by 0.10 for each subsequentcomputation, all the way up to a maximum of 1.00, and proceed to computealternative values of X ′ = (X1, . . . , Xn).

Three special cases of the model are discussed below. The first case assumes bothXC L = 0 and X F L = 0, and is the easiest case to discuss. This case corresponds tothe standard Markowitz mean-variance model. With our assumption that XC L =0, no liquidity motive arises for asset holdings. The second case assumes XC L =0. This case corresponds to the stereotypical asset–liability model. In this case,investors take into account the joint impact of asset and liability risks. In thethird case, we assume both XC > 0 and X F L > 0. The investments favored inthis case are not simply those that are highly correlated with pension liabilities,but those that are liquid as well. This concern over liquidity, as we will see,helps to explain why some institutional investors decide to invest in REIT stocksrather than illiquid real estate assets.

Case 1: Asset-Only Framework

With XC L = 0 and X F L = 0, liability considerations are inconsequential. Hence,the constraint in (2) becomes

P

{N∑

i=1

(1 + Ri − Ci )Xi ≤ 0

}≤ α, (7)

and the deterministic version of (7) becomes

N∑i=1

(1 + Ri − Ci )Xi + K (α)(X ′�X )1/2 ≥ 0. (8)

For X ′ = (X1, . . . , Xn) to be a solution to the maximization problem describedby (1), (8), (3), (4) and (5), it is necessary for X ′ = (X1, . . . , Xn) to yield thehighest possible

∑Ni=1 (1 + Ri − Ci )Xi for any given value of K (α)(X ′�X )1/2.

But here (X ′�X )1/2 is the standard deviation of the portfolio. Accordingly, if∑Ni=1 (1 + Ri − Ci )Xi is to be at a maximum for any given K (α)(X ′�X )1/2,

then∑N

i=1 (1 + Ri − Ci )Xi must also be at a maximum for any given level ofX ′ �X. Consequently, any portfolio which gives the highest possible expected(net) return for a given value of K (α)(X ′�X )1/2 must also give the highestpossible (net) return for a given value of X ′ �X.

It follows that, with XC L = 0 and X F L = 0, the optimal value of X ′ =(X1, . . . , Xn) is


X =

X1...

Xn

= �−1 A

A′�−1 A+

[I − �−1 AA′

A′�−1 A

]�−1γ, (9)

where

µ =

R1 − C1...

Rn − Cn

,

and where 1′ = (1, . . . , 1), A = [µ 1], δ = [k 1]′, k is a given value for theportfolio surplus return and γ ′ = (γ1, . . . , γn) is a Lagrange multiplier.3 AssumeXiγi = 0, Xi = 0 and γi > 0. Then the solution says that institutional investorsare happier with portfolios that have (1) higher returns and (2) less standarddeviation.

Next, consider the global minimum variance portfolio. This portfolio is

X =

X1...

Xn

= �−11

1′�−11+

[I − �−111′

1′�−11

]�−1γ. (10)

In this case, the solution shows that investors are happiest with the portfolio onthe frontier of all risky assets that gives the lowest possible variance. See Chun,Ciochetti and Shilling (2000) for more details. Here we would like to pointout that neither (9) nor (10) readily explains why most institutional investorshold very little of their investments in real estate. As seen, for example, inRosen (2001), Sanders (1999) and Geltner, Rodriguez and O’Connor (1995),mean-variance analysis generally suggests allocation to real estate of about 30to 35%, depending on the rate of return assumed. The data show, however, thatinstitutional investors tend to invest only about 2 to 3% of their assets in realestate. Nor can (9) or (10) easily explain why some institutional investors investin REIT stocks and others do not.

Case 2: Asset-Liability Framework with No Liquidity Constraints

Suppose now that X F L > 0, but otherwise the same assumptions of Case 1apply. In this case, the deterministic equivalent of (2) is

3 This presumes that the investor maximizes a linear combination of mean and variance,with a positive weight on mean and a negative weight on variance.


N∑i=1

(1 + Ri − Ci )Xi − (1 + ρF L )X F L + K (α)[(X ′�X ) + X2

F Lσ 2F L

− 2(X ′X F L )]1/2 ≥ 0. (11)

With∑N

i=1 Xi = 1, Xiγi = 0, Xi ≥ 0 and γi ≥ 0, the optimal value of X ′ =(X1, . . . , Xn) is

X =

X1...

Xn

= �−1 A

A′�−1 Aδ +

[I − �−1 AA′

A′�−1 A

]�−1γ

+[

I − �−1 AA′

A′�−1 A

]�−1X F L , (12)

where

µ =

R1 − C1 − ρF L X F L...

Rn − Cn − ρF L X F L

.

This has almost the same form as before, except the relation now containsthe vector of covariances between changes in X F L and asset returns. In thiscase, investors care about three attributes of their portfolios: (1) they wanthigher average return, (2) they want lower standard deviations and (3) theywant portfolios that perform well in periods when ρF L increases.4

The results also extend to the global minimum variance surplus return portfolio,which is given by

X =

X1...

Xn

= �−11

1′�−11+

[I − �−111′

1′�−11

]�−1γ+

[I − �−111′

1′�−11

]�−1X F L .

(13)

In this case, institutional investors will only want to buy assets whose returnsgo up when ρF L goes up.

Next, we simulate the model to examine the extent to which XC L = 0 and X F L >

0 influence portfolio choice. The results of the simulations are presented in

4 Other models stress the importance of duration matching in customizing pension planportfolios. See, for example, Leibowitz, Kogelman and Bader (1994) and Peskin (1997).


Panel A of Table 1. Here the stock returns are the returns on the S&P 500 andthe bond returns are the intermediate and long-term government bond returnsreported from Ibbotson Associates’ SBBI 1999 Yearbook. REIT returns are thetotal return from equity REITs reported by the National Association of RealEstate Investment Trusts (NAREIT). For returns on real estate, the principalsource of data is the quarterly NCREIF return index produced by the NationalCouncil of Real Estate Investment Fiduciaries (NCREIF). We use the Fisher,Geltner and Webb (1994) unsmoothing technique to unsmooth the NCREIFreturns. These quarterly returns are then annualized over the time period.

