The Cost of Captal of Cross-sted Frms Kees G. Koedjk, Mathjs A. van Djk ERIM REPORT SERIES RESEARCH IN MANAGEMENT ERIM Report Seres reference number ERS-2002-99-F&A Publcaton status verson October 2002 Number of pages 37 Emal address correspondng author c.koedjk@fbk.eur.nl, madjk@fbk.eur.nl Address Erasmus Research Insttute of Management (ERIM) Rotterdam School of Management Facultet Bedrjfskunde Erasmus Unverstet Rotterdam PoBox 1738 3000 DR Rotterdam, The Netherlands Phone: # 31-(0) 10-408 1182 Fax: # 31-(0) 10-408 9640 Emal: nfo@erm.eur.nl Internet: www.erm.eur.nl Bblographc data and classfcatons of all the ERIM reports are also avalable on the ERIM webste: www.erm.eur.nl
ERASMUS RESEARCH INSTITUTE OF MANAGEMENT REPORT SERIES RESEARCH IN MANAGEMENT BIBIOGRAPHIC DATA AND CASSIFICATIONS Abstract brary of Congress Classfcaton (CC) Ths paper analyzes the cost of captal of frms wth foregn equty lstngs. Our purpose s to shed lght on the queston whether nternatonal and domestc asset prcng models yeld a dfferent estmate of the cost of captal for cross-lsted stocks. We dstngush between () the multfactor ICAPM of Solnk (1983) and Sercu (1980) ncludng both the global market portfolo and exchange rate rsk prema, and () the sngle factor domestc CAPM. We test for the sgnfcance of the cost of captal dfferental n a sample of 336 cross-lsted stocks from nne countres n the perod 1980-1999. Our hypothess s that the cost of captal dfferental s substantal for frms wth nternatonal lstngs, as these are often large multnatonals wth a strong nternatonal orentaton. We fnd that the asset prcng models yeld a sgnfcantly dfferent estmate of the cost of captal for only 12 percent of the cross-lsted companes. The sze of the cost of captal dfferental s around 50 bass ponts for the U.S., 80 bass ponts for the U.K., and 100 bass ponts for France. 5001-6182 Busness 5601-5689 Accountancy, Bookkeepng 4001-4280.7 Fnance Management, Busness Fnance, Corporaton Fnance Journal of Economc terature (JE) HG 4028.C4 M M 41 G 3 G 15 G 31 F 31 Captal nvestments (Fnancal management) Busness Admnstraton and Busness Economcs Accountng Corporate Fnance and Governance Internatonal Fnancal Markets Captal Budgetng; Investment Polcy Foregn Exchange European Busness Schools brary Group (EBSG) 85 A Busness General 225 A Accountng General 220 A Fnancal Management 220 F Fnancal markets Gemeenschappeljke Onderwerpsontslutng (GOO) Classfcaton GOO 85.00 Bedrjfskunde, Organsatekunde: algemeen 85.25 85.30 Accountng Fnanceel management, fnancerng Keywords GOO Bedrjfskunde Bedrjfseconome Accountancy, fnanceel management, bedrjfsfnancerng, beslskunde Free keywords Kaptaalmarkt, Wsselkoersen, portfolo-analyse Cross-lstngs, cost of equty captal, foregn exchange exposure
The Cost of Captal of Cross-sted Frms Kees G. Koedjk Erasmus Unversty Rotterdam and CEPR Mathjs A. van Djk Erasmus Unversty Rotterdam Ths verson: October 2002 Correspondence Kees G. Koedjk Department of Fnancal Management (Room F4-21) Erasmus Unversty Rotterdam PO Box 1738 3000 DR Rotterdam THE NETHERANDS Phone: +31 10 408 2748 Fax : +31 10 408 9017 E-mal: c.koedjk@fbk.eur.nl We thank Peter Roosenboom and semnar partcpants at Erasmus Unversty Rotterdam for helpful comments. We are grateful to Rchard Speetjens for valuable research assstance. Any remanng errors are our own.
1 The Cost of Captal of Cross-sted Frms Abstract Ths paper analyzes the cost of captal of frms wth foregn equty lstngs. Our purpose s to shed lght on the queston whether nternatonal and domestc asset prcng models yeld a dfferent estmate of the cost of captal for cross-lsted stocks. We dstngush between () the multfactor ICAPM of Solnk (1983) and Sercu (1980) ncludng both the global market portfolo and exchange rate rsk prema, and () the sngle factor domestc CAPM. We test for the sgnfcance of the cost of captal dfferental n a sample of 336 cross-lsted stocks from nne countres n the perod 1980-1999. Our hypothess s that the cost of captal dfferental s substantal for frms wth nternatonal lstngs, as these are often large multnatonals wth a strong nternatonal orentaton. We fnd that the asset prcng models yeld a sgnfcantly dfferent estmate of the cost of captal for only 12 percent of the cross-lsted companes. The sze of the cost of captal dfferental s around 50 bass ponts for the U.S., 80 bass ponts for the U.K., and 100 bass ponts for France. Keywords Cross-lstngs, cost of equty captal, foregn exchange exposure JE subject codes G15, G31, F31
2 1. Introducton As many companes have become consderably more nternatonally orented over the past decades, foregn equty lstngs have ganed mportance as a strategc management tool. The number of nternatonal cross-lstngs n the U.S. has ncreased n recent years. Snce 1993, the total number of non-u.s. lsted companes at the NYSE has more than quadrupled to 471 as of October 14, 2002. The number of nternatonal stocks at Nasdaq has ncreased from 261 at the end of 1992 to more than 385 at October 14, 2002 (wth a peak of over 450 n August 2001). Snce 1996, the number of cross-lsted frms at the AMEX has about doubled to 52 as of October 14, 2002. The lterature on nternatonal cross-lstngs focuses on three man ssues. Frst, many studes have examned the effects of a cross-border lstng of a stock n terms of excess returns, lqudty, and rsk. Foerster and Karoly (1993) nvestgate Canadan stocks that lst n the U.S. and fnd a postve pre-lstng abnormal return, whle the 100-day post-lstng abnormal return s negatve. The lqudty of the stocks ncreases and the betas decrease on average. Werner and Kledon (1996) also fnd that lqudty ncreases for a sample of U.K stocks that have a cross-lstng at the NYSE. The authors fnd no effect for the rsk of the stocks. Joron and Schwartz (1986) compare the cost of captal and the rsk of 94 Canadan stocks that are dually lsted n the U.S. wth a benchmark sample of 655 Canadan stocks not lsted at an exchange n the U.S. They fnd that the cross-lsted companes have a lower cost of captal, but a hgher senstvty to U.S. market rsk than the benchmark frms. More recently, Doukas and Swtzer (2000) fnd a