Public Real Estate and the Term Structure of Interest Rates: A Cross- Country Study *

Transcription

1 Publc Real Estate and the Term Structure of Interest Rates: A Cross- Country Study * Alexey Akmov (correspondng author) Lancaster Unversty Management School, Department of Accountng & Fnance, Balrgg, Lancaster LA1 4YX, UK e-mal: a.akmov@lancaster.ac.uk Smon Stevenson Henley Busness School, Unversty of Readng, Whteknghts, Readng RG6 6UD, UK e-mal: s.a.stevenson@readng.ac.uk Maxm Zagonov Unversty of Toulouse, Toulouse Busness School, 0 Boulevard Lascrosses, Toulouse 31068, France e-mal: m.zagonov@tbs-educaton.fr Abstract Usng a varaton of the Nelson-Segel term structure model we examne the senstvty of real estate securtes n sx key global markets to unexpected changes n the level, slope and curvature of the yeld curve. Our results confrm the tme-senstve nature of the exposure and senstvty to nterest rates and hghlght the mportance of consderng the entre term structure of nterest rates. One ssue that s of partcular of nterest s that despte the fnancal crss the mportance of unantcpated nterest rate rsk weakens post 003. Although we examne a range of markets the emprcal analyss s unable to provde defntve evdence as to whether REIT and property-company markets dsplay heghtened or reduced exposure. Keywords securtsed real estate, nterest rate rsk, yeld curve modelng * Ths s the Authors Orgnal Manuscrpt of an artcle publshed n the Journal of Real Esate Fnance & Economcs on 13/01/015, avalable onlne: 1

2 Introducton The events of the last decade have brought nto sharp focus the mportance of the credt markets and how nterest rate dynamcs can feed through and mpact asset classes such as equtes. In a general context, Bernanke and Kuttner (005) argue that changes n nterest rates can affect stock prces n three prmary ways, namely the mpact on expected future dvdends, changes n the dscount rate and changes n the equty rsk premum. Much of the early lterature examned the overall equty market, or f sector specfc tended to concentrate on fnancal nsttutons (e.g., Chance and Lane 1980; Flannery and James 1984; Bae 1990; Madura and Zarruk 1995; Elyasan and Mansur 1998). However, n recent years an ncreasng number of papers have specfcally consdered and analysed publcly lsted real estate securtes such as REITs (Real Estate Investment Trusts). The corporate structure of publc real estate frms, together wth ther underlyng portfolos of real assets, provdes an deal forum to consder the senstvty of asset prces to changes n nterest rates. Ths s due to the multple channels through whch changes n nterest rates can mpact the operatng, fnancal and share prce performance of the frms. As a real asset, property s affected by the performance of the underlyng economy. Gven the key macroeconomc role that nterest rates have (Bernanke and Blnder 199), changes n rates may, f of suffcent magntude, have drect consequences on the operatng performance and cash flows of the frms. Ths may arse through the mpact on occupatonal demand and all of the related ssues that accompany ths, such as rental growth, occupancy and vacancy rates, and thus the overall ncome of the frms. Conceptual and theoretcal work, such as that of DPasquale and Wheaton (199) and Fsher (199) llustrate the nter-lnkages between economc actvty and the drect real estate market. Papers such as Lng and Naranjo (1997), and numerous papers subsequently, have emprcally demonstrated that nterest rates may represent a systematc rsk factor for real estate. Furthermore, nterest rate rsk can also be transmtted va the yelds used to captalse the ncome flows from the propertes underlyng the frms (Lzer and Satchell 1997). 1 1 Plazz et al. (008) examnes the dsperson across commercal real estate returns, fndng that rsk changes can be explaned by short-term nterest rates and a varable measurng the term-spread. 1

3 There are further reasons why real estate frms may respond dfferently to changes n monetary polcy compared to companes n other ndustres. One source of nterest rate exposure les wth the hgh degree of debt fnancng generally used n the real estate market (Bredn et al. 007). Ths s true not only for property companes but also for tax transparent vehcles such as REITs. Despte ther tax effcency REITs often contnue to employ substantal amounts of debt due to beng frms, havng hgher costs of equty, for growth and expanson purposes and for the opportuntes debt provdes to partcpate n large scale projects (e.g., Jaffe 1991; Chan et al. 003; Oo et al. 010). Accordngly, due to ther extensve use of debt, unexpected nterest rate fluctuatons wll affect the frms cost of fnance and market values. In addton, the use of leverage alters a company s cost-of-captal and therefore can affect the future avalablty of external debt facltes. Subsequent nterest rate changes and the nteractons between frms nvestment and fnancng actvtes may therefore be reflected n a company s share prce (Bernanke and Gertler 1995). Another reason for the potentally heghtened exposure of real estate stocks to nterest rates s specfc to the REIT sector and s concerned wth the mnmum mandatory dvdend payment. Gven the hgh-yeld status of REITs the mpact of nterest rate fluctuatons on the present value of dvdends s lkely to be greater n a REIT context (Bernanke and Kuttner 005). Gven the mportance of nterest rates, n both the specfc context of publc real estate and n the broader settng of the underlyng drect real estate market, ths paper emprcally address the senstvty of the publc real estate markets to nterest rate rsk and specfcally whch nterest rate factors are the most sgnfcant n terms of both returns and volatlty. We adopt a framework that captures the dynamcs of the entre term structure wth three latent factors, namely level, slope, and curvature. The results show the sgnfcance of both short and long-term nterest rates factors n explanng the returns of publc real estate. Not only the yeld curve parameters are sgnfcant n explanng returns, but n addton, volatlty s strongly and negatvely affected by changes n the yeld curve factors. Although the results vary across markets and over tme, they hghlght the mportance of consderng nterest rate rsk usng the entre term structure rather than concentratng on a sngle maturty. The study dffers from the prevous lterature n a number of respects. Frstly, unlke prevous contrbutons to the lterature we do not lmt our analyss to consder the senstvty to fluctuatons n a sngle nterest rate factor. Rather, we use a dynamc verson of the Nelson-Segel model whch captures the dynamcs of the entre term structure of nterest rates wth three latent factors representng the level, slope, and curvature of the yeld curve. Interest rate

