The Determinants of International Portfolio Holdings and Home Bias

Transcription

1 WP/04/34 The Determnants of Internatonal Portfolo Holdngs and Home Bas Hamd Faruqee, Shung L, and Isabel K. Yan

2 2004 Internatonal Monetary Fund WP/04/34 IMF Workng Paper Research Department The Determnants of Internatonal Portfolo Holdngs and Home Bas Prepared by Hamd Faruqee, Shung L, and Isabel K. Yan 1 Authorzed for dstrbuton by Tamm Bayoum February 2004 Abstract Ths Workng Paper should not be reported as representng the vews of the IMF. The vews expressed n ths Workng Paper are those of the author(s) and do not necessarly represent those of the IMF or IMF polcy. Workng Papers descrbe research n progress by the author(s) and are publshed to elct comments and to further debate. Despte the lberalzaton of foregn portfolo nvestment around the globe snce the early 1980s, the home-bas phenomenon s stll found to exst. Usng a relatvely new IMF survey dataset of cross-border equty holdngs, ths paper tests new structural equatons from a consumpton-based asset-prcng model on nternatonal portfolo holdngs. Usng of stock data allows us to provde new and clear-cut evdence on the determnants of nternatonal portfolo holdngs. The emprcal results show that an augmented gravty model performs remarkably well. The results ndcate that market sze, transacton cost, and nformaton asymmetry are maor determnants of cross-border portfolo choce. These fndngs shed lght on alternatve theores of nternatonal portfolo holdngs, especally on the transacton and nformaton cost-based explanatons of home bas. JEL Classfcaton Numbers: G11; G15 Keywords: Internatonal portfolo choce; Home bas; Transacton cost; Informaton cost Authors E-Mal Addresses: ecl@stanford.edu; efyan@ctyu.edu.hk; hfaruqee@mf.org 1 The authors are thankful to Assaf Razn, Ronald McKnnon, Lawrence Lau, Mchael Kumhof and the Internatonal Trade Semnar partcpants at Stanford Unversty for helpful comments. All remanng errors are our own.

3 - 2 - Contents Page I. Introducton...3 II. Lterature Revew...4 III. The Gravty Model of Internatonal Portfolo Holdng...5 A. The Theoretcal Model...5 B. The Econometrc Model of Internatonal Portfolo Holdngs...8 C. An Extenson...9 IV. Data Descrpton...10 A. Dependent Varables...10 B. Independent Varables...10 V. Emprcal Results...11 A. The Basc Gravty Model...11 B. Informaton Cost Explanaton and Return Chasng...12 C. Equty Prce Rato...13 D. Dvdend Yeld...13 E. Portfolo Dversfcaton of Internatonal CAPM...14 VI. Conclusons...14 References...24 Tables 1. Percentage of Domestc Equty Holdngs n Total Equty Holdngs at the End of Year 1997 and the Optmal Percentage Descrpton Statstcs Correlaton of Real Equty Return n Estmaton Results for the Equty-Holdng Equatons Estmaton Results for the Share-Rato Equatons...19 Fgures 1. Dstrbuton of the Internatonal Equty Holdngs of Australa, Austra, Belgum,,, and Fnland n Dstrbuton of the Internatonal Equty Holdngs of,,,, Malaysa, and the Netherlands n Dstrbuton of the Internatonal Equty Holdngs of New Zealand,,, Sngapore, Span, and n Dstrbuton of the Internatonal Equty Holdngs of the Unted Kngdom and the Unted States n

4 - 3 - I. INTRODUCTION Theores of cross-border portfolo holdngs are challenged by the "home-bas" puzzle, 2 whch refers to the fact that domestc nvestors hold too lttle of ther wealth n foregn assets as compared wth the predctons of standard portfolo theory. French and Poterba (1991) study nternatonal equty holdngs and fnd that there s too lttle cross-border dversfcaton gven the correlaton structure of the nternatonal equty markets, whch provdes great advantage for portfolo dversfcaton. In addton, Tesar and Werner (1995) study the excess return on a portfolo of foregn securtes compared wth a portfolo ncludng prmarly domestc securtes and fnd that there are sgnfcant gans to be made from nternatonal dversfcaton n all countres of ther sample (,, the Unted Kngdom and the Unted States) except. In Table 1, we compare the observed domestc portfolo share wth the predcton from the captal asset prcng model (CAPM). The classcal CAPM assumes symmetry across agents and countres and predcts that agents worldwde hold the same market portfolo n whch the fracton of nvestment at home s equal to the value of the domestc stock market relatve to the value of the world stock market. Hence, the benchmark world market portfolo weght w equals * mcp / mcp, where mcp s the market captalzaton of country. Column (1) n Table 1 reports the observed percentage of domestc equty holdngs n the total equty holdngs of each country at the end of year Column (2) n Table 1 reports the benchmark predcton from CAPM. Column (3) shows the dfferences. The data ndcate that the actual percentage of domestc asset holdngs s much larger than that of the CAPM benchmark predcton for all countres n the sample. For example, Austra's actual domestc equty holdngs are 300 tmes more than the predcton of CAPM. Ths observaton s puzzlng because t mples that the nvestors forgo potentally large gans from nternatonal dversfcaton. 3 Numerous theoretcal and emprcal studes have attempted to provde explanatons to ths home-bas phenomenon. However, exstng emprcal works have been mpeded by the problem of estmatng cross-border holdngs, whch are stock measures. In addton, most of the emprcal studes are restrcted to the Group of Seven (G-7) countres. Portes and Rey (2000) s one of the exceptons. They study the determnants of crossborder equty flows among 14 countres usng a trade-style gravty equaton. Nevertheless, as ponted out n Warnock (2001), flow data provde an naccurate measure of the cross-border equty holdngs, snce they are confounded by the turnover rate. Consderng the lack of relable measures of the turnover rate, the flow data provde lttle nformaton about the determnants of nternatonal asset holdngs (Lane, 2000; Warnock, 2001). Fortunately, n 1997, the IMF organzed a coordnated survey of cross-border equty holdngs n 29 countres. The results of ths survey provde a relatvely hgh-qualty stock measure of the blateral equty holdngs among these 29 countres at the end of Ths s the frst tme that econometrc crosssectonal analyss has been done on the blateral portfolo holdngs of a large sample of countres. The cross-border equty holdngs of varous countres n 1997 are shown n 2 For an early reference, see Levy and Sarnat (1970). 3 See Lews (1999) for a detaled dscusson of ths home bas puzzle.

