Tax evasion in new disguise? Examining tax havens' international bank deposits

Menkhoff, Lukas; Miethe, Jakob Article — Accepted Manuscript (Postprint) Tax evasion in new disguise? Examining tax havens' international bank deposits Journal of Public Economics Provided in Cooperation with: German Institute for Economic Research (DIW Berlin) Suggested Citation: Menkhoff, Lukas; Miethe, Jakob (2019) : Tax evasion in new disguise? Examining tax havens' international bank deposits, Journal of Public Economics, ISSN 0047-2727, Elsevier, Amsterdam, Vol. 176, pp. 53-78, https://doi.org/10.1016/j.jpubeco.2019.06.003 This Version is available at: https://hdl.handle.net/10419/240930 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by-nc-nd/4.0/ Tax evasion in new disguise? Examining tax havens’ international bank deposits∗ Lukas Menkhoff†and Jakob Miethe‡ Abstract Recent efforts to reduce international tax evasion focus on information exchange with tax havens. Using bilateral bank data for 1,397 country pairs in a balanced quarterly panel from 2003:I – 2017:IV, we first show that information-on-request treaties with tax havens reduce bank deposits in tax havens by 27.5%. Second, also deposits from tax havens in high tax countries decline after such treaties are signed, giving authorities a second angle to detect tax evasion. Both reactions dissipate overt time and treaties signed after 2010 trigger no further reactions. These results cannot be explained by deposit shifting alone and we find no evidence of transitioning into legality. Third, recent policy initiatives based on the automatic exchange of bank information lead to very similar initial reactions as earlier treaties, consistent with adjustments on the part of tax evaders. This suggests that tax evaders adapt to established information exchange treaties by using new disguises to hide their true income, and react again to new measures. These results cast doubt on the effectiveness of current forms of information exchange to tackle international tax evasion. JEL classification: H 26 (tax evasion and avoidance), F 38 (int’l financial policy) Keywords: Tax evasion; international information exchange treaties; international bank deposits; tax havens June 3, 2019 ∗We thank participants of the 73rd annual congress of the International Institute of Public Finance, 2017, the 15th INFINITY conference on International Finance, the 5th conference on the Shadow Economy, Tax Evasion and Informal Labor, the 2017 Annual Conference of the German Economic Association, the 2018 Annual Congress of the European Economic Association, as well as participants of several workshops and seminars, in particular Ron Davies, Marcel Fratzscher, Shafik Hebous, Niels Johannesen, Signe Krogstrup, Gian Maria Milesi-Feretti, Joel Slemrod, Tim Stolper, and Vadym Volosovych for comments and discussions at various stages of this project. Many thanks also to the editor and three anonymous referees for their thorough feedback. †Humboldt-University Berlin and DIW Berlin (German Institute for Economic Research), 10108 Berlin, Germany; e-mail: [email protected]; tel. ++49 (0)30 89 789 435. ‡Humboldt-University Berlin and DIW Berlin (German Institute for Economic Research), 10108 Berlin, Germany; e-mail: [email protected]; tel. ++49 (0)30 89 789 439. 1 This is the postprint of an article published in Journal of Public Economics 176 (2019), p. 53-78, available online at: https://doi.org/10.1016/j.jpubeco.2019.06.003 © <2021>. This manuscript version is made available under the CC-BY-NC-ND 4.0 license http://creativecommons.org/licenses/by-nc-nd/4.0/ Tax evasion in new disguise? Examining tax havens’ international bank deposits 1 Introduction Tax havens have a long-standing history. Since open borders allow for the international transfer of capital, there is an incentive to shift capital and earnings on this capital to places where taxes are relatively low and secrecy is high. While capital in tax havens is not necessarily illegal, it can be, for example, if interest income is not reported in the country where the recipients pay their taxes. Such international tax evasion is of great concern to policy makers for at least three reasons: first, it reduces tax revenues; second, it reduces effective taxation of the rich; and, third, the majority may lose trust in both the tax system and state institutions. Therefore, the fight against tax evasion is high on the international agenda. Whistleblowers and leaks of large datasets indicate that international tax evasion via tax havens remains a relevant problem. Bank deposits of non-banks are shifted into tax havens where the ownership of these deposits is hidden. While we know that such deposits initially reacted to information-exchange-on-request (IoR) agreements that are exclusively aimed at tax evaders (Hanlon et al., 2015; Johannesen and Zucman, 2014), important questions remain unanswered. Have these agreements eliminated tax evasion over time or is there evidence that a significant share of bank deposits in tax havens still evades taxation? If we find reactions of deposits in tax havens (outbound deposits); do we find the same pattern in bank deposits from tax havens in non-havens (inbound deposits)? Further, how do non-banks’ bank deposits in tax havens react to new types of international agreements that are based on the automatic exchange of bank information? These three questions are addressed by relying on a data set released in October 2016, when the Bank for International Settlements (BIS) made a significant portion of its bilateral locational banking statistics available, including bilateral deposit data. Using the bank deposits of non-banks reported in this data, the aforementioned three questions can be addressed for the first time, to the best of our knowledge. Results shed new light on international tax evasion: Interestingly, the ‘success’ of early IoR treaties does not show that they have realized their goal but rather that tax evaders seem to adapt to regulation by putting their bank deposits into new disguises. This major finding is supported by empirical examinations of the three questions. First, we extend the analysis of bank deposits in tax havens (outbound) by relying on the 2 event study approach of Johannesen and Zucman (2014). Their study is the first to analyze changes in bank deposits in tax havens by considering the impact of bilateral tax information exchange agreements. They had private access to a subset of the BIS data, going from 2002 to mid-2011 in the outbound direction, which are now largely public but with a longer time dimension. For a similar group of countries, we confirm that, despite having twice as many tax agreements by 2017, the effects of IoR treaties on bank deposits in tax havens are qualitatively the same. Going beyond their sample period, we find a – so far unreported – gradual decline of this effect for new IoR treaties over time, starting around 2010. That means non-banks’ bank deposits in tax havens no longer react to newly signed treaties, raising the question of whether treaties have successfully eliminated tax evasion or whether tax evaders have adapted to the new regulation by structuring their evasion schemes in a way that circumvents IoR treaties. Second, the BIS data allow us to address a gap in the literature by also considering the inbound direction, i.e. bank deposits of non-banks from tax havens in non-havens. ‘Inbound’ and ‘outbound’ are used from the perspective of the tax evader here who deposits outbound (in the tax haven) or inbound (from the tax haven to the non-haven). The reaction to IoR treaties in inbound deposits may seem surprising, given the fact that the money is already deposited in tax havens and the “true” owners are hidden. Still, the ownership structure of shell companies, private foundations, and trusts, as well as the connected bank accounts, is theoretically vulnerable to detection if, for example, the tax evader is documented as a beneficial owner. Thus, while single deposits cannot be followed through the tax haven cloud, the analysis of inbound deposits provides a second angle for tax authorities in home countries to tackle tax evasion. They can take the occurrence of such inbound deposits to investigate their ownership and look for evidence of illegal behavior. Third, in order to investigate the longer-term effects of information exchange on bank deposits in and from tax havens, we also assess newer forms of information exchange. In this respect, we analyze the OECD’s Mutual Competent Authority Agreement, which allows for automatic exchange of information (AEI) under its Common Reporting Standard (CRS) after a bilateral matching, as well as two US-Switzerland agreements and the US Foreign Account Tax Compliance Act (FATCA). If the original goal of regulators was realized, former tax evaders are compliant and the new agreements do not reduce bank deposits in tax havens. Our data show, however, the opposite: new agreements produce effects that are hauntingly similar to those of the older IoR treaties. This evidence speaks against the long-term success of bilateral information exchange, but is consistent with the hypothesis that tax evaders hide their funds 3 using new disguises and with other results in the literature that find no decline in aggregate offshore wealth (Alstadsæter et al., 2018; Pellegrini et al., 2016; Zucman, 2013). We also take into account amnesties and voluntary disclosure programs, showing that the results are not driven by their introduction. Unfortunately, we cannot analyze the longer-term effects of the AEIs due to their novelty and the resulting lack of observations. Literature. Information about tax havens is generally rare, including bank balance sheet positions. Due to such incomplete data, it is common practice in the literature on international financial integration to drop tax havens from the sample (as in Broner et al., 2013) or to control for them with a designated dummy (as in Lane and Milesi-Ferretti, 2008). Identifying or measuring tax evasion empirically is challenging, but there is progress, as Slemrod (2015) summarizes. One of the rare studies with access to data about bank positions in tax havens is Johannesen and Zucman (2014). This benchmark study for our work applies an event-study type approach to changes in banks deposits of non-banks in tax havens in relation to IoR agreements. This study is part of a strand of research working with tax haven data that, on top of tax haven bank deposits (see also Johannesen, 2014b), works with surveys of institutions registered in tax havens (Heckemeyer and Hemmerich, 2018) or leaked datasets (Caruana-Galizia and CaruanaGalizia, 2016; O’Donovan et al., 2019). Johannesen and Zucman (2014) find sizable significant reductions in these bank deposits as treaties threaten bank secrecy. While relying on their method, we use more recent and broader data. Thus, we can replicate their results, but find that it no longer holds during more recent years. Interestingly, however, similar results show up again when CRS agreements (with first activations in the last quarter of 2016) are analyzed. The extended data also allow constructing another clean falsification sample (non-haven to non-haven deposits) and to analyze inbound deposits. Relying on indirect identification strategies and data reported by non-havens, only Hanlon et al. (2015) provide a glimpse of the existence of tax evasion effects in inbound positions and their sensitivity to tax agreements. In a sample ending with 2008, they show that four IoR treaties between the US and tax havens negatively affected inward foreign portfolio investment from these tax havens. The advantage of such an approach is that it does not rely on tax havens as a source of data. For the first time in the literature, we are able to provide results based on both deposit data from tax havens (outbound) and non-havens (inbound) taken from the same dataset. The large international dimension also supports our identification strategy. 4 This paper proceeds in seven more sections: Section 2 sets out the identification via tax agreements and documents the institutional background. Section 3 introduces the empirical approach, including data. Results on the effect of IoR treaties on outbound deposits are presented in Section 4, while Section 5 provides analogous results for inbound deposits and Section 6 examines potential effects of more recent information exchange agreements. Robustness checks are presented in Section 7, while Section 8 provides conclusions. 