4 changes of dfferent maturtes are not ndependent due to the polcy transmsson mechansm from short to longterm rates. Not only do prevous studes suffer from the lmtaton that nterest rate proxes cannot be smultaneously ncorporated n a sngle model, but also the ratonale behnd the selecton of a specfc proxy s often of an arbtrary nature. However, Debold et al. (005) argue that nformaton on a set of nterest rates can be summarsed n a number of systematc rsk factors. Ths approach s superor to usng the term spread as t allows non-parallel shfts to be captured. Our study also extends prevous contrbutons by consderng the senstvty of publc real estate to unexpected nterest rate movements. Prevous studes n both the general fnance and real estate specfc lterature have largely consdered actual changes n nterest rates. Studes such as Bredn et al. (007, 011) and Xu and Yang (011) have, however, llustrated the mportance of consderng unexpected changes n both nterest rates and monetary polcy. Fnally, the paper extends a predomnantly U.S. focused lterature to examne sx key global markets, namely Australa, Hong Kong, Japan, Sngapore, the U.K. and the U.S. These markets account for approxmately 80% of nvestable publc real estate n terms of market captalsaton and volume traded (Psalts and Chubb 008). Ths sample also has the advantage n that t ncorporates a mxture of regulatory structures. Two of the markets (Australa, U.S.) operate under a REIT regme throughout the entre sample perod. The remanng markets have ntroduced a REIT structure durng the tme perod consdered n the study. However, the relatve mportance of REITs versus conventonal property companes dffers. For example, n Hong Kong the majorty of frms, ncludng the largest, have retaned a corporate structure. In contrast, n the U.K. the majorty of the large cap property companes converted to REIT status. It wll therefore be of nterest to see whether there are dscernble dfferences n the emprcal fndngs across the markets. The remander of the paper s structured as follows. The next secton dscusses some of the pertnent lterature and n dong so presents the core underlyng hypotheses to be consdered n the paper. Followng the lterature revew, we detal the data and present the adopted methodologcal framework. The man emprcal results and a number of robustness tests are then dscussed, pror to provdng concludng comments and observatons. Related Lterature and Hypothess Development As noted n the ntroducton, there have been an ncreasng number of papers over the course of the last decade to have examned the senstvty of publc real estate to nterest rates. Although the exact methodologcal framework 3

5 has dffered there are some commonaltes n the emprcal results reported. A number of papers have reported that REITs are predomnantly exposed to longer-term rates, for example Devaney (001) fnds heghtened senstvty to 10-year government bond yelds. Smlar fndngs are also reported by Chen and Tzang (1988) and He et al. (003). However, He et al. (003) emphasse that ther conclusons may be based by the nterest rate proxy chosen. Aganst ths background, Allen et al. (000) report sgnfcant senstvty of REITs to both short and long-term government bonds, as well as the stock market. They further argue that frms can adjust ther exposure to the market by changng fnancal leverage, but not the exposure to nterest rates. In a smlar ven, Stevenson et al. (007) use nterest rate proxes of three maturtes, namely the 1-month nterbank rate and the 10 and 15-year government bond yelds. Ther analyss of U.K. property companes s conducted for a perod charactersed by hstorcally low and stable nterest rates, yet the authors report sgnfcant results for all of the proxes examned, although the coeffcent on the short-term rate s of the unexpected postve sgn. Broadly, these fndngs are consstent wth Lzer and Satchell (1997) who estmated the response of U.K. property stocks under two dstnct regmes, fndng that frms were partcularly senstve durng the low nterest rate perod. In contrast, durng the hgh nterest rate regme property companes experenced sharp falls n value, felt to reflect ncreased uncertanty. Ths tme-varyng nature of the senstvty to nterest rates s also found n a number of studes that have adopted a GARCH (Generalsed Autoregressve Condtonal Heteroskedastcty) framework to consder the mpact on returns and volatlty (Devaney 001; Stevenson et al. 007). Furthermore, Low et al. (003), n ther analyss of the Sngaporean market, dentfy a systematc relatonshp wth nterest rate rsk but note that the prcng of that rsk s subject to market condtons. In addton to the aforementoned studes there s also a strand of the lterature that has adopted a dfferent methodologcal approach, one that comes from an asset prcng-factor model framework. Lang et al. (1995), for example, estmate a two-factor model that ncludes an nterest rate varable and the overall stock market. They observe only weak senstvty to government bond returns of dfferent maturtes. Studes such as Glascock et al. (000) have observed a reduced senstvty of U.S. Equty REITs to government bonds durng what s often referred to as the modern REIT era. Based on contegraton tests, the paper reports sgnfcant common long-term trends between REITs and bond returns over the full sample horzon, a result whch does not however hold post-199. In ths respect papers such as Glascock et al. (000), Swanson et al. (00) and Cheong et al. (009) all argue that structural shfts n the nterest rate senstvty of REITs have occurred at varous ponts n tme. Nonetheless, the two latter papers fnd that these shfts do not affect REIT's nterest rate senstvty generally and both document the 4

