Policy Research Working Paper 10345 The Impact of Tax Blacklisting Matthew Collin Macroeconomics, Trade and Investment Global Practice May 2023 Policy Research Working Paper 10345 Abstract This paper estimates the policy and economic impacts four years after the inception of the list. There is also no of a European Union–led effort to review and “blacklist” clear evidence that the listing exercise had any impact on jurisdictions based on their compliance with international offshore wealth or shifted profits, largely because the bulk of standards designed to curb corporate profit shifting and jurisdictions that host both of these were not targeted by the private tax evasion. Using a combination of regression dis- European Union. The results suggest that “coercive” efforts continuity and difference-and-difference methods, there is to reduce global tax evasion and avoidance will struggle evidence of only limited improvements in tax governance without better targeting and enforcement. This paper is a product of the Macroeconomics, Trade and Investment Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The author may be contacted at mattcollin@gmail.com. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team The Impact of Tax Blacklisting Matthew Collin ∗ † ‡ Keywords: International tax, blacklisting, tax havens, profit shifting, tax evasion JEL classification: G28, H24, H26, H87 ∗ EU Tax Observatory and Skatteforsk - Centre for Tax Tesearch, Email: mattcollin@gmail.com. I would like to thank Pierre Bachas, Carol Graham, Homi Kharas, Jan Loeprick, Tom Roth, Marijn Verho- even, Ludvig Weir and participants of both the International Online Public Finance Seminar Series and the EEA-ESEM 2921 Congress for helpful comments and suggestions. All errors are my own. † The first version of this working paper was published as https://www.brookings.edu/research/does- the-threat-of-being-blacklisted-change-behavior/ ‡ **This paper is a product of the WB Fiscal Policy and Sustainable Growth Unit under the Research and toolkits on international taxation project (P179248; TTL: Ana Cebreiro Gomez), which is financed by the WB GTP Program ( https://www.worldbank.org/en/programs/the-global-tax-program). Abbreviations AEOI Automatic Exchange of (tax) Information AML Anti-money laundering BEPS Base Erosion and Profit Shifting CbCR Country-by-country reporting CFC Controlled Foreign Company CIT Corporate Income Tax COCG Code of Conduct Group CRS Common Reporting Standard EFSD European Fund for Sustainable Development EFSI European Fund for Strategic Investments EOIR Exchange-of-information on Request EU European Union FATF Financial Action Task Force FHTP Forum on Harmful Tax Practices FSI Financial Secrecy Index IF Inclusive Framework MD Multi-dimensional MNE Multinational Enterprise OECD Organisation for Economic Cooperation and Development OFC Offshore Financial Center RD Regression discontinuity SF Stability factors 2 1 Introduction There has been a rapid shift in global tax governance in the past decade. This has been prompted by a recognition that there are two significant externalities driving the movement of financial assets and profits to offshore financial centers (OFCs). The first is financial secrecy, provided by offshore jurisdictions to clients who are able to obscure their ownership and potentially avoid taxation. Studies estimate the amount of wealth being held in offshore tax havens to be approximately 8% of all household wealth or 10% of global GDP, a significant portion of which goes unreported despite recent efforts to clamp down on evasion (Zucman 2013; Johannesen and Zucman 2014; Johannesen et al. 2020). The second externality is a set of corporate tax policies - a combination of rates, loopholes and lack of transparency - that create incentives for multinational enterprises (MNEs) to shift their profits away from high tax jurisdictions to lower tax ones where there is little economic activity of substance. In a recent study, Tørsløv, Wier, and Zucman (2022) estimate that up to 40% of global MNE profits are shifted to tax havens.1 Taken together, inducing a relatively small number of jurisdictions to reduce these negative spillovers could have sizable gains in global welfare (Slemrod and Wilson 2009). To collectively deal with these externalities, jurisdictions across the globe are in the process of committing to two separate OECD frameworks aimed at reducing international tax evasion and avoidance. The first of these is the OECD/G20 Base Erosion and Profit Shifting (BEPS) initiative, which is being taken forward by the international forum known as the Inclusive Framework (IF) on BEPS. The goal of the IF is to promote specific ac- tions and standards that will help countries tackle tax planning efforts by multinationals which lead to an erosion of the corporate tax base. At the very least, members of the IF are expected to adopt four minimum standards, built around reducing harmful tax prac- tices, combating tax treaty abuse, handling treaty disputes and arbitration and finally documenting transfer pricing. The last of these include country-by-country reporting (CbCR), the requirement for parent companies of multinationals to disclose significant details about their operations, profits and tax payments, which will then be exchanged between participating tax authorities.2 As of November 2021, 141 jurisdictions - around thirty of which are low or lower-middle income countries - have joined the Inclusive Frame- work as members, committing to adopting its standards. The majority of IF members recently endorsed a new global tax deal aimed at reducing the incentives to shift profits by introducing a global minimum tax of 15% as well as a reallocation of residual profits from the largest MNEs in operation to market jurisdictions. 1 The term tax haven is occasionally a contentious one, but they are generally thought of as jurisdictions that employ financial secrecy, low corporate tax rates, and/or preferential tax regimes to attract financial assets and investments from foreign entities and persons. 2 It should be noted that the BEPS Minimum Standards are only part of the entire package of reforms promoted by the OECD - and by themselves do not represent a sufficient set of policies for eliminating cross-border tax externalities. 3 The second initiative is the adoption of the OECD’s Common Reporting Standard (CRS) for the Automatic Exchange of Information (AEOI). Jurisdictions that adopt this standard will require financial institutions to report account information for non-resident taxpayers and for that information to be automatically exchanged between tax authori- ties in participating jurisdictions. Recent studies suggest that bilateral AEOI exchanges lead to a shift of offshore assets out of tax havens, although it is unclear how much of this presumably-untaxed wealth is repatriated or just moved elsewhere (Beer, Coelho, and Leduc 2019; Casi, Spengel, and Stage 2019; O’Reilly, Ramirez, and Stemmer 2019; Menkhoff and Miethe 2019; Bomare and Herry 2022). Approximately 120 jurisdictions have committed to exchanging under CRS, 40% of which which began their first ex- changes in 2017, another 40% in 2018 and the remaining between 2019 and 2024.3 The CRS framework is seen as an improvement from an older system of information exchange, known as exchange-of-information on request (EOIR), where tax authorities must make active requests for information on specific taxpayers. Despite this, there are still efforts to ensure that EOIR is being adequately implemented, as jurisdictions are reviewed through the Global Forum on Transparency and Exchange of Information for Tax Purposes. Despite this progress, out of concern for the impacts of profit shifting and tax evasion on its own tax base, beginning in 2017 the EU Commission began a careful review of 92 non-European jurisdictions to determine how compliant they were with international standards around tax transparency, fair taxation and adherence to the OECD’s Base Erosion Profit Shifting (BEPS) minimum standards. After the review and some dialogue with non-compliant countries, the EU released a ‘grey’ list of jurisdictions that were non- compliant with these standards, but had committed to make improvements, as well as a ‘black’ list of jurisdictions that were non-cooperative, that were to be subject to a number of EU countermeasures. In this paper, I investigate the impact that this process has had on the standards that the EU intended to enforce and on economic outcomes of the jurisdictions that were targeted. For the former, I rely on the process the EU used to select countries for consideration in its listing process to compare jurisdictions that scored just high enough to be considered with those that did not. Using a regression discontinuity specification, I find that countries that were selected into the EU’s review process were substantially more likely to be grey or blacklisted, but that there is mixed evidence that, to date, the process has substantially affected policy adoption. On average, index measures of global tax governance, based on the EU’s own goals, do not show large improvements. The main exception is for outcomes related to unfair tax regimes, where jurisdictions selected by the EU saw a large increase in the probability that their regimes had been inspected by the EU or the OECD and, as of the time of writing, that they no longer had any harmful regimes present. It also increased the chances that a jurisdiction was independently reviewed by the OECD Forum on Harmful Tax Practices, an indication 3 Over forty developing countries have indicated that they will implement the CRS, but have not yet set a date to do so. 4 that the EU policy expanded the number of jurisdictions that would ultimately be subject to some sort of review. There is weak evidence that the EU review process increased the number of BEPS minimum standards adopted. There is also some evidence that the EU process led to a sharp increase in the probability that the Global Forum rated a jurisdiction as “largely compliant” or better on its implementation of EOIR. The most robust and striking result from the analysis is the fact that countries selected into the process were substantially more likely to join the Inclusive Framework, thus weakly-committing themselves to implementing the BEPS minimum standards. In the short term this is a low-effort outcome: states can signal their willingness to reform at a later date. This means that even if the EU review process has not improved international tax governance by much in the short/medium term, it might do so in the long term as these commitments become more binding. It also has implications for the future of deliberation over new international tax rules: targeted countries were substantially more likely to endorse the new global tax deal. The second chief finding in this paper is that the EU tax blacklisting process has not had a discernible effect on economic outcomes of listed jurisdictions. To investigate this, I use a difference-in-difference strategy, comparing outcomes for jurisdictions that are added to the grey or blacklist to those that were not. Listed jurisdictions do not see any relative decline in their stock of offshore wealth, foreign direct investment or artificially-shifted profits, in tension with the intention of the listing exercise to induce them to adopt policies which should erode these outcomes. I also show, contrary to the purported policy goals of the EU, that the bulk of the world’s artificially-shifted profits reside in jurisdictions that were never targeted by the blacklisting process. Taken together, these results indicate that in order to be effective in reducing the sum of negative cross-border spillovers, the listing process would need to both be better targeted and substantially raise the costs of non-compliance. This paper makes several contributions. First, it is the first rigorous test of the impact of the EU’s efforts to improve tax governance worldwide. While it is easy to observe how countries included in the listing process have improved, we would not normally know how these countries would have improved without the EU’s intervention, particularly because there is ongoing pressure or assistance from a multitude of institutions (e.g. the OECD, US Government, IMF, World Bank) to improve global tax governance. Second, while most papers identify the impact of the recent evolution of anti-BEPS and anti-evasion policy,4 this paper instead investigates whether large regional bodies can effectively accelerate that evolution. A number of recent theoretical contributions have highlighted a coordination problem in the decision of tax haven jurisdictions to become more transparent (Elsayyad and Konrad 2012; Konrad and Stolper 2016). For example, Elsayyad and Konrad (2012) argue that incremental, sequential reform by some havens just increases the market power of other havens, reducing the chance of subsequent 4 For analyses of efforts prior to the BEPs era, see (Johannesen and Zucman 2014; Caruana-Galizia and Caruana-Galizia 2016; Omartian 2017; Johannesen, Langetieg, Reck, Risch, and Slemrod 2020). 5 reform. The authors argue that a “big push” multilateral agreement is thus necessary to successfully entice jurisdictions out of the tax haven business. While the OECD process arguably is facilitating that big push, this paper can be seen as a test of an attempt to expand the reach of that effort. Finally, this paper adds to an empirical literature on the effect of listing exercises on country policy-making and economic outcomes. Examples of the former include Morse (2019). who shows that those added to the Financial Action Task Force (FATF)’s ‘greylist’ of countries that lack compliance with international anti-money laundering (AML) stan- dards are significantly more likely to criminalize money laundering. Kelley and Simmons (2015) find that countries listed on the US State Department’s annual Trafficking in Person’s report are more likely to subsequently criminalize human trafficking. These em- pirical studies are backed up with case study evidence that jurisdictions are nudged into compliance by the threat of blacklisting (in both the space of anti-money laundering and in tax transparency), even when there are no explicit sanctions (Sharman 2009). In the space of tax, previous analyses of the impact of blacklists have focused on the response of multinational firms, rather than the tax haven itself (Grottke and Kittl 2016; Rusina 2020). For the latter, a number of studies have now examined the impact of listing exercises on economic outcomes (Case-Ruchala and Nance 2021; Morse 2022; Kida and Paetzold 2021). The rest of the paper proceeds as follows: Section 2 discusses the recent history of the EU review and listing process. Section 3 discusses the empirical approach I take in this paper, Section 4 presents the results on tax governance outcomes, Section 5 presents results on economic outcomes. Section 6 discusses the implications these results have for international tax governance as well as reasons the EU blacklisting process may not have a powerful effect on state behavior or on targeted jurisdictions. I conclude the paper with Section 7. 