Multinational Production and Corporate Taxes: A Quantitative Assessment

Transcription

1 Multinational Production and Corporate Taxes: A Quantitative Assessment Zi Wang Shanghai University of Finance and Economics December 28, 207 Abstract I study the impacts of international corporate tax competition and cooperation in the presence of multinational production (MP). I build a quantifiable multi-country general equilibrium model that incorporates salient features of international corporate tax system. The model delivers structural equations that can be used to estimate the key model parameter that determine how firms production sites respond to changes in corporate taxes. Calibrating the model to data on 40 countries, I characterize fully non-cooperative and cooperative corporate taxes and examine the welfare implications of these extremes. I find that countries have strong incentives to lower their corporate tax rates for domestic firms in order to gain from firm relocation at the expense of other countries. I also find that tax competition across eight major countries in MP activities can decrease their welfare by 2.85 percent on average, whereas tax cooperation among these countries can increase their welfare by 0.97 percent. Keywords: Multinational Production, corporate tax, tax competition JEL classification: F23, F42, F2, F3, O9 I am grateful to Stephen Yeaple for his continuous guidance and support. I thank Yuan Zi and Lixin Tang for extensive discussions and in-depth comments. All errors are my own. Correspondence: wang.zi@mail.shufe.edu.cn

2 Introduction In the era of globalization, corporate income tax is no longer solely a domestic policy since multinational enterprises (MNEs) account for a large fraction of global production and corporate tax rates have been shown to be influential for the spatial allocation of MNEs activities. International corporate tax competition and cooperation have drawn increasing attention from researchers and policy makers. In the presence of multinational production (MP), what are the welfare and distributional implications of corporate taxes? What corporate tax rates would countries impose if they did not have to fear any retaliation? What would occur if international tax treaties were abandoned? What would be the outcome if countries engaged in tax cooperation? These questions are key for understanding international corporate tax system. However, there is currently a lack of tractable models that connect theoretical answers to these questions to data, a gap that this paper aims to fill. In this paper, I quantify the aggregate effects of corporate taxes and analyze noncooperative and cooperative tax policies in the presence of MP. To achieve these, I develop a quantifiable multi-country general equilibrium that incorporates trade, MP, and salient features of the international corporate tax system. In the model, firms decide their production sites based on each country s corporate taxes, wage, productivity, market potential, and trade and MP frictions. I utilize bilateral corporate tax rates across over 30 countries and the corresponding multinational sales to estimate the key model parameter that determines how firms production sites respond to changes in corporate taxes across countries. Using the estimated model, I compute unilaterally optimal, Nash, and cooperative corporate tax rates in the 2007 world economy based on the observed bilateral trade and MP flows across 40 countries. My model extends the quantitative MP model developed by Arkolakis et al. (207) by incorporating source-based corporate taxes. 2 The model allows firms to produce outside of their home countries in order to serve destination markets at the lowest cost. A host country, as in practice, can impose different tax rates to firms originated from different countries. My model incorporates many countries with source-based corporate taxes, yet generating closedform gravity equations for trade and MP flows. Based on these structural gravity equations, For example, Huizinga and Voget (2009) show that international double taxation has substantial impacts on cross-border M&A. Barrios et al. (204) show that both host and parent country taxes have negative effects on the location of new foreign subsidiaries. 2 Source-based corporate taxation is the idea that MNEs income should be taxed where the income is earned. An alternative system is residence-based corporate taxation, which says that MNEs income should be taxed where the firms are normally based. The reason why I focus on source-based corporate taxes will be discussed in detail later. 2

3 I estimate model parameters and conducting counterfactual analysis. The key mechanism in my analysis is the economies of scale from firm relocation. In the presence of MP and trade frictions, firms tend to locate their headquarters together with production plants in countries where corporate tax rates are low for domestic firms but high for foreign firms. 3 This firm relocation increases a country s labor demand relative to other countries and thus improves its terms-of-trade. In addition to this term-of-trade effect, firm relocation to a country decreases the country s price index by increasing its access to final varieties (home-market effect). By manipulating corporate tax rates, countries boost firm relocation in order to gain from terms-of-trade and home-market effects at the expense of other countries. MP liberalization strengthens the motives of tax manipulation by shifting tax burdens to foreign firms and workers. The parameter that governs how firms production sites respond to changes in corporate taxes is key for the results. To estimate this parameter, I use equilibrium relationship from my model and data on bilateral corporate tax rates and MP flows in 200. My model generates a structural equation that predicts bilateral MP flows as a function of bilateral corporate tax rates, other frictions that imped MP, and source- and host-country-specific factors. I estimate the coefficient of corporate tax rates by fixed-effect regression, which yields an elasticity of MP entry probability with respect to corporate tax rates of This estimate is in line with the comparable elasticity that has been estimated in Arkolakis et al. (207) using alternative strategies. The outcomes of my counterfactuals also depend on the dispersion of firms productivities and the labor mobility between entrepreneurs who create firms and production workers. I calibrate these two elasticities from the literature and check the robustness of my counterfactual results with these parameters. Armed with structural parameters, I compute changes in equilibrium outcomes with respect to tax changes based on the observed bilateral trade and MP flows using the exact-hat algebra developed by Dekle, Eaton, and Kortum (2007). I first consider optimal corporate taxes, i.e., the corporate taxes countries would impose if they do not fear any retaliation. I find that each country can gain considerably at the expense of other countries by unilaterally imposing optimal tax rates that are much higher for foreign MNEs than domestic firms. The average optimal corporate tax rate for domestic firms is 5.6 percent, whereas the average optimal corporate tax rate for foreign MNEs is 56. percent. These taxes will lead to the average 2.09 percent welfare gain in the tax-imposing 3 Notice that the source-based corporate taxes are imposed by the host country after the production is made. That s why firms are more likely to place their production plants at home if their source countries impose lower corporate taxes to domestic firms. 3