6 economc sgnfcance of nterest rates. It s however mperatve that a dstncton s made between papers such as Glascock et al. (000) and Cheong et al. (009) who analyse fxed ncome ndex data and studes such as Devaney (001), He et al. (003) and Stevenson et al. (007) that base ther analyss on ether market nterest rates or bond yelds. Whle the aforementoned studes provde valuable nsghts nto the nterest rate exposure of publc real estate, they also frequently provde conflctng results due to the nconsstent choce of nterest rate proxy. Not only s the ratonale behnd the selecton of a specfc proxy often of an arbtrary nature, but prevous studes suffer from the lmtaton that nterest rate proxes cannot be smultaneously ncorporated n a sngle model. Furthermore, nterest rate changes of dfferent maturtes are not ndependent due to the polcy transmsson mechansm from short to long-term rates. Papers such as Swanson et al. (00) and He et al. (003) ncorporate a term-spread varable nto ther analyss. Used as a proxy for changes n the term structure, the varable s defned as the spread between long and short term nterest rates. Whlst t does capture some elements of the yeld curve, t does not necessarly reflect some mportant features of the term structure that can commonly occur. In partcular, such a measure does not capture changes n the shape of the yeld curve (e.g., S-type or humped shape) and other non-parallel shfts. Debold et al. (005) argue that nformaton concernng bond prcng s concentrated n a number of systematc rsk factors, whch tself mples a hgh correlaton between nterest rates of dfferent maturtes. A possble soluton s found n the yeld curve lterature. Nelson and Segel (1987) suggest an exponental components model where a large range of nterest rates can be descrbed through just three factors. In ths paper, we use the dynamc verson of Nelson-Segel model, as proposed by Debold and L (006): y t t t 1 e 1 e t 1 t t 3t e (1) t t Ltterman and Shenkman (1991) show that nformaton about of large set of nterest rates can be summarzed usng a number of common factors. Based on a Prncpal Components framework, the authors dentfy these factors as the term structure level, steepness, and curvature. 5

7 where y t ( ) s the yeld of a zero-coupon bond, wth tme-to-maturty τ at t. 1t, t, 3t are the parameters whch represent the level, slope and curvature of the yeld curve respectvely. Fnally, λ represents the exponental decay rate. Based on the model parametersaton above, the loadng on the level s equal to 1 and s 1t ndependent of the tme-to-maturty. Takng the lmt, t s easy to see that lm yt ( ) 1 t t and hence the yeld curve level can be seen as the long-term factor. A decrease n 1t would affect all yelds dentcally, thereby shftng the level of the yeld curve. The loadng on the slope parameter t s drven by the exponental functon, startng at 1 and decreasng monotoncally to zero wth ncreasng maturty. Therefore, the slope parameter may be seen as a short-term factor, and hence an ncrease n ths factor would amplfy short-rates more than long-term ones. In mathematcal terms, gven lm yt ( ) 1 t t t 0 y, t s easy to see that t ( ) yt ( 0) t. The loadng on the last parameter 3t (curvature) s also drven by the exponental functon, whch starts at zero (wth maturty 0 ), ncreases for the medum-term maturtes, and then decays back to zero as maturty further ncreases. Accordngly, 3t the yeld curve curvature can be seen as the medum term factor. The Nelson-Segel model fts the term structure usng a flexble, smooth parametrc functon based on a Laguerre functon. The model uses a number of parameters and provdes enough flexblty to capture a range of monotonc, S-type and humped shapes typcally observed n yeld curve data. Due to ts ablty to provde a good ft of the yeld curve, the model s advocated by Debold and L (006), Fabozz et al. (005), and t s wdely used by Central Banks (Bank for Internatonal Settlements 005). Snce the value of any stock, ncludng REITs and property companes, can be consdered as beng the present value of future cash-flows, a downward shft n the term structure of nterest rates, as captured by our level factor, should lead to hgher prces. We would also expect an nverse relatonshp between the slope, our proxy for short-term nterest rates, and prces. Consstent wth the deas of Mshkn (1996) and Renhart and Smn (1997), short-term rates wll be heavly nfluenced by Central Banks through nflaton targets and the settng of prme/base rates. Exstng emprcal evdence, that has documented an nverse relatonshp wth nterest rates of maturtes rangng from 1 month to 0 years, would support ths premse (e.g., Chen and Tzang 1988; Devaney 001; Swanson et al. 00; Stevenson et al. 007). However, none of these studes consder a range of nterest rates jontly. Therefore, usng the factors we extract from the dynamc Nelson-Segel model, we test the followng hypothess: 6

8 H 1 : There s an nverse relatonshp between publc real estate returns and the changes n level, slope and curvature of the yeld curve. One of the key features that dfferentates REITs from both property companes specfcally and stocks generally s the mnmum mandatory dvdend payment that s requred n most jursdctons. As prevously noted, ths hgh yeld nature would be expected to make REITs more senstve to nterest rates n the context of dscountng future expected dvdends (Bernanke and Kuttner 005). Furthermore, hgh dvdends and the coupon-lke nature of the underlyng rental ncome have been frequently dentfed as key reasons why REITs are felt to share some characterstcs wth fxed-ncome securtes (Cheong et al. 009). Gven these ssues we therefore antcpate a more pronounced reacton to changes n nterest rates n markets where a REIT regme was n place throughout the sample perod. Hence, we consder the followng hypothess: H : REIT shareholders should receve hgher compensaton for nterest rate exposure than both property specfc and general stock market holders. The results wll allow us to consder whch markets are more consstently exposed to nterest rate rsk. We also can examne whether countres wth establshed REIT markets are exposed to fluctuatons n nterest rates to a greater and more consstent degree. Data, model desgn and descrptve statstcs Interest Rate Data and Transformatons As dscussed n the prevous secton, one of the key objectves of the current paper s to examne the relatonshp between the yeld curve and publc real estate. Debold and L (006) propose a varaton of the Nelson and Segel (1987) parametrc approach to modellng the yeld curve. To estmate our nterest rate rsk factors (level, slope and 7