5 - 4 - Fgures 1 4. The fgures ndcate that there s a wde dsperson n nternatonal equty holdngs for the countres n the sample. Wthn the vast theoretcal and emprcal lterature, numerous explanatons have been offered for the determnants of nternatonal portfolo choce and home bas. Among others, they nclude nternatonal dversfcaton for hedgng country-specfc rsk, transacton costs n buyng and sellng foregn securtes, and nformaton asymmetres. The purpose of ths paper s not to add new explanatons but to apply the Coordnated Portfolo Investment Survey (CPIS) dataset to test the valdty of the aforementoned popular explanatons n the exstng lterature. We frst employ a hghly stylzed consumpton-based asset-prcng model that accommodates transacton costs. Ths model mples that nvestment n foregn countres can brng dversfcaton benefts but may be dscouraged by the transacton costs. The CPIS dataset allows us to test the relatonshp between blateral equty holdngs and transacton costs. In the emprcal study descrbed n ths paper, our sample ncludes 23 of those countres n the IMF CPIS dataset for whch relable macro data can be obtaned 4. We fnd that our model fts the data well. The market-sze, nformaton-costs and transacton-effcency varables can explan nearly 80 percent of the varaton of the cross-border equty holdngs. The structure of ths paper s as follows. Secton VI. represents a revew of the relevant lterature. Secton III. presents a structural model on the determnants of nternatonal portfolo holdngs and dscusses the mplcatons of the structural model on the emprcal regressons. Secton IV. descrbes the data used n ths study. Secton VI. reports the emprcal results and Secton VI. concludes. II. LITERATURE REVIEW The nternatonal CAPM of Frankel (1982) provdes an utlty-maxmzaton model of nternatonal asset dversfcaton. The nternatonal CAPM mples that the total portfolo rsk can be reduced by holdng foregn assets whose returns are negatvely correlated wth the returns of the home country assets. Ths suggests that the cross-border equty holdngs are () negatvely related to the degree of correlaton between the home and foregn assets () postvely related to the returns of the foregn assets. In ths paper, these mplcatons of the nternatonal CAPM are tested by ncludng the returns of the home and foregn assets as well as ther correlaton n the regresson. Our results support both mplcatons of the nternatonal CAPM. A new school of thought on the home bas puzzle focuses on the nformaton-based explanatons. Usng a smple model of nvestor preference and behavor, French and Poterba (1991) demonstrates that nformaton asymmetry can generate the same observed portfolo patterns as f the nvestors expect the domestc equty returns to be several hundred bass ponts 4 The 23 countres nclude Australa, Austra, Belgum,, SAR,, Fnland,,,,,, Malaysa, the Netherlands, New Zealand,,, Sngapore, Span,, Swtzerland, the Unted Kngdom, and the Unted States.

6 - 5 - hgher than the returns n the foregn markets. Gehrg (1993) models the nformaton asymmetry between domestc and foregn assets by usng a model where nvestors observe nosy sgnals of frms' returns wth dfferent degrees of precson. Domestc bas arses from better nvestor nformaton about domestc stocks and foregn nvestments appear on average more rsky. Hasan and Smaan (2000) derves the premum that an nvestor s wllng to pay to buy the full nformaton of the mean return vector and shows that ratonal nvestors would prefer home country domnstated portfolos over dversfed portfolos f the varablty of estmaton errors far exceeds the varablty of the mean return vector. On the emprcal sde, Frankel and Schmukler (1997)'s fndngs from country mutual fund data support the hypothess of asymmetrc nformaton, accordng to whch the holders of the underlyng assets have more nformaton about local assets than country fund share holders. Smlarly, usng data on foregn stock ownershp n, Kang and Stulz (1995) fnd that foregn nvestors overweght shares of frms whose nformaton are more readly avalable. These frms nclude large frms n the manufacturng ndustres and frms wth good accountng performance etc. They fnd that there s no evdence that foregn ownershp s related to the expected returns. Moreover, Portes and Rey (2000) shows that the gravty model explans the cross-border equty flows remarkably well as dstance serves as a good proxy for nformaton cost. In ths paper, blateral dstance and lngustc lason are used to measure the nformaton asymmetry between domestc and foregn nvestors. Further to the lterature on the nformaton-based explanatons, Black (1974) and Stulz (1981) develop equlbrum models of nternatonal asset prcng that explan home bas by ctng transacton cost frctons to nternatonal captal flows. However, Tesar and Werner (1995) suggest that transacton costs are unlkely to be an explanaton for home bas. The reason s that a hgher transacton cost on foregn nvestment should lead to lower turnover rates on the foregn components of the asset portfolos, but they fnd that the portfolo turnover rates are much hgher for foregn than domestc assets. Ther fndng s nfluental but controversal. Warnock (2001) suggests that ths under-weghted but overtraded puzzle n foregn equtes could be due to the ntrnsc problems n estmatng the cross-border holdngs (a stock measure) based on the captal flow data (a flow measure). Our regresson results show that transacton costs related varables such as blateral dstance, blateral phone costs, and communcaton nfrastructure are sgnfcant factors that affect nternatonal nvestment. III. THE GRAVITY MODEL OF INTERNATIONAL PORTFOLIO HOLDING In ths paper, we frst use a structural model to obtan the basc relatonshps between the nternatonal portfolo holdngs and ther determnants suggested n the lterature. A. The Theoretcal Model We relax the assumptons n Martn and Rey (1999) and extend ther model to N countres. Each country populates wth n rsk-averse mmoble dentcal agents. In the frst perod, each agent h { 1,..., n} n country s endowed wth y unts of a freely traded good (the numerare) and a rsky proect x h. They can ether consume the good or buy shares of proects developed by others. In the second perod, there are S exogenously