In addition, we need to measure changes in pension liabilities. These data areavailable from the Business Information File of Compustat, and they includepension assets, number of employees and, most importantly, the pension plan’svested benefit obligation (VBO) and projected benefit obligation (PBO). VBOrepresents the actuarial present value of all benefits earned by employees. PBOrepresents the actuarial present value of all benefits earned by employees tothat date plus the projected benefits attributable to future salary increases asdetermined by the plan’s benefit formula. Therefore, PBO includes VBO plusthe present value of future pension liabilities. For our model, XC L is equal toVBO liabilities and X F L is equal to PBO minus VBO liabilities. VBO and PBOliabilities are calculated for each pension plan based on information in Chun,Ciochetti and Shilling (2000). Then the annual rate of change in both PBO andVBO liabilities was calculated. All data are annual from 1989 to 1997.5

We assume a funding ratio of 1.03 (i.e., an XT L of 0.97).6 For the purposes ofour model, we have also assumed that there are no liquidation costs for stock,bonds, or REITs. For our estimate of real estate liquidation costs, we assumeliquidation costs of 100 basis points.7

5 The Business Information File of Compustat is the best available data set for measuringthe growth or cost of future pension liabilities, but it suffers from several limitations.One limitation of the Compustat data is the sample period. The data on pension liabilitiesare generally not available for most firms before 1989 because firms were not requiredto report their projected benefit obligations.6 The funding ratio is usually defined as pension plan assets divided by the presentvalue of total pension plan liabilities. From 1989 to 1997 the mean funding ratio fromour panel of companies, weighted by plan size, was 1.03, meaning the average pensionplan was slightly overfunded.7 It is hard to know how big real estate liquidation costs are. Most estimates are that therapid sale of real estate leads to price discounts of 20 to 25% relative to the orderly salethat might take place over several months. Work by Childs, Ott and Riddiough (1997)and Ciochetti and Riddiough (2000) calculates, for given default frequencies, loan lossseverities associated with commercial mortgage foreclosures of between 98 and 138basis points. Given this, we also simulated the model for Ci equal to 50 and 150 basispoints. The results are generally not sensitive to the size of the liquidation-cost parameter.


Table 1 � Theoretical pension plan allocations based on relative levels of current andfuture pension liabilities.

Long-Term Intermediate RealRisk Level Stock Bonds Bonds Estate REITs

Panel A: Pension with 0% current liabilities, 100% future liabilities

No risk aversion 100.00 0.00 0.00 0.00 0.0050% 57.37 42.63 0.00 0.00 0.0068% 39.59 59.37 0.41 0.00 0.6390% 22.46 54.52 10.91 12.11 0.0095% 16.13 62.52 0.00 21.35 0.0097.5% 13.39 63.59 0.00 23.02 0.0099% 10.69 64.66 0.00 24.65 0.00

Panel B: Pension with 40% current liabilities, 60% future liabilities

No risk aversion 100.00 0.00 0.00 0.00 0.0050% 53.94 46.06 0.00 0.00 0.0068% 26.54 53.13 17.55 0.00 2.7890% 17.85 52.05 22.45 7.20 0.4595% 12.34 50.99 27.93 8.74 0.0097.5% 10.23 54.97 25.22 9.58 0.0099% 8.42 57.89 21.37 12.32 0.00

Panel C: Pension with 60% current liabilities, 40% future liabilities

No Risk Aversion 100.00 0.00 0.00 0.00 0.0050% 52.51 47.49 0.00 0.00 0.0068% 10.58 37.19 47.58 0.00 4.6590% 9.68 34.46 48.27 4.85 2.7495% 8.84 34.04 49.89 7.23 0.0097.5% 8.13 32.70 52.19 6.98 0.0099% 7.55 32.24 54.74 5.47 0.00

Panel D: Pension with 100% current liabilities, 0% future liabilities

No Risk Aversion 100.00 0.00 0.00 0.00 0.0050% 49.79 45.99 0.00 0.00 4.2368% 7.61 38.33 44.86 0.00 9.2090% 5.02 37.88 48.44 1.71 6.9595% 4.63 36.23 50.83 3.27 5.0197.5% 2.53 36.01 55.24 2.51 3.7199% 1.39 35.66 60.86 0.00 2.09

Note: Real estate liquidation costs of 100 bp.


The results presented in Panel A of Table 1 assume 100% future liabilities. Theresults set out seven portfolios of stocks, long-term and intermediate bonds, realestate and REITs—one for no risk aversion and one for low risk tolerance, an α

of 0.01. Of the remaining five portfolios, three are low-risk portfolios, with anα between 0.025 and 0.10, and two are midrisk portfolios, with an α between0.32 and 0.50.

The results demonstrate that if the pension manager had no risk aversion, thenthe portfolio for a 100% future liability plan is 100% allocation to stock. How-ever, as risk aversion increases, the portfolio allocations change. A pensionwith all future liabilities becomes concentrated in bonds and real estate, witha small amount in stocks. With very high risk aversion, the allocations wouldbe 64.66%, 24.65% and 10.69%, respectively. Another important result wehave established is that REITs are not held in the portfolio except when riskaversion is relatively modest, and even then the allocation is quite small. Ac-cording to our model, pension plans with XC L = 0 have, on the average, lowlevels of REIT stocks, while they simultaneously have greater holdings in pri-vate real estate equities. Also, as we see, the results in Panel A of Table 1 areradically different from Rosen (2001), Sanders (1999) and Geltner, Rodriguezand O’Connor (1995). This discrepancy comes about in Rosen (2001), Sanders(1999), Geltner, Rodriguez and O’Connor (1995) and elsewhere when institu-tional investors rank portfolios according to their mean and variance of return.In our case, institutional investors tailor their asset holdings to hedge againstchanges in long-term liabilities.

Case 3: Asset-Liability Framework with Liquidity Constraints

For our final case, assume that XC L > 0 and X F L > 0. This model then impliesthat there may be times or states of the world in which institutional investorsmay prefer REIT stocks to real estate assets. The intuition is as follows. WithXC L > 0 and X F L > 0, investors care about four attributes of their portfolios:(1) they want higher average return, (2) they want lower standard deviations,(3) they want portfolios that perform well in periods when ρF L increases and(4) they need a certain amount of liquidity. Now, holding XT L fixed, increaseXC L for some institutional investors XC L < X ′

C L so that the liquidity constraintin (6) becomes binding. In this case, the typical investor is led to pick the bestpossible portfolio, trading off mean, variance and interest rate sensitivity, subjectto (6). As XC L increases, the incentive to invest in REIT stocks increases and theincentive to invest in real estate assets decreases. This is particularly so for thoseinstitutional investors who are unable to borrow (as most institutional investorsare) when returns are temporarily low. Consequently, for these reasons, someinstitutional investors may want portfolios that are more overweighted towardREIT stocks than toward real estate assets.


Our simulation results of the model in (1), (8), (3), (4) and (5) with XC F > 0and X F L > 0 are shown in Panel D of Table 1. We consider the same inputvalues for Ri , ρC L , �, and XT L as used above. We use the growth or changein PBO per worker to compute ρF L and .