9 curvature), we obtan from Bloomberg a seres of estmated soveregn bond zero-coupon yelds for twelve dfferent maturtes (3, 6, 1, 4, 36, 48, 60, 7, 84, 96, 108, 10, 40, 360 months). Bloomberg derves the zero-coupon yelds for the dfferent maturtes from observed bond prces by strppng the coupon bearng government notes and bonds usng a bootstrappng approach. 3 Descrptve statstcs are reported n Table 1. Yelds are then used as the ntal estmates on the left-hand sde of the Equaton (1). Followng Debold and L (006), Fabozz et al. (005) and Czaja et al. (009), we obtan estmates for the level, slope and curvature as follows. The dentfed seres of zero-coupon yelds are regressed on the factor loadngs and a constant usng cross-sectonal OLS (Ordnary Least Squares). Wth ths model parametersaton the factor loadngs on the rght-hand sde are calculated assumng a prefxed value of the decay parameter λ. Tam and Yu (008) and Banch et al. (009) retan the same value of the parameter as suggested n Debold and L (006), namely , reportng that the model satsfactorly fts the tme-seres for the countres they examne. A decay parameter of s equvalent to maxmsng the curvature, or medum-term factor, at 30 months. Yu and Salyards (009), however, llustrate usng a sample of both nvestment-grade and speculatve-grade bonds, that the optmal value of λ can vary substantally for dfferent set of bonds. Fgure 1 demonstrates how the factor loadngs n the model can be affected by the choce of λ. INSERT TABLE 1 INSERT FIGURE 1 Fg. 1 Factor loadngs of Nelson-Segel usng dfferent λ values In lght of ths varaton we evaluate the underlyng model separately for each country, n lne wth the approach adopted by Yu and Salyards (009). 4 Frstly, we estmate the model usng dfferent values of λ. Snce the λ 3 Bloomberg s soveregn bond yelds are generated by the Bloomberg Far Value (BFV) model. Ths model utlses well-prced (lqud) bonds wth smlar characterstcs such as ssuer, credtworthness, and embedded optons. Our sample utlses government ssued securtes whch are opton-free. The model also excludes same sector bonds wth sgnfcantly hgher or lower z-spreads. Therefore, concerns regardng thn tradng are mnmzed for our sample. 4 It should be noted that the varaton of the Nelson-Segel model used n ths study assumes a constant λ parameter across tme. Our approach closely follows the ntuton of Debold and L (006), where λ s set to be lnked to the 8

10 parameter corresponds to the maturty at whch the curvature, or medum-term factor, s maxmsed, we test the medum-term maturtes wthn the range we have avalable. The model wth the hghest goodness-of-ft s then selected. For Japan, Sngapore, the U.K. and the U.S. the best ft s found wth λ=0.0374, whch s equvalent to the curvature factor beng maxmsed at 48 months. For Hong Kong the λ fgure s , whch corresponds to 4 months. Only n the case of Australa do we obtan a λ parameter equal to that used n the Debold and L (006) paper. Ths not only supports the arguments of Yu and Salyards (009), but also justfes our decson to estmate λ for each country ndvdually. 5 In Fgure, we plot the yeld curve for randomly selected dates from the U.K. sample. The graph llustrates the ablty of the model to ft dfferent shapes of the yeld curve. The best ft s found n the normal shape yeld curve, wth the R beng 0.99 and the ftted lne beng almost ndstngushable from the orgnal. A more modest ft s notceable n humped, S-type and nverse yeld curves. Overall we fnd that on average, based upon the reported R fgures, the Nelson-Segel model explans 90% for each country n our sample. However, whlst the R does vary for the U.K. and U.S., t remans consstent throughout n the case of Japan. Fgure 3 graphcally dsplays the tmeseres of the three estmated parameters. INSERT FIGURES Fg. Yeld curves for selected dates alongsde ftted (model-based) counterparts medum-term nterest rate maturty (e.g., 30 months). One can estmate the model wth a tme-varyng λ smlar to Koopman et al. (010). However, f one employs a tme-varyng λ as a fourth latent factor, the medum-term parameter (curvature) would be no longer lnked to the observable medum-term maturty. To avod ntroducng an addtonal tme-varyng component n the yeld-curve model we opt to use the fxed λ model specfcaton. 5 In an earler verson of the paper, we estmated the model usng the approach of Debold and L (006), assumng a consstent decay parameter of Both specfcatons yeld statstcally dentcal results wth regard to the relaton between the Nelson-Segel factors and publc real estate markets. However, the selecton approach provdes a greater degree of confdence n the ndvdual country s results. The results from the earler verson of the paper are avalable from the authors upon request. 9

11 INSERT FIGURES 3 Fg. 3 Evoluton of the Nelson-Segel parameters, As suggested by Wllner (1996) and Czaja et al. (009) we convert the estmated factors nto ther frst dfferenced form r r r 1. Our study also extends prevous contrbutons by consderng the senstvty of publc real estate to unexpected nterest rate movements. Prevous studes n the real estate lterature have only consdered actual nterest rate changes. Ths s also true n the general fnance lterature, where only a number of papers have based ther emprcal tests on unexpected nterest rate changes (e.g., Flannery and James 1984; Bae 1990; Madura and Zarruk 1995; Faff and Howard 1999). The ratonale behnd the examnaton of unexpected changes s twofold. The frst ssue s based on the fact that f nomnal changes n rates are examned the analyss effectvely revolves around nflaton expectatons gven that the volatlty of real nterest rates s relatvely (Fama 1975; Nelson and Schwert 1977). Ths argument s reflected n prevous studes where researchers have employed a myrad of dfferent expectaton generatng processes to extract nterest rate nnovatons (Bae 1990; Madura and Zarruk 1995). Secondly, studes n the manstream lterature such as Bernanke and Kuttner (005) have hghlghted the mportance of adjustng for expectatons when consderng nterest rates and monetary shocks. These fndngs have also been supported n the context of both the U.S. REIT sector (Bredn et al. 007, 011) and global real estate markets (Xu and Yang 011) We estmate the unexpected changes n the yeld curve factors usng an Autoregressve Movng Average (ARMA) model. We model the unantcpated changes n the term structure factors as the dfference between the actual changes n the respectve factor at tme t and ones forecasted va an approprate ARMA(p,q) model: Lr,t Lu () s the set of Nelson-Segel factors, L and L where r,t 1t, t, 3t are the polynomal lag operators: L p p L 1 L L (3) 1 10

