The Impact of Endogenous Tax Treaties on Foreign Direct Investment: Theory and Empirical Evidence *
The Impact of Endogenous Tax Treaties on Foreign Direct Investment: Theory and Empirical Evidence *
Abstract and Keywords
This chapter sheds further light on the question of how tax treaties affect outward foreign direct investment (FDI). The chapter is organized as follows. Section A outlines the general equilibrium model of trade and multinational firms for studying the welfare and FDI effects of tax treaties. The theoretical hypotheses are summarized in Section B. Section C presents the database, the econometric methods, and the empirical results. Section D discusses the results in the light of previous research, and provides an extension regarding the time pattern of accumulation of the treatyinduced effect on FDI.
Keywords: FDI, taxation, tax treaty, international taxation, outward investments
INTRODUCTION
There is a large body of theoretical and empirical literature viewing taxation as an important determinant of a country’s locational attractiveness for investors (see Hines 1999; Gresik 2001; de Mooij and Ederveen 2003, for comprehensive surveys). One of the most visible obstacles to crossborder investment is the double taxation of foreignearned income. Double taxation arises if the same tax base (e.g., income or wealth) of a specific taxpayer (i.e., a person or enterprise) is taxed in two or more jurisdictions. Bilateral tax treaties, representing mutual agreements on the definition of tax liabilities, the assignment of taxing rights or the determination of withholding tax rates, are the most important way to circumvent double taxation of enterprises.
Worldwide, the importance of tax treaties has increased tremendously in the last decades. The number of treaties in force has risen from 100 in the 1960s to more than 2,500 today (see Easson 2000; Arnold, Sasseville, and Zolt 2002). Most of them are based on model conventions established by international organizations such as the Organization for Economic Cooperation and Development (OECD) or the United Nations, where the former accounts mainly for the interests of industrialized countries and the latter for those of less developed economies.
Apart from double taxation, tax treaties are usually motivated by additional objectives, depending on the specific economic and legal situation of the treaty partners. The preamble of the OECD Model Tax Convention on income and capital (henceforth OECD Model) emphasizes two major objectives: the alleviation of double taxation and the restriction of tax avoidance. The treaty partners are left (p.514) free to agree on one or both objectives.^{1} The U.S.–Canada treaty, for instance, focuses on “the avoidance of double taxation and the prevention of fiscal evasion with respect to taxes on income and on capital …” The coexistence of different aims highlights a fundamental conflict in tax treaty formation. As far as double taxation is concerned, tax treaties may on the one hand promote international production and investment activities. For instance, the assignment of taxing rights via tax exemption (Art. 23A OECD Model) or tax credits (Art. 23B OECD Model), or agreement on maximum withholding tax rates (e.g., for dividends paid from foreign subsidiaries), lower the overall tax burden of multinational enterprises. On the other hand, restricting tax evasion and tax avoidance, primarily through transfer pricing rules (Art. 9 OECD Model), antitreaty shopping rules, or the exchange of information (Art. 26 OECD Model), may induce negative effects on international investment.
The conventional view of tax lawyers and administratives is that tax treaties are desirable to attract foreign investment. In contrast, Dagan (2000) argues that tax treaties are primarily intended to redistribute tax revenue (from poor to rich countries), or to lower administrative costs (e.g., via the definition of tax terms between contracting countries). Dagan concludes that double taxation should be more effectively eliminated by unilateral arrangements, which are less expensive and more flexible than tax treaties. In fact, some countries (e.g., Chile in the early 1990s) have not signed tax treaties but have unilaterally offered multinational enterprises a similar environment with respect to tax rules. However, there are several aspects of tax treaties that cannot be replicated by a country unilaterally. Most importantly, tax treaties reassure international investors exante on how they are taxed with their overseas profits and this, in turn, may attract FDI.
Although international issues of taxation have been of growing concern in economics, tax treaties have attracted less attention so far. The majority of available studies examines theoretically the role of different tax assignment rules for international investment flows. There are three methods of international tax assignment: Under the credit method, foreignearned profits are taxed in both the parent and the host country, but the taxes paid abroad are deductible from the domestic tax liability.^{2} Under the exemption method, in contrast, domestic and foreign profits are taxed separately within their jurisdiction.^{3} A third method, (p.515) though not proposed in the OECD Model, is the deduction method, which allows a deduction of foreign taxes from the domestic tax base. Bond and Samuelson (1989), relying on a model with a capitalexporting and a capitalimporting country and discriminatory taxation (i.e., different tax rates for domestic and foreignearned income), demonstrate that tax rates will increase in equilibrium if the treaty partners agree on a credit system instead of the deduction method. This, in turn, induces a decline in investment flows to zero at the extreme (see Hamada 1966) for an earlier contribution). Davies and Gresik (2003) show that this result no longer holds under the assumption that subsidiaries are able to finance their operations in the host country. Considering nondiscriminatory taxation, Janeba (1995) finds that the home country sets its tax rate equal to zero, irrespective of the assignment of taxing powers. Davies (2003a), focusing on twoway capital flows, shows that there are always positive capital flows in equilibrium, regardless of which tax relief method is employed. In contrast to Janeba (1995), he finds that the home and host country set positive tax rates in equilibrium. In sum, the theoretical hypotheses with respect to the impact of tax treaties on international investment flows are ambiguous, so it remains an empirical question whether tax treaties promote or hinder foreign direct investment (FDI).
Empirical studies on the relationship between tax treaties and FDI are still scarce, and the available ones adopt methods that rely on the assumption that tax treaties are exogenous. Depending on whether a specification of FDI in logs or levels is chosen, previous studies find negative or no effects on FDI (Davies 2003b; Blonigen and Davies 2004a, 2004b; see Davies [2004] for an excellent survey). In this regard, Blonigen and Davies (2004a) indicate that empirical models of FDI should be formulated in logs rather than levels.^{4} However, based on such a specification recent research does not identify a significant impact of exogenous tax treaties on FDI (see Blonigen and Davies 2004a). For the average countrypair, the question then arises, whether implementing a treaty can be justified from an economic point of view.
This chapter sheds further light on the obviously relevant question of how tax treaties affect outward FDI, emphasizing two important issues that are not raised in previous research. First, it is reasonable to assume that new treaties are implemented only, if both treaty partners gain in welfare. Hence, tax treaties are endogenous from a general equilibrium perspective. Accordingly, it is a major concern under which conditions positive, treatyinduced welfare effects arise. Put differently, the determinants of tax treaties are at issue. Second, given that a treaty is implemented (i.e., given its positive welfare effects), the associated impact on FDI is of interest. Focussing on the double taxation aspect of tax treaties, we set up a twocountry general equilibrium model of trade and FDI, which is (p.516) closely related to the knowledgecapital model of the multinational enterprise (Carr, Markusen, and Maskus 2001; Markusen 2002; Markusen and Maskus 2002). The model is parameterized and the hypotheses on the impact of tax treaties on welfare and outward FDI (rather than foreign affiliate sales) are derived from numerical simulations. Abstracting from the potential administrative costreducing effect of tax treaties and tax avoidance issues,^{5} we find that treaties often have positive welfare effects, but the associated effect on FDI can even be negative without tax evading FDI.