The results in Panel D of Table 1 suggest that in a pension with all currentliabilities, the portfolio becomes very heavily weighted in long-term and in-termediate bonds (35.66% and 60.86% with high risk aversion), with smallallocations to stocks (1.39%) and REITs (2.09%). Interestingly enough, REITsare part of the asset allocation through most levels of risk aversion reachinga maximum allocation of 9.20%. Direct real estate is not a significant assetwith its highest allocation being only 3.27%, which is roughly consistent withobserved behavior.

Next note that as a pension plan goes from being all current liabilities to beingall future liabilities, the asset allocations become more concentrated in stock,long-term bonds and real estate, at the expense of REITs and intermediatebonds (see Panels B and C of Table 1). For real estate liquidation costs of 50and 150 basis points, it is possible to show that these results only change a fewpercent. Still, it should be noted that as the liquidity cost goes down, allocationsto real estate do, in fact, increase. It should also be noted that as the real estateallocations decrease, the allocations to REITs and intermediate bonds increase.These results suggest that liquid REIT stocks are a substitute for real estate.

Thus, to summarize this section, it is quite conceivable that liquidity-constrainedinstitutions prefer liquid REIT stocks to illiquid real estate assets and that, asthe level of current pension liabilities increases, liquid REIT stocks becomemore important relative to real estate assets as a source of portfolio returns forinstitutional investors. Moreover, the share of REIT stocks to total plan assetsshould be systematically larger the higher are real estate liquidation costs and thehigher is the correlation between REIT returns and the cost of pension liabilities.The remainder of this study examines portfolio weightings in various forms totest if liquidity constraints do indeed motivate institutional investors to holdmore of REIT stocks vis-a-vis direct investments in private real estate equitiesand to invest in more liquid, as opposed to illiquid, REIT stocks.

An Empirical Test of the Liquidity-Constraint Hypothesis

In the previous section, we developed a theoretical model looking at why someinstitutional investors invest in REIT stocks and others invest directly in pri-vate real estate equities. Now we want to test empirically the hypothesis thatpension plans with greater liquidity needs (i.e., greater current liabilities) willinvest more in REITs than plans with lower liquidity needs (i.e., higher futureliabilities).


To test this hypothesis, we estimate a Tobit model of REIT portfolio investment(accounting for heteroscedasticity) of the form

LPREIT = a0 + x B + ε if a0 + x B + ε > 0= 0 if a0 + x B + ε ≤ 0

(14)

E(ε2) = σ 2 = σ 20 eαZ , (15)

where x is a vector of explanatory variables, B is a vector of parameters, ε isan error term distributed N(0, σ 2), and Z is plan size (individual subscripts areomitted for notational convenience). The LHS variable in (14) is LPREIT =log(1 + PREIT), where PREIT is the level of REIT investment as a percent ofeach pension plan’s investment in real estate (including both direct and indirectinvestment in real estate). This percentage is zero for 33 out of the 441 pensionplans that hold some real estate; it is between zero and 100% for 157 pensionplans; and it is exactly equal to 100% for 251 pension plans. We transformour dependent variable by adding one to PREIT and then employing a logtransformation in an attempt to keep the extreme observations from dominatingthe regressions.

The RHS variables in (14) are defined as follows:

1. VBO is the pension’s current liabilities (per employee).

2. Cash refers to percent of assets in cash and short-term securities. Onewould expect institutional investors with large cash holdings to havebelow-average needs for liquid REIT stocks. One would therefore ex-pect a negative relation between REIT holdings and percent of assetsin cash and short-term securities.

3. Size is the pension plan’s assets. By all accounts, as pension plans getlarger, the barriers to investing directly in real estate, such as high searchand management costs, diminish as economies of scale are achieved.One would therefore expect a negative relation between REIT holdingsand pension plan size. Additionally, it is possible for large pensionfunds to become too large relative to the typical REIT in order to buyand sell REIT shares quickly without a loss. In this case, one wouldexpect nonlinearities in the relation between REIT holdings and pensionplan size; plans in the largest size strata experience low REIT holdingsrelative to those in the smallest size strata because of the limitations inREIT liquidity.

4. Small Cap is a measure of the percent of assets in small cap stocks.If institutional investors are small cap investors, then their portfoliosshould conceivably have more REIT securities than the portfolios of


those institutional investors that are strictly large cap investors, as themajority of REIT shares are small cap stocks.

5. D is a series of 0–1 indicator variables for industry group. The groupsconsist of Petroleum; Finance, Insurance and Real Estate; ConsumerDurables; Basic Industries; Food and Tobacco; Construction; CapitalGoods; Transportation; Utilities; Textile and Trade; Services; Leisure;and Technology. If firms in certain industries are growing and expandingfaster than other industries, one might expect that REIT securities shouldbe present in the minimum variance portfolio of certain industry groups,but not for others, depending on the specific values of ρC L , ρF L , XC L ,X F L , � and .

Because pension plans differ sharply in size, there is a potential for het-eroskedasticity in (14). Hence, we estimate this system of Equations (14) and(15) using maximum likelihood.8 Also, we take logs of all RHS variables. Stan-dardizing our LHS and RHS variables in this way so that it does not changewith units of measurements puts the equation on a common footing. Moreover,because the regressions are in a log-log specification, the coefficients can beinterpreted as elasticities.

The data used to estimate (14) were collected principally from MMD. MMDannually conducts surveys on the investment characteristics of over 60,000 U.S.pension plans, endowments and foundations. These data include a variety ofinstitutional detail, including total pension plan assets, allocation of plan assets,as well as information regarding the number and type of investment managers foreach asset class. With respect to real estate ownership, MMD reports both directand comingled holdings, and, importantly for this study, ownership of REITstocks. Unfortunately, the MMD data are incomplete, and several pension planinvestors lack data on REIT holdings. Missing observations were supplementedby the 13(f) filings whenever possible.

This procedure involved linking pension plan investors in MMD to their invest-ment advisor(s). This linkage allows one to estimate a pension plan’s holdingof REIT stocks by knowing each plan’s dollar investment with an investmentadvisor, the total amount of dollars being managed by each investment advisorand the total amount of REIT holdings held by each advisor. The latter is whatwe obtain from the 13(f) filings.

The SEC requires all institutions with over $100 million in discretionary fundsunder management and all positions in individual stocks greater than $200,000

8 See, for example, Petersen and Waldman (1981) and Greene (1993).


or 10,000 shares to report through 13(f) filings. These data are published quar-terly in CDA/Spectrum 13(f) Institutional Stock Holdings (Spectrum). Spec-trum categorizes institutional ownership as reported in 13(f) into five groups:(1) bank trust departments, (2) insurance companies, (3) investment compa-nies (i.e., mutual funds), (4) investment advisors and (5) other. These cat-egories are assigned by the predominant investment activity as reported byeach manager. Data are collected on both MMD and 13(f) filings for 1993 and1998.