12 L q q L 1 L L (4) 1 hence, L,S, C f are the unexpected changes n level, slope and curvature of the yeld curve. The use of only u, t one expectaton generatng process s supported by the fndngs of Bae (1990) who, usng three alternatve models to form expectatons, fnds dentcal results regardless of the model employed. 6 The approprate order of the autoregressve and movng average parameters used range from 0 to 5, dependng on the varables and markets consdered. One further adjustment s made to the nterest rate data. In order to assess the ndvdual mpact of each factor on returns we use an orthogonalsaton procedure smlar to that adopted by Czaja et al. (009). It recognses the level factor L for each country as the prmary drver of yeld curve changes, wth the slope C ranked second and thrd n mportance respectvely: S and curvature S C 1 1 L ŝ (5) L 3ŝ ĉ (6) We therefore orthogonalse the slope factor wth respect to the level, whlst also orthogonalsng the curvature wth respect to both yeld curve level and slope. The orthogonalsed seres L,ŝ, ĉ are then used as the exogenous varables n the emprcal framework detaled n the model desgn secton below. Publc Real Estate, Stock Market and Other Control Varable Data Our analyss s based on sx of the largest publc real estate markets n terms of market captalsaton and tradng volume, namely Australa, Hong Kong, Japan, Sngapore, the U.K. and U.S. The sample data s weekly n frequency and covers the perod January 1995 to December 01. Publc real estate ndces used are collected from the FTSE 6 See also Dnens and Stakouras (1998) who consder fve alternatve models to estmate unexpected nterest rate changes. 11

13 EPRA/NAREIT database. 7 Apart from the U.S. and Australa, the ndces used are not solely comprsed of REITs. Ths s especally so n the case of Hong Kong, where even after the ntroducton of REIT legslaton n 003 and the lstng of the frst trust n 005 (The Lnk REIT), property companes reman the domnant vehcle. For the remanng countres REIT regulatons came nto beng n 000 (Japan), 00 (Sngapore) and 007 (U.K.) respectvely. As well as the date at whch REITs were ntroduced there are also dfferences across the sx markets n terms of the amount of leverage REITs take on, as Fgure 4 llustrates. In addton, the specfc REIT regulatons vary across the sx markets. Whlst the majorty do have some lmt on the amount of debt a REIT can employ, there are no such restrctons n ether Australa or the U.S. However, as can be seen from Fgure 4, frms n those markets have lower leverage, on average, than n some countres wth a set lmt. In addton, the form of any restrcton vares. Whlst U.K. frms are restrcted on a cash flow bass by a mnmum Debt Servce Coverage Rato of 1.5, n other markets lmts are based on balance sheet data. For example, gearng s lmted n Sngapore to a maxmum of 35% of total captal. 8 Throughout the analyss we employ broad equty market ndces as control varables. The ndces used are the most common equty market benchmarks n each country, namely ASX00 (Australa), Hang-Seng (Hong Kong), Nkke 5 (Japan), Strats Tmes (Sngapore), FTSE-100 (U.K.), and the S&P500 (U.S.). The summary statstcs for the publc real estate and equty markets, together wth the nterest rate factors are reported n Table. Publc real estate delvered a better return performance over the perod of study, whle exhbtng hgher volatlty. A lack of normalty s a common feature n fnancal tme-seres and we observe t for almost every varable of nterest. In partcular, 7 An earler verson of ths paper used data from SNL Fnancal for the U.S. and Thomson Reuters Datastream for the remanng markets. The results appear to be robust and are not dependent on the choce of ndex provder. These fndngs support those of Serrano and Hoesl (009) who report hgh average correlatons between the man provders of publc real estate ndces. Serrano and Hoesl examned the 1990 to 007 perod and examned ndces produced by FTSE/EPRA NAREIT, Thomson Reuters Datastream, GPR and S&P/Ctgroup. 8 Ths fgure may be ncreased to as hgh as 60% f certan condtons, such as credt ratngs beng obtaned, are met. 1

14 most varables exhbt statstcally sgnfcant skewness and excess kurtoss, resultng n a formal rejecton of the Jarque-Bera normalty test. INSERT TABLE INSERT FIGURE 4 Fg. 4 REIT/publc real estate leverage (00-010). Source: SNL Fnancal Model Desgn The methodologcal desgn we adopt n ths paper s based upon the GARCH specfcaton frst adopted n the context of nterest rate senstvty by Elyasan and Mansur (1998) and subsequently used n REIT/publc real estate specfc papers such as Devaney (001) and Stevenson et al. (007). Let r, denote the return on the publc real t estate ndex of country. The GARCH(1,1) model can be descrbed as follows: r, 1, L, ŝ 3, ĉ 4,,t M 0 r (7) z (8) 0, 1, 1 1, 1 a a b (9) where r s the domestc equty ndex returns, L,ŝ, ĉ, t represent the set of unexpected Nelson-Segel factors. M z s a sequence of ndependent, dentcally dstrbuted random varables wth zero mean and unt varance, Φ s a condtonally normal heteroscedastc error term. Φ 1 s the nformaton set mplyng ~ N,t 1 0, avalable at tme t 1. 1,,,, 3, represent the exposure of the real estate sector to nterest rate rsk. The coeffcents are estmated usng a quas-maxmum lkelhood procedure wth the normal lkelhood functon and robust standard errors as suggested by Bollerslev and Wooldrdge (199). 13