The simulation results for welfare motivate an empirical model to explain the probability of implementing new tax treaties. Further, given positive welfare effects, the results indicate under which circumstances one should expect negative FDI effects from implementing a treaty. In econometric terms, we estimate the average treatment on the treated, that is, the effect of a tax treaty on FDI between countries that have actually implemented a treaty. We construct a bilateral panel data set for tax treaties over the period from 1985 to 2000. Applying different matching estimators and focusing on differencesindifferences, we estimate a negative effect of newly implemented tax treaties of about −34% on changes in bilateral OECD outward FDI.
The remainder of the chapter is organized as follows. Section A outlines the general equilibrium model. The theoretical hypotheses are summarized in Section B. Section C presents the database, the econometric methods, and the empirical results. In Section D, we discuss our results in the light of previous research, and provide an extension regarding the time pattern of accumulation of the treatyinduced effect on FDI. The last section concludes with a summary of the main findings.
A. A Numerically Solvable General Equilibrium Model of FDI and International Taxation
1. Households
There are two economies, referred to as countries 1 and 2 and indexed as {i, j} = {1, 2}. Both countries produce two goods, Z and X, Z is a homogeneous good produced at constant returns to scale by a competitive industry. Xgoods are differentiated in the usual Dixit and Stiglitz (1977) fashion. We consider the following firm types: national firms (NEs) sell on the local market and export to the other country, where the number of national firms of country i is denoted by n_{i}; horizontal multinational enterprises (MNEs) are running production plants in both countries, where h_{i} denotes the number of horizontal multinationals headquartered in i; vertical MNEs are able to unbundle the headquarters and the production plant, where v_{i} is the number of vertical multinationals with headquarters in i and production plants only in j. In contrast to horizontal MNEs, (p.517) vertical ones engage in goods trade. ${X}_{ij}^{n}$ are the exports of country i–based NEs to country j. Similar definitions apply for the other quantities, where the first subscript always denotes the home country of the firm whereas the second one refers to the country of consumption. X_{ic} denotes the consumption of X in country i, being a constant elasticity of substitution (CES) aggregate of the individual varieties. Consumer preferences are assumed to be a nest of the homogeneous Zgood and the differentiated Xgood. The symmetry of varieties within a group of goods allows us to formulate utility of country i (U_{i}) as follows:
Iceberg transport costs (τ) impede goods trade in both sectors. In quantity terms, one unit of consumption of an Xvariety in country j requires a firm in i to send (1 + τ) units. Similarly, one consumption unit of a foreign Zgood in j requires a firm to ship, that is, to produce, (1 + τ) units. For convenience, quantities of X are defined as (both NE and vertical MNE) firmspecific production for the respective foreign market, whereas Z_{ij} and Z_{ji} are normalized to represent consumed rather than produced quantities.
The consumer maximization problem can be solved in two steps. In the first step, each variety ${X}_{ij}^{k},k\in \left\{n,h,v\right\},$ needs to be chosen so as to minimize the cost of attaining X_{ic}, whatever the consumption of X_{ic} is. In the second step, consumers allocate income between the Zgood and the composite Xgood. Let ${p}_{ji}^{k}$ be the price of an X variety in country i produced by a typek firm headquartered in country j. The price for the homogeneous good, q_{i}, is indexed once, since all (indigenous and foreign) homogeneous goods consumed at a single location i must face the same price q_{i}. Further, s_{i} denotes the price aggregator, defined as the minimum cost of buying one unit of X_{ic} at prices of an individual variety:
2. Factor markets, production and income
Let ω_{Ki} and ω_{Li} denote the factor rewards for capital and labor, respectively. Assuming that Zproduction only uses labor (L), variable unit costs (i.e., marginal costs) c_{Zi} satisfy
There is monopolistic competition in the Xsector, where each firm produces under a Leontief technology, using both capital and labor. The countryspecific unit input coefficients of capital and labor are denoted by a_{Kxi} and a_{Lxi}, respectively. Additionally, national firms and horizontal and vertical MNEs require capital to setup plants α_{Kni}, α_{Khi}, α_{Kvi}, and they employ labor to produce firmspecific assets and blueprints (a _{Lni}, a _{Lhi}, a _{Lvi}) In line with the literature, we assume that firm specific fixed costs are highest for horizontal MNEs, slightly lower for vertical ones, and lowest for exporters: a_{Kni}ω_{Ki} + ωa_{Lni}ω_{Li} 〈 ωa_{Kvi}ω_{Ki} + a_{Lvi}ω_{Li} 〈 a_{Khi}ω_{Ki} + a_{Lhi}ω_{Li} Specifically, we set a _{Lvi} = a _{Lhi} = 1+δ, a _{Kvi} = 1+γ and a _{Khi} = 2+γ without loss of generality. δ is the additional labor requirement to organize a multinational network, and 1 + γ are the fixed costs country i’s MNEs have to incur to set up a foreign plant in j. As mentioned above, horizontal MNEs also run domestic production plants, which is reflected in a _{Khi} > a _{Kvi}.
Factor market clearing in country i for capital K_{i} and labor L_{i} requires
National firms in i have to bear fixed costs of ${a}_{Kni}{\omega}_{Ki}+{a}_{Lni}{\omega}_{Li}\u3008\text{}{a}_{Kvi}{\omega}_{Ki}+{a}_{Lvi}{\omega}_{Li}\u3008{a}_{Khi}{\omega}_{Ki}+{a}_{Lhi}{\omega}_{Li}$ and operating profits are taxed at the rate t_{i} (i.e., country i’s corporate tax rate), which is applied to profits earned in country i.^{6} The profit condition for NEs, therefore, implies
Vertical and horizontal MNEs are able to finance fixed costs via operating profits of both the domestic and the foreign activities. We maintain that the foreign subsidiary fully (and immediately) repatriates its profits to the domestic parent (see Hartman [1985] and Sinn [1993] for a discussion of alternative assumptions). Without a tax treaty, operating profits from foreign affiliates are exposed to double taxation. Let ${t}_{Mi}^{}$ denote the tax rate of country i applied to foreignearned profits of domestic multinational firms (i.e., the taxation of “permanent establishments” abroad indicated by superscript M), and ${t}_{Mj}^{}$ the withholding tax on repatriated profits levied by country j. The conditions for the number of MNEs are thus given by
From the perspective of an MNE with headquarters in country i, t_{j} is the corporate tax rate abroad, and, therefore, $\left(I{t}_{i}^{M}{t}_{j}^{\omega}\right)\left(I{t}_{j}\right)$ is equal to the overall tax burden on foreignearned profits under double taxation.^{7} Applying the exemption method in the tax treaty, ${t}_{i}^{M}$ becomes zero. In contrast, if both treaty (p.520) partners agree on the credit method, we have to replace ${t}_{i}^{M}$ by ${t}_{i}^{M}={t}_{i}^{{M}^{\prime}}{t}_{j}^{\omega}$ (assuming that ${t}_{i}^{M}>{t}_{j}^{\omega}$ ^{8}
All factors are owned by the households, so that consumer income (i.e., GNP) in country i is given by
The equivalence of total factor income (E_{i}, E_{j}) and demand in each economy implicitly balances international payments.