7 - 6 - determned and equally lkely states of nature. The rsky proect x h s an Arrow-Debreu securty wth payoff δ d h δ_{} n state, {1,...,S}, where δ = 1 f = and δ = 0 f. Ths assumpton captures the feature that dfferent proects and assets are mperfectly correlated so that assets are mperfect substtutes and dversfcaton mproves safety. For smplcty, we assume the dvdend d of the rsky proects n the same country are the same. All the proects h are traded n the compettve market wth exogenous prces. In the frst perod, the agents rase captal by sellng shares of ther proects and they buy shares of other proects. T s the total number of proects n the world. We assume the number S of the total states of the world s always bgger than T, whch means the market s not complete. Agents cannot elmnate all the rsk by holdng a portfolo of all traded assets. In some states of the world, there wll be no producton. In equlbrum, assume agents wll have no nterest n duplcatng a proect that has already been developed. Thus the total number of proects n the world s T = N n = 1 h = 1 We ntroduce nternatonal transacton costs n both trade and asset markets. In the frst perod, when agents trade assets the buyers of the assets bear the transacton cost. For example, h denote x h (or x h n short) the demand of agent h located n country for an asset developed h by agent h located n country. The amount pad by an agent h to buy xh asset sold on the stock market n country s p h h h (1 + τ ) where p h (or p n short) s the prce of a share of a x proect developed by agent h and τ s the transacton cost n asset markets between country and country. In our paper, we assume the transacton cost between two countres s symmetrc,.e. τ = τ. In the second perod, when the stochastc dvdend s shfted across border, an ceberg cost (transportaton cost) ψ s appled. If an agent n country holds an asset sold n country whch pays a dvdend d n perod 2, the shareholder n country wll receve only ( 1 ψ ) d per share. Consumpton goods and securty n the second perod s only dfferentated by geography transacton cost. By perfect competton and the symmetry n our model, the securtes wthn one country are homogeneous. We can wrte budget constrant for an agent h n country (home country) as, N 1, h + + (1 τ = 1 c ) n p x = y + p whereτ s the nternatonal transacton cost on fnancal market s.t. Each agent h n country (n home country) maxmzes the followng utlty, 1 1/ σ U = c + βe[ c /(1 1/ σ )] Max h1 x,..., x hn h h h 1, h h 2, h h τ >0 f and x h. τ =0 f =. Gven the descrpton of the payoff structure of the dfferent proects, the expected utlty of agent h s, E [ U h ] = c 1, h 1 + β T N = 1 [ n ((1 ψ ) d x h h ) 1 1/ σ 1 /(1 )] σ

8 - 7 - where ψ s the ceberg cost n trade s.t. ψ >0 f and ψ =0 f =. By the frst order condton, σ 1 h β σ d xh = ( ) σ T p h σ 1 β σ d (1 xh = ( ) σ T p (1 + σ 1 ψ ) for h σ τ ) The market clearng condtons are, N = 1 n x h = 1, = 1,... N Then we get the cross-border equty holdngs of from, eq = n n p x σ 1 σ 1 β σ d (1 ψ ) = nn p ( ) σ σ T p (1 + τ ) β σ σ 1 1 = ( ) ( mcp )( mcp )( TC )( R ) T p p (1) σ 1 σ where TC = (1 ψ ) /(1 + τ ) s the nternatonal transacton cost and R = d / p s the rate of return. mcp and mcp are the market captalzaton of country and respectvely. From ths equaton, we obtan the gravty model of the cross-border portfolo holdngs: the crossborder holdngs are postvely correlated wth the market captalzaton of country and ; the cross-border holdngs are negatvely correlated wth the nternatonal transacton cost n trade and fnancal market; there s a return-chasng behavor and p, p measures the prce of country 's asset relatve to country 's asset. By smple manpulaton, we obtan the followng equatons whch are equvalent to equaton 1, β σ σ 1 eq = ( ) nn ( TC )( R ) T (2) s p TC = s p TC (3) s R TC d = s R TC d (4) where s s the share of country s portfolo consstng of country 's equty, whch equal to eq /( mcp ). The nternatonal CAPM wthout transacton costs predcts that s = s = mcp / mcp. However, we observe s / s much smaller than 1. The mean value s 0.02 n our sample. In the emprcal secton of ths paper, we study whether transacton costs and asymmetrc nformaton can explan for the devatons of the actual blateral equty holdngs from the predcton of the nternatonal CAPM.

