A. Different Frameworks for EOI Over Time

Cross-border deposits can exist for a gamut of reasons, several of which can be orthogonal to tax evasion. Savers may want to invest elsewhere for passive diversification of foreign exchange, exploiting interest rate differentials, cross-border trade settlement, or migration linkages (such as remittance flows) (see for example Fornari and Levy, 2000; Masciandaro, 2004; Karpowicz, 2006; and Lane and Milesi-Ferretti, 2008). Notwithstanding, there is empirical evidence relating non-bank international deposits to tax rates on interest income and standards for domestic bank interest reporting, which is suggestive of tax evasion motives (Huizinga and Nicodeme, 2004).

Income earned on offshore assets is particularly vulnerable to imperfect compliance due to the additional layer of informational asymmetry between a foreign tax authority and an individual’s resident authorities. In this context, EOI arrangements can support the enforcement of residence-based taxation, with positive effects on production efficiency (Keen and Lighart, 2006), by increasing the likelihood of detection of cross-border tax evasion or money laundering.

The traditional model consisted in undertaking EOIR, if and when domestic tax authorities needed information located in another jurisdiction to properly enforce worldwide taxation of its residents.¹⁰ This model suffered from a range of limitations. First, there was no obligation for a jurisdiction to provide the information requested by another jurisdiction if the first-mentioned jurisdiction did not have an interest in the information requested for its own domestic tax purposes. In 2005, the OECD amended its Model Tax Convention by adding a new paragraph 4 to Article 26 to deal with this issue.¹¹ Second, EOIR required that the requesting jurisdiction present a sufficiently detailed request in which it needed to demonstrate that the information being requested is foreseeably relevant to the administration or enforcement of its domestic tax laws.¹² The need to meet this somewhat subjective threshold significantly limited the potential benefits from EOIR as tax evasion schemes are, by their very design, intended to ensure that information is concealed from domestic tax authorities. Third, EOIR has historically relied on bilateral relationships under a double tax treaty (DTT) or a tax information exchange agreement (TIEA) as the legal basis for undertaking the exchange. Access to information thus necessitated the successful conclusion of one of those instruments, which could take years in the making. And, in the case of a DTT, require tax concessions which may outweigh the expected benefits of obtaining access to the information being sought. While the number of bilateral relationships providing for EOIR has been steadily growing over the last few years, it remained challenging to ensure that all possible offshore locations were effectively covered under a given jurisdiction’s DTT/TIEA network (see Figure 1 which differentiates DTTs and TIEAS with and without paragraph 4).

EOIR under DTTs and TIEAs

Citation: IMF Working Papers 2019, 286; 10.5089/9781513519975.001.A001

Data source: Exchange of information portal, OECD.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

EOIR under DTTs and TIEAs

Citation: IMF Working Papers 2019, 286; 10.5089/9781513519975.001.A001

Data source: Exchange of information portal, OECD.

• Download Figure
• Download figure as PowerPoint slide

EOIR under DTTs and TIEAs

Citation: IMF Working Papers 2019, 286; 10.5089/9781513519975.001.A001

Data source: Exchange of information portal, OECD.

• Download Figure
• Download figure as PowerPoint slide

In light of these challenges, new models for undertaking EOI in tax matters emerged.¹³ These models focused on the automatic sharing of taxpayer information, which effectively amounts to an extraterritorial extension of the longstanding practice of using third-party reporting to support taxpayer compliance under self-assessment-based tax systems.

The first significant multilateral move towards a systematic framework for AEOI came from the European Union (EU). In June 2003, EU Member States reached an agreement to have banks and other financial institutions (“paying agents”) disclose information on interest income earned by non-resident individuals to the domestic tax authority, who would then automatically exchange that information with the tax authorities of the accountholder’s jurisdiction of residence.¹⁴ This landmark agreement, colloquially referred to as the EU Savings Directive (EUSD),¹⁵ came into effect on 1 July 2005. In order to help preserve a level playing field, similar arrangements providing for either reporting or withholding were concluded with select non-EU countries, as well as 10 dependent and associated territories.¹⁶ The EUSD was eventually repealed and superseded by parts of Council Directive 2014/107/EU which provides for the implementation of broader AEOI consistent with the “Common Reporting Standard” (discussed below) as of 1 January 2016.¹⁷

Amidst several banking scandals involving international tax evasion schemes—including with LGT Bank in Liechtenstein and, most notably, the Union Bank of Switzerland (UBS) and Crédit Suisse— and growing concerns that the qualified intermediary program that had been implemented in 2001 was being exploited (Byrnes and Munro, 2017), the US followed suit and made a much bolder move towards global application of AEOI. Enacted in 2010, the Hiring Incentives to Restore Employment (HIRE) Act included provisions—commonly referred to as FATCA—targeting non-compliance by US taxpayers (both US residents and citizens) using undeclared foreign accounts to conceal income. On the threat of a 30 percent withholding tax on US-source income, foreign financial institutions (FFIs) were either “encouraged” to directly enter into an FFI Agreement with the US Internal Revenue Service (IRS) to undertake certain due diligence, reporting and withholding requirements under FATCA or required to comply with the FATCA Intergovernmental Agreements (IGA) treated as in effect in their jurisdictions. FATCA requires FFIs to identify and report information about financial accounts held by US taxpayers, or by foreign entities in which US taxpayers hold a substantial ownership interest, such as the name, address, and taxpayer identification number of the account holder, the account number, account balance, and income (including interest and dividends), and sales proceeds. FATCA went into effect on July 1, 2014 and to date 113 jurisdictions have reached agreement with the US to undertake AEOI under FATCA IGAs.¹⁸