3. The public sector
The only source of income for the public sector are taxes on operating profits of firms. Tax revenues for country i can therefore be summarized as:
We consider three alternatives of public expenditure. First, governments pay lumpsum transfers to consumers. Second, governments use tax revenues to finance public infrastructure. By assumption, the production of public infrastructure only uses labor and lowers the capital needed to set up a plant (for this approach see Egger and Falkinger 2003; Kellenberg 2003). Let PI_{i} be the amount of public infrastructure provided by government in country i. The capital needed to set up a national firm in i is given by
A third possibility is to finance a public good PG_{i}, which directly enters the utility function of consumers (see, for example, Huizinga 1991; Lockwood 2003). (p.521) Again, it is assumed that the production of the public good only needs labor. The modified utility function in this case is given by:
4. Model parametrization
For our simulations, we use the following parameter values. World factor endowments are set at L = 100 and K = 50. To study the effect of changes in the absolute bilateral country size, we triple both countries’ factor endowments. For the additional labor requirement to organize a multinational network, we assume δ = 0:01. Foreignowned plants exceed the domestic capital requirement by γ = 0:1. The fixed input coefficients in the Xsector production are assumed at L_{xi} = 0:6 and K_{xi} = 0:4. We consider a value for the elasticity of substitution of ε = 4, which is similar to that one usually used in the literature (Markusen 2002). According to the United Nations World Trade Database, the share of manufacturing trade in the 1990s amounts to 70%–80% of total trade. Therefore, we assume an expenditure share for manufactures of α = 0:8. Motivated by the findings in Baier and Bergstrand (2001), the iceberg trade cost parameter is set at τ = 0:2. The scaling parameter used to model the importance of public expenditures for the capital usage in plant setup is set at β = 0:4. Regarding tax rates, we assume ${t}_{i}={t}_{i}^{M}={t}_{j}={t}_{j}^{M}=0.25,$, and ${t}_{j}^{\omega}={t}_{i}^{\omega}=0.05$ in the case of equal taxes. In the alternative scenario with unequal tax rates we assume ${t}_{i}={t}_{i}^{M}=0.25$ and ${t}_{j}=\text{}{t}_{j}^{M}=0.I.$.^{9}
B. Hypotheses
To analyze the impact of tax treaties on FDI, two issues are of special interest: (i) welfare effects of tax treaties—under which conditions are countries inclined toward implementing a bilateral tax treaty?; and (ii) FDI effects of tax treaties–does FDI rise or fall after implementing a tax treaty, given that the treaty increases welfare?
TAX TREATIES AND WELFARE: A bilateral tax treaty is implemented only if both countries gain in welfare.^{10} Figure 1 plots the signs of welfare effects of tax (p.522)
We infer the robustness of these results with respect to the alternative modes of government expenditures mentioned above (i.e., tax revenues are used to finance a public good, or they are distributed as a lumpsum transfer to consumers). Qualitatively, our theoretical findings on the relationship between welfare gains and factor endowment/country size differences remain unaffected.
To study the impact of bilateral country size on the treatyinduced welfare effects, we triple the world factor endowments ceteris paribus. The result is illustrated in Figure 2. Now, there are islands with positive welfare effects along the main diagonal. Moreover, the black ellipsoid associated with nonpositive welfare effects is narrower than in Figure 1. Hence, countries with similar relative factor endowments are more likely to implement a tax treaty, if both of them are larger. The reason for this lies in the rising dominance of horizontal MNEs in similarly endowed economies. If horizontal MNEs are prevalent in both countries, they will always gain from bilateral tax treaties. Notably, a change in world factor endowments does not affect the welfare effects of tax treaties under the alternative considered modes of public spending.
TAX TREATIES AND FDI: As mentioned in the introduction, the reduction of a multinational’s tax burden through the abolishment of double taxation should (p.524)
Figure 4 combines the information in Figures 1 and 3, and illustrates under which factor endowment configurations a treaty increases (reduces) a country’s outward FDI, given that a treaty will be implemented (i.e., increases welfare in both countries). In all nonblack areas in the figure, countries are willing to (p.525)
ASYMMETRIC TAXATION: So far, we assumed identical tax rates in the two economies. With asymmetric tax rates, the results concerning the welfare and FDI effects of tax treaties do not differ much (see the corresponding figures in the supplement, available upon request from the authors). This holds true irrespective of whether the credit or the exemption method is applied. Accordingly, we conclude that the derived hypotheses are robust enough to warrant empirical implementation.
