NBER WORKING PAPER SERIES INTERNATIONAL TAXATION AND CROSS-BORDER BANKING. Harry Huizinga Johannes Voget Wolf Wagner

Transcription

1 NBER WORKING PAPER SERIES INTERNATIONAL TAXATION AND CROSS-BORDER BANKING Harry Huznga Johannes Voget Wolf Wagner Workng Paper NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 Massachusetts Avenue Cambrdge, MA October 2012 We thank Ben Lockwood and Nadne Redel for comments on an earler draft and Ata Can Bertay for research assstance. The vews expressed heren are those of the authors and do not necessarly reflect the vews of the Natonal Bureau of Economc Research. NBER workng papers are crculated for dscusson and comment purposes. They have not been peerrevewed or been subject to the revew by the NBER Board of Drectors that accompanes offcal NBER publcatons by Harry Huznga, Johannes Voget, and Wolf Wagner. 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 Internatonal Taxaton and Cross-Border Bankng Harry Huznga, Johannes Voget, and Wolf Wagner NBER Workng Paper No October 2012 JEL No. F23,G21,H25 ABSTRACT Ths paper examnes emprcally how nternatonal taxaton affects the volume and prcng of cross-border bankng actvtes for a sample of banks n 38 countres over the perod. Internatonal double taxaton of foregn-source bank ncome s found to reduce bankng-sector FDI. Furthermore, such taxaton s almost fully passed on nto hgher nterest margns charged abroad. These results mply that nternatonal double taxaton dstorts the actvtes of nternatonal banks, and that the ncdence of nternatonal double taxaton of banks s on bank customers n the foregn subsdary country. Our analyss nforms the debate about addtonal taxaton of the fnancal sector that has emerged n the wake of the recent fnancal crss. Harry Huznga Tlburg Unversty 5000 LE Tlburg The Netherlands H.P.Huznga@uvt.nl Wolf Wagner Tlburg Unversty 5000 LE Tlburg The Netherlands wagner@uvt.nl Johannes Voget Unversty of Mannhem Mannhem Germany voget@un-mannhem.de

3 1. Introducton The nternatonal tax system tends to dscrmnate aganst foregn-owned banks. Specfcally, domestc banks are just subject to a local corporate ncome tax on domestc ncome, whle foregn-owned banks n addton may be subject to non-resdent dvdend wthholdng tax n the subsdary country and corporate ncome tax on repatrated dvdends n the parent country. Internatonal double taxaton potentally puts nternatonal banks at a compettve dsadvantage, wth mplcatons for the performance as well as the structure of the nternatonal bankng market. 2 Ths paper examnes emprcally the mpact of nternatonal taxaton on bank nterest margns and pre-tax proftablty as ndces of bankng-sector performance. Furthermore, we nvestgate how nternatonal taxaton affects bankng FDI n terms of foregn-bank assets and numbers, as measures of bankng-sector structure. Our study of the nternatonal taxaton of bankng offers nsghts that are nterestng from two man perspectves. Frst, n the aftermath of the fnancal crss of , many countres are thnkng of new taxes on ther fnancal systems to help prevent a next crss and also to rase the overall tax contrbuton of the fnancal sector. A man new tax beng consdered s the Fnancal Actvtes Tax, whch s a tax on a bank s combned profts and wage bll (see IMF, 2010). In a recent communcaton, the European Commsson (2010) has announced that t s conductng an mpact assessment study of the Fnancal Actvtes Tax (among other fnancal taxes), whch s potentally followed by a proposal for a European drectve to coordnate such taxaton n the EU. Our analyss of nternatonal ncome taxaton as appled to the bankng sector nforms about the lkely ncdence and dslocaton effects of a Fnancal Actvtes Tax, gven that the latter tax also s a tax on ncome derved from the fnancal sector. More drectly, we gan nsght nto the mpact of the corporate ncome tax when some frms n a country are subject to a dfferentally hgh level of tax due to nternatonal double taxaton. A second reason for studyng the nternatonal double taxaton of bankng s that t consttutes a barrer to further bankng market ntegraton. Regulatory barrers to nternatonal bankng have been reduced worldwde, but the drve at bankng market unfcaton has so far stopped short of elmnatng the nternatonal double taxaton of bankng ncome. Ths 2 Internatonal expanson only makes sense for a bank f ths provdes benefts that exceed the cost of nternatonal double taxaton. These benefts potentally nclude beng able to serve nternatonally actve customers, dversfcaton gans, economes of scale, access to agglomeraton benefts n nternatonal fnancal centers, and nternatonal regulatory arbtrage. See McCauley, McGure and von Goetz (2010), Claessens and Van Horen (2009), and Commttee on the Global Fnancal System (2010). 2

4 potentally explans why many countres bankng markets reman domnated by natonal banks, even f many bankng markets have become more nternatonal, as measured by the external assets and labltes of domestc banks as well as the ownershp of banks. 3 Our emprcal analyss of the mpact of nternatonal taxaton on banks s based on a sample of ndvdual banks n 38 countres durng the perod. We estmate that bank nterest margns almost fully reflect the addtonal nternatonal taxaton of dvdends pad by foregn subsdares. The ncdence of nternatonal taxaton thus appears to be on a bank s lendng and depostor customers. Ths result s robust to lmtng the sample to ntra-eu bankng, to lmtng the sample to banks that are foregn subsdares, and to adjustng the nterest margn for a bank s loan loss provsonng as a measure of credt rsk. We do not fnd that a bank s pre-tax proftablty s materally affected by nternatonal double taxaton of dvdend ncome. Ths may reflect that hgher nternatonal taxaton of a foregn subsdary s ncome trggers more outward proft shftng to the parent bank (and hgher tax-deductble costs to hde ths proft shftng). Consstent wth a proft shftng motve, we further fnd that a foregn subsdary s reported profts are postvely related to the corporate ncome tax n the parent country. Usng a gravty model approach, we nvestgate the mpact of nternatonal double taxaton on bankng FDI on a blateral aggregated bass. We fnd evdence that nternatonal double taxaton of dvdend ncome reduces bankng-sector FDI n terms of foregn-bank assets. In addton, we fnd that the number of foregn banks s sgnfcantly reduced by nternatonal double taxaton. The responsveness of the number of nternatonal bankng establshments to nternatonal double taxaton suggests that banks face a two-step nternatonal bankng decson: frst, they consder whether to set up a foregn subsdary and, second, they determne the prcng (and quantty) of ther foregn fnancal servces. The endogenety of the ntal FDI decson w.r.t. nternatonal double taxaton mples that the estmaton of the mpact of such taxaton on net nterest margns may be based. We apply a two-stage Heckman estmaton to net nterest revenue regressons to account for the possble endogenety of the FDI decson, confrmng a major pass through of nternatonal double taxaton nto hgher nterest margns. Taken together, 3 See Allen, Beck, Carlett, Lane, Schoenmaker, and Wagner (2011, Table 1.1) for nformaton on the development of the external assets and labltes of domestc bankng systems relatve to GDP of BIS reportng countres over the perod. 3