2 The EU review and listing process In the early 2010s, the European Commission came to the conclusion that several non- EU states (referred to as ‘third countries’) posed a continued threat to the tax base of EU Member States. Individual EU countries could respond to this threat bilaterally by enacting protective measures or imposing penalties, but because of the economic freedom individuals and firms faced within the bloc, they could easily re-route their affairs through any EU country that had not taken a tough stance against these third countries. The solution proposed by the European Commission was to come up with a unified approach to influencing the tax policy of third countries.5 This would in theory accomplish several things. First, it would amplify the pressure on third countries through the combined might of the entire bloc. Second it would raise the floor of anti-abuse and evasion actions 5 https://ec.europa.eu/taxation_customs/sites/taxation/files/resources/documents/ taxation/tax_fraud_evasion/c_2012_8805_en.pdf 6 taken by Member States by ensuring that at a minimum there would be a bloc-wide response to bad behavior. Finally, in contrast to the OECD’s efforts, it would allow the EU to target third countries it felt posed the biggest threat to its tax base, predominately those with which it shares close economic ties. This pan-European effort first emerged in mid-2015, when the EU published a list of non-cooperative tax havens as part of its “Action Plan for Fair and Efficient Corporate Taxation in the EU.”6 The list was presented as an amalgamation of watch lists maintained by each of the EU member states: if a jurisdiction was present on ten or more lists of member states, it was included in the published annex of the Action Plan. The EU’s 2015 list was quietly shuttered as the institution opted for a more systematic approach for creating a pan-European list.7 In order to establish an objective-seeming method for choosing which jurisdictions would be included, the European Commission released a scoreboard covering 194 non-European countries in September 2016.8 The scoreboard, which is discussed in more detail in Section 3.2, was devised to determine which jurisdictions were at the greatest risk of facilitating tax avoidance, ranked them according to three criteria: (i) the strength of their economic ties to the EU, (ii) their overall financial activity, and (iii) a series of ‘stability factors’ including corruption and regulatory quality. Those jurisdictions which ranked high enough all three criteria were selected for further scrutiny, although some, such as those designated by the UN as Least Developed Countries, were excluded from the review process. Others, such as a few tax havens the EU already had an existing transparency agreement with, were automatically included. The EU performed a very basic assessment for the eighty-six selected jurisdictions to determine whether there were risks with respect to transparency, preferential tax regimes or low tax rates. Following this, the EU’s Code of Conduct Group for Business Taxation (COCG) devised a set of criteria to screen jurisdictions for their adherence to international standards on tax transparency, fair taxation and anti-BEPS measures.9 This screening took place during the first half of 2017, after which the COCG then communicated directly with jurisdictions found to have deficiencies, asking for commitments to improve their tax governance by the end of the following year. In early December 2017, jurisdictions which were not able to make a credible commitment were added to EU’s first published “non-cooperative jurisdictions for tax purpose,” referred from hereon as the blacklist. In addition to this, in an annex, the EU also publishes a list of the jurisdictions it is working with to improve their adherence to international standards, which I will call the EU’s ‘grey list.’ Of the jurisdictions that were originally selected for the entire review and screening 6 https://ec.europa.eu/taxation_customs/business/company-tax/action-plan-corporate-taxation_ en 7 https://data.consilium.europa.eu/doc/document/ST-9452-2016-INIT/en/pdf 8 https://ec.europa.eu/taxation_customs/sites/taxation/files/2016-09-15_ scoreboard-indicators.pdf 9 These criteria are discussed in detail in Table A5 in the Online Appendix and below in Section 3.4. 7 Figure 1: Timeline of EU selection, review and listing Sep 2016 Dec 2017 Scoreboard released New grey/ and 92 jurisdictions blacklist Jun 2015 selected for review released Initial blacklist released Grey and black lists COCG reviews are updated jurisdictions every few months 2015 2016 2017 2018 2019 2020 2021 Oct 2017 EU asks 72 reviewed jurisdictions to address deficiencies Table 1: Number of jurisdictions reviewed and listed by the EU Ranked in Selected for Deficiencies raised Ever Ever Category scoreboard review by the COCG by COCG greylisted blacklisted 1. EU Member States (28) 0 0 0 0 0 2. Jurisdictions w/ EU transparency agreement (5) 5 5 3 4 0 3. Least Developed Countries (48) 48 1 1 1 1 4. Other non-EU jurisdictions (160) 141 86 68 67 29 Total 194 92 72 72 30 Note: Categories determined by EU during development of its scoreboard. Jurisdictions with existing EU tax transparency agreements at the time the scoreboard was developed include Switzerland, Liechtenstein, Andorra, Monaco and San Marino. Note that Vanuatu was initially excluded as an LDC, but was still reviewed as it was expected to soon be graduating from LDC status) Some jurisdictions that were excluded from the ranking exercise due to a lack of data were still subsequently reviewed. process, roughly 50% were included in the greylist in its first release and a further 15% were added to the blacklist. Figure 2 displays the number of countries listed on black and grey lists from December 2017 until early 2022. Over the course of 2018, the greylist swelled to over 60 jurisdictions as many jurisdictions committed to improving various tax governance outcomes, including several jurisdictions that successfully moved off of the blacklist. Many of these commitments were due in the 2019 calendar year, which led to a large reduction in jurisdictions on the greylist, some of whom moved to the blacklist due to continued noncompliance. The EU CCD continues to review and update that list up until today. If greylisted jurisdictions take too long in this implementation, they face the risk of being added to the blacklist. Blacklisted jurisdictions are those who are considered to be both non-compliant and non-cooperative. For these countries a number of EU sanctions will apply: (i) funding from a variety of EU Funding Instruments cannot be channeled through entities in listed countries, (ii) tax schemes or multinational activities routed through listed jurisdictions will be subject to additional reporting requirements by EU tax authorities, and (iii) member states have committed to including other specific sanctions, such as increased monitoring and audit risks or special withholding rates, although many of these were not implemented until the last two years. In addition to these, several European countries 8 Figure 2: Evolution of the EU greylist and blacklist over time Note: Figure shows number of jurisdictions on the EU’s list of non-cooperative jurisdictions for tax purposes (the blacklist) and those listed in its published reports (Annex II) under “state of play of the cooperation with the EU with respect to commitments taken to implement tax good governance principles” (the greylist). also announced that companies with subsidiaries based in blacklisted countries would not be eligible for government aid being provided in response to the ongoing COVID-19 pandemic. 3 Data and empirical approach I will be using two different empirical frameworks: a regression-discontinuity (RD) frame- work and a difference-in-difference framework. In this section I will discuss how these are specified, what running variables I will be using for the RD estimation, and the tax governance outcomes other economic outcomes I will be considering. 3.1 RD framework The main problem faced in estimating the impact of the review and blacklisting process is non-random selection into both. We know that the EU chose its three criteria for selection based on the assumption that they were positive correlated with a jurisdiction’s likelihood of facilitating tax avoidance that would affect EU member states. We also know that the final choice of the grey and black lists was determined endogenously by jurisdictions responding to the EU’s pressure to reform. So a simple comparison of selected countries with those that were not selected, or listed with unlisted countries, is likely to lead to biased estimates of the impact of the EU selection process. 9 However, I can take advantage of the fact that the EU’s selection process incorporated arbitrary cutoffs to determine eligibility, leading jurisdictions with similar scores to face very different outcomes solely because they fell on opposite sides of the eligibility thresh- old. In a regression discontinuity framework, the premise is that, absent any impact of the EU selection process, the relationship between a jurisdiction’s score on the criteria and its likelihood of adopting new tax transparency standards would be continuous as it crosses that threshold. Consider the following empirical specification. Let Yi be the tax governance outcome of interest for jurisdiction i. We are interested in the impact that selection into the EU review process, Si has on subsequent outcomes Yi . Consider the following reduced form equation, where: Yi = α1 [Di > 0] + α2 f (Di ) + α3 f (Di ) × [Di > 0] + Xi β + ϵi (1) Where Di indicates the jurisdiction’s distance to the cutoff used by the EU to deter- mine selection into the review process. f (Di ) is a function of that distance, allowed to vary in slope on either side of the cutoff. In the next subsection I will discuss how that distance measure is constructed. Xi is a vector of jurisdiction characteristics included as controls. In this specification, α1 estimates the effect of crossing the EU selection threshold has on the tax governance outcome of interest. However, not every jurisdiction that passed the threshold was eventually selected for review. So in addition to the reduced form specification above, we can use a ‘fuzzy’ regression discontinuity approach, where first we estimate the impact that crossing the threshold has on selection Si : Si = α1 [Di > 0] + α2 f (Di ) + α3 f (Di ) × [Di > 0] + Xi β + ϵi (2) and then use the selection outcome from (2) as an instrument for selection in the following equation: Yi = γ1 Si + γ2 f (Di ) + γ3 f (Di ) × Si + Xi β + ϵi (3) For estimation of equations (1) and (3), I will proceed as follows: I estimate treatment effects using local linear estimation, using bandwidth selection and bias-correction meth- ods outlined in Calonico, Cattaneo, and Titiunik (2014), Calonico, Cattaneo, and Farrell (2018) and Calonico, Cattaneo, Farrell, and Titiunik (2019). I do this separately for the reduced form impact of crossing the selection threshold and in the fuzzy RD framework, where the selection indicator in equation (2) is used as an instrument for Si in equation (3). For completeness, the results were also calculated using global quadratic and cubic functions,10 as well as a simple OLS regression where different orders of each EU selection 10 As noted in Gelman and Imbens (2019), higher order polynomials, particularly when the polynomial 10 indicator are entered into the equation separately (rather than aggregated as a single running variable). These results are presented in Table A6 in the Online Appendix. Next, the EU scoreboard data is discussed in detail as well as the steps used to construct the running variable Di , given the multidimensional nature of the data. 3.2 The EU scoreboard data As discussed above, the EU ranked jurisdictions according to three criteria: their strength of each jurisdiction’s economic ties with the EU, their level of financial activity, and the degree to which each is stable enough to be an attractive destination for funds. Each indicator was constructed using the following data: 1. Strength of ties: constructed using measures of the jurisdictions total trade with the EU, trade in services, both inward and outward FDI flows and the presence of foreign affiliates of EU-based companies 2. Financial activity: inward and outward dividends, interest payments, royalties, and FDI stocks 3. Stability factors: control of corruption and regulatory quality, as measured by the World Bank’s Worldwide Governance Indicators For each indicator, jurisdictions were given a rank which represented their highest rank across all measures used.11 Then within each indicator, these ranks were transformed into percentage scores, so that a jurisdiction that is - for example - ranked 8 out of 130 is assigned the percentage of score 8 130 × 100 = 6.15. Lower percentage scores are equivalent to higher ranks: so a jurisdiction that has a percentage score of 6 in the strength-of- ties indicator has a higher rank (and thus stronger ties) than a jurisdiction that has a percentage score of 7. The Commission then set cutoffs (60,40,70) for each indicator which reflected the priority it placed on each dimension. Jurisdictions with a percentage score lower than the cutoff in all three indicators were selected for further investigation.12 There were several exceptions to the EU selection process. Jurisdictions designated as Least Developed Countries (LDCs) by the United Nations were excluded on the grounds that they faced constraints in improving their tax governance, although they were still ranked. Several jurisdictions which already had a transparency agreement with the EU in place were automatically selected.13 Finally, and somewhat contentiously, the European Union excluded its own member states from the scoring and listing process. is global (defined over the entire sample, as described in the equation above) can lead to noisy estimates and bias. 11 The entire methodology is described here: https://ec.europa.eu/taxation_customs/sites/ taxation/files/2016-09-15_scoreboard-methodology_en.pdf 12 If data on only one or two indicators were available, jurisdictions were selected if they passed the cutoff for just those indicators. 