Page 1

2007

REAL ESTATE

ECONOMICS

V35 1: pp. 57–90

ComovementAfterJoininganIndex:

SpilloversofNonfundamentalEffects

Brent W. Ambrose,∗Dong Wook Lee∗∗and Joe Peek∗∗∗

This study considers the case of two overlapping categories in the context

of recent category models. Specifically, we examine whether investor senti-

ment and market frictions specific to one category can affect the returns on

assets belonging to the other category. With recent additions of several real

estate investment trusts (REITs) into general stock market indices as a natu-

ral experiment, we find support for spillovers of such nonfundamental effects,

as evidenced by the increased return correlation between REITs that remain

outside the index and the index stocks. Further analysis reveals that market

frictions play a greater role than investor sentiment.

Over the past several years, a literature has developed suggesting that asset

returns can comove more than their fundamentals alone would justify. Among

others, Barberis and Shleifer (2003) and Barberis, Shleifer and Wurgler (2005)

show that, when investors form asset categories and trade based on them, in-

vestor sentiment or market frictions associated with those categories can lead

to excess return comovement among assets in the same category. In the con-

text of their model, category-specific sentiment arises when investors focus on

differences between categories but not on within-category distinctions; in the

extreme, such sentiment can lead investors to focus on a single category. At the

same time, category-specific market frictions can retard the diffusion of new

information so that assets in different categories do not incorporate the new

information at the same speed.

Inthisstudy,weconsiderthecaseoftwooverlappingcategories.Thiscasecom-

plements the original category model by shedding light on possible spillovers

of nonfundamental effects (i.e., investor sentiment and/or market frictions) be-

tween categories. An understanding of such spillovers is crucial for analyzing

∗Department of Insurance and Real Estate, Smeal College of Business, Pennsylvania

State University, University Park, PA 16802 or bwa10@psu.edu.

∗∗Korea University Business School, Seoul, Korea 136-701 or donglee@korea.ac.kr.

∗∗∗Gatton College of Business and Economics, University of Kentucky, Lexington, KY

40506-0034 or jpeek0@uky.edu.

C ?2007 American Real Estate and Urban Economics Association

Page 2

58

Ambrose, Lee and Peek

markets in which asset categories are dynamic, and thus category overlaps are

likely.

An overlap can arise when an asset (which we refer to as the common asset)

joins a new category while maintaining its membership in its original category.

In this formulation, investor sentiment and market frictions that are specific to

one category can affect the returns on assets belonging to the other category.

Specifically,investorsinonecategorycanexpandtheirinvestmentinteresttothe

othercategory,resultingingreatercomovementbetweenthetwocategories.In-

vestor sentiment can also trigger mispricing in one category and consequently

invite pairs trading between the common asset and assets in the other cate-

gory, effectively transmitting sentiment across categories and creating excess

comovement.1From the market friction point of view, such weakening of dis-

tinctions between the two categories can make new information diffuse to both

categories at more similar rates, again leading to greater comovement.

We test this spillover implication using the recent introduction of several real

estate investment trusts (REITs) into the Standard and Poor’s (S&P) general

stock market indices (i.e., S&P500, S&P400 and S&P600 indices). This event

has several features that allow for a clean test. First, both the REIT sector and

the S&P market indices are well defined, so that it is straightforward to deter-

mine whether a certain stock is a member. As a result, an overlap between them

and the scope of possible spillovers can be accurately identified. Second, be-

cause the S&P market indices are well diversified across industries and sectors,

possible spillovers will be unidirectional, flowing from the index category to

the REIT category. Third, the REIT sector and the S&P market indices each

constitutes a category as defined in the original category models. In particular,

the common investment classifications divide the investment universe into four

broad asset categories consisting of cash, stocks, bonds and real estate. Thus,

representing the most liquid vehicle for real estate investment, REITs are con-

sideredtobeasinglehomogeneousinvestmentclass,eventhoughtheeconomic

fundamentals of individual REITs differ somewhat (e.g., Chui, Titman and Wei

2003).2Finally,becausetheindexadditionsarearecentevent,wecanutilizean

uncontaminated pre-addition benchmark period during which no REITs were

included in the index category, and thus no spillovers can possibly be present.

1Greenwood (2005) makes a similar point based only on economic fundamentals. As

longastradingpairsarefoundwithinthecategory,ourconjecturealsoisconsistentwith

astatisticalarbitragestrategy(e.g.,Gatev,GoetzmannandRouwenhorst2006)thatneed

not be justified by fundamentals.

2Foratextbooktreatmentofbasicinvestmentclassification,seeSection7.2.1inGeltner

and Miller (2001).