C. Data and DifferenceInDifference Estimation of the Endogenous Tax Treaty Effect on Bilateral Outward FDI
We estimate the effect of tax treaties on bilateral outward FDI of the OECD economies in the period from 1985 to 2000. Bilateral FDI stock data at an annual basis are from the OECD (Bilateral Investment Statistics Yearbook 2003). Detailed information on the implementation dates of tax treaties is available from the International Bureau of Fiscal Documentation (IBFD; Tax Treaties Database 2002, Release 3). Tables A1 and A2 in the Appendix provide details for those newly implemented tax treaties over the covered period that are part of the sample used in the empirical analysis. In order to isolate the effect of tax treaties on bilateral FDI, we focus on the marginal effect of implementing a new treaty on the change in FDI. Our theoretical model suggests tax treaty implementation as an endogenous event. Hence, there is self selection. To take account of this endogeneity, the proper specification of the decision to implement a new tax treaty (i.e., selection into treatment) is essential.^{13} Further, a differenceindifference approach is recommended to guard against timeinvariant, endogenous, unobserved effects. Such timeinvariant sources of bias arise, for example, if country (p.527)
Table a1. NUMBER OF OBSERVATIONS OF TREATED AND UNTREATED
Implementation year 
Treated 
Untreated 
Total 

1985 
3 
20 
23 
1986 
1 
31 
32 
1987 
3 
26 
29 
1989 
3 
34 
37 
1990 
4 
35 
39 
1991 
3 
40 
43 
1992 
5 
46 
51 
1993 
4 
56 
60 
1994 
3 
62 
65 
1995 
10 
57 
67 
1996 
6 
68 
74 
1997 
6 
73 
79 
1998 
7 
70 
77 
1999 
6 
60 
66 
2000 
3 
41 
44 
Total 
67 
719 
786 
SELECTION INTO TREATY FORMATION:
We compute differences between the average log FDI in the twoyear period after a new treaty was implemented and the average log FDI in the twoyear period before it was implemented. The same difference is calculated for the group of untreated country pairs. Finally, the formation of an appropriate control group is important. For each implementation year, we construct a biannual pretreatment and a biannual posttreatment period (the latter includes the implementation year). Only those country pairs that did not sign a treaty until the end of the posttreatment period are valid potential control units. We select the country pairs in the control group according to their similarity in the propensity score of the selection equation (a probit model). The effects of new tax treaties on FDI are estimated by several matching estimators (see below). We focus on the tax treaty effect on outward FDI for those country pairs that have actually implemented a new treaty. Hence, these estimates reflect the average treatment effect on the treated.
As indicated in the summary of the theoretical hypotheses, the welfare effects from implementing a bilateral tax treaty at the beginning of period $t+{I}^{\left(\Delta {T}_{ij,t+1}\right)}$ are determined by the following measures of average absolute and relative bilateral factor endowments in period t: total bilateral country size (approximated by the log bilateral $GDP,{G}_{ijt}=\mathrm{ln}\text{}\left(GD{P}_{it}+GD{P}_{jt}\right)$, the similarity in bilateral country (p.528) (p.529)
Table a2. NEWLY IMPLEMENTED TREATIES BETWEEN 1985–2000 (ONLY THE 67 TREATIES COVERED IN THE EMPIRICAL ANALYSIS ARE REPORTED)
OECD economy 
Partner economy 
Implementation year 

Australia 
Indonesia 
1993 
Austria 
Australia 
1989 
Austria 
South Africa 
1997 
Austria 
Slovenia 
1999 
Canada 
Sweden 
1985 
Canada 
Brazil 
1988 
Canada 
Inda 
1987 
Canada 
Hungary 
1995 
Canada 
South Africa 
1997 
Denmark 
Greece 
1993 
Finland 
Mexico 
1999 
France 
Turkey 
1990 
France 
Mexico 
1993 
France 
Venezuela 
1994 
France 
Panama 
1996 
Germany 
Philippines 
1985 
Germany 
Turkey 
1990 
Germany 
Mexico 
1994 
Germany 
Venezuela 
1997 
Iceland 
Canada 
1998 
Iceland 
Netherlands 
1999 
Iceland 
Poland 
2000 
Italy 
Bulgaria. 
1992 
Japan 
South Africa 
1998 
Korea, Rep. 
Philippines 
1987 
Korea. Rep. 
Indonesia 
1990 
Korea, Rep. 
Hungary 
1991 
Korea, Rep. 
Brazil 
1992 
Korea, Rep. 
Egypt 
1992 
Korea, Rep. 
Ireland 
1992 
Korea, Rep. 
Italy 
1993 
Korea, Rep. 
China 
1995 
Korea, Rep. 
Czech Republic 
1995 
Korea, Rep. 
Spain 
1995 
Korea, Rep. 
Romania 
1995 
Korea, Rep. 
Mexico 
1996 
Korea, Rep. 
Russia 
1996 
Korea, Rep. 
South Africa 
1997 
Korea, Rep. 
Portugal 
1998 
Netherlands 
Mexico 
1995 
Netherlands 
Venezuela 
1998 
Netherlands 
Argentina 
1999 
New Zealand 
Thailand 
1999 
Norway 
Greece 
1992 
Norway 
Australia 
1995 
Poland 
Ukraine 
1995 
Poland 
Hungary 
1996 
Portugal 
USA 
1996 
Portugal 
Bulgaria 
1997 
Portugal 
Poland 
1999 
Portugal 
Romania 
2000 
Switzerland 
Mexico 
1995 
Switzerland 
Thailand 
1997 
United Kingdom 
Turkey 
1989 
United Kingdom 
Mexico 
1994. 
USA 
Canada 
1985 
USA 
China 
1987 
USA 
Mexico 
1989 
USA 
Indonesia 
1990 
USA 
Spain 
1991 
USA 
Inda 
1991 
USA 
Israel 
1995 
USA 
Portugal 
1996 
USA 
Thailand 
1998 
USA 
Turkey 
1998 
USA 
South Africa 
1998 
USA 
Venezuela 
2000 
(p.530) Denoting the parameter estimates of the selection model by α’s, we expect the following coefficient signs from the above model. α_{G} > 0 indicates that the probability of implementing a new treaty rises with bilateral country size. The direct effects of S and R reflected in the signs of α_{S} and α_{R} should not be interpreted without reference to the interaction terms. Starting from the center of the factor box in Figure 1, a smaller relative factor endowment difference R is necessary as S gets smaller to obtain a positive welfare effect. Since S 〈 0 by definition, we expect α_{SxR} 〈 0. However, as S gets extremely small toward the lower left and upper right edges of the factor box, an increase in R is no longer associated with a higher likelihood of implementing a tax treaty. Accordingly, we expect α_{(SxR)} ^{2} 〈 0, because S ^{2} > 0. Hence, including S x R and (S × R)^{2} in addition to the main effects S and R helps to capture this nonlinear relationship between country size and relative factor endowment differences on the one hand and implementing a treaty on the other hand. In addition to these determinants, we include the parent and the host government expendituretoGDP ratios (g_{it}, g_{jt}) as two separate controls (World Bank, World Development Indicators 2004). These two variables indicate whether the parent and the host are hightax or lowtax countries.^{15} If the objective of tax treaties is primarily to prevent double taxation, we would expect a negative sign for the coefficient of g_{jt}. We report summary statistics for all variables in the Appendix (Table A3).
Formally, we specify selection into treatment (i.e., the probability of new tax treaty implementation) by the following probit model
Table a3. DESCRIPTIVE STATISTICS
Variables 
Mean 
Std. dev. 