We have tried to eliminate obvious errors in the MMD database and havecross-checked the data whenever possible. We have also compared reportedREIT holdings in MMD with the holdings inferred from the Spectrum database.This comparison shows that the two values are equal for about 50% of thesample. One-third of the holdings inferred from the Spectrum database appearto underreport actual holdings in MMD, while 17% appear to overreport. Themean squared prediction error is slightly under 2%. The correlation betweenthe reported and inferred holdings is 0.99. So, as nearly as we can tell, there isjust a slight underreporting error going from the Spectrum database, as we do,to the reported holdings in MMD. However, there is no evidence that the resultsto follow are substantially affected by these reporting errors. We considereddropping the missing observations from the sample. We rejected this idea,however, because of the potential loss in efficiency.

On the plus side, the merging of these two data sets allows us to categorize thedata into more precise groups. It also allows us to create a new category: pen-sion plan investors. In the process, we redefined the Other category in Spectrumto include investment advisors with non-pension-plan clients. There are 1,182investment advisors listed in Spectrum for 1998. Of these, 509 are strictlypension-plan advisors. The remaining 681 are generally advisors to endow-ments, foundations and high-wealth individuals, and some may be advisorsboth to pension- and non-pension-plan clients.

In Table 2, we describe the nature of institutional investment in REIT stocks forboth 1993 and 1998, as well as the change in investment characteristics overthis period. Statistics reported include the number of institutions respondingaccording to 13(f) requirements, the number of those institutions invested inREIT stocks, the percent of each category invested in REITs, dollar investmentby category, percent of total institutional investment and percent of the REITmarket.

As shown, the base sample is quite large, with 8,801 institutions reporting in1993 and 11,313 in 1998. Of these, 2,019 are shown to have REIT holdingstotalling $5.7 billion in 1993, comprising nearly 18% of the REIT market.Institutional allocation to REIT stocks increases dramatically by 1998, with


Table 2 � Institutional investment in REIT stocks.

BankPension Mutual Insurance Trust

Time Period Plans Funds Companies Dept. Other Total

1993Number of institutions 8,043 63 65 184 446 8,801Number invested in REITs 1,768 31 27 110 83 2,019Percent of category 22.0 49.2 41.5 59.8 18.6 22.9Dollars invested ($mm) 3,945 650 234 428 456 5,713Percent of institutional $ 69.1 11.4 4.1 7.5 8.0 100.0Percent of REIT market 12.3 2.0 0.7 1.3 1.4 17.7

1998Number of institutions 10,253 77 70 158 755 11,313Number invested in REITs 4,780 73 46 119 283 5,301Percent of category 46.6 94.8 65.7 75.3 37.5 46.9Dollars invested ($mm) 38,553 8,497 3,028 4,272 18,335 72,685Percent of institutional $ 53.0 11.7 4.2 5.9 25.2 100.0Percent of REIT market 27.9 6.1 2.2 3.1 13.3 52.6

Change 1993–1998 (%)Number of institutions 27.5 22.2 7.7 (14.1) 69.3 28.5Number invested in REITs 170.4 135.5 70.4 8.2 241.0 162.6Percent of category 111.8 92.7 58.3 25.9 101.6 104.8Dollars invested ($mm) 877.3 1,207.2 1,194.0 898.1 3,920.8 1,172.3Percent of institutional $ (23.3) 2.6 2.4 (21.3) 215.0 0.0Percent of REIT market 126.8 205.0 214.3 138.5 850.0 197.2

Note: Total REIT market capitalization: 1993 = $32.2 billion, 1998 = $138.3 billion.

5,301 institutional investors holding nearly $72.7 billion of REIT stocks, ornearly 53% of the outstanding market. Clearly the role of institutional investorsin the market capitalization of REIT stocks has increased significantly over thepast decade.

Pension plans are shown to be the dominant institutional investor in REITstocks for both 1993 and 1998, with $3.9 billion and $38.6 billion in sharesoutstanding, nearly a ninefold increase. Moreover, the REIT market share heldby pension plans is shown to have increased from 12.3% in 1993 to slightlyless than 28% in 1998, providing evidence in support of the five-or-fewer rulehypothesis. All other institutional categories also significantly increased theirexposure to REIT stocks over this time period. These data, of course, couldhide some more complicated interdependence.

We next stratify the sample by predominant REIT structure. We do so by col-lecting REIT-level data from SNL REIT Quarterly for the second quarters ofboth 1993 and 1998. For each REIT, we collect income and balance sheet data,


Table 3 � Institutional investment in REIT type.



1993Dollars invested ($mm) 3,945 650 234 428 456 5,713Equity REITs (%) 93.9 91.5 94.2 92.9 91.9 93.4Mortgage REITs (%) 4.4 3.0 2.8 3.6 6.5 4.3Hybrid REITs (%) 3.6 5.4 3.0 3.5 1.6 3.6

1998Dollars invested ($mm) 38,553 8,497 3,028 4,272 18,335 72,685Equity REITs (%) 95.9 94.3 97.0 95.0 94.0 95.2Mortgage REITs (%) 3.1 3.5 2.4 3.0 4.5 3.4Hybrid REITs (%) 2.0 2.2 0.6 2.0 1.6 1.9

Change 1993–1998 (%)Dollars invested ($mm) 877.3 1,207.2 1,194.0 898.1 3,920.8 1,172.3Equity REITs (%) 2.1 3.1 3.0 2.3 2.3 1.9Mortgage REITs (%) (29.5) 16.7 (14.3) (16.7) (30.8) (20.9)Hybrid REITs (%) (44.4) (59.3) (80.0) (42.9) 0.0 (47.2)


financial ratios, performance statistics and ownership structure. We also clas-sify each REIT as to type (i.e., equity, debt, or hybrid), as well as predominantproperty type held within each REIT (i.e., office, residential, retail, hotel, diver-sified, or other). Table 3 provides a summary of the data as stratified by REITtype. Equity REITs are by far the choice of institutional investors, comprising93.4% of institutional portfolio dollars in 1993, increasing to 95.2% in 1998.Mortgage REITs constitute 4.3% of institutional dollars in 1993, while hybridREITs encompass 3.6%. These two categories fell out of favor with institutionsover the study period, declining to 3.4% and 1.9% of institutional investmentsin REITs, respectively, in 1998.

Table 4 stratifies the sample by predominant property type. In 1993, retailREITs dominated the institutional portfolio, with an allocation of 52.0% of in-vestment dollars. This is followed by other, residential, and diversified REITs,with 17.2%, 16.0% and 12.5%, respectively. Interestingly, very little institu-tional money flowed to office or industrial REITs during this period, and nofunds were allocated to hotel REITs. By 1998, institutions significantly in-creased their positions in residential, hotel, industrial and office REITs, withoffice REITs accounting for 29.5% of REIT ownership, a 41-fold increase from1993. Hotel and industrial REITs, while not a large proportion of institutionalREIT ownership in 1998, nonetheless also experienced significant increasesin institutional attention. Retail REITs experienced the most dramatic decline


Table 4 � Institutional investment by REIT property type.