15 The general GARCH framework also allows us to examne the mpact of exogenous varables on the total rsk of real estate frms, as measured by the condtonal varance. We therefore extend the model (7) (9) usng two alternatve condtonal volatlty specfcatons wth respect to nterest rate varables. The frst specfcaton uses unexpected changes n nterest rates (Glosten et al. 1993). The second uses the condtonal volatlty of the nterest rate seres (Elyasan and Mansur 1998; Stevenson et al. 007): ˆ (10a) ˆ a0, a1,, t1 b1,, t1 g1, L, t1 g, s, t1 g3, c, t1 ~ ~ ~ (10b), t level slope curv a0, a1,, t1 b1,, t1 g1, h, t1 g, h, t1 g3, h, t1 Overall, the model offers the opportunty to examne a number of hypotheses regardng publc real estate returns and volatlty behavour. Frstly, Equaton (7) ncludes three nterest rate proxes, namely the level, slope and curvature. Usng a smple Wald coeffcent restrcton test we examne whether returns are senstve to all three yeld curve factors jontly (.e., 0 ). Secondly, we can use the augmented varance specfcatons (Equatons 1,, 3, 10a, 10b) and test whether our nterest rate proxes or ther condtonal volatltes have a sgnfcant jont effect on the volatlty of publc real estate: g 1, g g 0 and g ~ 1, g ~ g ~ 0 t,t 3, t t,t 3, t respectvely. Fnally, we examne whether our set of nterest rate factors affect jontly the returns volatlty of publc real estate: 1, g g g 0 and 1, g ~ g ~ g ~ 0., 3, 1,t,t 3, t, 3, 1,t,t 3, t Emprcal Results Publc Real Estate and Interest Rate Rsk We ntally estmate the four-factor GARCH model, as specfed n Equatons (7)-(9) for each market, wth the results reported n Table 3. We run the emprcal tests usng two equal sub-samples, wth a md-pont of year-end 00, n addton to the overall sample perod. For the full perod of study, we fnd that the level (the long-term nterest rate proxy) sgnfcantly nfluences publc real estate n both Australa and Sngapore, whlst Hong Kong and the U.S. are sgnfcantly senstve to the slope factor. In all cases the sgnfcant coeffcents are of the antcpated (negatve) sgn. Our results are consstent wth prevous REIT papers (Allen et al. 000; Devaney 001), that report 14

16 sgnfcant exposure to both short and long-term nterest rates. However, none of the analysed markets are sgnfcantly exposed to movements n the curvature factor, whch s the medum-term nterest rates proxy. One fndng that s of partcular nterest s that Japan exhbts postve senstvty to the slope factor over both the entre sample perod and n the frst sub-sample, mplyng that Japanese real estate stocks ncrease (decrease) n value followng an ncrease (decrease) n the short-term nterest rate. Such fndngs may, at least n part, be related to the prolonged perod of economc stagnaton and near-zero nterest rate polcy adopted n Japan durng the years we analyse. Therefore, an ncrease n nterest rates may have acted as a postve sgnal to the market f t was assocated wth mproved expectatons regardng economc condtons. INSERT TABLE 3 Another fndng that s qute notceable s that the results across the two sub-samples dffer qute substantally. Indeed, the Australan REIT market s the only one to be consstently affected by the changes n the yeld curve level at a statstcally sgnfcant level. Australa s also the only market to report jont sgnfcance of the three yeld curve factors across the dfferent sample perods. We however, fnd that the U.S. s senstve to all three factors n the second subsample. In contrast, the prevously reported exposure of the Hong Kong market to the changes n the slope factor s drven by the former sub-perod. Consstent wth Devaney (001) and Stevenson et al. (007) we confrm the tme-varyng nature of the nterest rate senstvty reported. Whle most prevous studes n the area looked at ndvdual markets and a sngle nterest rate proxy, we fnd that often nterest rate senstvty may occur n some markets (.e. the U.S. and the U.K.) and dsappear n the others (.e. Sngapore, Japan and Hong Kong). It should be noted that our methodologcal framework effectvely consders the excess nterest rate exposure of publc real estate, over and above the general equty market. Therefore, the results do not unequvocally mply that the publc real estate sectors n Hong Kong, Japan, and Sngapore dsplay no exposure to nterest rate movements n the latter sub-perod. In addton, there are other potental reasons behnd the fndngs reported. For nstance, the lack of sgnfcance n Sngapore may be possbly assocated wth the unque monetary polcy settng n the country. Durng the 1980s the Monetary Authorty of Sngapore (MAS) changed ther polcy objectve to controllng nflaton through the management of the Sngapore dollar aganst a basket of currences known as the trade-weghted exchange rate ndex (TWI). The TWI s now the man nstrument n conductng monetary polcy n Sngapore 15

17 nstead of the tradtonal short-term nterest rate. Exceptons do occur and are prmarly durng perods when the authortes needed to ensure suffcent lqudty n the money markets. Our fndngs are smlar n many respects to those reported n prevous studes. For example, the U.K. results support the fndngs of Stevenson et al. (007), n observng a postve but nsgnfcant relatonshp between property company returns and short-term nterest rates and a sgnfcant nverse relatonshp wth long-term rates. In the U.S. the mpact of long-term rates was prevously noted by Devaney (001) for Equty REITs. Overall, our fndngs suggest that nterest rates exposure should be modelled beyond a sngle nterest rate factor, and the factors from the Nelson-Segel yeld curve model are good canddates for such task. Addtonally, the results are consstent wth the general theoretcal formulaton of the relatonshp between stocks/reits and nterest rates. However, t would be hard to descrbe the results as homogeneous, whch s not surprsng gven the structural and legal dfferences across the nternatonal markets consdered. To fully consder the mpact of nterest rates on the volatlty of publc real estate we test the two augmented GARCH specfcatons detaled n Equatons (10a) and (10b). These specfcatons augment the base model wth frstly the one-perod lagged unexpected change n the level, slope and curvature factors and secondly ther respectve one-perod lagged condtonal volatltes. 9 The results are reported n Table 4 and t s evdent that volatlty s strongly and, n the man, negatvely affected by changes n the yeld curve factors n Australa, U.K. and U.S. Ths would suggest that unantcpated changes n the yeld curve factors translate nto lower volatlty of real estate returns n the subsequent perod. Effectvely, there s less volatlty clusterng snce the markets adjust ther expectatons more effcently usng nformaton from the money markets. Contrary to the results from the mean equaton (Table 3), the curvature factor s found to be consstently relevant n explanng volatlty behavour n the U.K. and the U.S. Ths may be due to the use of dervatves to hedge nterest rate exposure. Horng and We (1999) argue that nterest rate hedgng s the prmary reason for REITs employng dervatves. Ths s possbly due to 9 For the sake of brevty, we do not report the results for the mean equaton from the augmented volatlty GARCH models. The results are very smlar to those obtaned wth the base four-factor GARCH model reported n Table 3 and prevously dscussed. The complete results are avalable from the authors on request. 16