Page 3

Comovement After Joining an Index 59

Our first set of results is easily summarized. We begin by confirming for REITs

the prior general finding of an increase in the beta of stocks that are added to an

index.3Interestingly, these index REITs show virtually no change in their beta

with the portfolio of nonindex REITs (those that are never added to the index),

suggesting that, after some REITs were introduced into the index, the nonindex

REITs might have comoved more with the index category as well. It turns out

that the beta of nonindex REITs with the S&P market indices indeed increases

after some REITs were introduced. The magnitude of these beta increases is

not trivial; the beta of nonindex REITs, as well as that of index REITs, almost

doubles around the index addition event.4,5

At least two issues arise in interpreting the above results as evidence of the

spillover hypothesis. One is that the beta increase may be driven by factors

other than the index-related sentiment or friction. More specifically, given that

the index category is highly correlated with the general stock market, the ob-

served beta increase may be a result of the increased loading of nonindex

REITs on marketwide shocks or due to an increase in the marketwide shocks

themselves.

The other issue is endogeneity. Suppose that, in a general context, an over-

lap arises between two categories and subsequently all assets in one category

comove more with the other category. Then, instead of interpreting this as ev-

idence of the spillover effect, one could argue that the overlap coincides with,

or is even caused by, the increasing similarity between the two categories (see

Barberis, Shleifer and Wurgler 2005, p. 299). Hence, it could be argued that the

observed increase in the beta of nonindex REITs (as well as the beta of index

3We examine the period surrounding the time when several REITs are first added to

the index category. As mentioned earlier, this is to ensure that the pre-addition period

is free from any spillover effects. Although the results on index REITs are based only

on the six REITs that are first added to the index category, most of our analysis focuses

on the majority of REITs that never join the index category from which we obtain the

power of our tests.

4We report results based on daily and weekly frequency data. Monthly results are

similar in magnitude but weaker in statistical significance. However, a few available

observations during the study period (48 months) require some caution, and thus we

focus on the daily and weekly results in this article; the monthly results are available

upon request.

5To prevent any anticipation effects from affecting our results, we examine only those

REITsthatareneveraddedtoanyoftheS&Pmarketindices.Weseparatelyexaminedthe

portfolio of subsequently added REITs—constructed by keeping them in the portfolio

only until their actual addition—and found that the beta increase of this portfolio with

theS&Pmarketindicesisverysimilartotheonereportedinthestudy.Thissuggeststhat

the anticipation effects are at most marginal. We will revisit this issue in the subsection

“Return Beta of Nonindex REITs with the Index Category.”

Page 4

60

Ambrose, Lee and Peek

REITs) with the index category is a result of the ongoing increase in similarity

between the two categories.

To address these issues, we conduct an array of robustness checks. First, we

examine whether the sensitivity of the REIT sector to marketwide shocks has

increased,whichalsocanbeviewedasincreasedintegrationoftheREITsector

with the general stock market. To this end, we construct a portfolio with stocks

that are neither in the REIT sector nor in the index category and then purge

the index-related components from it.6Under this competing explanation, we

should observe an increase in the beta of nonindex REITs with this result-

ing residual portfolio return, because it still contains the marketwide shocks.

However, we find a decrease in the beta of nonindex REITs with this residual

return. Conversely, when we purge the marketwide shocks from the index re-

turn, nonindex REITs still show an increase in their beta with this alternative

residual index return.7It thus follows that possible increases in the sensitivity

oftheREITsectortomarketwideshockscannotexplaintheincreaseinthebeta

of nonindex REITs with the index category.

Another competing explanation is that the second half of the sample period

had more marketwide shocks, and consequently the comovement among all

stocks, including nonindex REITs and index stocks, increased. If this were the

case, then such an increase in marketwide shocks would have caused greater

commonalities among REITs and, more importantly, other stocks in the market

must have experienced a similar magnitude of commonality increase. Indeed,

the commonality of nonindex REITs, measured using the first five principal

components of their return covariance matrix, has increased during our sample

period. However, the commonality increase for stocks that are neither in the

REIT sector nor in the index category is far smaller, suggesting that contempo-

raneous increases in the marketwide shocks alone cannot explain the increase

in the beta of nonindex REITs with the index category.

Regarding the endogeneity issue, because the concern is about contemporane-

ous changes on the part of the index category in a particular way (i.e., additions

of stocks that happen to be similar to REITs or the growing importance of such

similar stocks that are already in the index category), we prescribe a simple

fix.8By reconstructing the index return using only those stocks that have been

6We do this by regressing daily returns of the portfolio on daily returns on each of the

three market indices.

7Alternatively, we use the Center for Research in Security Prices value-weighted index

in place of the portfolio of non-REIT, nonindex stocks and find similar results.

