Mobility Responses to Special Tax Regimes for the Super-Rich: Evidence from Switzerland

Abstract

We use a novel rich-list dataset to estimate the sensitivity of the location choices of super-rich foreigners to a special tax regime, under which wealthy foreigners are taxed based on their living expenses, rather than their true income and wealth. We are the first to show that, when some Swiss cantons repealed this controversial policy, their stock of super-rich foreigners dropped by 43% as a consequence. We find no response for the Swiss super-rich, who were unaffected by the policy change. The implied wealth and income tax mobility elasticities range between 28.4–32.2 and 1.4–1.5, respectively.

Switzerland is well known as a hiding place for large fortunes (Zucman, 2013; Alstadsæter et al., 2019; 2022) and for its mild tax climate, which attracts a considerable fraction of the global wealth elite. In April 2023, the number of billionaires per million inhabitants was four according to Forbes Magazine—more than twice that of the United States. According to estimates by Shorrocks et al. (2022), of the estimated 261,157 adults around the world with at least US$50 million in net worth in 2022, 1.22% were living in Switzerland—a country that is home to only 0.12% of the global adult population.

One important reason why Switzerland is so attractive for wealthy individuals from around the globe is a special tax privilege the small country offers to wealthy foreigners: the so-called expenditure-based taxation. Those eligible for this special tax regime pay taxes on their (and their spouse’s and dependants’) global living expenses, rather than on their true income and wealth. Living expenses are defined broadly and include all expenditures for food, clothing and housing, taxes and social security contributions (around 25,000 CHF per adult and year), alimony payments, remunerations paid to household employees (in cash and in kind), expenses for education and leisure (sports, travel, cultural events, hobbies), for health and wellness cures, and costs of keeping pets (riding horses, etc.), as well as maintenance and operating costs of cars, motorboats, yachts, aeroplanes, etc. Although it can be considered a special form of a consumption tax, the scheme is often referred to as ‘lump-sum taxation’ or ‘tax deals’. Indeed, in the past it was not always clear how strict tax administrations were in practice, whether they took into account, e.g., all global expenditures or rather just expenditures in Switzerland, or updated the base every year. To be eligible, taxpayers must further not have any kind of labour income earned in Switzerland (but they have to declare capital incomes earned in Switzerland), and they must not be Swiss citizens.

Like the banking secrecy, this tax scheme has a long tradition in Switzerland, dating back to the late nineteenth century. And just like the banking secrecy, these tax privileges came under fire internationally, as well as in the country itself, in particular in the aftermath of the global financial crisis of 2008. While other countries were concerned about erosion of their tax base due to such schemes, the domestic policy discussions concerned principles of fair taxation.

As a result, several of the Swiss states (called cantons) voted to abolish the practice, and eventually the conditions were at least tightened at the federal level. In particular, lower bounds introduced by the law should ensure a minimum level of taxation.¹ The scheme nevertheless remains highly attractive, in particular for super-rich foreigners. In 2018, a total of 4,557 taxpayers were still taxed according to their expenditures.²

Proponents of the policy argue that this scheme helps attract super-rich taxpayers from abroad who would otherwise not locate in Switzerland. Yet little is known about how the location choices of the super-rich depend on the controversial tax privileges Switzerland offers to wealthy foreigners. We fill this gap by examining the location choices of the super-rich—the tiny wealthy elite belonging to the top 0.01% of wealth holders in Switzerland—in response to expenditure-based taxation and its removal in some cantons. Because of the lack of access to administrative panel data on expenditure-based taxpayers, the location choice responses to this policy change have never been studied in previous research. We overcome this obstacle by using a newly compiled rich-list dataset (see Baselgia and Martínez, 2024), which contains the small, but highly relevant subset of approximately 150 super-wealthy foreigners who benefit from this practice each year.

