Asymmetric FDI and Tax-Treaty Bargaining: Theory and Evidence. April Revised March 2003

Transcription

1 Forthcoming: Journal Of Public Economic Aymmetric FDI and Tax-Treaty Bargaining: Theory and Evidence Richard Chiik and Ronald B. Davie April 2001 Revied March 2003 Abtract: Tax treatie are often viewed a a mechanim for eliminating tax competition, however thi approach ignore the need for bargaining over the treaty term. Thi paper focue on how bargaining can affect the withholding taxe et under the treaty. In a imple framework, we develop hypothee about pattern in treaty tax rate. A key determinant for thee pattern i the relative ize of bilateral foreign direct invetment (FDI) activity. In plauible ituation, more aymmetric countrie will negotiate treatie with higher tax rate. Thi theory i then teted uing 1992 data from U.S. bilateral tax treatie. Overall, the data upport the prediction that greater aymmetric FDI activity increae the negotiated tax rate. JEL Claification: F23, H25, K34. Key Word: Foreign Direct Invetment, Tax Treatie, Multinational Corporation, Bargaining, Withholding Taxe. Department of Economic DM-309C, Florida International Univerity, Miami, FL, 33199; Phone: (305) ; Fax: (305) ; chiikr@fiu.edu. Department of Economic, 533 PLC Building, 1285 Univerity of Oregon, Eugene, OR ; Phone: (541) ; Fax: (541) ; rdavie@oregon.uoregon.edu. We thank participant of the 2001 World Tax Competition, epecially John Mutti and William Randolph. We alo thank Gordon Hanon and eminar participant at the Univerity of Wetern Ontario, Oregon State Univerity, the Univerity of Kentucky, the 2003 American Economic Aociation Meeting, the Spring 2002 Midwet International Economic Group meeting, the Spring 2002 Royal Economic Society meeting, and the Fall 2001 Southeat Theory and International Economic Meeting. All error are entirely the reponibility of the author.

2 I. Introduction Tax treatie are often viewed a a remedy for tax competition. Under bilateral tax treatie, withholding taxe, tax definition, and relief method are choen jointly by the treaty partner. 1 In the rubric of game theory, tax treatie move taxation from non-cooperative tax competition to a cooperative etting. Becaue thee policie are now et cooperatively, it i tempting to believe that they eliminate tax competition. Thi preumption, however, i mileading ince the term of the treaty (and the ditribution of the gain from treaty formation) mut be bargained over. If countrie differ in their preferred treatie, then there i conflict within treaty formation itelf. In particular, if countrie differ in their deired treaty-pecified tax rate, there i a kind of tax competition a each country puhe for it preferred tax rate. 2 Recognizing the pattern of thi bargaining ha important implication for undertanding the potential of tax treatie. Thi paper make a firt attempt at modeling the conflicting goal in treaty formation by preenting a imple bargaining framework. The implication of the model are then teted uing 1992 data from bilateral tax treatie with the United State and within the OECD. We find that treatypecified withholding taxe vary in a ytematic way which i conitent with our imple bargaining model. In particular, our reult highlight the importance of difference in bilateral FDI activity between the two countrie. A the ize of thi aymmetry grow the cope for cooperation i decreaed and negotiated tax rate are higher. We find imilar reult for relative country ize. Thee finding indicate that it may be difficult for highly aymmetric countrie to negotiate a treaty, and in fact, our analyi ugget that countrie with highly aymmetric FDI activity are alo the leat likely to have a treaty. While tax treatie are rarely dicued in thi literature, when they are, they are typically preented a a mechanim of eliminating the inefficiencie created by tax competition. In fact, in the OECD (1997) model treaty, the claim i made that their goal i to reduce the inefficiencie caued by 1 For an excellent dicuion of the working of the OECD model tax treaty, ee Baker (1994). For additional dicuion on ome of the primary goal and iue of tax treatie, ee Blonigen and Davie (2002). 2 Thi type of tax competition differ coniderably from the tandard ort in which government trategically et taxe to influence foreign direct invetment (FDI). Wilon (1999) and Greik (2001) provide excellent urvey of thi literature. 1

3 tax competition and double taxation. In a model of unilateral capital flow, Janeba (1995) how that when taxe are uniform and either foreign tax credit or an exemption i ued to combat double taxation, there exit a et of mutually beneficial, harmonized tax rate. Since thi i a common proviion in tax treatie, Janeba ugget that thi provide a role for treatie. Davie (2003) demontrate that a imilar et of mutually beneficial, harmonized taxe exit when FDI flow are bilateral. Neither author, however, dicue how a particular rate i choen from thi et of mutually beneficial taxe. Thi i the firt goal of the preent paper. In addition to the mall economic literature on tax treatie, there alo exit work by international tax lawyer. Thee writing often portray treatie in a le-hopeful light than the economic tudie do. Dagan (2000), for example, pan the FDI efficiency gain a a myth. Intead, he argue that in U.S. treaty formation two other apect dominate policy development: reduction in tax loe overea and alleviation of adminitration cot. Radaelli (1997) alo ugget that U.S. treaty policy i not driven by a deire to improve efficiency, but rather to reduce tax evaion through mechanim uch a tranfer pricing. Other gain from treaty formation include information haring between government, dipute reolution mechanim, and coordinated policie on item uch a tranfer pricing and expene allocation. With thee argument in mind, we aume that a country can benefit from the treaty in two way. Firt, by negotiating a lower withholding tax, a country can lower what it invetor pay in overea taxe. Note that thi gain for one country i a lo to the other country. Thu, it can eaily be the cae that under the treaty one country net tax payment fall while the other rie. Thi i the ource of contention Dagan focued on. Second, each country experience an additional gain unrelated to the withholding tax level which arie from reduction in adminitrative and enforcement cot. In thi way, the treaty repreent an increae in total urplu for the two countrie and they mut agree how to plit thee gain between them. One way to tranfer urplu between the countrie i through the appropriate choice of a common withholding tax rate. 3 Uing a non-ymmetric bargaining olution (ee, for example, 3 When a firm invet overea, it typically doe o through a ubidiary that repatriate profit to it parent through dividend, interet, and royalty payment. Since thee payment are a cot to the ubidiary, they are not taxable by the hot country a part of the ubidiary' income. Neverthele, 2