2 Identification of international tax evasion In this section, we describe our identification of international tax evasion via tax havens, in line with typical procedures during our sample span. Section 2.1 lays out the dynamics of bank deposits in and from tax havens. Section 2.2 discusses the definition of the first target of such deposits, i.e. tax havens. Finally, Section 2.3 introduces into the international information exchange agreements that serve as our tool to identify evaded bank deposits. 2.1 Bank deposits in and from tax havens Transferring capital from a non-haven to a tax haven, called the ‘outbound deposit’ here, is a relatively simple matter for evaders. Setting up a bank account offshore can be done online for a small fee (see Sharman, 2010, for some real-life examples). As offshore wealth is typically held by very wealthy individuals, such fees are negligible (Alstadsæter et al., 2019). The most common way to transfer funds to that account is via an invoice for ‘consulting services’ carried out by the evader (see Zucman, 2013, 2014, for examples). As long as bank secrecy is upheld in the tax haven, this information is not exchanged with the home authorities of the depositing individual. Such deposits – protected by tax haven bank secrecy – can generate untaxed capital gains if the individual chooses to not self-report and evade taxes. Figure 1 provides a schematic visualization of how funds are deposited in and from tax havens. First, funds are moved into the tax haven (‘outbound’ from the perspective of a tax evader). Transfers between tax havens can take place, layers of secrecy can be added with a network of shell companies, and the funds can generate capital gains. Since it is unlikely that tax havens will themselves chase down tax evaders, we use positions between tax havens as a falsification sample where no threat of detection is expected (tax haven falsification). At some point, at least a fraction of the funds will be repatriated to a non-haven for various purposes (‘inbound’ from the perspective of the tax evader). Finally, positions between nonhaven countries should not be connected to international tax evasion of the same form we are 5 analyzing. Therefore, we use capital positions between non-havens as a second falsification group (non-haven falsification). — Figure 1 about here — As inbound deposits are less researched than outbound deposits, we want to clarify three issues in this regard: (i) Regardless of tax evasion motives or not, capital that is brought to tax havens will typically not be invested in the local economy of a tax haven but somewhere in the world economy. Inbound (bank) deposits are just one way to hold these assets and some of these bank deposits may be part of a tax evasion story. (ii) We do not expect a direct linkage between outbound and inbound deposits. Outbound deposits will also be used to buy assets other than inbound bank deposits and this can occur in different countries; however, there is some evidence of home bias in international capital flows (see Coeurdacier and Rey, 2013, for an overview). (iii) When funds that are linked to illegal activity, such as not declaring capital gains and, thus, evading taxation, are transferred to a bank account in a non-haven, the individual in question can hide their identity behind the veil of secrecy erected by tax haven shell companies that opened the bank account. However, there is a risk of detection here as the ownership structure of these funds might be exposed. Indeed, Sharman (2010) shows that it is increasingly difficult, although possible, to establish an evasion setup without providing identification at some point to the service provider in the tax haven. The World Bank, cooperating with the United Nations Office on Drugs and Crime (UNODC), records hundreds of cases of grand corruption in their ‘Stolen Assets Recovery Initiative’ with numerous cases including tax evasion using such schemes. Thus, qualitative evidence on such arrangements is available, due in part to convictions.1The US Internal Revenue Service provides another list of exemplary cases including such setups.2In many of these cases, bank accounts with non-haven banks were opened by shell companies or trusts domiciled in tax havens. Such schemes show up in the inbound deposits data we use and are part of; for example, the 6.5 billion USD French 1For example, in July 2007, Diepre Alamieyeseigha was sentenced to two years in prison for charges of false declaration of assets and money laundering. In order to transfer funds from his native Nigeria, where he was state governor, he used corporate vehicles in the Seychelles, the British Virgin Islands, the Bahamas, and South Africa, via which he transferred funds all over the world, including into a U.S. dollar account with UBS in London (van der Does de Willebois et al., 2011). Such funds show up as ‘inbound deposits’ by tax haven counterparties in the data we employ. 2The US Internal Revenue Service provides yearly examples of abusive tax schemes as well as the harsh penalties incurred when these are discovered. The list includes examples in line with shell companies in tax havens with bank accounts in non-havens, like those we use for identification: https://www.irs.gov/uac/examples-of-abusivetax-schemes-fiscal-year-2015 6 banks reported in deposits from the channel island of Jersey (98,069 inhabitants) in the middle of 2013 in our data. Before we test for these effects empirically, we provide information on the tax havens included in the analysis as well as the information exchange agreements that we use to identify evasion. 2.2 Tax havens Although there is consensus in defining a tax haven as a jurisdiction with low or zero tax rates on some income types, most definitions go further and restrict ‘tax havens’ to countries with high bank secrecy rules and low transparency regulations. Tax havens also score high on governance indicators, have relatively sophisticated communication infrastructure, and few natural resources (Dharmapala, 2008; Dharmapala and Hines, 2009a; Hines, 2010). The empirical literature commonly employs relatively unrestricted tax haven lists. For most research questions, including a de facto non-haven into the tax haven list leads to a more conservative estimation. Therefore, we follow this convention and use, as our baseline, a large list of tax havens obtained by combining the list of tax havens by Gravelle (2015), which already collects different sources (such as Hines and Rice, 1994), with that of Johannesen and Zucman (2014), which we also test separately. This tax haven classification is altered in extensive robustness checks, but results are unaffected. Table 1 summarizes key information on those tax havens for which comprehensive BIS data are available. A complete list of the 58 tax havens is provided in Appendix Table A1 and the few cases of disagreement over tax havens in different studies are summarized in Appendix Table A2. — Table 1 about here — As the first three columns document, the stereotypical small Pacific island is still pervasive and colonial ties remain important. Column 4 shows the number of tax information exchange treaties signed by tax haven countries. Column 5 shows the number of countries that report deposit data on that specific tax haven, revealing the availability of a large crosssectional dimension. Column 6 shows end-of-sample total liabilities reported against non-bank counterparties in the entire world to give an idea of financial size. Finally, columns 7 and 8 indicate if the respective tax haven is used in our outbound and/or inbound analysis. As an example: Guernsey, a British Crown Dependency with fewer than 70,000 inhabitants but more than 31 billion USD in total external bank liabilities, has signed 34 treaties with non-haven 7 countries during the 60 quarter sample period (2003:I – 2017:IV) that meet our requirements, as outlined in the next section. It reports bilateral data to the BIS and is thus used in our outbound sample, and twenty-four other reporting countries report data against Guernsey. To create a balanced panel, we drop all but three non-haven reporting countries that report positions against Guernsey (part of the inbound sample) and all but six tax haven reporting countries (part of the tax haven falsification sample). 2.3 Information exchange agreements Since tax evasion is, by definition, the illegal withholding of tax liabilities, international regulation attempts mainly focus on detecting the delinquent by agreeing to exchange information between tax authorities (OECD, 1998, 2000). Such treaties became popular after April 2009, when the G20 decided to sanction tax havens if they did not sign at least 12 such treaties (G20, 2009). Since the provision of services to facilitate tax evasion has benefits, at least in the form of service fees, tax havens have an incentive to maintain secrecy regimes. Cooperation is enforced by threat of economic sanctions, which presents the dilemma of choosing between compliance and secrecy. Konrad and Stolper (2016) model this dilemma of tax havens and predict a negative signaling effect (for tax evaders) if a tax haven shows compliance of some sort, a point we return to later. With political pressure, as of 2017, more than 3,000 bilateral tax information exchange treaties (TIEAs) have been signed. Bilicka and Fuest (2014) document that fears that tax havens would simply sign with 12 other tax havens or economically meaningless countries did not materialize: on average, treaties are signed between tax havens and non-havens that have strong economic ties. The provisions in these short treaties establish a procedure of bilateral information exchange upon request. There are several caveats: information has to be ‘foreseeably relevant’ (changed from the stronger ‘necessary’ prior to 2005), which implies knowledge of the identity of the evader; there can be refusal for public policy reasons (such as a request being ‘at variance’ with laws of the counterparty); and requests cannot be aimed at information that the requesting country can obtain itself (Christensen III and Tirard, 2016). Thus, information requests are rare in practice and necessitate a level of detail of information in the non-haven about the evader that would most likely imply trouble for her even without information exchange. Nevertheless, a threat of detection creates an incentive for evaders to react (see Dwenger et al., 2016, for a discussion of intrinsic versus extrinsic incentives). 8 of new information exchange treaties was signed between 2009 and 2012. This period includes significant balance sheet reduction in the international banking system. In order to alleviate concerns that our results are driven by this momentum of signatures (and not the precise treaty quarters), we carry out a placebo analysis where all 1,074 countrypairs in our four samples that have not signed an information exchange treaty are assigned a placebo-treaty. In order to mirror the signature momentum, we calculate a cubic spline over the quarters in which treaties are signed based on the total number of signatures in each year-quarter, for all countrypair groups and for both IoR treaty types. Based on these (normalized) splines, we draw placebo treaties for each countrypair in the respective group and build a placebo treatment variable based on these. Figure 4 plots, as an example, TIEAs and placebo-treaties for nonhaven – tax haven countrypairs with non-normalized splines. As can be seen, the placebos mirror the signature momentum of the OECD initiative. They start taking effect with that same concentration of treatments during the deleveraging period of 2009 – 2012. — Figure 4 about here — Column 1 of Table 3 uses this placebo treatment to test if it reduces the treatment effect we analyze. The coefficient is insignificant and economically almost zero with the results largely unchanged, thus making us confident that we do indeed capture a negative reaction to the exact information treaty used. — Table 3 about here — TIEAs vs. DTCs. So far, we follow Johannesen and Zucman (2014) by employing all possibilities of IoR by using both tax information exchange agreements (TIEAs) and double taxation conventions (DTCs) as a treatment variable. In order to go into further detail, Table 3 also shows results for differentiated treaties. Column 2 differentiates the two treaty types introduced above. Interestingly, only TIEAs drive the effect, not DTCs, which are insignificant both statistically and economically. This is in line with the aims of these treaty types. While TIEAs explicitly aim at information exchange to curb tax evasion, DTCs are more complicated treaties that cover a range of double taxation issues, with information exchange among many clauses. Further, as we