13 Switzerland, Liechtenstein, Andorra, Monaco and San Marino. 11 3.3 Running variable(s) To construct a running variable, I take two main approaches. The first is to construct a multidimensional running variable out of three percentage scores the EU used. The second is to use a percentage score that is the most binding for countries. I explain each of these in turn below: (i) Multidimensional (MD) running variable Normally, a regression discontinuity framework relies on a single running variable to determine assignment to treatment. In the case of the EU review process, I am faced with three, all of which need to exceed a specific threshold. To simplify the analysis, I collapse the three indicators into a single running variable, D, where D indicates the distance in percentile points the jurisdiction is from being eligible for review. 14 To construct a univariate running variable, I calculate the Manhattan distance from each jurisdiction to the boundaries of this selection space. This is the minimum distance across all dimensions a country must travel in order to cross the selection boundary, so that when a jurisdiction is outside the selection boundary: D = −1[(X − 60)1{X > 60} + (Y − 40)1{Y > 40} + (Z − 70)1{Z > 70}] Where 1{X > 60} is an indicator = 1 if X is above 60, and so on. When a jurisdiction is within the selection boundary, the relevant distance is the closest edge of the selection space: D = −min[X − 60, Y − 40, Z − 70] For example: a country with percentiles scores of (50,50,80) must move by (0,-10,-10) to reach the boundary, so faces a distance of -20 (in contrast to -14.4, the ‘as the crow flies’ Euclidean distance). By contrast, a country with percentile scores of (20,20,20) must only move by (0,-20,0) to reach the nearest boundary, which is the same as the Euclidean distance. In the Online Appendix I also present results using the Euclidean distance, and find that that it makes no qualitative difference to the headline results.15 Figure 3 displays the distribution of this collapsed multidimensional running variable, as well as a standard density test for manipulation. Centering the running variable around a three-dimensional cutoff an equivalent pro- cedure to what Wong, Steiner, and Cook (2013) describe as the ‘centering’ procedure for 14 Consider the first figure in the Online Appendix, which graphs each jurisdiction by its three percentage scores in three dimensional space. The purple cuboid represents the “selection space,” within which a jurisdiction is eligible for review by the European Union. For example, a jurisdiction with a percentage score of (59,39,69) is just inside this space where a jurisdiction with a percentage score of (61,41,71) is just outside. 15 See Table A6. 12 Figure 3: Distribution of running variable Histogram Estimated density Notes: Figure shows histograms and density tests for the multidimensional running variable (MD). collapsing a multidimensional RD into a single running variable RD. The main limitation to this approach is that it estimates a frontier average treatment effect (FATE) which is the weighted average of the univariate treatment effects defined over each indicator’s fron- tier (e.g. the treatment effect around crossing each threshold separately). The weights of the FATE are not scale invariant: for instance, re-scaling X by a positive number will decrease the weight placed on the univariate treatment effect for X. This makes the choice of scale for each indicator important, as it determines each indicator’s relative contribu- tion to the estimated treatment effect. For the analysis here, I take the EU’s construction of its indicators and its assignment process as given, keeping each indicator defined in percentiles, and assuming the indicators are comparable in percentile space. This is a strong assumption, as it assumes that moving up one percentage score in, for example, stability factors is equivalent to moving up one percentage score in financial activity. (ii) Stability factors (SF) running variable As both a robustness check, and to investigate possible heterogeneity in impact of the EU’s review process, I also estimate treatment effects using a univariate running variable defined solely over the ‘stability factors’ indicator. As shown in Table A1 in the Online Appendix, out of the three indicators, this one has the greatest predictive power in a RD setting, due to the fact that most jurisdictions around the stability factors threshold have already exceeded the other two thresholds. It is also the running variable for which jurisdictions appear the most balanced on pre-treatment observable characteristics, giving it one advantage over the multidimensional running variable. 13 3.4 Tax governance outcomes In choosing tax governance outcomes to consider in the analysis, I rely in part on those set out by the EU Council as criteria on which listed jurisdictions would be assessed.16 These are listed in detail in Table A5 in the Online Appendix, but can be broadly grouped into three categories: (1) tax transparency, (2) fair taxation and (3) anti-BEPS measures. Table 2 displays the full list of observable outcomes that I will use which either directly measure the EU’s criteria or reflect a reform process that a jurisdiction must follow in order to eventually comply with the EU’s criteria. Data on each of these outcomes was obtained directly from OECD online sources between November 2019 and April 2022. One difficulty with OECD reporting on tax governance reforms is a lack of panel data: in many situations it is only possible to observe the current state of affairs rather than precisely when a jurisdiction enacted a certain reform. This is one of the reasons an RD setting is preferable to that of a difference-in-difference framework. For my analysis of how the effects of the EU listing process have changed over time, when possible I have used the Internet Archive’s Wayback Machine17 to retrieve earlier instances of public data. For each of these categories, I have picked the most proximate, measurable outcomes relevant to the EU. One challenge is that one goal is imperfectly measured: under its “fair taxation” criteria, the EU requires that a jurisdiction has no harmful tax regimes in place. Both the EU’s COCG and the OECD’s Forum on Harmful Tax Practices (FHTP) conduct reviews of preferential regimes across many jurisdictions. Harmful regimes are detected through these reviews and jurisdictions typically (but not always, hence the blacklist) revise or abolish them so they are no longer considered harmful. However, we can only observe harmful regimes in jurisdictions that the COCG or the FHTP choose to review - there is no independent measure of harmful tax practices beyond what is considered by these two organizations. The former is driven primarily by which jurisdictions were selected for review by the COCG, while the latter is mainly driven by jurisdictions that have signed up to the Inclusive Framework. This means that harmful regimes may still exist in jurisdictions that have not received attention from either body. Without an objective measure for every jurisdiction, it is impossible to know the true effect of the EU on harmful tax regimes. In lieu of this, I have created an transparency-based outcome which is in the spirit of the EU objective, one in which a success is determined by whether a jurisdiction has been successfully examined and cleared of harmful tax practices. For this paper, a jurisdiction is considered ‘not harmful’ if it has been reviewed by either the COCG or the FHTP and, as of the latest COCG and FHTP reports, it has no active harmful regimes or has set a date for an upcoming rollback.18 Thus, if a jurisdiction has not been reviewed, or it has been reviewed, found to have harmful regimes and has not committed to abolishing them or rendering them benign, it will not be counted as a success in this outcome measure. 16 https://data.consilium.europa.eu/doc/document/ST-14166-2016-INIT/en/pdf 17 http://web.archive.org/ 18 This includes both preferential tax measures and offshore structures lacking substance requirements. 14 One of the successes that the EU takes credit for is nudging jurisdictions to eliminate harmful tax systems (European Commission 2020). In Section 6.1 I investigate these impacts further, although I am not able to put precise estimates on the number of harmful tax regimes eliminated as a result of the process. While I will report results for each of these outcome groups separately, to reduce concerns over multiple hypothesis testing, I construct indices for each outcome group. With the sole exception of the proportion of EU member states covered by an automatic exchange of information agreement (A.2), each of these outcomes is binary, so I construct the index as a simple mean of each outcome measure across the group. So, for example, if a jurisdiction has signed up to country-by-country reporting (C.1) but not yet adopted the MLI position (C.2) or published a MAP profile (C.3), its Anti-BEPS mean outcome will take a value of 1 3. One challenge in using a regression discontinuity framework in a cross-country setting is a general lack of power. As will be seen in the next section, the average effective number of observations being used varies between 40-80 countries. One concern is that low levels of power make it too difficult to reject the null hypothesis of no effect, meaning that modest impacts of the EU review and listing process will go undetected. I can allay these concerns somewhat by constructing additional outcomes where, if the EU’s involvement is having a sizable impact on international tax governance, we would expect large enough effect sizes that the minimum detectable effect (MDE) hurdle is likely to be cleared. For this, I construct one additional outcome measures: a variable equal to the number of proxies for the four minimum standard proxies a jurisdiction has imple- mented to date: (B.1) being cleared of harmful tax regimes, (C.1) CbCR commitment, (C.2) signing up to the MLI and (C.3.) publishing a MAP profile. This “minimum ef- fort’ outcome is intended to pick up even the slightest effort at meeting these basic tax governance outcomes. Given that the median number of these outcomes that have been reached is 3 for non-LDC jurisdictions selected into review by the EU and 0 for those that are not, an impact of 2 seems like a reasonable expectation, one that is also in line for the MDE for this outcome in the main reduced-form specification. 3.5 The impact of what? The above RD framework considers the reduced form impact of crossing the multidimen- sional selection threshold and subsequently being selected for review by the EU’s Code of Conduct group. Thus the treatment effect being identified is the aggregated effect of being selected into the review process, not necessarily the impact of blacklisting itself. There are several ways in which being selected might influence behavior. The first is that being selected increases scrutiny by the EU’s Code of Conduct group around the jurisdiction’s compliance with the tax governance standards, drastically increasing the probability a jurisdiction will be listed if it does not comply. Thus a jurisdiction might decide to comply in order to avoid 15 Table 2: EU-targeted tax governance outcomes and measures used in paper Outcome group Outcome measure Data sources A. Tax transparency A.1 Public commitment to exchange informa- OECD AEOI Por- tion under the Common Reporting Standard tal,a OECD Con- (CRS) for AEOI by 2020a as of May, 2022 vention website, b , OECD EOIR portalc A.2 Percentage of EU member states jurisdiction has a AEOI relationship with as of May, 2022 A.3 Commitment to (or signatory of) Multilateral Convention on Mutual Administrative Assis- tance (MAA) as of February, 2022 A.4 “Largely compliant” or better rating by Global Forum on EOIR as of May, 2022 B. Fair taxation B.1 Jurisdiction reviewed at least once by either EU COCG data on the EU Code of Conduct Group (COCG) or all regimes reviwed (BEPS Action 5) the OECD Forum on Harm- since 1998i ful Tax Practices (FHTP) and, as of April, 2021, no harmful regimes are present. C. Anti-BEPS C.1 (BEPS Action 13) Becoming a signatory of OECD CbCR the CbC MCAA as of May 2022 and MLI sites,e f OECDOECD C.2 (BEPS Action 6) Becoming a signatory of the MAP Profile listg Multilateral Convention to Implement Tax Treaty Related Measures to Prevent BEPS (MLI Position) as of May 2022 C.3 (BEPS Action 14) Publication of Mutual Agreement Procedure (MAP) profiles as of May 2022 Intermediate out- I.1 Membership in the Inclusive Framework on OECD IF member- comes BEPS as of November 2021 ship list d , EU Code of Conduct Group I.2 Membership in the Global Forum on Trans- Reports h , OECD parency and Exchange of Information for Tax Harmful Tax Prac- Purposes as of May, 2022 tices Peer Review Reportj I.3 Jurisdiction is greylisted by the EU in De- cember 2017 I.4 Jurisdiction is blacklisted by the EU in De- cember 2017 I.5 Jurisdiction is either grey or blacklisted by the EU in December 2017 a https://www.oecd.org/tax/automatic-exchange/crs-implementation-and-assistance/ crs-by-jurisdiction/ b https://www.oecd.org/tax/exchange-of-tax-information/Status_of_convention.pdf c https://www.oecd.org/tax/transparency/exchange-of-information-on-request/ ratings/ d https://www.oecd.org/tax/beps/inclusive-framework-on-beps-composition.pdf e https://www.oecd.org/tax/automatic-exchange/about-automatic-exchange/ CbC-MCAA-Signatories.pdf f http://www.oecd.org/tax/treaties/beps-mli-signatories-and-parties.pdf g https://www.oecd.org/tax/dispute/country-map-profiles.htm h https://www.consilium.europa.eu/en/policies/eu-list-of-non-cooperative-jurisdictions/ timeline-eu-list-of-non-cooperative-jurisdictions i https://data.consilium.europa.eu/doc/document/ST-9639-2018-REV-4/en/pdf j https://www.oecd.org/tax/beps/harmful-tax-practices-peer-review-results-on-preferential-regimes. pdf 16 being listed in the first place. If we observed an increase in compliance following selection, but no jurisdiction ended up on the grey or blacklists, we might infer that it was purely driven by the threat of being listed. The second channel is through the grey/blacklisting itself. If being listed is costly for jurisdictions, they will begin to comply to get off of it. If compliance only occurred after listing and only for listed jurisdictions, then we might infer that it was the listing itself that drove behavior. In practice, because most of the outcomes are observed at a single point in time, it will be difficult to disentangle these two effects. However, it is worth noting two things: (i) very few jurisdictions end up on the blacklist itself, suggesting most jurisdictions either comply early or agree with the EU that they will comply by a certain date, and (ii) as I will describe in Section 4, there is evidence that jurisdictions began to adopt some of these standards before the listing process actually began. For outcomes which are measured with a higher degree of frequency (as is the case with the economic outcomes considered in 5), a difference-in-difference strategy will be used to investigate the impact of the listing process separately than the review process. 