4 country and the average percent welfare loss in other countries. I then turn to analyze Nash corporate taxes among eight countries that have frequently been involved in tax conflicts. 4 I find that all eight countries incur welfare losses from tax wars in which their MP partners retaliate optimally. In tax wars, all countries simultaneously impose optimal taxes in order to spur firm relocation from other countries, which leads to a substantial MP decline and thereby welfare losses for everyone. The average welfare loss from tax competition is percent. Moreover, in a world where all MP costs are reduced by 0 percent from their calibrated levels, the average welfare loss from tax competition is percent. The efficiency loss from tax competition is magnified by MP liberalization. I further investigate cooperative corporate taxes across eight countries involved in tax competition, i.e., the taxes these countries would impose under efficient tax negotiation. The results show that tax cooperation decreases corporate tax rates dramatically in all participation countries, spurring global production and benefiting everyone. Tax negotiation starting from factual taxes yields a 0.97 percent welfare increase for each participant. Moreover, the structure of factual taxes is much closer to cooperative taxes than to Nash taxes. This result highlights the importance of international tax treaties that coordinate corporate taxes across borders. Finally, I show that the welfare gains from tax cooperation are magnified by MP liberalization. To the best of my knowledge, this is the first quantitative analysis of non-cooperative and cooperative corporate taxes in a multi-country environment. Theoretical work including but not limited to Zodrow and Mieszkowski (986), Wilson (986), Kanbur and Keen (993), Gordon and Mackie-Mason (995), Gordon and Hines (2002), Jonannesen (206), Devereux (204), and Devereux et al. (2008) analyzes welfare implications of corporate taxes for both tax-imposing countries and other countries. These models illustrate the qualitative effects of international tax competition on MP and welfare, but work with a small open economy or two countries. Hence, it is difficult to bring those models to data and assess their quantitative importance. In this paper, I incorporate corporate taxation into a multi-country model with trade and MP, calibrate the model to trade and MP data, and quantify the welfare effects of corporate taxes. My model builds on the quantitative MP model developed by Arkolakis et al. (207). A nice feature of their model is that they allow export-platform FDI but still obtain analytical gravity equations for aggregate trade and MP flows. However, Arkolakis et al. (207), together with other quantitative MP models such as Ramondo and Rodriguez-Clare (203), 4 They are Belgium, Canada, Germany, France, Britain, Ireland, the Netherlands, and the U.S. 4

5 Tintelnot (207), and Irarrazabal et al. (203), focus on the welfare effects of MP without considering corporate tax policies. I depart from their work by focusing on the welfare impacts of corporate tax and endogenize it by appealing to optimizing governments. My work is also inspired by the international commercial policy literature in the spirit of Bagwell and Staiger (999). In particular, I calculate optimal taxes, Nash taxes, and cooperative taxes just like trade policy researchers calculate optimal tariffs, Nash tariffs, and cooperative tariffs. The quantitative techniques I use here to compute Nash and cooperative taxes draw from the quantitative techniques in Ossa (204) that are used to study tariff wars and tariff talks. Ossa (206) also uses similar techniques to analyze subsidy competition among the U.S. states. My work is the first attempt in applying this technique to assess international corporate taxation. I estimate the elasticity of firms production locations with respect to corporate taxes to identify the key model parameter. This strategy is related to empirical studies on the effects of corporate taxes on the spatial allocation of MNEs activities. Barrios et al. (2009) show that host and parent country corporate taxes have a negative impact on the location of new foreign subsidiaries. Huizinga and Voget (2009) find that international double taxation has significant impacts on the parent-subsidiary structure of multinational firms created by cross-border M&A. Huizinga, Voget, and Wagner (204) find that international double taxation of foreign-source bank income reduces banking-sector FDI. My empirical estimates are consistent with their findings and are used as an input for my quantitative analysis. The remainder of the paper is organized as follows. In section 2, I present my theoretical framework describing the economic environment and the equilibrium given corporate tax rates. In section 3, I characterize the welfare-maximizing corporate taxes in special cases and the exact-hat algebra that can be used to compute the general equilibrium effects of corporate taxes. In section 4, I estimate the key model parameter that determines how firms production locations depend on corporate tax rates and calibrate the general equilibrium model to the 2007 world economy with 40 countries. Section 5 quantifies the welfare and distributional effects of corporate taxes. In section 6, I study the unilaterally optimal taxes. In section 7, I explore the consequences of tax wars and tax talks. Section 8 concludes. 2 A Model of MP and Corporate Taxation This section presents the theoretical framework that guides my quantitative analysis. My framework builds on quantitative MP framework of Arkolakis et al. (207) and extends 5

6 it to incorporate salient features of international corporate taxation. As I go through the model, I explain how corporate taxes in the model can be linked to policy tools used by governments in reality. My specification of source-based corporate taxes in the model is in line with an extensive literature of international corporate taxation. 5 I consider corporate tax rates chosen by a government with a welfare function developed by Ossa (204, 206). 2. Consumer Preferences I consider N countries in the world indexed by i, l, or n. Country i is endowed with a mass L i agents. Agents are immobile across countries and can work as workers (p) or entrepreneurs (e). An agent decides her position based on wages and agent-position-specific amenity shocks. Preferences for agent ν with position j {p, e} in country n depend on goods consumption c n (ν) and an idiosyncratic amenity shock b j n(ν): U j n(ν) = b j n(ν)c n (ν). () Final consumption is a CES aggregate over a continuum of differentiated goods: C n (ν) = [ q n (ν, ω) σ σ ω Ω n dω] σ σ, σ >. (2) The idiosyncratic amenity shocks b j n(ν) capture the idea that agents have heterogeneous preferences for positions. I assume that these amenity shocks are drawn independently across positions and agents from a Frechet distribution: G j n(b) = exp{ B j nb µ }, j {e, p}, (3) where the scale parameter B j n characterizes average amenities for position j in country n and the shape parameter µ controls the dispersion of amenities across agents in each position within country. 2.2 Production Each differentiated variety is produced by a firm using workers. After paying a fixed entry cost f e in terms of entrepreneurs, a firm is eligible to produce a differentiated variety under 5 The works include but not limited to Zodrow and Mieszkowski (986), Wilson (986), and Gordon and Hines (2002). Keen and Konrad (204) provides an excellent summary of these models. 6

