NBER WORKING PAPER SERIES WHAT MATTERS FOR FINANCIAL DEVELOPMENT? CAPITAL CONTROLS, INSTITUTIONS, AND INTERACTIONS. Menzie D.

Transcription

1 NBER WORKING PAPER SERIES WHAT MATTERS FOR FINANCIAL DEVELOPMENT? CAPITAL CONTROLS, INSTITUTIONS, AND INTERACTIONS Menze D. Chnn Hro Ito Workng Paper NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 Massachusetts Avenue Cambrdge, MA May 2005 Helpful comments were receved from the co-edtor Carlos Vegh, an anonymous referee, Joshua Azenman, Mchael Hutchson, Carl Walsh, Frank Warnock, Tom Wllett, partcpants at the Kennedy School Luncheon on Internatonal Economc Polcy, the USC development semnar, and the UCSC brown bag. We also thank Ashok Mody and Denns Qunn for provdng data. Fnancal support of faculty research funds of UC Santa Cruz are gratefully acknowledged.the vews expressed heren are those of the author(s) and do not necessarly reflect the vews of the Natonal Bureau of Economc Research by Menze D. Chnn and Hro Ito. All rghts reserved. Short sectons of text, not to exceed two paragraphs, may be quoted wthout explct permsson provded that full credt, ncludng notce, s gven to the source.

2 What Matters for Fnancal Development? Captal Controls, Insttutons, and Interactons Menze D. Chnn and Hro Ito NBER Workng Paper No May 2005 JEL No. F36, F43, G28 ABSTRACT We extend our earler work, focusng on the lnks between captal account lberalzaton, legal and nsttutonal development, and fnancal development, especally that n equty markets. In a panel data analyss encompassng 108 countres and twenty years rangng from 1980 to 2000, we explore several dmensons of the fnancal sector. Frst, we test whether fnancal openness can lead to equty market development when we control for the level of legal and nsttutonal development. Then, we examne whether the openng of the goods sector s a precondton for fnancal openng. Fnally, we nvestgate whether a well-developed bankng sector s a precondton for fnancal lberalzaton to lead to equty market development and also whether bank and equty market development complements or substtutes. Our emprcal results suggest that a hgher level of fnancal openness contrbutes to the development of equty markets only f a threshold level of general legal systems and nsttutons s attaned, whch s more prevalent among emergng market countres. Among emergng market countres, a hgher level of bureaucratc qualty and law and order, as well as the lower levels of corrupton, ncreases the effect of fnancal openng n fosterng the development of equty markets. We also fnd that the fnance-related legal/nsttutonal varables do not enhance the effect of captal account openng as strongly as the general legal/nsttutonal varables. In examnng the ssue of the sequencng, we fnd that the lberalzaton n cross-border goods transactons s found to be a precondton for captal account lberalzaton. Our fndngs also ndcate that the development n the bankng sector s a precondton for equty market development, and that the developments n these two types of fnancal markets have synergstc effects. Menze D. Chnn Department of Economcs Unversty of Wsconsn 7470 Socal Scence Buldng 1180 Observatory Drve Madson, WI and NBER mchnn@lafolette.wsc.edu Hro Ito Department of Economcs Portland State Unversty 1721 SW Broadway Portland, OR to@pdx.edu

3 1. Introducton Recent years have wtnessed a surge of nterest n ssues surroundng globalzaton, ncludng fnancal globalzaton. A seres of fnancal crses n the 1990s rekndled the debates over the effects of removng captal controls, whch led many observers to reconsder the balance between the costs and benefts of fnancal lberalzaton (Kamnsky and Schmukler, 2001a,b, 2002; Schmukler, 2003). 1 Many studes have adopted the poston that captal account lberalzaton can affect economc development through fnancal development; specfcally, lberalzed fnancal markets may contrbute to developng fnancal markets that provde funds to borrowers who may have productve nvestment opportuntes. 2 Theory suggests that captal account lberalzaton can lead to development of fnancal systems through several channels. Frst, fnancal lberalzaton may mtgate fnancal represson n protected fnancal markets, allowng the (real) nterest rate to rse to ts compettve market equlbrum (McKnnon, 1973; Shaw, 1973). Second, removng captal controls allows domestc and foregn nvestors to engage n more portfolo dversfcaton. These two ponts can be summarzed as that fnancal lberalzaton may reduce the cost of captal and ncreasng ts avalablty for the borrowers. Stulz (1999) shows that fnancal globalzaton reduces the cost of equty captal because of the reducton n the expected returns to compensate rsk as well as n agency costs (also, Henry, 2000; Bekaert et al., 2000, 2001). Thrd, not least, the lberalzaton process usually ncreases the effcency level of the fnancal system by weedng out neffcent fnancal nsttutons and creatng greater pressure for a reform of the fnancal nfrastructure (Claesens et al., 2001; Stulz, 1999; Stgltz 2000). Such an mprovement n fnancal nfrastructure may allevate nformaton asymmetry, decreasng adverse selecton and moral hazard, and further rasng the avalablty of credt. The lnk between fnancal lberalzaton and fnancal development s not unequvocal, however. It s often argued that to beneft from more open cross-border fnancal transactons, 1 In ths study we do not dscuss the merts of captal controls n the context of fnancal crses. For a revew, see Azenman (2002). Kletzer and Mody (2000) survey the debate n the context of self-protecton polces for emergng markets. Ito (2004) nvestgates the correlaton between fnancal lberalzaton and the output performance of crss-ht economes. 2 See for nstance Leahy, et al. (2001) for OECD-specfc results. Klen and Olve (2001) document the lnkage between fnancal development and economc growth for developed countres, and ts absence for less developed countres. Spegel (2001) examnes an APEC sample, whle Arteta, Echengreen and Wyplosz (2001) document the fraglty of many of these group-specfc results. IMF (2001, Chapter 4) surveys both the growth and fnance, and fnance and lberalzaton lteratures. For the most recent revew on fnance and growth, refer to Qunn, et al. (2002) 1