sgnfcantly postve stock market reacton to the announcement of a lstng n the U.S. by 79 Canadan frms n the perod 1977-1997. Ths s consstent wth the hypothess that nternatonal lstngs lead to a decrease n the rsk premum of frms operatng n mldly segmented markets. In an extensve survey of studes on crosslstngs, Karoly (1998) concludes that the evdence ndcates a favorable short-term stock prce reacton to the lstng, an mprovement n lqudty, and a consderably lower cost of equty captal. The evdence on longer term post-lstng stock prce performance s mxed. Second, the characterstcs of companes that lst ther shares abroad have been studed extensvely, as well as the motvatons for cross-lstng ther stock at a foregn exchange. Saudagaran (1988) examnes a sample of 223 companes that obtan a dual lstng n Canada, Europe, Japan, or the U.S. and fnds that large frms wth a hgh percentage of sales abroad are relatvely lkely to lst abroad. Smlarly, Pagano, Röell, and Zechner (2002) fnd that companes that lst abroad are relatvely large and have a hgh level of foregn sales and R&D
3 spendng. Bddle and Saudagaran (1989) conclude that frms are relatvely unlkely to lst at overseas exchanges wth strcter dsclosure regulatons than the home market. Karoly (1998) revews the recent evdence and concludes that strngent dsclosure requrements are the man obstacle to overseas lstngs. Fuerst (1998), on the other hand, argues that companes could use a cross-lstng at an exchange wth strct regulatons for sgnalng qualty. Thrd, a number of recent studes employ hgh-frequency data of cross-lsted securtes on dfferent exchanges to analyze prce dscovery of nternatonally-traded frms. Grammg, Melvn, and Schlag (2000) examne ntra-day quote data of three large German frms at the Frankfurt Stock Exchange and the NYSE. Ther results ndcate that prce dscovery manly occurs n the home market. Adjustment to exchange rate shocks (e.g. for mantanng the law of one prce) predomnantly takes place on the NYSE, however. Eun and Sabherwal (2002) study prce dscovery for a sample 62 Canadan frms lsted on the Toronto Stock Exchange and ether the NYSE, the Nasdaq or the AMEX. They fnd that whle prce dscovery prmarly occurs on the Toronto Stock Exchange for most frms, the U.S. market s contrbuton to prce dscovery s domnant for several stocks. The contrbuton of the U.S. exchange s postvely related to the U.S. share of tradng and negatvely related to the relatve spread sze n the U.S. We take another angle and focus on the cost of captal of nterlsted stocks. The purpose of ths paper s to shed lght on the queston whether nternatonal and domestc asset prcng models lead to a dfferent estmate of the cost of captal for a frm wth at least one lstng abroad. In a recent study, Stulz (1995) derves an expresson for the dfference n the estmaton of a frm s beta when computed wth the domestc CAPM as compared to the sngle factor ICAPM of Grauer, tzenberger, and Stehle (1976). The estmated cost of captal dfferental s an affne functon of ths so-called prcng error. Stulz uses data on the Swss multnatonal Nestlé and fnds a substantal prcng error. He concludes that the domestc CAPM may well provde an ncorrect estmate of the cost of captal for frms n small economes n general. Koedjk, Kool, Schotman, and van Djk (2001) derve statstcal tests for the prcng error between the domestc CAPM and the multfactor ICAPM of Solnk (1983) and Sercu (1980) ncludng both the global market portfolo and exchange rate rsk prema.
4 The ssue examned n ths paper s llustrated n fgure 1. The multfactor ICAPM s the mantaned hypothess. 1 A prcng error arses for an ndvdual frm f the drect approach of computng the cost of equty captal through the multfactor ICAPM leads to a dfferent result than the ndrect approach of usng the domestc CAPM. Our hypothess s that frms wth at least one nternatonal lstng exhbt a large prcng error. As mentoned above, several studes have shown that companes wth overseas lstngs have a large market captalzaton and a hgh percentage of sales abroad. These frms show a clear nternatonal orentaton and are therefore be expected to exhbt substantal exposure to the global rsk factors (ncludng exchange rates). Ths exposure cannot n general be captured n the nternatonal prcng of the local stock market ndex. Consequently, the drect estmate of the cost of captal of cross-lsted companes may well substantally devate from the ndrect estmate. We analyze a sample of 336 nterlsted stocks from nne dfferent countres over the sample perod 1980:02-1999:06. We fnd a sgnfcant prcng error between the domestc CAPM and the multfactor ICAPM for only 12 percent of the frms n our sample. The absolute dfference n the cost of captal for cross-lsted companes amounts to about 50 bass ponts for the U.S., 55 bass ponts for Germany, 90 bass ponts for Japan and 80 bass ponts for the U.K. Hence, we fnd lmted evdence supportng our hypothess that the prcng error s sgnfcant for frms wth nternatonal cross-lstngs. We show that these results are lkely to be due to strong country factors n the data, consstent wth the evdence of Heston and Rouwenhorst (1994) and Grffn and Karoly (1998). A potental explanaton for ths fndng s a lack of real captal market ntegraton (as opposed to fnancal captal market ntegraton) caused by cyclcal, structural, and nsttutonal country-specfc factors. De Ménl (1999) presents evdence that these country-specfc factors play a sgnfcant role n explanng corporate returns n Europe. Our evdence suggests that nvestors could explot the observed dfferences between countres for the purpose of portfolo dversfcaton. We compare our results for companes wth foregn lstngs to a benchmark sample of 2,957 companes that do not have nternatonal lstngs. Around 4 percent of these domestc frms show a sgnfcant prcng error. The estmated cost of captal dfferental amounts to 80 bass ponts on average for domestc stocks. 1 A number of recent papers do not reject the jont hypothess of the multfactor ICAPM ncludng currency rsk prema and captal market ntegraton for a varety of ndustralzed countres. We refer to secton 2 for a dscusson of the lterature on ths ssue.