Minimum 
Maximum 

Endogenous variables (changes):^{a} 

New tax treaties implemented 
0.09 
0.28 
0.00 
1.00 
Log change in bilateral outward FDI 
0.47 
0.80 
−3.17 
5.97 
Exogenous variables in selection equation (Initial period levels]:^{b)} 

Log bilateral GDP: G_{j} 
27.75 
1.33 
23.42 
30.27 
Log parenttohost similarity in GDP: S_{j} 
−1.88 
1.21 
−6.20 
−0.69 
Absolute difference in log parenttohost capitallabor ratio: R_{j} 
2.28 
1.83 
0.00 
11.90 
Interaction term: S_{ij} × R_{ij} 
−4.58 
5.57 
−39.40 
0.00 
Interaction term: (S_{ij} × R_{ij})^{2} 
51.93 
141.37 
0.00 
1552 44 
Parent government expenditure toGDP ratio: g 
18.66 
3.85 
9.75 
25.84 
Host government expenditure toGDP ratio: g 
15.63 
5.69 
3.14 
38.23 
Notes: 786 observations.  a) Change between 2year period after the treaty implementation and the 2year period before the implementation. i.e., the treaty dummy is set at 1. if a new treaty was implemented and 0, else. The change in FDI is the log difference between the 2year average of the period after and that one before the treaty was implemented.
Table 1 summarizes the parameter estimates of four different probit models. The first two reported models (referred to as the parsimonious ones) impose the restrictions ${\widehat{\alpha}}_{S\times \leftR\right}\text{}\u3008\text{}o$, whereas the third and the fourth model relax them. The second and the fourth model include country and time dummies. The country dummies especially improve the model fit considerably. This indicates that the propensity of signing a new treaty differs systematically between countries. Further, the restrictions on α_{(SxR)} and α_{(SxR)} ^{2} to be zero are rejected at conventional levels of significance. Accordingly, we use the results from the fourth model in Table 1 in the subsequent analysis. The parameter estimates of this model lend support to our theoretical hypotheses. First, the likelihood of implementing a treaty rises with bilateral country size, reflected by ${\widehat{\alpha}}_{{\left(S\times \leftR\right\right)}^{2}}\u3008o$ Further, 〈 0 and 〈 0 support our expectations regarding the functional relationship between relative and absolute bilateral factor endowment differences on the one hand and (p.532)
Table 1. SELECTION EQUATION (RATIFIED DOUBLE TAXATION TREATIES)
Explanatory variables^{a} 
Parsimonious specification 
Full model 


Excl. Group effects Est. Coeff. 
Incl. Group effects Est. Coeff. 
Excl. Group effects Est Coeff. 
Incl. Group effects Est. Coeff. 

Log bilateral GDP: G_{j} 
−0.04 
3.66 ** 
−0.06 
4.14 ** 
−0.76 
2.25 
−0.97 
2.46 

Log parenttohost similarity 
0.02 
1.61 * 
−0.05 
2.24 *** 
in GDP: S_{j} 
0.34 
1.96 
−0.39 
2.56 
Absolute difference in log parent 
0.03 
0.06 
0.02 
−0.33 * 
tohost capitallabor ratio: R_{j} 
0.96 
0.62 
0.20 
−1.84 
Interaction term: S_{ij} × R_{ij} 
– 
– 
−0.03 
−0.47 *** 
– 
– 
−0.29 
−3.06 

Interaction term: (S_{j} × R_{j}])^{2} 
– 
– 
0.00 
−0.02 *** 
– 
– 
−0.71 
−2.80 

Parent government 
−0.04 *** 
0.09 
−0.04 ** 
0.06 
expendituretoGDP ratio: g 
−3.42 
0.68 
−2.23 
0.48 
Host government 
−0.03 *** 
−0.12 ** 
−0.02 * 
−0.11 ** 
expendituretoGDP ratio: g 
−2.61 
−2.25 
−1.90 
−2.11 
Number of countrypairs 
786 
786 
786 
786 
Pseudo R^{2} 
0.04 
0.25 
0.03 
0.27 
Likelihood ratio tests (P>x^{2}}: Parent countries (18) 
– 
0.00 *** 
– 
0.00 *** 
Host countries (31) 
– 
0 02 ** 
– 
0.01 ** 
Time (14) 
– 
0.81 
– 
0.78 
(a) tvalues in italics below coefficients.
(***) significant at 1%;
(**) significant at 5%;
(*) significant at 10%
TAX TREATIES AND FDI:
One branch of econometric literature on the estimation of endogenous treatment effects suggests propensity score matching techniques to infer the treatment effect on the treated (see Rosenbaum and Rubin 1983; Heckman, Ichimura, and Todd 1998; Vella and Verbeek 1999). We follow this (p.533) line of research and apply several different matching estimators to our differenceindifference specification. For this, we need to compare the growth in FDI of those countries that implemented a treaty with those that did not but were similar in all other respects (as captured by similar propensity scores). A properly weighted sample of untreated country pairs serves as a group of control units.
We apply several different matching estimators. Nearestneighbor matching estimators select one (onetoone matching) or more (in our case, five) untreated control observations for each treated observation according to the propensity score, estimated in the above probit model. Each untreated observation potentially serves as a control for several treated ones. Under radius matching, all control units with a difference in propensity scores within a given radius are used. Kernel matching estimators make use of the whole sample of untreated observations, but they weight them according to the estimated kernel density weights. The obtained kernel estimates may be sensitive to the assumed density function and, more importantly, to the chosen bandwidth of the kernel. Local linear regression matching applies local linear weights rather than kernel weights. In general, there is a tradeoff between exact matches at the danger of low precision (under nearestneighbor, nearestfiveneighbors, small radius matching) and high precision at the danger of inexact matching (under wide radius matching or kernel matching).
For all matching estimators, it is important that the balancing property holds, that is, that the covariates are balanced between the treated and the control group. The balancing property is violated if the values of a covariate differ significantly between the treated and the matched control group. This is likely the case with many controls. Although the balancing property holds for our continuous variables according to ttests, it is violated for the country and time dummies. However, associated problems can easily be avoided by using these dummies not only in the selection equation but also as controls in the treatment regression after matching. If the balancing property holds, the average treatment effect on the treated can always be estimated by regressing the dependent variable (in our case, the change in FDI stocks) on the treatment dummy and a constant in a weighted leastsquares regression. The weights are determined by the chosen matching approach (for instance, in case of onetoone matching all treated and control observations exhibit weight one and all other observations weight zero; under kernel matching, the kernel weights are used). If the balancing property is violated, the problematic controls should simply be included in the weighted leastsquares regression in addition to the treatment dummy (see Blundell and Costa Dias 2002). In our case, this means regressing the change in log FDI stocks on the treatment dummy and the country and time dummies. The coefficient of the tax treaty dummy is an estimate of the average treatment effect on the treated.