1993Dollars invested ($mm) 3,945 650 234 428 456 5,713Office REITs (%) 0.7 0.0 1.8 0.6 1.3 0.7Residential REITs (%) 16.4 10.9 38.1 13.3 11.2 16.0Retail REITs (%) 54.9 50.6 42.0 43.1 42.8 52.0Hotel REITs (%) 0.0 0.0 0.0 0.0 0.0 0.0Diversified REITs (%) 13.3 8.4 4.5 13.6 14.7 12.5Industrial REITs (%) 1.8 0.1 0.4 1.5 1.4 1.6Other (%) 12.9 30.0 13.2 27.9 28.6 17.2

1998Dollars invested ($mm) 38,553 8,497 3,028 4,272 18,335 72,685Office REITs (%) 31.2 30.3 31.8 26.8 25.6 29.5Residential REITs (%) 23.3 19.2 21.1 19.7 20.4 21.8Retail REITs (%) 22.9 21.4 17.5 20.4 27.4 23.5Hotel REITs (%) 2.6 2.0 2.0 2.8 3.0 2.6Diversified REITs (%) 10.7 6.4 8.9 9.6 5.7 8.8Industrial REITs (%) 5.1 20.0 10.6 8.7 9.7 7.8Other (%) 4.2 0.7 8.1 12.0 8.2 6.0

Change 1993–1998 (%)Dollars invested ($mm) 877.3 1,207.2 1,194.0 898.1 3,920.8 1,172.3Office REITs (%) 4,357.1 n/a 1,666.7 4,366.7 1,869.2 4,114.3Residential REITs (%) 42.1 76.1 (44.6) 48.1 82.1 36.3Retail REITs (%) (58.3) (57.7) (58.3) (52.7) (36.0) (54.8)Hotel REITs (%) n/a n/a n/a n/a n/a n/aDiversified REITs (%) (19.5) (23.8) 97.8 (29.4) (61.2) (29.6)Industrial REITs (%) 183.3 1,990.0 2,550.0 480.0 592.8 387.5Other (%) (67.4) (97.7) (38.6) (56.9) (71.3) (65.1)


in institutional interest over this period, falling by nearly 55%. Much of thisdecline may be attributed to the oversupply and poor performance of the retailproperty sector during the mid- to late 1990s.

Table 5 provides a different perspective on institutional REIT ownership. Insteadof looking at the pattern of institutional ownership by type of REIT, we lookat the number of different REIT stocks held. We start out by first reportingboth the mean and median number of different REIT stocks held. We thendivide the sample into those investors holding up to 5, 5–10, 10–15, 15–20,20–50, 50–100 and over 100 different REIT stocks. Here we see that pensionplans tend to hold far more different REIT stocks than do institutions of othertypes. A typical mutual fund, for example, holds between 5 and 20 different


Table 5 � Number of different REIT stocks held by institutions.

Panel A: Mean and median number of REITs held

EndowmentsPension Mutual Insurance Bank Trust andPlans Funds Companies Dept. Foundations

1993 1998 1993 1998 1993 1998 1993 1998 1993 1998

Mean 14.8 38.3 5.9 20.2 4.7 21.9 6.6 23.2 9.9 27.4Median 7 21 3 9 2 4 3 8 5 10

Panel B: Distribution of holdings by percent of total institutions

EndowmentsPension Mutual Insurance Bank Trust andPlans Funds Companies Dept. Foundations

Numberof Shares 1993 1998 1993 1998 1993 1998 1993 1998 1993 1998

Over 100 0 13.2 0 2.7 0 6.5 0 5.0 0 6.851–100 1.1 8.4 0 9.6 0 8.7 0 9.2 0 8.821–50 28.2 29.0 6.5 17.8 3.7 6.5 6.4 16.0 15.6 19.916–20 7.5 7.2 3.2 5.5 3.7 9.7 7.3 4.2 7.5 6.911–15 6.0 6.4 0 11.0 3.7 2.2 5.5 6.7 11.3 7.46–10 15.6 10.4 19.3 12.3 11.1 13.0 12.7 20.1 14.6 16.81–5 41.7 25.7 71.0 41.1 77.8 54.4 68.2 38.7 50.9 33.6

REIT stocks. Insurance companies and bank trust departments average between4 and 23 different REIT stocks; endowments average between 7 and 27. Theaverage pension plan summarized in Table 5, by contrast, holds between 15 and40 different REIT stocks. This result is not surprising, since pension plans aregenerally more sensitive to transaction costs caused by large-percentage bid-askspreads for illiquid or low-priced REITs. One might also expect pension plansto take a relatively small position in a large number of different REITs, andthey indeed do so. While the largest position in a typical mutual fund accounts,for example, for about 42% of the market value of the entire mutual fund’sholdings of REIT stocks, the largest position in a typical pension plan accountsfor only 10% of the market value of the entire pension plan’s holdings of REITstocks. Furthermore, whereas about 4 or 5 REITs are required to account forhalf of the market value of the REIT stocks held by the typical mutual fund, thecomparable figure for pension plans is much larger, between 11 and 12 differentREIT stocks.

Motivated by these observations, we next ask the question whether institutionalinvestors invest more in REIT shares (and less in private real estate equities) be-cause of liquidity constraints. The results of estimating (14) and (15) are shownin Table 6. Model 1 reports the results when corrected for heteroskedasticity,


Table 6 � Model of REIT Ownership. Model: Log (1 + Percent in PREITs) =Log (RHS Variables).