Like the United States, Switzerland is a federation where each of the twenty-six federal states, the cantons, enjoy large freedom in tax matters. They set their own rates for the cantonal income and wealth tax, and they offer different deductions and tax reductions. To estimate the role the tax schemes play for location choices, we exploit the fact that between 2010 and 2014, in light of the aforementioned fairness concerns, five cantons repealed expenditure-based taxation from their cantonal tax codes following popular or parliamentary initiatives.³ Since income and wealth taxation is based on the canton and municipality of residence, this tax scheme is not available anymore to wealthy foreigners living in these cantons. Their local and cantonal taxes follow regular taxation.⁴ A national vote to abolish the practice altogether did not pass the ballots in 2014, and since then, no further attempts have been made to repeal expenditure-based taxation within individual cantons. However, requirements were tightened in 2016 in response to the controversy of the scheme (see Online Appendix Table B2 for details; for those already benefiting from the scheme prior to January 2016, the tightened requirements came into force as of tax year 2021).

To quantify the causal effect of the removal of expenditure-based taxation on the location choices of the super-rich, we employ two alternative identification strategies. We first estimate standard difference-in-differences models along with corresponding event studies. In a second approach, we follow the empirical strategy presented in Agrawal and Foremny (2019), which builds on Moretti and Wilson (2017), and that results from spatial equilibrium in a location choice model. Both our empirical approaches show that removing this preferential tax treatment reduces the stock of super-rich foreigners by approximately 43% five years after the abolition. Reassuringly, we find no effect on the location choices of Swiss-born super-rich taxpayers, who are not eligible for this preferential tax scheme and were therefore not affected by the cantonal repeal of the practice.

Using approximate estimates of the implied changes in effective tax rates, we then use our estimates to compute the implied mobility elasticity with respect to taxes. Our back-of-the-envelope calculations suggest that the elasticity of the stock of super-rich taxpayers in a canton with respect to the total net-of-tax rate on wealth lies in the range of 28.4–32.2. With respect to a revenue-equivalent tax on capital income (rather than on wealth), our estimates would imply an elasticity of the stock of super-rich taxpayers of 1.4–1.5.

Our paper contributes to the literature on spatial mobility of taxpayers in response to local tax differentials. Several studies have shown that taxpayers in Switzerland, especially the rich, tend to sort into low-tax cantons and municipalities (e.g., Schmidheiny, 2006; Schmidheiny and Slotwinski, 2018; Brülhart et al., 2022; Martínez, 2022; Köthenbürger et al., 2023). However, we are the first to study the effect of the abolition of expenditure-based taxation in Swiss cantons on the location choices of the super-rich, who belong to the global elite of high net worth individuals.

While other countries try to attract high-income taxpayers with tax privileges tailored to foreigners (e.g., Denmark (see Kleven et al., 2014) or the UK (see Advani et al., 2023)), those schemes differ substantially from the Swiss expenditure-based taxation, because, under the Swiss scheme, taxpayers are not allowed to earn any labour income in Switzerland. Expenditure-based taxation is also not limited to a certain number of years (such as, e.g., in the Danish case), nor is it tailored to a specific group of professionals like football players (Kleven et al., 2013), or star scientists (Akcigit et al., 2016). Rather, the target group are wealthy individuals or families who either pursue their economic activities abroad, or who are rentiers in the classic sense. Because of Switzerland’s pioneering position in the competition for super-rich taxpayers from around the globe, the Swiss context therefore provides a particularly fruitful setting to study the question of how sensitive the location choices of the super-rich global elites are to such preferential tax treatment.

Two recent papers are closely related to ours. Moretti and Wilson (2023) used the Forbes 400 rich list to study how the super-rich in the United States respond to differences in bequest taxation across US states. The number of Forbes 400 individuals fell by 35% in US states that still apply estate taxes compared to those that do not—yet estate taxes are conceptually very different to annual income and wealth taxes, limiting comparability.