4 fnancal systems need to be equpped wth reasonable legal and nsttutonal nfrastructure. In economes where the legal system does not clearly defne property rghts or guarantee the enforcement of contracts, the ncentves for loan actvtes can be lmted. Legal protectons for credtors and the level of credblty and transparency of accountng rules are also lkely to affect economc agents fnancal decsons. 3 Levne, Loayza, and Beck (2000) nvestgate whether the level of legal and regulatory determnants of fnancal development nfluences the development fnancal ntermedary sector. La Porta, Lopez-de-Slanes, Shlefer, and Vshny (hereafter LLSV, 1997, 1998) argue that the natonal legal orgn (whether Englsh, French, German, or Scandnavan) strongly affects the legal and regulatory envronment n fnancal transactons and explans cross-country dfferences n fnancal development. LLSV (1997, 1998) and Levne (1998, 2003) show that low levels of shareholder rghts are assocated wth poorly developed equty markets (especally n French cvl law countres). In contrast, Common law countres have hgh levels of shareholder rghts wth correspondngly hgh levels of equty market development (Claessens, et al., 2002 and Capro, et al., 2003), and that greater credtor rghts are postvely assocated wth fnancal ntermedary development. 4 Clearly, the lnk between fnancal lberalzaton and fnancal development s of great mportance to emergng market polcymakers, and naturally, one needs to examne the effect of lberalzng cross-border fnancal transactons n the context the nsttutonal settng. However, very lttle nvestgaton has been made to shed lght on the lnk n such a context. Ths s the topc of our study. In ths artcle, we extend our work (Chnn and Ito, 2002) focusng on the lnks between captal account lberalzaton, legal and nsttutonal development, and fnancal development. Snce our study s motvated by the development of equty markets through fnancal lberalzaton and ts contrbuton to economc development among emergng market countres n the 1990s, our focus s on the effect of captal account lberalzaton on the development of equty markets among less developed and emergng market countres. We conduct a panel data analyss encompassng 108 countres (ncludng 21 ndustralzed countres and 31 emergng market countres) and twenty 3 For the analyss of legal development on fnancal development, see Beck and Levne (2004), Claessens, et al. (2002), Capro, et al. (2003), and Johnson, et al. (2002). For a general dscusson on the mportance of legal and nsttutonal foundatons for fnancal development, see Bem and Calomrs (2001) and Stultz (1999). 4 Rajan and Zngales (2003), on the other hand, queston the lnk between legal orgns and cross-country dfference n fnancal development, and nstead stress the mportant role of poltcal forces n shapng polces toward fnancal markets and ther development. 2

5 years rangng from 1980 to In addton to searchng for the effects of each set of factors, we examne the oft-dscussed ssue of the sequence of lberalzaton. It has been often argued that countres need to lberalze ther goods market pror to lberalzng fnancal sector (McKnnon, 1991). Also, n order for fnancal systems to reap the beneft of fnancal lberalzaton, the systems themselves need to be developed up to a certan level, suggestng the mportance of a sequence of lberalzaton wthn the fnancal sector (Martell and Stulz, 2003). To test these clams, we examne whether the openng of the goods sector s a precondton for fnancal openng, and furthermore, nvestgate whether a well-developed bankng sector s a precondton for fnancal lberalzaton to lead to equty market development. Addtonally, we explore whether bank and equty market development complements or substtutes. Our emprcal results suggest that a hgher level of legal and nsttutonal development contrbutes both drectly and n an nteractve manner wth fnancal openness to the development of equty markets, but only f a country s equpped wth a reasonable level of legal and nsttutonal development, whch s more prevalent among emergng market countres than developng countres. A hgher level of bureaucratc qualty and law and order, as well as the lower levels of corrupton, may enhance the effect of fnancal openng n fosterng the development of equty markets. We also fnd that, among emergng market countres, the overall level of fnance-related legal/nsttutonal development ncreases stock market tradng volumes and enhances the effect of fnancal openness. However, the fnance-related legal/nsttutonal varables do not exhbt as strong an effect as the general legal/nsttutonal varables. In examnng the ssue of the sequencng, we fnd that the lberalzaton n cross-border goods transactons s found to be a precondton for captal account lberalzaton, n a result smlar to that obtaned by Azenman and Noy (2004). Our fndngs also ndcate that the development n the bankng sector s a precondton for equty market development, and that the developments n these two types of fnancal markets have nteractve effects. 2. An Econometrc Analyss of Openness, Insttutons and Fnancal Development The lnk between captal account openness, fnancal development, and legal/nsttutonal envronment has been nvestgated by Chnn and Ito (2002). In ths study, we demonstrated that fnancal systems wth a hgher degree of legal/nsttutonal development on average beneft more 3

6 from fnancal lberalzaton than those wth a lower one. Furthermore, the postve effect of legal/nsttutonal development seems to flow prmarly from the degree of shareholder protecton and accountng standards. In what follows, we extend our prevous study by employng updated data and also explorng more questons related to the lnk between captal account openness and fnancal development. More specfcally, we wll nvestgate the ssues relevant to the sequence of lberalzaton between fnancal and goods cross-border flows and the sequence of development n bankng and equty markets. 2.1 The Emprcal Specfcaton Frst, we reexamne the long-term effect of captal account openness on fnancal development n a model that controls for the level of legal and nsttutonal development. The model s specfed as: (1) FD FD = γ + ρfd + γ KAOPEN + γ L + γ ( L KAOPEN ) + X Γ + u, t t 5 0 t 5 1 t t 5 t 5 where FD s a measure of fnancal development; KAOPEN s a measure of fnancal openness; X s a vector of economc control varables; and L refers to a measure of legal or nsttutonal development. For the captal openness varable, we use the Chnn-Ito ndex whch s descrbed n greater detal n a later secton and the data appendx. The vector X contans macroeconomc control varables that nclude log per capta ncome n PPP terms, the nflaton rate, and trade openness, measured as the rato of the sum of exports and mports to GDP. In ths analyss, the set s kept farly small so as to retan some nterpretablty of the correlatons. Log per capta ncome s ncluded as there s a long lterature ascrbng fnancal deepenng, asde from the role of regulaton, to the ncreasng complexty of economc structures assocated wth rsng ncome. The nflaton rate s ncluded because t may dstort decson-makng. 5 In partcular, moderate to hgh nflaton may dscourage fnancal ntermedaton, and encourage savng n real assets. Fnally, trade openness s ncluded as an ad hoc control; many emprcal studes fnd a correlaton of trade openness wth any number of economc varables. The relatonshp between trade openness and fnancal openness wll be nvestgated more thoroughly n a later secton. t 5 Snce n most cases, the volatlty of nflaton rses wth the nflaton rate, the nflaton rate could be proxyng for ether or both of these effects. 4