5 The paper s organzed as follows. In secton 2 we brefly revew the CAPM, the ICAPM, and the prcng error testng methodology. Secton 3 provdes a descrpton of the data. We dscuss our emprcal results for nterlsted stocks and our benchmark sample of domestc stocks n secton 4. Secton 5 summarzes and concludes. 2. Methodology In ths secton we dscuss tests to evaluate whether the domestc CAPM yelds a sgnfcantly dfferent cost of captal than the multfactor ICAPM. The basc methodology s taken from Koedjk, Kool, Schotman, and van Djk (2001). In the Solnk-Sercu verson of the multfactor ICAPM, the systematc rsk factors are the global market portfolo and exchange rate factors. Assume a world wth N + 1 countres (currences). The ICAPM has N+1 systematc rsk factors: the global market portfolo and N exchange rates. The model can be expressed as E [ R ] r0 + E[ RG r0 ] d 1 + E[ S + r ι r0 ] d 2 =, (1) where R and R G are the return of asset and the global market, respectvely, expressed n the numerare currency. As the numerare currency we choose the home currency 0 of asset. S represents the vector of nomnal exchange rate returns of the other l = 1,..., N countres aganst currency 0. The vector r denotes the nomnal returns on the rsk-free asset n country l (l = 1,..., N). r 0 s the rsk-free rate n the numerare (home) country, and ι s a vector of ones. For a dervaton of equaton (1) we refer to Sercu and Uppal (1995). The global market beta and the exchange rate betas are defned as the regresson coeffcents d 1 and d 2 n R α + u, (2) = 1 + Z d + u = α1 + RG d 1 + S d 2 where Z = [R G, S ] and α 1 = r 0 (1 - d 1 ) + (r - ιr 0 ) d 2 s assumed to be constant. The specfc rsk u s orthogonal to Z. Ths verson of the ICAPM s the mantaned hypothess for the rest of ths paper. In order to estmate d we assume that the regresson parameters are constant wthn a partcular sample perod. The rsk prema on the global market and the currency factors may be tme varyng though. 2 Our emprcal tests wll be formulated n terms of hypotheses on the factor loadngs d for ndvdual stocks relatve to the global factors. We follow Stulz (1995) and consder the domestc CAPM as an alternatve model E = ] b, (3) [ R ] r0 + E[ R r0 2 See for example Dumas and Solnk (1995).
6 where R s the return of the local market ndex expressed n the numerare currency 0. The beta of the CAPM can be estmated n the regresson R α + R b + e, (4) = 2 The domestc CAPM posts a dfferent decomposton nto systematc and specfc rsk than the ICAPM. In order to compare the two models, we need to relate R to the global factors Z. Snce equaton (2) apples to every ndvdual stock, t also apples to the local market portfolo of every country. Applyng (2) to R we get R = α + Z d + u, (5) where u s orthogonal to Z. Substtutng equaton (5) nto (4) yelds 3 R = α + Z d b + u b + e, (6) where α 3 = α 2 + b α. Equatons (2) and (6) lead to the same decomposton of systematc and specfc rsk f the local specfc rsk e n equaton (4) s orthogonal to Z. In that case, the composte specfc rsk term u b + e s orthogonal to Z and equatons (2) and (6) are dentcal. But then the parameters n equatons (2) and (6) must be the same too, mplyng d = d b. (7) If the restrctons n equaton (7) hold, no prcng error results from usng the domestc CAPM nstead of the ICAPM. 3 We call a test for ths null-hypothess a prcng error test. It tests the orthogonalty between the global factors and the resduals from the domestc CAPM regresson (4). A smple way to mplement the test s to add the global nstruments Z to the domestc CAPM regresson, R = α 4 + R β + Z δ + ζ. (8) Under H 0 : δ = 0, we can see that α 4 = α 2, β = b, and ζ = e. The test for the null-hypothess δ = 0 s called the Prcng Error test. It tests the orthogonalty between the global factors and the resduals from the domestc CAPM regresson (4). If the restrcton holds, rsk that s specfc accordng to the domestc CAPM does not contan addtonal systematc rsk related to the global factors. Consequently, the domestc market portfolo contans all the nformaton that s relevant to prce assets. On the other hand, f rsk that s dversfable domestcally contans rsk that s systematc n the world market, the domestc CAPM ncorrectly gnores such rsk. The ICAPM wll requre a rsk premum, however. In that case, the domestc CAPM leads to a dfferent cost of captal than the ICAPM. 3 We assume that the parameter restrcton α 1 = α 2 + b α holds.
7 Rejecton of (7) can be due to ether the condton on the beta of the global market portfolo (d 1 = d 1 b ), the exchange rate betas (d 2 = d 2 b ), or both. If rejecton occurs because of volaton of the exchange rate restrctons d 2 = d 2 b, the mpact on the estmated cost of captal mght nevertheless be zero f requred foregn exchange rsk prema E[S + r - ι r 0 ] are zero. Therefore, whether only the frst restrcton n equaton (7) s rejected wthn the framework of the multfactor ICAPM s of nterest under the assumpton that exchange rate rsk prema are zero. In appendx A we show that the prcng error vector p = d b - d can be wrtten as a lnear combnaton of the parameter δ n equaton (8) p d d = Ι + σ 2 Ω 1 δ = Λδ, where Ω s the covarance matrx of Z and σ 2 s the varance of resduals u n equaton (5). We test the null-hypothess that the frst element of p s equal to zero. We call ths the Global Beta test. If the null-hypothess s rejected, the drect ICAPM beta d 1 wll dffer sgnfcantly from the ndrect beta d 1 b. An mportant assumpton n our analyss s that the multfactor ICAPM holds for every ndvdual stock and thus for the domestc market portfolo of every country. Hence, our tests can be nterpreted as an examnaton of the ssue whether the domestc CAPM wll produce an adequate estmate of a frm s cost of captal when the multfactor ICAPM s the correct model. The ssue of captal market ntegraton has receved a lot of attenton n the recent fnance lterature. Joron and Schwartz (1986) fnd that the uncondtonal sngle factor ICAPM does not accurately descrbe fluctuatons n Canadan stock returns for the perod from 1968 through 1982. They use a North