show in Figure 2, non-haven – tax haven countrypairs mostly signed TIEAs. Falsification samples. Columns 3 and 4 change the sample to the tax haven falsification group where we do not expect reactions if these jurisdictions are less likely to prosecute tax 15 evasion. This is confirmed by the data: both the parsimonious (column 3) as well as the specification differentiating treaty types (column 4) show insignificant results of information exchange amongst tax havens. Using data reported by non-havens allows us to compare the case of tax evasion via tax havens to the ‘normal’ behavior of deposits between non-havens. Columns 5 and 6 change the sample to this non-haven falsification group. Column 5 shows that the effect of information exchange-upon-request treaties actually changes sign when employed in this sample and is marginally significant. Column 6 again differentiates treaties. We do not claim to establish the direction of causality here, but there is evidence that bilateral tax treaties have a positive effect on capital positions (Blonigen et al., 2014) and, in our case, these effects are indeed positive. Thus, we interpret the findings in columns 5 and 6 as the ‘normal’ case in which tax evasion does not drive the results. These various falsification exercises make us confident that we have indeed identified a reaction of tax evaders to information exchange upon request in our main results. As we show in the following section, however, this effect fades over time. 4.3 The effect of treaties over time Our relatively long sample period allows us to compare reactions during different time periods. The initial results already hint at a qualitative change in the reaction to informationupon-request treaties over the course of our sample. Before we go into further detail, however, it is important to confirm random assignment, namely that early signatures are comparable to late ones. To the best of our knowledge, the only available study that investigates the characteristics of countrypairs signing IoR treaties is Bilicka and Fuest (2014). Replicating their data as closely as possible, we use the three yearly bilateral positions they analyze, namely: total export of goods and services from the IMF’s Direction of Trade statistics, the stock of foreign direct investment (fdi) taken from the OECD, and the stock of foreign portfolio investment (fpi) taken from the IMF’s Coordinated Portfolio Investment survey. Figure 5 shows these bilateral positions averaged for all countrypairs that sign a treaty at the time plotted on the horizontal axis. Before averaging, the data are transformed using the log equivalent inverse hyperbolic sine transformation to retain zero and negative values. — Figure 5 about here — The top panel shows that the tax haven – non-haven countrypairs (i.e. outbound) that sign new treaties toward the end of our sample exhibit similar, if not higher, bilateral positions 16 in trade and bilateral investment stocks compared to those countrypairs signing early in the sample. The middle panel shows that this holds for positions reported by non-havens against their tax haven signature counterparties as well (i.e. inbound). As the relevant point of comparison, the bottom panel shows the momentum of new signatures in the sample. It is not mirrored in the relative importance of the signing countrypairs, which we take this as evidence that, at least for observable characteristics, treaty signatures are not systematically non-random across our sample. Having established that countrypairs signing IoR are comparable over time, we find that the effect of treaties on tax haven deposits starts to approach zero around the year 2010. Figure 6 shows this development for outbound reactions of tax evaders for a five year period. We estimate our baseline results with a rolling window, limiting the sample to eight quarters before and after the time indicated on the horizontal axis, as well as showing effects of new IoR treaties within that time frame. We then plot the coefficient on a treatment variable collecting the most significant treatment lags of the baseline specification (k=−2 to k=2 around the treaty signature date) as well as their standard errors and indicate statistical significance with the shaded areas, light grey indicating 5% and dark grey 10% significance. — Figure 6 about here — 4.4 The short term effect of deposit shifting This effect of decreasing reactions to IoR treaties over time is not immediately intuitive. In light of increasing estimates of total international tax evasion (Pellegrini et al., 2016; Zucman, 2013) or offshore wealth (Alstadsæter et al., 2018), the effect of newer treaties should increase rather than decrease. This is especially true since Johannesen and Zucman (2014) show that one reaction to IoR treaties was to shift deposits to unaffected bilateral connections. As the counterparty non-haven, say France, signs more and more treaties with other tax havens, the tax haven that has not signed an IoR treaty with France experiences an increase in deposits from France. If that tax haven eventually signs and if tax evaders still use setups that are theoretically vulnerable to IoR treaties, the new treaty should trigger an even stronger reaction. Taking the results of these studies as motivation that IoR treaties do not reduce tax evasion in aggregate but may rather create deposit shifting into less regulated tax havens, we analyze this mechanism in our sample. Using all available information, as in Johannesen and Zucman 17 (2014),7we find some degree of deposit shifting, as shown in Figure 7. This figure plots the coefficients of a variable counting treaties the counterparty non-haven has signed with other tax havens. It is evident that the effect is rather short-lived and took place in anticipation to the bulk of IoR treaties. In the 4-year rolling window we employ, as before, the last sample where we cannot reject the null hypothesis of no deposit shifting runs from 2004 Q4 through 2008 Q4 with the middle of the sample (in this case 2006 Q4) plotted on the vertical axis. This indicates that the results established in the literature are present but also that deposit shifting is probably not the whole story, as the mixed results on deposit shifting by Johannesen and Zucman (2014) and Johannesen (2014b) indicate already. — Figure 7 about here — Focusing on the early years of IoR treaties, it looks as if these treaties may in fact motivate tax evaders to shift their capital into those tax havens that do not sign IoR treaties. However, starting at about 2008 there is a combination of effects of new IoR treaties with no deposit shifting and still no evidence of decreasing aggregate tax evasion. This suggests that something else is taking place: if capital income is not legalized and if it is not invested in tax havens that do not comply with IoR treaties, then we argue that this capital is put in other forms, a new disguise. An example of such a new disguise are shell companies with beneficial owners close to, but not identical to, the tax evader. Such structures alleviate the agency problem of financial services and allow the tax evader to retain an element of control but circumvent IoR treaties that require knowledge about the identity of the evader. The bank in the tax haven has no information about the actual tax evader that it could share with tax agencies.8 Already such simple setups successfully circumvent IoR treaties and would explain the 7This increases our bilateral countrypairs from the 557 in the balanced panel to 1,616 in an unbalanced panel. For completeness, we re-run the rolling outbound analysis of Figure 6 in this unbalanced sample and find the same petering out of IoR impacts. This is documented in Appendix 2, Figure A1. 8One example of such a setup are the daughters of German businessman Curt Engelhorn who were charged with tax evasion after the Paradise Paper leaks uncovered them as the beneficial owners of trusts set up for them years earlier (See https://www.tagesschau.de/ausland/paradisepapers/paradisepapers-125.html, last accessed April 16, 2019). Similarly, the father of the international football star Lionel Messi was the ultimate beneficiary of shell companies exposed in the Panama Papers that led to tax fraud charges (See https://panamapapers.sueddeutsche.de/articles/57021852a1bb8d3c3495b438/, last accessed April 16, 2019). Such structures were addressed to some extent in 2013, three years after the treatment effect we report fades out: The guidelines on politically exposed persons (PEPs) of the financial action task force (FATF) recommends as best practice that “if a customer or beneficial owner is identified as a family member or close associate of a PEP, then the requirements for PEPs should also apply accordingly” (FATF Guidance Politically Exposed Persons, Recommendations 12 and 22, section C.34, p. 10, Last accessed April 16, 2019 at http://www.fatfgafi.org/media/fatf/documents/recommendations/Guidance-PEP-Rec12-22.pdf). 18 petering out of our treatment effect and the short lived deposit shifting effect. Unfortunately, we have no quantitative measure to evaluate such structures directly.9After adjustments have taken place, no further reactions to IoR treaties are to be expected and our treatment effect dies out even though treatment variation is still substantial after 2010. In effect, IoR treaties have lost their usefulness in making tax evasion visible. The next opportunity that allows a glimpse into the behavior of international tax evaders is the international introduction of automatic exchange of bank information. Before turning to this more recent initiative, we shortly complete the picture of international tax evasion by analyzing inbound deposits from tax havens in non-haven countries. 5 Results of information exchange-upon-request treaties on inbound deposits In this section, we compare the results of the last section with reactions of deposits by tax haven counterparties in non-havens. This allows us to analyze inbound deposits based on the same data source as outbound deposits with identical treatment variables. We analyze the inbound deposits here in the same order as the outbound deposits in the preceding Section 4. Baseline result. Our first hypothesis is that there is a reaction of inbound deposits to treaties that mirrors the outbound evasion effects shown above. Column 1 of Table 4 shows that a treaty reduces deposits by tax haven counterparties in non-haven countries by 37%, quantitatively in line with the inbound effect reported by Hanlon et al. (2015) for US portfolio liabilities. At first glance, this result looks as if residents in tax havens use non-haven countries to evade taxes and are afraid that their government will use the availability of information exchange to detect just that. Since we define tax havens as countries that uphold strong secrecy rules and relatively low tax rates, this is unlikely. The result is meaningful once viewed through the lens of the findings in the last section. The capital deposited by tax haven counterparties does not only constitute capital from citizens of these jurisdictions but includes that of foreign depositors who are originally residents of non-havens and deposit some of these funds in nonhavens disguised as tax haven residents. — Table 4 about here — There is reason to believe that reactions in inbound deposits happen with some lag. As funds are already invested, re-arranging their structure takes time. Indeed, although the sign is 9The sham corporation variable in Johannesen and Zucman (2014) is not applicable here as we are not necessarily talking about positions from another tax haven. The new structure can well be in the same jurisdiction as the original deposit. 