Global Treaties and Domestic Politics: Do Bilateral Investment Treaties Constrain Taxation in Developing Countries?

Abstract

Bilateral investment treaties (BITs) define a set of investors’ rights and allow foreign investors to sue countries for policies that breach them. Political scientists typically study the extent to which that chastens policymaking in the context of health and environmental rules. We extend that literature to tax policy. We argue and show that BITs limit developing country governments’ tax intake. We further argue that umbrella clauses are an important driver of these dynamics. Umbrella clauses extend BITs’ protection to side agreements that are often more constraining than the BITs themselves and are typical in developing country resource extraction projects. Umbrella clauses’ tax-inhibiting effects are significant because resource revenues are important to developing countries’ public finance systems. Any assessment of BIT’s developmental benefits should consider their fiscal impact alongside their potential to attract capital inflows.

Los Tratados Bilaterales de Inversión (TBI) definen un conjunto de derechos de los inversores y permiten a los inversores extranjeros demandar a los países por el uso de políticas que incumplan estos tratados. Los politólogos suelen estudiar hasta qué punto afecta esto a la formulación de políticas en el contexto de las normas sanitarias y ambientales. En este artículo, ampliamos esta literatura al ámbito de la política fiscal. Argumentamos y demostramos que los TBI limitan la recaudación de impuestos por parte de los Gobiernos de los países en desarrollo. Además, argumentamos que las cláusulas paraguas son un importante impulsor de esta dinámica. Las cláusulas paraguas amplían la protección que ofrecen los TBI a otros acuerdos paralelos, los cuales a menudo son más restrictivos que los propios TBI y son típicos de los proyectos de extracción de recursos que se llevan a cabo en los países en desarrollo. Los efectos inhibidores de impuestos de las cláusulas paraguas son significativos debido a que los ingresos procedentes de los recursos naturales resultan de gran importancia para los sistemas de finanzas públicas de los países en desarrollo. Cualquier evaluación que se lleve a cabo con respecto a los beneficios del desarrollo de los TBI debe considerar su impacto fiscal y no solo su potencial para atraer ingresos de capital.

Les traités bilatéraux d'investissement (TBI) définissent un ensemble de droits des investisseurs et permettent aux investisseurs étrangers de poursuivre des pays en justice quand des politiques les enfreignent. Généralement, les politologues étudient leur effet calmant sur l’élaboration de politiques dans le cadre des règles sanitaires et environnementales. Nous prolongeons cette littérature pour y intégrer la politique fiscale. Nous affirmons et montrons que les TBI limitent les recettes fiscales des gouvernements des pays en développement. Nous affirmons en outre que les clauses parapluie favorisent considérablement ces dynamiques. Elles élargissent la protection des TBI aux accords connexes, souvent plus contraignants que les TBI eux-mêmes, et se retrouvent généralement dans le développement de projets d'extraction de ressources nationales. Les effets limitatifs des clauses parapluie sur la fiscalité sont importants, car les recettes liées aux ressources sont déterminantes pour les systèmes financiers publics des pays en développement. Toute évaluation des avantages des TBI en termes de développement devrait tenir compte de leur effet fiscal ainsi que de leur attractivité potentielle pour les entrées de capitaux.

Introduction

The international political economy literature often sees bilateral investment treaties (BITs) as catalysts for global convergence toward investor-friendly policies.¹ BITs allow investors to sue governments for treaty nonconforming policies. Fear of costly arbitration can lead to “regulatory chill,” where governments do not implement otherwise desirable laws and regulations because they might run counter to BIT-related protections. These dynamics are widely written about in environmental and health policy.² The political science literature has written less about whether BITs deter corporate taxation.³ We suspect that this is because BITs are meant not to do so.⁴ BITs often have explicit tax carve-outs,⁵ and arbitration panels in tax-related challenges have been reluctant to equate tax policy changes with BIT-prohibited takings.⁶ As the tribunal in EnCana vs. Ecuador put it: “From the perspective of expropriation, taxation is in a special category. In principle a tax law creates a new legal liability on a class of persons to pay money to the State in respect of some defined class of transactions, the money to be used for public purposes. In itself such a law is not a taking of property” (p. 51).⁷