Advani et al. (2023) studied international migration responses in the context of the UK ‘non-dom’ system. In contrast to large elasticities documented in earlier papers, they found hardly any out-migration response one year after limitations on the ‘non-dom’ system were imposed. They estimated that less than 5% left the UK in response and reported a semi-elasticity with respect to the flow (not stock, as we do) of 0.26, ruling out magnitudes larger than 0.4. Unfortunately, their estimates are constrained by two limitations: (i) they are limited to the year immediately following the policy change, thereby not allowing for a longer response time, something we find is important; (ii) they only include out-migration, without taking into account changes in inflows or the overall number of affected taxpayers, as we do. We show that super-rich foreigners newly entering our rich-list data are significantly less likely to reside in cantons that abolished the practice than in those that still offer it. This is important for tax policy: not attracting new super-rich taxpayers has dynamic implications for the tax base, even if the (initial) outflow of super-rich is small.

Yet even setting these methodological differences aside, it seems likely that the estimates for the UK case are lower than those for the Swiss case, due to the different rules governing the respective tax privileges. The ‘non-dom’ system is tailored to individuals with investment incomes earned abroad. ‘Non-dom’ taxpayers do not need to be super-rich, they may earn labour incomes in the UK (indeed, many have employment income as their main income source), and they are allowed to possess UK nationality. In addition, the UK reform affected long stayers that had been in the UK for at least fifteen of the last twenty years or those born in the UK to a UK father. These taxpayers are presumably more strongly attached to the UK than the super-rich foreigners we study are to Switzerland. In line with the conclusion by Kleven et al. (2020), these contrasts highlight why elasticity estimates likely differ substantially across countries and institutional settings.

The remainder of this article is organised as follows. Section 1 describes the institutional setting, followed by a description of the data in Section 2. In Section 3, we present our empirical analysis. Section 4 concludes.

1. Institutional Background

1.1. Income and Wealth Taxation in Switzerland

The Swiss tax system is characterised by strong fiscal federalism. Taxation is residence based, such that the canton and municipality of residence have a strong influence on the effective tax burden. Cantons tax income and wealth, setting their own tax rates. The tax base is harmonised across cantons and is very comprehensive: all income and wealth earned or held within or outside of Switzerland enter the tax base.⁵ No distinction between the source of income and wealth is made.

At the federal level, only income is taxed (besides corporate taxes and indirect taxes like VAT). The cantons collect taxes for all three government layers. Married taxpayers have to file jointly and are hence considered as one tax unit.⁶

1.2. Expenditure-Based Taxation for Wealthy Foreigners

Wealthy foreigners without Swiss citizenship who take residence in Switzerland, but do not earn any labour income in Switzerland can opt for a preferential tax treatment known as expenditure-based or lump-sum taxation (sometimes mistakenly referred to as ‘tax deals’). This preferential tax scheme is explicitly aimed at attracting wealthy foreigners to Switzerland. Swiss citizens are not eligible. While expenditure-based taxpayers can earn labour income abroad, they cannot earn any type of labour income from within Switzerland. A French tennis player, for example, could not play the Basel ATP without having to give up the preferential tax treatment, as this would be considered work. A foreign CEO may be eligible if she works for a foreign firm, but not if she works for a firm domiciled in Switzerland. As married couples always file jointly in Switzerland, both spouses have to fulfil the requirements.

The scheme has been in place in different cantons since the late nineteenth century and was introduced at the federal level in 1934. Similar tax regimes exist in the UK (known as the ‘non-dom’ system, dating back to 1799), Belgium, Austria and Italy. Under the ‘non-dom’ system, however, eligible taxpayers are allowed to work in the UK, but claim their permanent domicile to be outside of the UK. Investment income from abroad is only taxed when transferred into the UK. Under the Swiss system, eligible taxpayers claim Switzerland as their main domicile, but are not allowed to earn labour income within Switzerland. Incomes earned abroad can be transferred freely to Switzerland.