5 our results ndcate that nternatonal double taxaton ncreases nterest margns, as t causes foregn-owned frms to reduce ther host-country supply of fnancal servces by way of FDI. Several papers have prevously examned the tax and non-tax determnants of bank nterest margns and proftablty. For a sample of banks n 80 countres over the years , Demrgüç-Kunt and Huznga (1999) fnd that nterest margns and pre-tax proftablty are negatvely related to offcal reserves, whch represent a form of mplct taxaton. Interest margns and proftablty are further postvely related to the local corporate tax rate. The estmated coeffcent on the corporate tax rate n the proftablty regresson s consstent wth a full pass-through of the corporate tax to bank customers. Demrgüç-Kunt and Huznga (2001) extend ths analyss to dstngush between domestcally owned and foregn-owned banks. The proftablty of foregn-owned banks s found to rse relatvely lttle wth the local corporate tax rate, whch can be explaned by nternatonal proft shftng or by the nternatonal double tax relef provded by parent countres. The present paper goes beyond Demrgüç-Kunt and Huznga (2001) by ncludng n the analyss both host and parent country taxaton payable by foregnowned banks, thus accountng for nternatonal double taxaton. 4 An extensve lterature, surveyed by Ederveen and de Mooj (2006), examnes the mpact of taxaton on FDI. Several authors have prevously found a role for parent-country taxaton to affect the locaton of FDI. For US multnatonals, Kemsley (1998) fnds that the host country tax only affects the rato of US exports to foregn producton over the perod f the multnatonals fnd themselves n excess credt postons. Analogously, a role of parent-country taxaton n affectng FDI nto the Unted States s found by Hnes (1996) who shows that foregn countres wth worldwde taxaton nvest relatvely much n US states wth hgh state taxes. Ths reflects that multnatonals located n countres wth worldwde taxaton may be able to obtan foregn tax credts for US state corporate ncome taxes. Egger, Loretz, Pfaffermayr, and Wnner (2009) construct an effectve tax rate on a blateral bass that reflects overall host and parent country taxaton, and they fnd that ths blateral effectve tax rate has a negatve mpact on 4 Demrgüç-Kunt, Laeven and Levne (2003) examne the mpact of bank regulatons, market structure and natonal nsttutons on the bank net nterest margns for a sample of banks from 72 countres over the years whle not consderng taxaton. Martnez Pera and Mody (2004) examne how foregn bank partcpaton affects nterest margns of Latn Amercan banks durng the perod , dstngushng between ndvdual-bank and bankng-system foregn ownershp. Maudos and Guevara (2003) examne the mpact of bank market power on nterest spreads n sx large European bankng markets n the perod Valverde and Fernandez (2007) examne the mpact of a bank s actvty mx on bank margns n Europe. 4

6 blateral FDI stocks after controllng for host and parent country unlateral effectve tax rates. Barros, Huznga, Laeven and Ncodème (2012) examne how nternatonal double taxaton affects foregn subsdary locaton, fndng that parent country corporate ncome taxaton dscourages subsdary locaton. Huznga and Voget (2009) fnd a negatve mpact of nternatonal double taxaton on headquarter locaton followng nternatonal M&As usng ndvdual deal as well as aggregated data. The present paper examnes the mpact of nternatonal double taxaton on FDI n the bankng sector only, usng nformaton for all banks rather than just for those that are newly formed through M&As. Focusng on the bankng sector has the advantage that bank-level data allow us to dentfy a prce response (through nterest margns) and a quantty response (through FDI) to nternatonal double taxaton. Ths paper s organzed as follows. Secton 2 descrbes the nternatonal tax system, and t provdes some summary nformaton on the nternatonal tax rates that apply to our sample of banks. Secton 3 presents the emprcal results on the mpact of nternatonal taxaton on bank nterest margns and proftablty. Secton 4 n turn presents results on how nternatonal taxaton affects bankng sector FDI. Ths secton also examnes whether the results on nterest margns and bank proftablty are robust to controllng for the potental endogenety of FDI. Secton 5 concludes. 2. The nternatonal taxaton of banks 2.1 The nternatonal tax system In ths secton, we descrbe the nternatonal tax system that apples to a bank owned by some foregn parent bank. We consder the addtonal nternatonal taxaton that s leved on the subsdary s dvdend and also ts nterest payments to outsde nvestors on the assumpton that these payments are frst made to the parent frm whch then passes them on to fnal nvestors. Thus, we wll assume that the parent bank pays out any dvdends receved from the foregn subsdary as dvdends to nvestors, whle any nterest receved s pad out as nterest. We examne the nternatonal tax system as t apples to dvdend and nterest payments n turn. 5 A bank s ncome s subject to the local corporate ncome tax before t can be pad out as dvdends. For a domestc bank located n country, the corporate ncome tax t s the only tax on 5 See Huznga, Laeven and Ncodème (2008) for an alternatve descrpton of the nternatonal tax treatment of the debt and equty fnance of a multnatonal frm. 5