9 - 8 - log( TC B. The Econometrc Model of Internatonal Portfolo Holdngs Assume the nternatonal transacton cost ) = log( TC ) + log( TC ) Trade fnancal TC s n Cobb-Douglas form. Hence, = βx where X s a K 1 vector whch contans the logarthm of transacton cost varables. β s a K 1 coeffcent vector. The transacton costs varables employed n ths paper nclude the dstance ( dst ), the recprocal of blateral openness ( 1 / bopen ), the number of phonelnes ( phonelne ) and the phone costs ( phone cos t ). Snce the sample countres had elmnated almost all captal controls on portfolo nvestment n the 1990s, no captal control ndex s explctly ncluded n the regresson analyss. Takng logarthm of equaton 1 we derve the econometrc equaton: log( eq ) = constant + β log( mcp ) + β log( mcp ) + β log( dst ) β log(nf ormaton cost var ables ) + β log( transacton cost var ables ) β log( return var ables ) + β log( prce ) + β log( prce ) + ε (5) The "log" operator denotes natural logarthm. The dependent varable eq s the stock of country equty held by resdents of country at the end of It s taken from the coordnated survey of cross-border equty holdngs organzed by the IMF. All equatons nclude a constant term. When analyzng the nternatonal holdng of equtes, we used the begnnng-of-perod market captalzaton mcp and mcp to represent the fnancal sze of country and country respectvely. We use dst to approxmate the tradng cost (ncludng transacton cost and nformaton cost). However, t s dffcult to fnd an accurate measure correspondng to the concept of equlbrum prce p and p ) of the model. Snce p and p are endogenously related to the market captalzaton mcp ( and mcp ( mcp = N = 1 n x h p ), we cannot consstently estmate the coeffcents of mcp and mcp n equaton 5 f we omt the prce varables and put them nto the error term. For ths reason, we use GDP as a measure for the economc agent number n n country and estmate equaton 2 usng the followng econometrc equaton: log( eq ) = cons tan t + β log( GDP ) + β log( GDP ) + β log( dst ) + β log(nf ormaton cost 4 + β log( transacton cos t 5 + β log( return var ables ) + ε var ables ) var ables ) However, t s possble that some countres are fnancally less developed and may not use fnancal market to dversfy ther rsks. For example, the prce-economy rato (mcp/gdp) of was 0.25 n 1997, whch was much smaller than 's 2.7. Hence, for some cases, GDP may not be able to reflect fully the number of the economc agents who partcpate n the 3 (6)

10 - 9 - equty market. For ths reason, we can nstead estmate equaton 4 as follows whch does not nvolve the prce varables: log( s / s ) = cons tan t + β log( dst + β log( transacton cos t + β log( return + β log(nf ormaton cos t / dst var ables varables varables / return ) / nf / transacton cos t varables ormaton cos t ) + ε varables ) var ables ) (7) Snce t s not easy to obtan good measure of dvdend d n country ether, we put t nto the error term n the regresson. However, although d s exogenous, t s correlated wth the return varables through the market clearng condton. Because of ths, we need to use nstrumental varables for the return varables n order to estmate the parameters consstently. Snce the return varables are nonlnear functons of the transacton cost varables (e.g. dst ) whch are not lkely to correlate wth the dvdend varables, we can use the square of the logarthm of those transacton cost varables as the nstrumental varables for return varables and obtan consstent estmates of equaton 4 usng 2SLS. C. An Extenson In the above structure model, we assume that the proects n dfferent countres do not overlap wth each other. Ths assumpton can be easly relaxed. Suppose that dfferent countres have some proects that overlap wth each other 5. In ths case, the measure of economc scale n country should be the proects whch do not overlap wth country 's. We defne GDP + * β = GDP /(1 corr ), where corr s the correlaton of returns between country and for the part of the country 's proects whch do not overlap wth home country 's. Under ths specfcaton, the blateral correlaton of returns s ntroduced nto all of the above regressons. Table 3 shows the matrx of cross country correlatons of real equty return n In the emprcal secton, we wll frstly estmate a "core" equaton whch ncludes the three varables n equaton 2 ( GDP, GDP and dst ). Then we wll explctly add other transacton cost varables to the estmaton equaton. Table 4 reports the results of regressons n whch the dependent varables are log( eq ). All regresson estmates are Whte heteroskedastcty-consstent estmates. 5 It s assumed that the transacton costs of nternatonal nvestments are nonzero so that t s always proftable to hold the home country's proects even though the same proects are σ 1 σ σ 1 σ avalable n other countres. In other words, t s assumed that ( d / p ) TC < ( d / ph ).Ths assumpton s nnocuous because, otherwse, no one wll hold the home country proects at equlbrum.

11 IV. DATA DESCRIPTION The cross-border equty data s from a survey of nternatonal portfolo holdngs coordnated by the IMF. The countres that we have ncluded n our sample are Australa, Austra, Belgum,, (SAR),, Fnland,,,,,, Malaysa, the Netherlands, New Zealand,,, Sngapore, Span,, Swtzerland, the Unted Kngdom, and the Unted States. Table 2 provdes the descrptve statstcs of all the varables. The data sources and defntons are summarzed as follows: A. Dependent Varables eq : equtes of country held by the resdents of country (n mllons of dollars) n Source: IMF, "Results of the 1997 Portfolo Investment Survey". Snce, SAR, and Swtzerland are ncluded n the survey as destnaton countres only but not as nvestng countres, the data set conssts of 20 nvestng countres' holdngs of 23 destnaton countres' equtes at the end of The equty data s thus a 20 by 23 matrx, whch gves 460 observatons n total. s : t equals to eq / mcp where mcp s the total market captalzaton of equty n country n 1997 (n bllons of dollars). Source: MSCI. B. Independent Varables Market Sze: GDP : the Gross Domestc Product at current prce of country (n mllons of dollars). NumFrm : the number of publcly lsted companes n country. Informaton Cost Varables dst : physcal dstance between the captal ctes of country and. When =, we take dst_{}=1km. Source: Ln : a dummy varable whch equals to 1 f the offcal languages are the same n country and, 0 otherwse. Source: phonelne : the number of man phone lnes n use per 1000 nhabtants n country. Source: "The World Compettveness Yearbook", 1997 and 1999 ssues, publshed by IMD. Transacton Cost Varables phone cos t : the per-mnute nternatonal phone costs to call from country to country durng busness hours. Source:

12 Asset Returns and Asset Prces averreturn eq : average real annual return of equty n country (t ncludes the changes n the stock market ndces and the dvdend whch s assumed to be renvested monthly). The real return data s adusted for the nflaton rate and exchange rate movement n country to take nto account the nfluence of exchange rate changes (currency rsk) and nflaton on portfolo return. The th real return s calculated as r = [( 1+ )(1 + e ) /(1 + π )] 1 where s the nomnal return of an asset n ts own currency, e s the rate of apprecaton of the home currency relatve to the currency of the destnaton country and π s the rate of nflaton n the destnaton country. Source: return data s from MSCI monthly (enttled "MSCI" before 1996 and enttled "EAFE And World Prospectve" startng 1996), nflaton rate and exchange rate data are from IFS lne..ae.. and..xzf.. respectvely. PE : The average prce-earnng rato (P/E rato) of equty n the destnaton country dvded by that of the source country. Source: Global Fnancal Data -- MSCI P/E rato. Openness of the Asset Market bopen : (blateral export of goods and servces + blateral mport of goods and servces) /2( n mllons of dollars ) between country and, normalzed by the average GDP of country and country. Source: "Drecton of Trade Statstcs" 1997 ssue, IMF. Portfolo Dversfcaton correq : The correlaton of real equty return n country and between 1987 and Source: same as the varable averreturneq above. V. EMPIRICAL RESULTS A. The Basc Gravty Model Frst we estmate the core equaton whch ncludes the market sze and dstance varables: log( eq ) = cons tan t + β 1 log( GDP ) + β2 log( GDP ) + β3 log( dst ) + ε (8) In ths equaton, the GDP varables capture the market szes of the nvestng and destnaton countres whle the dstance varable provdes a frst-step approxmaton of the nformaton cost. Ths core equaton s a basc form of the gravty model whch s actually a smplfed form of equaton 2 derved from the theoretcal model. The nformaton costs are expected to be postvely correlated wth dstance as longer dstance mples weaker busness and cultural lnks, and the costs of face-to-face talk are hgher. Thus dstance s also a good proxy for the nformaton cost n the fnancal market. The estmaton result s reported n Column (1) of Table 4. Equaton 2 of the theoretcal model predcts that the equty holdng of country n country ( eq ) s drectly and postvely proportonal to the market szes ( n, n ) and negatvely proportonal to the nformaton costs. In partcular, the model predcts that the cross-country equty holdngs are negatvely correlated

13 wth dstance. The regresson results show that all the coeffcents of the varables are hghly sgnfcant wth the expected sgn. Moreover, The estmated coeffcents of the market sze varables are statstcally equal to one, whch ndcates that the cross-border equty holdngs are drectly proportonal to the market sze, as mpled by the theoretcal model. Furthermore, the postve relatonshp between cross-border portfolo holdngs and domestc market sze s consstent wth Lane (2000)'s emprcal fndng that countres wth larger domestc market sze tend to holdng greater quantty of foregn assets 6. These three ndependent varables can explan more than 64% of the varance of the cross-secton equty holdng data. Ths result shows that market szes matter whch s consstent wth the theory of Martn and Ray (1999). Ths fndng also supports the nformaton asymmetry explanaton of the home bas puzzle. In the next step, explct transacton cost and nformaton cost varables are added to the model. B. Informaton Cost Explanaton and Return Chasng We now explctly nclude more nformaton and transacton cost varables ( Ln, phontlne, phonelne, phone cost, and bopen ) and equty return varable of the destnaton country ( realreturn ) nto the regresson to estmate equaton 2 of the theoretcal model 7. The regresson then becomes log( eq ) = cons tan t + β log( GDP ) + β log( GDP ) + β log( dst ) + β Ln 4 + β log( phone cos t 7 + β log( phonelne ) + β log( phonelne ) 5 1 ) + β log( bopen ) + β log( realreturn ) + ε Ln s the dummy varables of lngustc and hstorcal cultural lnks between country and country. The lngustc ndcator measures the current communcaton cost. Ln s also an ndcator for the accumulaton of mutual knowledge between two countres. The estmates n column (2) of Table 4 show that the coeffcent of Ln s sgnfcantly postve as suggested by the theory. As we wll see, Ln s a very robust determnant of equty holdngs. The coeffcent 9 3 (9) 6 Calvo and Mendoza(1999) explan for ths postve relatonshp by usng fxed costs n acqurng nformaton about the nvestment condtons n a gven country 7 A possble alternatve measure of transacton cost s to construct a transacton cost ndex usng the prncpal egenvector dentfed by the prncpal component analyss (PCA). The prncpal egenvector assgns weghts to the sx transacton cost varables (ncludng dstance) so as to come up wth an ndex (a weghted sum) whch can best explan the percentage of varatons of the sx varables. Based on the PCA, 34 percent of the total varaton can be explaned by a transacton cost ndex formed by the frst prncpal component. The varables (n log) ranked accordng to the absolute magntude of ther weghts are: blateral openness (-0.557), phone cost (0.554), dstance (0.519), phone lne of the destnaton country (-0.284), phone lne of the nvestng country (-0.164) and lngustc (-0.059).