We argue and show that BITs do systematically deter taxation in developing countries. We consider two distinct causal pathways through which they might do so. Most directly, BITs plausibly deter taxation by allowing investors to claim a tax was applied in a discriminatory way or that it violated a right to fair and equitable treatment (FET). Firms do occasionally bring these cases and sometimes win them. Yet these types of claims are relatively rare, have a low success rate,⁸ and, as reflected in the EnCana decision quoted above, arbitration panels have generally held tax-related cases to an exceptionally high bar.⁹

This paper focuses on umbrella clauses as an alternative causal pathway through which BITs might deter taxation. Umbrella clauses are included in about 46 percent of BITs¹⁰ and are so-called because they extend the “umbrella” of treaty protection to other agreements that a firm makes with a government. This allows BIT-protected firms to elevate contract violations to treaty violations, even if the offending policy is otherwise BIT-compliant. Umbrella clauses are particularly relevant for developing country resource extraction projects, which are often protected by “stabilization agreements” that lock in terms of investment, including allowable levels of taxation.¹¹ Those stabilization agreements often lack access to investor-state dispute settlement (ISDS), such that in the absence of an umbrella clause, firms often pursue contract violations in local courts under local law,¹² which exposes firms to local politics and limits these contracts’ enforceability.¹³ Pursuing claims via BIT-provided ISDS avoids those complications.¹⁴ Pursuing contract violations as BIT violations can be preferable even when the contract gives access to ISDS. Doing so allows firms and their lawyers to bundle contract-based claims with a broader set of co-occurring treaty violations in the same arbitral setting. Bundling in this way is cheaper, allows firms to pressure host country governments more effectively, and reduces the possibility of inconsistencies arising from arbitrations across multiple forums.¹⁵Walde (2005, 200) and others have argued that governments consider treaty violations higher stakes commitments than firm-to-state contracts and, thus, a more effective deterrent to nonconforming policies. In this way, BITs do not deter taxation solely or even primarily because they deter or prohibit specific tax policies, but because firms can use their umbrella clauses to more effectively enforce contracts that do. That combination of factors plausibly disciplines governments by pushing contract disputes onto investor-favorable terms.¹⁶ Because those contracts (especially in developing countries) are likelier to contain tax-specific terms, umbrella clauses should have a particularly large effect on taxation in developing countries with substantial resource production.

We support our theory by analyzing tax revenue data from over 100¹⁷ developing countries between 1996 and 2020. We find that BITs deter taxation. In line with our proposed causal mechanism, these effects are larger and more precisely estimated in resource-producing countries whose BITs’ include formal umbrella clauses. The results are robust to various measures for our key variables, different control variable sets, and instrumental variable regression. The effects we identify are also substantively significant. For a country-year at the 75th percentile of oil production,¹⁸ increasing the number of BITs with umbrella clauses from the 25th to the 75th percentile (equivalent to moving from 2 to 10) reduces tax revenue as a percentage of GDP by 1.25 percentage points, which is roughly 50 percent of a standard deviation.¹⁹

Our results suggest that BITs’ economic impacts are more significant and multifaceted than is commonly appreciated. BITs’ developmental promise is that they help developing countries attract Foreign Direct Investment (FDI) that political risk might otherwise deter. However, BITs’ policy deterrent effects extend beyond regulatory freezes and into a de facto convergence around low resource taxation, potentially undermining their development bona fides. Developing countries often rely on resource taxation due to their limited ability to tax income or consumption.²⁰ A complete accounting of BITs’ effects should consider not just whether they attract FDI but whether they attract enough to compensate for their tax effects. In a more general sense, it suggests that we should not conceptualize BITs as a tradeoff between separable notions of “economic benefits” and “sovereignty costs.” In fact, BITs’ “sovereignty costs” extend far into the economic arena.

Finally, BIT’s impact on taxation is important precisely because it is often crafted and believed not to have such an impact. The constraints BITs place on policymakers are typically legitimated by governments submitting to them voluntarily, in what Dolowitz and Marsh (2000) call a “negotiated transfer” (p. 15) of regulatory authority.²¹ In that view, BITs are controversial less for the restrictions they place on policymakers as for the extent to which those restrictions exceed governments’ expectations. BITs’ effect on taxation significantly stretches the negotiated element of the negotiated transfer.²²

The remainder of the paper is as follows: The second section discusses BITs and our theory; the third section presents the data and research design; and fourth section shows the empirical results. The fifth section concludes.