19 already negative, we do not find consistently significant results before the fourth lag (column 2). After one year, deposits remain lower, which is in line with a negative signaling effect of compliance (Konrad and Stolper, 2016). We confirm this lagged reaction in a dynamic differences-in-differences treatment analysis in the robustness section. Extended results. In column 3, we introduce the bilateral control variable capturing the relative importance of the bilateral connection in international bank claims (financial weight). Again, this variable exhibits a significant and positive coefficient. Mirroring the reform momentum of our treatment variable, column 3 also introduces the placebo treatment. The coefficient is insignificant and economically close to zero. The differentiation between TIEAs and DTCs provided in column 4 confirms that the inbound effect is also driven by TIEAs. Addressing potential concerns regarding the tax haven list, column 5 reduces the counterparty tax havens to those on the Johannesen and Zucman (2014) list and column 6 additionally excludes Switzerland. As for outbound deposits, we carry out a comprehensive test of sample effects dropping any combination of one or two counterparties, countries, yearquarters, and years. Not one of the almost 3000 resulting specifications is insignificant or leads to qualitatively different coefficients (see robustness section). The effect over time. The final step in showing some symmetry of outbound and inbound bank deposits via tax havens is to compare the developments of inbound reactions for new treaties each in rolling windows. Figure 8 shows this development for inbound reactions. Again, we estimate our baseline results with a rolling window, limiting the sample to eight quarters before and after the time indicated on the horizontal axis and comparing new signatures within these sub-samples. The plotted treaty coefficient is based on the most relevant lags found in the baseline (k=4 : k=6 around the treaty signature date) and its standard errors are shown in light grey (95% confidence interval) and dark grey (90% confidence interval). Except for a slight lag in the inbound reactions pointed out above, these results are virtually identical to those we find in a completely different sample for outbound deposits. Between 2011 and 2012, exchange-upon-request treaties lose significance and the treatment effect converges toward zero. — Figure 8 about here — We take the results in this section as evidence that funds deposited by tax haven counterparties in non-haven countries could be connected to tax evasion. Quantitatively, our results in both directions are in line with the effect of the EU savings directive on Swiss deposits 20 analyzed in Johannesen (2014b), who establishes reactions of deposits in the range of 30% to 40% (15% to 30% for other tax havens) and those of Hanlon et al. (2015) for inbound portfolio liabilities. It is especially striking how similarly these reactions evolve over time: this is not an automatic effect of the congruency of inbound and outbound bank deposits, as we only use liability data reported against non-bank counterparties. Instead it shows that not all funds deposited by tax evaders end up in tax haven sinks (though some will, as Garcia-Bernardo et al. (2017), show with the example of corporate ownership structures), rather some return to non-haven countries. There are two interpretations for the finding of reduced reactions over time. Either treaties are indeed successful in decreasing tax evasion or evaders increasingly rely on other options to evade taxes. Since, if anything, calculations of aggregate evaded capital or offshore wealth, as provided by Zucman (2013), Pellegrini et al. (2016), and Alstadsæter et al. (2018), point to increasing evasion over time, the first interpretation is not plausible. Instead, it seems likely that tax evaders channel their wealth via non-complying tax havens or create new disguises for their funds. In the following, we test an implication of the latter hypothesis. 6 Results on recent information exchange agreements The recent declining effect of IoR treaties on bank deposits in both directions, i.e. outbound in tax havens as well as inbound in non-havens, may be caused by tax evaders adjusting how they set up their deposits. If this is the case, we should again expect effects of recent far-reaching agreements based on an automatic exchange of information (Section 6.1). We test this first for the CRS (Common Reporting Standard) in Section 6.2 and then for a set of further agreements specific to the US in Section 6.3. 6.1 The motivation for examining recent information exchange agreements Criticism concerning information exchange upon request was pervasive, coming to light even as IoR treaties were introduced (Kudrle, 2008). Information upon request presupposes that most of the information necessary to convict a tax evader is already known to authorities, including, most importantly, their identity. There are two international attempts to introduce automatic information exchange on foreign nationals: FATCA and the CRS. The Foreign Account Tax and Compliance act (FATCA) is bilateral in nature, with the US asking foreign countries to enter bilateral treaties, starting in 2010. With the threat of effectively being excluded from US capital markets by a 30% withholding tax, foreign financial institutions 21 must report the identities of American account holders who they serve (Johannesen et al., 2018). These requirements are so restrictive that FATCA is criticized for overreaching US competences in foreign countries (Michel and Rosenbloom, 2011), especially since intra-US competition has led to states like Delaware offering tax haven type services themselves (Dyreng et al., 2013) and tax havens are already complaining about such ‘hypocrisy’ (European Parliament, 2016). However, it should be pointed out that FATCA’s implementation regulations significantly reduced these requirements. The US Senate’s Permanent Subcommittee on Investigations published a report outlining several feasible evasion strategies circumventing FATCA (United States Senate, 2014). As of July 2018, there are 127 signed FATCA agreements, 63 of which are included in our balanced panel.10 Inspired by the US model of automatic information exchange, the OECD, with the support of the G20, initiated the Common Reporting Standard (CRS) and the multilateral competent authority agreement (MCAA) under which automatic exchange of information was introduced. Information exchange is based on a bilateral matching process, which again leaves ample loopholes for evaders as many countrypairs have not agreed on such a matching. As in the case of IoR treaties, 26,000 bilateral agreements would be needed for a complete network, while only 2,243 were activated during our sample period; 336 can be matched to bilateral deposit data in our balanced panel. Under these agreements, comprehensive information is automatically exchanged annually on investment income, account balances, and proceeds from the sale of financial assets. These data are reported by banks, custodians, brokers, investment vehicles, and insurance firms regarding accounts held by individuals and entities, including trusts and foundations (OECD, 2016). Thus, the CRS is significantly more comprehensive in scope and its automatic nature introduces a real threat of detection that might induce evaders to react. Conceptually, these measures again aim for introducing an information exchange between tax havens and other countries, thus making deposits and their owners potentially known to relevant tax authorities. The fact that new agreements exist can be seen as indication that authorities are not fully satisfied with their earlier initiatives. If IoR treaties had fully succeeded, then policy makers would have the best world – from their perspective – as tax evasion would be a marginal phenomenon. In the case of new disguises, however, the regulation of tax havens 10These are recorded at: https://www.treasury.gov/resource-center/tax-policy/treaties/Pages/FATCA.aspx, last accessed April 16, 2019. In order to keep the results comparable to Sections 4 and 5 and to be able to test both results in the same setting, we remain in the balanced panel used before. A wider balanced panel starting at a later point in time than 2003 shows consistent results, however. 22 has largely failed and a new set of rules would be appropriate. In order to shed light on the latter mechanism, we significantly extend our database of information exchange treaties with the activated bilateral CRS relationships, signed FATCA agreements, and two dummies for US-Switzerland agreements. This extended database is analyzed in the same way IoR treaties are analyzed. If these new agreements reduce nonhaven bank deposits in tax havens, while recent IoR treaties do not, this indicates that the form of tax evasion may have changed. 6.2 The reaction of non-bank deposits in tax havens to CRS The Common Reporting Standard (CRS) contains automatic information exchange and, thus, is much stricter, as seen from the perspective of a tax evader. Interestingly, bilateral activations of automatic information exchange relationships under the CRS lead to a decline in outbound deposits similar to the one we observed during the early years of IoR treaties. Table 5 column (1) shows a 43% reduction in reaction to CRS activation with the financial weight variable again being significantly positive. — Table 5 about here — It has to be considered that the introduction of the CRS and its automatic exchange of information coincided with a number of domestic amnesties and voluntary disclosure programs. As the setup varies from country to country, with only a subset of countries implementing these, we do not attempt to explain their occurrence (see Bayer et al., 2015, for such a discussion). Important in our context is their varying success. Argentina broke records with 116 billion USD flowing back due to the amnesty,11 while the South African program has, essentially, been ignored at the time of writing. If successful, at least a part of the CRS effect we find could be driven by such amnesties (see Langenmayr, 2017, for a discussion). To investigate this, we collected data on 18 of such programs12 and add them as a dummy variable in column 2 of Table 5. The resulting effect is insignificant and economically almost zero. A caveat applies here: Several of the programs under consideration are still ongoing at the time of writing. Past experience shows that self-reporting peaks just before such programs end. Therefore, it is likely that we do not capture their full effect in these results. Still, the effect of CRS activation does 11“Argentina rakes in $116.8 billion from hidden assets amnesty” – Bloomberg April 4, 2017, https://www.bloomberg.com/news/articles/2017-04-04/argentina-rakesin-116-8-billion-from-hidden-assets-amnesty, last accessed April 16, 2019. 12This list includes one or more amnesties in Argentina, Australia, Brazil, Brazil, India, Indonesia, Israel, Kenya, Mexico, Pakistan, Russia, South Africa, South Korea, Turkey, and the United Kingdom. 23 not seem to be driven by amnesties and Langenmayr (2017) shows, that voluntary disclosure programs can even increase tax evasion. This indicates that a significant portion of capital shifted in reaction to CRS activation is not legalized. We also add the earlier signed IoR treaties and the placebo used above, both only marginally change the CRS coefficient. In the robustness section, we test for anticipation effects and can show that it is indeed the activation of the bilateral CRS exchange and not the earlier signature of the multilateral agreement that drives these results. We also test for interaction effects with previous IoR treaties and find none. Finally, we run tests in our two falsification groups, i.e. we test whether CRS activation affects bank deposits, either between tax havens or between non-havens. The expectation is that the CRS coefficients again should be insignificant. Indeed, columns (3) to (6) in Table 5 show the expected insignificant and economically small CRS coefficients. By contrast, the coefficients on the financial weight are significantly positive and the coefficients on earlier treaties between non-havens remain marginally significantly positive, just as they were for the isolated analysis of IoR treaties. Overall, we find the exact same pattern in empirical results as we did for information exchange based on IoR treaties. Bank deposits from non-haven countries to tax havens go down significantly if AEI relationships between these countries are activated in the form of the CRS. At the same time, the falsification samples are unaffected. While we see a very clear effect for the outbound deposits, it is currently not reasonable (keeping in mind the lagged effect of IoR inbound deposits) to analyze the inbound deposits, because CRSs were only activated at the end of our sample, as Figure 2 shows. These results point to a worrying tendency: Reactions to the CRS mirror those to IoR treaties, thus suggesting the CRS is not the game changer it is often touted to be. The still bilateral exchange of information leaves loopholes open but the CRS can also be circumvented directly. The “new disguise” that we described above and that protects against IoR treaties by naming a close relative as the beneficial owner will react to CRS activations. Suddenly, large sums would appear held in the name of an owner who was not previously suspected of tax evasion and therefore not targeted by information