14 of phonelne also shows up to be sgnfcantly postve and phone cost sgnfcantly negatve. Ths s exactly as predcted by the model snce more phone lnes and lower phone cost mples lower nformaton and transacton costs between nvestors' home countres and the foregn countres. The estmates are reported n column (2) of Table 4. After addng these varables, the coeffcents of dst and the market sze varables reman hghly sgnfcant. To analyze the return chasng motve mpled by the nternatonal CAPM, we nclude the real return of the destnaton countres ( realreturn ) nto the regresson. Ths varable s only margnally sgnfcant at the 10% sgnfcance level. It provdes an evdence that the returnchasng hypothess of portfolo choce alone s not suffcent to explan for the cross-border equty holdngs and the home bas puzzle. C. Equty Prce Rato The equty prce rato ( p / p ) enters equaton 3 of portfolo holdng and we approxmate t by the relatve P/E rato n the nvestng and the destnaton countres. Our model predcts that a hgh relatve prce n the destnaton country wll result n a hgh equty holdng there. Ths s an evdence of the "herdng" effect that was found n the stock market --- a hgh prce now ncreases people's expectaton about future prce and thus encourages people to nvest more n the stock market. Equaton 3 of the theoretcal model s estmated as follows: log( s / s ) = cons tan t + β log( PE / PE ) + β log( dst / dst ) + β Ln 3 + β log( phone cost 5 + β log( phonelne / phonelne ) 4 1 / phone cos t 2 ) + β log( bopen 6 ) + ε As the equty prce rato may cause endogenety problem, nstrumental varables are used for t n the regresson. Based on the market clearng condton, the equty prce rato s nonlnearly related to the transacton cost varables and market sze varables. Because of ths, varables related to transacton costs and market sze (ncludng 2 log( GDP ), log( GDP ), log( dst ), log( dst ), log( NumFrms ) and log( NumFrms ) are used as nstrumental varables for equty prce rato to consstently estmate the parameters. The estmates are reported n column (1) of Table 5. The coeffcent of the equty prce rato s sgnfcantly postve as predcted by equaton 3 of the theoretcal model. Also, the transacton costs varables (ncludng dst / dst, Ln, phonelne / phonelne and bopen ) are sgnfcant and of the correct sgns. D. Dvdend Yeld The dvdend rato ( d / d ) enters equaton 4 of portfolo holdng n the theoretcal model. Nevertheless, snce t s not easy to obtan a good measure of the country-wde dvdend, we nclude the dvdend n the error term of the regresson. However, although d s exogenous, t s correlated wth the return varables through the market clearng condton. In regard of ths, nstrumental varables are used for the return varable n order to estmate the parameters consstently. As the return varables are nonlnear functon of the transacton cost varables (e.g. dst_{}^{}) whch are not lkely to correlate wth the dvdend varables, we use the square of (10)

15 the logarthm of those transacton cost varables as the nstrumental varables for the return varable and obtan consstent coeffcents for equaton <ref>equaton4</ref>. The estmaton results are reported n column (3) of Table 5. The estmaton results show that transacton cost are negatvely related to the rato of cross-border to domestc equty holdngs, as ndcated by equaton <ref>equaton4</ref> of the theoretcal model. Also, the real return varable shows up to be sgnfcantly postve, whch supports the predcton of equaton 4 of the model. E. Portfolo Dversfcaton of Internatonal CAPM As the nternatonal CAPM also mples a negatve relatonshp between nternatonal asset holdng and the degree of correlaton between the returns of home and foregn assets, we add the correlaton of home and foregn asset returns ( corr ) nto regressons 9, 10 and equaton 4. The results are reported n column (3) of Table 4, column (2) and column (4) of Table 5 respectvely. The correlaton varables are all sgnfcantly negatve at the 1 percent sgnfcance level, whch supports the portfolo dversfcaton mplcaton of the nternatonal CAPM. VI. CONCLIONS We fnd strong evdence of home bas n equtes n 20 countres despte the general relaxaton of controls on foregn portfolo nvestments by developed countres n the early 1980s. Our emprcal fndngs shed lght on the relevance of competng explanatons of nternatonal portfolo choce and the home-bas puzzle. In our fndngs, nternatonal portfolo holdngs are determned by market sze, transacton costs, and nformaton costs. The estmaton results also support explanatory return-chasng behavor and portfolo dversfcaton as mpled by the nternatonal CAPM. These results suggest that nternatonal nvestng behavor s determned by multple factors, whch helps explan why sngle-factor models are nadequate n solvng the home-bas puzzle. Comparng competng explanatons, we fnd that the gravty model performs best, wth most of the explanatory power comng from the fnancal-market-sze and dstance varables. Therefore, we conclude that fnancal market sze and nformaton asymmetry are maor determnants of nternatonal portfolo choce and home bas. Ths emprcal fndng s consstent wth the mplcatons of the theoretcal model.

16 Table 1. Percentage of Domestc Equty Holdngs n Total Equty Holdngs at the End of Year 1997 and the Optmal Percentage Country Actual Percentage of Domestc Equty Held by Domestc Investors (1) Benchmark Percentage (based on CAPM) (2) Home Bas (Actual mnus benchmark) (3) Australa Austra Belgum Fnland Malaysa Netherlands New Zealand Sngapore Span Unted Kngdom Unted States Source(s): CPIS, IMF (1997).