BITs and Policy Constraints

BITs reduce FDI’s political risk by defining investor rights and allowing investors to protect those rights through ISDS.²³ BITs’ protections vary, but most guarantee treatment on par with domestic firms (National Treatment) or firms from other countries (Most Favored Nation). Most also define a minimum standard of treatment, including protections against direct and indirect expropriation and an often less clear requirement of FET.²⁴ BITs can play an important role in promoting FDI in developing countries, but the arbitration cases are often unpleasant. BIT-related challenges undermine states’ reputations as good investment locations.²⁵ The cases are typically expensive, and judgments can be unpredictable. That unpleasantness is partly why BITs are effective. The unpleasantness of being sued (regardless of the case’s outcome) dissuades governments from enacting policies that risk arbitration. BITs’ capacity to dissuade can lead to a regulatory chill, in which governments avoid otherwise meritorious policies for fear of legal consequences. The regulatory chill literature typically focuses on health and environmental policies where the extent of BITs’ protection is often unclear, and the government’s reluctance to act is particularly controversial.

We show that BITs can also deter taxation, which is a more substantial encroachment on state sovereignty. Bird-Pollan (2018, 119) argues that BITs’ encroachment into taxation threatens their social legitimacy because taxation “is at the very heart of what it means to govern autonomously.” That sensitivity seems to be appreciated by BIT designers, as many BITs go far out of their way to exclude tax policies.

The most straightforward mechanism through which BITs can encroach on taxation is through cases claiming that a particular tax was discriminatorily applied or violated a right to FET. Firms do bring these cases and sometimes win. Indeed, one of the more frequently written about cases—Micula vs. Romania—is an example of a successful FET-based challenge to a tax subsidy change. However, as Pelc (2017) and Kerner and Pelc (2022) show, FET-based challenges are often viewed as speculative and have a notably and increasingly low success rate. Moreover, the bar to establishing that a tax violates a firm’s treaty-defined rights is exceptionally high, for reasons reflected in the EnCana decision quoted in the introduction. Arbitration panels have dismissed many cases brought over claims of excessive taxation, even to the point in Feldman vs. Mexico of taxes so high that they put the claimant out of business.²⁶ Firms can challenge tax laws that are applied discriminatorily, but taxes, even high, unexpected taxes, are otherwise typically BIT-compliant.

BITs more directly shape tax outcomes through umbrella clauses that lend the force of the treaty to side agreements that, unlike BITs, often contain provisions that define allowable levels of taxation without reference to whether governments apply those taxes in a discriminatory or nondiscriminatory manner. Firms with umbrella clause-protected contracts can pursue violations as treaty violations rather than through whatever dispute settlement procedure the contract establishes.

Umbrella clauses do not grant firms the right to elevate claims to treaty violations, but tax-related claims often can. Questions about umbrella clauses’ applicability typically revolve around the clauses’ scope and whether the alleged breach arises from the exercise of sovereign rather than commercial power.²⁷ Umbrella clauses’ scope varies as a matter of the text itself and in panels interpretation of them.²⁸ The umbrella clause in most older US BITs is notably general: “Each Party shall observe any obligation it may have entered into with regard to investments.”²⁹ That language opens states to expansive interpretations of umbrella clauses’ applicability. Some have argued that umbrella clauses create a “mirror effect” through which any breach of any obligation is necessarily a treaty violation. The tribunal in Eureko B.V. vs. Poland maximally interpreted the Netherlands-Poland BIT’s umbrella clause.³⁰ They write that “The plain meaning—the “ordinary meaning”—of a provision prescribing that a State “shall observe any obligations it may have entered into” with regard to certain foreign investments is not obscure. The phrase, “shall observe” is imperative and categorical. “Any” obligations is capacious; it means not only obligations of a certain type, but “any”—that is to say, all—obligations entered into with regard to investments of investors of the other Contracting Party.”³¹