requests. However, the CRS can be circumvented by acquiring dual citizenship and tax residence in a tax haven. If the evader moves their funds to a bank account in that tax haven with local citizenship and does not disclose their original citizenship to the tax haven bank, such an account is not reported in the 24 Figure 1: International bank deposits of tax evaders 31 Table 1: Tax havens with significant available deposit data Jurisdiction population Affiliation Treaties with nonhavens Deposit time series (balanced panel: non-havens; tax havens) total liabilities (m. USD 2017:IV) outbound inbound Austria 8,711,770 EU 6 28 (8; 7) 72196 - 1 Bahamas 327,316 C.o.N 21 28 (6; 7) 35356 - 1 Bahrain 1,378,904 - 18 26 (3; 7) 54294 - 1 Belgium 11,409,077 EU 3 29 (7; 6) 103478 1 1 Bermuda 70,537 British o.T. 27 28 (6; 5) 1986 - 1 Cayman Islands 57,268 British o.T. 25 29 (6; 7) 268102 - 1 Chile 17,650,114 - 2 27 (5; 4) 2715 1 1 Guernsey 66,297 British C.D. 34 24 (3; 6) 31805 1 1 Hong Kong 7,167,403 China S.A.R. 14 27 (8; 7) 458182 - 1 Ireland 4,952,473 EU 12 28 (8; 6) 66194 1 1 Isle of Man 88,195 British C.D. 28 23 (2; 6) 25430 1 1 Jersey 98,069 British C.D. 32 25 (4; 6) 59786 1 1 Luxembourg 582,291 EU 10 29 (8; 6) 148886 1 1 Malaysia 30,949,962 C.o.N 6 27 (7; 5) 28186 - 1 Panama 3,705,246 - 17 29 (6; 5) 22344 - 1 Singapore 5,781,728 - 11 28 (7; 7) 254465 - 1 Switzerland 8,179,294 - 12 29 (8; 6) 454075 1 1 Notes: Shows tax havens that report BIS statistics or appear as counterparties of at least 10 reporting countries in the balanced panel (2003:I - 2017:IV). Population in column 2 and affiliations in column 3 are taken from the CIA World Factbook. Affiliations are abbreviated as follows: ‘British C.D.’s are the British Crown Dependencies, ‘British o.T.’s the British overseas Territories, ‘C.o.N.’ is the Commonwealth of Nations, and ‘China S.A.R.’s are special administrative zones of China. The number of information exchange treaties signed with non-havens in column 4 is based on the OECD Exchange of Tax Information Portal; narrowed down to treaties signed after 2003:I, meeting OECD standards and including the updated less stringent requirements for information exchange. The first number in column 5 shows the frequency of each tax haven in the time series reported by all other reporting countries. The first number in parentheses shows the number of non-havens reporting data in the balanced panel (inbound sample), the second indicates the number of other tax havens reporting against this specific tax haven (tax haven falsification sample). Column 6 shows the total liabilities reported by that tax haven against the rest of the world (not bilaterally) and Columns 7 and 8 show if the tax haven is included in the inbound and/or outbound sample. A comprehensive table including all tax havens listed in recent studies is available in Appendix Table A1. 32 Figure 2: Unique information exchange relationships over time TIEAs DTCs CRS 2005 2010 2015 0 20 40 60 0 20 40 60 0 200 400 600 800 number of treaty signatures tax haven with tax haven tax haven with non−haven non−haven with non−haven Notes: First depicts the number of signatures of tax information exchange agreements (TIEAs, top panel) and double taxation conventions (DTCs, middle panel). These treaties specify information-exchange-on-request (IoR). Both treaty types are based on data from the OECD Exchange of Tax Information Portal; narrowed down to treaties signed in the balanced panel (with the last treaty available signed in 2016:II), meeting OECD standards and including the updated less stringent requirements for information exchange. The number of activations of the Common Reporting Standard (CRS, bottom panel) is based on data provided in the OECD Automatic Exchange Portal. CRS activations are shown till the end of the sample: 2017:IV. Shading is based on the countrypair type with treaties between tax haven shown in black, treaties between non-havens shown in light grey and treaties signed between tax havens and non-havens shown in dark grey. The underlying tax haven list combines those of Gravelle (2015) and Johannesen and Zucman (2014) as detailed in Table 1. 33 Table 2: Reaction of deposits in tax havens (outbound) to IoR treaties Dependent variable: log(deposits) outbound outbound outbound outbound outbound outbound baseline JZ(2014) JZ(2014) JZ(2014) JZ(2014) tax haven list without sample time close list Switzerland only replica (1) (2) (3) (4) (5) (6) IoR treaty signed −0.275∗∗∗ −0.303∗∗∗ −0.327∗∗∗ −0.359∗∗∗ −0.384∗∗∗ −0.133∗∗ (0.078) (0.087) (0.087) (0.095) (0.090) (0.062) IoR treatyk=−1−0.146∗−0.138∗∗ −0.178∗∗ (0.075) (0.070) (0.076) IoR treatyk=−2−0.130∗−0.117∗∗ −0.146∗∗ (0.066) (0.059) (0.065) Financial weight 0.554∗∗∗ 0.493∗∗ 0.593∗∗ (0.199) (0.194) (0.258) countrypair f.e. Yes Yes Yes Yes Yes Yes year-qtr f.e. Yes Yes Yes Yes Yes Yes Observations 33,420 28,682 26,184 18,364 17,267 16,523 R20.080 0.081 0.084 0.085 0.105 0.120 Adjusted R20.063 0.061 0.064 0.064 0.074 0.089 Notes: Autocorrelation and heteroscedasticity robust standard errors in parentheses. The treatment variable ‘IoR treaty signed’ takes value 1 if a bilateral treaty specifying information exchange on request has been signed. Column 1 shows the baseline effect of IoR using the entire balanced panel, the restrictive treaty definition, and the full tax haven list as discussed in the main text. Column 2 introduces two leads and the weight of the bilateral integration in bank claims relative to all other countries that report data against the counterparty nonhaven (Financial weight) to show that our results are not driven by omitted developments in the banking sector. In column 3, we reduce the sample to reports by jurisdictions categorized as tax havens in Johannesen and Zucman (2014) – “JZ(2014)” for brevity – thus dropping reports by Ireland. In column 4 we additionally drop the reports by Switzerland. Column 5 changes the treaty and tax haven definitions to our baseline but only uses the sample time available to Johannesen and Zucman (2014). Column 6 shows the effect in a sample limited to the length, treaty restrictiveness, and tax haven definition used by Johannesen and Zucman (2014). The dependent variable in all columns are data on time series of deposits by non-haven counterparties in tax haven banks (outbound sample). The balanced sample consists of 557 countrypairs reported by 8 reporting tax havens against a combined 146 non-havens over 60 quarters (2003:I - 2017:IV). * denotes 10% significance, ** 5% significance, and *** 1% significance. 34 Figure 3: Sample robustness outbound ● ● ● ●●● ● ● ●● ●● ● ●● ●● ●● ● ● ● ●● ●● ●● ●●● ●● ● ● ● ● ●● ●● ● ●● ● ●● ● ● ● ● ● ● ●● ●●●●● ● ●● ● ● ● ●●● ● ● ● ● ●●● ● ●● ● ● ●●● ●● ● ●●●● ●●● ● ● ●●● ●●● ●● ●● ● ● ● ●●● ●●● ● ●● ●● ●●● ●● ●● ●● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ●● ● ● counterparty country year−quarter year −0.4 −0.3 −0.2 −0.1 0.000 0.002 0.004 0.006 0.0 0.1 0.2 0.3 0.0003 0.0004 0.0005 0.0006 0.0007 0.000 0.002 0.004 0.006 estimate p−value of estimate dropped cases ●1 2 none (baseline) Switzerland dropped Notes: The four panels plot the robustness of our results to changes in the sample. In all panels, the black rectangle shows the baseline results with the estimate plotted on the horizontal axis and its p-value plotted on the vertical axis. The top panel shows results from estimations dropping one (dark grey circle) or any combination of two (light grey cross) counterparties at a time. The second panel drops one or all combinations of two reporting countries at a time with the triangles indicate having dropped Switzerland. The third panel drops one or any combination of two year-quarters and the final panel drops one or any combination two entire years from the sample. 35 Figure 4: TIEAs and placebos non-haven – tax havens ● ● ●●●●●●●●● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●●●●●● 2003 2006 2009 2012 2015 0 2 4 6 8 10 12 number of TIEAs signed ●●●●●●●●●●●●●●●●●● ● ●● ●● ● ● ●●● ● ● ● ●● ● ● ●● ●●● ●●● ●●● ● ●●● ●●●●●●●●●● 2003 2006 2009 2012 2015 0 5 10 20 30 number of placebos drawn Notes: The top panel plots the number of TIEAs signed between tax havens and non-havens over the sample period in the circles. Each circle denotes the number of treaties (vertical axis) signed during the respective quarter (horizontal axis). The solid line depicts a cubic spline calculated over these treaties. This spline is normalized and used to draw a number of placebo treaties which mirror the time momentum of the TIEAs. These placebo treaties are depicted in the bottom panel. Again, the circles represent the number of placebo treaties per quarter. The solid line in the bottom panel is a spline based on the placebos and only shown for comparability. Since the group of countrypairs that have not signed treaties is much larger than those which have, the vertical axes of both panels differ. The placebo treaties are then randomly assigned to the countrypairs which have not signed a bilateral treaty. 36 Table 3: Treaty differentiation and falsification results Dependent variable: log(deposits) tax haven non-haven falsification falsification outbound outbound sample sample (1) (2) (3) (4) (5) (6) IoR treaty signed −0.294∗∗∗ −0.064 0.205∗ (0.082) (0.123) (0.122) Placebo assigned −0.003 (0.045) TIEA signed −0.505∗∗∗ −0.253 0.294 (0.094) (0.200) (0.196) DTC signed 0.100 0.067 0.194 (0.122) (0.132) (0.135) Financial weight 0.553∗∗∗ 0.568∗∗∗ 0.525∗∗∗ 0.508∗∗∗ 0.385∗∗∗ 0.384∗∗∗ (0.199) (0.200) (0.196) (0.195) (0.131) (0.131) countrypair f.e. Yes Yes Yes Yes Yes Yes year-qtr f.e. Yes Yes Yes Yes Yes Yes Observations 28,682 28,682 11,133 11,133 27,431 27,431 R20.080 0.088 0.083 0.086 0.152 0.152 Adjusted R20.061 0.068 0.061 0.064 0.135 0.135 Notes: Autocorrelation and heteroscedasticity robust standard errors in parentheses. The treatment variables takes value 1 if a bilateral treaty specifying information exchange on request has been signed. Column 1 adds the placebo treatment to the IoR specification. Column 2 separates the treatment variable into double taxation conventions (DTC) and tax information exchange agreements (TIEA). Columns 3 and 4 change the sample to the tax haven falsification group and repeat columns 1 and 2 in this sample. Columns 5 and 6 change the sample to the non-haven falsification group and repeat the same exercise. The sample in columns 1 and 2 are data on time series of deposits by non-haven counterparties in tax haven banks (outbound sample) resulting in 557 countrypairs reported by 8 reporting tax havens against a combined 146 non-havens. The dependent variable in columns 3 and 4 are data on time series of deposits by tax haven counterparties in tax haven banks (tax haven falsification sample). This sample consists of 197 tax haven - tax haven countrypairs. The dependent variable in columns 5 and 6 are data on time series of deposits by non-haven counterparties in non-havens banks (non-haven falsification sample). This sample consists of 473 countrypairs. All series run over 60 quarters (2003:I - 2017:IV). * denotes 10% significance, ** 5% significance, and *** 1% significance. 37 Figure 5: Average positions of countrypairs at time of IoR agreement ● ● ● ● ● ● ● ●● ● ● ● ● ●●●● ● ● ● ●● ● ● ● ● outbound inbound 2002 2005 2008 2010 2012 2015 0 5 10 15 20 25 0 5 10 15 20 25 inv. hyperb. sine of million USD ●exports fdi fpi ●● ● ● ● ● ● ● ● ● ●● ● 0 50 100 150 2002 2005 2008 2010 2012 2015 treaty signatures Notes: The three panels compare the yearly development of bilateral positions at the time of IoR agreements in the outbound (top panel) and inbound (middle panel) directions with signature momentum over time shown in the bottom panel. The top two panels show the development of the same variables used by Bilicka and Fuest (2014) namely exports of goods and services, the stock of foreign direct investment (fdi), and the stock of foreign portfolio investment (fpi) from the reporting country to the counterparty with which a treaty is signed. All variables are in million USD and transformed using the log-equivalent inverse hyperbolic sine transformation to retain zeros and negative stocks. The momentum visible in the bottom panel is not mirrored in the top and middle panel, suggesting that newly signed treaties are signed between countrypairs of similar bilateral importance as those that signed treaties earlier. 