8We thank an anonymous referee for this suggestion.

Page 5

Comovement After Joining an Index 61

in the index during the entire sample period and holding their weights at their

initialvalues,wecanpreventanychangesonthepartoftheindexcategoryfrom

affecting our results. With this reconstructed index return, we still observe an

increase in the beta of nonindex REITs. Although the statistical significance

weakens in some cases, we remain confident about our inference due to the

magnitude of the beta change.

Another way of ensuring robustness of our results is to examine the pattern

of the beta change. If our results are due to the increased similarity between

the index and REIT categories, then the change in the beta of nonindex REITs

with the index category should rather be gradual, and thus the beta change

should be detectable even before the overlap between the REIT and the index

categories arises (i.e., before the first index additions of REITs). Also, if our

results are simply a transient phenomenon, then the increase in the beta will

eventually be reversed over time. We find that the beta of nonindex REITs with

the index category remains unchanged until after the first index addition event.

Subsequently, their index beta increases, and this increase persists as more

REITs join the index category, suggesting that the observed spillover effects

are not temporary. The REIT sector in general appears to experience similar

changes in other aspects of its trading environment. For example, changes in

both the institutional holdings of REITs and REIT mutual fund flows exhibit

similarities with those from S&P500 stocks until after the first index addition

event. Subsequently, they increase substantially relative to S&P500 stocks,

making these alternative explanations less of a concern.

In addition to stock returns, we examine the trading volume of REITs relative

to the index category. Admittedly, it is difficult to make inferences from trading

volume results, because we do not know the exact composition. However, the

spillover hypothesis clearly implies that trading volume should become more

highly correlated between the two categories, even if it does not increase gen-

erally. Indeed, we find that the trading volume of nonindex REITs becomes

more highly correlated with the trading volume of the index category. How-

ever, changes in the REITs volume itself are inconclusive, because the results

are not robust to using different measures of trading volume.

We then investigate the relative importance of the two possible components

of the spillover effect: investor sentiment and market frictions. As in Barberis,

Shleifer and Wurgler (2005, p. 309), we use the Dimson (1979) beta to quan-

tify the contribution of market frictions to the observed beta increase. We find

that the increase in the beta of nonindex REITs with the large-cap index is

attributable primarily to mitigated market frictions, as is evidenced by no sig-

nificant change in the Dimson beta. In contrast, the Dimson beta of nonindex

REITs with the mid-cap or small-cap stock index categories continues to show

Page 6

62

Ambrose, Lee and Peek

a significant change. With the reconstructed index category returns (using only

stocks that have been in the index during the entire sample period and hold-

ing their weights at their initial values), the Dimson beta does not increase,

suggesting that nonindex REITs comove more with those incumbent stocks

because they now incorporate new information at more similar rates. These

results are consistent with the notion that large firms tend to incorporate new

information faster than smaller firms (e.g., Lo and MacKinlay 1990, Chordia

and Swaminathan 2000).

To put our results in perspective, we examine their implications for a mean-

variance optimizer who uses REITs for diversification purposes. Specifically,

weconductthemean-variancespanningtestofHubermanandKandel(1987)to

answerthequestionofwhetheraninvestorwhoalreadyholdsawell-diversified

portfolio of non-REIT equities (i.e., the three S&P market indices omitting

any REITs) can improve the mean-variance frontier by adding some REITs

to the portfolio. Although REITs remain a useful diversification instrument

throughout the study period, the results show that the diversification potential

of REITs is weaker following the introduction of REITs into the S&P indices.

The magnitude of the diversification reduction is equivalent to increasing the

annualized portfolio return standard deviation by up to 2% for the same mean

return.

Thisarticleproceedsasfollows.Thenextsectiondiscussestherelatedliterature.

The third section discusses in more detail the REIT sector as our experimental

setting.Thenextthreesectionspresenttheempiricalresults,andthelastsection

concludes.

Related Literature

A substantial literature about nonfundamental determinants of asset price co-

movements exists both at the aggregate and the individual asset levels and in

both domestic and international settings. Although the initial research question

on this issue was whether shocks in one country can affect other countries that

are economically unrelated to the original country (e.g., Forbes and Rigobon

2002, Bekaert, Harvey and Lumsdaine 2002, Bekaert, Harvey and Ng 2005),

interesting results are often found with individual assets or within a single

country.9For example, Froot and Dabora (1999) find a surprisingly low return

correlationbetweentwostocksthathaveclaimsonexactlythesamefuturecash

flows, whereas Pindyck and Rotemberg (1990) find a persistent price comove-

ment among commodities that are more or less unrelated.