Other umbrella clauses are more precise, and the trend has been toward more precision over time.³² For example, the umbrella clause included in the 2001 Mexico–Austria BIT specifies that “obligations” mean “written contracts:”³³ “Each Contracting Party shall observe any other obligation it has assumed in writing, with regard to investments in its territory by investors of the other Contracting Party. Disputes arising from such obligations shall be settled under the terms of the contracts underlying the obligations.”³⁴ The current US model BIT so precisely limits its umbrella clause (Article 24) that UNCTAD no longer classifies it as such, even though it continues to protect contracts covering resource investments.³⁵ Several panels have also interpreted umbrella clauses in more limited ways. The tribunal in SGS vs. Pakistan (2003)³⁶ finds that the “mirror effect” interpretation was unworkably “far-reaching in scope.”³⁷  El Paso Energy International Company vs. The Argentine Republic suggests that it undermines state sovereignty and that extending umbrella clauses’ reach into “an ordinary commercial contract entered into by the State or a State-owned entity … ” would have “far-reaching consequences … quite destructive of the distinction between national legal orders and the international legal order.”³⁸

Notably, none of the debates around umbrella clauses’ applicability suggest that they would not protect stabilization agreements. Stabilization agreements that define allowable levels of taxation are formal contracts in which the state is acting in a sovereign capacity. Even El Paso finds that umbrella clauses “will cover additional investment protections contractually agreed by the State as a sovereign—such as a stabilization clause—inserted in an investment agreement.”³⁹ Thus, even in the more minimal interpretations, umbrella clauses are powerful tools for investors to challenge contractual breaches, including those related to taxation. The underlying issues surrounding stabilization agreements’ enforceability through umbrella clauses are relatively cut and dry, especially compared to tax-related cases brought under FET standards or claims of legitimate expectations. For that reason, umbrella clauses are likely a powerful deterrent that can dissuade or limit resource taxation in the first place.

Stabilization agreements are widespread in resource extraction.⁴⁰ Resource extraction often involves direct contracting with the state and significant upfront capital allocations to be recouped slowly over time. The revenues themselves are often politically fraught. Resource revenues’ centrality to public sector finance means that delegating production decisions to private firms can become politically untenable.⁴¹ Moreover, resource extraction can generate periods of windfall profits from resources that may be perceived as public, even if private firms own the exploitation rights. Such investments pose substantial political risks, and investors use stabilization agreements to mitigate that risk by locking in the terms of their investment.

The most severe forms of stabilization agreements are “freezing clauses” that prohibit states from applying any legislative changes to the covered investment project. The most stringent freezing clauses apply whether or not policy changes carry any form of discriminatory intent or consequence.⁴² These clauses are prevalent in resource extraction contracts with developing country governments. Freezing clauses in contracts with developed-country governments often allow more leeway if new laws are applied generally and are nondiscriminatory. Less severe stabilization agreements often take the form of “economic equilibrium clauses,” which only require compensation for compliance costs associated with new legislation.⁴³ We are unaware of a comprehensive database of such agreements or their terms. However, Shemberg’s (2009) study of 88 stabilization agreements indicates that tax-related carve-outs were rare and entirely limited to OECD host countries.⁴⁴

The above is familiar information; firms have been using umbrella clauses to enforce the tax-related terms of the stabilization agreements for many years.⁴⁵ However, political scientists underappreciate the impact of these dynamics on BITs' capacities as a policy deterrent. BITs’ terms might not apply directly to (nondiscriminatory forms of) taxation, but firms can use their umbrella clauses to enforce side contracts that do. That makes BITs’ potential to affect taxation especially notable in developing countries with substantial resource extraction projects. This expands the policy area where we expect BIT-catalyzed convergence toward corporate-friendly policies.⁴⁶ If BITs affect taxation by amplifying the consequences of stabilization agreements, developing country governments should collect less taxes to the extent BITs are in place. Those effects should be especially large where and when a large portion of theoretically taxable revenue accrues in the resource extraction sector.

The above motivates two hypotheses. The first is that BITs should have an unconditionally negative effect on taxation to the extent that firms’ capacity to challenge taxes as FET or similar violations acts as a deterrent. The second is that BITs should reduce tax intake when those BITs include umbrella clauses and when the host economy produces revenues in industries where stabilization agreements with tax provisions are standard.

 

Our focus on umbrella clauses’ capacity to limit taxation in sectors of the economy where stabilization clauses are standard has a significant advantage: It is a relatively specific causal story that we can identify in the data. In contrast, support for H1 could indicate support for our theory or that BITs affect taxes through noncontingent clauses, or it could capture the consequences of shifting economic activity into multinational corporations (MNCs), where it can more easily escape taxation.⁴⁷ Whether BITs’ impact on taxation applies broadly or in the narrower terms we have theorized is ultimately an empirical question we address below.