7 ncome pad out as dvdends at the corporate level. Table 1 ndcates the statutory corporate tax rate on corporate proft n 2008 for the 38 countres n ths study, whch n addton to many European countres ncludes Australa, Canada, Japan, Mexco, New Zealand, South Korea and the Unted States. 6 Dvdends pad out by a foregn subsdary located n country can be subject to a nonresdent dvdend wthholdng tax e w leved by the subsdary country. Blateral dvdend wthholdng taxes for our sample of countres n 2008 are presented n Table 2. Among longstandng EU member states, nonresdent dvdend wthholdng taxes for payments to parent frms are zero on account of the EU Parent-Subsdary Drectve. Non-EU countres such as Canada, Japan, New Zealand, and the Unted States mantan non-zero dvdend wthholdng taxes n a consderable number of cases. The parent country may or may not tax any ncome generated abroad. In case the parent country operates a terrtoral or source-based tax system, t effectvely exempts foregn-source ncome from taxaton. The effectve tax on ncome generated n country and pad out as e dvdends n country p then s t + w 1 t ), and the addtonal tax on account of foregn ( e ownershp, denoted τ, equals w 1 t ). ( Alternatvely, the parent country operates a worldwde or resdence-based tax system. In ths nstance, the parent country subjects ncome reported n country to taxaton, but t generally provdes a foregn tax credt for taxes already pad n country to reduce the potental for double taxaton. The OECD model treaty, whch summarzes recommended practce, gves countres the choce between an exempton and a foregn tax credt as the only two ways to releve double taxaton (OECD, 1997). The foregn tax credt reduces domestc taxes on foregn source ncome one-for-one wth the taxes already pad abroad. The foregn tax credt can be ndrect n the sense that t apples to both any wthholdng tax and the underlyng subsdary-country corporate ncome tax, or t s drect and apples only to the wthholdng tax. In ether case, foregn tax credts are generally lmted to prevent the domestc tax lablty on foregn source ncome from becomng negatve. In the ndrect credt regme, an nternatonal bank wll pay no corporate ncome tax n e the parent country, f the parent tax rate t p s less than t w 1 t ). The nternatonal bank then ( 6 The sample s restrcted to OECD countres and countres of the European Economc Area due to data avalablty. 6

8 has unused foregn tax credts and s sad to be n an excess credt poston. Alternatvely, t p e e exceeds t w t w. In that nstance, the bank pays tax n the parent country at a rate equal to e e the dfference between t p and t w t w. The effectve, combned tax rate on the dvdend ncome then equals the parent country tax rate, t p. To summarze, wth the ndrect credt system, the effectve rate on ncome generated n country, s gven by max [ t p, t w t w ], and the e addtonal tax on account of foregn ownershp τ, equals max [ t t w (1 t ) ]. Wth a drect p e, foregn tax credt, the nternatonal bank pays no corporate ncome tax n the parent country, f e the parent tax rate t p s less than w. In the more common case where t p exceeds e e w, the bank e nstead pays tax n the parent country at a rate equal to (1 t )( t w ). The effectve, two- e country tax rate now s gven by t (1 t )max[ t, w ], and the addtonal nternatonal tax τ, e equals (1 t )max[ t, w ]. A few countres wth worldwde taxaton do not provde foregn tax p credts, but nstead allow foregn taxes to be deducted from the multnatonal s taxable ncome. e In the scenaro, the effectve rate of taxaton on dvdends s gven by (1 t )(1 w )(1 t ), e and equals ( 1 t )[1 (1 w )(1 t p )]. p p 1 p Columns 2-4 of Table 1 provde nformaton on the double taxaton rules appled to ncomng dvdends n Several countres are seen to dscrmnate between nternatonal tax treaty partners and non-treaty countres. We have collected nformaton on the exstence of blateral tax treates to assess the relevant double tax relef method. Also, several EU countres are seen to offer relatvely generous double tax relef for ntra-eu dvdends. Next, we consder the addtonal nternatonal taxaton that may apply to nterest payments by a foregn subsdary bank that reach fnal nvestors va an nternatonal parent bank. Interest expense on debt s generally deductble from taxable corporate ncome n the subsdary country, but the subsdary country may levy a non-resdent wthholdng tax d w on nterest payments to the parent bank n country p. As seen n Table 3, blateral nonresdent wthholdng taxes on nterest on nterest payments to related partes tend to be zero n the EU on account of the Interest and Royaltes Drectve, even f non-eu countres such as Canada, Japan and the Unted States frequently levy postve nonresdent nterest wthholdng taxes. 7

9 The parent country generally apples corporate ncome tax to the parent bank s nterest recept from ts foregn subsdary. As before, the parent country has three man optons regardng double tax relef: () an exempton, () a foregn tax credt, or () a deducton. For each of these three cases, an addtonal nternatonal tax rate on nterest on account of the subsdary s foregn ownershp can be derved, and formulae are presented n Table 4. Columns 5 and 6 of Table 1 provde nformaton on the double taxaton rules applcable to ncomng nterest from treaty and non-treaty sgnatory countres, respectvely. As seen n the table, most countres provde a foregn tax credt (to be appled to any nonresdent nterest wthholdng tax), a few countres allow a deducton n the absence of a tax treaty, and no country exempts foregn nterest ncome. 2.2 Internatonal taxaton of banks n the sample Data on ndvdual banks are taken from Bankscope. Ths data source provdes accountng data on banks worldwde n a standardzed format. In addton, Bankscope contans data on ownershp relatonshps among banks. For each bank, Bankscope provdes nformaton on major owners (and also nformaton on any owned subsdares). Our am s to have a sample of all the bankng establshments that operate n a country, and for each establshment provde nformaton on majorty foregn ownershp, f any. To construct a comprehensve sample of the banks n a country, we nclude unconsoldated parent frms and all subsdares. The ownershp nformaton provded for subsdares s then used to see f there s a corporate major shareholder and to fnd out where such a major shareholder has ts resdence. Our country coverage s lmted to the countres for whch we have collected tax nformaton as lsted n Table 1. Thus, we nclude banks that are located n one of these countres and that have majorty owners resdent n one of these countres. Our sample covers the years Table 5 provdes a breakdown of our sample of banks by the country of locaton. Banks located n the US comprse 46% of the sample, or 4462 US banks n an overall sample of Other countres wth at least 400 observatons are France, Germany, Italy, Luxembourg and Swtzerland. Table 5 also provdes nformaton on the share of assets held by foregn-owned banks. The foregn-bank asset share s on average 9.5% nternatonally. The foregn ownershp share by assets s very hgh n the Baltc states (96.5% n Estona, 49.7% n Latva, and 80.0 n Lthuana) and also n Luxembourg (67.2%), whle t s lowest n the US at 1.2%. 8

