Too high a price? Tax treaties with investment hubs in Sub-Saharan Africa

Definitions of “investment hubs” or “tax havens” differ across institutions and over time, and there is no single globally accepted definition. The OECD has developed a list of havens as part of its work on harmful tax practices in 1998 and has since revised the list and approaches to define havens. Other efforts at defining the concept have been made by researchers, legislators, CSOs and regional and international organizations. A close link also exists to the discussion and definition of offshore financial centers (OFC), providing financial services to nonresidents.

Several CSOs, including the Tax Justice Network (2015), Oxfam, ActionAid (2016) are publishing lists of tax havens, tax treaty policy decisions and examples of cost. In the G20/OECD-led base erosion and profit shifting (BEPS) process (2015), several measures that will help reduce the risks of tax treaty abuse have been agreed.

See Sect. 2 for details.

Between 1985 and 2015, Mauritius concluded 15 DTTs with economies in SSA. Due to data limitations, we evaluate the effect of 11 treaties.

Unless double tax treaties stipulate otherwise, withholding taxes are typically levied on the gross income payments to foreign entities, thus limiting the incentive to engage in tax arbitrage. Where administrative capacity to address profit shifting is limited, withholding taxation can therefore act as “a second line of defense” in safeguarding the domestic tax base (Balabushko et al. 2017).

The IMF’s Coordinated Direct Investment Survey (CDIS) aims to collect comprehensive data on bilateral inward and outward direct investment positions. Many economies in Sub-Saharan Africa do not report on the survey and coverage among those that do is often not complete from 2009 to 2015. Examples include: Benin, Burkina Faso, Cape Verde, Ivory Coast, Eswatini, Mali, Namibia, Niger, Rwanda, Senegal and Togo.

As argued further in the empirical section, any potential endogeneity issues would most likely result in an underestimation of effects. To the extent that investment hubs actively selected their treaty partners based on prospective returns, revenue and FDI effects ascribed to treaty conclusion would be upward biased since both measures would have also increased in the absence of any causal effect of treaties.

Estimates based on macro-data are potentially biased, due to the inability to control for confounding factors, such as firm-level assets, when measuring tax minimization responses. However, micro-data might also be biased if coverage of firm-level data is limited. For instance, where no or incomplete accounting information is available, availability could be correlated with more aggressive tax optimization. Given the limited coverage of firm-level information and bilateral transaction information for Sub-Saharan African countries (Platform for Collaboration on Tax 2017), our analysis is based on macroeconomic aggregate information, which is available for most countries, but tends to be less accurate when aiming at identifying tax driven effects.

Notably, a review of available firm-level data for Sub-Saharan Africa indicates that more than 50% of MNE groups operating in Sub-Saharan Africa out of the USA include an entity in Mauritius in the group structure. However, the information in commercial databases is very limited for most countries on the continent (Platform for Collaboration on Tax 2017).

A recently concluded global investor survey ranks taxation below other factors such as political stability, the legal and regulatory environment, and market size (WBG 2017).

Interestingly, Braun and Zagler (2014) analysis suggests that, in the context of developing countries, foreign aid may sometimes be an instrument to compensate for asymmetries in their tax treaties.

See Beer et al. (2019) for a review of the empirical literature on international tax avoidance by MNEs.

See blog post by Jim Brumby and Michael Keen (2016) referencing a presentation by Stephen Shay: http://​blogs.​worldbank.​org/​governance/​tax-treaties-boost-or-bane-development.

In recent years, the role of tax treaties has been reevaluated, most notably in the context of work led by the OECD aimed at avoiding or countering the abuse of tax treaty provisions. Treaty anti-abuse provisions to counter treaty shopping comprise a minimum standard under the OECD/G20 BEPS initiative. More specifically, the minimum standard requires: (i) inclusion of a clear statement in tax treaties of the intention to avoid non-taxation or reduced taxation through tax evasion or avoidance, including through treaty shopping, and (ii) the adoption of one or both of a limitation of benefits (LOB) rule or a principal purpose test (PPT) rule, supplemented as needed to deal with conduit arrangements.

The allowable business activities of a GBC (category 1) include, among others, assets management, consultancy services, financial services, insurance, logistics, or marketing. All those activities are non-location specific and thus easily transferable within any MNE. A GBC is regarded as tax resident in Mauritius if they can demonstrate that their management and control is situated in Mauritius.

The so-called Panama (or Mossack Fonseca) and Paradise papers and related stories pursued by the International Consortium of Investigative Journalists. See, for example, ICIJ (2017).

Mauritius signed its first treaties with EU countries in the early 1980 s. Starting in the 1990 s, Mauritius expanded its network with other investment hubs, including Barbados, Cyprus, Malta, Luxembourg, Guernsey, Monaco, the Seychelles, and Singapore, and with African countries, thus enhancing its attractiveness as an investment hub for the region. In addition to the 43 effective treaties, 31 DTTs are being negotiated, signed, or are awaiting ratification. An important initial driver of the growth of Mauritius’s offshore sector was a double tax treaty signed with India in 1982 that has been recently renegotiated to gradually remove its capital gains tax exemption from 2017 to 2019.

As an extension, we also assess the effect of treaty conclusion with investment hubs more generally, by redefining our treatment variable to include DTTs between SSA economies and countries where the sum of inward and outward FDI stocks exceeds two times the domestic GDP in 2016 (for a similar definition see Damgaard and Elkjaer 2017). The extended list of countries meeting this definition includes Malta; Cyprus; Hong Kong SAR, China; Luxembourg; Singapore; Ireland; Switzerland; the Netherlands; Barbados; and the Seychelles. However, as none of these countries has concluded more than 5 treaties within our sample period in the region, they do not lend themselves for a separate empirical evaluation.