Data, Measurement, and Research Design

We test our hypotheses using data from between 117 and 154 developing countries,⁴⁸ observed between 1996 and 2020. We focus on post-1996 data because the first tax-related ISDS case stemming from BIT commitments occurred in 1995 (Goetz vs. Burundi). Therefore, our entire sample exists in a world where the threat to bring such a case is credible.⁴⁹

Dependent Variable

Our dependent variable measures tax revenue (excluding social contributions) as measured by the Government Revenue Dataset.⁵⁰ We measure the tax revenue data as a percentage of GDP.⁵¹ While the bulk of the variation in both dependent variables occurs across units, there is sufficient variation within units to support our fixed effects analyses.⁵²

Independent Variables

Our key independent variable is the cumulative number of BITs in force that include ISDS provisions. We construct our BITs measure using UNCTAD’s International Investment Agreements database.⁵³ We supplement that dataset by coding an additional 665 treaties.⁵⁴ The variable ranges from 0 to 111 in our primary sample, with an average of 15.⁵⁵ A general BIT measure characterizes the impacts of the BIT-ISDS regime. However, hypothesis two more specifically implicates BITs with umbrella clauses. We identify those using the data in the “Mapping of IIA Content” at the Investment Policy Hub of UNCTAD. The average country-year is a party to about six BITs with a formal umbrella clause and ten BITs without one.⁵⁶

We proxy for taxable resource revenue using a country’s total oil income per capita, taken from Haber (2023).⁵⁷ We rescale that variable to represent $1,000 of oil income.⁵⁸ As with our tax revenue measure, oil income mainly varies cross-sectionally. However, this measure is sensitive to price and production fluctuations, which generate ample within-unit variation that we can leverage to test our theory in a fixed-effects context.⁵⁹

Control Variables

We report three sets of models: one with no control variables, one with political controls, and a third with additional macroeconomic controls. This strategy balances concerns about omitted variable bias and posttreatment variable bias.

Our “political” controls account for the possibility that BITs identify governments that are, for separate reasons, inclined to ratify more BITs and tax less, both of which are at least vaguely in line with a country’s commitment to Washington Consensus policies. These additional controls are a binary indicator of whether a country has an active IMF arrangement for at least five months of a particular year⁶⁰ and a measure of capital account openness.⁶¹ We also include controls that account for the possibility that BITs and tax policies are jointly determined by state capacity.⁶² Underdeveloped institutions may signal to investors that a country poses a political risk and incentivize governments to sign BITs. Underdeveloped institutions may also limit governments' ability to tax. We include a measure of public sector corruption⁶³ and an indicator of judicial independence.⁶⁴ We also control for (the log of) a country’s population⁶⁵ and whether a country is a party to the Energy Charter Treaty.

Our “economic” controls account for potentially confounding macroeconomic effects. We include these controls in separate regressions because they raise the possibility of posttreatment bias.⁶⁶ Those variables are the log of GDP per capita in constant US dollars;⁶⁷ imports as a share of Gross Domestic Product;⁶⁸ per capita GDP growth⁶⁹ and (logged) FDI stock as a percentage of GDP.⁷⁰ The results of our three models—no controls, political controls, and political and macroeconomic controls—all yield similar results. Oster’s (2019) test for sensitivity to omitted variables finds no evidence of it.

We capture the effects of time-invariant factors using country-fixed effects. We model time using a linear time trend and its square.⁷¹ We lagged all independent variables by one year. We estimate our models using OLS, with Driscoll–Kraay standard errors to adjust for autocorrelation.⁷²

Our main empirical models take the following form:

EQ 1: Tax revenue (percent GDP)_{i, t} = α₁ + α₂BITs_{i, t-1} + Χ_i,t-1 + time + time² + country fixed effects + ε_t

EQ 2: Tax revenue (percent GDP)_{i, t} = α₁ + α₂BITs_{i, t-1} + α₃BITs_{i, t-1}*Oil Income_{i, t-1} + Χ_i,t-1 + time + time² + country fixed effects + ε_t

H1 suggests that α₂ should be negative in EQ1, i.e., BITs reduce tax revenue. H2 suggests a negative estimate of α₃, indicating that BITs’ should limit taxation more in countries with more oil income.