7 A seres of regressons s conducted for each of the fnancal development varables (FD), whch nclude prvate credt creaton (PCGDP), stock market captalzaton (SMKC), stock market total value (SMTV), all measured as a rato of GDP, and stock market turnover (SMTO). Also for the seres of regressons wth dfferent fnancal development measures, we also nclude each of the nne legal/nsttutonal varables and ts nteractve term wth the captal account openness ndex. The nne legal/nsttutonal varables nclude those whch are pertanng to the general development of legal systems or nsttutons as well as those pertanng partcularly to fnancal transactons. Further dscussons about the legal/nsttutonal varables are presented n the data secton. In order to avod problems of endogenety assocated wth short term cyclcal effects, we specfy our model as a growth rate on levels regresson, akn to a panel error-correcton model wth non-overlappng data. That s, we only sample data every fve years between 1980 and 2000, and use the fve-year average growth of the level of fnancal development as the dependent varable and the ntal condtons for tme-varant explanatory varables, ncludng the ntal level of the fnancal development ndcator, for each fve-year panel The Data The data are drawn from a number of sources, prmarly the World Bank s World Development Indcators, the IMF s Internatonal Fnancal Statstcs, and the databases assocated wth Beck, Demrgüc-Kunt, and Levne (2000). The analyss s based upon data orgnally recorded at an annual frequency, over the perod, coverng 108 countres. Detals are reported n Appendx Fnancal Development Measures The measures of fnancal development are extracted from the dataset of Beck, Demrgüc-Kunt, and Levne (2000). PCGDP, the rato of prvate credt from depost money banks to the prvate sector, represents the overall development n prvate bankng markets. 7 Whle ths varable s examned for purposes of comparson, our focus s prmarly on the development of equty markets development, for whch we use three varables as the measures: SMKC (stock market captalzaton), SMTV (total value of stocks traded), and SMTO (stock market turn over 6 Tme fxed effects are also ncluded n the model to control for possble tme-specfc exogenous shocks. 7 Whle many researchers use M2 or lqudty labltes (M2Y and LLY n our data set), we only report results for PCGDP as our focus s on equty market development, and also because the correlaton between M2Y or LLY and 5

8 rato). We can consder SMKC as the measure of the sze of equty markets and SMTV and SMTO as the measure of the actveness of equty markets. Regardng the measurement ssues surroundng fnancal development, we must make two more notes. Frst, n ths study, we do not look nto offshore markets as part of fnancal development, and therefore, focus merely on the development of domestc equty markets. Although we have wtnessed through the Asan fnancal crss that some emergng market countres such as Korea and Thaland tred to complement ther domestc markets by developng offshore markets and allowng foregn nvestors, mostly hedge funds, to actvely engage, there have not developed so much lterature regardng these ssues, manly owng to the recentness of the development of hedge funds and offshore markets. 8 Furthermore, the relatonshp between onshore and offshore funds has not been rgorously nvestgated (except for Km and We, 2002) due to data unavalablty. 9 Hence, we do not attempt to lnk offshore funds to domestc equty markets. Secondly, our analyss does not consder the effect of captal account lberalzaton on the overseas lstngs by nternatonal frms. Several authors have dscussed the process of emergng market companes mgratng to mature markets by lstng ther shares and tradng n New York or other major stock markets. The mplcatons of ths process s under debate. Claessens et al. (2002) and Levne and Schmukler (2003) provde evdence that mgraton of tradng from domestc to nternatonal markets allows frms to reap from nternatonal portfolo dversfcaton gans, whle contrbutng to the shrnkage of local markets. On the other hand, Karoly (2003) shows that cross-lstngs n the form of Amercan Depostary Recepts (ADRs) nether facltate nor hnder local market development, but that t may be an outcome of the declnng market condtons and not a cause of them. 10 Gven the ambguty of the effects, and data lmtatons, we focus on the development of domestc equty markets. PCGDP s qute hgh (84.9% and 81.9%, respectvely),. 8 The exceptons nclude Fung and Hseh (2001), ther other works, Brown and Goetzmann (2001), and Brown et al. (1998). 9 Km and We study the behavor of the Korean offshore market, and fnd that the Korean offshore funds trade more aggressvely than onshore markets; but they do not engage n postve feedback tradng unlke ther onshore counterparts; and that they do herd, but not so much as the onshore funds n the U.S. or U.K. whch tends to herd more durng a crss. 10 Sarkssan and Schll (2002) show that nternatonalzaton of tradng s constraned by geographcal and nsttutonal/cultural famlarty, makng the gans from nternatonal portfolo dversfcaton small. Ther analyss shows that overseas lstngs are not conducted to overcome nvestor home bas, but reflect the bas. 6