Amercan market ndex as the only prced rsk factor. Ths can be nterpreted as evdence aganst ntegraton of the Canadan and U.S. equty markets. Harvey (1991) tests whether the condtonal sngle factor ICAPM s consstent wth the behavor of stock returns n 17 countres over the perod 1969-1989. Harvey concludes that the hypothess of condtonal mean varance effcency cannot be rejected for most countres. The model s restrctons are rejected for Japan at the 5% level and for the U.S. at the 10% level, however. As s noted by e.g. Bekaert and Harvey (1995), t s dffcult to nterpret the jont hypotheses tested n these studes. Bekaert and Harvey (1995, p. 404) formulate the ntrcacy of nterpretng Harvey s (1991) results as follows: Is the rejecton n Japan a result of usng a one factor model, a functon of Japanese stock prces devatng from ther fundamental values (neffcency), or an mplcaton of mposng the null hypothess of complete market (9)
8 ntegraton?. More recently, De Sants and Gérard (1997) present evdence that global market rsk s equally prced across countres n a condtonal framework. The paper analyzes the world s eght largest equty markets over the perod 1970-1994. The hypothess that the prce of country-specfc rsk s not dfferent from zero s not rejected. Ths s consstent wth the sngle factor ICAPM and wth nternatonal captal market ntegraton. Several studes examne market ntegraton n the context of a multfactor ICAPM, n whch the assumpton of purchasng power party s relaxed. Dumas and Solnk (1995) reject the hypothess that currency rsk s not prced for Germany, Japan, the Unted Kngdom, and the Unted States n the perod January 1970 to December 1991. They argue that the condtonal multfactor ICAPM domnates the sngle factor ICAPM. De Sants and Gérard (1998) drectly test the restrctons mposed by the condtonal multfactor ICAPM usng stock market ndces of Germany, Japan, the Unted Kngdom, and the Unted States n the perod 1973-1994. Ther specfcaton of the nternatonal asset prcng models ncludes three currency rsk factors related to the Deutsche mark, the Japanese yen, and the Brtsh pound. The analyss provdes strong evdence for a model that ncludes both global market rsk and currency rsk factors. Country-specfc rsk s not prced, whch suggests markets are ntegrated. Vassalou (2000) fnds that foregn exchange rate rsk s prced n the returns of ndvdual securtes from 10 countres n the perod 1973-1990. The ssue whether captal markets n Japan and the U.S. can be consdered ntegrated has been extensvely studed n the lterature. Usng an uncondtonal multfactor ICAPM (wthout currency rsk factors), Gultekn, Gultekn, and Penat (1989) do not fnd evdence of segmentaton between the Japanese and the U.S. markets n the four years after the major lberalzaton n the Japanese captal market n December 1980. Campbell and Hamao (1992) fnd some evdence for common movements n Japanese and U.S. stock returns, whch suggests at least partal ntegraton. However, stock returns are not well explaned by a constant-beta sngle factor ICAPM. Evaluatng tests for captal market ntegraton s dffcult. Rejectons of the ntegraton hypothess for Japan and the U.S. n early studes may reflect the fact that these studes employ sngle factor versons of the ICAPM and consequently gnore devatons from PPP. ater studes that relax the assumpton of absolute PPP, e.g. De Sants and Gérard (1998) and Vassalou (2000), do fnd evdence n favor of the jont hypothess of the multfactor ICAPM and market ntegraton for a varety of countres over a recent sample perod.
9 3. Data We use monthly data for nne ndustralzed countres: Australa, Canada, France, Germany, Japan, the Netherlands, Swtzerland, Unted Kngdom, and the Unted States. Nomnal exchange rates for all countres are taken from the nternatonal Fnancal Statstcs (IFS) tape (lne ae). We analyze the perod 1980:02-1999:06. The market weghted local equty ndces and the market weghted global equty ndex are from Morgan Stanley Captal Internatonal (MSCI). Data on ndvdual stocks n ths study s obtaned from Datastream. We have downloaded stock prces, dvdend yelds, and dvdends of frms that are ncluded n the Datastream equty lsts. If dvdends are unavalable, the dvdend yeld s used. If nether dvdend data nor dvdend yelds are avalable, the stock s excluded from the sample. We also exclude stocks that have not been contnuously lsted over the whole perod. Furthermore, the data s fltered for data errors; stocks wth outler observatons are excluded from the sample. 4 Table 1 presents summary statstcs for local and global stock market (MSCI) returns, and exchange rate returns. Returns are measured as logarthmc dfferences and gven n percentages per month. The average domestc market return n local currency ranges from 0.63 for Japan to 1.51 for the Netherlands. Correspondng standard devatons vary between 6.20 for Australa and 4.32 for the U.S.. Columns seven and eght contan summary statstcs of the MSCI world market portfolo expressed n local currency. Agan, Japan s an outler wth an exceptonally low average return related to a substantal apprecaton of the Yen. Correlatons between local and global stock market returns n U.S. dollars are provded n panel A of table 2. Domestc stock markets generally move together, though far from perfectly. Correlatons range from 0.31 (Australa and Canada versus Japan) to 0.73 (Canada versus the U.S). The Japanese stock market appears to have relatvely low correlatons wth the rest of the world. Panel B of table 2 shows correlatons between U.S. dollar exchange rate changes. They range from 0.10 for the Canadan versus the Japanese exchange rate to 0.99 for the blateral rates for Germany and the Netherlands. The European currences appear to move roughly up and down together. Japan, Australa, and Canada have more dosyncratc dollar exchange rate movements, wth low correlaton both among 4 These are stocks wth average annual returns larger than 200%, stocks wth a local beta smaller than 0.1, and nfrequently traded stocks whch have a zero return for more than twenty percent of the observatons.