10 Table 6 provdes nformaton on the local and nternatonal tax burdens on banks by country of resdence. The host country corporate ncome tax on average s 36.1% for all banks. The average dvdend double tax, correspondng to the expressons n Table 4, s calculated as 0.8% for all banks. The average nterest double tax s further shown to be postve only for some banks located n Cyprus and Swtzerland. In the emprcal work below, we wll only consder the nternatonal double taxaton of dvdend ncome, gven the dearth of observatons where the nternatonal double taxaton of nterest ncome s postve. The fnal two columns of Table 6 provde nformaton on the average nternatonal taxes for only the sample of foregn-owned banks. For these banks, the average dvdend double tax amounts to 3.5%, to suggest that foregn-owned banks on average face a 10% hgher tax than domestc banks that are only subject the local corporate tax rate. 3. Bank nterest margns and proftablty In ths secton, we examne how the nternatonal taxaton of banks affects bank nterest margns and bank proftablty. A bank s pre-tax profts are defned by the followng accountng dentty Pre-tax profts = Net nterest ncome + Net other operatng ncome Loan loss provsons Overhead. In the emprcal work, we wll use nterest ncome and proftablty measures scaled by total bank assets. Thus, net nterest ncome over assets s a bank s net nterest ncome dvded by total assets. Net nterest ncome over assets has a sample mean of 2.8%, as seen n Table 7. Smlarly, Pre-tax profts over assets s the rato of a bank s pre-tax profts to total assets. Ths proft varable reflects varaton n all the varous tems n a bank s ncome statement, ncludng ts net nterest ncome. The mean of ths varable s1.3%. Pre-tax profts over assets can be splt nto Taxes over assets and Post-tax profts over assets wth mean values of 0.4% and 1.0%, respectvely. The Taxes over assets varable reflects the taxes pad by the reportng bank, and hence they exclude any corporate taxes to be pad by an nternatonal parent bank and any nonresdent dvdend wthholdng taxes. These latter taxes are to be pad out of the dvdends dstrbuted by the bank. 7 7 The Taxes over assets varable smlarly excludes any nonresdent nterest wthholdng taxes that are to be pad out of nterest pad by the bank and receved by nonresdents. 9

11 The margn and proftablty varables wll be explaned by several tax rate varables n the emprcal work. Among these, the local or host country corporate ncome tax has a mean of 35.2% for the observatons n our sample. Next, the parent country tax s the corporate ncome tax rate n the parent country n case a bank s foregn-owned. Ths varable s set to zero n case of domestc ownershp. The mean of the parent country tax varable s seen to be 7.4%. Internatonal double taxaton of dvdend ncome has a mean of 0.8%. The mpact of bank taxes on bank net nterest revenue and proftablty reflects the extent to whch these taxes are shfted onto bank customers and other related partes through dfferent prce settng. In practce, banks may be able to shft some of ther taxes to bank retal customers, other bank lablty holders, bank employees and further provders of bankng nputs. For nstance, a bank could shft some of ts taxes to ts retal customers n the form of a hgher lendng rate and a lower depost rate, gvng rse to hgher net nterest revenues and hgher pretax proftablty. Bankng taxaton may also affect the recorded net nterest revenue and proftablty as a result of nternatonal proft shftng wthn a multnatonal bank. Hgher host country taxaton and dvdend double taxaton, n partcular, provde ncreased ncentves to shft profts to an nternatonal parent bank, mplyng lower recorded proftablty of a foregn subsdary bank. Internatonal proft shftng thus may lead to a less postve or even negatve relatonshp between the taxaton of subsdary profts and recorded subsdary profts. At the same tme, recorded subsdary profts may be postvely related to the parent country corporate tax rate, f hgher parent country taxaton causes a multnatonal frm to shft profts from the parent bank towards ts foregn subsdares. Several bank-level and country-level varables are ncluded n the analyss as controls. Assets s the log of total bank assets n real terms to control for bank sze. The rato of earnng assets to total bank assets s the share of a bank s assets that generates nterest or dvdend ncome, and t proxes for a bank s focus on nterest-generatng actvtes as opposed to feegeneratng actvtes. Foregn bank sgnals ownershp by foregn shareholders wth at least 50% ownershp. Foregn ownershp potentally affects net nterest revenue and proftablty on account of dfferent nterest margns and proftablty n an economc sense as well as on account of nternatonal proft shftng. Bank market share s a bank s total loans as a share of all loans provded by banks located n a certan country. A hgh bank-level market share may gve rse to 10