Empirical Results and Discussion

BITs’ Impact on Tax Revenue

Our first set of models—Models 1–3 in table 1—considers the unconditional relationship between BITs and rax revenue. Model 1 omits control variables beyond the time trends and country-fixed effects. Model 2 introduces our political controls, and Model 3 adds our macroeconomic control variables. All three models support H1. The coefficient estimates for our BITs variable are consistently negative and statistically significant. Tax revenue declines as governments sign more BITs.

The negative correlation of BITs with tax revenue in these models is consistent with our theory but not dispositive. BITs could, for example, reduce tax intake by shifting economic activity to MNCs, where it can more easily escape taxation. H2, which suggests that BITs should reduce tax revenues more when they contain umbrella clauses and when resource income increases, tests our theory better. Models 4–12 in table 2 test this hypothesis.

Models 4–6 interact BITs (regardless of whether they contain an umbrella clause) with oil income. As before, Models 4–6 differ only in the progressive introduction of more control variables. All three models produce similar results. The estimated coefficient on BITs is always negative and always statistically significant. That suggests that BITs correlate with less tax intake even when oil income equals 0, which is the case for roughly 50 percent of the sample. More significantly, the interaction terms between BITs and oil income are also consistently negative and statistically significant, implying that BITs’ negative effect on tax intake increases as countries produce more oil. Models 7–12 replicate Models 1–6 but focus only on BITs with umbrella clauses. As expected, focusing on just those treaties reveals a larger negative relationship between BITs and tax intake. The coefficient on BITs is consistently negative and statistically significant in Models 7–10. The coefficients on BITs with umbrella clauses and their interaction with oil income are substantially larger than in models that considered BITs generally.

We represent these results graphically on the top row of figure 1. Graphs 1a (which relates to Model 5) and 1b (which relates to Model 11) show the estimated conditional effects of BITs on tax revenue with 95 percent confidence intervals across the observed range of oil income. The rug plot at the bottom of the graphs indicates the density of observations at different values of oil income. BITs’ estimated effect on tax intake is consistently negative and statistically significant, and in both cases, it is increasing as oil income increases. These figures show that BITs’ effect on tax intake is exaggerated at high levels of oil income when considering BITs with umbrella clauses. The 95 percent confidence intervals are wide at the highest levels of oil income, where data is relatively sparse, but the differences between these estimates are always apparent. Consider the differences between BITs’ effect on taxation in a non-oil-producing country-year and the average oil-producing country-year, which in this sample produces about $520 of oil income per capita, corresponding to 0.52 on the X-axis of these graphs. Model 5 indicates that BITs’ effect on tax intake is 50 percent larger for the average oil-producing country-year than for a country-year with no oil income. Model 11 suggests that BITs with umbrella clauses’ effect on tax intake is 136 percent larger for the average oil-producing country-year than for a country-year with no oil.

Table 2 shows the results of several robustness checks. Models 13–15 consider that our theory should be sensitive to the amount of BIT-protected revenues generated in a country. Intuitively, BITs’ capacity to limit taxation should affect tax intake only to the extent that MNCs operate in a host state and that there are thus BIT-protected revenues to tax. Governments’ heightened ability to tax revenues that do not exist is purely theoretical and should not affect outcomes. If BITs’ impact on tax intake was insensitive to MNCs’ economic presence, it would cast doubt on our explanation.⁷³ We test that proposition by interacting our measure of BITs with (logged) FDI stock as a percentage of GDP, using data taken from UNCTAD.⁷⁴ As anticipated, Models 13–15 estimate a negative interaction term between BITs and the FDI stock, indicating that BITs’ negative relationship to tax intake is more significant where MNCs are most present.

The bottom row of figure 1 graphically represents these results (and all other results in table 2). Graph 1c plots BITs’ estimated effect on tax intake (and a 95 percent confidence interval) across the observed FDI stock/GDP range. Graph 1c suggests that BITs’ have a statistically insignificant and negative effect on taxes for country-years beyond the 20th percentile of FDI/GDP. That effect increases dramatically as FDI/GDP increases. BITs’ tax-inhibiting effects are 137 percent larger for a country-year at the 75th percentile of FDI/GDP than for a country-year at the 25th percentile of FDI/GDP. BITs appear to counterintuitively increase tax intake at very low levels of FDI/GDP, but subsequent models suggest that this is a modeling artifact of FDI/GDP's linear operationalization.