3.6 Validity of RD 3.6.1 Density tests Density tests of the two main running variables are presented in Figure A1 in the Online Appendix. A standard test for manipulation of the running variable fails to reject the null of no manipulation in each case. It would be unusual for there to be signs of manipulation in this context, as the EU’s construction of its percentage scores mechanically smooths the distribution, making any bunching near a particular cutoff unlikely. 3.6.2 Balance One way we can be more assured that the regression discontinuity approach is valid is to check for balance in our outcomes of interest and for a variety of economic indicators prior to the EU selection process. Tables A2 and A3 in the Online Appendix check balance for pre-treatment outcomes of interest. This includes all the measures used by the EU to construct its three selection indicators,19 the selection indicators themselves, as well as Log(GDP per capita) from 2015 and, when available, the jurisdiction’s score from the 2015 edition of the Tax Justice Network’s Financial Secrecy Index (FSI), and the jurisdictions corporate income tax (CIT) rate as of 2015. While there is good balance across most indicators for the stability factors cutoff, the multidimensional cutoff has a small imbalance in Log(GDP per capita), so I will be including it as a control in my main specification. The stability factors cutoff does not display this imbalance. 19 These are indicators sourced from Eurostat, the IMF, and UNCTAD, and are averaged over the five year period preceding the EU selection (2011-2016). 17 3.7 Pseudo-cutoffs I also test for effects using alternate, pseudo cutoffs for three of the main robust outcomes of interest (grey/blacklisting, IF membership and whether a jurisdiction successfully has its preferential regimes reviewed). For each of these I find no consistent evidence that effects are present when I am using an alternate cutoff. These are presented in Figure A2 in the Online Appendix. 3.8 First stage performance As discussed above, being selected by the EU implies many possible treatments. Juris- dictions that are selected for review are then assessed for risk factors which determine which to prioritize. Then, through their subsequent interactions with the EU, ‘risky’ ju- risdictions decide whether they are willing to commit to implementing the above reforms. Those that do so are added to the EU’s greylist until the reforms are complete (or they are upgraded) and those that do not are blacklisted. Because the selection between the two lists is endogenous, the treatment effect implied by crossing the threshold captures the entire impact of the EU review process, including some probability the jurisdiction was eventually listed. The top panel (i) of Figure 4 displays the results of the local linear RD specification estimating the reduced form effect of crossing the MD threshold on the probability of being selected for review and on subsequent grey or blacklisting. Note that the probability of selection is only lower than one in the multidimensional specification due to the inclusion of LDCs in the sample. Jurisdictions that cross the MD threshold are 60-75 percentage points more likely to be subsequently grey or blacklisted. When selection is instrumented for, the point coefficients indicate that grey or blacklisting is all but determined by selection.20 In addition to these results, Figures A6 and A7 in the Online Appendix demonstrate that crossing the multidimensional threshold also results in a much higher chance that the EU communicated to the jurisdiction about their deficiencies ahead of their subsequent blacklisting. 3.9 Differences-in-differences estimation and economic outcomes For a small subset of governance outcomes and for all economic outcomes I will use a differences-in-differences approach. This will have a few advantages. First, it will allow me to estimate treatment effects that are no longer restricted to being around the regression discontinuity cutoff, thus incorporating effects on more tax havens (which tend to be underrepresented within the usual RD bandwidth). Second, it will allow me to differentiate the impact of the listing process more carefully from the impact of the review process. Finally, it will allow for the inclusion of European jurisdictions, including European tax havens, as controls. For estimations where the treatment of interest is being 20 The SF specification also shows strong and significant results, albeit slightly smaller than the multi- dimensional specification. 18 Figure 4: Impact of crossing the selection threshold on first stage outcomes and main tax governance outcomes (i) First stage outcomes (a) Selected for review by EU (b) Black or greylisted by EU (ii) Tax Governance Outcomes (c) Tax transparency (d) Fair taxation (e) Anti-BEPS Notes: Each figure shows the results of a local linear regression-discontinuity estimate, without controls, of the (reduced form) effect of crossing the EU selection threshold on (a) selection into the review process and (b) being grey or blacklisted anytime between December, 2017 and February 2022, and the three main tax governance outcomes (c) tax transparency (d) fair taxation and (e) anti-BEPs policies. Actual values are shown in Columns (1) of Table 3). Running variable is the multidimensional (MD) cutoff described above. 90% confidence intervals shown. Bins chosen using mimicking variance evenly-spaced (ESMV) method (Calonico, Cattaneo, and Titiunik 2015). added to either the EU’s grey or blacklist, I will use the doubly-robust method outlined by (Callaway and Sant’Anna 2021) to account for weighting and bias issues generated in difference-in-difference settings with staggered treatment. The first outcome I will consider is offshore deposits controlled by foreigners, as defined and published by the Bank of International Settlement’s Locational Banking Statistics (LBS). Foreign-held deposits have been shown in a large number of contexts to react to tax transparency policy (Casi, Spengel, and Stage 2019), and so both the threat of the upcoming adoption of these policies (as suggested by being added to the grey list) or scrutiny from jurisdictions back home (as suggested by holding assets in a blacklisted jurisdiction) might reduce the incentive to hold deposits in targeted jurisdictions. The second is both the stock and flow of foreign direct investment (FDI), as measured by UNCTAD. This is to account both for changes in FDI that might manifest if (i) the stigma of the listing leads firms to do business elsewhere, (ii) uncertainty around or 19 changes in tax policy make inward investment less worthwhile (Jardet, Jude, and Chinn 2022), or (iii) the increased scrutiny of listed jurisdictions by European tax authorities make investment by European firms less attractive. As a significant amount of both inward and outward FDI in tax havens constitutes ‘phantom FDI,’ (Damgaard, Elkjaer, and Johannesen 2019) cross-border positions that involve little economic substance, these positions should be particularly sensitive to listing. Finally, I look at two outcomes intended to capture an end-goal of the EU listing process: the degree to which firms operating in targeted jurisdictions enjoy low effective tax rates, and the degree to which those jurisdictions play host to shifted profits. I will discuss these data below in Section 5. 4 Results on tax governance outcomes The bottom panel of Figure 4 displays the main results of the RD specification for each of the three main tax-governance outcomes: while there is weak evidence of small effects on Tax Transparency and Anti-BEPs outcomes, they are not statistically significant. However we see large effects on the Fair Taxation outcome, which is wholly defined by being reviewed by either the EU or OECD for harmful tax practices and, at present, having no harmful regimes in operation. Table 3 shows both the reduced form and fuzzy RD results for both cutoffs with and without controls. For Tax Transparency, selection by the EU into the review process changes the average result by between 0.24 and 0.43, depending on the specification, none of which are significant. It is worth noting that in the MD specification, selection into the EU review process has a large, statistically significant impact on being compliant with EOIR standards, of up to 76 percentage points. The estimated effects are much smaller and insignificant for the SF specification, which may reflect heterogeneity in the treatment effect. For Fair Taxation, selection increases the chance of having at least one regime reviewed no longer having harmful regimes present by near certainty21 For Anti-BEPS outcomes, selection increases the mean outcome by 22-33 percentage points, on average, although it is not significant. Selection into the EU review process has enormous implications for IF membership, increasing the probability of joining by between 60-100 percentage points, depending on the specification. As a result of this, jurisdictions selected for review were also highly likely to agree to the OECD/G20 two-pillar solution for international tax reform. By contrast, there is no robust evidence that selection increases the probability of signing up to the Global Forum, an alternative outcome I consider. Some of these outcomes, while statistically significant, are still sizable effects. In many circumstances (although not in the case of Fair Taxation) these results are slightly 21 Many of these results predict increases of more than 100 percentage points, an artifact of the linear probability model. 20 Table 3: Impact of EU review process on targeted outcomes Multidimensional Stability factors cutoff (MD) cutoff (SF) Reduced Fuzzy Reduced Fuzzy form RD form RD (1) (2) (3) (4) (5) (6) (7) (8) (A) Tax transparency: Mean outcome 0.32 0.14 0.43 0.24 0.24 0.15 0.29 0.27 (0.28) (0.24) (0.42) (0.39) (0.29) (0.32) (0.42) (0.46) (i) CRS committment by 2023 0.13 -0.13 0.23 -0.18 0.34 0.28 0.49 0.46 (0.34) (0.30) (0.50) (0.45) (0.27) (0.27) (0.38) (0.43) (ii) % of EU countries covered by AEOI 0.17 -0.055 0.19 -0.20 0.29 0.26 0.30 0.32 (0.29) (0.24) (0.46) (0.45) (0.28) (0.30) (0.43) (0.44) (iv) Largely compliant on EOIR 0.59** 0.55** 0.76** 0.84** 0.25 0.19 0.29 0.27 (0.26) (0.23) (0.36) (0.38) (0.31) (0.34) (0.45) (0.53) (iii) Signed up to MAA 0.40 0.31 0.60 0.47 0.068 -0.018 0.14 -0.056 (0.32) (0.28) (0.47) (0.49) (0.39) (0.39) (0.48) (0.50) (B) Fair taxation: Mean outcome 0.85*** 0.95*** 1.33*** 1.51*** 0.86** 0.88*** 1.20*** 1.22*** (0.20) (0.22) (0.14) (0.23) (0.43) (0.21) (0.19) (0.26) (i) Reviewed w/ no harmful regimes 0.85*** 0.95*** 1.33*** 1.51*** 0.86** 0.88*** 1.20*** 1.22*** (0.20) (0.22) (0.14) (0.23) (0.43) (0.21) (0.19) (0.26) (C) Anti-BEPS: Mean outcome 0.20 0.088 0.33 0.13 0.21 0.29 0.29 0.32 (0.25) (0.20) (0.41) (0.31) (0.30) (0.29) (0.36) (0.36) (i) Signed up to Cbcr 0.35 0.16 0.56* 0.23 0.24 0.25 0.41 0.34 (0.24) (0.20) (0.31) (0.28) (0.36) (0.38) (0.50) (0.48) (ii) Signed up to MLI 0.0035 -0.058 -0.12 -0.13 -0.14 -0.13 -0.21 -0.11 (0.33) (0.31) (0.60) (0.50) (0.41) (0.46) (0.58) (0.60) (iii) Published MAP profile 0.24 0.089 0.55 0.24 0.60 0.74** 0.67 0.70 (0.26) (0.24) (0.55) (0.44) (0.57) (0.35) (0.52) (0.46) Intermediate outcomes: Participation in IF 0.59** 0.72*** 1.08* 1.29** 1.17*** 1.20*** 1.57* 1.50** (0.23) (0.25) (0.61) (0.62) (0.33) (0.34) (0.81) (0.67) Signed up to global deal 0.57** 0.67*** 1.05* 1.13* 1.17*** 1.20*** 1.54* 1.49** (0.23) (0.24) (0.57) (0.57) (0.33) (0.34) (0.83) (0.66) Signed up to Global Forum -0.015 0.063 0.012 -0.0084 0.61 0.70 0.82 0.89 (0.23) (0.19) (0.34) (0.35) (0.48) (0.49) (0.65) (0.70) Ever greylisted 0.72*** 0.59*** 1.08*** 1.00*** 0.47 0.44 0.89*** 0.76*** (0.17) (0.16) (0.058) (0.11) (0.32) (0.27) (0.12) (0.20) Ever blacklisted 0.19 0.033 0.36 -0.00086 0.081 0.0018 0.13 0.033 (0.20) (0.19) (0.32) (0.29) (0.22) (0.21) (0.31) (0.30) Ever grey or blacklisted 0.77*** 0.60*** 1.10*** 0.98*** 0.60* 0.54* 0.97*** 0.88*** (0.17) (0.17) (0.058) (0.11) (0.31) (0.28) (0.057) (0.16) Other outcomes: Number of minimum standards 1.21 0.98 2.17* 1.42 1.61 1.96** 2.09* 2.24* (0.83) (0.68) (1.27) (0.98) (1.16) (0.96) (1.11) (1.19) Controls X X X X Observations 194 194 194 194 177 177 177 177 Avg Effective # Obs 75 64 44 57 43 40 43 45 Avg Bandwidth 15.1 13.6 8.68 11.8 9.68 9.06 9.71 10.2 Max Bandwidth 22 21 11 15 14 14 16 14 Min Bandwidth 12 10 7 9 6 6 8 7 Notes: Each cell shows the estimated impact of selection into the EU review process on a different outcome, using bandwidth-selection, bias-corrected, robust methods outlined in Calonico, Cattaneo, and Titiunik (2014), Calonico, Cattaneo, and Farrell (2018) and Calonico, Cattaneo, Farrell, and Titiunik (2019). Each column indicates a different specification, and each column pair indicates a different running variable (e.g. stability factors, multidimensional). Mean outcome indicates the average of all sub-outcomes listed under a category. Fuzzy RD results use review by the EU Code of Conduct Group as the treatment, instrumented by crossing the RD threshold. Reduced form results ignore actual assignment to treatment and just use the crossing of the RD threshold as treatment. ∗ p < 0.10,∗∗ p < 0.05,∗∗∗ p < 0.01 21 underpowered, with minimum detectable effects (MDEs) of around 0.34-1 for the mean outcomes at 80% power and 90% confidence.22 However, it is worth pointing out that only the Fair Taxation result shows