12 market power, leadng to hgher net nterest revenue and proftablty. Alternatvely, a hgh market share could reflect bank effcency, resultng n low nterest margns to the extent that bank customers reap the benefts of hgher bank effcency. Among the country-level controls, the natonal foregn ownershp share s the share of assets of foregn-owned banks n total bankng system assets. A hgh share of foregn ownershp natonally suggests free entry of foregn banks, possbly reducng nterest margns and proftablty. Natonal top fve market share s the share of loans of the top fve lendng banks n total loans provded n a country. A hghly concentrated lendng market, as ndcated by a hgh top 5 lendng share, may explan hgh nterest margns and proftablty. GDP per capta s the log of real GDP per capta. Industral growth rate s the growth rate of ndustral producton. Strong ndustral growth may mply hgh loan demand, pushng up net nterest revenue and proftablty. Inflaton rate s the rate of change n the consumer prce ndex. Hgh nflaton may ncrease net nterest revenue, f lendng rates more accurately reflect nflaton than depost rates. Fnally, real nterest rate s the money market nterest rate mnus the nflaton rate. Hgh real nterest rates may reflect plentful opportuntes to nvest proftably, pushng up nterest margns and bank proftablty. Table 8 shows the results of regressons of the net nterest ncome over assets varable. The regressons nclude host country and year fxed effects, and standard errors are robust to clusterng at the bank level. In regresson 1, the host country tax obtans a negatve coeffcent of that s statstcally nsgnfcant. 8 The falure of the host country corporate tax to lead to hgher nterest margns could reflect that hgher host country taxaton nduces outward proft shftng. Alternatvely, t reflects that the ncdence of the corporate ncome n open economes s largely on labor, gvng rse to lower wages. Consstent wth ths, Arulampalam, Devereux, and Maffn (2012) estmate that an exogenous rse n the corporate tax of 1$ would reduce the wage bll by 49 cents. In regresson 1, the double dvdend tax obtans a coeffcent of that s sgnfcant at the 5% level. Thus, some of the ncdence of nternatonal double taxaton appears to be on the foregn subsdary s lendng and depostor customers and other supplers of funds. Internatonal 8 The coeffcent for the tax rate reflects the long-run effect on the net-nterest margn. An explct modelng of short-run dynamcs would requre longer tme seres. See also Verbeek (2008, p ). 11

13 double taxaton, unlke host country taxaton, s not prmarly shfted to labor, as double taxaton only apples to a specfc set of foregn bank owners rather than to banks generally. In regresson 1 the assets varable obtans a coeffcent of that s sgnfcant at the 1% level. Ths may reflect that bg banks deal wth large customers that obtan favorable nterest rates. The bank-level foregn ownershp dummy enters wth a negatve coeffcent of that s sgnfcant at 5%. Ths could equally reflect that foregn-owned banks tend to deal wth sophstcated customers, or alternatvely that they have to offer more attractve nterest terms to ther customers on account of lack of nformaton or dstrust. Net nterest ncome over assets s further postvely and sgnfcantly related to the bank s own market share, as a large loan market share may enable t to exercse market power n the loan market. Among the macroeconomc varables, the net nterest ncome relatve to assets s postvely and sgnfcantly related to the growth rate of ndustral producton and to the rate of nflaton. The estmated coeffcent of for the double tax varable n regresson 1 mples a certan sharng of the ncdence of addtonal nternatonal taxaton between the bank and ts customers. To evaluate ths, let n be the net-of-tax net nterest margn calculated as (1-t)b where t s the combned natonal and nternatonal tax rate on net nterest ncome and b s the pre-tax nterest margn. Tax revenue s denoted r and s equal to tb. A change n the combned tax rate t changes the bank return by dn/dt = -b + (1-t)*db/dt, whle the change n revenues s gven by dr/dt=b + t*db/dt. The share of the ncdence on the bank s computed as ( dn/dt)/(dr/dt), whle the share of the ncdence on bank customers equals (db/dt)/(dr/dt). To do the calculaton, we set b to ts sample mean of 0.028, t s the combned summed mean dvdend double tax and mean host country tax of 0.36 (the sum of and 0.008), and db/dt s the estmated coeffcent of The share of the ncdence of a hgher nternatonal double tax on the bank s now calculated to be 13.8%, wth the remanng share of 86.2% of the ncdence beng borne by the bank s loan customers and depostors. Member states of the EU do not mpose dscrmnatory restrctons on ntra-eu foregn bankng and they subscrbe to a common set of basc mnmum standards of bank regulaton n areas such as captal adequacy and depost nsurance. 9 In the EU, however, there stll s some nternatonal double taxaton of dvdend ncome, as several EU member states contnue to tax the worldwde ncome of ther resdent multnatonal banks. A sample just of EU banks (located 9 The European Unon s Second Bankng Drectve of 1989 allows EU banks to freely operate throughout the EU. 12

14 n the EU and wth EU parent frms, f any) provdes an nterestng settng to test for the mpact of nternatonal taxaton on nterest margns and proftablty, gven that EU bankng markets are otherwse relatvely unform. In regresson 2, we restrct the sample to only EU banks. The host country tax enters ths regresson wth negatve coeffcent that s nsgnfcant, whle the double tax varable obtans a coeffcent of that s sgnfcant at the 5% level. Next, n regresson 3 we restrct the sample to foregn owned banks, reducng the sample from 9731 observatons n regresson 1 to 2135 observatons. The coeffcent for the host country tax s estmated to be nsgnfcant, whle the coeffcent of for the double tax varable s sgnfcant at the 5% level. Regresson 4 ncludes bank fxed effects rather than country fxed effects to account for possble unobserved bank heterogenety. The host country tax now receves a postve coeffcent that s nsgnfcant, whle the double tax varable s estmated wth a coeffcent of that s sgnfcant at 5%. Fnally, n regresson 5 the dependent varable s the nterest margn varable adjusted for concurrent loan loss provsonng, to reflect dfferences across banks n the rskness of ther credt portfolos. The host country tax now enters wth a negatve coeffcent of that s sgnfcant at 5%. A negatve coeffcent s consstent wth a proft shftng motve. Banks, n partcular, have an ncentve to transfer questonable credts to hgh-tax countres before any loan loss provsons are taken to beneft from the tax deductblty of such provsonng (or of the mpled subsequent wrte-offs) at the hgher tax rate. Table 9 presents results of regressons of bank proftablty and of the level of host corporate ncome taxes pad. Specfcally, the dependent varables n regressons 1-3 are pre-tax profts over assets, taxes over assets, and post-tax profts over assets, respectvely. In all three regressons, the host country tax s seen to enter wth negatve coeffcents that are statstcally sgnfcant, consstent wth an nternatonal proft shftng motve. In partcular, reported pre-tax proftablty would declne wth proft shftng on account of the profts actually shftng abroad, and also on account of the varous costs the bank ncurs to mplement and hde the proft shftng. 10 The dvdend double tax does not enter any of the three regressons wth a statstcally sgnfcant coeffcent, as hgher ncome resultng from a pass through of the tax to bank 10 These varous costs are not reflected n the nterest margn varable, whch can explan a less negatve response of net nterest ncome to the host country tax n the presence of proft shftng that for the case of pre-tax proftablty. 13