4 Myeron, p.390, 1991), we olve for the jointly choen tax rate a a function of relative bargaining trength, relative FDI activity, and non-treaty tax policie. Our econd goal i to then tet the implication of thi olution uing 1992 U.S. and OECD data. We perform thi etimation uing affiliate ale data, FDI tock data, and intrument developed from recent work by Carr, Markuen, and Maku (2001). We find trongly ignificant reult for the effect of aymmetry in FDI activity on the negotiated tax rate. Our reult ugget that a ale from overea affiliate become unbalanced, the negotiated tax will rie. The ign of the coefficient are conitent with a ituation in which change in tax revenue are highly important to countrie. Our proxie for bargaining ize do not perform a expected, but do reinforce the idea that more aymmetric countrie negotiate higher tax rate. Thee reult are robut acro our data et and under both Tobit etimation and the ue of intrumental variable. Finally, ince treaty-negotiated rate are only oberved for countrie with treatie, we tet whether ample election i driving our reult. Uing Heckman (1979) two-tep method, we find that our reult hold even after controlling for ample election. The remainder of the paper i a follow. Section II preent the bargaining model and develop ome hypothee for our etimation. Section III dicue our data and etimation procedure. Reult are found in Section IV. Section V conclude. II. Bargaining in Tax Treatie In order to develop tetable hypothee, in thi ection we develop a model of bargaining over the treaty-pecified withholding tax rate. While thi model i admittedly tylized, we ue it to explore the conflict likely to arie in treaty formation and to anticipate what reult might be found in the data. Since the treatie in quetion are bilateral, conider a etting with two countrie, home and foreign. Mirroring the data, each country invetor produce at home and abroad. Home dometic production i hk ( Z) and it overea production function i h ( Z) where K i home capital tock and Z are it capital outflow. Similarly, foreign dometic production i f K ( Z ) and it overea mot hot government capture part of thi parent income through withholding taxe levied on thee repatriation. Tax treatie reduce withholding taxe by pecifying maximum allowable tax rate. 3

5 production i f ( Z ). The price of output i contant and equal to one for all four type of production. In thi one period model, all overea profit are repatriated. Upon repatriation, home invetor mut pay a withholding tax to the foreign country jut a foreign invetor mut pay the home withholding tax. Without a treaty, the home withholding tax i τ and the foreign withholding tax i thee non-treaty rate remain the ame for non-treaty countrie, we treat them a exogenou parameter. 4 Following the treaty convention, under a treaty each country chooe the ame τ. Since withholding tax τ. 5 It i alo ueful to note that among U.S. treatie, τ i no greater than either country non-treaty rate. Although we do not explicitly model it, we take the new view on the effect of withholding taxe which wa initiated by Hartman (1985). Thi theory poit that withholding taxe will have no effect on the ize of overea operation by a mature ubidiary. 6 Thi occur becaue, given the initial parental capital injection, retained earning preent a le expenive ource of invetment than repatriated and re-exported fund. Sinn (1993) formalized thi reult and alo found that while withholding taxe do not affect the ize of a mature ubidiary, they can impact the initial parental injection of equity. However, a hown by Weichenrieder (1996), even thi effect on the initial equity injection need not influence FDI activity if there exit paive invetment option in the hot country. Grubert (1998) extended the Hartman-Sinn reult to a etting in which profit can be repatriated through dividend, royaltie, and interet payment and found reult imilar to Sinn. Furthermore, a demontrated by Althuler and Grubert (1996), there exit cotly triangular trategie which enable firm to achieve the equivalent repatriation without actually repatriating fund from the hot country. Both Grubert (1998) and Grubert and Mutti (1999) provide empirical reult conitent with 4 Our model could eaily be extended to include both home and foreign corporate income taxe. Since thee do not change under tax treatie, they would cancel out in the bargaining olution. Thu, their incluion would not alter the model prediction. 5 If intead of a common tax rate each country chooe it own treaty-pecified tax, then under the Hartman-Sinn analyi there exit a continuum of home and foreign taxe which achieve the ame ditribution of rent a any given common tax. 6 It i important to note that the Hartman/Sinn reult indicate that withholding taxe hould be irrelevant for the ize of overea operation, not that other taxe uch a corporate income taxe hould be irrelevant. A wealth of evidence, uch a that provided by Grubert and Mutti (1999), ugget that thee other taxe do affect FDI activity while withholding taxe do not. 4