9KarolyiandStulz(2003)haveanexcellentsurveyofinternationalfinancialcontagion.

Page 7

Comovement After Joining an Index 63

In a single-country setting, the inclusion of stocks into a general market index,

such as the S&P500 index, has been widely used as an experimental setting

on the following two grounds. First, the stock addition event is claimed to

be information free, meaning that it has little to do with the fundamentals of

added stocks.10Second, the presence of a large group of investors who have

an investment interest in index stocks is expected to create an economically

meaningful effect on the returns of assets in the index category from demand

shocksassociatedwithfluctuationsinfundflows.Infact,numerousstudieshave

found evidence that fund flows into and out of well-defined index categories

are correlated with fund returns.11

Inthecontextofindexadditions,BarberisandShleifer(2003)and,inparticular,

Barberis,ShleiferandWurgler(2005)formalizeinamodelandthenempirically

confirm two nonfundamental effects that can cause excess return comovement.

Specifically, when investors form asset categories and trade based on them,

category-specific investor sentiment or market frictions can arise and cause

assets in the same category to comove more than their economic fundamentals

alone would justify. Category-specific sentiment can cause investors to focus

on differences between categories but not on within-category distinctions and

attimestofixateonasinglecategory.Category-specificmarketfrictions,onthe

otherhand,canaffectthediffusionofnewinformationsothatassetsindifferent

categories do not incorporate the new information at the same speed. With at

least one of these two effects operative, assets can exhibit excess comovement

with other assets simply because they belong to the same category.

One assumption in the models of Barberis and Shleifer (2003) and Barberis,

Shleifer and Wurgler (2005) is that an asset can be in only a single category.

This assumption can be appropriate in some cases. For instance, a stock cannot

belong to both growth and value categories at the same time. However, this

10See, for instance, Harris and Gurel (1986), Shleifer (1986), Beneish and Whaley

(1996), Lynch and Mendenhall (1997), Kaul, Mehrotra and Morck (2000) or Wurgler

and Zhuravskaya (2002). More recently, however, evidence has been presented that

such index addition events are not entirely information free, for example, serving as

an indicator of increased future earnings (Denis et al. 2003). In addition, Dhillon and

Johnson(1991),inanearlierstudy,alsointerprettheirfindingofanincreaseintheprices

of nonconvertible bonds issued by firms that are added to the S&P index as evidence

of an information effect associated with an index addition. This debate, however, is

not particularly relevant for our analysis, because we focus on stocks that are never

added to the index. In any case, the index inclusion event does not necessarily affect the

covariance of economic fundamentals among different stocks, even if it is associated

with change in fundamentals on the part of the added stock.

11These include, among others, Vijh (1994), Warther (1995), Edwards and Zhang

(1998), Fortune (1998), Goetzmann, Massa and Rouwenhorst (2000), Cha and Lee

(2001), Edelen and Warner (2001), Karceski (2003), Goetzmann and Massa (2003) and

Ling and Naranjo (2003).

Page 8

64

Ambrose, Lee and Peek

assumption becomes less realistic once one allows for different types of cate-

gories. In the context of the index inclusion event, even after a stock is added

to an index, the stock does not leave its original industry. Moreover, some, al-

though not all, industries have a large group of investors whose investments are

dedicated to stocks in that industry and thus constitute their own category. This

type of category is different from the index category, thus remaining relevant

even after the index addition.

Experimental Setting: Introduction of REITs into the S&P

Market Indices

This study considers the case of a category overlap between an index category

and an industry category that can arise when a stock from an industry is newly

addedtoabroader-basedmarketindex.Albeitmorerealistic,ananalysisofsuch

anoverlapandpossiblespillovereffectsisdifficultbecausefewindexadditions

offer an appropriate experimental setting. Specifically, an ideal setting should

be such that (1) the addition is the first from its industry and (2) the stock’s

industry itself is a well-defined category whose characteristics are different

from those of the index category.

REITssatisfytheseconditions.ThefirstadditionsofREITsintogeneralmarket

indices, namely the S&P500, S&P400 and S&P600 indices, are a recent event,

soweareabletoutilizeapre-additionbenchmarkperiodduringwhichnoREITs

were in the index category, and thus no spillover effects were present. Further-

more,becausetheirinceptionin1960asaninvestmentvehicletoefficientlyown

commercialrealproperties,REITshaverepresentedaseparateinvestmentclass

in broader asset allocation decisions. Even though the economic fundamentals

of individual REITs differ to some extent, their distinctiveness compared to or-

dinary stocks make the REIT sector a unique and homogeneous category (e.g.,

Chui, Titman and Wei 2003). Few other industries satisfy these conditions.12

The characteristics of REITs have changed over time. Until recently, REITs’

market capitalization clearly placed them in the small-cap spectrum. For exam-

ple, in 1990 the total equity REIT market capitalization was only $8.3 billion

(in constant 2005 dollars).13However, by 2005, the total equity REIT market

capitalization had increased to $301.5 billion.