such a high degree of stability across these specifications (as well as alternate specifications in the Online Appendix). Still, it is possible that choosing mean outcomes sets the bar too high relative to the desired outcomes of the EU process. Recall that the EU review process was designed to uncover deficits in standards. For example, some jurisdictions may not have signed up to country-by-country reporting, where others may not have implemented AEOI. Focusing on average improvements across all indicators may be ignoring these improvements on the margin. To set a lower bar, as described above, I also included an indicator of how many proxies for the four minimum standards a jurisdiction has implemented: review and non-existence of harmful tax practices, CbCR, the MLI position, or published a MAP profile. While this does not, in itself, improve power, we might expect the impact of the EU review to have a much larger impact on this outcome. The results using the MD specification suggest this is not the case: it finds that selection increases the number of minimum standard met by 1-1.46 percentage points, a result that is not significant. However, the SF specification indicates that selection may have had an impact of roughly 2.4 minimum standards. Examining the results from each individual outcome suggest this is primarily being driven by the harmful regime outcome (B.1) and the publication of MAP profiles (C.3). Why are these results different for the stability factors cutoff? Recall that, while the MD specification estimates a frontier average treatment effect (FATE), a weighted average of each of the three selection indicators, the SF specification only estimates a local average treatment effect (LATE) for jurisdictions on either side of that particular cutoff. This suggests there is some heterogeneity in the effect of the EU selection over different jurisdictions. One possibility is that jurisdictions close to the stability factors cutoff were initially more responsive to EU attention: the estimated impact of EU selection on grey/blacklisting in the SF specification is lower, suggesting that jurisdictions around this cutoff may have done more to improve their tax governance outcomes to avoid being listed. Table A6 in the Online Appendix shows the same results using a global RD specifica- tion with varying polynomial functions, as well as a flexible specification where each EU criteria is entered in as a separate function (a la Dell (2010)). In these results, the impact of selection in the EU review process on Fair Taxation is significant and robust across nearly every specification, a result driven entirely by the impact of the EU process on participation in the Inclusive Framework. While at times both the Tax Transparency and 22 A competing concern is that the small number of effective observations in many of these regressions may lead to artificially-small standard errors. In Table A4 in the Online Appendix I show that the results are robust to the use of the wild bootstrap procedure from Cameron, Gelbach, and Miller (2008) implemented using the Stata module boottest| (Roodman, Nielsen, MacKinnon, and Webb 2019). 22 Figure 5: Impact on Inclusive Framework membership across time (a) January, 2017 (b) November, 2021 (c) RD results across time (d) Event study estimates of impact of selection Notes: Subfigures (a) and (b) shows the robust, bias-corrected estimate of the effect of crossing the RD threshold on the probability of being a member of the Inclusive Framework both before and after the review period. Subfigure (c) shows the impact for every period available in the sample for both the multidimensional and the stability factors cutoffs. Red shaded area indicates the period during which the EU reviewed selected jurisdictions and issued communications to invite jurisdictions to adopt better tax governance standards. Blue shaded area indicates the beginning of the grey and blacklisting period. Left column displays estimates using multidimensional cutoff and right column displays estimates using the stability-factors cutoff. 90% confidence intervals. Subfigure (d) shows estimates from event study specification (equation (5)) using monthly data. Standard errors are clustered at the jurisdiction level. 95% confidence intervals shown. Anti-BEPS outcomes are both positive and significant, they are not consistently so across the various specifications. Only the Fair Taxation and Inclusive Framework outcomes show a high degree of consistency. 23 4.1 Effects over time Given that selection into the EU review process implied several different treatments occur- ring at different stages, it is worth exploring the dynamics of these effects. For example, Figure 5 displays the basic local linear RD result of the impact of crossing the MD thresh- old on Inclusive Framework membership in January 2017, as the EU review process kicked off, versus the end of 2019. From this graph, it is clear that there has been a substantial increase in IF membership for jurisdictions just to the right of the cutoff, but little change on the left. To dig into this further, in this section, I look at membership in the Inclusive Frame- work at several discrete points in time, constructing a panel of IF membership using information from OECD-published membership lists, OECD announcements of specific join dates, and reports from several of the “Big Four” consultancy firms. Figure 5(c) displays the results from this exercise for each of the three outcomes, using both the MD and SF specifications. The period of the EU review process is highlighted in red, after which the EU began publishing its grey and blacklists. The impact of crossing the threshold on IF membership is statistically indistinguishable from zero prior to the selection process.23 However, during the review process the effect begins to increase and is large and significantly positive during the summer of 2017, several months prior to the publication of an EU list. This was the period during which the EU was intensely reviewing jurisdictions to determine if they would be eligible for listing, suggesting that it was the EU review process itself, as well as the upcoming threat of the grey and blacklist publication, that induced jurisdictions to join the Inclusive Framework. As discussed above, the regression discontinuity estimates provide an estimate around the relevant cutoff, but no further. To investigate the general impact of the review process, I consider a standard event-study design framework: −2 ∑ ∑ L Yit = αi + αt + βj 1{t − t∗selected = j } + βj 1{t − t∗selected = j } + ϵit (4) j =−K j =0 where Yit is the outcome of interest for jurisdiction j at time t and αi and αt are jurisdiction and period fixed effects, respectively. {t − t∗selected = j } are event study dummies equal to one when an observation is j periods prior to its treatment, and zero if not or if the jurisdiction was never treated. For the treatment of interest (being selected into the EU review process), treatment happens concurrently for all treated units, so all the identifying variation comes from comparisons between selected jurisdictions and non- selected jurisdictions. Thus, the identifying assumption is that there are parallel trends between these two units. Figure 5(d) presents the results from the impact of the EU review process on Inclusive 23 There are some signs of imbalance in the MD specification prior to the review process, but these recenter around zero as of January, 2017, which is when the review process formally began. 24 Framework membership. The few pre-period estimates we can observe suggest little in the way of parallel trends. Following the beginning of the treatment period (November 2016, when the screening criteria were announced, following by actual screening in January), the Inclusive Framework membership rate of those selected begins to increase, exceeding 30 percentage points by the end of the series (two years later on). Note that the estimated effect is roughly 20 percentage points at 11-12 months following the selection into the EU review process, which is when the listing began. The fact that an effect opens up even before grey and blacklisting starts is in line with the RD results across time presented in the previous section. To understand the average impact, I run a standard two-way fixed effects difference- in-difference model of the form: Yit = αi + αt + βSelectedi + ϵit (5) The results from this specification are presented in Table 4. The average difference-in- difference effect, presented in column (1) is a 27 percentage point increase in the probabil- ity of Inclusive Framework Membership. Columns (2) and (3) show this effect oscillates between 16-31 percentage points if I restrict the sample to the same rough bandwidths used in estimating the RD results above (this essentially compares the evolution of IF membership for jurisdictions just above and below the two RD cutoffs). Given the limita- tions of DiD in this setting, these results should be taken as indicative but not conclusive evidence of the average treatment effect on the treated (ATT). Columns (4)-(6) consider the change if we restrict the treatment group to jurisdictions that were selected but not ultimately listed as of December 2017, those that were selected and then greylisted, and those that were selected and then blacklisted. In all cases the control group are those that were not selected. These groups are non-random, so some care must be exercised in interpreting these coefficients: jurisdictions that were selected by the EU for review and would go on to be grey or blacklisted (columns (5)-(6)) saw a faster relative increase in Inclusive Framework membership than those who were selected for review but were not ultimate listed (Column (4)). This framework presents several challenges: many of the tax governance outcomes of interest became actionable only shortly before or even during the EU review process. For example, while some jurisdictions privately committed to joining a few months prior, the first official meeting and public list of Inclusive Framework Members occurred at the end of June 2016, just 4-6 months before the screening process began. This limits my ability to establish parallel trends over a long period, as the outcome of interest (IF membership) effectively did not exist before this period. The MLI was only adopted in late 2016, making comparisons of jurisdictions prior to the screening process impossible. Both CbCR and MAA adoption show clear signs of a failure of parallel trends assumption, or not enough lead time to establish causal effects. The same is true for difference-in-difference estimates of the effect of the grey or blacklisting itself. The one outcome that does not display clear 25 Table 4: Difference-in-difference estimates of impact of EU selection/review on Inclusive Framework membership (1) (2) (3) (4) (5) (6) Multi-dimensional Stability factors Treatment = Treatment = Treatment = Full bandwidth bandwidth reviewed but reviewed and reviewed and sample <= 16 <= 10 not listed greylisted blacklisted Selected for review by EU 0.27*** 0.16** 0.31** 0.16** 0.30*** 0.39*** (0.043) (0.075) (0.12) (0.063) (0.057) (0.090) R2 0.827 0.833 0.817 0.908 0.844 0.872 Observations 9,585 3,555 2,025 6,705 7,560 6,210 # jurisdictions 213 79 45 149 168 138 Notes: Table presents different-in-difference estimates of impact of being reviwed by the EU process on the probability a jurisdiction joins the inclusive framework. Columns (2) and (3) restrict the sample to the same rough bandwidth used in the regression discontinuity design. Column (4) restricts the treated sample to jurisdictions that are reviewed but not (initially) subsequently listed in Dec 2017. Column (5) restricts the treated sample to those that are reviewed and subsequently greylisted in Dec 2017. Column (6) restricts the treated sample to those that are reviewed and subsequently blacklisted in Dec 2017. Standard errors clustered at the jurisdiction level. ∗ p < 0.10,∗∗ p < 0.05,∗∗∗ p < 0.01 pre-trends, CbCR commitment, at best shows small long term effects of a 20% percentage point increase, in line with the RD estimates (albeit significant)24 . 5 Economic impacts and allocation of profits Figure A11 in the appendix show the staggered difference-in-difference estimates (using the doubly-robust method from Callaway and Sant’Anna (2021) ) of the impact of listing on foreign-held deposits and both the stock and flow of foreign direct investment (as a % of GDP). There is no evidence of any shift following the listing of a jurisdiction. If anything, there is a large, but insignificant, positive impact of listing on foreign-held deposits, running contrary to the expected impact. Table 5 shows the full range estimates using (i) an ordinary difference-in-difference estimation, (ii) the two-stage difference in difference estimator from Gardner (2021), and (iii) the doubly-robust method from Callaway and Sant’Anna (2021). It also shows sep- arate results for when the treatment indicator is the first time a jurisdiction is added to either list, the first titme it is greylisted, or the first time it is blacklisted. For the FDI results, which encompass a large number of jurisdictions, it also estimates the results when the sample is restricted to jurisdictions within the average regression discontinuity bandwidth. Across these specifications, there is no significant evidence that listing leads to a decline in any of the main outcomes of interest. 5.1 The location of profits and assets Looking beyond immediate policy and economic impacts, did the introduction of the blacklist curb profit shifting to tax havens in aggregate? The dearth of high quality panel data on profits means that only tentative answers are available. Figure 6 displays 24 See Figure A5 in the Online Appendix 26 Figure 6: Profits booked in tax havens before and after the EU blacklisting began Notes: Figure shows aggregate estimates of artificially-shifted profits, foreign and local profits for all tax havens across three categories. Estimates taken from (Tørsløv, Wier, and Zucman 2022; Wier and Zucman 2022). estimates of aggregate shifted profits, foreign profits and local profits taken from Tørsløv, Wier, and Zucman (2022) for tax haven jurisdictions for the period 2015-2018. These are divided by (i) those that were blacklisted anytime between the end of 2017 and mid-2018, (iii) those that were greylisted but never blacklisted during this period and (iii) those that were never put on a list during this time. Jurisdictions that escaped listing continued their upward trend on the amount of artificially-shifted profit that was declared there. Those that were greylisted actually saw an increase following 2017, despite having pledged to adopt new policies that should, in the long term, make them less attractive for parking profits. Blacklisted countries saw a slight decline. This could be a sign that blacklisting does make jurisdictions less attractive for multinationals to park profit, but more research is needed to fully understand if this effect is causal. Finally, Figure 6 makes it clear that regardless of the year being considered, most shifted profits are allocated to jurisdictions that never ended up on the grey nor blacklist. 