Recognizing the benefits of implementing an AEOI framework on a truly global (and multilateral) basis, and mindful of the need to standardize FATCA-style due diligence procedures so as to minimize implementation costs for the financial sector, other countries soon followed suit in moving towards a new AEOI regime. In September 2013 at the St. Petersburg Summit, G20 Leaders committed to AEOI as the “new global standard” and endorsed the OECD’s work to develop the Common Reporting Standard (CRS) which provides for multilateral exchange of financial account information in tax matters. In July 2014, the OECD released the final version of the CRS, which contains the reporting and due diligence procedures that underpin AEOI under the new standard. Much like FATCA from which it largely draws, the CRS has broad scope and covers many financial institutions such as custodial institutions, depository institutions, investment entities and specified insurance companies. It also requires financial institutions to look through passive entities to identify the relevant controlling persons, the “beneficial owners”. Aside from a few mostly technical details, the main difference between the CRS and FATCA resides in the fact that the CRS was designed for multilateral implementation with full reciprocity between participating jurisdictions.¹⁹ It marks an unprecedented international effort to combat international tax evasion; as of November 2018, 101 jurisdictions committed to start AEOI under the CRS framework by 2017 or 2018, and an additional seven countries committed to doing so in 2019 or 2020.²⁰

B. Skepticism Over the Impact of AEOI Frameworks

While AEOI under the EUSD, FATCA and the CRS were all meaningful steps in the fight against international tax evasion, questions have emerged as to whether these frameworks have or would truly be effective in achieving their underlying objectives of detecting and deterring tax evasion.

For example, from the EUSD’s inception, there were persistent concerns that it would be far less successful at preventing tax evasion as first hoped as it could be circumvented in a number of ways. For instance, as the EUSD initially only focused on accountholders who are individuals, it could be avoided through the use of trusts, foundations, or other entities. Other sources of concerns pertained to the possible relocation of assets outside the EUSD-covered jurisdictions to avoid reporting or withholding, or their relocation to other investments not covered under the EUSD’s relatively narrow scope.

There is broad skepticism that FATCA has generated anything close to the amount of revenue initially anticipated. FATCA was projected to generate $8.7 billion in revenue between fiscal years 2010–2020 (Joint Committee on Tax), a yearly average of $792 million. The limited information that is publicly-available suggests that FATCA may not have lived up to its hype. For example, since the launch of the Offshore Voluntary Disclosure Program, a partial amnesty program that allows persons not under audit to disclose unreported offshore activities and benefit from reduced civil and criminal penalties, in 2009, more than 56,000 taxpayers paid a total of $11.1 billion in back taxes, interest and penalties.²¹ However, the bulk of these proceeds are believed to come from anti-money laundering penalties applied against individuals’ foreign assets for failure to file the Report of Foreign Bank and Financial Accounts (FBAR) form, instead of being based on non-compliance with tax obligations (Byrnes and Munro, 2017). Also, while the IRS has spent approximately $380 million to implement and administer the FATCA program, a recent report from the Treasury Inspector General for Tax Administration (TIGTA) concludes that the IRS is still not prepared to enforce compliance with FATCA.²²

In the case of the CRS, participating jurisdictions may choose to allow its financial institutions to apply a threshold such that preexisting entity accounts below US$ 250,000 are not subject to review. Hence, tax evaders using sham entities in a jurisdiction availing themselves of this option may remain undetected by fragmenting their holdings into multiple entity accounts before the entry into force of the enabling domestic legislation. Another technique to conceal beneficial ownership (which rests on a 25 percent threshold, the international standard set by the Financial Action Task Force), would be to dilute interest in a given passive entity between related individuals (e.g., between spouses, children, grandchildren, etc.). For new individual accounts, due diligence procedures contemplate that FIs can rely on customers’ self-certification (and the FI’s confirmation of its reasonableness).²³ A concern arises with the increasing number of countries offering “citizenship/residence by investments” schemes,²⁴ which would seemingly allow these individuals to avoid reporting in their true residence jurisdiction by documenting nexus in the countries where citizenship was purchased (e.g., by providing a passport and a utility bill).²⁵

These claims all appear to raise legitimate concerns about the different challenges that these AEOI frameworks may face in being effective in combatting offshore tax evasion. However, they are largely based on anecdotal evidence and a priori analysis of the underlying due diligence and reporting obligations. To more systematically analyze the impact of the EUSD, FATCA and CRS on tax evasion in offshore jurisdictions, we turn to empirical analysis.

A. Measuring the Impact of EOI

Previous empirical work quantified the impact of EOI by estimating variants of the regression specification:

where, deposit_ijt are foreign-owned deposits in country i held by residents of country j in time t, T_iJt is an indicator variable taking the value one if an EOI agreement between i and j is effective in year t, λ_t is a time-specific fixed effect, μ_ij is a pair-specific fixed effect and ε_ijt is an idiosyncratic error. By including a set of time-specific fixed effects, the coefficient β can be interpreted a DD-based estimate of the treatment effect: it captures the relative change in the deposits of countries where an EOI agreement became effective that exceeds the relative change in the control group where a similar agreement is not in effect²⁶

The BIS locational data provides information on the deposits held in 39 reporting countries (banks) owned by non-residents in over 200 counterparty jurisdictions (savers). By classifying banks and savers into offshore and non-offshore jurisdictions, the data can be seen to consist of four different subsets (Table 1). Previous studies have focused on different subsets to identify the treatment effect of EOI. For instance, JZ2014 use the subset of banks in offshore jurisdictions and savers from non-offshore jurisdictions (subset C in Table 1) to estimate equation (1).²⁷ In this case, the implicit control is the progression of deposits held in offshore jurisdictions by non-offshore savers whose country has not concluded an EOI agreement. In contrast, given the focus on the treatment effect of EOI agreements on round-tripping, Menkhoff and Miethe (2017) use the subset of non-offshore banks and offshore savers (subset B). The implicit control group in this case are the deposits held in non-offshore jurisdictions by savers in offshore countries where there are no EOI agreements in place.