12Therefore, our spillover hypothesis does not warrant a more general intraindustry

effect study. In fact, Barberis, Shleifer and Wurgler (2005) conduct an intraindustry

analysis and find little impact on other stocks in the same industry as those added to the

S&P500 Index. Their sample period is from 1976 to 2000, a period that precedes the

2001 introduction of REITs into the S&P indices so that no REITs are included in their

sample.

13Source: National Association of Real Estate Investment Trusts (www.nareit.org).

Page 9

Comovement After Joining an Index 65

As the REIT industry has matured, the correlation and comovement of REITs

with the broader stock market has changed. For example, Ibbotson and Siegel

(1984)estimatedthecorrelationofrealestateandtheS&P500at−0.06overthe

periodfrom1947to1982,whileRosen(2001)documentsacorrelationbetween

REITsandtheS&P500of0.2overthe1993–2000period.Inaddition,anexten-

sive literature examines REIT performance and characteristics compared with

the general market. For example, Sanders (1998) finds that REITs performed

no better than the general stock market on a risk-adjusted basis and that REIT

performanceismostsimilartosmallstocksandhigh-yieldbonds.Similarly,ev-

idence presented by Clayton and MacKinnon (2003) suggests that REITs may

not be fully integrated with the broader stock market. In contrast, Li and Wang

(1995) find that REIT pricing is similar to that of other stocks, implying that

the REIT market is integrated with the broader stock market.14Most recently,

Chiang, Lee and Wisen (2005) examine the temporal stability of REIT market

betas. Their analysis finds only weak evidence (using a single-factor model) of

a downward trend in equity REIT betas.

Although numerous studies have examined the comovement of REITs with the

broader stock market, no study has examined the implications of adding REITs

to a broader market index and the possible roles of investor sentiment and

market frictions in the comovement between the REIT and the index sectors.

Given the low historical correlation between REIT returns and those of the

general stock market, the diversification potential has been one of the primary

advantages of investing in REITs. Therefore, our study will also examine how

the diversification potential of REITs has been affected by index additions.

Return Betas of REITs

Return Beta of Index REITs with the Index Category

We begin our analysis by confirming for REITs the prior general finding of an

increaseinthebetaofstocksaddedtoanindexwiththatindex(Barberis,Shleifer

andWurgler2005).Thedistinguishingfeatureofourtestisthatweexaminethe

veryfirstadditionsofstocksfromaspecificindustrytoageneralstockindexso

that the pre-addition benchmark period is free from any spillover effects from

prior additions. Specifically, we examine two adjacent periods around October

9, 2001, when six REITs were added to one of the S&P general market indices

(i.e., S&P500, 400 or 600). Prior to this date, none of these widely followed

14Other studies confirming the integration of REITs with the broader stock market

include Ambrose, Ancel and Griffiths (1992), Gyourko and Keim (1992), Li and Wang

(1995), Giliberto and Mengden (1996), Chaudhry, Myer and Webb (1999), Ling and

Naranjo (1999) and Glascock, Lu and So (2002).

Page 10

66

Ambrose, Lee and Peek

indices contained any REITs. Subsequent to the initial additions on October 9,

2001, another 15 REITs had been added as of the end of 2003.15

Weestimatethefollowingfirmfixed-effectpanelequationovertheperiodfrom

January 1, 2000 to December 31, 2003:

Ri,t= αi+ β1AFTER + β2RS&P−i,t+ β3(RS&P−i,t∗ AFTER) + υt,

where Ri,tis the period t (either daily or weekly16) log return of asset i (one of

thesixREITsaddedtooneoftheS&PindicesonOctober9,2001),andRS&P−i,t

is the period t log return of the value-weighted portfolio with all stocks in the

corresponding S&P index except for those six REITs. We also exclude from

this value-weighted portfolio all REITs that are subsequently added to any of

the indices. AFTER is a dummy variable taking a value of 1 for observations

occurringafterOctober9,2001and0otherwise.Incalculatingteststatistics,we

use the generalized method of moments (GMM) method to account for serial

correlationandheteroscedasticity.17Inordertocontrolforanyeffectsassociated

with the tragic events on September 11, 2001, we estimate Equation (1) with

and without the returns from September 2001.