10 themselves and relatve to the European countres. In panel C of table 2, correlaton coeffcents between local and global stock market returns expressed n U.S. dollars versus blateral nomnal exchange rate changes aganst the U.S. dollar are reported for each par of countres. Correlatons between stock returns and exchange rate changes are generally relatvely low, wth the excepton of the correlatons between the domestc stock market return of a country and the return of ts currency aganst the U.S. dollar. One could argue that the prcng error between the domestc CAPM and the sngle factor ICAPM (wthout currency rsk factors) wll tend to be small, as the domestc market portfolos are relatvely hghly correlated wth the global market portfolo. 5 When currency rsk factors are omtted from the analyss, the dfference between the drect and the ndrect approach of computng the cost of captal may be small for companes from countres whch local stock market s hghly correlated wth the global market. In our analyss, however, we explctly ncorporate exchange rate rsk factors nto the ICAPM. As s mentoned n secton 2, several recent studes, e.g. Dumas and Solnk (1995) and De Sants and Gérard (1998), present evdence that currency rsk s prced for frms from a varety of countres. We argue that n the presence of multple rsk factors exposure to local market rsk cannot generally be expected to capture the (multdmensonal) exposure to the global factors. Therefore, we expect to fnd a substantal prcng error for the cross-lsted frms n our sample, as these are probably hghly exposed to nternatonal rsk factors. The low correlatons between the local market portfolos and the eght blateral exchange rates reported n table 2 corroborate ths argument. Table 3 reports the number of stocks ncluded for each country after the selecton procedures. The total sample conssts of 3,293 stocks wth a complete seres of returns for the perod 1980:02-1999:06. The frst and second columns of table 3 show the number of crosslsted companes, respectvely the number of purely domestc stocks for each country. Our sample conssts of more than 300 companes wth cross-lstngs and almost 3,000 domestc frms. The other four columns of table 3 show the number of nterlsted and domestc stocks for two subperods, 1980:02-1989:12 and 1990:01-1999:06. The number of nterlsted stocks s roughly the same for all sample perods. Ths s probably related to the fact that whle the number of cross-border lstngs has ncreased sharply n the last decade, the study of Pagano, Röell, and Zechner (2002) suggests that the rse n cross-lstngs was less marked n the late 5 Ths supposton s, however, questoned by the analyss of Stulz (1995), who fnds a consderable prcng error between the domestc CAPM and the sngle factor ICAPM of Grauer, tzenberger, and Stehle (1976) for the Swss multnatonal Nestlé.
11 1980s and early 1990s. The total amount of domestc stocks vares wdely, however. Our man emprcal analyss focuses on cross-lsted stocks. We use our sample of domestc stocks as a benchmark n order to assess to what extent the prcng error of cross-lsted stocks dverges from those of domestc companes. 4. Emprcal Results In ths secton we dscuss our emprcal analyss of companes wth nternatonal lstngs as well as domestc frms. Secton 4.1 examnes the prcng error results. In secton 4.2 we present a varance decomposton analyss that explores the contrbuton of both local and global factors to the returns of cross-lsted stocks. Ths decomposton provdes a plausble ratonale for our prcng error test results. Fnally, as a related ssue we examne the exchange rate exposure of nterlsted frms n secton 4.3. 4.1 Prcng Error As prevous studes ndcate that frms wth nternatonal lstngs are predomnantly nternatonally orented, our hypothess s that these corporatons have a consderable prcng error. The frst column of table 4 presents rejecton percentages of the Prcng Error test for nterlsted companes. Ths test examnes whether the frm s cost of captal s dfferent when estmated wth the domestc CAPM nstead of the multfactor ICAPM. We fnd a sgnfcant prcng error for approxmately 12 percent of the 336 frms. It s nterestng to note that companes wth a sgnfcant prcng error are typcally from the large countres n our sample, such as Germany, Japan, the U.K., and the U.S. The fourth column of table 4 contans rejecton frequences of the Global Beta test. Ths test evaluates the sgnfcance of the frst element of the prcng error, also referred to as the beta error. The beta error s computed as the dfference between the drect beta (the multfactor ICAPM beta d 1 ) and the ndrect beta (the global beta of the local market d 1 multpled by the CAPM beta b ) of a frm. The beta error s sgnfcantly dfferent from zero for 7.44 percent of the cross-lsted frms. In addton, table 4 shows rejecton frequences of the Prcng Error test and the Global Beta test for two subperods. For the perod 1980:02-1989:12, the Prcng Error test rejects for 4.19 percent of the 336 frms n the sample and the rejecton frequency of Global Beta test s equal to 11.98 percent. The hypothess of no prcng error s rejected for 26 out of 334
12 nterlsted companes n the subperod 1990:01-1999:06. The Global Beta test rejects for 7.49 percent of the frms. The fact that the hypothess that the prcng error s equal to zero s rejected for a smlar number of frms over the two subsamples suggests that the assumpton that betas are not tme-varyng does only have a margnal mpact on our results. Table 5 shows the average, the average of the absolute value, the standard devaton, the mnmum, and the maxmum of the beta error for our sample of cross-lsted stocks. The average beta error s depcted n the frst column of table 5 and s relatvely close to zero. 6 The second column shows that the absolute beta error amounts to around 0.1 for most countres, varyng from 0.056 for the Germany to 0.142 for Canada. The average of the absolute beta errors of all nterlsted frms n the U.S. s equal to 0.067. The (dscrete) annual return on the global market portfolo over the sample perod was 15.2 percent when expressed n U.S. dollars. The one-month rsk free rate amounted to 7.8 percent on average. Consequently, the global market rsk premum n U.S. dollars was equal to approxmately 7.4 percent. The mpled cost of captal dfferental between the CAPM and the ICAPM s then 50 bass ponts on average for U.S. frms. 7 In cost of captal terms the beta error amounts to 53 bass ponts for Germany, 90 bass ponts for Japan, 80 bass ponts for the U.K., and 112 bass ponts for Canada. Averaged over all countres, the mpled cost of captal dfference s approxmately 80 bass ponts for nterlsted stocks. Table 6 depcts the results of both prcng error tests for our benchmark sample of domestc stocks. On average, the Prcng Error test rejects for 4.40 percent of the frms. Ths number vares only slghtly across countres. Column 4 of table 6 shows the rejecton percentages per country of the Global Beta test. Ths test detects a sgnfcant beta error for 2.44 percent of the domestc corporatons. Table 6 also presents test results for subperods, whch are remarkably smlar to the results for the whole sample perod. Summary statstcs of the frst element of the prcng error for domestc frms are presented n table 7. The average beta error s depcted n the frst column and s close to zero, as expected. The second column 6 The value-weghted sum of the ICAPM betas equals unty. Also, each local market s prced correctly by the ICAPM, accordng to the nternatonally undversfable rsks of that portfolo. By constructon the market weghted average prcng error s equal to zero. Ths means that for an ndvdual frm the CAPM and the ICAPM mght gve dfferent cost of captal but on average, (value-weghted) domestc prcng provdes the correct cost of captal. Note that the above characterstcs only hold n a world where both local and global market ndexes are measured perfectly ncludng all ndvdual stocks. Non-zero average prcng errors arse frst because we do not use all stocks ncluded n the local and global MSCI ndces, and second because we present equally weghted averages. 7 In the absence of currency rsk prema (and n the absence of devatons from the restrcton α 1 =α 2 + b α ) the dfference (d b - d )E[R G - r 0 ] would gve an estmate of the cost of captal dfference between the domestc and the nternatonal CAPM.