As the name suggests, the tax base for these taxpayers is not their true income and wealth, but their total annual living expenses. These are defined comprehensively and include the cost of living for themselves and their dependants (living in Switzerland or abroad), personal expenses, expenses for house personnel and maintenance, as well as other recurring expenses around the world, e.g., for private jets, yachts, holiday homes or large estates and lands abroad. Cantonal and federal tax laws define some minimum values for the cost of living—and, hence, the tax base. While there are written rules and guidelines regarding the estimation of expenses, authorities assess the tax base case by case. The income tax base is then replaced with the estimated expenses. For the wealth tax, a multiple of the expenses, typically a factor of 20, serves as the tax base in most cantons. Online Appendix Table B2 gives an overview of all specifics in each canton.⁷

Importantly, expenditure-based taxpayers differ from regular taxpayers only in terms of the tax base. The standard tax rates defined in the cantonal and federal tax laws are applied. Foreigners have an incentive to opt for this form of taxation if their overall living expenses are lower than their true income and/or the multiple of the expenses is significantly lower than their total global net wealth. Unfortunately, we lack data that would allow us to quantify by how much the true tax bases are undervalued under this scheme.⁸ The scheme can further significantly reduce taxpayers’ costs of tax filing and compliance. In 2018, 4,557 persons—slightly less than 0.1% of all taxpayers—were subject to expenditure-based taxation in Switzerland.

1.3. Abolition of Expenditure-Based Taxation across Cantons

Expenditure-based taxation has become the subject of heavy criticism over the past decade, both from outside and within the country. In light of these discussions, several cantons proposed to abolish this practice, usually holding a popular vote.⁹ Zürich (2010), Schaffhausen (2012), Appenzell Ausserrhoden (2012), Basel Stadt (2014) and Basel-Landschaft (2014) adopted corresponding proposals and removed the option of expenditure-based taxation. Seven other cantons held a popular vote between 2011 and 2014 that did not find a majority (see Online Appendix Figure B1). Online Appendix Table B1 lists dates and results of all the popular votes held. At the national level, a popular vote to abolish expenditure-based taxation was rejected by 59.2% in 2014. After this date, there were no further attempts to repeal the practice at the cantonal level.

2. Data

In light of the highly cherished Swiss tax secrecy, neither the Federal Tax Administration nor any cantonal tax administration is willing to grant access to any (micro) data on expenditure-based taxpayers.¹⁰ To shed light on the effect of expenditure-based taxation, we therefore turn to Swiss rich-list data, compiled by Baselgia and Martínez (2024). This dataset is based on the annual rankings of the 300 richest individuals and families in Switzerland by the business magazine BILANZ (1989–2020), the equivalent of the Forbes 400 list in the United States. It contains the 300 richest individuals and families with net market wealth well above 100 million CHF. We describe this newly compiled panel dataset in much detail in our companion paper (see Baselgia and Martínez, 2024).

For the analysis in this paper, we use the number of super-rich per year and canton for the period 1999–2020, further distinguishing between foreign-born and Swiss-born super-rich—as only rich foreigners are affected by the policy change. In any given year, roughly half of all super-rich in our data are foreign-born. Online Appendix Table B3 provides further descriptive statistics on the characteristics of Swiss-born and foreign-born super-rich. While the rich lists do not contain information on the tax status of super-rich foreigners, we assume based on their large wealth and general eligibility that foreign-born individuals on the list are subject to expenditure-based taxation. Online Appendix Figure 2 shows the geographical distribution of the super-rich across cantons. Importantly, we focus on the super-rich while they are in Switzerland. We do not follow those who leave Switzerland using, e.g., foreign rich lists. When a super-rich person departs Switzerland, they drop from our sample.

3. Responses in Location Choices of Super-Rich Foreigners

3.1. Difference-in-Differences Estimation

To quantify the causal impact of the elimination of expenditure-based taxation on the location decisions of the super-rich, we first apply a difference-in-differences (DD) strategy and estimate event studies, showing how the dynamic effect of abolishing the policy played out over time. First, we estimate various specifications of a two-way fixed-effect (TWFE) DD model of the type

$$\begin{eqnarray} \ln N_{c,t} = \beta ^{DD}\cdot \Gamma _{c,t} + \theta _c + \theta _t + \hat{\Theta }_g^{pre} \cdot t + \Psi X_{c,t} + \epsilon _{c,t}, \end{eqnarray}$$