7 monopolistic competition. A firm can potentially produce outside of their source countries and set its foreign affiliates as export platforms. The unit cost for firm ω originated from country i producing in country l is c il (ω) = γ il w l ϕ i (ω)z l (ω), (4) where ϕ i (ω) is the core productivity of firm ω and z l (ω) is the production-location-specific productivity, w l is the wage of workers in country l, and γ il is the iceberg MP cost. As in Melitz (200), we assume that the core productivity, ϕ i (ω), is drawn from a Pareto distribution: P r(ϕ i (ω) ϕ) = G i (ϕ) = T i ϕ θ, ϕ T θ i, θ > max{σ, }. (5) After drawing core productivities, the firm decides which countries to serve. To serve country n, firm ω incurs a fixed marketing costs f n in terms of country n s workers. Firms also incur an iceberg trade cost τ ln for shipping goods from country l to n. After choosing its destination countries, firm ω observes a vector of production-locationspecific productivities (z l (ω)) N l= and chooses its production sites.6 z l (ω) is drawn independently across firms and production sites from a Frechet distribution: P r(z l (ω) z) = exp{ A l z ɛ }, z > 0, ɛ > θ, (6) where A l is the level of production productivity in l and ɛ characterizes the dispersion of productivities across potential production sites. With large ɛ, a firm draws similar productivities across production sites, which means that it may change production sites due to small changes in relative corporate tax rates across countries. Therefore, parameter ɛ reflects the extent to which firms production sites change in response to corporate tax changes. 2.3 Corporate Taxes In this subsection, I derive the firms profits conditional on their production sites and introduce source-based corporate income taxes. The unit cost for firm ω from country i to 6 Following Arkolakis et al. (207), I assume that there is no fixed cost in establishing a production site. This assumption leads to an analytical gravity equation of trade and MP flows. 7

8 produce in country l and sell to country n is c iln (ω) = τ ln c il (ω) = ξ iln ϕ i (ω)z l (ω), where ξ iln = γ il w l τ ln. The standard results for monopolistic competition imply that the pre-tax profits of firm ω from country i serving market n from its plant in country l can be expressed as π iln (ω) = σ σ σ c iln (ω) σ X n P σ n, σ = σ σ, where X n is the total expenditure of country n, and P n is the price index in country n. The corporate tax system in my model is an abstraction of the complex international corporate tax architecture in the real world. 7 Consider a firm with its headquarters located in country i. The profits made by its affiliate in country l are subject to the local corporate income tax before it can be paid out as dividend. I denote the local corporate tax rate in country l as ct l, which is identical for all firms operating in country l. According to Huizinga, Voget, and Wagner (204), for a domestic firm located in country l, this local corporate tax is the only tax on income paid out as dividends at the corporate level. When post-tax corporate income is paid out as dividends from host country l to source country i, it is subject to a nonresident dividend withholding (NRDW) tax, nrdw il, levied by the host country. This tax is an additional tax at the corporate level on account of foreign ownership. Notice that nrdw il is source-country-specific, meaning that the host country can effectively impose different corporate tax rates on MNEs from different source countries. According to Huizinga, Voget, and Wagner (204), all 38 major economies levy source-country-specific corporate taxes. In practice, NRDW tax rates are often determined by bilateral tax treaties. In my quantitative analysis, I consider the local corporate tax rate ct l and NRDW tax rates (nrdw il ) N i= as the policy tools for country l s government. 8 Equivalently, the government of country l decides a vector of tax rates, ( κ il ) N i=, which consists of corporate tax rates for firms originated from countries i =,..., N: κ il := ct l + ( ct l ) nrdw }{{ il. (7) } gross NRDW rate 7 The international corporate tax system in reality has been characterized in detail by Huizinga and Voget (2009) and Huizinga, Voget, and Wagner (204). 8 In reality, dividends earned abroad can also be subject to taxes in MNEs source country. These residence-based taxes are less prevalent and smaller in size than source-based corporate taxes. For simplicity, in this paper I only discuss sourced-based corporate taxes levied by host countries. 8