5.2 Effective tax rates paid by European banks in tax havens Finally, I examine whether the EU’s review or listing of jurisdiction has had an impact on effective tax rates. Multinationals have faced falling effective tax rates over the past 27 two decades, a result of both the increased booking of profits in tax havens as well as tax competition between different jurisdictions (Garcia-Bernardo, Janskỳ, and Tørsløv 2022). Panel data on effective tax rates is scarce, particularly following the period that the EU listing process kicked off. Efforts to produce country-by-country estimates using firm level data, such as ORBIS, are beholden to the fact that it takes a few years before coverage is substantial enough to produce meaningful estimates. Estimates for EU MNEs would be possible using EU CbCR data were it available, but as of yet aggregate data has not been made possible past 2017. However, as part of the Capital Requirements Directive IV, EU banking groups with greater than €750m in consolidated turnover are required to publish CbCR data, and previous research (cite) has combined and published this information for the period of 2014-2020. Following the approach outlined in Aliprandi, Baraké, and Chouc (2021), jurisdiction-by-year effective tax rates can be calculated using the following formula: ∑ T axesit ET Rir = ∑ (6) P rof itsit where T axesit indicate the aggregate taxes paid by all reporting banks in a given jurisdiction i in year t and P rof itsit indicate all profits declared by all reporting banks in the same jurisdiction-year.25 I contrast this with average statutory corporate tax rates, as measured by the Tax Foundation.26 Figure 7 shows event-study estimates from estimating, in turn, the impact of a juris- diction being selected into review by the COCG, being greylisted, or being blacklisted, on average statutory corporate tax rates and the effective tax rates paid by European banks. There is no discernible effect of the review on either statutory or effective rates. Greylist- ing has a positive and significant (4.6 percentage points in the post-treatment period) on statutory rates, but no apparent effect on the effective rates paid by banks. This is consistent with jurisdictions raising their rates in response to the attention generated by the list, but use other means to reduce the tax that (banks) ultimately have to pay. By contrast, blacklisting has no significant impact on either rate (although for the effective tax rate measure, parallel trends do not appear to hold, so the results themselves are not easily interpreted). 25 Banks that declare negative profits for a given jurisdiction year are excluded from this calculation. 26 The Tax Foundation’s database of statutory CRRs is an amalgamation of OECD, KPMG and Bloom- burg reports, as well as individual research. https://taxfoundation.org/publications/corporate-tax-rates- around-the-world/ 28 Figure 7: Impact of the EU review/listing on corporate tax rates Notes: Figure shoes event-study estimates of the impact of (i) the EU COCG review, (ii) the first greylisting or (iii) the first blacklisting of a jurisdiction on its average statutory corporate tax rates or the effective tax rate paid by European banks. All estimates use Callaway and Sant’Anna (2021)’s doubly-robust difference in difference method. 95% confidence intervals. 6 Discussion of the EU listing exercise’s impact 6.1 Impact on the composition of the IF membership and implications for future global tax rules One of the few consistently-robust results from the above exercise is that the EU review process and the threat of the blacklist has had a positive impact on the probability that affected jurisdictions join the Inclusive Framework. How different is the Inclusive Framework thanks to the EU’s efforts? One limitation of the RD approach is that, while the impact of the review process had at least an estimate impact of around 70% percentage points in membership around the threshold in the RD specification, we are unable to say much about the impact on jurisdictions away from that threshold. We can, however, take the ITT estimates from the difference-in-difference estimation as a tentative lower bound of the effect of the EU selection on Inclusive Framework membership. At the time of writing, the IF is currently considering the implementation of two major reforms to international tax rules. The first of these, Pillar I, aims to reallocate taxing rights over non-routine profits to market jurisdictions, based on sales in those jurisdictions. Pillar II aims to establish a global minimum tax regime. As of early 2020, the IF has endorsed framework for Pillar I but has yet to finalize the design of Pillar II. The ongoing COVID-19 Pandemic has led to a postponement of the next Inclusive Framework meeting from the summer of 2020 until October. Let us ponder, for a moment, the impact a EU-driven 27% increase in the probability of Inclusive Framework membership would have had on the membership. 92 jurisdictions 29 were ultimately reviewed by the European Union. A 27% increase in IF take up implies that roughly 25 jurisdictions joined the Inclusive Framework directly as a result of the peer review process. Back-of-the-envelope calculations suggest that this has not led to a large shift in the composition of the Inclusive Framework. For example, assuming no heterogeneity in the EU selection treatment effect, the average GDP per capita (in nominal 2018 dollars) for an IF member is only $370 lower than it would have been without the EU’s involvement. The average Tax Justice Network Financial Secrecy Index value of IF members is currently 291, versus 288 in the counterfactual. The EU review and listing process has made the IF more representative of poorer countries and those who potential contribute to global profit shifting, but only slightly. The EU made two choices in designing its review process that have arguably had a large impact on the IF’s composition. The first is in the setting of its thresholds for selection into the review process. The median GDP per capita, in 2018 nominal dollars, of jurisdictions within 10 percentile points of the cutoff is approximately $4,200. The second is excluding least developed countries (LDCs) from the review process. Taken together, a significant number of developing countries that might have been induced to take part in the Inclusive Framework are left out of the current negotiations. One of the factors that led the EU to exclude LDCs and set thresholds where it did was the fact that many developing countries do not currently have the capacity to implement the BEPS minimum standards. However, while it is not clear that the EU review process would have led the same pressure to join as it did for many tax havens, one potential benefit ceded by the design of the review process was greater involvement in the IF by developing countries, although there is an ongoing debate as whether their participation in the IF has translated into greater taxing right for developing countries (Christians and Van Apeldoorn 2018; Hearson 2020). Despite this, the EU process also increased the total number of jurisdictions - including developing countries - that ultimately endorsed the Two Pillar Solution. 6.2 Eliminating harmful tax regimes One important outcome of the BEPS process and a focus of the EU listing effort is the reduction of harmful tax practices. The results in the previous section suggest that the EU selection and listing process led to an increase in the number of jurisdictions that have been cleared of any harmful tax practices. But what impact did it have on the actual number of harmful regimes that were overturned? The EU claims that “over 120 harmful regimes have been eliminated worldwide, thanks to the EU listing requirements.”27 This is a difficult assertion to test, due to the way we observe the presence of harmful tax regimes. Currently, the data only allow us to observe the evolution of harmful regimes that are reviewed by the EU COCG and the OECD 27 Source: https://ec.europa.eu/commission/presscorner/detail/en/ip_20_262 30 Figure 8: The impact of EU selection on the probability of detection and rollback of harmful regimes (a) Reviews conducted by the EU Code of Conduct Group (b) Reviews conducted by the OECD Forum on Harmful Tax Practices Notes: Each figure shows the results of a local linear regression-discontinuity estimate, without controls, of the (reduced form) effect of crossing the EU selection threshold on the probability a jurisdiction has at least one preferential regime reviewed, the probability that at least one harmful regime is detected during a review, and the probability that at least one regime has been rolled back after a review. Panel (a) uses data solely on reviews conducted by the EU’s Code of Conduct Group. Panel (b) uses data solely on reviews conducted by the OECD Foreum on Harmful Tax Practices. Running variable is the multidimensional cutoff described above. 90% confidence intervals shown. Bins chosen using mimicking variance evenly-spaced (ESMV) method (Calonico, Cattaneo, and Titiunik 2015). FHTP, but not those that were not subject to review. It is impossible to know the precise impact of the EU’s efforts without observing the entire universe of regimes across all jurisdictions, as some jurisdictions that were not selected may have chosen to amend or abolish harmful tax regimes. The EU’s statement is based on the decisions by reviewed or listed jurisdictions to comply, but we do not know what compliance would have looked like in a world in which those reviews would not have taken place, because we would not have been able to observe it. This complicates our ability to say anything concrete about the causal impact of the EU process on the number of harmful regimes that have been rolled back, either using the regression discontinuity framework or even exploiting changes over time. Stepping away from the net impact of its own review process, whatever the impact the EU exercise has had on the number of harmful regimes struck down, it extends beyond the efforts of the Code of Conduct Group itself. To illustrate this, consider Figure 8, which shows that crossing the multidimensional cutoff significantly increases the probability that a jurisdiction is (i) reviewed, (ii) is reviewed and has at least one harmful regime detected and (iii) is reviewed and rolls back at least one harmful regime that was detected in a review. The Figure shows the results separately for reviews conducted by the EU Code of Conduct Group (Panel A) and the OECD Forum on Harmful Tax Practices (Panel B). 31 Fuzzy RD results with bias-correction and controls are presented in Table 6. The results indicate that selection into the EU review-and-listing exercise led to an increase in the probability of review by both the EU COCG and the OECD FHTP by roughly the same degree. If anything, the effects on the probability of OECD FHTP are slightly stronger. Results for the detection of harmful regimes and rollbacks of regimes are not generally significant, but consistent with a world in which the EU’s review process led to a crowding in of effort by the OECD. This is likely generated by the fact that the EU exercise sharply increased the proba- bility of a jurisdiction joining the Inclusive Framework, which itself leads to a review by the FHTP. A reasonable lower bound of the net effect of this spillover can be constructed if we make some basic assumptions about the impact of IF membership on the detection of harmful tax regimes. Currently, the latest report from the Inclusive Framework on the review of harmful tax practices revealed that roughly 205 regimes were found to be harmful or were amended too not be harmful, or have been abolished or are in the process of being so.28 That suggests that for each IF member, 205 141 ≈ 1.45 harmful tax regimes are detected and amended or struck down. If the EU review process led to 25 additional jurisdictions joining the IF, this implies - holding the detection rate constant - that an additional 36 regimes, or 17% of the total reviewed by the FHTP, were detected and reformed as a result. This is a lower band for two reasons: I am using the lowest bound estimate of the impact on IF membership and I am not counting any additional impacts the EU listing process may have had beyond the increase in IF members, such as the efforts by the COCG. 6.3 Limitations of the RD approach and possible heterogeneity in treat- ment effects One potential limitation of the research design in this paper is the fact that a regression discontinuity design is useful for identifying effects in the immediate area of the cutoff. This allows me to identify what happens when there is a large, discontinuous change in EU pressure for otherwise-similar jurisdictions. There are two-related concerns with this approach. The first is that there may be little actual effort by the EU to pressure jurisdictions close to the cutoff relative to those further away. The second is that, even if the EU is pressuring jurisdictions close to the cutoff, it may be putting even more effort into pressuring jurisdictions further away from the cutoff. This is a reasonable concern to have: in Figure A9 in the Online Appendix I show that most ‘traditional’ tax havens are not well-represented within the bandwidth of the RD estimate.29 If the European Union decides to exert more effort on this small number of 28 https://www.oecd.org/tax/beps/harmful-tax-practices-peer-review-results-on-preferential-regimes. pdf 29 For my measure of traditional tax havens, I use the list of jurisdictions blacklisted by the EU in 2015, a list compiled by Johannesen and Zucman (2014) as well as a list of OFCs constructed by Garcia-Bernardo, Fichtner, Takes, and Heemskerk (2017). All three show higher levels of representation further outside the typical bandwidth. 