(1)

where |$\ln N_{c,t}$| is the log number of (foreign-born) super-rich living in canton |$c$| at time |$t$|⁠,¹¹ and |$\Gamma _{c,t}$| is a treatment interaction dummy, indicating the post-reform period in an abolition canton. Treatment is defined in the year prior to the statutory removal (e.g., |$\Gamma _{ZH,2009}=1$| for the canton of Zurich), to allow for responses as early as in the year the abolition was voted on. Our parameter of interest, |$\beta ^{DD}$|⁠, captures the effect of the abolition of expenditure-based taxation on the number of (foreign-born) super-rich in the treatment cantons compared to cantons that did not abolish the preferential tax scheme. Thus, |$\beta ^{DD}$| provides an estimate of the percentage change (⁠|$\approx \exp (\beta ^{DD})-1$|⁠) in the number of foreign-born super-rich related to the removal of the preferential tax treatment.¹² Since the effective tax burden is lower under expenditure-based than under regular taxation (i.e., if |$\Gamma _{c,t }= 0$|⁠), we expect a negative sign for |$\beta ^{DD}$|⁠.

In (1) |$\hat{\Theta }_g^{pre} \cdot t$| controls for different pre-trends between the two groups |$g$| of treatment and control cantons. Specifically, we apply the ‘pre-trend specification’ suggested by Goodman-Bacon (2019): we regress our outcomes on all the fixed effects and control variables and a differential linear trend for treated and non-treated cantons, respectively, using only years before any of the cantons were treated (i.e., up to 2008, as Zurich voted in 2009) as illustrated in (2):

$$\begin{eqnarray} \ln N_{c,t} = \Theta _{\text{pre},g} \cdot t + \theta _c + \theta _t + \Psi X_{c,t} + \epsilon _{c,t} \quad \text{for } t \le 2008 . \end{eqnarray}$$

(2)

Following this procedure, we therefore obtain a common linear trend for the treated cantons and a separate common linear trend for the control cantons. We construct the residuals |$\ln \tilde{N}_{c,t}$| from this regression for the whole sample and use these residuals as the outcome in our main regression specification

$$\begin{eqnarray} \ln \tilde{N}_{c,t} = \beta _{DD} \cdot \Gamma _{c,t} + \theta _c + \theta _t + \Psi X_{c,t} + \nu _{c,t} . \end{eqnarray}$$

(3)

Vector |$ X_{c,t}$| adds the following time-varying canton controls in logarithms: the top average net-of-tax rate on wealth, the top average net-of-tax rate on income, net-of-tax rates on inheritance for children and non-related individuals, respectively, and the share of foreigners in the total population (see Online Appendix A for details). SEs are clustered at the canton level, the level of treatment.

The reform should not have had a direct effect on the location decisions of the Swiss-born super-rich, as they were never eligible for expenditure-based taxation. We therefore conduct the same analysis separately for the affected foreign-born and the unaffected Swiss-born super-rich, in which case we expect |$\beta ^{DD} = 0$|⁠.¹³

The central identifying assumption for this DD analysis is parallel trends in the outcome of interest between the treated and non-treated cantons in the post-treatment period if the treated cantons had not been treated. This assumption cannot be tested directly, but pre-treatment trends serve as a reference. Figure 1(a) shows the trends in the number of super-rich for foreign-born living in treatment (black circles) and control cantons (grey diamonds) on a logarithmic scale. While abolition cantons were home to only relatively few super-rich, their log number was increasing in both treatment and control cantons prior to the repeals starting in 2010. Afterwards, we see a clear trend reversal for the foreign-born super-rich in the treatment cantons, but not in the non-treated cantons. Importantly, most of the treated cantons had not yet implemented the repeal in 2010. As a result, we do not observe an immediate downward jump in Figure 1(a), but rather a gradual decline in the number of foreign-born super-rich, reflecting the staggered nature of the policy implementation in the cantons. Figure 1(b) shows the trends for the Swiss-born super-rich. While abolishing cantons had apparently not been among the favourites for wealthy foreigners, almost 45% of all Swiss-born super-rich lived in those cantons in 1999. But, for the Swiss-born super-rich, we see a declining trend in the decade prior to 2010. The gap between the treatment and control cantons widens continuously over the entire sample period, including the post-treatment period.