Table 1.

Subsets of Locational Data and Measured Effects

Table 1.

Subsets of Locational Data and Measured Effects

		Saver
		Non-offshore	Offshore
Bank	Non-offshore	A	B
	Offshore	C	D

View Table

Table 1.

Subsets of Locational Data and Measured Effects

		Saver
		Non-offshore	Offshore
Bank	Non-offshore	A	B
	Offshore	C	D

CSS2018 use a slightly different approach and combine subsets A and C to look at deposits held in all banks, including offshore and non-offshore, by non-offshore savers. To identify the treatment effect of the CRS, they include an interaction between T_ijt and an indicator which classifies banks into offshore and non-offshore jurisdictions as an additional regressor in equation (1).

Exploiting information of subsets A and C allows to control for factors that are unrelated to tax evasion but potentially correlated with EOI agreements. For instance, increased legal certainty provided by DTTs might attract additional FDI and thus increase foreign-owned deposits. Treatment effects identified on basis of subset C alone would confound this positive effect from increased legal certainty, in this example, with the potentially negative effect from an increased risk for tax evaders. In contrast, the DD approach based on the wider sample captures the positive effect from legal certainty, which also impacts deposits in subset A, and allows differentiating this effect from the negative effect from an increased risk for tax evaders, which only affects deposits in subset C. Notably, if a given EOI framework leaves deposits in the non-offshore sample unaffected, then the DD approach using the wider sample identifies the same treatment effect which could be obtained by using subset C alone.

We employ a generalized version of the methodology used by CSS2018 by estimating

on the sample of deposits held in all banks (offshore and non-offshore) by savers located in non-offshore jurisdictions (subsets A and C in Table 1).²⁸ The indicator Off_t takes the value of one if the bank is classified as an offshore jurisdiction according to JZ2014 in their online appendix.²⁹ We use four indicator variables for the existence of a specific EOI agreement (EOI4, EUSD, FATCA, and CRS), and the indicator variable AEOI, which takes the value of one if any of the automatic exchange of information frameworks exists. Furthermore, we include two sets of time-specific fixed effects, one set for offshore banks ${\lambda }_t^{Off}$ and one for non-offshore banks ${\lambda }_t^{Non-off}$,³⁰ and a set of pair-specific fixed effects μ_ij. We use robust standard errors (Arellano, 1987) in all estimations that account for clustering at the bank-saver level and allow for arbitrary heteroscedasticity.

By including two sets of time-specific fixed effects, the coefficient β₂ can be interpreted as a triple differenced estimate of the treatment effect. It captures the relative effect of EOI agreements on offshore banks which exceeds the effect of EOI agreements on non-offshore banks. Note that the same estimator could be obtained by estimating equation (1) separately for subsets A and C, and then taking their difference. We expect the coefficient to take a negative value for effective EOI agreements.

In a second set of estimations, we apply the logic of equation (2) to test the effect of EOI agreements on round-tripping. Specifically, using the subset of deposits held by all savers (offshore and non-offshore) in non-offshore banks (subsets A and B), we re-estimate equation (2) where now Off_j takes the value of one if saver j is an offshore jurisdiction. We again expect the coefficient to take a negative value for effective EOI agreements.

B. Identifying Offshore Banks

Effective EOI agreements are only expected to reduce deposits in offshore jurisdictions where income is hidden, but not in other countries. The classification of countries into offshore centers and non-offshore centers is thus critical for measuring EOI’s impact. While an excessively broad definition of offshore centers will result in a downward bias of EOI’s measured impact, too narrow a definition may result in either a downward or upward bias.³¹

Given the lack of a universally accepted list of offshore centers, we employ a data-driven stochastic classification procedure to test the sensitivity of our findings. The approach builds on the idea that EOI reduces foreign-owned deposits in offshore jurisdictions but not (or much less so) in non-offshore jurisdictions. Country-specific estimates on the sensitivity of cross-border deposits thus provide information on a country’s likely offshore status. We implement the idea using a finite mixture model (A.P. Dempster, N.M. Laird, and D.B. Rubin, 1977, and Quandt and J. Ramsey, 1978), where the impact of EOI and the probabilistic classification of countries are estimated jointly in an iterative procedure.³²

To formalize the idea, assume that cross-border deposits take the following functional form:

where, β^cb,cs captures the response to EOI agreements conditional on whether the bank is an offshore center (cb = off) or not (cb = on) and whether the saver is an offshore center (cs = off) or not (cs = on). Similarly, time fixed effects may differ, depending on the classification of banks and savers. As before, we include pair-specific fixed effects β_ij and β_ijt represents an idiosyncratic error.

Next, we denote by p_i the probability that country i is an offshore center and assume this probability is a logistic function of a predictor vector z_it = (z_1it, z_2it,...):

which includes both time-varying and time-constant variables. Following Hines and Dharmapala (2014), we include the WB’s governance indicators, measures of a country’s size (GDP, population), English language use, and UN membership as variables that potentially help explaining the offshore status of countries. In addition, we include a measure of trade-openness (the sum of imports and exports as a share of GDP), FDI stocks (as a share of GDP), tax rates (corporate and personal), a financial secrecy index (which is only available for one year), and dummy variables that indicate whether a country was including in an existing list of offshore centers.