32 jurisdictions, particularly those with closer ties and a higher impact on the tax base of EU member states, the RD estimates may be understanding the true effect.30 Conversely, if these jurisdictions receive more bilateral pressure and pressure from organizations like the OECD, the additional impact of EU effort may actually be lower than it is for jurisdictions around the cutoff. The first concern should be mitigated by the fact that I find significant effects on the probability of being listed (shown in Figure 4) and on the probability of the EU reviewing the jurisdiction’s tax regimes. In addition to this, in Figures A5 and A6 in the Online Appendix I show that both the risk criteria the COCG used to prioritize investigation and deficiencies they discovered and communicated to jurisdictions are largely positive and significant in the area around the cutoff, indicating that the EU was paying attention to these jurisdictions. To investigate the second concern (higher levels of effort away from the cutoff), I consider several different indicators. First, in Figure A6 in the Online Appendix, I consider how the three risk indicators that the COCG used to prioritize their investi- gation evolve as we move further away from the cutoff. There, I do find evidence that the COCG was somewhat likely to flag jurisdictions outside of the RD bandwidth for transparency/exchange-of-information problems than those inside, although that likeli- hood collapses as the distance reaches its furthest point. I find strong evidence that the COCG was more likely to flag a jurisdiction for having no effective corporate income tax, but no evidence they were more likely to flag a jurisdiction for having a preferential tax regime. In Tables A6 and A7 I show that the EU was just as likely to pressure jurisdic- tions within the RD bandwidth into addressing deficiencies, although there are a few areas (AEOI, MCAA and Global Forum participation) where the EU was slightly less likely to bring up deficiencies for those within the bandwidth than those immediately outside. These are only indicators of risk, not subsequent effort, but we do observe effort by the EU on one dimension: the review of potentially harmful tax regimes by the Code of Conduct Group (COCG). Figure 9 displays the probability of a jurisdiction having at least one regime reviewed by the COCG between 2015-2019. It shows this across both of the main running variables I use in this paper (the MD and SF cutoffs, panels (a) and (b)) as well as across the jurisdiction’s ‘strength-of-ties’ measure (panel (c)) for jurisdictions that were selected into the EU review. For the multidimensional cutoff, there is not strong that the EU focused more on juris- dictions away from the cutoff point than those close - however there is a slight decline for those furthest from the cutoff. For the SF cutoff, however there are substantial increases in the probability a harmful regime is found for those further for the cutoff. One potential counterfactual is the probability of review by the OECD Forum on Harmful Tax Practices (FHTP), which could, far away from the cutoff, represent the ‘status quo’ of how likely it is a jurisdiction would be reviewed if the EU had not been 30 Evidence from analyses of information exchange agreements, for example, show that tax havens are more likely to sign up to agreements with jurisdictions with close economic ties (Bilicka and Fuest 2014). 33 involved. This counterfactual is imperfect because, as shown in previous sections, we know that the EU nudged many jurisdictions into the Inclusive Framework, which dramatically increased the probability they would be reviewed by the FHTP. But it is still worth investigating how FHTP effort behaves further from the cutoff: in most cases we see it tracking or exceeding that of the EU. This is consistent with a world in which global pressure on traditional tax havens exceeded that of the EU, particularly for the those which are the most economically significant. In summary, while there is ample evidence the EU did put pressure on jurisdictions close to the cutoff, I cannot fully rule out that its effort changed for jurisdictions further away. These concerns aside, is the set of jurisdictions for which the RD estimates are defined over a meaningful set of jurisdictions? As discussed above, they are less likely to be labelled as tax havens in the literature and are on average less economically important to the EU than those further away from the cutoff. Table 7 considers the jurisdictions that fall within the largest bandwidth estimated in all the RD specifications from the main analysis. Roughly 20% of these have been identified previously as a haven, either by the EU or in the literature. Almost 70% of these jurisdictions have been included in the latest edition of the Financial Secrecy Index that is published by the Tax Justice Network. Although many of these are not ranked high on the Index, roughly 25% of them are ranked in the top 30 and the average Secrecy Score (a measure of financial secrecy independent of the size of the jurisdiction’s financial sector) is on par with the average jurisdiction in the Financial Secrecy Index. Even if most of the jurisdictions within the RD bandwidth are not substantial players in the offshore economy, the limited gains observed in the RD estimates are still worth considering. Tax evasion and avoidance is often characterized by a ‘race to the bottom,’ where people and firms move their assets and operations away from jurisdictions that improve their governance to less scrupulous places (for example, the deposit shifting behavior observed in Johannesen and Zucman (2014)). Thus there are benefits to inducing jurisdictions to embrace tax policies which reduce the scope for negative cross-border externalities even if those jurisdictions have yet to develop into fully-fledged tax havens. 6.4 Limitations of the EU listing process In the long run, jurisdictions that have joined the Inclusive Framework will eventually need to improve their tax governance, until these efforts are superseded by the recent efforts by the OECD to establish new ground rules under Pillars I and II. While this analysis cannot fully rule out small improvements, in the medium term, it does not appear that the EU’s review and listing exercise has led to substantive changes in international tax governance for the group of jurisdictions that were originally targeted. Nor does it seem to have changed the economic reality for targeted jurisdictions, nor the propensity of multinationals to shift profits there. 34 Why might this be the case? There are a number of reasons why the whole listing exercise may not have had a large impact. 6.5 Poor targeting of the listing regime The goal of the EU listing exercise was to curb the revenue spillovers that tax havens had on European countries and the world more broadly. However, while the listing exercise itself has largely been focused on a jurisdiction’s adherence to tax transparency and BEPS minimum standards, it has not taken into account the full risk that a given jurisdiction poses to global tax revenue. If the goal is to reduce the externalities that jurisdictions pose, the listing process would need to prioritize those that are host to a greater share of hidden wealth or shifted profits. This does not appear to have been the case in practice. Figure 10 displays the proba- bility that a jurisdiction was reviewed, greylisted or blacklisted by the EU at any point in time, graphed against a number of different observable characteristics. The first, Figure 10(a), shows that across 50 jurisdictions that were home to shifted profits in 2015, those that hosted $100b worth of shifted profits were no more likely to be grey or blacklisted than those host to $100m. Of the estimated $1.1 trillion that was shifted to havens in 2018, approximately 1.3% went to blacklisted jurisdictions, 33% went to greylisted ju- risdictions, while 65% went to jurisdictions that appeared on neither list that year (as demonstrated in Figure 11. Slightly over a third of all shifted profits were located in EU countries, exempt from the listing process entirely. As can be seen in Figure 10(b) the EU also appeared to be no more likely to list jurisdictions that score high on measures of financial secrecy that take into account the size of the financial sector,31 itself a predictor of offshore deposits. This is concordant with the ongoing debate around the EU’s unwillingness to sanction some of the larger jurisdictions that are thought to be destinations for hidden wealth. For example, as it already acquires global taxpayer information on its residents and nonresidents through the Foreign Account Tax Compliance Act (FATCA), the US has been hesitant to adopt the CRS (Noked 2019). But while FATCA provides obvious benefits to the US, the lack of reciprocity in most of its inter-governmental agreements means that FATCA does not provide the EU with the same level of information it would obtain through the CRS on its own tax residents. Despite considering the US to not be providing a reciprocal form of AEOI, the EU has not yet added the country to its greylist, nor taken any other form of countermeasure. Poor countries were less likely to be reviewed or listed. This is by design, as least- developed countries (LDCs) were explicitly excluded from the review exercise. However, as can be seen in Figure 10 (c), poorer countries were much more likely to be listed conditional on being reviewed (the difference between the blue and the green/black lines 31 The Tax Justice Network’s Financial Secrecy Index is the product of a secrecy score and the size of that jurisdiction’s financial sector. 35 grows with income). This indicates that the EU was significantly more likely to list developing countries that did not qualify as LDCs. Slow movement on EU-specific countermeasures As described in Section 2, the EU stipulated a number of potential countermeasures that could be applied to listed jurisdictions. This included preventing EU funds, such as those from the EFSD and EFSI, from being channeled through entities in listed jurisdictions. While there was push-back from the European Parliament and from civil society groups on European Investment Bank funds being channeled through tax havens in the years leading up to the introduction of the new blacklist, it is unclear at this point in time what proportion of investment funds actually followed this route. Thus it is hard to know whether these restrictions are likely to have been damaging for listed jurisdictions. When the blacklist was first released in late 2017, EU member states also agreed to increase transaction monitoring and audits for transactions and taxpayers associated with listed jurisdictions. However, it is unclear to what extent these practices were ac- tually implemented prior to 2021, when member states began committing to implement more stringent legislative measures for blacklisted jurisdictions such as controlled foreign company (CFC) rules or withholding taxes. As of the time of writing, the EU notes that twenty-one member states now operate some form of legislative or administrative defensive measure. The EU has also amended its directive on the mandatory reporting of certain cross-border arrangements, DAC6, to explicitly refer to the blacklist, starting from mid-2020. It is thus possible that even though the impact of the blacklist was muted during the period studied in this paper, its effectiveness in terms of policy success and economic bite will be increased in the future. Unequal enforcement and moving deadlines While the EU has made its criteria for evaluating compliance clear, in practice it uses substantial discretion in choosing which jurisdictions are upgraded to its blacklist. In some cases these are due to extraordinary circumstances, such as eight Caribbean jurisdictions which were given extra time after they suffered a series of devastating hurricanes in mid- 2017. It also, for greylisted countries, typically agrees on a timeline for improvement. Target dates for implementing new tax governance measures can be as far as a year into the future, somewhat dampening the urgency behind reform. By contrast, the Financial Action Task Force gives jurisdictions very short time frames (three months at a time) to improve their anti-money laundering standards, when considering their inclusion or position on their grey and blacklists, which gives governments much less leeway to stall. 36 6.6 External pressure beyond the EU blacklist Another explanation is a general trend of improvement in international tax standards, outside that being driven by the EU’s efforts. Adoption of the BEPS minimum standards and AEOI is driven by a number of factors ranging from multilateral pressure from the OECD, bilateral pressure from trading partners, and reputational issues that extend be- yond the EU blacklist. It is possible that these trends leave less room for any additional impact of the blacklist. 7 Conclusion In this paper I have considered the impact that the European Union’s review and sub- sequent listing of jurisdictions for noncompliance with tax governance standards has had on that compliance. In summary, using both regression discontinuity and difference-in- difference frameworks, I found limited evidence that the EU review process led to large improvements in tax governance outcomes or any impact on economic outcomes. I do find strong, significant and persistent effects of the EU’s efforts on membership in the In- clusive Framework, an important outcome for both the detection of harmful tax practices by the OECD and for the eventual adoption of the Two Pillar Solution. This study highlights the limitations of efforts by regional bodies to unilaterally pres- sure other governments into reducing cross-border spillovers, particularly when that pres- sure is not accompanied quickly by clear, visible countermeasures. While the EU has clearly had a positive impact on the progress towards better international tax gover- nance, the lack of sizable effects across many dimensions suggests that these efforts need to be reassessed to ensure a higher level of impact and, perhaps, credibility. It also, given the fact that the bulk of shifted profits reside in jurisdictions that were not targeted by the listing exercise, suggests that reformulation of the selection criteria is warranted, perhaps to include EU member states themselves. 