13 shows that the absolute beta error s approxmately 0.1 for most countres, varyng from 0.077 for the U.S. to 0.123 for France. The mpled cost of captal dfferental s equal to 57 bass ponts for the U.S., 75 bass ponts for Germany and Japan, 70 bass ponts for the U.K., and 106 bass ponts for Swtzerland. On average, the estmated cost of captal dfferental for domestc stocks s very smlar to the dfferental for nterlsted stocks. Hence, the evdence ndcates that the prcng error s very nfrequently sgnfcantly dfferent from zero for domestc frms as well. Overall, our prcng error results provde lttle evdence for our hypothess that the prcng error s economcally and statstcally large for cross-lsted frms. The percentage of frms wth a sgnfcant prcng error s only slghtly larger for cross-lsted companes than for domestc frms. Secton 4.2 attempts to explore these results by decomposng the varance of a cross-lsted stock nto local and global factors. The am of ths analyss s to assess the margnal contrbuton of the global market ndex and the currency factors to the explanatory power of the domestc stock market portfolo. 4.2 Varance Decomposton In ths secton we nvestgate how much of the rsk that s specfc n the local market s systematc n the global captal market. We assess the respectve contrbutons of the local market, the global market and the vector of exchange rate changes to an ndvdual asset s return n a varance decomposton analyss. Ths analyss may shed lght on our fndng that the domestc CAPM leads to a dfferent estmate of a frm s cost of captal than the multfactor ICAPM for a small percentage of the frms wth foregn lstngs n our sample. The decomposton assesses how much the global market ndex and the currency rsk factors add to the local market ndex as a measure of systematc rsk n the CAPM. We consder the regresson R 5 + R b + η Z h ξ, (10) = α + where η Z represents the resdual vector from regressng Z on R. In equaton (10) we can estmate the margnal contrbuton of the global factors to the explanatory power of the regresson condtonal on the contrbuton of the local market. Under the null hypothess that the prcng error s equal to zero, the global rsk factors are fully accounted for by the local market ndex. Equaton (10) s a smple reparametrzaton of equaton (8), but drectly yelds the addtonal explanatory power of the global factors Z. Takng the varance of both the left and the rght hand sde of equaton (10), the varance of stock can be decomposed as
14 Ω d d Ω ω = b ω + h (Ω + )h + σ. (11) 2 2 2 2 ω In equaton (11) the total varance of stock (denoted by ω 2 ) s decomposed nto systematc local market rsk (related to the varance ω 2 of the local market return), addtonal global rsk n Z that s orthogonal to the local market (related to the covarance matrx Ω of Z) and specfc rsk σ 2. Note that the contrbuton of the global factors should be zero under the null hypothess that the cost of captal dfferental s equal to zero. That s, the estmate of h must equal zero under the null hypothess. Fgure 2 presents the average varance decomposton of all cross-lsted frms per country. The varance decomposton for a country s a weghted average of the decompostons for all ndvdual frms n that country wth the specfc rsk of these frms as weghts. Obvously, the margnal contrbuton of the global factors Z across frms n one country s very small on average. Whle the exchange rate rsk factors exhbt some explanatory power, the contrbuton of the global market ndex s trval. Fgure 2 thus confrms our fndng that the domestc CAPM and the multfactor ICAPM yeld a dfferent estmate of the cost of captal for a relatvely small percentage of frms. The varance decomposton analyss ndcates sgnfcant country effects n nterlsted stock returns, consstent wth the evdence of Heston and Rouwenhorst (1994) and Grffn and Karoly (1998). Fgure 2 suggests that nterlsted frms wthn one country share a common exposure to the global market and currency factors. The exposure to global factors appears to be captured n the nternatonal prcng of the local market ndex, ndcatng that the local market s a suffcent statstc for measurng a frm s senstvty to global factors. Ths means that even n ntegrated markets the prcng error s very small for most frms, because the local market factor can serve as a proxy for the omtted global factors n the domestc CAPM. A sgnfcant prcng error arses only for frms that have sgnfcantly devatng exposure from the average frm n ther country. Our evdence ndcates that the senstvty of stocks wth nternatonal lstngs to global factors does not substantally devate from the exposure of the average frm n the market ndex of ther home country. The results n sectons 4.1 and 4.2 ndcate that a frm s rsk profle s closely lnked to ts home country. Ths holds for the large majorty of the cross-lsted frms n our sample. A tentatve explanaton of ths fndng s related to what De Ménl (1999) calls lack of real captal market ntegraton. De Ménl (1999) fnds that both cyclcal, structural, and nsttutonal country-specfc factors sgnfcantly contrbute to the explanaton of cross- 2
15 country dfferences n ROA for large non-fnancal frms. More n partcular, he fnds sgnfcant effects for the level of captal deepenng and for the regulatory envronment. Wth respect to the latter, De Ménl ponts to labor and product market regulaton as sgnfcant determnants of frm performance. In ths respect, all frms wthn the same country face smlar constrants and opportuntes. In short, wth a lack of real captal market ntegraton and substantal cross-country dfferences n market regulaton, a country s fortunes and the fortunes of the frms operatng n ths country are closely ted together. It may be true that certan frm characterstcs such as sze and degree of nternatonal actvtes play a role n explanng the devatng exposure of a frm relatve to the local market. Further research s requred to examne ths ssue. Increasng harmonzaton of regulatory polces as s happenng n the EU wll reduce these structural dfferences. In the same ven, ncreasng real ntegraton wll reduce cyclcal dfferences. For the tme beng, substantal dfferences reman between countres and frms across countres. These dfferences could be used by ndvdual nvestors for the purpose of portfolo dversfcaton. Note that the lack of real ntegraton s separate from the ssue of fnancal ntegraton. Because we take the ICAPM as the null-hypothess, we mplctly assume that stock markets are fully ntegrated. Consequently, our results have no mplcatons for the fnancal ntegraton of nternatonal captal markets. In secton 4.3 we present another way to llustrate the mportance of country factors n the returns of nterlsted stocks. We show that the prcng error tests n ths paper are very smlar to the well-known tests for foregn exchange rate exposure. We employ varous exposure tests for our sample of nterlsted frms. Our analyss ndcates that currency exposure test are smlarly affected by country factors as our tests for prcng errors. 4.3 Country Factors and Exchange Rate Exposure In ths secton we analyze foregn exchange rate exposure for cross-lsted companes. Adler and Dumas (1984) defne exchange rate exposure as the mpact of exchange rate movements on the value of a frm. We test for currency exposure of ndvdual companes n the tmeseres regresson 8 R = γ 0 + R γ 1 + S γ 2 + ε. (12) 8 Recent papers n the lterature, e.g. Joron (1990), Bartov and Bodnar (1994), and He and Ng (1998), base ther tests on an analogous regresson, but use a trade-weghted exchange rate ndex.