38 Figure 6: Rolling estimation window: Outbound −0.4 −0.2 0.0 2007 2008 2009 2010 2011 2012 signature effect Notes: This figure shows the reaction of deposits in tax havens by non-haven counterparties (outbound) over time. The solid black line shows coefficients of regressions in a rolling window of +/- 8 quarters around the year-quarter plotted on the horizontal axis. The coefficient captures the effect of the most significant lags of IoR signatures (k = -2 : k = 2). The dark grey area denotes 10% significance while the light grey area denotes 5% significance in pointwise confidence bands. Estimated coefficients are plotted on the vertical axis and the horizontal line denotes a 0 effect. 39 Figure 7: Deposit shifting reactions 2005 2006 2007 2008 −0.04 0.00 0.04 0.08 deposit shifting effects pvalues > 10 % < 5 % Notes: This figure shows the reaction of deposits in tax havens by non-haven counterparties (outbound) to treaties the counterparty non-haven signs with other tax havens in an unbalanced panel. Bar lengths show coefficient sizes of regressions in a rolling window of +/- 8 quarters around the quarter plotted on the horizontal axis. The deposit shifting variable is constructed by adding the IoR treaties a counterparty non-haven has signed with other tax havens. Shading denotes significance and no estimate has a p-value between 0.1 and 0.05. 40 Trinidad & Tobago 1,220,479 C.o.N - 25 (3; 2) - - 1 Turks & Caicos Islands 51,430 British o.T. 13 20 (3; 3) - - 1 U.S. Virgin Islands 102,951 USA - - (-; -) - - - Uruguay 3,351,016 - 18 28 (3; 4) - - 1 Vanuatu 277,554 C.o.N 8 19 (3; 3) - - 1 Vatican City 1,000 - - - (-; -) - - - Notes: Shows all countries that show up on tax haven lists of recent studies. Population in column 2 and affiliations in column 3 are taken from the CIA World Factbook. Affiliations are abbreviated as follows: British C.D.s are the British Crown Dependencies, British o.T.s the British overseas Territories, C.o.N. is the Commonwealth of Nations, and China S.A.R.s are special administrative zones of China. The number of information exchange treaties signed with non-havens in column 4 is based on the OECD Exchange of Tax Information Portal; narrowed down to treaties signed after 2003:I, meeting OECD standards and including the updated less stringent requirements for information exchange. The first number in column 5 shows the frequency of each tax haven in the time series reported by all other reporting countries. The first number in parentheses shows the number of non-havens reporting data in the balanced panel (inbound sample), the second indicates the number of other tax havens reporting against this specific tax haven (tax haven falsification sample). Column 6 shows the total liabilities reported by that tax haven against the rest of the world (not bilaterally) and Columns 7 and 8 show if the tax haven is included in the inbound and/or outbound sample. 47 Table A2: Disagreements in tax haven lists Tax.Haven Glautier & Bassinger (1987) Hines & Rice (1994) OECD (2000) Dharmapala (2008) Johannesen & Zucman (2014) Gravelle (2015) Andorra 1 1 1 1 1 Anguilla 1 1 1 1 1 Aruba 1 1 1 1 Austria 1 1 Belgium 1 Chile 1 Costa Rica 1 1 1 Dominica 1 1 1 1 1 Ireland 1 1 1 1 Jordan 1 1 1 Lebanon 1 1 1 Macau 1 1 1 1 Malaysia 1 Maldives 1 1 1 1 Malta 1 1 1 1 1 Marshall Islands 1 1 1 1 1 Mauritius 1 1 1 Monaco 1 1 1 1 1 Nauru 1 1 1 1 Netherlands 1 Niue 1 1 1 1 Samoa 1 1 1 1 San Marino 1 1 1 1 Seychelles 1 1 1 1 St. Lucia 1 1 1 1 1 Tonga 1 1 1 Trinidad & Tobago 1 U.S. Virgin Islands 1 1 1 1 Uruguay 1 Notes: All of the surveys shown in this table include the following list of countries as tax havens: Antigua and Barbuda, Bahamas, Bahrain, Barbados, Belize, Bermuda, British Virgin Islands, Cayman Islands, Cook Islands, Curacao*, Cyprus, Gibraltar, Grenada, Guernsey**, Hong Kong, Isle of Man, Jersey**. Liberia, Liechtenstein, Luxembourg, Montserrat, Panama, Saint Kitts and Nevis, Saint Vincent and the Grenadines, Singapore, Sint Maarten (Dutch part)*, Switzerland, Turks and Caicos Islands, and Vanuatu. * Curacao and Sint Maarten (Dutch part) are included as the “Netherlands Antilles” in some publications, they separated on the 10th of October 2010. ** Guernsey and Jersey are included as the “Channel Islands” in some publications. 48 Appendix 1 Reactions to US policy measures The US led several initiatives targeting Swiss bank secrecy, completing two intergovernmental agreements as a result. While Johannesen and Stolper (2018) as well as Johannesen et al. (2018) analyze these and related measures in detail, we limit our analysis to the two intergovernmental treaties that we expect to be important enough to generate reactions in aggregate bilateral deposit data. The expectation about the respective coefficient signs are straightforward in two cases: the first is an agreement between both governments that the largest Swiss bank, UBS, which was at that time involved in a criminal case of tax evasion in the US, would reveal the identity of 4,450 customers to US authorities. This was the first official undermining of the Swiss banking secrecy in history and, as it was detrimental for tax evaders, we expect a negative coefficient sign. The case is reversed for the US-Swiss bank program that allows Swiss banks to apply for an amnesty by reporting previously hidden deposits of US nationals. As of April 2019, 81 banksA1 have signed up for this program. They will potentially report deposits that were previously hidden positions. Indeed, empirical results confirm these expectations, as Table A3, column 1 shows. However, since these treaties only act on one bilateral connection in the sample (the Switzerland – United States countrypairs constitutes 60 observations in each sample), we do not want to overinterpret the results. Of the 50 countries in our balanced panel that have signed FATCA, eight report liabilities against the US and, thus, results should be more reliable. The coefficient sign is negative and highly significant for the outbound direction (see columns 3 and 4), indicating that deposits from the US to tax havens are reduced due this agreement. In summary, for these US agreements, we conclude that we find the same kind of reactions of outbound deposits as for the earlier IoR treaties. This also holds when we control for the latter (columns 2 and 4) and the pattern is partly also seen in inbound deposits. The latter is most obvious for the two US-Swiss agreements. Regarding FATCA, the inbound coefficient is insignificant. This effect is largely expected because the inbound deposits are diluted due to the US-focused nature of these treaties. While, in the case of CRS activations and IoR treaties, we can analyze a world-wide inbound sample looking at positions from all tax havens to all nonhavens, FATCA only allows looking for direct deposits back in the US. However, such deposits do not need to be deposited in the country where they came from; here the US. Accordingly, we do not measure an inbound reaction to the agreement. The same logic applies to direct roundA1See https://www.justice.gov/tax/swiss-bank-program for a list of participating banks. Last accessed April 16, 2019 49 tripping deposits at the countrypair level, where the matching is also weak for IoR treaties, as we demonstrate in Appendix 3. Another effect going into this same direction, but not referring to tax havens, is indicated by the mounting evidence that some of the activities usually attributed to tax havens are increasingly carried out directly in non-haven countries. Dyreng et al. (2013) show intra US competition along such lines, focusing on the role of Delaware as a domestic tax haven. Sharman (2010) provides qualitative evidence showing that financial service providers in OECD countries make it possible to set up shell companies with bank accounts without any personal identification. Representatives of Bermuda, the Cayman Islands, and the Isle of Man pointed out this ‘hypocrisy’ during the 2016 London corruption summit, arguing that the focus on small jurisdictions is outdated (European Parliament, 2016). This allegation is unsubstantiated: our research shows robust and strong evasion effects in tax havens. The shell companies set up by Sharman (2010) via firms in OECD countries are also mostly incorporated in small, affluent, and well-governed islands, as characterized by Dharmapala and Hines (2009b) and Hines (2010). Still, this outside option for evaders is a potential avenue for future research. 50 Table A3: Reactions to US specific measures Dependent variable: log(deposits) inbound inbound outbound outbound (1) (2) (3) (4) 2009 US-CH Agreement −0.170∗∗∗ −0.257∗∗∗ −0.094∗∗∗ −0.119∗∗∗ (0.057) (0.070) (0.034) (0.034) US-Swiss Bank Program 0.322∗∗∗ 0.332∗∗∗ 0.596∗∗∗ 0.576∗∗∗ (0.097) (0.098) (0.134) (0.121) FATCA signed 0.162 0.124 −0.503∗∗∗ −0.496∗∗∗ (0.126) (0.127) (0.156) (0.141) IoR treaty signed −0.370∗∗∗ −0.274∗∗∗ (0.124) (0.078) countrypair f.e. Yes Yes Yes Yes year-qtr f.e. Yes Yes Yes Yes Observations 10,200 10,200 33,420 33,420 R20.183 0.196 0.076 0.082 Adjusted R20.164 0.177 0.059 0.064 Notes: Autocorrelation and heteroscedasticity robust standard errors in parentheses. Column 1 shows the effect of US specific regulation attempts: two bilateral treaties with Switzerland (the US-CH Agreement and the USSwiss Bank Program) that only act on this bilateral dimension as well as FATCA, which was signed with 127 countries. In column 2 we control for the international IoR treaties previously used. Columns 3 and 4 replicate this analysis in the outbound sample. The dependent variable in columns 1 and 2 are data on time series of deposits by tax haven counterparties in non-havens banks (inbound sample). The sample consists of 170 countrypairs reported by 11 reporting non-havens against a combined 44 tax havens over 60 quarters (2003:I - 2017:IV). The dependent variable in columns 3 and 4 are data on time series of deposits by non-haven counterparties in tax haven banks (outbound sample). This sample consists of 557 countrypairs reported by 8 reporting tax havens against a combined 146 non-havens. * denotes 10% significance, ** 5% significance, and *** 1% significance 51 Appendix 2 Robustness checks These robustness analyses are organized along four themes. To test the robustness of our results on IoR treaties, we split the analysis into, first, outbound deposits, second, inbound deposits. We also provide treatment graphs of a dynamic differences-in-differences exercise which confirms our results in both directions. Third, we run an extensive robustness check over different tax haven lists since the rough binary classification into tax haven and non-haven could affect all results reported here. Finally, we investigate the relationship between the CRS and IoR treaties and show that the declining effect of IoR treaties over time is not driven by later CRS agreements. We also show robustness of the CRS results and confirm that the CRS reactions are close to identical for countrypairs that have or have not participated in previous policy initiatives. Replication of the outbound analysis. Since the outbound analysis is inspired by Johannesen and Zucman (2014), we continue the robustness analysis by replicating their results with our data. Thus, Table A4 uses their preferred lag order specification in order to compare effects directly. Column 1 shows the results using our sample, tax haven list, and treaty specification. Column 2 again reduces the sample length to 2003:IV-2011:II, and shows stronger impacts, consistent with the fading treatment effect we discuss in the main text. — Table A4 about here — In order to replicate the Johannesen and Zucman (2014) results, we first change the tax haven list to that used by the authors (column 3) before also employing their treaty variable, which they made available (column 4). The treatment variable now includes a number of treaties that were not reviewed or did not meet the OECD peer review standard as well as domestic law changes triggering information exchange. The results in column 4 are quantitatively much closer to those of Johannesen and Zucman (2014). Thus, it is not the sample length or the tax haven list that differentiates our quantitatively larger results, but rather the differing list of information exchange treaties. Since the goal of this study is not to evaluate the OECD initiative quantitatively but to employ information exchange treaties to identify tax evasion, these results confirm our choice of a very restrictive treaty definition as a lens to detect tax evasion. OECD and G20 meetings. As we point out in the main text, IoR treaties were signed following active pressure by the OECD and the G20. In Table A5 we show results testing the hypothesis that these meetings might have a signaling effect dominating treaty signatures. 