We denote the probability that both bank i and saver j are offshore centers by $q_{i,j}^{off,off}=p_i{p}_j$ and the probability that these countries are both not offshore centers by $q_{i,j}^{on,on}=(1-p_i)(1-p_j)$. Further, we denote by f(log(deposit_ijt) |θ^cb,cs,x_ijt) the density of logarithmic cross-border deposit stocks conditional on the parameter vector θ^cb,cs = (β^cb,cs,λ^cb,cs) and explanatory variables x_ijt. While bilateral deposit stocks follow the same distributional form, a normal distribution, differences in the underlying parameters imply four different distributions of cross-border deposits. Combining the above, the density of bilateral deposit stocks between i and j is given by:

where, $\theta =({\theta }^{off,on},{\theta }^{off,off},{\theta }^{on,off},{\theta }^{on,on})andq=(q_{ij}^{off,off},q_{ij}^{off,off},q_{ij}^{on,off},q_{ij}^{on,on})$.

We obtain the parameter vector θ and the probability vector p = {p_it} by maximum likelihood estimation using an EM-algorithm.

Intuitively, the iterative procedure consists of three steps.

• First, we estimate equation (3) with weighted least squares for each of the four subsets depicted in Table 1, where the weights reflect the prior probability of an observation belonging to any one of the four groups. To test whether the response to AEOI differs, we estimate these four equations in a SUR framework.

• Second, we compute posterior probabilities that a country is an offshore center, based on last round’s parameter estimates. This step identifies outliers of the current classification by contrasting observed reactions with prototypical reactions. We restrict posterior probabilities to be constant across time.

• Third, we use the computed posterior probabilities to estimate equation (4). This step allows refining measured posterior probabilities by explaining the measurements with a vector of control variables that should be related to the likelihood of being an offshore center. We then use these predicted probabilities as a prior for next round’s iteration.

The iterative process ends when posteriors do not change from one round to the next. In one of the specifications, we effectively skip the third step by including a vector of country-specific fixed effects in estimating equation (4). Predicted and computed posteriors are thus equivalent. Annex 1 provides further details.

A. Effect on Deposits in Offshore Jurisdictions

Table 3 summarizes our baseline results. The first four specifications examine the effect of EOI4 agreements, while columns 5 and 6 estimate the effect of all EOI agreements. All estimations include a set of pair-specific fixed effects and time-specific fixed effects. A more flexible time-fixed effects structure that allows for different trends among offshore banks and non-offshore banks which did not conclude an EOI agreement is used from specification 3 onward.

Table 3.

Baseline Results

*, **, and *** indicate significance at the 10, 5, and 1 percent level. Fully robust standard errors in brackets.

Table 3.

Baseline Results

Dependent variable: log deposits, Sample: all banks, non-offshore savers (except Model 1)
Explanatory	Model 1	Model 2	Model 3	Model 4	Model 5
EOI4	0.08 [0.069]	0.329*** [0.071]	0.207*** [0.073]	0.205*** [0.071]	0.195*** [0.072]
EOI4*Offshore		-0.431*** [0.095]	-0.127 [0.100]	-0.129 [0.099]	-0.114 [0.099]
EUSD				0.660*** [0.070]
FAT				0.067 [0.228]
CRS				0.258*** [0.045]
EUSD*Offshore				-0.514*** [0.101]
FAT*Offshore				-0.457 [0.280]
CRS*Offshore				-0.351*** [0.071]
AEOI					0.525*** [0.049]
AEOI*Offshore					-0.535*** [0.075]
One set time FE	Yes	Yes	No	No	No
Two sets time FE	No	No	Yes	Yes	Yes
AdjR	0.052	0.067	0.076	0.086	0.084
Obs	98374	237313	237313	237313	237313

*, **, and *** indicate significance at the 10, 5, and 1 percent level. Fully robust standard errors in brackets.

View Table

Table 3.

Baseline Results

Dependent variable: log deposits, Sample: all banks, non-offshore savers (except Model 1)
Explanatory	Model 1	Model 2	Model 3	Model 4	Model 5
EOI4	0.08 [0.069]	0.329*** [0.071]	0.207*** [0.073]	0.205*** [0.071]	0.195*** [0.072]
EOI4*Offshore		-0.431*** [0.095]	-0.127 [0.100]	-0.129 [0.099]	-0.114 [0.099]
EUSD				0.660*** [0.070]
FAT				0.067 [0.228]
CRS				0.258*** [0.045]
EUSD*Offshore				-0.514*** [0.101]
FAT*Offshore				-0.457 [0.280]
CRS*Offshore				-0.351*** [0.071]
AEOI					0.525*** [0.049]
AEOI*Offshore					-0.535*** [0.075]
One set time FE	Yes	Yes	No	No	No
Two sets time FE	No	No	Yes	Yes	Yes
AdjR	0.052	0.067	0.076	0.086	0.084
Obs	98374	237313	237313	237313	237313

*, **, and *** indicate significance at the 10, 5, and 1 percent level. Fully robust standard errors in brackets.

Overall, the results confirm a strong and statistically significant effect of automatic exchange of information agreements in reducing cross-border deposits into offshore jurisdictions. The measured effect implies that such deposits decreased by between 35 percent (for the CRS) to 50 percent (for the EUSD) following the introduction of agreements to undertake AEOI. In contrast, EOIR reduced deposits in offshore jurisdictions by around 12 percent, although this effect is not statistically significant in most specifications.