6 the Hartman/Sinn reult. In light of thi work, we operate under the aumption that the ize of overea operation (and the Z and Z ) are exogenou to the withholding tax and are therefore determined outide the model. 7 Note that under thi aumption, in equilibrium, one would expect the non-treaty rate to equal one ince countrie can increae their hare of the inbound FDI tax bae without affecting the ize of that bae. Thi reult can be eliminated by extending the model to a etting in which, due to non-tax bae cot of taxation uch a cotly enforcement, equilibrium withholding tax rate would be le than one even under the Hartman-Sinn reult. Since our goal i to decribe the treaty-negotiated taxe rather than the non-treaty rate, we et thi iue aide and ue the current, more direct model. Without lo of generality we label our countrie uch that home ha relatively more overea output: h ( Z) > f ( Z ) (1) In thi ene, the home country i large relative to foreign. Note that thi doe not correpond with the tandard trade definition of large. We define h Z f Z ( ) ( ) a the degree of aymmetry, o that an increae in h ( Z) or a decreae in f ( Z ) increae the aymmetry of FDI. Although Z and Z are equilibrium level of FDI, given the Hartman/Sinn aumption they are determined exogenouly to the environment we conider. An alternative interpretation of our model i that it capture only a ingle-period naphot of a more general dynamic environment. In thi way, the FDI level contitute an optimal repone to the previou tax level a well a to the expected current and future reduction from the tax treatie. We concentrate, here, on the negotiated tax for thee given FDI level. We explore thi richer dynamic environment, whereby FDI level gradually increae and treaty-tax rate gradually decreae over time, in a companion paper (Chiik and Davie, forthcoming). 8 There, we derive the ame aymmetry effect a we note here. We preent the impler naphot 7 Since the Hartman/Sinn reult arie in part becaue of the firm ability to defer dometic taxe until repatriation, it may be unwarranted to impoe it in a one period model. However, if the preent etting i thought of a a ingle period of a longer, intertemporal interaction, then it i not unreaonable to aume their reult within that period. A our goal i to develop ome tetable prediction for a croection of data rather than to retate the Hartman/Sinn analyi, we proceed with the current formulation. 8 For example, the original draft of the U.S.-Canadian treaty lowered the royalty withholding tax to 15%. Renegotiation in the late 1970 reduced thi to 10% and according to Price-Waterhoue 5

7 verion in the current paper, in order to leave room for the empirical verification, and we direct the intereted reader to that work. In line with Bond and Samuelon (1989), Janeba (1995), and other, we aume that government maximize national income. In our bilateral context, a country national income i the um of the home-controlled production and net tax revenue. Thu, home national income without a tax treaty i: Y h h τ h τ f = + +. (2) Under the tax treaty, two change occur to home national income. Firt, with movement to a common withholding tax τ, net tax revenue can change. Second, there i an additional non-revenue gain imply from being part of a treaty. Thi non-revenue gain can repreent reduction in enforcement cot due to increaed inter-governmental cooperation, reduction in the wateful triangulation activitie decribed by Althuler and Grubert (1996), or Dagan (2000) adminitrative aving. In order to keep thi effect a general a poible, we imply repreent thee gain by Φ( h, f ) for home and Φ ( h, f ) for foreign. Both of thee function are non-decreaing in both of their argument, uch that greater FDI activity (either outbound or inbound) can lead to greater non-revenue gain from the treaty. To eae the development of the intuition for our main reult, for the moment we aume that the inbound effect i zero, i.e. that Φ =Φ = 0. Thi would be the cae if each country i only h f concerned with aving adminitrative and enforcement cot and reducing the tranfer pricing loe aociated with it own outbound FDI. Not urpriingly, when thi aumption i relaxed additional interaction are introduced which lead to le clear-cut reult. However, a i hown below, under plauible condition imilar reult can be found even in thi more general cae. Incorporating thee change under the treaty, home income can be written a: Y = h+ h τ ( h f ) +Φ ( h, f ) (3) Corporate Taxe: A Worldwide Summary, it wa eliminated entirely in Additionally, the parental dividend tax fell from 15 to 10 to 5 percent over the ame period. See our companion paper for further example of treatie that exhibit falling withholding taxe over time, often a a reult of renegotiation. 6

8 which i again the um of worldwide ale and net tax revenue with the addition of the non-tax revenue gain from being part of the treaty. Since home i the large country, note that under the treaty it collect negative net tax revenue. Combining equation (2) and (3), home gain from the treaty i: Y Y = τ h τ f τ( h f ) +Φ ( h, f ) (4) which i the change in net tax revenue plu the non-tax gain from the treaty. Thi mirror Dagan (2000) belief that reduction in net tax loe and adminitrative cot are the primary concern for the (relatively large) U.S.. Similar to home, foreign gain from the treaty i: Y Y = τ f τ h + τ( h f ) +Φ ( f, h ) (5) From the third term in equation (4) and (5), we can ee the conflict between countrie over the treatypecified tax rate τ, ince increaing τ hift gain from the large home country to the mall foreign country. Becaue of thi, home prefer a lower τ while foreign prefer higher tax rate. Since the treaty contitute a Pareto improvement it mut be individually rational for both countrie, therefore, τ i contrained to the et τ h τ f Φ τ h τ f +Φ,, h f h f with the two countrie preferring the oppoite end of thi interval. 9 Anecdotally, the neceity of a mutually-beneficial treaty i illutrated by the U.S. treaty with Hondura which eliminated all withholding taxe. Hondura felt that, ince nearly all FDI flowed from the U.S. to Hondura, the treaty only benefited the U.S., which i akin to a τ outide of thi range. Thi led Hondura to cancel the treaty in 1966, ten year after it implementation (Diamond and Diamond, 1998). 9 Thi et of mutually-agreeable tax rate i comparable to thoe found in Janeba (1995) unilateral FDI model and in Davie (2003) bilateral FDI model. In thoe paper, they dicard the Hartman/Sinn aumption and aume endogenou capital flow. By harmonizing tax rate under a treaty, urplu i created by improving capital market efficiency, which can alo be repreented by our Φ and Φ. They find that there i a range of tax rate which achieve thi reult and that the two countrie prefer oppoite end of thi range. Neither author, however, dicue the method by which a treaty arrive at a particular tax rate from the range. 7