Adjusted for Unadjusted forHeteroskedasticity Heteroskedasticity

Variables Model 1 Model 2 Model 3 Model 4

Intercept −0.2247 −0.1765 −0.2403 −0.2269(0.1028) (0.0997) (0.0815) (0.0824)

Log of VBO per employee 0.0031 0.0024 0.0034 0.0025(0.0010) (0.0013) (0.0008) (0.0011)

Log of percent in cash −0.0020 −0.0016 −0.0022 −0.0020(0.0017) (0.0019) (0.0015) (0.0016)

Log of percent in small cap stocks 0.0008 0.0009 0.0009 0.0010(0.0019) (0.0018) (0.0017) (0.0017)

Log of pension plan assets −0.0321 −0.0302 −0.0332 −0.0312(0.0135) (0.0136) (0.0129) (0.0131)

Log of pension plan assets squared 0.0010 0.0010 0.0012 0.0011(0.0006) (0.0006) (0.0005) (0.0005)

FIRE −0.0229 −0.0232(0.0138) (0.0135)

Consumer durables −0.0041 −0.0043(0.0095) (0.0091)

Basic industries 0.0041 0.0044(0.0093) (0.0089)

Food and tobacco −0.0002 −0.0004(0.0121) (0.0105)

Construction 0.0008 0.0010(0.0144) (0.0142)

Capital goods 0.0003 0.0003(0.0091) (0.0086)

Transportation −0.0023 −0.0027(0.0141) (0.0137)

Utilities −0.0069 −0.0072(0.0094) (0.0089)

Textile/trade 0.0111 0.0114(0.0945) (0.0106)

Services −0.0119 −0.0125(0.0175) (0.0169)

Leisure 0.0064 0.0068(0.0131) (0.0125)

Technology −0.0226 −0.0235(0.0399) (0.0337)

Summary statisticsR-square 0.13 0.18 0.17 0.22Log likelihood −53.27 −55.62 −80.12 −81.24

Note: Standard errors are in parentheses.


while Model 3 reports the results assuming homoskedastic errors. The results arevery similar. First, the level of current liabilities is significantly positively relatedto REIT holdings at the 95% confidence level. Second, there is a quadratic andsignificant relation between REIT holdings and plan asset size. Third, the signson percent in cash and percent in small-cap stocks are as expected, but neitherone is significant.

We tested another specification of the model using the pension plan’s two-digitStandard Industrial Classification (SIC) code. The results of this model aresimilar to the first model, except none of the SIC codes were significant inexplaining pension plan allocations to REITs. See Models 2 and 4 in Table 6.

The cross-sectional results in Table 6 suggest a U-shape relation between REITholdings and plan size. The linear term is negative and the squared term ispositive for plan size, implying that REIT holdings decrease at a decreasing ratewith plan size at first, then increase thereafter. The lesson is that investors withlow wealth may choose not to invest in direct investments in private real estateequities because the benefits of diversification are not sufficient to compensatethem for the acquisition cost, and so many choose instead to invest in REITstocks. But as plan size increases, more and more investors choose to investin direct investments in private real estate equities. Thus, their REIT holdingsdecrease with plan size. However, beyond some point, the pattern changes, withlarge investors choosing to invest more in REIT stocks. We interpret this changein REIT holdings as consistent with the notion that large investors tend to viewREIT stocks as tactical investments. The alternative hypothesis is that REITstocks are not liquid in a way that matters to large pension funds. Support forthe latter hypothesis would require a negative squared term for plan size, whichis easily rejected by the data.

A second result from Table 6 is that as liquidity needs increase, measuredby the level of current liabilities, pension plans are more likely to invest inREIT stocks. This result is consistent with the theoretical model developed inthe second section. Furthermore, the results imply that studies of real estateportfolio behavior should include measures of liquidity constraints and shoulduse a risk constraint on wealth losses and a liquidity constraint on types of assetsto explain the cross-sectional differences in institutional investors’ propensitiesto invest in REIT securities.

An Empirical Test of Clientele Effects

Having established that institutional investments in REIT stocks are affected byliquidity constraints, we now turn to the question: Do the large positions heldby institutional investors lead them to demand REIT stocks with large-market


capitalizations? There is now considerable amount of evidence that institutionalinvestors do not invest in the same common stocks as everyone else. Instead theyprefer common stocks with large-market capitalization and thick markets. Thisbehavior stems from liquidity and transaction-cost motives in Schwartz andShapiro (1992) and from “prudence” considerations in Del Guercio (1996) andGompers and Metrick (2001), but it generally reflects the fact that institutionalinvestors work within their own set of regulatory and economic restraints.

Just as for the holdings of REIT stocks, we run a Tobit model on the holdings ofinstitutional investors for each REIT in 1993 and 1998 to study the preferencesfor REIT stock characteristics. The model is shown in (16) and (17).

LIO = a0 + x B + ξ if b0 + xβ + ξ > 0= 0 if b0 + xβ + ξ ≤ 0

(16)

E(ξ 2) = σ 20 eαW , (17)

where the latent variable, IO, the level of institutional ownership as a percentageof the total outstanding shares for the REIT, is transformed by LIO = log(l +IO), x is a vector of explanatory variables, β is a vector of parameters, andW is REIT size. To the extent that institutional investors invest in the sameREIT stocks as everyone else, we would expect IO to be somewhat constantacross all REIT stocks. In this case, the results of running Tobit regressionsof LIO on REIT stock characteristics should yield a nonnegative intercept andzero coefficients on all other variables. In general, of course, if the preferencesof institutions and individuals are different, LIO should vary systematicallywith some variable like REIT size (a measure of liquidity) or REIT leverage(a measure of prudence)—a relationship that Del Ouercio (1996) and Gompersand Metrick (2001) test for common stock.

For these regressions, the RHS variables include measures of prudence consid-erations as well as liquidity considerations:

1. Dividend yield. Dividends from the preceding year divided by the priceat the beginning of the year.

2. Payout ratio. We use dividends from the preceding year divided byearnings for the preceding year.

3. Leverage. Leverage is measured as the ratio of total debt to total equity.

4. Stock price volatility. The variance of monthly prices over the previoustwo years.

5. Size. Price per share times shares outstanding.


6. Log of size, squared.

7. Price per share.

8. Share turnover. Volume divided by shares outstanding.

Items 1–4 are proxies for prudence considerations. Items 5–8 are proxies forliquidity considerations. The model allows for one set of parameters to deter-mine both the probability of investing in REITs and the density of additionalinstitutional demand for REIT shares. The model is estimated in log–log spec-ification. The variance of the error term is modeled in a fairly flexible way.That is, we assume that the error term varies directly with REIT size (marketcapitalization). The model is estimated twice, once for observations in 1993and once for observations in 1998.

Our data on REIT institutional ownership are drawn from 13(f) filings. We notethat in 1993 the distribution of IO (our approximate LHS variable) has a modeclose to zero and a fairly short tail to the right. About 50% of these REITs haveno holdings by large institutional investors, and 75% have IO levels less than15%. In 1998, IO increases across all REITs. The mode is still near zero, but itslopes downward much more slowly than in 1993. Only about 10% of REITshave no institutional ownership, and the top quartile has IO levels over 57%,with the highest at just over 84%.

In Table 7, we set out six sets of results: cross-sectional estimates of the modelfor 1993 and 1998 for all institutions, and then for pension plans, mutual funds,insurance companies, bank trust departments and endowments and foundations.The findings are summarized by four conclusions.