9 Table 1 reports the growth rates of fnancal development measured n the varables used n ths study. Inspecton of the table reveals that durng the 1990s, all subsample groups experenced the most rapd development n equty markets, measured along several dmensons, ncludng sze (SMKC) and transactons actvty (SMTV and SMTO). Ths s true despte the retrenchment n the equty markets of less developed and emergng market countres durng the second half of the decade. Hence, the development of equty markets has progressed albet n a haltng fashon for the last three decades Captal Openness Index the Chnn-Ito Index It s well known that t s extremely dffcult to measure the extent of openness n captal account transactons (e.g., Echengreen, 2002; Edson et al., 2002). Although many measures exst to descrbe the extent and ntensty of captal account controls, the consensus s that such measures fal to fully capture the complexty of real-world captal controls for a number of reasons. 11 Frst, conventonal measures of quantfyng captal controls (or fnancal openness) sometmes fal to account for the ntensty of captal controls. The most promnent example of such measures nclude bnary varables based upon the IMF s categorcal enumeraton reported n Annual Report on Exchange Arrangements and Exchange Restrctons (AREAER). 12 Second, IMF-based varables are too aggregated to depct the ntrcacy of actual captal controls. Captal controls can dffer dependng on the drecton of captal flows (.e., nflows or outflows) as well as the type of fnancal transactons targeted. 13 Thrdly, t s almost mpossble to dstngush between de jure and de facto controls on captal transactons. Captal control polces are often mplemented wthout explct polcy goals to control the volume and/or type of captal flows. Conversely, as Edwards (1999) dscusses, t s often the case that the prvate sector crcumvents captal account restrctons, nullfyng the expected effect of regulatory captal controls. Therefore, researchers 11 See Edson and Warnock (2001), Edwards (2001), and Edson et al. (2002) for dscussons and comparsons of varous measures on captal restrctons. Dooley (1996) provdes an extensve lterature revew and Neely (1999) presents a descrptve overvew of captal controls. 12 There are bnary varables created based on a set of on-off clarfcaton, whch ncludes an ndcator varable for the exstence of multple exchange rates (k 1 ); restrctons on current account (k 2 ); captal account transactons (k 3 ); and a varable ndcatng the requrement of the surrender of export proceeds (k 4 ). k 3 s the one often used for captal controls. In 1996, the classfcaton method n the AREAER changed and these four categores became more dsaggregated as an effort to reflect the complexty of captal controls polces. 13 Ths ssue s somewhat allevated by the recent dsaggregaton n the AREAER of the k3 category nto 13 subcategores. Johnston and Tamrsa (1998) created the tme seres of captal controls based on the new 13 dsaggregated components n the AREAER. However, ther data seres are not suffcently long; t only covers years after Most recently, Mnane (2004) constructed a set of ndces to measure the ntensty of captal controls, based on an approach akn to Johnston et al., but extendng the data back to 1983 for 34 countres. 7

10 often refer to fnancal ntegraton among countres and nterpret t as de facto restrctons on captal transactons (See De Gregoro (1998) and Rajan (2003)). In ths study, we rely upon the captal account openness ndex, KAOPEN developed by Chnn and Ito (2002). Ths ndex s the frst prncple component of the four IMF bnary varables dscussed above. 14 One of the merts of the KAOPEN ndex s that t attempts to measure the ntensty of captal controls, nsofar as the ntensty s correlated wth the exstence of other restrctons on nternatonal transactons. By the nature of ts constructon, one may argue that the KAOPEN ndex measures the extensty of captal controls because t may not drectly refer to the strngency of restrctons on cross-border transactons, but to the exstence of dfferent types of restrctons. However, measurng the extensty of captal controls may be a good proxy to the measure of ntensty of captal controls. 15 Ths pont can be made more concrete by consderng a country wth an open captal account. It may stll restrct the flow of captal by lmtng transactons on the current account restrctons or other systems such as multple exchange rates and requrements to surrender export proceeds. Alternatvely, countres that already have closed captal accounts mght try to ncrease the strngency of those controls by mposng other types of restrctons (such as restrctons on current account and requrements for surrender of trade proceeds) so that the prvate sector cannot crcumvent the captal account restrctons. Another mert of ths ndex s ts wde coverage (more than 100 countres) for a long tme perod (1970 through 2000). 16 Appendx 2 explans how KAOPEN s constructed. Table 2 presents the averages of KAOPEN and IMF bnary varables for the full sample perod of and each decade. Not surprsngly, the group of ndustralzed countres have mantaned a hgh level of captal account openness over the full sample perod whle emergng market countres seem to have slghtly more open captal accounts than less developed 14 For the extenson of the four bnary classfcatons after 1996, we followed Mody and Murshd (2005). 15 One mght thnk of the Qunn (1997) ndex as the measure of the ntensty of captal controls. The Qunn ndex s a composte measure of fnancal regulaton that ranges from 0 to 14, wth 14 representng the least regulated and most open regme. The bulk of the ndex s based upon Qunn s codng of the qualtatve nformaton contaned n the AREAER pertanng to k 2 and k 3, augmented by nformaton regardng whether the country n queston has entered nto nternatonal agreements wth nternatonal organzatons such as the OECD and EU. A complete tabulaton for the OECD members exsts, but the coverage for the less developed countres s much less extensve. The correlaton between the Qunn ndex and the Chnn-Ito ndex s found to be 83.9%, suggestng that KAOPEN s proxyng the ntensty of captal controls. The correlaton between the aforementoned ndex by Mnane (2004), whch s based on more dsaggregated AREAER nformaton on captal controls, and the Chnn-Ito ndex s found to be 80.2%. 16 The Qunn ndex s avalable for the OECD members between 1958 and 1997, but the coverage for the less developed countres s lmted to certan years (1958, 1973, 1982, 1988, and 1997). 8