The importance of controlling for factors that are unrelated to tax evasion is evidenced by the results in Columns 1 and 2. Following JZ2014, column 1 estimates the effect of EOI4 relationships using only the subset of offshore banks and non-offshore savers (subset C). This specification suggests that deposits in offshore jurisdictions increased by 8 percent following the introduction of an EOIR relationship. This finding is inconsistent with earlier evidence on a negative impact of EOIR (for instance, in the order of 12 percent in JZ2014). Column 2 adds the sample of non-offshore banks and identifies the effect on tax evasion by interacting the treatment variable with the offshore indicator. Using this broader sample, we find that EOIR increased deposits in non-offshore jurisdictions by around 33 percent. Relative to this benchmark, the effect in offshore jurisdictions is much lower (-44 percent) and statistically significant, suggesting that EOIR reduces income concealed in offshore jurisdictions that signed such an agreement. However, this second specification presumes that non-treated banks followed the same time trajectory.

The specification presented in Column 3 allows for more flexibility by introducing a separate set of time fixed effects for offshore and non-offshore banks. Using this specification, the estimated effect of EOIR in addressing tax evasion in offshore jurisdictions remains negative (-12 percent) but loses statistical significance.

Columns 4 and 5 estimate the effect of both EOIR and AEOI. As before, we use a triple differenced estimator to identify the treatment effect of EOI agreements on tax evasion. By expanding the set of explanatory variables, the measured effect of EOI4 remains relatively stable at around -12 percent but is also consistently not statistically significant. In contrast, the effects of AEOI range from -35 percent for the CRS percent to -51 percent for the EUSD, with both results being highly statistically significant. At -45 percent, the measured effect of FATCA is sizeable but not statistically significant. Based on a Wald test, we cannot rule out that the effects of FATCA, CRS, and the EUSD are identical and thus increase estimation precision in column 6 by estimating the effects of these agreements jointly. Using the combined AEOI indicator, the measured effect is also very large (-53 percent) and significant at the 1 percent level.

B. Robustness

Table 4 reports a series of robustness checks, using bank deposits between 2002 and 2018. By restricting the number of years, we lose around 18 percent of the baseline sample observations but potentially increase the accuracy of our estimated effects by reducing the total number of estimated coefficients. All specifications include two sets of time fixed effects as well as pair-specific (bank-saver) fixed effects. Throughout the specifications, we include additional explanatory variables to test the sensitivity of our findings. To improve readability, we do not show the first order terms (non-interacted EOI agreements) in the table even though these are included in the estimations.

Table 4.

Robustness Checks

*, **, and *** indicate significance at the 10, 5, and 1 percent level. Fully robust standard errors in brackets. All specifications include separate time-fixed effects for offshore locations and non-offshore locations. The non-interacted EOI agreements are included in the estimations but not presented in this table for simplicity.

Table 4.

Robustness Checks

Dependent variable: log deposits; Sample: all banks, non-offshore savers
	Model 1	Model 2	Model 3	Model 4	Model 5
EOI4*Offshore	-0.094 [0.101]	-0.082 [0.101]	-0.097 [0.094]	-0.1 [0.094]	-0.097 [0.094]
EUSD*Offshore	-0.452*** [0.097]
FAT*Offshore	-0.291 [0.277]
CRS*Offshore	-0.320*** [0.069]
AEOI*Offshore		-0.461*** [0.066]	-0.254*** [0.061]	-0.254*** [0.061]	-0.252*** [0.061]
Log(Saver Gdp)				-0.001 [0.007]	0.001 [0.007]
Log(Bank Gdp)				0.274*** [0.090]	0.273*** [0.090]
Log(FX rate)				-0.057 * [0.034]	-0.057 * [0.034]
Saver Estimate Voice					-0.237*** [0.046]
Bank Estimate Voice					0.01 [0.076]
Bank-specific time trends	No	No	Yes	Yes	Yes
AdjR	0.031	0.03	0.086	0.087	0.088
Obs	194673	194673	194673	194673	194673

*, **, and *** indicate significance at the 10, 5, and 1 percent level. Fully robust standard errors in brackets. All specifications include separate time-fixed effects for offshore locations and non-offshore locations. The non-interacted EOI agreements are included in the estimations but not presented in this table for simplicity.

View Table

Table 4.

Robustness Checks

Dependent variable: log deposits; Sample: all banks, non-offshore savers
	Model 1	Model 2	Model 3	Model 4	Model 5
EOI4*Offshore	-0.094 [0.101]	-0.082 [0.101]	-0.097 [0.094]	-0.1 [0.094]	-0.097 [0.094]
EUSD*Offshore	-0.452*** [0.097]
FAT*Offshore	-0.291 [0.277]
CRS*Offshore	-0.320*** [0.069]
AEOI*Offshore		-0.461*** [0.066]	-0.254*** [0.061]	-0.254*** [0.061]	-0.252*** [0.061]
Log(Saver Gdp)				-0.001 [0.007]	0.001 [0.007]
Log(Bank Gdp)				0.274*** [0.090]	0.273*** [0.090]
Log(FX rate)				-0.057 * [0.034]	-0.057 * [0.034]
Saver Estimate Voice					-0.237*** [0.046]
Bank Estimate Voice					0.01 [0.076]
Bank-specific time trends	No	No	Yes	Yes	Yes
AdjR	0.031	0.03	0.086	0.087	0.088
Obs	194673	194673	194673	194673	194673

*, **, and *** indicate significance at the 10, 5, and 1 percent level. Fully robust standard errors in brackets. All specifications include separate time-fixed effects for offshore locations and non-offshore locations. The non-interacted EOI agreements are included in the estimations but not presented in this table for simplicity.

Columns 1 and 2 repeat the baseline regressions used in columns 4 and 5 in Table 3 (omitting coefficients on the non-interacted EOI agreements). The estimated effects remain largely unchanged using the restricted sample: EOIR exerts a small and statistically non-significant impact on offshore deposits while all AEOI frameworks exert a strong negative impact (still statistically insignificant for FATCA only). The second column again restricts the effects of AEOI agreements to be uniform, with the measured effect being only slightly smaller than in the baseline estimations (-46 percent).