Models 16–18 in table 2 further explore the differences between BITS with and without umbrella clauses. H2 suggests that our theorized relationship should be stronger when BITs have umbrella clauses. This is particularly true for our hypothesized interactive effect, which should be exclusive to BITs with umbrella clauses. Models 16–18 test that claim directly by separating our BITs variable into BITs that include an umbrella clause and BITs that do not. We interact each of those variables with our measure of oil income and expect that the interaction should be statistically significant and negative for BITs with umbrella clauses but insignificant for BITs without umbrella clauses. That is what we find. The coefficient on BITs without umbrella clauses is consistently negative, suggesting that BITs correlate with low taxation. The interaction between oil income and BITs without umbrella clauses is never statistically significant, inconsistently signed, and always very close to 0. There is no evidence that BITs without umbrella clauses’ effect on taxes increases with oil income. However, the interaction between oil income and BITs with umbrella clauses is consistently negative and statistically significant. These findings strongly support H2.

Figure 1d graphs these results, showing a negative interaction between oil income and BITs with umbrella clauses similar to what we found in Model 11. More notably, these results point to two separate causal channels linking BITs and tax intake. BITs without umbrella clauses correlate with reduced tax intake. A separate and more significant negative correlation associated with umbrella clauses applies only to oil-producing states. The constituent term coefficient for BITs without umbrella clauses in model 17 is −0.07, and the conditional effect for BITs with umbrella clauses is larger than that for the typical oil-producing country.⁷⁵

Models 19–21 in table 2 allow umbrella clauses’ interaction with oil income to be nonlinear. Our theory suggests that umbrella clauses’ effect should be larger at higher levels of oil income but is otherwise mute to functional form. To allow for more flexibility, we replace our linear measure of oil production with a binned measure that separates country-years into six categories. The lowest category includes country-years that produce no oil, and we separate the remaining country-years into quintiles based on oil production. Thus, the second lowest category indicates a country-year whose oil production was greater than 0 but below the 20th percentile of oil-producing country-years; the next category includes country-years with oil production between the 20th and 40th percentile of oil-producing country-years, and so on. The binned operationalization makes fewer assumptions about functional form and more naturally captures nonlinear relationships.⁷⁶ We use the lowest bin—country-years with no oil revenue—as the residual category. The coefficient on BITs with umbrella clauses indicates their effect on taxes in the absence of oil revenues, and the various interaction terms represent the additional extent to which BITs depress taxation in country-years with oil income.

These results suggest relationships similar to those we reported in table 1. The most notable exception is that they suggest BITs with umbrella clauses only impact tax intake for countries with oil incomes above the 40th percentile of oil-producing countries, which is a higher threshold than early estimates. Interestingly, there is mixed evidence in these regressions as to whether that underlying relationship is, in fact, linear. Models 19 and 20 suggest a high degree of linearity, such that BITs’ tax depressive effects rachet up by a roughly constant amount for country-years with oil income between the 60th and 80th percentiles and again for country-years above the 80th percentile. Model 21, which includes all the macroeconomic controls, suggests that BITs with umbrella clauses depress tax intake in country-years above the 40th percentile in oil income, but there is little evidence that the impact continues to increase beyond the 40th percentile. Graph 1e in figure 1 graphically represents the results of Model 19.

Our final set of models addressed potential endogeneity using an instrumental variable approach. Signing and ratifying BITs is a political decision that may relate to tax policies in ways not fully accounted for above. We use two instruments for our BITs variable in these regressions. The first measures BITs ratified in economically “similar” states. We measure “similarity” by similar export profiles. We follow Elkins, Guzman, and Simmons’s (2006) construction of export product similarity, which calculates cross-national correlations for each year across ten export-related indicators from World Development Indicators (WDI).⁷⁷ We then form a weighting matrix from those correlations to calculate the number of BITs in economic competitors. That measure increases as countries with similar export profiles sign BITs. We do this procedure separately for BITs and BITs with umbrella clauses. The number of ratified BITs should be (and in practice is) correlated with the number of ratified BITs in countries with similar export profiles. Governments sometimes ratify BITs because countries with a similar industrial structure have done so out of fear that they will be left behind in a competition for capital if they do not follow suit. The exclusion restriction is plausible: It is not apparent why BITs in countries with similar export profiles affect taxation other than indirectly through their effect on BITs.⁷⁸ This instrument’s focus on export profiles captures competition for capital stemming from hard-to-alter structural characteristics, which is not a key explanation for the race to the bottom in capital taxation—or, in other words, for low levels of taxation. To the contrary, direct effects on taxation are more plausibly driven by geographic proximity, perceptions of riskiness, and other labels found to drive capital investors (Brooks, Cunha, and Mosley 2015). We base our second instrument on Arias, Hollyer, and Rosendorff (2018) and Rosendorff and Shin (2012), which instrument BITs with the cumulative number of noneconomic United Nations Educational, Scientific and Cultural Organization (UNESCO) conventions a state has ratified. The measure captures an exogenous propensity to join international institutions. We use the interaction of our instruments with oil income as an instrument for the interaction of BITs with oil income.