11 countres. 17 Interestngly, both ndustralzed and emergng market countres acheved much more rapd fnancal lberalzaton between the 1980s and 1990s than less developed countres, especally makng the dfference between emergng market and less developed countres n captal account openness become qute large by the 1990s Measures of Legal/Insttutonal Development The legal/nsttutonal varables can be categorzed nto two groups. The frst group contans the measures related to the general development of legal systems and nsttutons, namely LEGAL1, Corrupt, LAO, and BQ. LEGAL1 s the frst prncpal component of the other three varables, and we treat ths varable as a representatve measure of the general level of legal/nsttutonal development. 18 Corrupt, LAO, and BQ measure the level of corrupton, law and order, and the qualty of the bureaucratc system, respectvely. All of these data seres ncluded n LEGAL1 are obtaned from the ICRG database. In these ndexes, hgher values ndcate better condtons. The data seres are avalable for the perod of 1984 through 1997, but are ncluded as the perod-average n order to mantan comparablty wth the LEGAL2 varables. The summary statstcs for the LEGAL1 varables are reported n Table 3. In the table, we can observe that the wde gap n the level of legal and nsttutonal development between ndustralzed and less developed countres, and also that emergng market countres have acheved a hgher level of legal development wthn the group of less developed countres. The second group of legal varables CREDITOR, ENFORCE, SHRIGHTS, and ACCOUNT pertan specfcally to fnancal transactons, and are obtaned from LLSV (1998). CREDITOR refers to the level of credtor protecton, whle ENFORCE s an ndex of the effectveness of the legal system n enforcng contracts. SHRIGHTS s a measure of shareholder protecton, and ACCOUNT s an ndex of the comprehensveness of company reports. LEGAL2 s the frst standardzed prncpal component of the four varables, and therefore depcts the overall development of the legal system governng fnancal transactons. 19 The summary statstcs for these LEGAL2 varables are reported n Table 4. Before dscussng the results, we make the followng two observatons. The frst pertans to the temporal nature of these nsttutonal varables. Although we use panel data specfcatons n 17 The cross-sectonal average of KAOPEN for the full sample perod s zero by constructon. 18 The frst egenvector for LEGAL1 was found to be (Corrupt, LAO,BQ) = (0.574, 0.580, 0.578), ndcatng that the varablty of LEGAL1 s not merely drven by any partcular seres. 19 The frst egenvector for LEGAL2 was found to be (Credtor, Enforce, Shrghts, Account) = (0.206, 0.670, 0.095, 9

12 the followng analyses, the data on legal/nsttutonal development are cross-sectonal n nature,.e., they are tme-nvarant. However, the ncluson of these tme-nvarant factors do not pose a substantal problem for our analyss, snce these characterstcs represented by the legal/nsttutonal varables are lkely to change only very slowly. 20 Moreover, we focus manly on the effect of fnancal openness on fnancal development, but not the effect of legal/nsttutonal development per se. In other words, rather than sheddng lght on how the development of nsttutons and legal systems affects fnancal development, we examne how the effect of fnancal openness changes dependng upon the envronment of nsttutons and legal systems. Therefore, tme-varaton of the legal/nsttutonal varables s not crtcal to our study. The second ssue s sample sze. Whle the orgnal panel encompasses 108 countres, the data set based on LLSV spans less than 50 countres. 21 There s mnmal mpact on the coverage of the ndustralzed countres, but the sze of the LDC sample s substantally reduced. Consequently, our LDC sample n ths porton of the analyss essentally becomes the emergng market group prevously defned. 22 Hence, for the regressons wth the frst group of legal/nsttutonal varables (.e., LEGAL1, Corrupt, LAO, and BQ), we wll present results for the full sample and the subsamples for less developed countres (LDC) and emergng market countres (EMG) whereas for those wth the second group of legal varables, we report results for the full sample and a subset ttled the LDC/EMG category, whch s manly composed of EMG countres. 2.3 Emprcal Results The regresson results for the model specfed n equaton (1) are reported n Tables 5-1 through 5-4. We focus on the coeffcents of KAOPEN t-5 (frst row), the legal varable (second row), and the nteractve term between the legal varable and KAOPEN t-5 (thrd row). Our observatons wll manly focus on the regressons wth equty market development measures and 0.707). 20 Stulz (1999) and Stgltz (2000) argue that fnancal globalzaton puts pressure on governments to mprove legal systems and nfrastructure for fnancal markets. However, to our knowledge, there s no emprcal evdence for the causalty. Also, as prevously mentoned, the ICRG legal varables are avalable snce 1984, whch also creates practcal data constrants for us to use tme-varyng varables for legal and nsttutonal development. 21 More specfcally, the data set contans LEGAL data for 38 countres, CREDITOR for 46 countres, ENFORCE for 48 countres, SHRIGHTS for 47 countres, and ACCOUNT for 40 countres. 22 The LLSV cross-sectonal data set used for ths paper ncludes only three countres whch are not ether ndustralzed or emergng market countres by our defnton. 10

13 especally those of less developed and emergng market countres. 23 Table 5-1 reports the regresson results for the models wth LEGAL1. We can see that when the regressons are controlled for the general development of legal systems and nsttutons (LEGAL1), fnancal openness (KAOPEN t-5 ) contrbutes to fnancal development n equty markets, measured by stock market total values, n both LDC and EMG subsamples. In these models, the nteractve effect between fnancal openness and legal development s also detected. Sgnfcant coeffcents for the nteractve term are also found n the models wth other measures of equty market development for both LDC and EMG subsamples. However, we must be careful about how to nterpret the overall effect of captal account openness because t depends on the level of legal development. 24 That s, gven equaton (1), the total effect of fnancal openness can be shown as: Total Effect of KA openness ( + ) KAOPEN = γ γ, 1 L 3 t 5 where L s the mean of a measure of legal development. For example, when we examne the regresson specfcatons for the fnancal development measured n stock market total value for LDC and EMG groups n Table 5-1 (columns [7] and [11]), the total effect of a one-unt ncrease n KAOPEN s calculated to be for less developed countres and for emergng market countres f we use each subsample s average of LEGAL1, 0.74 for the former group and 0.28 for the latter. Thus, although the estmated coeffcents for KAOPEN and the nteracton terms are not sgnfcantly dfferent between LDC and EMG groups, the dfference n the general level of legal and nsttutonal development makes the effect of KAOPEN on equty market development qute dfferent between the subsample groups; whle openng captal accounts, on average, leads to a lower rate of development n equty markets among less developed countres, t leads to a more rapd rate of equty market development for emergng market countres. As another example, Peru, an emergng market country n our sample, ncreased ts fnancal openness level from 1.84 to 2.27 between 1990 and Gven ts LEGAL1 level of 1.65, the ncrease n fnancal openness 23 The defnton of emergng market countres reles upon the Internatonal Fnancal Corporaton s (IFC) ndces. The group of emergng market countres n ths study refers to the countres whch were ncluded n ether IFC s Global, Investble, or Fronter Index as of By ths defnton, there are 31 EMG countres n our sample. However, as has been mentoned prevously, for the regressons wth the LEGAL2 (fnancal legal) varables, because of data avalablty, there s no dstncton between LDC and EMG, but there s a subgroup LDC/EMG. 24 In case of the regressons wth LEGAL1, the fact that the varable can be negatve for a lower value (see Table 3) also contrbutes to the complexty n the nterpretaton. 11