Columns 3 to 5 allow for bank-specific time trends in addition to the two sets of time-specific fixed effects (our preferred specification). By allowing for more flexibility, the measured effect of AEOI agreements is substantially reduced to 25 percent. The associated standard error implies that, in 95 percent of the observed cases, the bank deposits held in offshore jurisdictions by non-offshore residents declined by between 13 and 37 percent. This measured effect remains unaffected by including GDP in the bank and saver locations, as well as the foreign exchange rate between the bank location and the US dollar, as explanatory variables in column 4, and by adding a governance indicator in column 5.

Overall, our results substantiate earlier evidence on the impact of AEOI but offers a nuanced view on the impact of EOIR. JZ2014 report that EOIR agreements reduce deposits in offshore jurisdictions by an average of 12 percent. While our analysis confirms the magnitude of these effects, the variability of the measured effect increases when using a broader timespan such that we cannot reject the assumption that the effect of EOIR in reducing deposits in offshore jurisdictions may be null. This finding supports the conclusion of Menkhoff and Miethe (2017) that the effects of EOIR may have weakened over time. As for AEOI, Johannesen (2014) shows reactions of Swiss deposits in the order of 30–40 percent in response to the introduction of the EUSD and, most recently, CSS2018 also find a statistically significant impact of the CRS on cross-border bank deposits in offshore jurisdictions (14 percent).³⁹

C. Effect on Countering Round-Tripping

To investigate the effect of EOI relationships in countering round-tripping schemes, that is, indirect investments made by a resident in his home country through an offshore center, we next analyze the sample of deposits held in non-offshore jurisdictions by savers from offshore jurisdictions (subset B). ⁴⁰ This analysis provides an indirect test of EOI’s overall impact, insofar as it builds on the assumption that concealed income from residents in non-offshore jurisdictions in routed through offshore jurisdictions and reinvested in the investor’s residence country.⁴¹

Table 5 presents the results. Overall, we find that the presence of an EOI relationship reduces deposits held in non-offshore locations by residents of offshore centers. This finding, which appears most statistically significant for EOIR, would suggest a reduction in round-tripping behavior. However, we find a reverse effect of FATCA agreements on the cross-border deposits from offshore centers into the US: following the conclusion of FATCA agreements with offshore jurisdictions, deposits in the US increased by roughly the same amount by which other non-offshore jurisdictions witnessed a decline; by 20 percent, possibly due to the non-reciprocal nature of FATCA compared to other types of AEOI arrangements, as explained further below.

Table 5.

Round-Tripping Results

*, **, and *** indicate significance at the 10, 5, and 1 percent level. Fully robust standard errors in brackets. All specifications include separate time-fixed effects for offshore locations and non-offshore locations.

Table 5.

Round-Tripping Results

Dependent variable: log deposits; Sample: non-offshore banks, all savers
	Model 1	Model 2	Model 3	Model 4	Model 5
EOI4*Offshore	-0.206 * [0.121]	-0.202 * [0.121]	-0.261*** [0.117]	-0.265*** [0.117]	-0.277*** [0.115]
EUSD*Offshore	-0.204 [0.150]
FATCA*Offshore	0.22 [0.161]	0.234 [0.156]	0.242 [0.147]	0.247 * [0.147]	0.241 * [0.144]
CRS*Offshore	-0.144 [0.123]
Non_FATCA*Offshore		-0.190 * [0.115]	-0.174 [0.113]	-0.173 [0.114]	-0.16 [0.112]
Save Gdp				-0.007 [0.005]	-0.006 [0.005]
Bank Gdp				0.654*** [0.168]	0.607*** [0.167]
Log(Bank FX)				0.845*** [0.192]	0.974*** [0.195]
Saver Voice					-0.277*** [0.052]
Bank Voice					0.702*** [0.100]
AdjR	0.105	0.103	0.133	0.133	0.135
Obs	177621	177621	177621	177621	176440

*, **, and *** indicate significance at the 10, 5, and 1 percent level. Fully robust standard errors in brackets. All specifications include separate time-fixed effects for offshore locations and non-offshore locations.

View Table

Table 5.

Round-Tripping Results

Dependent variable: log deposits; Sample: non-offshore banks, all savers
	Model 1	Model 2	Model 3	Model 4	Model 5
EOI4*Offshore	-0.206 * [0.121]	-0.202 * [0.121]	-0.261*** [0.117]	-0.265*** [0.117]	-0.277*** [0.115]
EUSD*Offshore	-0.204 [0.150]
FATCA*Offshore	0.22 [0.161]	0.234 [0.156]	0.242 [0.147]	0.247 * [0.147]	0.241 * [0.144]
CRS*Offshore	-0.144 [0.123]
Non_FATCA*Offshore		-0.190 * [0.115]	-0.174 [0.113]	-0.173 [0.114]	-0.16 [0.112]
Save Gdp				-0.007 [0.005]	-0.006 [0.005]
Bank Gdp				0.654*** [0.168]	0.607*** [0.167]
Log(Bank FX)				0.845*** [0.192]	0.974*** [0.195]
Saver Voice					-0.277*** [0.052]
Bank Voice					0.702*** [0.100]
AdjR	0.105	0.103	0.133	0.133	0.135
Obs	177621	177621	177621	177621	176440

*, **, and *** indicate significance at the 10, 5, and 1 percent level. Fully robust standard errors in brackets. All specifications include separate time-fixed effects for offshore locations and non-offshore locations.