9 JPN MLT BRA IND BEL CHN ITA IDN DEU USA MEX AUS PRT RUS FIN GRC DNK KOR CZE GBR FRA SVN SWE EST ESP CAN TUR AUT LUX SVK POL NLD ROM LVA LTU BGR TWN HUN IRL CYP CAN USA AUS TUR ESP KOR FRA LUX DEU ITA JPN NLD GBR SWE AUT LTU HUN MEX TWN SVN SVK RUS ROM PRT POL MLT LVA IRL IND IDN FIN EST DNK CZE CYP BRA BEL GRC BGR CHN Factual Local Tax Rate % Factual Gross NRDW Rate % (a) Local (b) Gross NRDW Figure : Corporate Tax Rates in 2007 (Notes: In Panel (b), the dot refers to the median of the gross NRDW rates. The upper bar of the box refers to the 75 quantile of the gross NRDW rates and the lower bar of the box refers to the 25 quantile of the Gross NRDW rates.) Figure illustrates the effective corporate tax rates 9 across 40 countries in Patterns in the data are consistent with the specification of corporate taxes in my model: Panel (a) of Figure shows that local corporate tax rates vary considerably across countries, ranging from 0% for Cyprus to 39.8% for Japan. Moreover, Panel (b) of Figure suggests that the gross NRDW rates are sizable 0 and vary substantially across source countries in most of the host countries. In other words, countries do levy source-country-specific corporate taxes. 2.4 Firm s Problem I proceed by describing the firm s problem of global production and sales, taking corporate taxes as given. The timing of events described in Section 2.2 is summarized by Figure 2. Notice that the firm decides its destination markets before drawing production-site-specific productivities (z l ). As a result, two firms originated from country i with the same core productivity will have the same set of destination markets. This assumption simplifies the firm s problem and ensures model tractability. According to my specification of corporate taxes, the post-tax operating profit of firm ω 9 The effective tax rate is the average rate at which a firm is taxed. Devereux (2008) argues that the effective average tax rates play a role in firms choices of production sites. The data comes from Huizinga, Voget, and Wagner (204) and KPMG s Corporate and Indirect Tax Rate Survey which will be discussed in detail later. 0 China is an exception. In 2007, China s effective corporate tax rate for Chinese domestic firms was 33%. However, the effective corporate tax rate for foreign MNE affiliates operating in China was 20%. China subsidized foreign MNEs by lowering their corporate tax rates. After 2008, China s effective corporate tax rate is 25% for all firms operating in China. 9

10 Figure 2: The Timing of Events originated from country i to serve market n from its plant in country l is given by: π iln (ω) = ( κ il ) σ σ σ c iln (ω) σ X n P σ n. (8) Firm ω from country i will serve market n if and only if E (zl (ω)) [ N ] π iln (ω) w n f n. i= Now I specify the firm s problem in choosing its production sites. Equation (8) suggests that corporate taxation is equivalent to an increase in firms unit costs, the extend of which can be given by κ il = ( κ il ) σ. (9) Then a firm originated from country i will choose its production location to serve market n by minimizing its tax-adjusted unit cost: l(ω) = arg min k=,...,n { κ ik ξ ikn z k (ω) }. (0) Figure 3: Firm s Decision on Production Sites Figure 3 illustrates the firm s problem in Equation (0). It suggests that a firm decides its For simplicity, I assume that the fixed marketing cost is not tax-deductible. 0

11 production sites based on corporate tax rates, (κ il ), together with factor prices, productionsite-specific productivities, and trade and MP costs. Consequently, corporate taxes are key in shaping MNEs global production and thus income distribution across countries. 2.5 Aggregate Trade and MP Flows In this subsection, I solve the firm s choices of production sites and destination markets and aggregate these decisions to solve for the general equilibrium. Let ζ iln be the probability that a firm originated from country i serves country n by its affiliate at country l. Utilizing the property of Frechet distribution, I have ζ iln = A l(ξ iln κ il ) ɛ k A. () k(ξ ikn κ ik ) ɛ Notice that the corporate tax is an income tax which does not affect individual prices directly. In other words, the corporate tax affects a firm s choices for production sites, but not its prices conditional on production sites. As a result, the probability that a firm originated from country i will serve country n by its affiliate at country l is not equal to the share of pre-tax sales for firms from i to n through l. Let X iln be the pre-tax sales of firms from country i to country n from their affiliates in country l. Let X in be the total pre-tax sales of firms from country i to country n. Then ψ iln := X iln X in = ζ iln κ σ il N k= ζ iknκ σ ik = A l (ξ iln κ il ) ɛ κ σ il N k= A. (2) k(ξ ikn κ ik ) ɛ κ σ ik Rewriting Equation (), I have ζ iln = as in Arkolakis et al. (207), then ψ iln = ζ iln. ψ ilnκ σ il N k= ψ iknκ σ ik. Notice that if κ il = for all (i, l) Similarly, the corporate tax affects the expected unit cost of firm ω from country i to serve market n only through its choice of production sites. For convenience, I define Φ in = [ N k= ] ɛ A k (ξ ikn κ ik ) ɛ, Ψ in = N k= ζ ikn κ σ ik, (3) where Φ in is the the post-tax expected cost of firms from country i to serve country n, and Ψ in adjusts the corporate taxes. The expected unit cost of firm ω originated from country i

12 to serve market n is given by where γ = [ Γ ( ɛ (σ ) ɛ c in (ω) = E z (min{c iln (ω)}) = γ l ϕ i (ω) Φ inψ σ in, (4) )] σ and Γ(.) is the gamma function. The pre-tax sales of firms originated from i to market n, X in, is determined by their pre-tax expected costs: λ in := X in X n = M i T i Φ θ in Ψ in N h= M ht h Φ θ hn Ψ, (5) hn where M i is the mass of firms in country i. Intuitively, the sales of firms originating from country i in country n increases with respect to the mass of firms in country i and decreases with respect to the expected unit serving costs. The price index in country n is determined by the expected unit serving costs from all home countries: P θ n θ (σ ) = θσ σ θ γ θ σ θ (σ ) [ wn F n X n ] θ (σ ) σ N i= M i T i Φ θ in Ψ in. (6) 2.6 General Equilibrium In this subsection, I define the general equilibrium in my model. The aggregation in the previous subsection delivers the following expression for the pre-tax tri-lateral trade flows: X iln = ψ iln λ in X n. Moreover, monopolistic competition and Pareto distribution of core productivity imply that the total fixed marketing cost associated with the sales of firms originated from country i to market n equals to δ X in Ψ in where δ = θ (σ ). θσ The incomes and expenditure can then be expressed by aggregate trade and MP flows. First, the wage income for workers in country i can be expressed as ( w i L p i = ) X kin + δ σ k,n k X ki Ψ ki. (7) Entrepreneurs in country i own firms originated from country i. 2 The entrepreneurial 2 In reality, multinational firms may be owned by investors from foreign countries. However, the assumption here is a good approximation due to the home bias of portfolio holdings. 2