15 customers and other nput provders may be offset by ncreased outward proft shftng and ts assocated costs. To further check for the presence of proft shftng, we next nclude the parent country tax varable n regressons 1-3 of Table 9, wth the results reported as regressons 4-6. A hgher parent country tax potentally leads to hgher reported proftablty at pertnent foregn subsdares, as ths provdes a parent bank wth the ncentve to shft profts to ts foregn subsdares. In regresson 4, pre-tax proftablty ncreases sgnfcantly wth the parent country tax, consstent wth a proft shftng ncentve. Correspondngly, n regresson 5 corporate taxes pad n the host country are postvely and sgnfcantly related to the parent country corporate tax, reflectng that hgher reported pre-tax profts lead to a hgher host country tax lablty. In regresson 6, post-tax proftablty s also related postvely to the parent country corporate tax, but the relatonshp s not statstcally sgnfcant. The proftablty and taxaton varables n regressons 4-6 are qualtatvely related to the host country tax and dvdend double tax varables as n regressons FDI n the bankng sector The prevous secton examned how nternatonal taxaton affects net nterest revenue and bank proftablty gven the domestc and foregn ownershp of banks. The evdence s consstent wth a sgnfcant pass-through of nternatonal taxaton nto hgher nterest margns. Wth elastc demand for fnancal servces, hgher net nterest revenues relatve to assets can only be acheved by cuttng back the volume of fnancal servces. Thus, nternatonal taxaton should have a dscernble mpact on the quantty of fnancal servces provded by foregn-owned banks. In ths secton, we estmate the mpact of nternatonal taxaton on the volume of foregn-provded fnancal servces as well on the number of nternatonal bankng establshments. In addton, we provde two-stage Heckman estmaton of some of our bank nterest margn and proftablty regressons. Ths accounts for the potental endogenety of the FDI decson to nternatonal double taxaton. Our volume varable s the aggregate assets of foregn-owned banks n a partcular country as owned by corporate enttes n another country. Aggregate assets of foregn banks on a blateral bass are expected to declne wth dvdend double taxaton, f the average bank cuts back ts actvtes n countres where dvdends are subject to double nternatonal taxaton. As an 14

16 alternatve bankng FDI varable, we also consder the number of banks n a partcular country owned by corporate enttes n another country. Internatonal taxaton may prevent the establshment of foregn ownershp relatonshps between certan pars of countres, or t may cause hghly taxed banks to sell ther foregn subsdares to brng about a more tax effcent ownershp structure, thereby reducng the number of foregn owned banks. We apply a gravty model to estmate the mpact of nternatonal taxaton on our ndces of bankng-sector FDI. Prevously, We (2000), Evenett (2003), and Buch, Klenert and Toubal (2004) have used the gravty model to explan FDI. Further, D Govann (2005) and Huznga and Voget (2009) have appled the gravty model to the volume of cross-border M&as, whle Portes and Rey (2005) have estmated a gravty model of trade n fnancal assets. The gravty model relates our measures of cross-border bankng to natonal and nternatonal tax rates and to a range of non-tax controls. Among these controls, we nclude standard gravty model varables such as the blateral dstance, contguty (a dummy varable sgnalng that two countres have a common border), and common offcal language (a dummy varable sgnalng that two countres have a common offcal language). Also ncluded are host and potental parent country GDPs whch are expected to be postvely related to blateral bankng FDI. Fnally, we nclude ndces of host and parent countres regulatory qualty, and ndces of ther use of captal controls. Inward bankng FDI may be related negatvely and postvely to host-country and parent-country regulatory qualty respectvely, f bankng FDI s drven by a need for a parent bank to be located n a country wth relatvely hgh regulatory qualty. Captal controls generally may dscourage bankng FDI. Table 11 shows summary statstcs for the varables n our bankng FDI regressons. Followng the modelng of trade flows n Santos Slva and Tenreyro (2006), Table 11 shows estmaton results of Posson regressons, where the dependent varable s ether the total assets of foregn-owned banks or the number of foregn-owned banks on a blateral aggregate bass. 11 The regressons nclude host country, parent country, and year fxed effects, and errors 11 Slva and Tenreyro (2006, p. 645) ndcate that the Posson estmator s consstent f E[y x ] = exp(x β) where y s the dependent varable and x are the ndependent varables. The correspondng regresson n ther paper relates the level of y to the natural logarthm of each element of x. Correspondngly n Table 10, the number of foregnowned bank and foregn owned assets are reported n levels, whle dstance, host GDP, and parent GDP are n logs. The estmated coeffcents for the logged rght-hand-sde varables are nterpreted as elastctes, whle the coeffcents on other varables ncludng the tax varable are nterpreted as sem-elastctes. Negatve bnomal regressons are not consdered as an alternatve to the Posson regressons because Bosquet and Boulhol (2010) pont 15