Column 1 estimates the treatment effect for each EOI framework separately. In contrast to the direct test of EOI’s effect shown in subsection A, the indirect test does not suggest that AEOI is more effective in reducing tax evasion than EOIR. The coefficients on EOI4, EUSD and CRS indicate that deposits decreased by between -14 and -20 percent. While a Wald test confirms that these point estimates do not differ statistically from each other, only EOIR is significantly different from zero at the 10 percent level. Nonetheless, we note that the reduction in deposits from offshore savers in non-offshore banks under EOIR is consistent with an attempt to mitigate triggering an EOI request by the non-offshore jurisdiction (an incentive that would not be present under AEOI). Surprisingly, however, the effect of FATCA is positive, implying that cross-border deposits in the US increased by 23 percent following the conclusion of FATCA agreements with offshore jurisdictions.

Column 2 estimates effects of the EUSD and CRS jointly by including a combined indicator variable NON_FATCA. The measured effect is now statistically significant at the 10 percent level and suggests a reduction in the order of 19 percent. Columns 3 to 5 add control variables, with little impact on the measured treatment effects: EOIR continues to exert a statistically significant negative impact on the deposits in non-offshore jurisdictions; the effect of the EUSD and CRS (again measured jointly) become slightly lower in magnitude and measured with less precision, which is likely due to less time variation compared to EOI4.

Our measured effect of EOIR agreements on deposits in non-offshore jurisdictions is smaller than the effect of -36 percent reported by Menkhoff and Miethe (2017), which is likely driven by differences in the classification of offshore jurisdictions as we further discuss below. CSS2018 find that the introduction of the CRS induced a reduction in cross-border deposits held in traditional offshore locations. However, the authors show that such wealth has not been repatriated but instead relocated mostly to the US, which they argue has emerged as a potentially attractive location for cross-border evasion. Our results are not inconsistent with this finding, as they point to the US FATCA being the only AEOI agreement which has not led to a reduction in deposits from offshore centers to a non-offshore country (here, the US) following the introduction of such agreements. One rationale for the observed response is that FATCA IGAs concluded by offshore centers provide certainty to evaders located in third countries that investing in the US through these centers avoids the application of FATCA withholding. At the same time, the existence of a FATCA IGA de facto preserves anonymity for the offshore saver, since the US does not reciprocate in terms of identification and reporting of beneficial owners. Finally, the US may also be an attractive jurisdiction in which to hold bank deposits in light of its offering of a zero percent withholding tax rate on related interest payments to nonresident aliens.⁴²

D. Finite mixture model results

Our previous assessment of EOI’s impact in addressing international tax evasion was based on the JZ2014 classification of offshore jurisdictions. We next test the sensitivity of our findings by estimating the probability of being an offshore center and the sensitivity of cross-border deposits to the introduction of EOI agreements simultaneously, using a finite mixture model. While any given jurisdiction’s usefulness to serve as an offshore center likely depends on bilateral-specific characteristics, such as the existence of double tax treaties, we follow existing classification schemes by ruling out this possibility. That is, we restrict a country’s estimated posterior probability of being an offshore center to be constant across all counterparty jurisdictions. In contrast to existing classification schemes, however, we allow in some specifications for the countries’ probabilistic offshore status to change over time.

Table 6 provides the results. The upper part of the table depicts the relative effect of EOI agreements on offshore banks which exceeds the effect of EOI agreements on non-offshore banks (i.e. triple-differenced treatment effects). The lower part of the table presents estimation results on the drivers of the data-driven classification. We allow posterior probabilities to vary annually in the first two specifications (Columns 1 and 2) and restrict them to be constant across time in subsequent specifications.

Table 6.

Finite Mixture Model Results

Notes: Upper table provides estimated treatment effects of EOI4 and AEOI. Lower table provides results of logit regression, explaining offshore jurisdiction status. *, **, and *** indicate significance at the 10, 5, and 1 percent level. Robust standard errors in parentheses.

Table 6.

Finite Mixture Model Results

Time-varying prior	Yes		No
Parameter estimation
Column	(1)	(2)	(3)	(4)	(5)	(6)
E0I4	0.019 [0.089]	0.021 [0.089]	-0.057 [0.104]	0.008 [0.104]	-0.013 [0.083]	0.08 [0.118]
AEOI	-0.254*** [0.063]	-0.249*** [0.063]	-0.249*** [0.081]	-0.163** [0.081]	-0.330*** [0.058]	-0.240*** [0.086]
Probability estimation
log(Gdp)	0.282*** [0.062]	0.272*** [0.058]	0.879*** [0.401]	1.412*** [0.369]
log(Population)	-0.323*** [0.062]	-0.320*** [0.058]	-0.382 [0.308]	-0.873*** [0.305]
Secrecy	0.226 [0.472]		3.641 [2.603]	4.660** [2.291]
Governance	0.016 [0.015]	0.014 [0.015]	0.001 [0.055]
CIT rate (in percent)	-0.014 * [0.008]	-0.014 * [0.008]	-0.146*** [0.059]	-0.253*** [0.062]
English	0.790*** [0.184]	0.765*** [0.179]	3.293*** [1.120]	3.619*** [1.174]
UN member	-0.516*** [0.236]	-0.500*** [0.222]	-4.002*** [1.395]	-4.377*** [1.223]
PIT rate (in percent)	-0.021*** [0.005]	-0.022*** [0.005]	0.001 [0.033]
FDI	0.084*** [0.038]	0.082*** [0.034]	0.629*** [0.237]	4.504*** [1.266]
Trade	0 [0.001]		-0.024*** [0.009]	-0.080*** [0.019]
JZ list					0.826 [0.622]
Hines Rice list					1.712*** [0.755]
Zorome					0.26 [0.694]
Casi Spengel					0.138 [0.793]
Country-specific fixed effect	No	No	No	No	No	Yes
Null Dev	1358.643	1365.123	98.681	105.671	157.21
Resid Dev	1234.054	1239.685	58.767	44.544	121.21
Res DF	2000	2002	111	113	117

Notes: Upper table provides estimated treatment effects of EOI4 and AEOI. Lower table provides results of logit regression, explaining offshore jurisdiction status. *, **, and *** indicate significance at the 10, 5, and 1 percent level. Robust standard errors in parentheses.