14 would reduce the growth rate of stock market total value by 4.1% pont annually. Argentna, another EMG country, experenced a smaller ncrease of 3.15 n ts KAOPEN varable (from 1.16 to 2.00), but because ts LEGAL1 level s 0.175, hgher than Peru s, ts SMTV s predcted to grow at an addtonal 1.9% annually. Gven that SMTV grew at 3.00% annually for emergng market countres durng the 1990s (Table 1), ths acceleraton n the growth rate s sgnfcant. Thus, openng captal accounts can be effectve for fnancal development only f a country has attaned a threshold level of legal and nsttutonal development. Table 6 makes ths pont clear. In ths table, row [A] shows the total effect of a one-unt ncrease n KAOPEN calculated usng the estmates from the regresson model wth stock market total value for the subsample of less developed countres, evaluated at the average values of the legal varable for LDC and EMG (shown n row [B]). Row [C] shows the threshold level of the legal varable, above whch a one-unt ncrease n captal account openness has a postve mpact on equty market development. Thus we can conclude that, n order for captal account openness to contrbute to the development of equty markets, countres must be possessed of a level of legal/nsttutonal development greater than LEGAL1 = Hence, emergng market countres (whose average value of LEGAL exceeds the 0.68 threshold), wll on average beneft from openng ther captal accounts. On the other hand, less developed countres wll hamper ther equty market development by openng captal accounts. In our dataset, Tunsa has a value of LEGAL1 closest to the threshold level (ts value s 0.683). The countres margnally above the threshold level nclude Inda and Morocco ( and 0.566, respectvely) among others, whle those closely below nclude Mexco and Iran ( and 0.738, respectvely) among others. Fgure 1 depcts ths clear dfference n the correlaton between captal account openness and equty market development. The upper scatter dagram (a) shows the relatonshp between the fve-year change n the KAOPEN level fve years pror to the tme perod (.e., change between t-10 and t-5) and the level change n SMTV between t-5 and the contemporary tme perod (t) for the countres whose LEGAL1 value s above the threshold level, whle dagram (b) s for those wth lower LEGAL1 varables than the threshold. Not surprsngly, we can observe that, for the countres wth above-threshold levels of general legal or nsttutonal development, KAOPEN and SMTV have a postve relatonshp. For the countres wth below-threshold levels of legal development, there s no dscernable relatonshp between captal account openng and equty market development. Also, n these countres, equty markets appear to be far from actve 12

15 regardless of the degree of fnancal lberalzaton. Table 5-2 summarzes the results from the regressons that are run wth each of the components of LEGAL1 (.e., Corrupt, LAO, and BQ) ncluded ndvdually and nteractvely. For the sake of brevty, the table shows only the coeffcents of the fnancal openness varable, the legal/nsttutonal varable, and the nteracton term. Across the dfferent models wth dfferent measures of fnancal development, the sgnfcance of the estmated coeffcents appears to be qualtatvely the same as those of the regressons wth LEGAL1. Table 6, agan, helps our nterpretaton of the overall effect of KAOPEN for the models wth SMTV. Generally, we can surmse that lberalzng captal accounts may lead to development n equty markets only when the measures aganst corrupton or law and order are hgher than the threshold levels (52.2 and 54.5, respectvely). In other words, ther measures need to be as well-establshed as n emergng market countres n order to reap the beneft of captal account lberalzaton. When we control for the level of bureaucratc qualty, fnancal openness seems to leads to fnancal development among both less developed and emergng market countres (because of both subsample groups hgher average values than the threshold level), but ts effect s hgher for the latter group. Interestngly, we can observe that the coeffcent of KAOPEN t-5 alone has a negatve sgn wherever the nteractve term has a sgnfcant coeffcent, suggestng that openng fnancal markets alone may lead to underdevelopment of equty markets, but t can be avoded only f the countres are equpped wth a reasonable level of legal/nsttutonal development. When we focus on the effect of legal development relatng partcularly to fnancal transactons (LEGAL2 and LLSV varables), the fndngs are not as decsve as n the prevous cases (results shown n Tables 5-3 and 5-4). 25 Usng LEGAL2 as the legal/nsttutonal varable (Table 5-3) n the LDC/EMG subgroup, ts nteractve effect wth LEGAL2, and the LEGAL2 level term are sgnfcant n the determnaton of the development n stock market total value. Table 6 shows that the total effect of fnancal openness s postve on average n the LDC/EMG subgroup (.e., the subgroup s average level of legal development s hgher than the threshold level). In other models for the LDC/EMG subgroup, however, nether fnancal openness nor ts nteractve term seems to affect bankng or equty market development n a sgnfcant fashon. Among the components of the fnancal legal development varable (LEGAL2), shown n 25 As prevously dscussed, due to the data avalablty of the LLSV varables, there s only one subsample for non-ndustralzed countres, whch we call LDC/EMG. 13