The amendments included among others the introduction of withholding tax on interest earned in South Africa and introduced a new tie-breaker clause to avoid double non-taxation. Withholding rates in the treaty remain, however, more attractive than the regular domestic rate. See Davis Tax Committee report on preventing tax treaty abuse, p.73-76; http://​www.​taxcom.​org.​za/​docs/​New_​Folder3/​8%20​BEPS%20​Final%20​Report%20​-%20​Action%20​6.​pdf.

In 2013, with significant increases in withholding rates among others.

TJN argued that the treaty violates the Kenyan constitution by going against principles of good governance, sustainability, and accountability.

The International Consortium of Investigative Journalists (ICIJ), for instance, points toward structures that have been used to shift profits out of African economies toward Mauritian holding companies with limited or no meaningful functions (see ICIJ 2017). The Mauritian Economic Development Board’s official response submits that the tax advantages accorded were required for the viability of the project in question (see: EDB, 2018).

Based on leaked documents from the law firm Conyers Dill and Pearman (ICIJ, 2019).

If location i operated a worldwide tax system, we would need to distinguish two cases. First, if the tax credit in the resident country was too small, the tax cost of income repatriations would remain at the same value \( \omega_{i}^{\text{div}} \). With sufficient tax credit in the resident country, the tax cost of income repatriation in the form of dividends would change to \( \frac{t_{i} - t_{s}}{1 - t_{s}} \), a grossed-up tax differential.

So, while we model the financing decision in each resident location as a discrete choice, the project is ultimately financed with both financing sources. We therefore also neglect thin-capitalization rules.

For instance, if affiliates in location \( i \) provide debt and no withholding taxes apply on interest payments to this location, the effective tax cost of income repatriation would read \( \tau_{i } = t_{i} - t_{\text{s}} \), where \( t_{i} \) denotes the corporate income tax rate in country \( i \). So, if the entire project was financed through this location, the after-tax return would simplify to \( \pi = \left[ {f\left( K \right) - rK} \right]\left( {1 - t_{\text{s}} } \right) + rK\left( {1 - t_{i} } \right) - rK \), where the first term represents corporate profit in the source country, the second term is profit in the resident country, and the third term represents (non-deductible) financing costs in the resident country.

For simplicity, we assume in the theoretical model that these costs are constant across all resident locations. In the empirical section, however, we allow them to vary between investment hubs and other locations.

Arguably costs of treaty shopping have escalated due to the recent emphasis in the OECD/G20 BEPS initiative on introducing anti-abuse provisions that require sufficient substance for an entity to qualify for treaty benefits. However, these rules still require effective implementation in sometimes challenging political economies, where political connection may drive tax evasion (Rijkers et al. 2017).

Sufficiency follows from concavity of the production function. Proofs are available upon request.

For another context where avoidance opportunities have a positive real effect, see, e.g., Grubert and Slemrod (1998).

At the limit, where treaty-shopping cost tends to zero, the tax cost of income repatriation to location i needs to equal the average tax cost of income repatriation to any other location, implying that \( \omega_{i}^{*int} = t_{\text{s}} + \bar{\tau }_{ - i} \).

Control of corruption and governance effectiveness.

See Crivelli et al. (2016) for a similar approach.

Besides optimization behavior, core assumptions of the theory include that marginal treaty-shopping costs are proportional to the amount of rerouted income and that investments are financed with debt when made from low-tax jurisdictions.

Malta; Cyprus; Hong Kong SAR, China; Luxembourg; Singapore; Ireland; Switzerland; the Netherlands; Barbados; and the Seychelles.

In unreported estimations, we confirm that the measured effect of DTTs with Mauritius remains unchanged when controlling for bilateral investment treaties.

However, the bias vanishes with increasing time observations and our panel covers up to 25 years.

We expect this inequality as the parameter is inversely related to MNEs’ overall costs of treaty shopping.

The DD estimates indicate that total revenue losses due to investment hub treaties are around 20 percent of CIT revenue.

Beer et al. (2019) estimate that MNEs’ tax avoidance might reduce CIT revenue in the USA by 17 percent; Clausing (2016) puts this number between 19 and 30 percent.

Under the G20/OECD base erosion and profit shifting (BEPS) initiative, there is ongoing pressure to improve some of the anti-abuse provisions in Mauritius’ tax treaties. Mauritius is a member of the BEPS Inclusive Framework and has therefore committed to implement the BEPS minimum standards. Further, Mauritius signed the Multilateral Convention (MLI) on July 5, 2017, which will accelerate adoption of anti-abuse rules. Several reforms to the taxation of the Mauritian global business sector were announced recently, which are aimed at implementing international standards targeting harmful tax practices. However, core provisions of concern for many of its treaty partners are the withholding rates agreed on dividend, interest, and royalty payments, which are not (yet) covered under these initiatives.

Matching and regression-based estimators both presume that all confounding factors are observable or constant across time. While matching estimators relax parametric assumptions, thus potentially increasing the validity of causal inference (Ho et al. 2007), they require that the counterfactual outcome is a convex combination of observed outcomes among the control group. In contrast, regression-based analyses are less restrictive in this regard, allowing the extrapolation of counterfactuals. Finally, while the regression-based approach presumes a common trend between treated and controls, our matching-based approach allows trends to differ between countries. Consequently, results may differ.

A popular approach to quantify similarity and assign weights is the difference in the predicted likelihood of receiving treatment, the propensity score (Rosenbaum and Rubin 1983). In panel settings, where the set of control units changes over time, propensity score matching requires estimation of a separate binary outcome regression for each period. This strategy is infeasible in our context, as in several years only one country concluded a treaty with Mauritius.