13 income comes from firms post-tax profits: w e i L e i := M i w e i f e = l,n [ ] X in σ κ σ il X iln δζ iln, (8) Ψ in where the first equality comes from the definition of entrepreneurial incomes and the second equality comes from the free entry condition. Total income of an agent equals her wage income plus tax revenues transfered by the government. I assume that corporate tax revenue are evenly allocated to domestic workers through lump-sum transfers. The aggregate expenditure in country i is then: X i = w i L p i + we i L e i + Λ i, (9) where Λ i := k,n σ ( ) κ σ ki Xkin (20) is the tax revenue in country i. Finally, I solve labor allocation within country. Agents choose to work in the position that gives them the highest utility. Therefore, L e i = B e i (W e i ) µ B e i (W e i )µ + B p i (W p i )µ L i, L p i = L i L e i, (2) where W j i is the real income of an agent with position j {e, p} in country i: W e i = we i + Λ i /L i, W p i = w i + Λ i /L i. (22) P i P i Equation (2) suggests that the labor mobility between production workers and entrepreneurs is determined by the shape parameter of amenity distribution, µ. Labor allocation is more responsive to changes in real income if µ is larger. This can be seen more clearly from two extremes: when µ = 0, L p i and L e i are exogenously given, whereas when µ, agents can freely move across positions so that W p i I then define the general equilibrium as follows: = W e i for all i. Definition Given (T i, L i, A i, τ ln, γ il ; f e, B e i, B p i ; θ, ɛ, σ, µ) and (κ il), the equilibrium consists of (w i, w e i, P i, X i, L p i, Le i, M i ) such that. w i satisfies labor market clearing condition (7). 3

14 2. X i is given by the current account balance condition (9). 3. wi e and M i is determined by free entry condition (8). 4. L i and L e i are determined by labor allocation in Equation (2) and (22). 5. P i is given by the price equation (6). 2.7 The Government s Objective Function In this subsection, I specify the government s objective function to characterize its incentives of setting corporate tax rates. Following the literature of optimal commercial policies, 3 I assume that the government maximizes the welfare of agents in its own country. The welfare measure in my model is straightforward: since the idiosyncratic amenity shocks are drawn independently from a Frechet distribution, the expected utility of an agent in country i before observing her amenity shocks can be expressed as U i = [B e i (W e i ) µ + B p i (W p i )µ ] µ. (23) In other words, although production workers and entrepreneurs may have different real income ex post, all agents within a country have the same expected utility before the realization of amenity shocks. This ex ante utility is a natural welfare measure. As mentioned above, the policy tools for country l s government are its corporate tax rates, ( κ il ) N i=, for firms from all countries, or, equivalently, (κ il ) defined by Equation (9). Therefore, the problem of country l s government is to maximize U l by choosing (κ il ). Notice, however, that U l does not only depend on (κ il ) but also on other countries corporate tax rates. To define the government s problem rigorously, I need to specify the governments interaction. In my quantitative analysis below, I examine three extreme scenarios that can shed light on tax competition and coordination in reality. First, I consider the case in which a government decides its welfare-maximizing corporate tax rates, taking other countries corporate tax rates as given. This unilaterally optimal tax scheme characterizes the government s incentives underlying its factual tax system. Second, I consider the Nash equilibrium in which countries retaliate optimally with each other. This case resembles a tax war in which all tax treaties are abandoned. Finally, I consider tax negotiations starting at factual taxes 3 See, for example, Ossa (204, 206) 4

15 following a Nash bargaining protocol. This case follows the specification of trade talks in Ossa (204) and resembles fully efficient tax negotiations. 2.8 Tax Avoidance and Evasion In my model, I assume that corporate taxes only affect firms real activities, i.e. their global production and sales. This assumption is in line with the workhorse models of international tax competition which include but not limited to Zodrow and Mieszkowski (986) and Wilson (986). However, as summarized by Keen and Konrad (203), firms can shift their profits to countries with low corporate tax rates by transfer pricing, by financial structuring, or by judicious choice of organizational form. How do tax avoidance and evasion affect the way in which we should think about international tax competition and coordination? In the appendix, I extend my model to incorporate MNEs profit-shifting behaviors in a two-country world. The model shows that the fraction of shifted profits is positively correlated with the tax difference but negatively correlated with the profit-shifting cost. Moreover, MNEs profit-shifting has no direct impacts on workers wage; but it affects welfare through its impacts on tax revenues and firm entry. The quantitative importance of profit-shifting depends crucially on the institutional costs of shifting profits across borders. The main reason why I do not incorporate profit-shifting in the baseline framework of my quantitative analysis is the lack of data that can be used to calibrate profit-shifting costs across countries. 4 A minor reason is the difficulty in extending the two-country model with profit-shifting into a multi-country model. I leave these extensions to future work. 3 Characterizing the Equilibrium In this section, I characterize how a welfare-maximizing government would set its corporate tax rates in a multi-country world. I restrict my analysis to special cases than can be explored analytically or through simple numerical examples. These cases illustrate how, in the presence of MP, terms of trade and home market effects shape governments incentives for manipulating corporate taxes. They also shed light on the basic forces behind my quantitative analysis in Section 6 and Section 7. Moreover, I derive the system of equations that can be used to compute the changes in equilibrium outcomes with respect to changes 4 To calibrate these costs, I need data on the magnitudes of profits shifted by transfer pricing and other methods for each pair of countries. 5