9 We appeal to the generalized Nah bargaining olution to derive the reult from the bargaining proce. Thi technique indicate that the olution can be found by chooing a τ which maximize a weighted product of the two countrie gain from treaty formation. Thu, τ mut atify: τα ( ) arg max ( Y Y) α ( Y Y ) 1 α (6) where α repreent the relative bargaining power of the home country. 10 The firt order condition for thi problem can be written a: ( ) α ( ) α Y Y ( f h ) + (1 ) Y Y ( h f ) = 0, (7) in which the firt term i negative. The maximizing τ i unique. After ome implification the negotiated tax can be written a: τ h τ f + (1 α) Φ αφ τ = h f (8) Note that when α = 1 home ha all of the bargaining power and the choen tax i equal to home mot preferred tax rate. Similarly, a α approache zero, the tax approache foreign optimal tax rate. Subtituting the treaty tax rate into equation (4) and (5) we ee that the olution ditribute income between the countrie in the following way: and Y Y = α( Φ+Φ ) (9a) Y Y = (1 α)( Φ+Φ ). (9b) Hence, the non-revenue gain are plit according to each country bargaining power. By chooing the appropriate tax rate, income i hifted from one country to another uch that both are willing to agree to the treaty. In thi fahion, a ide payment i built into the treaty itelf and i reflected in the agreed upon tax rate. Note that if there are no non-revenue gain from treaty formation, then there i no cope for treaty formation. From equation (8), we can derive the following et of comparative tatic a well a our main reult. 10 Hence, unle α = ½ we are abandoning Nah (1953) ymmetry axiom. 8

10 Propoition 1: If Φ =Φ = 0, then the negotiated tax rate i increaing in the aymmetry of h f outbound FDI level. Furthermore, the comparative tatic effect of τ, τ, α, h, and f on τ are: τ f = < 0 τ h f (10a) τ h τ = h f > 0 τ Φ Φ = < 0 α h f (10b) (10c) τ τ τ + (1 α) Φ h = > 0 h h f (10d) τ τ τ+ αφ f = < 0 f h f (10e) Proof: Firt remember that without lo of generality we label the countrie o that h > f. The comparative tatic then follow from manipulation of the partial derivative of equation (8). Furthermore, if h > f, then an increae in the aymmetry of FDI level correpond to an increae in h and/or a decreae in f and, therefore, from equation (10d) and (10e), τ i increaing along with thi aymmetry.! The intuition behind Propoition 1 i a follow. An increae in the home non-treaty rate mean that the foreign country ave more in tax payment for a given treaty tax rate. Thi foreign windfall i a cot for the home country. To return to the bargaining olution it i, therefore, neceary to tranfer urplu from foreign to home, which i achieved by lowering τ. The intuition for a change in τ i imilar. A noted above, when home ha more bargaining power, it i able to puh more trenuouly for it deired low tax rate, yielding a negative derivative. The comparative tatic effect of FDI activity on τ can be decribed a follow. The firt two term in equation (10d) how the difference between foreign non-treaty and treaty tax rate. A 9

11 home overea invetment rie, thi increae home gain from a tax reduction. At the ame time, thi lower foreign gain from the treaty. Thi neceitate tranferring income from home to foreign to return to the bargaining olution, a reult which i achieved by raiing τ. Thi effect i reinforced by the econd term, which repreent change in the non-tax gain from treaty formation. An increae in non-tax gain for the home country generate a larger total urplu from the treaty, (1-α) percent of which mut be tranferred to foreign. Equation (10e) indicate that an increae in f S ha the oppoite effect. Since h > f, an increae in the aymmetry of FDI level i generated by an increae in h S and/or a decreae in f S. Thi increaed aymmetry in FDI level affect the threat point in the bargaining problem and a thee threat point become more aymmetric the negotiated tax rate mut increae. Under the more general formulation for Φ and Φ, equation (10d) and (10e) become: τ τ (1 α) α h h = τ + Φ Φ h h f (11a) and f = τ τ τ+ αφ (1 α) Φ f h f f (11b) Here, both the third and fourth term repreent change in the non-tax gain from treaty formation. When h rie, thi increae total urplu from the treaty by Φ +Φ, α percent of which will go to home. Since Φ arie in home directly, to again atify the Nah bargaining olution home mut h tranfer the difference between thi amount and home hare of the total rie in urplu to foreign. Note that if Φ i ufficiently enitive to h S or if home bargaining power i ufficiently large, then a rie in h S may require a tranfer to home, i.e. a reduction in τ. In thi cae, the comparative tatic in (11) are ambiguou. An alternative way of recognizing thi ambiguity i that a rie in h S increae both h h Y Y andy Y through the non-revenue treaty gain in equation (7). Since thee move in the ame direction, to determine whether it i neceary to move income to home or to foreign it i neceary to 10

12 compare the relative magnitude of thee change, i.e. compare Φ with Φ. Thi leave u with two h h ituation in which we can unambiguouly ign thee comparative tatic: when revenue change are larger than the non-revenue change or when a rie in a country' outbound invetment increae the non-revenue gain generated within it border by more than it increae that country' hare of total non-revenue gain. Thee condition are ummarized by Propoition 2. Propoition 2: Sufficient condition for an increae in the aymmetry of FDI level to generate an increae in the treaty tax rate are that: a) revenue effect are larger than non-revenue effect, or b) that Φ α > and that Φ 1 α h h Φ Φ f f 1 α >. α An intereting extenion of the above theory i to allow the non-revenue gain from treaty formation to depend on the treaty pecified tax rate. If the non-revenue gain partly reflect reduction in wateful tax evaion or enforcement, then we might expect that lower taxe would correpond to le wate and greater gain from the treaty o that both Φ and Φ are negative. Under thi modification, we can write the firt order condition from the bargaining problem a: ( ) τ α ( ) τ τ α Y Y ( f h +Φ ) + (1 ) Y Y ( h f +Φ ) = 0. (12) Equation (12) i the counterpart of equation (7) with one key difference, the introduction of two new term, α( Y Y ) Φ and τ (1 α)( Y Y) Φ which are both negative. Thee new term demontrate that τ when lower taxe increae the non-revenue gain from the treaty, additional downward preure i placed on the tax rate. To olve for the comparative tatic in (12), note that ince α (0,1), 0, Y Y > 0, and f h +Φ τ < 0, it mut be the cae that h f +Φ τ > 0. Uing thi, it i τ Y Y > 11