17 Table 2. Descrpton Statstcs Varables Mean Std. Dev. Mn. Max. Equtes Holdngs (n mllon D) GDP of the 20 nvestng countres (n mllon D) GDP of the 23 destnaton countres (n mllon D) Phone cost ( n D) Dstance (n km) Lngustcs ( dummy: 0 or 1) Phone-lnes of the 20 nvestng countres (per 1,000 n nhabtants) Phone-lnes of the 23 destnaton countres (per 1,000 n nhabtants) Blateral openness (rato) Correlaton of equty returns between home and destnaton countres Real return of equty holdngs (percent) P/E rato of equtes Source(s): CPIS, IMF (1997).

18 Table 3. Correlaton of Real Equty Return n AU AT BE CA HK DK FI FR GM IE IT JP AU AT BE CA HK DK FI FR GM IE IT JP 1.00 MY NL NZ NO PT SG SP SE SZ AU AT BE CA HK DK FI FR GM IE IT JP MY NL NZ NO PT SG SP SE SZ Note: AU (Australa), AT (Austra), BE (Belgum), CA (), HK (HongKong SAR), DK (), FI (Fnland), FR (), GM (), IE (), IT (), JP (), MY (Malaysa), NL (Netherlands), NZ (N.Zealand), NO (), PT (), SG (Sngapore), SP (Span), SE (), SZ (Swtzer.), (Unted Kngdom), (Unted States)

19 Table 4. Estmaton Results for the Equty-Holdng Equatons Dependent Varable log( eq (1) (2) (3) ) Constant ( ) ** (-9.090) ** (-8.300) ** log(gdp ) (22.480) ** (19.290) ** (17.950) ** log(gdp ) (17.180) ** (19.300) ** (19.207) ** log(dst ) (-9.102) ** (-7.810) ** (-8.228) ** Ln (5.365) ** (5.805) ** log(phonelne ) (7.344) ** (7.490) ** log(phonelne ) (-0.318) (-0.297) log(phonecost ) (-2.812) ** (-3.110) ** log(bopen ) (7.290) ** (7.392) ** log(realreturn ) (1.620) (1.928) correq (-2.635) ** R F statstcs Notes: 1. The numbers n parentheses are the t-statstcs. Standard errors are heteroskedastcty consstent. 2. "*" means the t-statstc s 5% sgnfcant. 3. "**" means the t-statstc s 1% sgnfcant.

20 Table 5. Estmaton Results for the Share-Rato Equatons Dependent varable log( s / s (1) (2) (3) (4) ) Constant (0.513) (3.015) ** (0.722) (3.305) ** log(pe /PE ) (3.060) ** (3.315) ** log(dst /dst ) (-5.924) ** (-6.941) ** (-6.048) ** (-7.192) ** Ln (2.967) ** (3.350) ** (2.803) ** (3.406) ** log(phonelne / phonelne ) (1.502) (1.492) (0.362) (0.296) log(phonecost / phonecost ) (-3.903) ** (-4.029) ** (-3.372) ** (-3.451) ** log(bopen ) (5.067) ** (5.497) ** (5.151) ** (5.646) ** log(realreturn / realreturn ) (2.733) ** (2.912) ** correq (-3.560) ** (-3.724) ** R F statstcs Note: 1. The numbers n parentheses are the t-statstcs. Standard errors are heteroskedastcty consstent. 2. "*" means the t-statstc s 5% sgnfcant. 3. "**" means the t-statstc s 1% sgnfcant.

21 Fgure 1. Dstrbuton of the Internatonal Equty Holdngs of Australa, Austra, Belgum,,, and Fnland n 1997 Swtzerland Span Sngapore Netherlands Malaysa Fnland Belgum Austra Swtzerland Span Sngapore Netherlands Malaysa Fnland Austra Australa Swtzerland Span Sngapore Netherlands Malaysa Fnland Belgum Austra Australa Australa mllon D Belgum mllon D mllon D Swtzerland Span Sngapore Netherlands Malaysa Fnland Belgum Australa Swtzerland Span Sngapore Netherlands Malaysa Fnland Belgum Austra Australa Swtzerland Span Sngapore Netherlands Malaysa Belgum Austra Australa Austra mllon D mllon D Fnland mllon D Source(s): CPIS, IMF (1997).

22 Fgure 2. Dstrbuton of the Internatonal Equty Holdngs of,,,, Malaysa, and the Netherlands n 1997 Swtzerland Span Sngapore Netherlands Malaysa Fnland Belgum Austra Australa Swtzerland Span Sngapore Netherlands Malaysa Fnland Belgum Austra Australa Swtzerland Span Sngapore Netherlands Fnland Belgum Austra Australa mllon D mllon D Malaysa mllon D 613 Swtzerland Span Sngapore Netherlands Malaysa Fnland Belgum Austra Australa Swtzerland Span Sngapore Netherlands Malaysa Fnland Belgum Austra Australa Swtzerlan Span Sngapore mllon D Malaysa 344 Fnland Belgum Austra Australa mllon D Netherlands mllon D Source(s): CPIS, IMF (1997).