16 in corporate tax rates. 3. Special Cases 3.. Optimal Tax under Autarky It is instructive to consider the case in which trade and MP costs are infinite, i.e. γ il for all i l and τ ln for all l n. In this case, the welfare-maximizing corporate tax can be given as: Proposition 2 (Autarky) Let κ be the corporate tax rate that maximizes the welfare of a closed-economy. Suppose that µ and B p = B e. Then κ = 0. Proposition 2 indicates that in the closed economy where workers are homogeneous, the welfare-maximizing government should not collect corporate taxes. This is not surprising since Dixit and Stiglitz (977) shows that this monopolistic competition equilibrium is constrained efficient. Therefore, to rationalize non-zero corporate taxes in this setting, it must be (i) in an open economy where a country manipulates its corporate taxes and gains at the expenses of others, or (ii) with imperfect labor mobility. Both cases will be discussed later in my quantitative assessment A Two-Country Example I now discuss the welfare-maximizing corporate taxes in a two-country world. Suppose that i =, 2, τ ln = τ, and γ il = γ for all i l. For simplicity, I assume that µ, σ θ, and B p i = Bi e = L i = T i = A i = for all i. The welfare, as mentioned above, is measured by the real income per capita, W i := X i /P i. Suppose that τ = γ =. I consider optimal corporate taxes of country, (κ, κ 2 ), given country 2 does not collect corporate taxes. The following result shows that country can manipulate the terms of trade and home market effects via corporate taxes: Lemma 3 Suppose that i =, 2, τ ln = γ il = for all (i, l, n), µ, σ θ, and B p i = Bi e = L i = T i = A i = for all i. Suppose that κ 2 = κ 22 =. Normalize w =. If in equilibrium min{m, M 2 } > 0, then. Terms-of-trade: w 2 κ 2 < 0, w 2 κ > 0. 6

17 2. Home-market effect: M κ < 0, M κ 2 > 0 and M 2 κ > 0, M 2 κ 2 < 0. The first result in Lemma 3 suggests that the wage in country 2 decreases with country s corporate tax rates for firms from country 2, and increases with country s corporate tax rates for domestic firms. By raising corporate taxes for foreign MNE affiliates and reducing corporate taxes for domestic firms, country increases its relative wage and, due to the constant markup, improves its terms of trade. The second result in Lemma 3 shows that by raising corporate taxes for foreign MNE affiliates and reducing corporate taxes for domestic firms, country attracts firm relocation and gains from home market effects. This welfare gain would be magnified when τ > since consumers can gain from the increase in firm masses through the access of cheaper varieties. Motivated by the terms-of-trade and home market effects in Lemma 3, the optimal corporate taxes in my two-country example can be characterized by the following proposition: Proposition 4 (Optimal Corporate Taxes) Suppose that i =, 2, τ ln = γ il = for all (i, l, n), µ, σ θ, and B p i = B e i = L i = T i = A i = for all i. Suppose that κ 2 = κ 22 =. Let W (κ, κ 2 ) be the welfare of country under (κ, κ 2 ) and (κ, κ 2) = arg max κ,κ 2 W (κ, κ 2 ). Then I have. If I impose that κ = κ 2, i.e. country has to set a uniform corporate tax rate for domestic and foreign firms, then κ := κ = κ 2 >. 2. For each κ = κ 2 > under which min{m, M 2 } > 0, W (κ,κ 2 ) κ < 0. Proposition 4 implies that if country has to set a uniform corporate tax rate for all firms, then its optimal corporate tax rate is strictly positive. Unlike under autarky, in an open economy zero corporate tax is no longer optimal. The government can gain from collecting positive corporate taxes at the expense of other countries. Moreover, the second result in Proposition 4 suggests that a country has incentives to impose higher corporate tax rates for foreign MNE affiliates than for domestic firms. This is consistent with the term-of-trade and home market effects characterized by Lemma 3. For γ > or τ >, I do not have analytical characterization of optimal taxes. Therefore, I consider the following numerical example: θ = σ = 4, ɛ = 6, and γ = τ =.5. Again, I assume that country 2 does not collect corporate taxes. Figure 4 depicts the optimal corporate tax rates with respect to γ, τ, ɛ, and θ. In each case, I change one parameter and keep other parameters at their baseline levels. This exercise characterizes how optimal tax rates change with respect to trade and MP frictions and other key model parameters. 7

18 (a) γ (b) τ (c) ɛ (d) θ Figure 4: Optimal Corporate Taxes of Country (Notes: the optimal corporate tax rates of country refer to the corporate tax rates of country that maximizes country s welfare, given country 2 has zero corporate taxes. The solid line represents the optimal tax rate of country for domestic firms, whereas the dash line represents the optimal tax rate of country for foreign firms.) 8