13 τ traightforward to verify that although the expreion for τ they match the ign of thoe given in Propoition 1. τ τ, are now more complicated, τ α, Unfortunately, without impoing further retriction, it i impoible to ign the marginal effect of FDI. A with Propoition 2, part of thi difficulty arie from comparing the relative change in Φ and Φ with repect to FDI. An additional problem, however, i that thee change are alo τ affected by the treaty tax. In particular, the ign of h, depend on the ize of the cro-derivative Φ and τ, h τ, h Φ relative to each other. A comparable difficulty exit for igning the comparative tatic effect of foreign FDI. Neverthele, if thee additional effect are relatively mall, then the reult of Propoition 2 carry through, that i, increae in FDI aymmetrie increae treaty-pecified tax rate. With thee prediction in hand, we now turn to data on U.S. bilateral treatie to tet their plauibility. III. Empirical Methodology and Data To tet the prediction of our theory, we ue two data et, both from The firt data et conider the U.S. and it bilateral tax treaty partner. We form two ubample of the U.S. data, one that ue affiliate ale and one that ue FDI tock a the meaure of FDI activity. Our econd data et ue the FDI tock between OECD member countrie. 12 Since treatie affect four different withholding taxe, for each country pair we conider four different tax rate: that on dividend paid to the parent, that on non-affiliated dividend, that on non-financial interet payment, and that on indutrial royalty payment. Although we believe the above model decribe the tradeoff in treaty formation, we can only olve for an explicit functional form under the mot retrictive aumption. Therefore, rather 11 An earlier draft of the paper alo ued 1997 affiliate ale data for the U.S. and it treaty partner. Thi data et included more treatie than the 1992 verion, however, it lacked the neceary control for IV etimation. Since the reult from that data match the preented reult, we omit them for pace. Thee additional reult are available upon requet. 12 Affiliate ale information wa not available for a reaonably large number of OECD countrie. 12

14 than etimate a variant of the tructural equation (8), for our baeline reult we etimate the following reduced-form equation: ijk ( hij, fij, ik, jk, ij, Dk ) τ = τ τ τ α (13) where i i the home country, j i the foreign country, and k i the type of withholding tax. The firt five right hand ide variable are defined a in the theory, that i, h i the value of overea FDI production by the relatively large country, f i the value of overea FDI production by the relatively mall country, etc.. The final term, D k, i a contant plu a et of dummy variable for the parental dividend tax, the unrelated dividend tax, and the royalty tax. Note that ince we are uing all four taxe imultaneouly, our coefficient are bet interpreted a the relation between the independent variable and the overall level of treaty taxe rather than pecific type of withholding taxe. 13 For meaure of h and f, we ue data drawn from two ource. For the U.S. data et, we ue either affiliate ale of non-financial intitution in the hot country or the tock of FDI in the hot country, both of which can be obtained from the Bureau of Economic Analyi webite. 14 We ue two meaure becaue of potential problem with uing affiliate ale. Firt, withholding taxe are not applied to ale but to repatriation, therefore, ale may not cloely approximate the repatriated value of FDI. While tock are uceptible to the ame criticim, we hope that uing two meaure that yield imilar reult alleviate concern. Second, ale are a flow value of invetment and may reflect hortrun variation that doe not correpond to the longer run conideration of treaty formation. Since the 13 In reult not reported here, we alo ran eparate regreion for each of the four tax type. With the exception of the regreion uing the withholding tax on non-parental dividend, thi procedure yielded etimate imilar in ign and magnitude to thoe reported for the aymmetry variable, the foreign tax rate, and our bargaining power meaure. For the interet and royalty withholding tax regreion, thee etimate were generally ignificant when uing either affiliate ale or FDI tock. For the parental dividend regreion, the tock meaure gave u ignficance at the tandard level wherea ale yielded ignificance only in the 20 to 30% range. Since the tock meaure increaed the number of obervation from 21 to 28, thi ugget that combining the tax rate into a ingle regreion improve the etimate ignificance while not dramatically altering their ign or magnitude. To check thi, we alo did pairwie combination of the parental dividend, interet, and royalty taxe. Thi yielded imilar coefficient but increaed ignificance. The non-parental dividend tax regreion yielded a ign reveral for home FDI. However, none of the coefficient from thi regreion even approached ignificance. Additionally, the etimated coefficient on the aymmetry variable were an order of magnitude maller than thoe from the other regreion. Since thi tax mot likely applie to portfolio invetment and not FDI, thi i not epecially urpriing. Thee alternative reult are available upon requet. 14 A of the time of thi paper, thi webite i 13