1. Moderate preference for liquidity in 1993. The results generally suggestthat institutions have a moderate preference for liquidity in 1993, andnot much else. See the first column of Table 7. Here we find a positiveand significant coefficient of stock price, which is used as proxy fortransaction costs. The theory is the higher the stock price, the lowerthe percentage bid-ask spread. The point estimates imply that a 10%increase in stock price increases institutional ownership by 6%. Of thisincrease, 3.3% is generated by a change in the probability of investing inREIT shares and 2.7% is generated by a marginal change in additionalinstitutional investment spending.9

9 Effects are calculated at the fraction of the sample investing in REIT shares and atmean values of the xs. The change in the probability of investing in REIT shares iscalculated by ∂ F(z)/∂x = f (z)β/σ, where z = xβ/σ, f (z) is a unit normal density, andF(z) is the cumulative normal distribution function. The marginal change in additional

InstitutionalInvestors’Preferences

forR

EIT

Stocks

589

Table 7 � Tobit regressions of institutional ownership. Model: Log (1 + IO) = Log (RHS Variables).

Type of Institution

All Pension Mutual Insurance Bank Trust Endowments/Institutions Plans Funds Companies Departments Foundations

Variables 1993 1998 1993 1998 1993 1998 1993 1998 1993 1998 1993 1998

Intercept −69.6 −3.70 −62.7 −5.28 −289.4 −3.76 −5.55 −5.34 −95.5 −6.15 −35.3 −3.49(82.2) (2.83) (71.4) (2.63) (219.2) (2.53) (29.3) (1.91) (75.8) (1.05) (39.5) (2.23)

Log of yield 0.26 −0.76 −0.11 −0.89 1.81 −0.79 0.71 −0.68 −0.17 0.16 −0.51 −1.01(4.00) (0.49) (3.42) (0.46) (3.61) (0.48) (1.48) (0.31) (1.34) (0.18) (1.70) (0.39)

Log of payout of FFO −2.03 −0.45 −1.71 −0.41 −1.83 0.07 −0.75 0.10 −0.51 0.10 −0.67 −0.84(1.45) (0.48) (1.25) (0.46) (1.29) (0.45) (0.54) (0.32) (0.44) (0.18) (0.59) (0.38)

Log of debt to equity −1.35 0.16 −1.26 0.23 −0.41 0.03 0.19 0.13 −0.25 0.2 −0.68 0.36(0.83) (0.21) (0.71) (0.19) (0.75) (0.19) (0.30) (0.12) (0.27) (0.07) (0.34) (0.16)

Log of volatility −1.51 −0.21 −1.44 −0.07 −0.76 −0.11 −0.26 −0.17 −0.67 −0.26 −0.62 0.07(2.02) (0.23) (1.73) (0.22) (1.68) (0.22) (0.65) (0.15) (0.61) (0.09) (0.81) 0.07

Log of market value 9.29 2.78 8.48 3.24 42.7 1.63 −0.24 1.91 14.3 1.51 4.84 2.87(14.2) (0.82) (12.3) (0.76) (33.9) (0.72) (4.90) (0.53) (11.7) (0.29) (6.73) (0.63)

Log of market value −0.37 −0.21 −0.34 −0.26 −1.62 −0.11 −.002 −0.4 −0.55 −0.11 −0.18 −0.22squared (0.58) (0.06) (0.50) (0.06) (1.31) (0.05) (0.18) (0.04) (0.45) (0.02) (0.27) (0.05)

Log of price 3.98 0.32 3.51 0.34 2.55 0.28 0.77 −0.13 0.96 0.05 1.65 0.38(2.02) (0.31) (1.71) (0.29) (1.74) (0.28) (0.74) (0.19) (0.64) (0.11) (0.84) (0.24)

Log of share volume −0.44 0.69 −0.21 0.64 −0.66 0.86 0.03 0.39 −0.01 0.14 −0.04 0.54(1.04) (0.17) (0.89) (0.16) (0.89) (0.16) (0.30) (0.11) (0.31) (0.07) (0.42) (0.14)

Summary statisticsR-square 0.50 0.56 0.52 0.60 0.38 0.64 0.38 0.49 0.39 0.61 0.49 0.66Log likelihood −63.8 −163 −60.6 −154 −45.9 −148 −29.3 −91.6 −31.9 −43.2 −41.7 −133

Note: Standard errors are in parentheses.


2. Strong preference for liquidity and some signs of prudence in 1998. Theresults generally show that institutions have very strong preferences forthe liquidity variables, high trading volume and higher market values.The results for the prudence variables are somewhat mixed. The signon volatility is as expected, but the signs on yield, payout ratio andleverage, while insignificant, are not.

3. Significant differences in preferences exist between 1993 and 1998. Thetheoretical basis for this change in behavior appears to lie in transaction-cost considerations. When IO levels for REIT stocks are small (lessthan 20%), institutions have a very weak preference for liquidity. Then,as institutions increase their share of the market, they demand greaterliquidity. These findings are similar to those in Gompers and Metrick(2001).

4. Tests involving the equality of coefficients of different Tobit regressionsare inconclusive. Generally speaking, no significant differences amongpension plans, mutual funds, insurance companies and bank trust de-partments were noted in 1993, but there is some support for pensionplans behaving differently in 1998. One notable exception is for banksin 1998. In this case, the coefficient on volatility is significantly nega-tive. This result is similar to Del Guercio (1996), where she finds thatbanks have higher preferences for prudence variables.

Below, Stansell and Coffin (2000) make similar points regarding the set offactors that can lead to cross-sectional variation in institutional ownership ofREIT stocks. They generally find preferences for REITs with large-marketcapitalizations. Our study extends their analysis by looking at preferences overall REITs and by type of investor. We also extend their analysis by examininginstitutional preferences for diversification.

Several additional versions of the model were also tested. These tests includedwhether institutional investors view their investments in REIT securities asstocks. If viewed as stocks, one would expect institutions to increase theirinvestments in REIT securities automatically as REITs become a larger shareof the small stock market, making the estimates in Table 7 spurious.10 Overall,

institutional investment spending is calculated by ∂IO∗/∂x = [β/(1 + IO)][1 − z f (z)/F(z) − f (z)2/F(z)2].10 To test this hypothesis, we divide institutional investors in our sample into those thathave a small-cap stock portfolio and those that do not. We then reestimate the relationbetween institutional ownership and REIT characteristics separately for each subsample.We generally conclude that splitting the sample into two groups does little to help explaininstitutional REIT ownership.


we find that both types of investors have a strong and very significant preferencefor REITs with large capitalizations and thick markets.