(1)

We report results in Table 1, Panel A. The betas under the column heading

“before” correspond to β2in Equation (1) and reflect the comovement of the

REITs with the S&P indices prior to their inclusion in the index. The be-

tas reported under the column heading “after” correspond to the period from

October10,2001toDecember31,2003andarecalculatedasthesumofβ2and

β3in Equation (1). As expected, we find a significant increase in the REIT’s

betas after they are included in the S&P indices. The coefficients indicate that

the comovement of REITs with the S&P index stocks almost doubled after

their inclusion into the index. Furthermore, our results remain robust to using

a weekly return interval (the increase in beta is even larger), and the inclusion

of the returns from September 2001 does not qualitatively alter our results or

their interpretation.

Return Beta of Index REITs with Nonindex REITs

We now examine return comovements between the six index REITs and other

REITs that remain outside those indices during the entire study period. We

15A list of those index REITs is provided in Table A.1 in the Appendix.

16Weekly returns are from Tuesday closing to Tuesday closing, so that a new week

starts on October 10, 2001.

17Because we estimate a linear model and use all variables in the model as instruments,

the GMM coefficient estimates are identical to the corresponding ordinary least squares

coefficient estimates. However, their standard errors will be corrected for heteroscedas-

ticity and autocorrelation. See, for example, Greene (2000, p. 487).

Page 11

Comovement After Joining an Index 67

Table 1 ? Changes in beta of index REITs.

Without September 2001

Observations

With September 2001

Observations

BeforeAfter Diff. (p)BeforeAfter Diff. (p)

Panel A. Beta of Index REITs with the S&P Index

Daily

Weekly

0.246

0.173

0.449

0.562

(0.000)

(0.000)

0.274

0.211

0.449

0.562

(0.000)

(0.000)

Panel B. Beta of Index REITs with Nonindex REITs

Daily

Weekly

1.054

0.974

1.063

1.113

(0.923)

(0.124)

1.037

0.943

1.063

1.113

(0.743)

(0.052)

Note: Panel A reports the average beta of six REITs that are first added to one of the

three value-weighted S&P market indices (S&P500, S&P400 or S&P600) on October

9, 2001. The “before” beta is for the period from January 1, 2000 to October 9, 2001,

whereas the “after” beta is for the period from October 10, 2001 to December 31,

2003. The estimation is done using a firm fixed-effects panel regression of the log

return of the six REITs on a dummy variable for the period subsequent to October 9,

2001, the log return of the corresponding S&P index and their interaction term. The

“before” beta is the coefficient for the S&P index, and the “after” beta is the sum of

the coefficient on the S&P index and the coefficient on its interaction term with the

post-October 9, 2001 dummy variable. Observations are at either a daily or weekly

frequency. Weekly returns are from Tuesday close to Tuesday close, so that a new week

starts on October 10, 2001. The S&P indices are reconstructed without any REITs.

The estimation period is from January 2000 to December 2003. Autocorrelation- and

heteroscedasticity-consistent p values in parentheses are for the coefficient on the

interaction term that tests for a difference between the two betas. The p values for the

coefficient for the S&P index (i.e., “before” beta) are suppressed to save space, but they

are always significant at least at the 5% level. Panel B reports the average beta of the

six REITs with the value-weighted portfolio of other REITs that are never in any of the

three market indices during the entire sample period (January 2000 to December 2003).

continue to use the six REITs that are first added to one of the S&P indices

in order to avoid using, as a pre-addition benchmark period, a period during

which a spillover effect from earlier additions may be present. Therefore, we

change the preceding regression specification (Equation (1)) only by replac-

ing the portfolios of S&P index stocks with the portfolio of nonindex REITs.

The sample nonindex REITs are the constituents of the S&P REIT index (not

included in the general market indices).18

Specifically,weestimatethefollowingfirmfixed-effectpanelequationoverthe

period from January 1, 2000 to December 31, 2003:

18The S&P REIT index consists of 100 REITs, covering more than 80% of the securi-

tized U.S. real estate market.

Page 12

68

Ambrose, Lee and Peek

Ri,t= αi+ β1AFTER + β2Rnon Index REITs,t

+β3(Rnon Index REITs,t∗ AFTER) + υt,

(2)

where Ri,tis the period t (either daily or weekly) log return of asset i (one of

the six REITs added to one of the S&P market indices on October 9, 2001),

and Rnon Index REITs,tis the period t log return of a value-weighted portfolio of

REITsthatarenotinanyoftheS&Pgeneralmarketindices.Fromthisportfolio,

we exclude all REITs that are subsequently added to any of the S&P market

indices. We continue to use the GMM method to produce test statistics in order

to account for serial correlation and heteroscedasticity.