View Table

Table 6.

Finite Mixture Model Results

Time-varying prior	Yes		No
Parameter estimation
Column	(1)	(2)	(3)	(4)	(5)	(6)
E0I4	0.019 [0.089]	0.021 [0.089]	-0.057 [0.104]	0.008 [0.104]	-0.013 [0.083]	0.08 [0.118]
AEOI	-0.254*** [0.063]	-0.249*** [0.063]	-0.249*** [0.081]	-0.163** [0.081]	-0.330*** [0.058]	-0.240*** [0.086]
Probability estimation
log(Gdp)	0.282*** [0.062]	0.272*** [0.058]	0.879*** [0.401]	1.412*** [0.369]
log(Population)	-0.323*** [0.062]	-0.320*** [0.058]	-0.382 [0.308]	-0.873*** [0.305]
Secrecy	0.226 [0.472]		3.641 [2.603]	4.660** [2.291]
Governance	0.016 [0.015]	0.014 [0.015]	0.001 [0.055]
CIT rate (in percent)	-0.014 * [0.008]	-0.014 * [0.008]	-0.146*** [0.059]	-0.253*** [0.062]
English	0.790*** [0.184]	0.765*** [0.179]	3.293*** [1.120]	3.619*** [1.174]
UN member	-0.516*** [0.236]	-0.500*** [0.222]	-4.002*** [1.395]	-4.377*** [1.223]
PIT rate (in percent)	-0.021*** [0.005]	-0.022*** [0.005]	0.001 [0.033]
FDI	0.084*** [0.038]	0.082*** [0.034]	0.629*** [0.237]	4.504*** [1.266]
Trade	0 [0.001]		-0.024*** [0.009]	-0.080*** [0.019]
JZ list					0.826 [0.622]
Hines Rice list					1.712*** [0.755]
Zorome					0.26 [0.694]
Casi Spengel					0.138 [0.793]
Country-specific fixed effect	No	No	No	No	No	Yes
Null Dev	1358.643	1365.123	98.681	105.671	157.21
Resid Dev	1234.054	1239.685	58.767	44.544	121.21
Res DF	2000	2002	111	113	117

Notes: Upper table provides estimated treatment effects of EOI4 and AEOI. Lower table provides results of logit regression, explaining offshore jurisdiction status. *, **, and *** indicate significance at the 10, 5, and 1 percent level. Robust standard errors in parentheses.

Overall, the sensitivity analysis confirms the impact of AEOI. We find that cross-border deposits decrease by an excess of between 16 and 33 percent in offshore centers, as identified by the EM algorithm, following the introduction of an AEOI agreement. These effects are statistically significant at the one percent level. In contrast, the measured impact of EOIR on offshore centers does not differ statistically from the measured impact on non-offshore centers. Results presented in the first four columns of Table 6 show that non-sovereign, English-speaking countries, with low tax rates (both personal and corporate), large FDI stocks, small trade flows and high financial secrecy, are more likely to be categorized as an offshore center by the EM algorithm. While we do find that offshore centers do have higher governance indicators, as highlighted by Dharmapala and Hines (2009), we do not find a statistically significant relationship.

The first four columns estimate a country’s offshore status based on country-specific variables. Columns 1 and 2 allow posterior probabilities to vary on a yearly basis. Here, personal income tax rates are a better predictor for a country’s offshore status than corporate income tax rates. For instance, increasing the personal income tax rate by 10 percentage points reduces the odds of being an offshore center by almost 20 percent. Countries that use English as an official language are more likely to be offshore centers, as are non-sovereign states, which are not a UN member. Taken together, the estimations imply that non-sovereign English-speaking countries are more than 3.5 time more likely to be an offshore center than being not an offshore center. Notably, while Dharmapala and Hines (2009) also find a positive association between English language use and offshore center status, they do not find a statistically significant relationship between UN membership and offshore status. The second specification drops the weakest predictors of offshore center status, leaving the estimated effects of the remaining predictors largely unchanged. When restricting posterior probabilities to be constant across time (Columns 3 and 4), estimated effects on the predictor variables are larger in magnitude and tend to be more significant. As before, larger FDI stocks (as a share of GDP) signal offshore center status, as do smaller trade flows. The CIT rate is a better predictor for the offshore status in these specifications. When dropping the worst predictors of the third specification, we find that the secrecy score becomes a significant predictor of offshore center status.

As the choice of predictor variables impacts measured effects, we abstain from using country-specific macro variables in Columns (5) and (6). Specifically, we use the offshore center lists compiled by JZ2014, Hines and Rice (1994), Zoromé (2007), and CCS2018 to predict a country’s offshore status in column (5). While all lists are positively related to the endogenously determined posterior probability of being an offshore center, only the list compiled by Hines and Rice (1996) is a significant predictor. Finally, the estimation presented in column (6) uses a vector of country-specific fixed effects to estimate posterior probabilities of being an offshore center. Since posterior probabilities are, by construction, constant across time, this specification effectively skips the predictor stage and directly uses country-specific posterior probabilities as an input for next round’s iteration. The final specification identifies 20 countries as offshore centers. Even though this classification is much narrower than the classification used in our baseline analysis, measured effects of AEOI are very comparable.