Table 2 summarizes our findings for the various matching estimators based on the fourth probit model in Table 1. We report three sets of average treatment effects on the treated. The first one does not condition on the controls of the (p.534)
Table 2. THE IMPACT OF TAX TREATY RATIFICATION ON BILATERAL OUTWARD FDI (MATCHING ESTIMATES) (AVERAGE TREATMENT EFFECT ON THE TREATED)
Excl. country & time dummies 
Incl. country & time dumnies 
Incl. all controls of the probit 


Estd. Effect 
% change 
Estd. Effect 
% change 
Estd. Effect 
% change 

Descriptive comparison (tax treaties exogenous) 
0.002 
0.62 
0.18 
16.86 
0.19 
18.11 
Standard error 
0.12 
12.34 
0.12 
10.24 
0.12 
10.12 
Onetoone matching 
0.40** 
34.17 
0.53** 
42.44 
0.41** 
34.45 
Standard error 
0.17 
10.81 
0.20 
11.63 
0.17 
11.32 
Five nearest neighbor matching 
0.26* 
23.57 
0.30 ** 
26.22 
0.30** 
26.58 
Standard error 
0.15 
11.14 
0.13 
9.60 
0.12 
8.86 
Radius matching 
0.21 
19.99 
0.21 ** 
19.14 
0.21** 
19.15 
Standard error 
0.14 
10.77 
0.10 
8.08 
0.09 
7.34 
Kernel matching 

Epanechnikov kernel: bandwidth = 0.06 
0.23 * 
21.35 
0.20 ** 
18.75 
0.21** 
19.20 
Standard error 
0.14 
10.75 
0.10 
8.05 
0.09 
7.28 
Epanechnikov kernel; bandwidth = 0.01 
0.24 
22.45 
0.22 ** 
19.90 
0.22** 
20.28 
Standard error 
0.15 
11.51 
0.11 
8.73 
0.10 
7.83 
Epanechnikov kernel; bandwidth = 0.8 
0.12 
12.15 
0.26 *** 
23.24 
0.24*** 
21.82 
Standard error 
0.13 
11.06 
0.10 
7.35 
0.09 
7.00 
Gaussian kernel; bandwidth = 0.06 
0.21 
19.63 
0.21 ** 
19.32 
0.21** 
19.26 
Standard error 
0.13 
10.72 
0.10 
7.99 
0.09 
7.29 
Uniform kernel: bandwidth = 0.06 
0.24 * 
21.71 
0.21 ** 
19.60 
0.22** 
20.13 
Standard error 
0.14 
10.73 
0.10 
7.91 
0.09 
7.17 
Local linear regression matching 
0.22 
20.69 
0.18* 
17.16 
0.20** 
18.23 
Standard error 
0.14 
11.09 
0.10 
8.54 
0.09 
7.55 
Note: The tangible and intangible investment to sales ratios are logistically transformed; the selection probit equation includes firm size in terms of employment in 1993, the foreign ownership dummy, and the average exports to sales ratio 19971998. Tangible and intangible investment is measured in percent of sales and averaged over the periods 19992001. ** significant at 5%.
On top of the matching results, the descriptive comparison reflects the parameter estimate of a regression of the log difference in FDI on the change in the tax treaty dummy (all standard errors in Table 2 are heteroskedasticityrobust). In the absence of endogenous selection, this provides a consistent differenceindifference estimate of both the treatment effect and the standard error. In contrast, the standard error of the tax treaty parameter estimated in a fixedeffects panel is likely downward biased (Bertrand, Duflo, and Mullainathan 2004). The covariates of the probit are jointly significant in the secondstep weighted leastsquares regression of the log change in FDI on the tax treaty dummy. Accordingly, we consider the third block of results in Table 2 as the preferred one. Note that the tax treaty coefficients cannot be interpreted at face value. Instead, one should take account of the underlying semilogarithmic regression framework (see Kennedy 1981). We compute the percent changes in FDI associated with these estimates following van Garderen and Shah (2002). This obtains the corresponding percent change in FDI.
Treating tax treaties as an exogenous variable, we do not find any significant impact on FDI. However, the onetoone matching estimates indicate that newly implemented tax treaties reduce FDI significantly. This points to endogenous selection into treatment leading to an underestimation of the tax treaty effect on FDI in absolute value. As long as we condition on the country and time dummies in the secondstep weighted leastsquares regression of FDI on the tax treaty dummy, tax treaty implementation is found to reduce FDI significantly. This effect must be interpreted as an average treatment effect on the treated. It indicates the change in outward FDI for two countries that have actually implemented a tax treaty. The general result of a negative tax treaty effect is neither affected by the choice of the matching procedure nor by different assumptions for bandwidth and density functions in case of kernel matching. The effect is highest under onetoone matching in absolute value. There, the control units are closest to the treated ones (“exact” matching). The corresponding tax treaty effect is more reliable than that one estimated from “inexact” but potentially more efficient procedures such as kernel matching or radius matching. As we enlarge the control group by considering not only the nearest neighbor but also other control units, the weighted estimated impact gets smaller in absolute value. The smallest estimate is obtained from the local linear regression procedure. In sum, this indicates that the loss in matching quality from using more control units is relatively big as compared to the associated gain in efficiency in our sample.
D. Discussion and extension
How do the obtained estimates relate to the results in previous research? As in Blonigen and Davies (2004a), a log specification in our sample leads to an insignificant tax treaty effect on outward FDI, if we assume that treaties are exogenous (p.536) (see the first line of results in Table 2).^{16} In contrast, we rigorously account for the endogeneity of tax treaties by specifying an empirical selection model, which is guided by the hypotheses derived in a general equilibrium model of trade and multinationals. In our case, ignoring selfselection leads to insignificant and much lower point estimates in absolute value. Further, the data are typically unbalanced and the implementation of new treaties is unequally spaced over time (i.e., the spell lengths before and after new treaties are implemented differ across country pairs with new treaties). This complicates the interpretation of the treatment coefficient, since the effect may implicitly reflect a weighted shortrun and longrun impact. Only if the pretreatment and treatment periods are of the same length for all treated and untreated country pairs in the sample (in our design, two years for each) is a differenceindifference interpretation justified (see also Kyriazidou 1997).
Our estimates of the treatment effects on the treated in Table 2 indicate that the expectation of a negative effect of new tax treaties is justified as concluded in the inferior levelspecifications in Blonigen and Davies (2004a, 2004b). However, our estimates underpin the importance of the endogeneity of tax treaties for an unbiased inference. In our sample of countries, we consider the onetoone matching result of an expected reduction in FDI growth by 34% in the third block of results in Table 2 as most reliable among the reported results.