19 Optimal Taxes w.r.t. MP and trade costs: Panel (a) of Figure 4 shows that the optimal corporate taxes are decreasing with respect to MP cost γ. As γ decreases, country 2 s firms and production workers will bear more corporate taxes of country due to MP expansion. As a result, country benefits from raising corporate tax rates for both foreign and dometic firms. In contrast, the optimal corporate taxes are increasing with respect to trade cost τ. This is intuitive through the perspective of proximity-concentration trade-off. Higher trade costs will stimulate MP and thus mitigate the corporate tax burden to foreign firms and production workers. Optimal Taxes w.r.t. ɛ: Panel (c) of Figure 4 shows that the optimal corporate tax rate for foreign firms is decreasing with respect to the dispersion of productivities across production sites, ɛ. Notice that ɛ governs how firms production locations respond to changes in corporate tax rates. Large ɛ implies that a small change in corporate tax rates will lead to substantial relocation of firms production sites. As a result, countries have incentives to lower the tax rates for foreign MNEs in order to prevent their relocation. In other words, when firms production sites are very responsive to tax changes, countries compete head-tohead in attracting MNEs. Optimal Taxes w.r.t. θ: Panel (d) of Figure 4 shows the optimal corporate tax rate for domestic firms decreases in the dispersion of core productivity, θ. When θ gets larger, domestic firm entry would be more responsive to tax cuts since the distribution of core productivity is more concentrated. As a result, country gains more from boosting firm entry by lowering its corporate tax for domestic firms. 3.2 General Equilibrium Effects of Corporate Taxation Due to large number of parameters, it would be difficult the calibrate the entire general equilibrium model. However, Dekle, Eaton, and Kortum (2007) derive an algorithm that can compute the changes of equilibrium outcomes from the observed trade flows, trade elasticity, and parameter changes in interest. studies. It is now the standard device in quantitative trade In this paper, we are interested in how much changes in (κ il ), associated with changes in (γ il ) and (τ ln ), lead to changes in (U i ). Suppose I have the values of (ɛ, θ, σ, µ, α i ) in hand and observe (ψ iln, λ in, κ il, X n, Lp i L i ). Let y be the level of any variable y after change and ŷ := y /y. Then based on the equilibrium system in Definition, I can solve the changes in equilibrium outcomes, (ŵ i, ŵ e i, ˆX i, ˆL p i, ˆP i, ˆM i ), with respect to policy changes, (ˆκ il ). The details of this exact-hat algebra is presented in the appendix. 9

20 4 Empirical Estimates and Calibration In this section, I bring my model to the world economy in 2007 with 40 major countries. 5 The parameters needed for quantitative assessment include elasticities (θ, ɛ, µ, σ), trilateral trade flows (X iln ), and corporate income tax rates (κ il ). My empirical implementation is arranged as the following. First, I calibrate (θ, µ, σ) from the literature. Then I estimate ɛ from the equilibrium conditions. Finally, I calibrate unobserved trilateral flows X iln from bilateral trade and MP flows. The calibration of (σ, θ, µ) is summarized in Table. As shown in Section 3, the shape parameter of core productivities, θ, is crucial for the welfare implications of corporate taxes. In addition to the baseline calibration reported in Table, I test the sensitivity of my results with respect to θ in the quantitative analysis. Parameter Table : Calibration of (σ, θ, µ) Source σ = 4 Profit share. Arkolakis et al. (207) θ = 4 Unrestricted gravity equation. Arkolakis et al. (207) µ = 3 Labor mobility w.r.t. real income. Hsieh et al. (206) 4. Structural Estimates for ɛ The dispersion of productivities across production sites, ɛ, characterizes to what extent changes in corporate tax rates affect MNEs production sites. A large ɛ implies that a small change in trade or MP cost can lead to substantial production relocation of MNEs. As shown in Section 3, the welfare impacts of corporate taxes are sensitive with respect to ɛ. ɛ can be estimated using the structural gravity equation delivered by my model. From Equation (2), I have X MP il := n { X iln = κ ɛ (σ ) il γ ɛ il } {Al } { k,n w X } ɛ ikn l k,n A k(ξ ikn κ ik ) ɛ κ σ. (24) ik Taking logs on both sides of Equation (24) and expressing iceberg MP costs in terms of 5 These economies are Australia, Austria, Belgium, Bulgaria, Brazil, Canada, China, Cyprus, Czech, Germany, Spain, Estonia, Finland, France, UK, Green, Hungary, Indonesia, India, Ireland, Italy, Japan, Korea, Lithuania, Luxembourg, Latvia, Mexico, Malta, Netherlands, Poland, Portugal, Romania, Russia, Slovakia, Slovenia, Sweden, Turkey, Taiwan, and US. 20

21 gravity variables, I have the following equation for estimating ɛ: log X MP il = ɛ (σ ) σ log( κ il ) ɛδ γ dist log(dist il) ɛδ γ gravgrav il + δ s i + δ h l + u il, (25) where dist il is the physical distance between the source country i and host country l, grav il are gravity dummy variables such as contiguity, common language, and common legal system, δ s i is the source-country fixed effect, δ h i is the host-country fixed effect, and u il is a measurement error. The identification of ɛ comes from the variation of bilateral MP sales with respect to bilateral corporate tax rates, controlling for other bilateral MP barriers and source- and host-country fixed effects. To estimate ɛ using Equation (25), I employ data on bilateral MP flows and corporate tax rates. The bilateral MP flows come from Ramondo, Rodriguez- Clare, and Tintelnot (205). They combine the bilateral MP sales from UNCTAD with M&A data from Thomson and Reuters to impute MP sales across major economies. Their data is the average level over The bilateral corporate tax rates come from Huizinga, Voget, and Wagner (204). They collect effective average corporate tax rates across major OECD countries over For countries that are not included in Huizinga, Voget, and Wagner (204), I take the effective average corporate tax rates from KPMG s Corporate and Indirect Tax Rate Survey. To estimate ɛ, I utilize the corporate tax rates in 200. The summary statistics are presented in the appendix. The results in Table 2 suggest that corporate taxes do matter to MP sales: one-percentagepoint increase in bilateral corporate tax rate would reduce the bilateral MP sales by.42%. Moreover, the bilateral MP flow will decrease with respect to physical distance but increase if the source and host countries share common borders or common legal origins. The coefficients of the gravity terms are in line with the estimates of gravity equations for MP flows. 6 Table 2 suggests that ɛ (σ ) =.42. Since σ = 4, I have ɛ = This estimate is σ comparable to the coefficient estimated by the restricted gravity equation in Arkolakis et al. (207). Using the multinational sales of the U.S. MNEs and tariff data, they estimate the elasticity of firm production entry with respect to iceberg MP costs as For example, Antras and Yeaple (204) documents that a % increase in distance is associated with 0.57% fall in bilateral affiliate sales. 7 See Arkolakis et al. (207) for details. 8.4 is their estimate using Poisson PML estimator. OLS leads to the estimate as