Fig. 1.

Parallel Trends: Number of Super-Rich in Treated and Non-Treated Cantons.

Note: Panel (a) shows the number of foreign-born super-rich living in cantons that eventually abolished expenditure-based taxation (black circles) and those that did not (grey diamonds). Likewise, panel (b) shows the number of Swiss-born super-rich living in treatment (black circle) and non-treatment (grey diamonds) cantons. Note that the y axis is on a logarithmic scale. The dashed vertical line in 2010 indicates the year in which the first canton (Zurich (ZH)) abolished expenditure-based taxation; further abolitions took place in 2012 and 2014 (see Online Appendix Table B1 for details).

3.1.1. DD results

Table 1 shows the TWFE DD estimates following (1) to (3) using standard OLS for four different samples. The abolition should only have an effect in the sample of foreign-born super-rich (panel A). Note that because, with the data at hand, we observe only the richest of all expenditure-based taxpayers, most of the variation we capture comes from the repeal in the largest Swiss canton of Zurich (home to about 18% of the Swiss population). This can also be inferred from Online Appendix Figure B2.b. In the sample of Swiss-born super-rich (panel B), we would not expect to see any effects.¹⁴ Given that foreign-born make up almost half of all the super-rich, we also run the estimation on the full sample of the super-rich in the BILANZ (1989–2020) rich list (panel C) to see whether responses by foreign-born super-rich are large enough to be reflected in the full sample. In panel D, finally, we estimate (1) for the period 2003–20 using an even broader population: all taxpayers with a net wealth exceeding 10 million CHF as reported in official wealth tax statistics.¹⁵ This sample should in principle include the majority of all expenditure-based taxpayers, and hence all cantons are included in the analysis. At the same time, this sample also contains many unaffected taxpayers. We therefore take this as an additional test for a potential response by unaffected taxpayers.

Table 1.

Abolition of Expenditure-Based Taxation—Static DD Estimation Results.

Model	(1)	(2)	(3)	(4)	(5)	(6)
Panel A: foreign-born super-rich, 1999–2020
|$\beta ^{DD}$|	−0.31***	−0.58***	−0.57***	−0.57***	−0.55***	−0.56***
	(0.10)	(0.11)	(0.11)	(0.11)	(0.11)	(0.11)
|$\beta ^{CS}$|	−0.37***	−0.51***	−0.47***	−0.49***	−0.82***	−0.57***
	(0.08)	(0.08)	(0.06)	(0.06)	(0.20)	(0.10)
No. of observations	411	411	411	411	411	411
Panel B: Swiss-born super-rich, 1999–2020
|$\beta ^{DD}$|	−0.26***	0.11	0.13	0.11	0.00	0.01
	(0.08)	(0.07)	(0.08)	(0.08)	(0.07)	(0.07)
|$\beta ^{CS}$|	−0.13**	0.07	0.10*	0.11*	−0.04	−0.11
	(0.06)	(0.07)	(0.06)	(0.06)	(0.12)	(0.20)
No. of observations	466	466	466	466	466	466
Panel C: all super-rich, 1999–2020
|$\beta ^{DD}$|	−0.27***	0.02	0.02	0.01	−0.03	−0.02
	(0.08)	(0.08)	(0.09)	(0.09)	(0.08)	(0.08)
|$\beta ^{CS}$|	−0.11	0.04	0.07	0.09	−0.11	0.01
	(0.07)	(0.07)	(0.07)	(0.06)	(0.18)	(0.12)
No. of observations	506	506	506	506	506	506
Panel D: rich taxpayers, 2003–20
|$\beta ^{DD}$|	−0.33***	0.00	0.01	−0.02	−0.04	−0.10
	(0.08)	(0.08)	(0.07)	(0.07)	(0.07)	(0.07)
|$\beta ^{CS}$|	−0.20***	0.01	0.01	0.00	−0.01	−0.03
	(0.04)	(0.04)	(0.03)	(0.03)	(0.05)	(0.06)
No. of observations	468	468	468	468	468	468
Controls	(1)	(2)	(3)	(4)	(5)	(6)
Canton fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Canton-treatment linear trend	No	Yes	Yes	Yes	Yes	Yes
Top average wealth tax rates	No	No	Yes	Yes	Yes	Yes
Top average income tax rates	No	No	No	Yes	Yes	Yes
Inheritance tax rates	No	No	No	No	Yes	Yes
Share of foreigners	No	No	No	No	No	Yes