17 are clustered at the host country level. We n turn consder the overall nternatonal sample and the sample of only ntra-eu bankng relatonshps. In regresson 1 for the overall sample, the dependent varable s total assets of foregn-owned banks on a blateral bass. The dvdend double tax obtans a sgnfcantly negatve coeffcent of Ths estmated for the dvdend double tax mples that a 1 percentage pont ncrease n ths varable reduces blateral FDI by 7.2%, whch s economcally sgnfcant gven a mean dvdend double tax of 3.5% for foregnowned banks as seen n Table 6. In regresson 2 the dependent varable s the number of cross-border banks. The dvdend double tax s sgnfcantly negatve wth a coeffcent of Ths suggests that a one percentage pont ncrease n the dvdend double tax reduces the number of cross-border banks on a blateral bass by 3.3%. 12 Ths estmated coeffcent of s less negatve than the estmated coeffcent n the correspondng foregn-bank assets regresson 1. Ths suggests that a hgher dvdend double tax leads to both fewer and smaller cross-border banks. 13 Next, regressons 3 and 4 of Table 12 reproduce the frst two regressons of ths table for the ntra-eu sample. When FDI s measured n terms of cross-border bankng assets, the estmated coeffcent for the dvdend double tax n regresson 3 for the ntra-eu sample of s more negatve than the correspondng coeffcent of n regresson 1 for the wder sample. Thus, ntra-eu bankng FDI appears to be relatvely senstve to nternatonal taxaton, perhaps because EU banks from dfferent countres offer smlar servces gvng rse to hgh demand elastctes at foregn-owned banks nsde the EU. On the other hand, when FDI s measured n terms of subsdary banks abroad, then the estmated coeffcent on the dvdend double tax for the ntra-eu sample n regresson 4 of s very smlar to the estmate of n regresson 2 for the wder sample, although the coeffcent s now nsgnfcant n the smaller sample. The responsveness of the number of nternatonal bankng establshments to nternatonal double taxaton suggests that banks face a two-step nternatonal bankng decson: frst, they consder whether to set up a foregn subsdary and, second, they determne the prcng (and out that negatve bnomal regressons wth a contnuous dependent varable are scale-dependent. Instead, employng robust errors accommodates devatons from the Posson dstrbuton. 12 Consstent wth ths, Barros, Huznga, Laeven, and Ncodème (2012) report evdence that the nternatonal locaton decsons of multnatonal frms reflect nternatonal double taxaton of corporate ncome. 13 Verbeek (2008, p. 250) ponts out that a selecton bas does not arse f selecton depends upon the exogenous varables only. Hence, a sgnfcant effect of dvdend double taxes on the number of cross-border banks does not mply a selecton bas for the nterest margn regressons n the prevous secton. 16

18 quantty) of ther foregn fnancal servces. The endogenety of the ntal FDI decson w.r.t. nternatonal double taxaton mples that the estmaton of the mpact of such taxaton on net nterest margns may be based, f the errors at the frst-stage FDI stage are correlated wth the errors at the second-stage prcng stage. To conclude ths secton, we apply a two-stage Heckman estmaton to several of the net nterest revenue, proftablty and taxaton regressons from Tables 8 and 9 where the frst-stage regressons are probt specfcatons estmatng postve blateral bankng FDI correspondng to the FDI regresson 1of Table 11. Varables at the second stage are aggregated at the blateral natonal level on a yearly bass. 14 The results of the secondstage regressons are reported n Table 12. Regresson 1 re-estmates the net nterest revenue regresson 1 of Table 8, yeldng an estmated coeffcent for the double tax varable of that s sgnfcant at the 1% level, somewhat hgher than the correspondng estmate of n Table 8. Regresson 2 reproduces the net nterest revenue adjusted for loan loss provsonng regresson 5 of Table 8, gvng rse to an estmated coeffcent of for the double tax varable that s sgnfcant at 1%. Regressons 3-5 of Table 12 redo the pre-tax proftablty, taxes pad, and posttax proftablty regressons 1-3 of Table 9. Reported pre-tax and post-tax bank proftablty are negatvely and sgnfcantly related to the host country tax rate n regressons 3 and 5 of Table 12 consstent wth a proft shftng motve, but the relatonshp between taxes pad and the host country tax rate s statstcally nsgnfcant n regresson 4. Overall, our estmaton results n the net nterest revenue regressons of Tables 8 and 12 and the FDI regressons of Table 11 are consstent n that the dvdend double tax has statstcally sgnfcant effects n both settngs. Taken together, our results ndcate that the dvdend double tax ncreases margns, as t causes foregn-owned frms to reduce ther host-country supply of fnancal servces. 5. Conclusons Internatonal double taxaton s a remanng barrer to nternatonal bankng market ntegraton. As a result of such taxaton, nternatonal banks may face hgher corporate ncome taxaton than domestc banks that operate n the same bankng market. Internatonal double taxaton thus provdes for varaton n the taxaton of banks wthn countres as well as across 14 The selecton model cannot be estmated at the frm level because there s no sample of potental parent frms but only a sample of potental parent countres. 17

19 countres. In ths paper, we estmate the prcng response as reflected n nterest margns and the quantty response as reflected n bankng-sector FDI - to varaton n nternatonal double taxaton. We fnd that the nternatonal double taxaton of the dvdend ncome of nternatonal banks s almost fully reflected n hgher nterest margns. Thus, nternatonal banks appear to have enough prcng power to pass on ther nternatonal tax burden to local bank customers. As the revenue of ths tax n part accrues to the parent country treasury, the parent country corporate tax appears to be partally exported to the host country bankng market. To be able to rase prces, however, banks are shown to restrct the supply of fnancal servce n bankng markets subject to hgher nternatonal double taxaton of dvdends. Specfcally, blateral aggregate FDI n terms of foregn bank assets s shown to declne wth the nternatonal double taxaton of dvdend. The senstvty of bankng-sector FDI to nternatonal double taxaton mples that such taxaton dstorts the nternatonal bankng market. Specfcally, the nternatonal ownershp of banks subject to hgh nternatonal double taxaton s dscouraged. True ntegraton of the nternatonal bankng market requres that dscrmnatory taxaton of nternatonal banks s elmnated. Ths mples that countres elmnate nonresdent dvdend wthholdng taxes and exempt the foregn-source ncome of ther resdent multnatonal banks from domestc taxaton. In our larger data set, the average rate of nternatonal double taxaton of dvdend ncome, reflectng both nonresdent wthholdng taxaton and home country corporate ncome taxaton, amounts to a substantal 3.5%. In the EU, nonresdent dvdend taxes on ntrafrm dvdend payments have been elmnated by the Parent-Subsdary Drectve, but parent country corporate ncome taxes generally reman. Specfcally, EU countres that contnue to tax corporate ncome on a worldwde bass are Bulgara, Greece, Ireland, Poland, Portugal and Romana. The Unted Kngdom swtched to a terrtoral tax system n Worldwde, the US s a major country that contnues to tax corporate ncome on a resdence bass. Our results have mplcatons for the debate on any addtonal taxaton of the fnancal sector followng the fnancal crss of The IMF (2010) dscusses a range of optons for new tax nstruments that would ncrease the tax burden on the fnancal sector. These nclude a Fnancal Actvtes Tax, whch s a levy on a bank s combned profts and wage bll, and a Fnancal Stablty Contrbuton, whch taxes a bank s labltes net of ts nsured deposts. Our results concernng nternatonal ncome taxaton as appled to the bankng sector nform 18