15 FDI tock i a tock meaure of FDI activity, it idetep thi problem. For the OECD data et, we ue the tock of FDI a reported in the OECD International Direct Invetment Statitic Yearbook. Note that ince thi only report outward FDI for OECD member, it i only poible to contruct the neceary bilateral FDI meaure when both countrie are OECD member. Becaue of thi, the U.S. data preent a broader election of countrie while the OECD data include obervation for which the U.S. i not one of the two treaty partner. On the other hand, ince the OECD data i between only developed economie, it i poibly a better fit for the mature FDI tory of Hartman/Sinn. It hould alo be noted that due to cro-country variation in definition and reporting requirement, the OECD meaure of FDI tock are poibly noiier than the BEA meaure. The year 1992 i ued becaue it i the mot recent year for which both the OECD data and many of our control variable are available. It hould be noted that if the ale or tock meaure report the action of a ingle firm, then the BEA cenor thi data, deleting ome treaty partner from our ample. Thi i a greater problem for ale than the tock data, allowing u to increae our obervation by one-third in the tock regreion. In order to claify countrie a home or foreign, we compare the relative FDI activity of the two countrie for each year that bilateral data wa available and deignated the one which had higher activity in the mot number of year home. In the U.S. data, with a few exception, thi coding mean that the U.S. play the part of the home country. Similarly, in both the U.S. and the OECD data et Japan wa alway a home country which i not urpriing given Japan traditional barrier to inbound FDI. To tet the enitivity of our reult to thi coding, we alo ue a gravity pecification in which, rather than uing the home and foreign FDI meaure eparately, we ue the um of FDI activity and the quared difference between home and foreign FDI. Thi gravity method of dealing with aymmetrie i common in the empirical literature on trade and FDI. 15 Data on non-treaty rate are obtained from the Price-Waterhoue Corporate Taxe - A Worldwide Summary (1992). Thi ource wa alo ued to determine whether a country ue credit or exemption to relieve the double taxation of foreign earned profit. For the U.S. data et, ince the U.S. i almot alway the home country and all U.S. non-treaty withholding tax rate were 30%, the 15 Recent example include Feentra, Markuen, and Roe (2001), Boughea, Demetriade, and Morgenroth (1999), and Brainard (1997). 14

16 home non-treaty tax i nearly contant for the U.S. regreion. The information on the treaty-pecified tax rate i drawn from the treatie themelve a reprinted in Diamond and Diamond (1998). We alo obtain information on the initial year of treaty enforcement from thi ource. Thi i ued to create a treaty age variable which i defined a the number of year ince the firt treaty wa formed between two countrie a of Note that the tax rate information we ue i for the treaty in force a of 1992, thee rate may differ from the initial verion of the treaty that form our treaty age variable. A a meaure of the home country bargaining power, we ue the home country hare of the total gro dometic product (GDP) of the two countrie. 17 Thi proxy i baed on the idea that a country with a larger economy will have more way in the negotiation. One rationale for thi preumption i that a mall country might chooe to appeae a large one in the hope of future conceion on other international agreement uch a trade pact. Data on real GDP come from the Penn-World Table, which are detailed in Summer and Heton (1991). In the gravity pecification, we replace thi meaure with the um of GDP and the difference in GDP quared. In the theory, we make great ue of the Hartman/Sinn reult that overea affiliate ale are unreponive to the withholding taxe. Thi aumption need not hold in the data and we therefore ue a Hauman tet for endogeneity. While gravity model uch a Brainard (1997) have been popular pecification for affiliate ale, they were developed more in repone to the data than to the theory of the multinational enterprie (MNE). Intead we develop our intrument uing recent work by Carr, Markuen, and Maku (2001) and Markuen and Maku (2001), both of whom etablih empirical pecification of FDI activity that are arguably more grounded in the formal theorie of multinational firm. A noted by Blonigen, Davie, and Head (2002) there i a mipecification in thi framework regarding kill variable and we therefore ue their alternative abolute-value pecification. Carr, Markuen and Maku ue their empirical model to examine affiliate ale of U.S. firm in other 16 Although Diamond and Diamond (1998) do not lit a bilateral treaty of FDI between France and Japan, Price-Waterhoue (1992) doe lit treaty tax rate. Therefore, thi country pair i not included in our OECD regreion uing treaty age. If thi treaty i eliminated from all OECD regreion, our reult remain nearly identical. 17 Earlier draft of the paper alo ued the home relative GDP and the home and foreign GDP a two eparate, independent variable a meaure of bargaining power. Thee alternative yielded imilar reult for our other independent variable and are available upon requet. 15

17 countrie and foreign affiliate ale in the U.S. over the period and find that their unretricted pecification fit their data quite well. Blonigen, Davie, and Head how that thi pecification alo perform well uing both U.S. FDI tock and OECD FDI tock data. Detail of our intrumental variable etimation are found in the appendix. Here, we merely note that the modified Carr, Markuen and Maku pecification doe reaonably well in capturing the variation in affiliate ale with R 2 for home and foreign affiliate ale of.9740 and.9196 repectively. Summary tatitic for our data are found in Table A1 of the appendix. Table A2 lit the treaty partner ued in our etimation. In addition to the variable in (13), in ome pecification we conider two additional explanatory variable: treaty age and double tax rule. By uing a cro-ectional approach, we are teting for ytematic variation in the long-run equilibrium of the bargaining game between countrie rather than the marginal effect of change in our explanatory variable. Becaue of thi, in the baeline pecification, we may mi out on long-run effect of our variable on the treaty-pecified tax rate. Specifically, there are two concern that one might have relating to treaty age. Firt, the relative FDI activity in 1992 may not reflect the ituation when the treaty wa originally igned. If thi i the cae, we would expect no ignificant relationhip between FDI in 1992 and treaty-pecified taxe. Thi concern i mitigated omewhat by the fact that treatie can and do get renegotiated. 18 Thu, if the current ituation differ highly from when the treaty wa initially formed, one would expect that thi would lead to a renegotiation. Therefore, we expect that the current verion of the treaty hould be at leat partially reflective of the current FDI ituation. Another poible influence of treaty age i that when countrie have a long hitory of cooperation, thi may impact their tax treaty negotiation ince they may feel more integrated. If thi i the cae, the large country may be more willing to implement a treaty with lower tax rate regardle of FDI aymmetrie ince it may gain conceion on other front. Alternatively, there may exit inertia between countrie with long hitorie that make 18 See footnote 8 for an example of a treaty that ha been renegotiated over time. See our companion paper (Chiik and Davie, forthcoming) for a further explanation of the ource of thi renegotiation and for everal other example of renegotiated treatie. A we develop there, and explain briefly in ection II of thi paper, the renegotiation take place in repone to change in FDI o that the currently pecified treaty tax rate can be conidered a an optimal repone to the 1992 FDI level. 16