23 Fgure 3. Dstrbuton of the Internatonal Equty Holdngs of New Zealand,,, Sngapore, Span, and n 1997 Swtzerlan Span Sngapore Netherland Malaysa Fnland Belgum Austra Australa Swtzerlan Span Sngapore Netherland Malaysa Fnland Belgum Austra Australa Swtzerlan New Zealand Sngapore Netherland Malaysa Fnland Belgum 113 Austra3 Australa mllon D mllon D Span mllon D Swtzerlan Span Sngapore Netherland Malaysa Fnland Belgum Austra Australa Swtzerlan Span Netherland Malaysa Fnland Belgum Austra Australa Swtzerlan Span Sngapore Netherland Malaysa Fnland Belgum Austra Australa mllon D Sngapore mllon D mllon D Source(s): CPIS, IMF (1997).

24 Fgure 4. Dstrbuton of the Internatonal Equty Holdngs of the Unted Kngdom and the Unted States n 1997 Swtzerland Span Sngapore Netherland Malaysa Fnland Belgum Austra Australa Swtzerland Span Sngapore Netherland Malaysa Fnland Belgum Austra Australa Unted Kngdom mllon D Unted States mllon D Source(s): CPIS, IMF (1997).

25 References Adler, Mchael, and Bernard Dumas, 1983, Internatonal Portfolo Choce and Corporaton Fnance: A Synthess, Vol. XXXVIII, No. 3, pp Anderson, James, 1979, A Theoretcal Foundaton for the Gravty Equaton, Amercan Economc Revew, Vol. 69, No. 1, pp Brennan, Mchael, and H. Cao, 1997, Internatonal Portfolo Investment Flows, Journal of Fnance, Vol. 52:5, pp Calvo, G., Mendoza, E., 1999, Ratonal Contagon and the Globalzaton of Securtes Markets, NBER Workng Paper No (Cambrdge, Massachusetts: Natonal Bureau of Economc Research). Clemens, Salm, 1999, Internatonal Asset Dversfcaton and the Exchange Rate Rsk Premum, Workng Paper. Coakley, Jerry, F. Kulas, and R. Smth, 1998, The Feldsten-Horoka Puzzle and Captal Moblty: A Revew, Internatonal Journal of Fnance and Economcs, Vol. 3, pp Coordnated Portfolo Investment Survey, 1997, IMF (Washngton: Internatonal Monetary Fund). Dumas, Bernard, and R. Uppal, 1998, Global Dversfcaton, Growth and Welfare wth Imperfectly Integrated Markets for Goods, IMF, Workng Paper. Frankel, J., 1982, A Search of the Exchange Rsk Premum: A Sx-Currency Test Assumng Mean-Varance Optmzaton, Journal of Internatonal Money and Fnance, Vol. 1, pp Frankel, Jeffrey and S. Schmukler, 1997, Country Funds and Asymmetrc Informaton, CIDER Workng Paper No. C97-087, pp (Berkeley, CA: Center For Internatonal And Development Economcs Research). French, Kenneth, and J. Poterba, 1991, Investor Dversfcaton and Internatonal Equty Markets, Amercan Economc Revew, Vol. 81, pp Gehrg, Thomas, 1993, An Informaton Based Explanaton of the Domestc Bas n Internatonal Equty Investment, Scand. J. Economcs, Vol. 95:1, pp Gordon, Roger H., and Lans A. Bovernberg, 1996, Why s Captal So Immoble Internatonally? Possble Explanatons and Implcatons for Captal Income Taxaton, Amercan Economc Revew, Vol. 86, pp

26 Grnblatt, Mark, and M. Keloharu, 1999, What Makes Investors Trade?, UCLA, Workng Paper. Hasan, Ifekhar, and SmaanYusf, 2000, A Ratonal Explanaton for Home Country Bas, Journal of Internatonal Money and Fnance, Vol. 19, pp Isard, Peter, 1977, How Far Can We Push The Law of One Prce? Amercan Economc Revew, Vol. 67, Kang, Jun-Koo, and R. Stulz, 1997, Why Is There Home Bas? An Analyss of Foregn Portfolo Equty Ownershp n, Journal Fnancal Economcs, Vol. 46:1, pp Lane, Phlp and T. Dubn, 2000, Internatonal Investment Poston: A Cross- Sectonal Analyss, Journal of Internatonal Money and Fnance, Vol. 19:4, pp Lews, Karen, 1999, Tryng to Explan Home Bas n Equtes and Consumpton, Journal of Economc Lterature, Vol. XXXVII, pp Martn, Phlppe, and H. Rey, 1999, Fnancal Super-Markets: Sze Matters for Asset Trade, No. 2232, Pp (London, : Centre For Economc Polcy Research. Dscusson Paper Seres). Obstfeld, Maurce, and K. Rogoff, 2000, The Sx Maor Puzzles n Internatonal Macroeconomcs: Is There a Common Cause?, NBER Workng Paper No. 7777, pp (Cambrdge, Massachusetts: Natonal Bureau of Economc Research). Pesent, Paolo, and E. Wncoop, 1998, An Internatonal Benchmark Portfolo: The Role of Non-Traded Goods, NBER, Workng Paper. Portes, Rchard, and H. Rey, 2000, The Determnants of Cross-Border Equty Flows: The Geography of Informaton, CEPR. Tesar, Lnda, and I. Werner, 1995, Home Bas and Hgh Turnover, Journal of Internatonal Money and Fnance, Vol., 14(4), pp Tmmermann, Allan, and D. Blake, 2000, Internatonal Investment Performance: Evdence from Insttutonal Investors' Foregn Equty Holdngs, Workng Paper. Warnock, Francs E., 2001, Home Bas and Hgh Turnover Reconsdered, Board of Governors of the Federal Reserve System, Internatonal Fnance Dscusson Papers Number 702.