16 Table 5-4 (whch agan only shows the coeffcents of nterest), the level of credtor protecton seems to be an mportant factor for fnancal development n both bankng and equty sectors whle no nteractve effect s detected. When contract enforcement s used as the legal/nsttutonal varable, n all models wth equty market development, we can detect a postve nteractve effect and a negatve solo effect of the level of fnancal openness. 26 Among the models wth shareholder protecton, the level of share holder protecton seems to matter for stock market captalzaton whle the level of fnancal openness contrbutes to stock market turnover. The models usng accountng standards as the legal/nsttutonal measure do not ndcate any effect of fnancal openness on fnancal development. Our fndngs suggest that n order for fnancal openness to help develop equty markets, the level of legal and nsttutonal development s crucal. If a country tres to develop ts equty markets whle t s not equpped wth well-developed legal systems or nsttutons, openng captal accounts may even harm the development of equty markets. Also, we fnd that t s the development of general legal systems and nsttutons, not of those specfc to fnancal transactons, that s crucal for a country to beneft from openng ts fnancal markets. 2.4 Robustness Checks Here, we examne whether our baselne results are senstve to outlers. Concerns about the mpact of outlers flows from two ssues. Frst, n addton to the usual measurement error present n macroeconomc data, t s lkely that the data for fnancal development s subject to even greater measurement errors. Second, these fnancal development ndcators may unntentonally capture fnancal bubbles, although the use of fve year changes may serve to mtgate ths concern. As a pont of reference, t s useful to note that n many studes of lendng booms as fnancal crses ndcators, changes n lendng or stock market szes over a shorter wndow, of between 2 to 4 years are, often used. 27 Nonetheless, we nvestgate whether the regresson results are beng dstorted by data outlers. In order to conserve space, we merely summarze the results and our observatons below. Frst, usng the orgnal annual data, we exclude the observatons of fnancal development varables f ther annual growth rates are two standard devatons away from the mean n both 26 The patterns of the sgnfcance of the coeffcents are no concdence. By constructon, the components of ENFORCE are closest to LEGAL1 varables as shown n data appendx. 27 See Corsett, Pesent, and Roubn (1998), Chnn, Dooley and Shrestha (1999), Kamnsky (2003), Kamnsky, Lzondo and Renhart (1998), Kamnsky and Schmukler (2001b), and Sachs, Tornell and Velasco (1996). 14

17 drectons, and re-estmate the same sets of regressons. 28 The excluson of outlers shrnks the observaton sze by a relatvely small degree, about 0 11%. 29 Generally, n the re-estmated results (not reported), the magntude of the estmated coeffcents often becomes smaller, but so do the standard errors, especally for the models wth stock market related measures for the LDC and EMG subgroups. Therefore, not only does the statstcal sgnfcance of the coeffcents reman qualtatvely unchanged, but n addton some of the coeffcents whch were prevously nsgnfcant become sgnfcant. Interestngly, n many models the adjusted R-squared ncreases. Ths s most apparent n the models wth LEGAL2 and ACCOUNT. In ths analyss for the LDC/EMG group, the coeffcent of LEGAL2 s now sgnfcant for the model wth SMKC, and both the KAOPEN varable and the nteractve term are statstcally negatve and postve, respectvely, for the models wth SMKC and SMTV. The same exercse s then repeated, but ncreasng the range of outler excluson by droppng the observatons f ther annual growth rates are larger than one and a half standard devatons away from the mean n both drectons. Ths excluson shrnks the sample sze qute substantally (sometmes as much as 40%), although the results are largely unchanged. The ft and the sgnfcance of the coeffcents even mproves for the models for the LDC/EMG wth LEGAL2 and ACCOUNT. Two conclusons flow from ths exercse. Frst, the key fndngs of the analyss are not drven by outlers. Second, some of the fndngs related to the legal/nsttutonal varables could have even been obscured by the effects of outlers. 2.5 Reverse Causalty? One may reasonably ask f fnancal development s what allows countres to mplement fnancal lberalzaton polcy, rather than the reverse. Whle we have worked wth non-overlappng, fve-year wndow panels n order to mtgate problems assocated wth smultanety, t may stll be worthwhle to nvestgate whether countres need to develop ther fnancal systems before undertakng captal account lberalzaton. Conversely, f we can show that reverse causalty s rrelevant, that wll be evdence that countres can develop ther fnancal markets by exogenously decdng to open ther fnancal markets. 28 Snce we are dealng wth a set of non-overlappng fve year panels, n essence the only data for 1980, 1985, 1990, 1995, and 2000 are affected by the removal of outlers. 29 Excluson of the outlers takes place more for the models wth stock market total value and stock market turnover than those wth prvate credt creaton and stock market captalzaton, whch reflects that the former group are more subject to market volatlty than the latter. The rate of excluson s generally hgher for the subgroup of emergng 15

18 To explore the above queston, we change the specfcaton n equaton (1) by exchangng the places of KAOPEN t-5 and fnancal development measures (PCGDP, SMKC, SMTV, and SMTO); the left-hand sde varable s now the fve-year average growth n KAOPEN whle the ndependent varables of our focus now becomes the fnancal development varable and the nteractve terms between the legal/nsttutonal varables and the fnancal development varables. We run regressons specfed as follows, usng non-overlappng data and ncludng each of the legal/nsttutonal varables: (2) KAOPEN t KAOPEN t 5 = ( L FD ) φ 0 + ϕkaopent 5 + φ1fdt 5 + φ2l + φ3 t 5 + X t 5Φ + ν t. The coeffcent of our nterest s φ 1; A sgnfcantly postve φ 1 would ndcate that the above OLS regresson results ental smultaneous causalty,.e., fnancal development leads to fnancal openness. The regresson results (not reported) show that across the regressons wth dfferent fnancal development measures as well as legal/nsttutonal varables, the coeffcents, φ 1 s, are mostly statstcally nsgnfcant or sgnfcantly negatve, ether of whch s aganst the null hypothess that fnancal development leads to fnancal openness. The sgnfcantly negatve φ 1 s are found n the regressons that have the dependent varable of the equty market development measures. One of the possble explanatons for the negatve φ 1 may be that a rapd growth n equty markets, n terms of the sze of the markets (SMKC) or the lqudty of the markets (SMTV or SMTO) s sometmes assocated wth fnancal crses, and that polcy makers lower the degree of fnancal openness durng perods of crss (Ito, 2004). At the very least, we can conclude that our estmaton results are not subject to obvous smultanety ssues (as n Bekaert, et al.,2001). 3. Investgatng the Sequence of Lberalzaton 3.1 Trade and Captal Account Lberalzaton One suggestve result we must dscuss s the frequent fndng of statstcally sgnfcant and postve coeffcents on the trade openness varable n the above estmaton results based on equaton (2). 30 To the extent that our trade openness varable reflects trade lberalzaton, our results are germane to the queston f trade openness s a precondton for fnancal openng. Ths s market countres for the same reason. 16