As mentioned above, we have so far required the pretreatment and the treatment period to be of equal length (two years). If the treatment effects do not take place immediately due to adjustment costs, varying the window size of the treatment period may shed further light on the shortrun versus longrun effects of implementing a new tax treaty on the treated. To assess the accumulation pattern of the tax treaty impact on outward FDI over time, we change the window size of (p.537)
CONCLUSION
Two primary goals of bilateral tax treaties are the elimination of double taxation of crossborder activities and the prevention of tax avoidance and evasion. Accordingly, the net impact of implementing a tax treaty on foreign direct investment is not clearcut.
(p.538) This paper analyzes the influence of bilateral tax treaties on bilateral stocks of outward FDI, where we focus on the double taxation issue but not on tax evasion. We set up a general equilibrium model of trade and multinational firms to study the welfare and FDI effects of tax treaties. This model motivates an empirical specification to explain the implementation of tax treaties. The selection into treatment (i.e., a positive treatyinduced change in welfare) is considered as an endogenous event. Based on this selection model, we estimate the tax treaty implementation effect on FDI using various differenceindifference propensity score matching approaches. In our sample of OECD outward FDI over the period from 1985 to 2000, we identify a significant negative and reasonable treatment effect on the treated, which is robust with respect to the matching estimator choice. A negative tax treaty effect on FDI can be explained by our general equilibrium model, if tax revenues are spent for public infrastructure to reduce plant setup costs. However, it may also indicate that the tax avoidance aspect of bilateral tax treaties is present in our sample.
References
Bibliography references:
Arnold, B. (2004). “Tax treaties and tax avoidance: the 2003 revisions to the commentary to the OECD model,” Bulletin for International Fiscal Documentation, 58, pp. 244–260.
Arnold, B.J., J. Sasseville and E.M. Zolt (2002). “Summary of the proceedings of an invitational seminar on tax treaties in the 21st century,” Bulletin for International Fiscal Documentation, 56, pp. 233–245.
Baier, S.L. and J.H. Bergstrand (2001). “The growth of world trade: tariffs, transport costs, and income similarity,” Journal of International Economics, 53, pp. 1–27.
Baier, S., Dwyer, G.P. and R. Tamura (2002). “How important are capital and total factor productivity for economic growth?,” unpublished manuscript, Clemson University.
Bertrand, M., E. Duflo and S. Mullainathan (2004). “How much should we trust differencesindifferences estimates?,” Quarterly Journal of Economics, 119, pp. 249–275.
Blonigen, B. A. and R. B. Davies (2004a). “The effects of bilateral tax treaties on U.S. FDI activity,” International Tax and Public Finance, 11, pp. 601–622.
Blonigen, B.A. and R.B. Davies (2004b). “Do bilateral tax treaties promote foreign direct investment?,” in J. Hartigan, ed., Handbook of International Trade: Economic and Legal Analysis of Laws and Institutions (Oxford: Blackwell Publishers), forthcoming.
Blundell, R. and M. Costa Dias (2002). “Alternative approaches to evaluation in empirical microeconometrics,” Portuguese Economic Journal, 1, pp. 91–115.
Bond, E.W. and L. Samuleson (1989). “Strategic behaviour and the rules for international taxation of capital,” The Economic Journal, 99, pp. 1099–1111.
Carr, D.L., Markusen, J.R. and K.E. Maskus (2001). “Estimating the knowledgecapital model of the multinational enterprise,” American Economic Review, 91, pp. 693–708.
Dagan, T. (2000). “The tax treaties myth,” New York University Journal of International Law and Politics, Summer, pp. 939–996.
(p.539) Davidson, R. and J.G. MacKinnon (1993). Estimation and Inference in Econometrics (New York: Oxford University Press).
Davies, R.B. (2003a). “The OECD model tax treaty: tax competition and twoway capital flows,” International Economic Review, 44, pp. 725–753.
Davies, R.B. (2003b). “Tax treaties, renegotiations, and foreign direct investment,” Economic Analysis and Policy, 32, pp. 251–273.
Davies, R.B. (2004). “Tax treaties and foreign direct investment: potential versus performance,” International Tax and Public Finance, 11, pp. 775–802.
Davies, R.B. and T.A. Gresik (2003). “Tax competition and foreign capital,” International Tax and Public Finance, 10, pp. 127–145.
de Mooij, R.A. and S. Ederveen (2003). “Taxation and foreign direct investment: a synthesis of empirical research,” International and Tax Public Finance, 10, pp. 673–693.
Devereux, M.P. and R.G. Hubbard (2003). “Taxing multinationals,” International Tax and Public Finance, 10, pp. 469–487.
Dixit, A.K. and J.E. Stiglitz (1977). “Monopolistic competition and optimum product diversity,” American Economic Review, 67, pp. 297–308.
Easson, A. (2000). “Do we still need tax treaties?,” Bulletin for International Fiscal Documentation, 54, pp. 619–625.
Egger, H. and J. Falkinger (2003). “The role of public infrastructure for firm location and international outsourcing,” unpublished manuscript, University of Zürich.
Hamada, K. (1966). “Strategic aspects of taxation on foreign investment income,” Quarterly Journal of Economics, 80, pp. 361–375.
Hartman, D. (1985). “Tax policy and foreign direct investment,” Journal of Public Economics, 26, pp. 107–121.
Gresik, T.A. (2001). “The taxing task of taxing transnationals,” Journal of Economic Literature, 39, pp. 800–838.
Heckman, J., H. Ichimura, and P. Todd (1998). “Matching as an econometric evaluation estimator,” Review of Economic Studies, 65, pp. 261–294.
Hines, J.R. (1999). “Lessons from behavioral responses to international taxation,” National Tax Journal, 52, pp. 305–322
Huizinga, H. (1991). “National tax policies towards productinnovating multinational enterprises,” Journal of Public Economics, 44, pp. 1–14.
Janeba, E. (1995). “Corporate income tax competition, double taxation treaties, and foreign direct investment,” Journal of Public Economics, 56, pp. 311–326.
Kellenberg, D. (2003). “The provision of public inputs and foreign direct investment,” Center for Economic Analysis Discussion Paper No. 0311 (Department of Economics, University of Colorado at Boulder).
Kennedy, P.E. (1981). “Estimation with correctly interpreted dummy variables in semilogarithmic equations,” American Economic Review, 71, p. 801.
Kyriazidou, E. (1997). “Estimation of panel data sample selection model,” Econometrica, 65, pp. 1335–1364.
Lockwood, B. (2003). “Imperfect competition, the marginal cost of public funds and public goods supply,” Journal of Public Economics, 87, pp. 1719–1746.
Markusen, J.R. (2002). Multinational Firms and the Theory of International Trade (MIT Press: Cambridge, MA).