16 The null-hypothess of the test for currency exposure can be formulated as H 0 : γ 2 = 0. Ths test s called the Exposure test. It uses a subset of the orthogonalty condtons n equaton (8). As shown n secton 2, testng for a prcng error bols to examnng whether a set of nstrumental varables s orthogonal to the resduals from the domestc CAPM regresson (4). The Exposure test can also be nterpreted as a prcng error test as t analyzes whether any systematc currency rsk can be fltered out from the rsk of a frm that s dversfable domestcally. As suggested n secton 4.2, foregn currency exposure as estmated n equaton (12) may (n part) be captured by the domestc market factor, as most frms wthn a country exhbt a common exposure to global factors. In order to control for ths country factor effect we also run the alternatve regresson R = c0 + S c1 + η G c2 + η c3 + ν, (13) where η G s the resdual vector from regressng R G on an ntercept and S. Smlarly, η s the resdual vector from regressng R on an ntercept, R G and S. By orthogonalzng R, we want to accomplsh that the coeffcent on S does not merely reflect the devatng exposure of frm from the average currency exposure of all frms n the country. The test of c 1 = 0 s called the Total Exposure test. A thrd alternatve test for exchange rate exposure we consder s based on the regresson R α + u (14) = 1 + RG d 1 + S d 2 Note that equaton (14) s the same as equaton (2). The test of H 0 : d 2 = 0 looks for sgnfcant Currency Betas. Several recent studes n the lterature, e.g. Joron (1990), Bartov and Bodnar (1994), and He and Ng (1998), hardly fnd any evdence of sgnfcant foregn exchange rate exposure n a varety of samples. Bartov and Bodnar argue that these results may be partly due to sample selecton crtera. We expect to fnd consderable exposure to exchange rates n our sample of cross-lsted companes, as a hgh percentage of ther sales are realzed abroad. Table 8 shows the percentage of nterlsted frms wth sgnfcant exposure to foregn exchange rates. Column 1 depcts the rejecton percentages of the Exposure test, whch s very smlar to the tests used n the recent lterature. Ths test s rejected for 25 percent of the crosslsted companes. Ths result s comparable to e.g. He and Ng (1998), who fnd sgnfcant exposure for 25 percent n a sample of Japanese frms. The rejecton percentages of the Total Exposure test as depcted n column 2 of table 8 are mportantly hgher than those of the
17 Exposure test. Almost 82 percent of the cross-lsted stocks exhbt sgnfcant exposure. Smlar fgures are obtaned wth the Currency Betas test. Ths mples that the results from the Exposure test are strongly affected by the country factors n the data. When we control for ths effect, we fnd strong evdence for our hypothess that the value of stocks wth overseas lstngs s hghly senstve to fluctuatons n exchange rates. 9 It could be argued that the stock prces of cross-lsted companes can be expected to be relatvely responsve to exchange rate shocks, as the absence of arbtrage opportuntes mples that stock prces n the home and foregn are equal when expressed n a common currency. We contend, however, that ths type of currency adjustments plays a role at a much hgher frequency than we consder. Moreover, the evdence of Grammg, Melvn, and Schlag (2000) suggests that the hgh-frequency adjustment to exchange rate shocks may well be born by the prce n the dervatve market, whle we only study stock returns n the home market. 5. Conclusons As companes become more and more nternatonally orented, nternatonal lstngs are an ncreasngly mportant part of a frm s long-term strategc polcy. Two man ssues can be dstngushed n the lterature on stocks wth overseas lstngs. The frst strand of the lterature focuses on the queston whether the stock market performance, the lqudty, and the cost of captal of a company change as a consequence of lstng abroad. The second strand examnes the motvatons and features of companes that obtan an overseas lstng. We focus on the queston whether nternatonal and domestc asset prcng models lead to dfferent estmates of the cost of captal for nterlsted companes. We examne the socalled prcng error, whch s lnearly related to the computed cost of captal dfferental, for a sample of monthly data for 336 cross-lsted frms from nne major ndustralzed countres from 1980 to 1999. We dstngush between: () the multfactor ICAPM of Solnk-Sercu ncludng both the global market portfolo and exchange rate rsk prema, and () the sngle factor domestc CAPM. Our hypothess s that the prcng error s consderable for nterlsted frms, as they are relatvely nternatonally orented. We fnd a sgnfcant cost of captal dfferental for only 12 percent of the cross-lsted corporatons, however. The cost of captal dfferental between the 9 Estmaton results for subperods are qualtatvely smlar. They are not reported n ths paper but are avalable from the authors on request.
18 domestc CAPM and the ICAPM amounts to 50 bass ponts for the U.S., 75 bass ponts for the U.K., and 100 bass ponts for France. Our analyss thus provdes lttle evdence n favor of our hypothess that companes wth an overseas lstng exhbt a relatvely large prcng error. Usng a varance decomposton analyss we demonstrate that ths results are probably due to strong country factors n the data. Frms wthn a country generally exhbt a smlar exposure to nternatonal currency and stock market factors. Snce such average exposure s captured n the nternatonal prcng of the local stock market ndex, the CAPM nduces a prcng error only for frms that have sgnfcantly devatng exposure. Most companes can therefore rely on the domestc CAPM for the computaton of ther cost of captal. A tentatve explanaton for the strong country-specfc factors n ndvdual stock returns s a lack of real captal market ntegraton, due to both cyclcal, structural, and nsttutonal country-specfc factors. As asset returns contan large country-specfc components, nvestng n dfferent ndustres wthn one country s nsuffcent to reap all the benefts of portfolo dversfcaton n a global settng. In that sense, our evdence renforces the home bas puzzle. Further research s requred to examne these ssues.