20 especally about the lkely ncdence and dslocaton effects of a Fnancal Actvtes Tax gven that the latter tax also s a tax on ncome derved from the fnancal sector. Our emprcal results specfcally suggest that a Fnancal Actvtes Tax could well be largely passed on to bank customers, and lead to sgnfcant dslocaton effects of bankng actvty. Ths outcome s more lkely f a Fnancal Actvtes Tax vares wdely across countres and possbly wthn countres n case t were leved on a bank resdence-bass. Wthn-country varaton n Fnancal Actvtes Tax can be avoded f t leved on a consumpton destnaton bass, as recommended by the IMF. Whle there are clear parallels between corporate ncome taxaton and a Fnancal Actvtes Tax, t remans uncertan to whch extent our results on nternatonal corporate ncome taxaton and bankng carry over to a Fnancal Actvtes Tax. 19

21 References Allen, Frankln, Thorsten Beck, Elena Carlett, Phlp Lane, Drk Schoenmaker, and Wolf Wagner, 2012, Cross-Border Bankng n Europe: Implcatons for Fnancal Stablty and Macroeconomc Polces, CEPR polcy report. Arulampalam, Wj, Mchael Devereux, and Gorga Maffn, 2012, The drect ncdence of corporate ncome taxes on wages, European Economc Revew 56, Barros, Salvador, Harry Huznga, Luc Laeven, and Gaëtan Ncodème, 2012, Internatonal taxaton and multnatonal frm locaton decsons, forthcomng n Journal of Publc Economcs. Bosquet, Clement, and Herve Boulhol, 2010, Scale-dependence of the negatve bnomal pseudomaxmum lkelhood estmator, mmeo. Buch, Clauda M, Jorn Klenert, and Fard Toubal, 2004, The dstance puzzle: On the nterpretaton of the dstance coeffcent n gravty equatons, Economcs Letters 83, Commttee on the Global Fnancal System, 2010, Long-term ssues n nternatonal bankng, CGFS Papers No. 41. Claessens, Stjn, and Neeltje Van Horen, 2009, Beng a foregner among domestc banks: Asset or lablty?, De Nederlandsche Bank Workng Paper 224. Demrgüç-Kunt, Asl and Harry Huznga, 2001, The taxaton of domestc and foregn bankng, Journal of Publc Economcs 79, Demrgüç-Kunt, Asl and Harry Huznga, 1999, The determnants of commercal bank nterest margns: some nternatonal evdence, World Bank Economc Revew 13, Demrgüç-Kunt, Asl, Luc Laeven and Ross Levne, 2003, Regulatons, market structure, nsttutons, and the cost of fnancal ntermedaton, Journal of Money, Credt and Bankng 36, , D Govann, Julan, 2005, What drves captal flows? The case of cross-border M&A actvty and fnancal deepenng, Journal of Internatonal Economcs 65, Ederveen, Sjef and Ruud de Moojj, 2006, What a dfference does t make? Understandng the emprcal lterature on taxaton and nternatonal captal flows, Economc Papers No. 261, European Commsson, Drectorate-General for Economc and Fnancal Affars, Brussels. Egger, Peter, Smon Loretz, Mchael Pfaffermayr, and Hannes Wnner, 2009, Blateral effectve tax rates and foregn drect nvestment, Internatonal Tax and Publc Fnance 16, European Commsson, 2010, Communcaton on the taxaton of the fnancal sector, Brussels. 20

22 Eurostat, 2004, Structures of the Taxaton Systems n the European Unon, Offce for Offcal Publcatons of the European Communtes, Luxembourg. Evenett, Smon J., 2003, Do all networks facltate nternatonal commerce? U.S. law frms and the nternatonal market for corporate control, Journal of the Japanese and Internatonal Economes 17, Gwartney, James, Robert Lawson, Herbert Grubel, Jakob de Haan, Jan-Egbert Sturm, and Eelco Zandberg, 2009, Economc Freedom of the World: 2009 Annual Report, The Fraser Insttute, Vancouver, BC. Head, Keth, and Therry Mayer, 2002, Illusory Border Effects, CEPII Workng Paper No Hnes, James, R., 1996, Altered states: Taxes and the locaton of foregn drect nvestment n Amerca, Amercan Economc Revew 86, Huznga, Harry, Luc Laeven, and Gaëtan Ncodème, 2008, Captal structure and nternatonal debt shftng Journal of Fnancal Economcs 88, Huznga, Harry and Johannes Voget, Internatonal taxaton and the drecton and volume of cross-border M&As, Journal of Fnance 64, IBFD, 2009a, Corporate Investment Income, IBFD Publcatons, Amsterdam. IBFD, 2009b, Corporate Taxaton n Europe, IBFD Publcatons, Amsterdam. IBFD, 2009c, IBFD Tax Treates Database, IBFD Publcatons, Amsterdam. IBFD, 2009d, Tax News Servce, IBFD Publcatons, Amsterdam. Internatonal Monetary Fund, 2010, A far and substantal contrbuton by the fnancal sector, Fnal report for the G-20, Washngton, DC. Kaufman, Danel, Aart Kraay, and Massmo Mastruzz, 2009, Governance Matters VIII, World Bank Polcy Research Workng Paper Kemsley, Deen, 1998, The effect of taxes on producton locaton, Journal of Accountng Research 36, KPMG Internatonal, 2009, KPMG's Corporate Tax Rate Survey , KPMG Internatonal. Loretz, Smon, (2008), Corporate taxaton n the OECD n a wder context, Oxford Revew of Economc Polcy, 24,