18 REITs havng no ncentve to use dervatves for tax purposes. However, the evdence would appear to ndcate that the use of dervatves for hedgng purposes s n realty relatvely lmted. INSERT TABLE 4 Wth the excepton of Sngapore, we observe that the mpact on volatlty shfts from short-term rates to long-term rates and vce versa. Ths mples that nformaton about the yeld curve factor changes s an mportant determnant of volatlty. Moreover, the mpact may be hard to capture usng conventonal sngle maturty models as the relatonshp s tme-varyng. Therefore, we test whether our nterest rate proxes, and ther condtonal volatltes, have a sgnfcant jont effect on volatlty. Wald test results on the jont sgnfcance of the yeld curve factors are sgnfcant. We also fnd that real estate frms n all sx markets are affected drectly by the factors changes rather than ther volatltes. Asymmetrc Responses to Changes n the Yeld Curve Gven that our ntal fndngs do ndcate a degree of tme-varaton we expand the analyss to consder whether there are possble asymmetres present n how returns respond to the yeld curve factors. Based upon the analyss contaned n papers such as Bernanke and Kuttner (005), Bredn et al. (007) and Xu and Yang (011) we conduct two alternatve tests, whch consder whether the senstvty observed s dependent on frstly the sgn and secondly the sze of the unexpected change n the yeld curve factors. For the frst test, consderng the sgn of the unantcpated change, we construct two dummy varables for each yeld curve factor as follows: varables L S C D, D, D take the value of 1 f there s a postve change n the level, slope or curvature respectvely and 0 otherwse; varables L S C D, D, D take the value of 1 f there s a negatve change n the respectve yeld curve factors and 0 otherwse. Therefore, we test the followng specfcaton 10 : 10 The use of sgn dummes n nteracton wth the yeld curve factors resolves the multcollnearty problem commonly assocated wth the use of dummy varables. 17

19 r L L S S 0, 1, L 1, L, S, S (11) C C M 3, C 3, C ψ r a 0, a g 1,, S 1 1 b 1, S 1 1 g, g S 1, L 1 1 S 1 L 1 g 3, g C 1, L 1 1 C 1 L 1 g 3, C 1 C 1 (1) The estmaton results from Equaton (11) are reported n Table 5. Overall, the evdence wth respect to asymmetry s strongest n Australa and the U.S., wth evdence ndcatng a sgnfcant reacton to all three yeld curve factors. We fnd that U.S. REITs and the Hong Kong sector both react negatvely to unexpected ncreases n the slope of the yeld curve, whch s assocated wth short-term rates. We fnd that postve changes n the slope factor, whch would mply a contractonary monetary polcy, drve prces downwards. Therefore, our ntal multfactor model results for U.S. REITs, as reported n Table 3, are lkely to be drven by the flatter slope of the yeld curve, whle the steeper slope, as an ndcaton of economc expanson, has lttle effect on REIT returns. In Australa and Japan we fnd the opposte effect for postve slope changes. Returns n Australa also react negatvely to ncreases n the level, consstent wth the ntal results for Australa and the vew that they are drven by unexpected ncreases n short and long-term rates rather than reductons. INSERT TABLE 5 The second test consders whether the sze of unexpected changes n the yeld curve factors affects the nterest rate senstvty observed. Xu and Yang (011) provde some evdence n support of ths vew n ther analyss of how U.S. monetary shocks can affect nternatonal publc real estate markets. To analyse ths ssue we sort the absolute values of the changes n each yeld curve factor n ascendng order and assgn two dummy varables, f, D and f, D, for each factor f. D f takes the value of 1 f the yeld curve factor change s greater than the medan factor change n absolute terms and zero otherwse, D f, 1 D f,. As a result, our model takes the followng form: 18

20 r 0, 1,, L S L, S, 1, 3, L C L, C,, S 3, C S, C, M ψ r (13) a 0, a g g g 1, 1,, 3, L S C b 1, L, 1 S, 1 C, 1 1 g g g 1,, 3, L S C L, 1 S, 1 C, 1 (14) The results are reported n Table 6 and reveal an nterestng dfference between the response n returns and volatlty. It can be seen that volatlty s partcularly senstve to changes n the yeld curve factors, wth sgnfcant results reported for Australa (slope and curvature), U.K. (slope and curvature), Hong Kong (level), and Sngapore (slope). In contrast, returns are prmarly affected by the magntude of the change n the yeld curve factors. In partcular, large unexpected changes n the level are nversely related to returns n Australa and Sngapore. Furthermore, large unexpected changes n the slope affect Hong Kong, Sngapore and the U.S. Our results are consstent wth those reported by Xu and Yang (011) who fnd that nternatonal publc real estate markets returns are mostly exposed to large changes n monetary polcy. When we consder the Wald tests, the estmates for the large unexpected changes n the level and the slope of the yeld curve are sgnfcantly dfferent from the coeffcents relatng to changes. Fnally, t s of nterest that n contrast to the base lnear model, reported n Table 3, we do n ths nstance report statstcally sgnfcant nterest rate exposure n the case of Sngapore. The orgnal specfcaton faled to detect any sgnfcant nterest rate senstvty. INSERT TABLE 6 Robustness Tests The fnal component of the emprcal analyss conssts of a varety of alternatve specfcatons and tests n order to consder the robustness of the results reported thus far. In the frst nstance we ntroduce two addtonal control varables, namely global stock market returns and tradng volume. The global equty market s proxed by the MSCI World ndex returns r and s ncluded n addton to the domestc equty market benchmarks used thus far. The W tradng volume data s collected from Thomson Reuters Datastream. Consstent wth prevous studes, we fnd that 19