18 them le likely to renegotiate an exiting treaty even if one country could reap greater reward from renegotiation. To invetigate thee poibilitie, we will examine the effect of treaty age, both by itelf and interacted with other variable, on the treaty-pecified tax rate. In addition to conidering treaty age effect, we alo control for double tax rule. When a payment i received by a parent firm, it may face parent country taxe in addition to the hot corporate and withholding taxe. The burden of thi parent country tax depend both upon tatutory rate and the double tax rule. In our ample, all treaty partner offer either a limited foreign tax credit or exempt foreign earned profit from dometic taxation. If the parent country offer a credit for hot taxe, then if the combined hot taxe lie below the dometic corporate income tax, the firm marginal effective tax rate i driven by the parent country tax, not hot taxe. 19 Regardle of the relative tax burden, thi i not cae when the parent country offer exemption. Becaue of thi, firm tax avoidance trategie may be more reponive to the withholding tax when it operate under exemption. A uch, thee firm may have a greater incentive to avoid hot taxe, impoing greater enforcement and monitoring cot on government. Thi implie that, all ele equal, treaty gain may be more enitive to the treaty tax when one or both ignatorie ue exemption. While thi implie greater gain from lower negotiated taxe, thi effect may be tempered depending on whether the additional gain tend to accrue to the home or foreign country. To invetigate thee potential effect, we create an exemption dummy variable equal to one if a parent country ue exemption and zero otherwie. We ue thi to etimate a verion of (13) that include both the exemption dummy and it interaction with the FDI meaure. Before turning to our etimation reult, three iue deerve mention. Firt, ince thee data are available for more than one year, it i tempting to ue a panel data pecification. Unfortunately, during the period for which bilateral FDI data i available, there i inufficient within-treaty variation in the treaty-pecified tax rate for thi approach to be ueful. Second, although the treaty tax rate do vary acro type of withholding taxe within a country pair our other variable do not vary within our ingle year ample. Thi preclude the ue of country-pecific fixed effect for the U.S. data et. It 19 Thi tatement i only approximate ince it ignore other hot taxe and additional credit determination iue uch a income baket. 17

19 can alo caue clutering effect, a dicued by Kloek (1981), which can lead to undertated tandard error. Therefore, we correct for clutering on country pair when calculating our tandard error. Finally, treaty tax rate are only oberved for countrie with treatie. Therefore, it i neceary to ak how thi ample election impact our reult. We do thi uing Heckman (1979) two-tep proce that etimate treaty-pecified tax rate conditional on the exitence of a treaty. Note that thi i only poible for the U.S. data et ince there are treatie in place for all of the OECD country pair with available bilateral FDI data. IV. Reult Table 1 through 3 preent our baeline reult uing the U.S. ale, U.S. tock, and OECD tock meaure of FDI activity repectively. In each table, Column 1 preent OLS etimate uing the actual FDI data while Column 2 preent the reult uing the intrument for FDI. Since there are no negative withholding taxe, Column 3 of each table report reult uing the Tobit etimation procedure that correct for a dependent variable retricted to non-negative value. Column 4 of each table report the reult from the gravity pecification. Finally, ince it i poible to ue a fixed effect approach in the OECD data et, Column 5 of Table 3 alo report the reult when uing countrypecific dummy variable. In addition to the reported independent variable, each of thee regreion include a contant a well a dummy variable for the parental dividend, unrelated dividend, and royalty tax. Thee etimate are omitted for pace and are available upon requet. Finally, all tandard error are calculated uing White (1980) conitent method. The non-tobit error are alo corrected for clutering on country pair uing the method decribed by Rodger (1993). Regardle of which ample we ue, we find very imilar reult. For both the U.S. and the OECD data et, home FDI i poitively correlated with the treaty tax while foreign FDI i negatively correlated with the treaty tax. Thee reult are alway ignificant for affiliate ale and uually ignificant for the FDI tock meaure. Since the treatie in our ample lower tax rate, thee reult are conitent with two ituation: revenue effect dominate or, a decribed in Propoition 2, the parent marginal non-revenue gain i larger than the hot. Since thee variable coefficient have 18