52 Column 1 shows the outbound baseline while the following columns alter the underlying sample used as shown by the decreasing number of observations. In column 2, we exclude the time frame during which the OECD initiative on harmful tax practices was actively endorsed and promoted by G20 meetings (2008:III-2010:I). In line with other robustness checks, this sample change does not affect our results much. Tax evasion, however, has stayed on the political agenda of the G20 until today. In Column 3, we therefore exclude all quarters following a G20 meeting in our sample period. We also exclude all IoR treaties signed during these quarters entirely from the analysis thus reducing the total number of IoR treaties in our balanced panel from 251 to 185. Still, results hold qualitatively. Columns 4-6 repeat these exercises for inbound deposits. Together with the treatment analysis reported further below, we take this as evidence that our effects are not driven by events such as G20 meetings but rather by the treaty signatures themselves. — Table A5 about here — Unbalanced panel. In order to verify the validity of using an unbalanced panel to show deposit shifting effects, we replicate the main outbound effect of a decreasing effect of IoR treaties. This is a direct copy of Figure 6 in the main text with the only difference being that the entire available locational banking statistics as reported by tax havens is used. Figure A1 shows that both the significant negative effect of IoR treaties as well as the decreasing effect over time show up here as well. — Figure A1 about here — Sample period. To test the robustness of results concerning inbound deposits, we first test whether the results we find are an artifact of the sample period or hold over sub-samples. Table A6 reports the results with column 1 repeating the baseline from Table 4 for easy comparison. Column 2 shows results for a sample limited to 2003:IV - 2011:II, which is the time period used by Johannesen and Zucman (2014). Our main result holds, as seen by the significant and economically meaningful reaction in the fourth to the sixth lags. If anything, reactions are stronger. Column 3 shows results with the financial crisis excluded, which we start with 2007:II and the Bear Stearns fund failures, running through 2008:IV, when the US started emerging from the crisis. Again, the results are virtually unchanged, which shows that they do not depend on the sample period. — Table A6 about here — 53 Macro controls. Due to the rich bilateral nature of our dataset and limited macroeconomic data availability for tax havens, macroeconomic control variable have proven mostly meaningless for our analysis. Quarterly bilateral data starting in 2003 is usually not available for the type of jurisdiction introduced in Table 1: even basic macroeconomic data is hard to find for the typical tax haven. To show this, column 4 includes a number of such control variables in the spirit of Hanlon et al. (2015): namely the log of the number of telephone landlines per 100 persons as well as the population in the counterparty tax haven and the growth rate of GDP per capita as well as the population in the reporting non-haven. This data is taken from the yearly World Development Indicators dataset and linearly interpolated to match our quarterly frequency. As it is neither bilateral, nor available for all tax havens, the sample is reduced. As expected, results are unchanged and some coefficients are insignificant. Bank claims. Next, we check the identification strategy by testing for effects of information exchange in the parts of bank balance sheets where we do not expect tax evasion to have a significant impact. So far, we have used bank liabilities against international nonbank counterparties to proxy deposits of firms and households. Indeed, if such deposits are used for tax evasion and if we have identified tax evasion correctly, we should find no discernible effects for bank claims. These claims include, for example, loans to other banks and do not include deposits (BIS, 2013). In non-haven countries, from which this data are reported, we do not expect bank lending to react significantly to decreased tax evasion and banks themselves have not been convicted in the kind of personal tax evasion cases we identify. We find results neither for overall claims (column 5) nor for loan claims (column 6). The fact that claims are lower after more than three years (lag 13 and later) is consistent with our interpretation of a general negative signaling effect of treaties on the attractiveness of the tax haven that signs them. Inbound sample effects. Replicating the results of Figure 3 in the main text for inbound deposits, Figure A2 shows the same four panels, namely the fluctuation of estimates and the associated p-values around the baseline (the black square). In the top panel, we sequentially drop one (dark grey circles) and any combination of two (light grey crosses) counterparties from the sample and plot the re-estimated treatment effects and their p-values. Having dropped Switzerland as a counterparty is again highlighted with triangles. The second panel does the same for the country dimension. The third panel shows the effects of dropping one or two yearquarters at a time and the bottom panel finally shows dropping one or two entire years from the panel. Inbound results are even more robust than outbound results with no specification turning insignificant and estimates remaining stable in economic size. Neither particular time 54 periods nor countries drive our inbound results. The inbound results are quite robust to dropping particular tax havens since we draw on reports against a total of 44 tax havens. — Figure A2 about here — More treaties. In order to establish the robustness of our results to our choice of treaties, we broaden our very restrictive definition of which treaties constitute a credible threat of detection. Table A7 compares the baseline (column 1) to the inclusion of the 56 treaties signed within the balanced panel dimension that do not include paragraphs 4 and 5 but were reviewed by the OECD and met the standard (column 2). Then, we include another 19 treaties that were not reviewed at the time of analysis (column 3). Finally, we also include those 25 treaties that were reviewed, but failed to meet the OECD standard (column 4). Results are broadly consistent. — Table A7 about here — Treatment analysis. Here, we verify that the anticipation effects in outbound deposits and the lagged effects in the inbound deposits are no artifact of the strong fluctuations in international financial data. The treatment analysis used here allows us to visually compare developments in the treatment groups to aggregate developments while taking into account distance to IoR treaties explicitly. For outbound deposits, we run a dynamic differences-indifferences analysis of the following form: log(depositsi jkt) = αi j +γt+θk+ K=−4 ∑ k=−20 βk(IoRtreatyk i j,k)+ K=20 ∑ k=−2 τk(IoRtreatyk i j,k)+εi jkt (3) The notation follows the main text: subscripts i j denote the countrypair, tthe respective year-quarter, and kthe distance from a treaty. The treatment fixed effects θkare based on this distance and the coefficients we plot are interpretable relative to the omitted treatment distance period, in the outbound case k=−3. We use the placebos introduced in section 4.2 as the k=0 distance to ‘treaty’ for the control group countrypairs that have not signed IoR treaties. We focus on 20 quarters before and after treaty signatures to see if the pre-treatment trend is reasonably stable. The same setup is run for inbound deposits, this time taking k=0 as the point of comparison, as no anticipation was visible there. Figure A3 shows the resulting treatment coefficients and associated standard errors plotted over the distance to treatment on the horizontal axis. Both outbound (top panel) and inbound 55 (bottom panel) graphs show a clear drop after IoR treaties. Pre-treatment trends aren’t flat but reasonably stable for macroeconomic international financial data aggregated for the bilateral connection. As our panel results indicate, inbound deposits mirror the reaction but not the anticipation effect that we observe in outbound deposits. Both results, however, are clearly not driven by single lags of the treatment. These results show average reactions for all treaties in the sample and are thus not to be confused with the reactions over time we show in the main text where we analyze treaties sequentially. — Figure A3 about here — Tax haven list. The controversies surrounding tax haven lists are very relevant for the study at hand, it determines the assignment of our inbound and outbound samples. These data can change considerably when changing the tax haven list. For outbound deposits, table A8 reports results for different lists that are commonly used in the literature (see Table A2 for an overview of different lists). Column 1 repeats the baseline results before we change the tax haven dimension in the counterparties to the list provided in Gravelle (2015) (column 2). Results remain qualitatively the same. — Table A8 about here — A tougher test is the reduction of the tax haven list to the absolute minimum on which recent studies agree. Therefore, we compile a tax haven list with the consensus candidates of Bilicka and Fuest (2014); Dharmapala (2008); Dharmapala and Hines (2009b); Gravelle (2015); Hanlon et al. (2015); Hines and Rice (1994); Johannesen and Zucman (2014); OECD (2000). This means that we remove 21 countriesA2 from the tax haven list used so far. Column 3 shows that our results hold despite losing almost half of our observations. To go even further, we now include the tax havens dropped from the tax haven list as non-haven countries. Macao is now in the same category as France and Sweden when it comes to facilitating tax evasion. Column 4 that in the outbound direction, results are not affected by this change. Table A9 repeats the same exercise for inbound deposits. Since the tax haven dimension in this direction is the reporting country level, we rely on a much smaller number of tax havens. Here, we have a larger number of counterparty tax havens (44) and can thus differentiate the OECD (2000) list (column 2) from the Gravelle (2015) list (column 3). Moving to the A2These include Aruba, Austria, Belgium, Chile, Costa Rica, Ireland, Jordan, Lebanon, Macao, Malaysia, Maldives, Mauritius, Nauru, Niue, Samoa, San Marino, Seychelles, Tonga, Trinidad and Tobago, Urugay and the US Virgin Islands. 56 Figure A2: Sample robustness inbound ●● ● ●● ● ● ● ● ●●● ● ● ● ● ●●● ●● ● ● ● ● ●●● ● ● ● ● ●● ● ● ●●● ●● ●● ● ●●● ●●●●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● counterparty country year−quarter year −0.5 −0.4 −0.3 0.000 0.005 0.010 0.015 0.020 0.00 0.03 0.06 0.09 0.12 0.0020 0.0025 0.0030 0.0035 0.001 0.002 0.003 0.004 0.005 estimate p−value of estimate dropped cases ●1 2 none (baseline) Switzerland dropped Notes: The four panels plot the robustness of our inbound results to changes in the sample. In all panels, the black rectangle shows the baseline results with the estimate plotted on the horizontal axis and its and p-value plotted on the vertical axis. The top panel shows results from estimations dropping one (dark grey circle) or any combination of two (light grey cross) counterparties at a time with the triangles indicate having dropped Switzerland. The second panel drops one or all combinations of two reporting countries at a time. The third panel drops one or any combination of two year-quarters and the final panel drops one or any combination two entire years from the sample. 63 Table A7: Robustness of inbound results to treaty definition Dependent variable: log(deposits) baseline balanced signature + no para 4/5 + not reviewed inbound inbound inbound inbound (1) (2) (3) (4) IoR treatyk=1:k=3−0.167∗−0.356∗∗∗ −0.371∗∗∗ −0.387∗∗∗ (0.095) (0.123) (0.125) (0.124) IoR treatyk=4:k=6−0.331∗∗∗ −0.331∗∗∗ −0.325∗∗∗ −0.341∗∗∗ (0.107) (0.113) (0.112) (0.111) IoR treatyk>6−0.425∗∗∗ −0.383∗∗∗ −0.377∗∗∗ −0.393∗∗∗ (0.143) (0.136) (0.135) (0.134) countrypair fixed effects Yes Yes Yes Yes year-qtr fixed effects Yes Yes Yes Yes Observations 10,200 10,200 10,200 10,200 R20.197 0.195 0.195 0.197 Adjusted R20.179 0.177 0.177 0.178 Notes: Autocorrelation and heteroscedasticity robust standard errors in parentheses. The treatment variables ‘IoR treaty’ take value 1 k quarters from the IoR treaty, k=1 : k=3 for example takes value one for the first three quarters after a treaty. Column 1 repeats the baseline. Column 3 includes treaties signed in the sample period that did not include the paragraphs relaxing the requirement to request information to ‘foreseeably relevant’ from ‘necessary’ information. Column 4 additionally includes treaties signed in the sample period that were not reviewed by the OECD and column 5 includes treaties that were reviewed but did not meet the standard. The dependent variable are data on time series of deposits by tax haven counterparties in non-havens banks (inbound sample). The sample consists of 170 countrypairs reported by 11 reporting non-havens against a combined 44 tax havens over 60 quarters (2003:I - 2017:IV). * denotes 10% significance, ** 5% significance, and *** 1% significance. 