22 Table 2: Estimating the Gravity Equation of MP Flows Dependent Variable: log X MP il log( κ il ).42* (.80) log(dist il ) -.85*** (.09) Contiguity.37** (.7) Common Language. (.5) Common legal origin.84*** (.0) Source f.e. Host f.e. R-squared 0.89 N. of Obs. 678 (Notes: the gravity equation is estimated by the fixed-effect regression. i refers to the origin country and l refers to the host country.) 4.2 Calibrating Trilateral Flows from Bilateral Flows As discussed above, conducting counterfactual analysis using the exact-hat algebra requires data on trilateral trade flows (X iln ). Since (X iln ) is not directly observed in the data, it has to be computed from the observed bilateral trade and MP volumes through the lens of my model. Let T il = (M i T i ) θ A ɛ l γ il w l. Then X iln can be expressed as X iln = ψ iln λ in X n = h r [ T ɛ il τ ɛ ln κ [ɛ (σ )] il k ɛ T hr τ rn ɛ κ [ɛ (σ )] hr ] θ ɛ ɛ T ik τ ɛ kn κ ɛ ɛ ik [ k T ɛ hk τ ɛ kn κ ɛ hk ] θ ɛ ɛ X n. (26) Armed with parameters (θ, ɛ, σ), corporate tax rates ( κ il ), and total expenditure (X n ), I can compute X iln from ( T il, τ ln ) by Equation (26). By definition, trade volume from country l to n is Xln T R = i X iln and MP sales from country i to l is Xin MP = n X iln. Therefore, ( T il, τ ln ) can be calibrated by matching the model-generated bilateral trade and MP flows to the data. I use data on trade, MP, and corporate tax rates for 40 countries in I aggregate sectoral bilateral trade flows in WIOD 2007 to construct the aggregate bilateral trade flows across 40 countries. There are no direct observations on bilateral MP sales in I assume that the bilateral MP sales are proportional to the bilateral FDI stock from UNCTAD. 22

23 Ramondo, Rodriguez-Clare, and Tintelnot (205) have shown that their imputed MP flows have strong positive correlation with FDI stocks. Finally, I obtain bilateral corporate tax rates in 2007 from Huizinga, Voget, and Wagner (204) and KPMG s Corporate and Indirect Tax Rate Survey. The calibration strategy based on Equation (26) enables my model to exactly match the observed trade and MP flows. Therefore, to examine the model-fit, I compare the calibrated trade and MP costs with the observed barriers for trade and MP. Notice that τ ln is calibrated directly by solving Equation (26). Moreover, by assuming that MP costs are symmetric, I calibrate γ il by γ il = Til Tli T ii Tll. Figure 5 shows that the calibrated iceberg trade and MP costs are strongly increasing with physical distances across countries. Moreover, I regress the calibrated trade and MP costs on distance and dummies for common border and common language. The results in Table 3 shows that having common borders decreases trade costs substantially but has no significant impacts on MP costs. In contrast, using common language decreases MP costs greatly but has much less impacts on trade costs. The regression results in Table 3 are in line with the empirical regularities of bilateral MP flows summarized by Antras and Yeaple (204). Iceberg Trade Cost (in log) Distance (in log) Iceberg MP Cost (in log) Distance (in log) (a) τ (b) γ Figure 5: Model-Fit: Calibrated Trade and MP Costs vs. Distance (Notes: the iceberg trade and MP costs are calibrated by solving Equation (26). I normalize τ ii = γ ii =. The domestic trade and MP costs are excluded from the figure.) 5 Welfare and Distributional Effects of Corporate Taxes In this section, I examine the welfare and distributional effects of corporate taxes using the calibrated model. As discussed in Section 2, the welfare of country i is measured by the 23

24 Table 3: Model-Fit: Calibrated Trade and MP Costs vs. Gravity Variables Dependent Variables log(τ ln ) log(γ il ) Distance (log) 0.5*** 0.4*** (0.020) (0.0068) Contiguity -0.8* (0.097) (0.034) Common Language -0.2** -0.29*** (0.099) (0.034) R-squared N. of Obs Average Costs Standard Deviation (Notes: The domestic trade and MP costs are excluded.) expected utility of workers in country i before observing their amenity shocks in Equation (23). In addition to welfare, I also explore the impacts of corporate taxes on the income of entrepreneurs relative to production workers, measured by r i := W i e W p i, (27) where Wi e and W p i are defined by Equation (22). This measure of relative income helps us to understand whether a country is specialized in creating firms or production. It also depicts the income distribution between firm owners and workers, which is at the center of many policy debates. 5. Elimination of Corporate Taxes I first examine the welfare and distributional effects of eliminating corporate taxes. Figure 6 illustrates the consequences of eliminating corporate taxes. I consider two scenarios (i) eliminating all corporate taxes, and (ii) eliminating NRDW so that all firms operating in the same country face the same corporate tax rate. Panel (a) of Figure 6 shows that most countries in this exercise gain from the elimination of all corporate taxes. This is mainly due to the increase in firm entry and the expansion of MP. However, about half of the countries lose from eliminating NRDW due to the decline in tax revenue. Panel (b) shows that the elimination of all corporate taxes increases the incomes of entrepreneurs relative to production workers in all countries, whereas the elimination of NRDW has ambiguous effects 24