37 Table 5: Impact of listing on foreign-held deposits (Difference-in-difference estimates) Log(Foreign-held deposits, millions) Log(Inward FDI stock, as % of GDP) Log(Inward FDI flow, as % of GDP) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) Treatment: First listing 0.178 0.172 0.0676 0.346 0.0242 -0.0577 0.0230 -0.0586 0.00382 0.0449 0.0328 0.0674 0.00673 0.0163 0.00733 0.0163 0.0715 0.0964 0.110 0.103 (1.16) (1.14) (0.34) (0.79) (0.16) (-1.07) (0.15) (-1.09) (0.10) (1.21) (0.70) (1.50) (0.07) (0.12) (0.08) (0.12) (0.50) (0.51) (0.56) (0.36) First greylisting 0.207 0.214 0.0235 0.202 0.0348 -0.0189 0.0317 -0.0200 0.00776 0.0431 0.0310 0.0563 0.000663 0.0145 -0.000244 0.0115 0.00401 0.00290 -0.0202 -0.0594 (1.47) (1.55) (0.19) (0.61) (0.24) (-0.38) (0.22) (-0.40) (0.19) (1.23) (0.68) (1.27) (0.01) (0.11) (-0.00) (0.09) (0.03) (0.02) (-0.12) (-0.26) ATT -0.108 -0.109 0.142 0.497 0.0489 -0.0649 0.0545 -0.0717 0.0463 0.0934∗ 0.0523 0.103∗ 0.0281 0.0972 0.0243 0.0803 0.117 0.220 0.100 0.278 (0.152) (0.146) (0.196) (0.461) (0.143) (0.0766) (0.146) (0.0786) (0.0290) (0.0385) (0.0373) (0.0459) (0.133) (0.144) (0.137) (0.148) (0.208) (0.328) (0.235) (0.367) 38 Max # Jurisdictions 49 49 49 49 Max # Observations 1372 1372 1372 1372 Period Quarter Quarter Quarter Quarter Year Year Year Year Year Year Year Year Year Year Year Year Year Year Year Year Controls? X X X X X Method DiD Gardner CSA CSA DiD DiD Gardner Gardner CSA CSA CSA CSA DiD DiD Gardner Gardner CSA CSA CSA CSA Bandwidth restriction? X X X X X X X X Notes: This table shows results of various difference-in-difference specifications of the impact of EU listing on the the log of foreign-held deposits held in a jurisdiction, as measured by the BIS Locational Banking Statistics. Each row indicates a different treatment: first listing is whenever a jurisdiction is first added to either the black or greylist, first greylisting indicates the first time a jurisdiction is added to the greylist (excluding all jurisdictions that have only ever been blacklisted), with first blacklisting indicating the first time a jurisdiction is added to the blacklist (excluding all of those who have) only ever been greylisted. Each column indicates a different estimation approach, where DiD = two-way fixed effects, Gardner indicates (Gardner 2021), and CSA indicates (Callaway and Sant’Anna 2021) Regressions with bandwidth restriction limit the sample to those within 16 bandwidth points of the RD cutoff (restriction not made for deposit regressions due to low number of units). Robust standard-errors in parentheses. + p < 0.10,∗ p < 0.05,∗∗ p < 0.01,∗∗∗ p < 0.001 Table 6: Impact of EU selection on probability of detection and rollback of harmful regimes Jurisdiction Harmful regime Harmful regime reviewed detected rolled back (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) EU EU OECD OECD EU EU OECD OECD EU EU OECD OECD Selected by EU 0.844*** 0.805*** 1.233*** 1.466*** 0.502 0.180 0.721** 0.144 0.690** 0.140 0.504 0.225 (0.300) (0.290) (0.208) (0.218) (0.326) (0.308) (0.328) (0.265) (0.325) (0.277) (0.326) (0.311) Controls X X X X X X Observations 194 179 194 179 194 179 194 179 194 179 194 179 Notes: Outcome is = 1 if jurisdiction has had at least one regime reviewed by an entity, whether it had at least one regime reviewed and at least one harmful regime was detected, and whether it had at least one regime reviewed and at least one regime was rolled back. Results are divided by whether the reviewing entity was the EU Code of Conduct Group (COCG) or the OECD Forum on Harmful Tax Practices (FHTP). Results use the multidimensional running variable. Coefficient is estimated impact of selection into the EU review process on a different outcome, using bandwidth-selection, bias-corrected, robust methods outlined in Calonico, Cattaneo, and Titiunik (2014), Calonico, Cattaneo, and Farrell (2018) and Calonico, Cattaneo, Farrell, and Titiunik (2019). ∗ p < 0.10,∗∗ p < 0.05,∗∗∗ p < 0.01 39 Figure 9: EU effort in reviewing and detecting preferential regimes does not appear to exceed that of the OECD as we move further away from the cutoff (a) EU scrutiny relative to OECD scrutiny, by jurisdiction’s multidimensional (MD) score (b) EU scrutiny relative to OECD scrutiny, by jurisdiction’s stability factors (SF) scores (c) EU scrutiny for selected jurisdictions compared to OECD scrutiny, by jurisdic- tion’s score on “Strength of ties” measure Notes: Each figure shows the results of a local polynomial estimate of the probability a juris- diction has at least one preferential regime reviewed by the EU COCG or the OECD FHTP against that jurisdiction’s (a) score on the multidimensional running variable, (b) its score on the stability factors running variable and (c) that jurisdiction’s strength of ties score, conditional on being selected by the EU. Also shown is the approximate width of the local RD estimate bandwidth. 95% confidence intervals shown. 40 Figure 10: The probability of being reviewed and listed by the EU Notes: Figure (a) is a local polynomial estimate of the probability of ever being reviewed, greylisted or blacklisted against the total amount of shifted profits (in 2016), for jurisdictions that are home to a positive amount of shifted profits as estimated by Tørsløv, Wier, and Zucman (2022). Figure (b) displays the same probabilities against a jurisdiction’s Financial Secrecy Index Score (2020) (Tax Justice Network 2020). Figure (c) displays the same probabilities against GNI per capita in 2016. Figure 11: Profits shifted to blacklisted and greylisted jurisdictions in 2018 Notes: Bar graph shows the total amount of positive shifted profits estimated by Tørsløv, Wier, and Zucman (2022) in 2018 for countries that were (i) blacklisted that year (ii) greylisted that year but not blacklisted and (iii) neither.. 41 References Aliprandi, G., M. Baraké, and P.-E. Chouc (2021). Have European Banks left tax haven? Evidence from country-by-counry data. Ph. D. thesis, Eu-Tax. Beer, S., M. D. Coelho, and S. Leduc (2019). Hidden treasure: The impact of automatic exchange of information on cross-border tax evasion. Technical report, International Monetary Fund. Bilicka, K. and C. Fuest (2014). With which countries do tax havens share information? International Tax and Public Finance 21(2), 175–197. Bomare, J. and S. L. G. Herry (2022). Will we ever be able to track offshore wealth? evidence from the offshore real estate market in the uk. Callaway, B. and P. H. Sant’Anna (2021). Difference-in-differences with multiple time periods. Journal of Econometrics 225 (2), 200–230. Calonico, S., M. D. Cattaneo, and M. H. Farrell (2018). On the effect of bias estimation on coverage accuracy in nonparametric inference. Journal of the American Statistical Association 113 (522), 767–779. Calonico, S., M. D. Cattaneo, M. H. Farrell, and R. Titiunik (2019). Regression dis- continuity designs using covariates. Review of Economics and Statistics 101(3), 442–451. Calonico, S., M. D. Cattaneo, and R. Titiunik (2014). Robust nonparametric confidence intervals for regression-discontinuity designs. Econometrica 82 (6), 2295–2326. Calonico, S., M. D. Cattaneo, and R. Titiunik (2015). Optimal data-driven regression discontinuity plots. Journal of the American Statistical Association 110 (512), 1753– 1769. Cameron, A. C., J. B. Gelbach, and D. L. Miller (2008). Bootstrap-based improvements for inference with clustered errors. The Review of Economics and Statistics 90 (3), 414–427. Caruana-Galizia, P. and M. Caruana-Galizia (2016). Offshore financial activity and tax policy: evidence from a leaked data set. Journal of Public Policy 36 (3), 457–488. Case-Ruchala, D. and M. Nance (2021). Discipline without punishment: illicit finance, blacklisting, and the ideational sources of compliance in global financial governance. Technical report, Working Paper. Casi, E., C. Spengel, and B. Stage (2019). Cross-border tax evasion after the common reporting standard: Game over? ZEW-Centre for European Economic Research Discussion Paper (18-036). Christians, A. and L. Van Apeldoorn (2018). The oecd inclusive framework. Bulletin for International Taxation, April/May. 42 Damgaard, J., T. Elkjaer, and N. Johannesen (2019). What is real and what is not in the global FDI network? International Monetary Fund. Dell, M. (2010). The persistent effects of peru’s mining mita. Econometrica 78 (6), 1863–1903. Elsayyad, M. and K. A. Konrad (2012). Fighting multiple tax havens. Journal of in- ternational Economics 86 (2), 295–305. European Commission (2020). Tax policies in the european union. Technical report, European Commission. Garcia-Bernardo, J., J. Fichtner, F. W. Takes, and E. M. Heemskerk (2017). Uncovering offshore financial centers: Conduits and sinks in the global corporate ownership network. Scientific Reports 7 (1), 1–10. Garcia-Bernardo, J., P. Janskỳ, and T. Tørsløv (2022). Decomposing multinational corporations’ declining effective tax rates. IMF Economic Review , 1–44. Gardner, J. (2021). Two-stage differences in differences. Technical report, Working paper. Gelman, A. and G. Imbens (2019). Why high-order polynomials should not be used in regression discontinuity designs. Journal of Business & Economic Statistics 37 (3), 447–456. Grottke, M. and M. Kittl (2016). First the stick, then the carrot? a cross-country evaluation of the oecd’s initiative against harmful tax competition. Technical report, Passauer Diskussionspapiere-Betriebswirtschaftliche Reihe. Hearson, M. (2020). Corporate tax negotiations at the oecd: What’s at stake for de- veloping countries in 2020? Jardet, C., C. Jude, and M. D. Chinn (2022). Foreign direct investment under uncer- tainty: evidence from a large panel of countries. Technical report, National Bureau of Economic Research. Johannesen, N., P. Langetieg, D. Reck, M. Risch, and J. Slemrod (2020). Taxing hidden wealth: The consequences of us enforcement initiatives on evasive foreign accounts. American Economic Journal: Economic Policy 12 (3), 312–346. Johannesen, N. and G. Zucman (2014). The end of bank secrecy? an evaluation of the g20 tax haven crackdown. American Economic Journal: Economic Policy 6 (1), 65–91. Kelley, J. G. and B. A. Simmons (2015). Politics by number: Indicators as social pressure in international relations. American journal of political science 59 (1), 55– 70. Kida, M. and S. Paetzold (2021). The Impact of Gray-Listing on Capital Flows: An Analysis Using Machine Learning. International Monetary Fund. 43 Konrad, K. A. and T. B. Stolper (2016). Coordination and the fight against tax havens. Journal of International Economics 103, 96–107. Menkhoff, L. and J. Miethe (2019). Tax evasion in new disguise? examining tax havens’ international bank deposits. Journal of Public Economics 176, 53–78. Morse, J. (2019). Blacklists, market enforcement, and the global regime to combat terrorist financing. International Organization, Forthcoming. Morse, J. C. (2022). The Bankers’ Blacklist: Unofficial Market Enforcement and the Global Fight against Illicit Financing. Cornell University Press. Noked, N. (2019, June). Should the united states adopt crs? The Michigan Law Review . Omartian, J. D. (2017). Do banks aid and abet asset concealment: Evidence from the panama papers. Available at SSRN 2836635 . O’Reilly, P., K. P. Ramirez, and M. A. Stemmer (2019). Exchange of information and bank deposits in international financial centres. Roodman, D., M. Ø. Nielsen, J. G. MacKinnon, and M. D. Webb (2019). Fast and wild: Bootstrap inference in stata using boottest. The Stata Journal 19 (1), 4–60. Rusina, A. (2020). Name and shame? evidence from the european union tax haven blacklist. International Tax and Public Finance. Sharman, J. C. (2009). The bark is the bite: International organizations and blacklist- ing. Review of International Political Economy 16 (4), 573–596. Slemrod, J. and J. D. Wilson (2009). Tax competition with parasitic tax havens. Journal of Public Economics 93 (11-12), 1261–1270. Tax Justice Network (2020). Eu blacklists uk’s crown jewel tax haven while letting other tax havens off the hook. Technical report, TJN. Tørsløv, T. R., L. S. Wier, and G. Zucman (2022). The missing profits of nations. Technical report, National Bureau of Economic Research. Wier, L. S. and G. Zucman (2022). Global profit shifting, 1975-2019. Technical report, National Bureau of Economic Research. Wong, V. C., P. M. Steiner, and T. D. Cook (2013). Analyzing regression-discontinuity designs with multiple assignment variables: A comparative study of four estimation methods. Journal of Educational and Behavioral Statistics 38 (2), 107–141. Zucman, G. (2013). The missing wealth of nations: Are europe and the us net debtors or net creditors? The Quarterly journal of economics 128 (3), 1321–1364. 44 Table 7: Jurisdictions within the maximum RD bandwidth Distance FSI Secrecy to Rank Score 2015 Jurisdiction cutoff (2020) (2020) EU Blacklist Tax haven* Sink OFC** Cameroon -19.9 53 71 Kyrgyzstan -18.5 Philippines -18.4 60 63 Argentina -17 94 55 Lebanon -16.5 26 64 X Belarus -16.1 Tajikistan -15.6 Ukraine -15.1 70 65 Papua New Guinea -14.6 New Zealand -14.1 57 59 Kenya -13.7 24 76 Marshall Islands -12.2 49 70 X X X Azerbaijan -12.2 Brunei Darussalam -11.5 125 78 X Palau -11.5 Nicaragua -10.8 French Polynesia -10.5 Russian Federation -10.3 44 57 Bolivia -9.8 91 79 Gabon -9.5 Moldova -9.3 Suriname -8.5 Pakistan -8.4 100 55 Dominican Republic -6.9 107 59 Honduras -6.4 Kuwait -5.4 28 71 Egypt -5 46 71 Kazakhstan -4 108 64 Mexico -3.6 80 53 Guyana -3.1 X Ghana -1.5 117 52 Cote d’Ivoire -.7 Uruguay 0 90 57 China, Macao SAR .1 31 65 X Viet Nam .3 37 74 Taiwan 2 13 66 X Curacao 2.6 96 75 X Peru 2.7 101 57 Belize 2.7 110 74 X X X Trinidad and Tobago 3.7 127 65 Armenia 4 Swaziland 4.1 Indonesia 4.1 79 51 India 4.6 47 48 Albania 5.6 China 5.7 25 60 Costa Rica 6.2 83 62 Colombia 6.3 102 56 Korea, Republic of 6.9 21 62 Oman 7 Fiji 7.5 Saudi Arabia 7.5 45 67 Isle of Man 8.1 43 65 X Mongolia 8.5 Guernsey 9.2 11 71 X X Maldives 9.9 74 80 X Tunisia 10.9 78 66 Morocco 11 72 68 Thailand 12.3 17 73 Seychelles 13.3 95 70 X X X Jamaica 13.7 Brazil 14.2 73 52 Jersey 14.7 16 66 X X Chile 14.9 82 56 Qatar 15.5 20 77 Panama 16.2 15 72 X X Samoa 17.1 86 75 X X Botswana 18.1 113 62 Cabo Verde 18.5 Bosnia and Herzegovina 19 Japan 19.3 7 63 Jordan 19.9 42 78 45 Notes: List excludes LDCs. * Jurisidictions listed as tax havens in Johannesen and Zucman (2014). ** Jurisdictions listed as ‘sink’ offshore financial centers in Garcia-Bernardo, Fichtner, Takes, and Heemskerk (2017). FSI Rank = Rank in Tax Justice Network’s Financial Secrecy Index. Secrecy Score = FSI’s measure of financial secrecy.