Panel B of Table 1 shows that the “before” beta (β2, the coefficient for

Rnon Index REITs) is very close to 1 (and, although the p values are not shown,

does not differ from 1). Furthermore, this relationship between index REITs

and nonindex REITs does not experience any meaningful change after October

9, 2001. That is, the “after” betas (the sum of β2and β3) are not significantly

different from the “before” betas. We find virtually no difference in the results

between daily returns and weekly returns, although for the weekly return series

that includes the September 2001 observations, the difference between the be-

fore and after subperiods is close to statistical significance (p value = 0.052).

If index REITs were affected by the tragic events on September 11 differently

than nonindex REITs, perhaps due to their differential exposures to related real

estate markets, then their correlation would be dampened during the “before”

period and thus could show an increase after October 9, 2001.19To avoid any

such spurious results, we therefore exclude the September 2001 observations

in the remaining analysis.20

The results in Panel B indicate that the index REITs were almost perfectly

correlated with the value-weighted portfolio of nonindex REITs before being

included in the S&P indices, and they indicate that this relationship does not

change even after those REITs are introduced into an S&P index. Given the

increaseinbetaoftheindexREITswiththeportfolioofotherstocksintheS&P

index (shown in Table 1, Panel A), this suggests that the returns of nonindex

REITs might also have become more highly correlated with the returns of the

19The analysis by Kallberg, Liu and Pasquariello (2005) of REIT returns after the

September 11, 2001 terrorist attacks indicates that the REITs with significant exposure

totheNewYorkofficemarketdidexperiencepositiveabnormalreturnsaftertheattacks,

but the effect had dissipated by November 2001.

20Barberis, Shleifer and Wurgler (2005) similarly exclude October 1987 observations

from their analysis. Their sample period ends in 2000.

Page 13

Comovement After Joining an Index 69

index category. We therefore directly examine the beta of nonindex REITs with

the index category in the next subsection.

Return Beta of Nonindex REITs with the Index Category

We now directly examine the return comovements of nonindex REITs with

the S&P general market indices. As in the previous section, we exclude from

this nonindex REIT group all REITs that are subsequently added to any of the

S&P market indices, and we estimate the following regression over the period

from January 1, 2000 to December 31, 2003, omitting the September 2001

observations:

Rnon Index REITs,t= α + β1AFTER + β2RS&P−REITs,t

+β3(RS&P−REITs,t∗ AFTER) + υt,

(3)

where Rnon Index REITs,tis the period t (either daily or weekly) log return of a

portfolioofREITsthatneverenteranyoftheS&Pmarketindices.Weconstruct

thenonindexREITportfoliobothasavalue-weightedportfolioandasanequally

weighted portfolio. As in Equation (1), RS&P-REITs,tis the period t log return of

a value-weighted portfolio with all stocks in the relevant S&P index, excluding

any REITs. We continue to use the GMM method to produce test statistics in

order to account for serial correlation and heteroscedasticity.

Panels A and B of Table 2 report the results based on the value-weighted and

equal-weightednonindexREITportfolios,respectively.AsinTable1,thebetas

under the column heading “before” correspond to β2in Equation (3), while the

betas reported under the column heading “after” correspond to the period from

October 10, 2001 to December 31, 2001 and are the sum of β2and β3in

Equation (3). As with the index REITs, we find a significant increase in the

nonindex REITs’ betas with the index categories for both the daily and the

weekly return series.

The observed increase in the beta of nonindex REITs with the index category

is consistent with the spillover hypothesis that index-related investor sentiment

or market frictions affect the returns of nonindex REITs through the newly

included index REITs. This result avoids the debate on the information content

of the index addition event (i.e., informational vs. information free; see, e.g.,

Denis, McConnell, Ovtchinnikov and Yu (2003)), because we focus on the

REITs that are not added to the general S&P indices. Furthermore, our focus

on those REITs that are never added to the index makes it unlikely that the

Page 14

70

Ambrose, Lee and Peek

Table 2 ? Changes in beta of nonindex REITs.

With the S&P500With the S&P400 With the S&P600

Diff.

(p)

Diff.

(p)

Diff.

(p) Before AfterBefore AfterBefore After

Panel A. Beta of Nonindex REITs (Value Weighted)

Daily

Weekly 0.155

0.194 0.339 (0.000) 0.173

0.430 (0.046) 0.113

0.378 (0.000) 0.177

0.468 (0.007) 0.104

0.378 (0.000)

0.458 (0.004)

Panel B. Beta of Nonindex REITs (Equally Weighted)

Daily

Weekly 0.195

0.2200.376 (0.000) 0.194

0.458 (0.052) 0.142

0.420 (0.000) 0.212

0.497 (0.007) 0.144

0.431 (0.000)

0.495 (0.003)

Note: This table reports the beta of nonindex REITs (those that are never in any of the

three S&P market indices—S&P500, S&P400 or S&P600—during the entire sample

period from January 2000 to December 2003) with one of the market indices. The

“before” beta is for the period from January 1, 2000 to October 9, 2001, whereas

the “after” beta is for the period from October 10, 2001 to December 31, 2003.