19 Appendx A In ths appendx we show that the prcng error of the CAPM as compared to the multfactor ICAPM of Solnk-Sercu can be expressed as a lnear combnaton of the parameter δ n the regresson (8) n the text Z R R ζ δ β α + + + = 4. (A1) Ths s equaton (8) n the paper. The moment condtons of equaton (A1) can be wrtten as Ω = Ω Ω Ω d b d d 2 2 ω δ β ω, (A2) where Ω s the (N+1) (N+1) covarance matrx of Z, ω 2 s the varance of R, and d s the vector of regresson parameters n regresson (5) n the text u d Z R + + = α, (A3) for the local market portfolo. The covarance between Z and R s therefore equal to Ωd. Solvng for δ from the second lne of (A2) we get d d β δ =. (A4) Substtutng ths expresson nto the frst lne of equaton (A2) gves 2 2 2 p d b d d d d b σ ω ω β Ω = Ω Ω =, (A5) where p = d b - d s the prcng error and σ 2 s the varance of resduals u. Substtutng ths expresson for β back nto equaton (A4) yelds p d d Ω + Ι = 2 σ δ. (A6) Equaton (A6) can be rewrtten as d d p δ σ 1 2 Ω + Ι =. (A7) Note that d, Ω, and σ 2 are unrelated to asset and are treated as exogenous.
20 References Adler, M., and B. Dumas, 1983, Internatonal Portfolo Choce and Corporaton Fnance: A Synthess, Journal of Fnance, 38, 925 984. Adler, M., and B. Dumas, 1984, Exposure to Currency Rsk: Defnton and Measurement, Fnancal Management, 13, 41 50. Bartov, E., and G.M. Bodnar, 1994, Frm Valuaton, Earnngs Expectatons, and the Exchange Rate Exposure Effect, Journal of Fnance, 49, 1755 1785. Bekaert, G., and C.R. Harvey, 1995, Tme-Varyng World Market Integraton, Journal of Fnance, 50, 403 444. Bddle, G., and S. Saudagaran, 1989, The Effects of Fnancal Dsclosure evels on Frms Choces among Alternatve Foregn Stock Exchanges, Journal of Internatonal Fnancal Management and Accountng, 1, 55 87. Campbell, J.Y., and Y. Hamao, 1992, Predctable Stock Returns n the Unted States and Japan: A Study of ong-term Captal Market Integraton, Journal of Fnance, 47, 43 69. De Ménl, G., 1999, Real Captal Market Integraton n the EU: How Far Has t Gone? What Wll the Effect of the Euro Be? (wth dscusson), Economc Polcy, 28, 167 204. De Sants, G., and B. Gérard, 1997, Internatonal Asset prcng and Portfolo Dversfcaton wth Tme-Varyng Rsk, Journal of Fnance, 52, 1881 1912. De Sants, G., and B. Gérard, 1998, How Bg s the Premum for Currency Rsk?, Journal of Fnancal Economcs, 49, 375 412. Doukas, J., and.n. Swtzer, 2000, Common Stock Returns and Internatonal stng Announcements: Condtonal Tests of the Mld Segmentaton Hypothess, Journal of Bankng and Fnance, 24, 471 502. Dumas, B., and B. Solnk, 1995, The World Prce of Exchange Rate Rsk, Journal of Fnance, 50, 445 479. Eun, C.S., and S. Sabherwal, 2002, Cross-Border stngs and Prce Dscovery: Evdence from U.S. sted Canadan Stocks, Journal of Fnance, forthcomng. Foerster, S.R., and G.A. Karoly, 1993, Internatonal stngs of Stocks: The Case of Canada and the U.S., Journal of Internatonal Busness Studes, 24, 763 784.
21 Foerster, S.R., and G.A. Karoly, 1998, The Effects of Market Segmentaton and Investor Recognton on Asset Prces: Evdence from Foregn Stock stngs n the U.S., workng paper, Oho State Unversty. Fuerst, O., 1998, A Theoretcal Analyss of the Investor Protecton Regulatons Argument for Global stng of Stocks, workng paper, Yale Unversty. Grammg, J., M. Melvn, and C. Schlag, 2000, Prce Dscovery n Internatonal Equty Tradng, workng paper, Arzona State Unversty. Grauer, F..A., R.H. tzenberger, and R.E. Stehle, 1976, Sharng Rules and Equlbrum n an Internatonal Captal Market Under Uncertanty, Journal of Fnancal Economcs, 3, 233 256. Grffn, J.M., and G.A. Karoly, 1998, Another ook at the Role of the Industral Structure of Markets for Internatonal Dversfcaton Strateges, Journal of Fnancal Economcs, 50, 351 373. Gultekn, M.N., Gultekn, N.B., and A. Penat, 1989, Captal Controls and Internatonal Captal Market Segmentaton: The Evdence from the Japanese and Amercan Stock Markets, Journal of Fnance, 44, 849 869. Harvey, C.R., 1991, The World Prce of Covarance Rsk, Journal of Fnance, 46, 111 157. He, J., and.k. Ng, 1998, The Foregn Exchange Exposure of Japanese Multnatonal Corporatons, Journal of Fnance, 53, 733 753. Heston, S.., and K.G. Rouwenhorst, 1994, Does Industral Structure explan the Benefts of Internatonal Dversfcaton?, Journal of Fnancal Economcs, 36, 3 27. Joron, P., 1990, The Exchange Rate Exposure of U.S. Multnatonals, Journal of Busness, 63, 331 345. Joron, P., and E. Schwartz, 1986, Integraton vs. Segmentaton n the Canadan Stock Market, Journal of Fnance 41, 603 616. Karoly, G.A., 1998, Why Do Companes st Shares Abroad?: A Survey of the Evdence and Its Manageral Implcatons, Fnancal Markets, Insttutons & Instruments, 7, Number 1. Koedjk, C.G., C.J.M. Kool, P.C. Schotman, and M. A. van Djk, 2001, The Cost of Captal n Internatonal Fnancal Markets: ocal or Global?, CEPR Dscusson Paper, 3062, forthcomng (2002) Journal of Internatonal Money and Fnance. Pagano, M., A.A. Röell, and J. Zechner, 2002, The Geography of Equty stng: Why Do Companes st Abroad?, Journal of Fnance, forthcomng.