Markusen, J.R. and K.E. Maskus (2002). “Discriminating among alternative theories of the multinational enterprise,” Review of International Economics, 10, pp. 694–707.
(p.540) Mutti, J. and H. Grubert (2004). “Empirical asymmetries in foreign direct investment,” Journal of International Economics, 62, pp. 337–358.
Rosenbaum, P.R. and D.B. Rubin (1983), “The central role of the propensity score in observational studies for causal effects,” Biometrika, 70, pp. 41–55.
Sinn, H.W. (1993.) “Taxation and the birth of foreign subsidiaries,” in H. Heber and N. van Long, eds., Trade, Welfare, and Economic Policies: Essays in Honor of Murray C. Kemp (University of Michigan Press: Ann Arbor, MI), pp. 325–352.
van Garderen, K.J. and C. Shah (2002). “Exact interpretation of dummy variables in semilogarithmic equations,” Econometrics Journal, 5, pp. 149–159.
Vella, F. and M. Verbeek (1999). “Estimating and interpreting models with endogenous treatment effects,” Journal of Business and Economic Statistics, 17, pp. 473–478.
Wooldridge, J.M. (2002). Econometric Analysis of Cross Section and Panel Data (MIT Press: Cambridge, MA).
Notes:
(*.) This chapter was reprinted with permission from Canadian Journal of Economics. The chapter was originally published as “The impact of endogenous tax treaties on foreign direct investment: theory and empirical evidence”, Canadian Journal of Economics 39(3), 2006, 901–931. The authors are grateful to Ron Davies and the seminar participants at the 2004 IIPF Congress in Milan for helpful comments and suggestions. Financial support from the Austrian Fonds zur Förderung der wissenschaftlichen Forschung (FWF, grant no. P17028G08) is gratefully acknowledged.
(1.) The aim of preventing tax avoidance was introduced recently in the 2003 revision of the OECD Model. In the 1963 and 1977 versions of the OECD Model, tax evasion was limited to fraud or criminal tax evasion (see Arnold 2004).
(2.) If the foreign tax burden is higher than the domestic one, the tax rebate is limited to the tax level of the home country. The limitation may be applied either for a single country (percountrylimitation) or for all countries (overalllimitation).
(3.) Consequently, this reduces the domestic tax base by the amount of foreign profits, which, in turn, lowers the overall tax burden in the presence of a progressive income tax system. In practice, this ‘splitting advantage’ is reduced via an exemption with progression.
(4.) Mutti and Grubert (2004) significantly reject level specifications versus log specifications based on a RESET test motivated by Davidson and MacKinnon (1993) in the context of taxation and FDI.
(6.) In terms of the OECD Model, national firms do not hold a “permanent establishment” (i.e., “a fixed place of business through which the business of an enterprise is wholly or partly carried on”; Art. 5 OECD Model) in foreign markets, and are therefore, not subject to taxation abroad.
(7.) Several OECD countries also apply various forms of unilateral methods to reduce the burden of double taxation, such as flat rates, deductions from the tax base or exemptions of foreignearned profits. However, in most cases such practices do not completely eliminate double taxation.
(8.) If $t+{I}^{\left(\Delta {T}_{ij,t+1}\right)}$, a firm would accumulate an “excess foreign tax credit.” In this case, the credit is typically limited to ${t}_{i}^{M}$ yielding ${t}_{j}^{\omega}={t}_{i}^{\omega}=0$. Consequently, the overall tax burden under the credit method is equal to that of the exemption method (see Hines 1999 for further details) The credit method is applied, for instance, by the United States, the United Kingdom, Japan, Greece, Ireland and Spain, while the exemption method is used by Germany, France and Italy. We do not refer to the deduction method, since it is not proposed in the OECD Model and it is typically applied unilaterally (see Devereux and Hubbard 2003 for a welfare analysis).
(9.) The choice of these rates is motivated by the fact that withholding tax rates on crossborder payments of dividends are of similar size in most bilateral tax treaties.
(10.) Here, we rule out the possibility that the welfare gain of one party is used to compensate the welfare loss of the other. Further, it is reasonable to assume that marginal administrative costs of treaty formation rule out the implementation of a treaty at zero welfare gains.
(11.) Note that we focus on the endogeneity of bilateral treaties, leaving the unilateral setting of tax rates exogenous. Bond and Samuelson (1989), Janeba (1995), and Davies (2003) investigate the strategic tax rate setting under different assignment rules (i.e., credit, exemption, and deduction). The case of endogenous tax treaties under exogenous tax rates seems justified in the medium run. For instance, in the dataset used in the empirical analysis below, the number of newly implemented tax treaties is five times higher than that one of changes in the statutory corporate tax rates.
(12.) Graphically, imagine a relative factor price line with a negative slope. Two points on this line are relevant: the consumption point on the diagonal of the factor box, and a second point located at the border between the white and the black area. The segment of the factor price line between these two points is smaller for countries of dissimilar size than for countries with identical size.
(13.) Wooldridge (2002, 637) points out that endogeneity is an issue in panel data, if the treatment (in our case, implementing tax treaties) is correlated with timevarying unobservables that affect the response. The same argument applies for differenceindifference analysis. However, this correlation takes place in a theoretical general equilibrium model such as the one outlined above.
(14.) We are grateful to Scott Baier for kindly providing the data.
(15.) Alternatively, we used corporate tax rates available from the Office of Tax Policy Research (World Tax Database). However, the results are insensitive with respect to this choice.
(16.) In a sample of OECD parent countries and a knowledgecapital specification using data over the period 1983–1992, Blonigen and Davies (2004b) estimate a significant negative impact of new tax treaties on levels of bilateral outward FDI stocks and flows. Their preferred specification is a model with fixed countrypair effects, and the identified tax treaty effect is extremely large in absolute values. In Table 5 of Blonigen and Davies (2004b), they report a new tax treaty parameter of −2597:6 in the FDI outbound stock specification in column 1. Given the average level of OECD outbound stocks of 3378:13 reported in their Table 2, this means that FDI stocks should be expected to shrink by almost 77%. In column 3 of Table 5, they run a fixed countrypair specification which includes the lagged FDI stock as a regressor. The reported new tax treaty parameter in this model is −2212:3 and the coefficient of the lagged dependent is 0:308, implying a longrun treaty effect of −3197:0, that is, a reduction in FDI stocks by 97%. These results are also consistent with the levels regressions in Blonigen and Davies (2004a), finding new treatyinduced reductions in U.S. FDI of 77% for outward stocks and of 88% for inward stocks. However, the effect on U.S. FDI in a preferable specification in logs is insignificant.
(17.) We are not able to consider longer periods because of the associated substantial loss of observations and, in particular, of tax treaty–implementing country pairs.