Table 3 shows model estimates replicating table 1, except that they are estimated using a two-stage least squares estimator.⁷⁹ The first-stage F-statistics are well above traditional instrument strength benchmarks, and the Hansen J statistics suggest no evidence that they violate the exclusion restrictions. These models produce results similar to those of our OLS model. The consistently negative interaction terms in table 3 suggest that BITs (especially BITs with umbrella clauses) depress taxation to a greater degree in countries with large amounts of oil income. All three interaction terms are statistically significant and negative, and all are substantively larger than in the OLS models reported in table 1. These models do not indicate that BITs have an unconditionally negative effect on tax intake. The coefficients on our uninteracted BIT variables in Models 22–24 are always correctly signed but never statistically significant. In the interacted models (Models 25–33), the coefficient on the BITs variable (indicating BITs’ effect on taxes in the absence of oil income) is never statistically significant. Our instrumental variables regressions thus suggest a negative relationship between BITs and tax revenue limited to country-years with substantial oil revenue. These findings support H2 but not H1.

Additional Robustness

The online appendix reports the results of several additional alternative model specifications. These include a replication of our main models using a random effect (rather than a fixed effect) estimator; models using discretized versions of BIT variables rather than the continuous operationalizations that we use in the main models; models using data that extends back to 1980 and thus well before ISDS was a significant factor in MNC-host country power dynamics; regressions that explicitly model the spatial diffusion of BITs in countries with similar export profiles;⁸⁰ and models that control for oil prices (Haber 2023), aiming to capture the major source of fluctuation in natural resource revenue and taxation of those revenues. We also report models that include a control variable for the corporate tax rate,⁸¹ which can control for drivers of low taxation more broadly (which we excluded from our main models due to excessive missing data). Our results are robust throughout.

Conclusion

BITs limit host countries’ ability to generate tax revenues from their resource rents by making it easier for firms to enforce the tax-related terms of their contracts. The more resource rents are available in an economy, the more BIT obligations correlate with a lower tax intake. The effect is statistically significant and qualitatively substantial.

These findings are notable for several reasons, including that they suggest that scholars tie their understanding of BITs’ macroeconomic impacts too narrowly to their impact on FDI flows. We rarely evaluate BITs for macroeconomic effects beyond that. Our arguments suggest the need for a more holistic approach. Any accounting of BITs’ developmental impacts should consider their effect on the government’s ability to tax, especially in countries where tax revenue depends on foreign direct investment in the oil and gas sectors. The relevant question for BITs is not whether they attract FDI, but whether they attract enough of it to compensate for the fiscal constraints they place on policymakers.

Footnotes

Author Biography

Cristina Bodea is a Professor in the Michigan State University Political Science Department. She writes about central banks, global finance, gender, and political economy.

Jia Chen is an Associate Professor in the School of International and Public Affairs at Shanghai Jiao Tong University. His research focuses on the politics of foreign direct investment and the international investment regime.

Andrew Kerner is an Associate Professor in the Michigan State University Political Science Department. He writes about global finance and other political economy topics.

Fangjin Ye is an Associate Professor in the School of Public Economics and Administration at Shanghai University of Finance and Economics. He studies international and comparative political economy, with a focus on investment, trade, and inequality.

Notes

Author’s Note: Authors are listed in alphabetical order and have contributed equal work. The authors thank the ISQ editors and reviewers for their comments. Early versions of this paper were presented at PEIO (Bern, Switzerland, 2017) and at the CEU (Department of Public Policy seminar, 2017). Replication material is available at https://dataverse-harvard-edu.libproxy.ucl.ac.uk/dataverse/isq.

References