We also tested whether institutions with and without private real estate invest-ments have the similar preferences for REIT shares. Institutions that invest inprivate real estate investments generally ought to have better knowledge aboutreal estate returns than other investors, and they may try to use their under-standing of real estate returns to improve REIT portfolio performance. Theirpreferences for REIT stock characteristics may therefore be significantly dif-ferent from those of other investors.11 The results (not reported here) suggest,however, that LIO is related to the other RHS variables in both Tobit regressionsin the same way. A Chow test fails to reject the hypothesis that the estimatedcoefficients in the model are equal. There is therefore little evidence that privatereal estate investments influence institutional REIT ownership.

Finally, we tested whether institutions that own a large number of different REITstocks have the same preferences as do other investors.12 It appears from theseregressions (not reported here) that larger REIT investors are more discerning,although, in both cases, the variable with the most explanatory power is marketcapitalization.

Conclusion

This study makes available detailed statistics on REIT portfolio investment andthe share of the REIT market held by pension plans, mutual funds, insurancecompanies, bank trust departments and endowments and foundations between1993 and 1998, directs attention to the differing preferences of these investorsfor certain types of REIT shares between these two time periods, and notessome influences affecting REIT portfolio investment in 1998, compared withother institutional investors.

To explain why some institutional investors invest in REIT stocks while otherinstitutions invest directly in private real estate equities, we develop a model

11 The analogy here is the REIT IPO market. Better-informed investors submit buy ordersfor underpriced issues, but they do not submit orders for overpriced issues. Noting thatfew institutional investors participated in REIT IPOs in the 1970s and 1980s, Ling andRyngaert (1997) use this argument to help explain why REIT IPOs experienced nearlya 3% decline on the first day of trading in the 1970s and 1980s, while experiencingslightly less than a 4% increase on the first day of trading in the period 1991–1994.12 To test this hypothesis, we reestimated regressions separately for institutions that ownmore than 20 different REIT stocks and institutions that own less than 20 different REITstocks.


in which institutions maximize expected surplus return (asset minus liabilities)subject to a risk constraint on wealth losses and a liquidity constraint on types ofassets. This formulation captures in a simple way the fact that most institutionalinvestors like their asset portfolios to do particularly well in the event that therate of return on pension liabilities increases and the need for wealth becomesmore important. The model also captures in a simple way the notion that aninstitution’s desire for liquidity increases as the size of its current liabilitiesincreases.

Two implications of the model follow. (1) For some institutions, the desire forliquidity brings about a clear decline in investment in illiquid assets, like realestate, and an increase in investment in liquid REIT stocks. (2) There should bea bias on the part of institutional investors for large, liquid REITs. Otherwise,there is no real gain to investing in REIT stocks.

The evidence in the paper shows that liquidity-constrained institutions do in-deed prefer liquid REIT stocks to illiquid real estate assets. Somewhat limitedevidence further suggests that institutional investors have a strong preferencefor REIT shares with greater market capitalization and greater liquidity. Thismakes the demand for REIT stocks appear less random. It also supports theimportance of investor clienteles for understanding REIT asset pricing.

The authors thank David Geltner, participants at conferences held by the American RealEstate and Urban Economics Association in New York and by the Real Estate ResearchInstitute, and three anonymous referees for their helpful comments. We are grateful toJay Josey for providing data.

References

Below, S.D., S.R. Stansell and M. Coffin. 2000. Institutional Investment in REIT Com-mon Stocks: An Examination of the Prudent Man Investment Hypothesis. Journal ofReal Estate Portfolio Management 6(2): 113–130.Chan, S.H., W.K. Leung and K. Wang. 1998. Institutional Investment in REITs: Evidenceand Implications. Journal of Real Estate Research 16(3): 357–374.Childs, P.D., S.H. Ott and T.J. Riddiough. 1997. Bias in an Empirical Approach toDetermining Bond and Mortgage Risk Premiums. Journal of Real Estate Finance andEconomics 14(3): 263–282.Chun, G., B.A. Ciochetti and J.D. Shilling. 2000. Pension Plan Real Estate Investmentin an Asset/Liability Framework. Real Estate Economics 28(3): 467–491.Ciochetti, B.A. and T.J. Riddiough. 2000. Timing, Loss Recovery, and Economic Per-formance of Foreclosed Mortgages. Working Paper. University of North Carolina.Del Guercio, D. 1996. The Distorting Effect of the Prudent-Man Laws on InstitutionalEquity Investments. Journal of Financial Economics 40(l): 31–62.Downs, D.H. 1998. The Value in Targeting Institutional Investors: Evidence from theFive-or-Fewer Rule Change. Real Estate Economics 26(4): 613–649.


Fisher, J.D., D.M. Geltner and R.B. Webb. 1994. Value Indices of Commercial RealEstate: A Comparison of Index Construction Methods. Journal of Real Estate Financeand Economics 9(2): 137–164.Geltner, D.M., J. Rodriguez and D. O’Connor. 1995. The Similar Genetics of Public andPrivate Real Estate and the Optimal Long-Horizon Portfolio Mix. Real Estate Finance12(3): 13–25.Gompers, P.A. and A. Metrick. 2001. Institutional Investors and Equity Prices. QuarterlyJournal of Economics 116(l): 229–259.Graff, R.A. and M.S. Young. 1997. Institutional Investor Impact on Equity REITPerformance. Real Estate Finance 14: 31–39.Greene, W.H. 1993. Econometric Analysis. Macmillan: New York.Leibowitz, M.L., W. Kogelman and L.N. Bader. 1994. Funding Ratio Return. Journalof Portfolio Management 21(l): 39–47.Ling, D.C. and M. Ryngaert. 1997. Valuation Uncertainty, Institutional Involvement,and the Underpricing of IPOs: The Case of REITs. Journal of Financial Economics43(3): 433–456.Markowitz, H. 1959. Portfolio Selection: Efficient Diversification of Investments. JohnWiley & Sons, Inc.: New York.Peskin, M.W. 1997. Asset Allocation and Funding Policy for Corporate-SponsoredDefined-Benefit Plans. Journal of Portfolio Management 23(2): 66–73.Petersen, D. and D. Waldman. 1981. The Treatment of Heteroskedasticity in the LimitedDependent Variable Model. Mimeo. University of North Carolina at Chapel Hill.Rosen, K.T. 2001. Real Estate Investment Trusts: A Safe Haven in Volatile FinancialMarkets. Lend Lease Rosen Real Estate Securities LLC: Berkeley, CA.Sanders, G. 1999. An Updated Look at Asset Allocation: Private and Public Real Estatein a Multi-Asset Class Portfolio. The Real Estate Finance Journal 14: 5–13.Schwartz, R. and J. Shapiro. 1992. The Challenge of Institutionalization of the EquityMarket. A. Saunders, editor. Recent Developments of Finance. New York SalomonCenter: New York.

Date post:	10-Feb-2017
Category:	Documents
Upload:	vuongbao
View:	214 times
Download:	0 times

Institutional Investors' Preferences for REIT Stocks

Documents