Coefficient estimates are from a regression of the log return of the portfolio of nonindex

REITs on an intercept, a dummy variable for the period subsequent to October 9,

2001, the log return of the value-weighted S&P market index (S&P 500, S&P400

or S&P600) and their interaction term. The “before” beta is the coefficient for the

S&P index, and the “after” beta is the sum of the coefficient on the S&P index

and the coefficient on its interaction term with the post-October 9, 2001 dummy

variable. Observations are at either a daily or weekly frequency. Weekly returns

are from Tuesday close to Tuesday close, so that a new week starts on October

10, 2001. The S&P indices are reconstructed without any REITs. The estimation

period is from January 2000 to December 2003. We exclude September 2001 from

the estimation period. Autocorrelation- and heteroscedasticity-consistent p values in

parentheses are for the coefficient for the interaction term that tests for a difference

between the two betas. The p values for the coefficient for the S&P index (i.e., “before”

beta)aresuppressedtosavespace,buttheyarealwayssignificantatleastatthe5%level.

increase in the beta is due to the anticipation by investors that they will soon be

added to the general market indices.21

Still, one might be concerned that the beta increase may be driven, at least in

part, by factors other than the index-related sentiment or friction. For example,

21To further examine the anticipation effect, we constructed a portfolio of REITs that

are subsequently added to the index by keeping them in the portfolio only until their

actual additions. This portfolio shows a very similar increase in the beta with the index

category, suggesting that anticipation effects are not playing a major role; results are

available upon request. Although it is possible that all REITs, including those all-time

nonindex REITs, were subject to the anticipation effect, it would be more reasonable to

thinkthatthenumberofsubsequentadditionsisratherlimited,andthusanyanticipation

effect would be reflected in REITs that are subsequently introduced into the index

category.

Page 15

Comovement After Joining an Index 71

a general increase in the sensitivity of nonindex REITs to the general stock

market (which can be viewed as greater integration of the REIT sector with

the stock market in general) could lead to greater return comovement between

REITs and the S&P indices, because the indices and the broader market are

highly correlated. In addition, the beta increase could result from an increase

in marketwide shocks. Perhaps more importantly, the results in Table 2 do not

address possible endogeneity associated with the event in question (i.e., index

additions of several REITs to the index category). In the following subsec-

tions, we present a number of robustness checks in an attempt to rule out these

alternative explanations.

Robustness

Return Beta of Nonindex REITs with the General Stock Market

In this subsection, we examine whether possible increases in the sensitivity

of the nonindex REITs to marketwide shocks explain the prior result showing

an increase in their beta with the S&P indices. This alternative explanation

is plausible because the S&P index categories account for a large part of the

entire stock market. Our test for this explanation involves a two-step process.

First, we purge any marketwide shocks from the S&P indices by estimating the

following regression over the sample period:

RS&P−REITs,t= φ0+ φ1Rnon REIT,non S&P,t+ ηt,

where Rnon REIT,non S&P,trepresents the daily log return of the value-weighted

portfolio of all non-REIT common stocks that are not in any of the three S&P

market indices.22As before, RS&P-REITs,tis the daily log return of a value-

weighted portfolio with all stocks in the S&P index, excluding any REITs. The

residual,ηt,fromthisequationthusrepresentstheS&Pindex-relatedshocksthat

are orthogonal to broader marketwide shocks. In the second step, we estimate

the following regression:

(4)

Rnon index REITs,t= α + β1AFTER + β2ηt+ β3(AFTER × ηt) + εt,

where Rnon index REITsrepresents the log of the nonindex REIT returns, ηtis

the residual from the estimation of Equation (4) and AFTER is the dummy

(5)

22Theaveragenumberofstocksinthis“remainder”portfolioisslightlylessthan4,400,

and not surprisingly most of them (about 3,700 stocks) are small in the sense that their

market capitalization is below the 30th percentile of the New York Stock Exchange

(NYSE)stocks.However,theportfolioalsocontainsslightlymorethan100stockswhose

market capitalization is above the 70th percentile of the NYSE stocks. In addition, the

market capitalization of those large-cap stocks accounts for more than 50% of the total

market capitalization of this remainder portfolio, while the small-cap stocks compose

lessthan20%.Becausewearevalueweightingthosestocksinconstructingtheportfolio,

our results cannot be driven solely by small stocks.