64 Figure A3: Dynamic Differences-in-Differences −0.5 0.0 0.5 −20 −10 0 10 20 distance coefficients Outbound −1.0 −0.5 0.0 0.5 −20 −10 0 10 20 distance coefficients Inbound Notes: The two panels test the robustness of our main results outbound (Tables 2) and inbound (Table 4) in a dynamic differences-in-differences specification. The top panel shows outbound deposits around IoR treaty signatures (vertical line) against the baseline k=−3 as explained in the text. The bottom panel shows inbound deposits against the baseline k=−1. All series are standardized at the countrypair level. The dark grey area denotes 10% significance while the light grey area denotes 5% significance of the associated coefficients. Estimated coefficients are plotted on the vertical axis and the horizontal line denotes a 0 effect. 65 Table A8: Robustness of results to changes in tax haven list: outbound Dependent variable: log(deposits) Gravelle consensus consensus baseline 2015 tax havens both outbound outbound outbound outbound (1) (2) (3) (4) IoR treaty signed −0.287∗∗∗ −0.324∗∗∗ −0.356∗∗∗ −0.350∗∗∗ (0.082) (0.083) (0.082) (0.077) IoR treatyk=−2−0.147∗∗ −0.164∗∗ −0.148∗∗∗ −0.143∗∗∗ (0.066) (0.067) (0.056) (0.053) IoR treatyk=−1−0.162∗∗ −0.187∗∗ −0.175∗∗∗ −0.160∗∗ (0.074) (0.075) (0.066) (0.063) countrypair f.e. Yes Yes Yes Yes year-qtr f.e. Yes Yes Yes Yes Observations 33,420 25,560 22,980 25,500 R20.081 0.086 0.097 0.101 Adjusted R20.063 0.068 0.079 0.084 Notes:Autocorrelation and heteroscedasticity robust standard errors in parentheses. The treatment variables ‘IoR treaty’ take value 1 k quarters from the IoR treaty, k=−1 for example denoting the quarter before a treaty. Column 1 repeats the outbound baseline, column 2 limits the tax haven list to that proposed by Gravelle (20015). Column 3 (consensus tax havens) limits the tax havens to those agreed on by the authors cited in the text and detailed in Appendix A2. Column 4 (consensus both) adds the countries thus excluded from the tax-haven list to sample as non-haven countries. The dependent variable are data on time series of deposits by non-haven counterparties in tax haven banks (outbound sample). The baseline sample consists of 557 countrypairs reported by 8 reporting tax havens against a combined 146 non-havens over 60 quarters (2003:I - 2017:IV) and is changed as the tax haven list changes over the columns. . * denotes 10% significance, ** 5% significance, and *** 1% significance. 66 Table A9: Robustness of results to changes in tax haven list: inbound Dependent variable: log(deposits) OECD Gravelle consensus consensus baseline 2000 2015 tax havens both inbound inbound inbound inbound inbound (1) (2) (3) (4) (5) IoR treatyk=1:k=3−0.167∗−0.281∗∗∗ −0.267∗∗∗ −0.300∗∗∗ −0.146 (0.095) (0.107) (0.102) (0.108) (0.097) IoR treatyk=4:k=6−0.331∗∗∗ −0.459∗∗∗ −0.436∗∗∗ −0.484∗∗∗ −0.262∗∗ (0.107) (0.128) (0.120) (0.129) (0.124) IoR treatyk>6−0.425∗∗∗ −0.523∗∗∗ −0.514∗∗∗ −0.557∗∗∗ −0.340∗ (0.143) (0.170) (0.160) (0.173) (0.188) countrypair f.e. Yes Yes Yes Yes Yes year-qtr f.e. Yes Yes Yes Yes Yes Observations 10,200 7,080 8,220 6,840 9,240 R20.197 0.214 0.215 0.219 0.103 Adjusted R20.179 0.194 0.195 0.198 0.081 Notes: Autocorrelation and heteroscedasticity robust standard errors in parentheses. The treatment variables ‘IoR treaty’ take value 1 k quarters from the IoR treaty, k=1 : k=3 for example takes value one for the first three quarters after a treaty. Column 1 repeats the inbound baseline, column 2 limits the tax haven list to that proposed by the OECD (2000), column 3 to that used by Gravelle (20015). Column 4 (consensus tax havens) limits the tax havens to those agreed on by the authors cited in the text and detailed in Appendix Table A2. Column 5 (consensus both) adds the countries thus excluded from the tax-haven list to sample as non-haven countries. The dependent variable are data on time series of deposits by tax haven counterparties in non-havens banks (inbound sample). The baseline sample consists of 170 countrypairs reported by 11 reporting non-havens against a combined 44 tax havens over 60 quarters (2003:I - 2017:IV) and is changed over the columns as tax haven lists are changed. * denotes 10% significance, ** 5% significance, and *** 1% significance. 67 Figure A4: Sample robustness of CRS results ● ● ● ● ●● ● ●●● ● ● ●● ●● ● ●● ● ● ● ●●●● ● ● ●●● ●● ● ● ●● ●● ●● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ●●●● ● ●● ●●● ●● ●● ● ● ● ● ●● ● ●● ●● ●● ● ● ●● ● ●● ●● ● ● ● ● ●● ● ● ●●● ● ● ● ●● ●●● ● ●● ●● ● ●● ● ●●● ●● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● counterparty country year−quarter year −0.55 −0.45 −0.35 0.000000 0.000002 0.000004 0.000006 0.000008 0.000 0.001 0.002 0.003 0.000000050 0.000000075 0.000000100 0.000000125 0.000000150 0.000000175 0.0000000 0.0000002 0.0000004 0.0000006 0.0000008 estimate p−value of estimate dropped cases ●1 2 none (baseline) Switzerland dropped Notes: The four panels plot the robustness of the CRS results to changes in the sample. In all panels, the black rectangle shows the baseline results with the estimate plotted on the horizontal axis and its and p-value plotted on the vertical axis. The top panel shows results from estimations dropping one (dark grey circle) or any combination of two (light grey cross) counterparties at a time with the triangles indicate having dropped Switzerland. The second panel drops one or all combinations of two reporting countries at a time. The third panel drops one or any combination of two year-quarters and the final panel drops one or any combination two entire years from the sample. The year 2017 is omitted from the bottom panel since CRS activations only started in the fourth quarter of 2016. 68 Table A10: CRS/MCAA anticipation in late treaty signatures Dependent variable: log(deposits) outbound outbound outbound outbound outbound outbound (1) (2) (3) (4) (5) (6) CRS activation −0.356∗∗∗ −0.303∗∗∗ −0.303∗∗∗ −0.302∗∗∗ −0.289∗∗∗ (0.078) (0.078) (0.074) (0.074) (0.074) MCAA −0.088 (0.069) IoR signed before 2013:I −0.366∗∗∗ −0.344∗∗∗ −0.344∗∗∗ −0.346∗∗∗ (0.085) (0.083) (0.083) (0.083) IoR signed after 2013:I 0.188 0.127 0.195 (0.133) (0.770) (0.199) IoR signed after 2013:I * 0.091 MCAA signature cpair (0.778) IoR signed after 2013:I * 0.024 CRS activation cpair (0.257) IoR after 2013:I followed 0.141 by but set to 0 after CRS (0.153) IoR signed after 2013:I, 0.195 not followed by CRS (0.199) countrypair f.e. Yes Yes Yes Yes Yes Yes year-qtr f.e. Yes Yes Yes Yes Yes Yes Observations 33,420 33,420 33,420 33,420 33,420 33,420 R20.078 0.079 0.084 0.086 0.086 0.086 Adjusted R20.061 0.061 0.066 0.069 0.069 0.068 Notes:Autocorrelation and heteroscedasticity robust standard errors in parentheses. The treatment variables take value 1 after signature or activation. Column 1 shows the effect of CRS activation without controls, column 2 adds a dummy indicating if the countrypair has signed up to the multilateral competent authority agreement (MCAA) after 2014:III within which the CRS can be activated. Column 3 shows differentiated effects for IoR treaties signed before and after 2013:I with the latter interacted with MCAA signature countrypairs (columns 4) and bilateral CRS activation countrypairs (column5). Column 6 finally splits the resulting interaction into two variables. The first captures treaties signed after 2013:1 followed by CRS activation but with the the CRS quarters set to zero. The second takes value 1 after IoR treaties signed after 2013:I for countrypairs that did not activate the CRS during the sample period. The dependent variable are data on time series of deposits by non-haven counterparties in tax haven banks (outbound sample). The baseline sample consists of 557 countrypairs reported by 8 reporting tax havens against a combined 146 non-havens over 60 quarters (2003:I - 2017:IV) and is changed as the tax haven list changes over the columns. * denotes 10% significance, ** 5% significance, and *** 1% significance. 69 Table A11: CRS activations and other policy initiatives Dependent variable: log(deposits) outbound outbound outbound outbound outbound outbound IoR signed no IoR signed (1) (2) (3) (4) (5) (6) CRS activation −0.356∗∗∗ −0.338∗∗∗ −0.351∗∗∗ −0.329∗∗∗ −0.308∗∗∗ (0.078) (0.116) (0.128) (0.117) (0.075) IoR treaty signed −0.255∗∗∗ −0.233∗∗∗ −0.243∗∗∗ (0.075) (0.075) (0.077) CRS * IoR 0.059 (0.136) EU STD −0.178∗∗ (0.074) FATCA signed −0.462∗∗∗ (0.144) all AEI −0.235∗∗∗ (0.057) countrypair f.e. Yes Yes Yes Yes Yes year-qtr f.e. Yes Yes Yes Yes Yes Observations 33,420 33,420 7,260 26,160 33,420 33,420 R20.078 0.083 0.140 0.072 0.086 0.084 Adjusted R20.061 0.065 0.118 0.054 0.068 0.067 Notes: Autocorrelation and heteroscedasticity robust standard errors in parentheses. The treatment variables take value 1 after signature or activation. Column 1 shows the effect of CRS activation without controls, column interacts this variable with IoR treaty signatures. Column 3 shows CRS effects in a sample reduced to such countrpyairs that have signed IoR treaties during our sample time and column 4 shows the sample of countrypairs that have not. Column 5, shows results on all policy initiatives considered in this text and in column 6, all initiatives that include automatic information exchange (AEI) are lumped into one dummy variable that takes value one if either the EU savings tax directive, FATCA, or the CRS are signed or activated (CRS) respectively. The baseline sample consists of 557 countrypairs reported by 8 reporting tax havens against a combined 146 non-havens over 60 quarters (2003:I - 2017:IV) and is changed in columns 3 and 4 as pointed out above. * denotes 10% significance, ** 5% significance, and *** 1% significance. 70 Appendix 3 Linkages between outbound and inbound deposits Having examined outbound and inbound deposits above, we now analyze relations between these two. In a world of frictionless capital markets, there does not need to be any relation between outbound and inbound effects as the capital that was transferred to a tax haven may be reinvested anywhere in the world again. In fact, however, Hanlon et al. (2015) demonstrate, for their small sample of 4 IoR treaties and ending in 2008, that there is a home bias in portfolio investment to the US, where funds typically move back to the same country where they started. While we cannot track specific capital positions, we can provide some arguments supporting the claim that there is also home bias in international tax evasion. The rational reason behind this may be that many tax evaders mainly live in one country where they can enjoy the benefits of their capital. To support this idea we plot, for all available countrypairs that signed a treaty and for which BIS data is available in both directions, the relative changes in outbound and inbound bank deposits around an information exchange treaty. In the case of home bias, i.e. a relation between these deposits, the changes should be lined up close to a 45-degree line in Figure A5, which has outbound changes on the x-axis and respective inbound changes on the y-axis. This expected relation is indeed evident in the data. Since this graphical representation again includes positions that have nothing to do with evasion, the fact that a positive correlation is visible is additionally striking. 71 Figure A5: Deposit drops in reaction to tax haven - non-haven IoR treaties Notes: Calculated based on the baseline results. The graph shows changes in deposits in relation to deposit levels around treaties for tax haven - non-haven countrypairs which signed a treaty and for which BIS data is available in both directions. This limits the representation to countrypairs that include havens for which deposit data has been released (see Table 1). To calculate the outbound drop, we take into account the anticipation effects we find and deduct deposits in havens by non-havens in k+1 from those in k−2 with t being the quarter of signature. For the inbound drop, we deduct deposits in non-havens by haven counterparties in k+6 from those in k+3 as our baseline results suggest. We show both values as a ratio of mean deposits in the same window (k:k+3) starting with the signature date. The solid line is the line of best fit. 72