21 tradng volume has a substantal tme trend. Therefore, n order to ensure that the data s statonary we follow prevous studes such as Campbell et al. (1993) and Hutson et al. (008) and de-trend the data as follows: log( Volume ) a0 a TIME a TIME u (15), 1,, The de-trended non-negatve volume seres vol s defned as the resduals u. We augment the mean equaton of our baselne GARCH specfcaton as follows: u mnus the mnmum value of M W 0, 1, L, ŝ 3, ĉ 4, r 5, r 6, vol (16) r The results are presented n Table 7 and are broadly consstent wth the ntal fndngs that were reported n Table 3. The specfcaton that ncludes the global market returns and volume reveals that changes n the level of the yeld curve have a stronger mpact on U.S. REITs. We also fnd no support for the weak sgnfcance of the yeld curve factors n Japan. Prevously they have been reported to have an unexpected postve sgn. The lack of statstcal sgnfcance n Japan s, however, consstent wth Yourougou (1990) who reports no sgnfcant senstvty durng perods of low nterest-rates. INSERT TABLE 7 The second robustness test consders and ncorporates nto the analyss the mpact of two major events durng the sample perod; namely the Asan Fnancal Crss (A) and the Global Fnancal Crss (G) a decade later. We restrct the A perod from July 1997 untl the end of June 1998 n lne wth Gerlach et al. (006) and Kho and Stulz (000). We follow the Global Fnancal Crss (G) tmelne from Longstaff (010), who defnes the perod as beng from August 007 through to December 008 when the U.S. government started authorsng lendng through the Troubled Asset Relef Program (TARP). In order to explctly consder whether ether of these events sgnfcantly alters our fndngs we re-specfy the GARCH models to ncorporate a dummy varable relatng to the two crses. Ths adapted specfcaton can be dsplayed as follows: 0

22 r 0, 1, L 1, L, S, S (17) 3, C 3, C ψ X where the error term usng the nformaton set Ω for country avalable at tme t 1 s condtonally normal and heteroscedastc and expressed as z,t and,t 1 ~ N0,, t Ω. dentcally dstrbuted random varables wth zero mean and unt varance. z s a sequence of ndependent, X s the vector of the control varables that nclude market ndex returns, world ndex returns and de-trended tradng volume. Varables L, S, and C are the unexpected changes n the yeld curve level, slope and curvature, all calculated at tme t as the dfference between the actual changes n the yeld curve level/slope/curvature and the ones forecasted va the approprate specfcaton of the ARMA model. The Fnancal Crss dummy, D, takes the value of unty durng the Asan Fnancal Crss of and the Global Fnancal Crss of The volatlty equaton for publc real estate s nterchangeably augmented n two alternatve ways: a 0, a 1, 1 g b, 1, S 1 1 g g, 1, S L 1 1 g 1, 1 L g 1 3, C 1 1 g 3, C 1 1 (18a) a 0, a 1, 1 g b, 1, h S 1 1 g g, 1, h h S 1 L 1 g 1, 1 h g L 1 3, h C 1 1 g 3, h C 1 1 (18b) The equalty of the nterest rate coeffcents durng the crss and non-crss perods are tested usng Wald tests. The estmaton results are reported n Table 8. The specfcatons (17)-(18a,b) serve as a further check to the ntal senstvtes reported n Tables 3 and 4. The yeld curve factors enter the equatons alongsde nteracton dummes. The estmates of the factors for the non-crss perods n Table 8 are consstent wth what we report for the full sample n Table 3. Interacton crss varables add lttle to the results from the prevous sectons as they appear to be largely nsgnfcant n explanng publc real estate return behavour. The lack of sgnfcance of the nterest rate factors durng the crss perods s consstent wth Lzer and Satchell (1997) and Lzer et al. (1998) who reported no senstvty n ether the U.S. or U.K. to nterest rate changes durng the perods of hgh volatlty. 1

23 Wth respect to volatlty, the results are largely consstent and mostly support the fndngs prevously reported and dscussed. In addton, we fnd that durng the two crses, volatlty n Japan and the U.S. are affected by changes n the long-term nterest rate factor, whlst n Australa volatlty s lnked to nterest rate volatlty. Interestngly, changes n the yeld curve level factor have a calmng effect on the total rsk of publc real estate n these three countres. The fndngs on volatlty are supported by the Wald test for asymmetry between crss and non-crss perods. INSERT TABLE 8 Concludng Comments Despte the evdent channels through whch nterest rates can potentally mpact upon publc real estate, the majorty of the exstng lterature has focused predomnantly upon the U.S. REIT sector. Ths work extends that lterature by consderng sx of the largest global markets. The analyss also extends the lterature by consderng a range of nterest rate factors that more completely capture the dynamcs of the yeld curve and by focusng on unantcpated nterest rate senstvty. Overall, the results reveal the sgnfcance of both short and long-term nterest rates factors n explanng the returns of publc real estate wth the results dsplayng a hgh degree of robustness. In lne wth theoretcal expectatons, all sx markets exhbt sgnfcant senstvty at one tme or another. However, we do fnd that senstvty does vary across markets and over tme. It s however, hard to specfy whether any observed varaton n the results can be attrbuted to the legal structure n place. Specfcally, the results do not clearly ndcate that markets wth a REIT regme n place throughout the sample perod dsplay heghtened senstvty. The results do provde mportant mplcatons for nvestors' assessment of nterest rate rsk and senstvty n publc real estate markets, hghlghtng the mportance of nterest rates n such markets. Furthermore, n contrast to prevous studes whch have concentrated on a sngle nterest rate factor, the emprcal results also llustrate the mportance of consderng movements across the entre term structure of nterest rates. One area that the paper does not consder s contagon effects n the yeld-curve factors n the dfferent markets. One recent paper has consdered ths ssue n the context of the U.S. and Canada (Wong et al. 011). Ths s an aspect that does warrant attenton n