20 oppoite ign, our etimate imply that a rie in FDI aymmetry lead to higher negotiated tax rate. We find comparable reult when uing the intrumented FDI meaure. Furthermore, Hauman tet reject the hypothei that there are ytematic difference between the regreion uing the actual FDI data and their contructed counterpart. Thi ugget that endogeneity i not driving our reult. 20 The Tobit reult find imilar effect of aymmetrie, indicating that truncation of tax rate i not reponible for our finding. In fact, for the U.S. tock pecification, the Tobit etimation increae the overall ignificance of the FDI meaure relative to the OLS pecification. Thee reult hold up even after introducing country-pecific fixed effect into the OECD pecification, although tandard error rie o that only the home FDI tock ha a ignificant coefficient. 21 In any cae, even here we find that FDI aymmetrie between treaty partner play a role in treaty negotiation. The reult that FDI aymmetrie increae taxe i confirmed by the U.S. gravity pecification where we find a poitive, trongly ignificant coefficient on quared FDI difference. Thi indicate that at leat for the U.S. our reult are not contingent on our coding of the mall and large countrie. In the OECD data et, we do not find ignificant coefficient on our FDI meaure in the gravity pecification. However, ince the OECD non-gravity reult mirror the U.S. non-gravity reult, the difference between the U.S. and OECD gravity reult may be due to the gravity tranformation exaggerating the additional noie in the OECD data. For affiliate ale, the magnitude of the coefficient on home and foreign affiliate ale center around and repectively. Thi indicate that an increae in home affiliate ale of $1 billion would increae the level of the negotiated tax by.0629 percentage point (i.e. an increae from 5% to %). An equivalent increae in foreign affiliate ale would lower the level of the negotiated tax by.0939 percentage point. While thee magnitude eem mall, conider them in the following light. The U.S. had roughly the ame level of affiliate ale in Canada and the U.K. in 1992, however, Canada had only half the ale in the U.S. that the U.K. did. If Canadian ale roe to thoe of the U.K., our etimate predict a drop in the Canadian treaty tax applied to dividend paid to the 20 In an additional battery of tet in which FDI wa the dependant variable, we found that FDI wa not driven by treaty-pecified withholding taxe. Thee are available on requet. 21 Note that in thi pecification, ince the only Icelandic treaty for which FDI data wa available wa it treaty with the U.S., our ample ize i reduced by four. 19

21 parent of approximately 6.5 percentage point. Noting that thi tax rate i 10% in the Canadian treaty but only 5% in the U.K. treaty, thi ugget economic meaningful effect from aymmetric FDI flow. In mot of our pecification, the home hare of GDP i ignificantly and poitively correlated with the treaty-pecified rate, that i, a home get relatively large, tax rate rie. Thi i the oppoite of our prediction for home bargaining power. One explanation for thi i that our expectation that larger countrie hold more bargaining power i incorrect. For example, ince relatively mall countrie do not upport large international military operation, they are far more likely to hot a U.S. military bae than the U.S. i to hot one of their bae. The threat of expulion might tip the balance in favor of the mall country reulting in higher tax rate. However, when uing the gravity pecification, we conitently find a negative coefficient on quared GDP difference. Thu, in the gravity pecification, a rie in the home GDP relative to foreign reduce treaty-pecified tax rate. Since thi conflict with the home hare of GDP reult, we feel that the more likely explanation i that we imply have poor proxie of bargaining power, highlighting the need for additional reearch on the determinant of bargaining power in international agreement. In any cae, thee etimate ugget that, imilar to FDI aymmetrie, higher GDP aymmetrie are linked to higher negotiated taxe. Turning our attention to the non-treaty tax rate, we find that the foreign non-treaty tax ha a poitive coefficient in all the regreion and i generally ignificant. Thi mirror our theory prediction that when the foreign country ha an initially high tax rate, the average negotiated tax i alo higher. Contrary to the theory, the coefficient on the home non-treaty tax i alo poitive although primarily ignificant only in the U.S. regreion. One poible reaon for thi i that all the variation in the U.S. regreion come from thoe few cae in which the U.S. i the mall, foreign country. A uch, thi variable may imply be capturing other variable pecific to thoe few countrie that have greater FDI in the U.S. than the U.S. doe in them. One factor that argue againt thi i that in the OECD reult, the home non-treaty tax i only ignificant when country-pecific dummie are included. An alternative explanation i that a high non-treaty tax indicate a country with a preference for government revenue relative to firm profit regardle of it relative ize. Thu, large countrie 20

22 that deire large tax revenue may puh for treatie with high taxe, even at the cot of efficiency or tranferring additional urplu to the relatively mall country. Such a rationale would alo be conitent with the coefficient on the foreign non-treaty tax. Treaty Age Effect We now modify the baic pecification by including treaty age, both on it own and interacted with our FDI variable. When we include both FDI and interaction between FDI and treaty age, we only find ignificant coefficient when uing U.S. affiliate ale. Here, the interaction term mirror the above reult, that i increae in FDI aymmetrie increae tax rate. At the ame time, however, the coefficient on home and foreign FDI revere ign, although only the home FDI coefficient i ignificant. Thee ign reveral and overall inignificance are likely due to collinearity between our interacted and non-interacted variable with FDI meaure. To tet thi, in the even numbered column, we drop the baic FDI term and ue only the interacted FDI term. Now we find ignificant reult that match thoe from Table 1 through 3, that i, FDI aymmetrie ignificantly increae tax rate with a tronger effect for older treatie. The treaty age variable itelf i alway negative, but never ignificant. Thi i true even if we exclude all interaction term uggeting that while older treatie may involve lower tax rate, thi link i tenuou. In any cae, when including treaty age effect, we again find that greater FDI aymmetrie are linked to higher treaty-pecified tax rate. Double Tax Relief Effect In Table 5, we turn our attention toward the poible effect of a country double tax rule. To examine thi, we include our exemption dummy both alone and interacted with the FDI variable. 22 In the U.S. data, the home country alway offer credit. Becaue of thi, we only ue a foreign exemption dummy. Note that ince only the foreign country FDI i operating under exemption, we expect an inignificant coefficient on the interaction between exemption and home FDI. For the OECD data, we interact a parent country exemption dummy with it outbound FDI and 22 The coefficient for other exemption interaction were alway inignificant and are therefore omitted from the reported pecification. 21