19 the optmal sequence queston, whch has been rased by many, ncludng McKnnon (1991). In fact, lberalzaton n goods markets s often clamed to be a precondton for fnancal lberalzaton polcy (Tornell et al., 2004). We emprcally explore ths hypothess by employng a smple model that accounts for the determnants of fnancal openness or restrctons. As stated n equaton (3), we model fnancal openness s the functon of government budget surplus, nternatonal reserves, trade openness, and per capta GDP. 31 (3) KAOPENt = ξ + ξ1kaopent 5 + ξ2tradeopent 5 + Zt 1 t 5Ξ + ν t 0, where Z t-1 t-5 s a vector of macroeconomc control varables, namely, government budget surplus, nternatonal reserves, and per capta GDP. 32 The ratonale for usng these macro varables follows the past lterature. Grll and Mles-Ferrett (1995) showed that a hgher level of restrctons on captal flows s emprcally lnked to hgher rates of nflaton, a hgher share of segnorage n total taxes, lower real nterest rates, and a hgher share of government consumpton n GDP. Ther fndng mples that captal controls appear to have strong fscal mplcatons,.e., countres wth a less developed tax system tend to mplement captal controls as the source of government revenue as well as the remedy to captal flows caused by the nflaton-drven dstortons n the fnancal markets. In nvestgatng the emprcal determnants of captal controls, Johnston and Tamrsa (1998) fnd that countres tend to mplement captal controls, the more prevalent the balance of payments concerns are, the hgher real nterest rates and real exchange rates, 33 and the larger the sze of the government defct as a share of GDP. We select two varables the government budget 30 Agan, due to space consderatons, the regresson results are not reported. They are avalable upon request. 31 The emprcal model also controls for regonal dfferences by regonal dummes. In Glck, Guo and Hutchson (2004), an emprcal model of captal account lberalzaton s proposed. Our set of explanatory varables overlaps, but does not match, thers. In part the dfference arses from ther more emprcally orented motvaton for model selecton. 32 The varable for gross nternatonal reserves s a proxy to the balance of payments stuaton of the countres and s measured by gross nternatonal reserves n months of mports. The lower gross reserves n months of mports, the hgher prevalence of balance of payments concerns are. The data are extracted from the World Bank s World Development Indcators. Others have used nternatonal reserves as a rato to M2 as a proxy to the balance of payments stuaton. The regresson results shown n ths paper are qualtatvely unchanged f nternatonal reserves as a rato to M2 are used. 33 Ther theoretcal predcton, whch contrasts wth that of Grll and Mles-Ferrett, s that countres use captal controls to pursue nconsstent nternal and external balances smultaneously such as the case where outflow controls are mplemented to avod nomnal currency deprecaton pressures wthout tghtenng of monetary condtons. When such a threat of currency crss arses, the real nterest rates or real exchange rates tends to be hgher. See Leblang (1997) for more dscussons on the determnants of captal controls. 17

20 surplus and nternatonal reserves snce they are most commonly argued as the determnants of captal controls. 34 We also nclude per capta GDP to control for the level of development of the economc system. Snce these varables are supposed to control for the general trend of macro varables, they are ncluded as the 5-year average pror to the tme perod t (as shown as (t-1 t-5) n the regresson results table). Wth these macro varables, we test whether trade openness s a precondton for fnancal openng by ncludng a fve-year lagged varable for the trade openness measure. Although we used n the prevous analyses, opn, the sum of exports and mports dvded by GDP, as the trade openness varable, we use a dfferent varable to measure openness n trade flows. By constructon, the varable opn measures the openness n good transactons n terms of not only economc factors but also regulatory measures such as tarffs and quotas. As such, we use another varable TRADEOPEN whch s a recprocal of the dutes mposed on both mports and exports. 35 In order to mnmze the possblty of two-way causalty, we employ a non-overlappng panel data analyss as we dd n prevous analyses. Whle the macro varables are ncluded as the fve-year average, both KAOPEN and TRADEOPEN are ncluded as the ntal condtons of each fve-year panel. In the emprcal analyss, we focus on the coeffcent ξ 2 to see f the openness n goods trade can be a precondton for fnancal openng. The frst three columns of Table 7 report the regressons results. Whle the average budget surplus and GDP per capta enter sgnfcantly, but not nternatonal reserves, the trade openness varable seems to contrbute to the level of fnancal openness n the full sample and the subsample for the less developed countres, ndcatng that the openness n goods transactons s a precondton for fnancal openness. Columns (4) through (6) of Table 7 show the results when we swtch the places for KAOPEN and TRADEOPEN n the regresson, n order to see whether the reverse causalty also holds. 36 We can see that n both the full sample and the subsamples, the 34 In addton to the above varables, we attempted to nclude varables that refer to the government engagement n segnorage, such as the nflaton rate or the reserve rato. However, these measures of segnorage can cause multcollnearty n regresson analyss due to ts correlaton wth the level of government budget surplus, the reverse of whch s often the reason for segnorage. Therefore, we decded not to nclude segnorage-related varables n our regresson model. 35 Import and export dutes as a rato to mports and exports, respectvely, are avalable from the World Bank s WDI. The weghted average of these varables s calculated usng the share of mports and exports n total trade and then subtracted from 100 to show the openness. Hence, the hgher (or close to 100) TRADEOPEN s, the less dutes mposed on trade flows n both drectons,.e., the more open trade flows are by regulaton. Note the varable s ncluded n log form. 36 The varable for nternatonal reserves s not retaned because t lacks a theoretcal motvaton. 18