Note: This table shows the estimation results of the model described in (1) to (3) using OLS. Here, |$\beta ^{DD}$| represents the coefficient of interest from a standard TWFE estimation, while |$\beta ^{CS}$| refers to the (static) estimates of the treatment effect based on the new estimator by Callaway and Sant’Anna (2021). Panel A uses the number of foreign-born super-rich in our BILANZ data set as the dependent variable. More detailed results for this sub-sample, including estimation coefficients on the control variables, are shown in Online Appendix Table B4. Analogously, panel B employs the number of Swiss-born super-rich (detailed results are reported in Online Appendix Table B5). Panel C utilises the full sample of super-rich. Panel D employs an alternative sample, namely, the number of rich taxpayers (i.e., taxpayers with a net wealth greater than 10 million CHF). In columns (2) to (6), we control for pre-trends using de-trended residuals as the outcome variable (see the text for details). SEs are clustered by canton and are shown in parentheses, below the coefficients. |$^\ast\ p\lt .10$|⁠, |$^{\ast \ast}\ p\lt .05$|⁠, |$^{\ast \ast \ast}\ p\lt .01$|⁠.

Column (1) reports the |$B^{DD}$| estimates with only time and canton fixed effects. The estimates suggest that eliminating expenditure-based taxation reduces the number of super-rich by 23%–28% across all sub-samples |$(\beta ^{DD} \in [-0.26, -0.33] )$|⁠. For panels B to D, however, this result is entirely driven by the differential pre-trend between treated and non-treated cantons. Once we control for these trends (column (2)), the effect of the reform vanishes: the coefficient changes sign and/or is very close to zero with large confidence intervals. Only for the foreign-born super-rich (panel A), who were effectively affected by the policy change, the coefficient remains statistically significant after controlling for differential pre-trends, and increasing in magnitude to |$\beta ^{DD}=-0.58$|⁠. Sequentially adding time-varying controls (columns (3)–(6)) only marginally changes the results, which remain significant at the 1% level across all specifications. In our preferred specification including all controls, we find an economically large negative effect of around −43% |$(\beta ^{DD}=-0.56)$|⁠.

Overall, the static DD specifications suggest that super-rich foreigners have been responsive to the abolition of expenditure-based taxation. The removal of expenditure-based taxation reduced the number of foreign-born super-rich by 42%–44% (⁠|$\beta ^{DD} \in [-0.55, -0.58]$|⁠; see footnote 12) in the affected cantons, while it had no effect on the location choices of Swiss-born super-rich—just like one would expect, given the nature of the tax policy. The response of rich, foreign-born taxpayers, however, seems to be too small to be detected in larger samples that include all the super-rich. As we discuss below, these results prove very robust to potential concerns regarding our estimation strategy.

3.1.2. Robustness of DD results

Our results are robust to a series of adjustments and alterations in the empirical estimation. First, note that our baseline results on the foreign-born super-rich are not purely driven by the inclusion of canton-treatment time trends, but hold in all specifications including controls even in the absence of such trends (see Online Appendix Table B6). The estimated coefficient of interest drops to −0.31 in the full model without trend (compared to −0.58 with trend), but is likewise statistically significant at the 1% level.

Second, in our setting, the timing of treatment varies across cantons, and the treatment could be heterogeneous across cantons; hence, the static TWFE estimator may potentially be biased (De Chaisemartin and d’Haultfoeuille, 2020; Goodman-Bacon, 2021; Borusyak et al., 2024; see De Chaisemartin and d’Haultfoeuille, 2023 and Roth et al., 2023 for a summary of the fast-growing literature on TWFE DD estimation). In Table 1, we therefore also report |$\beta ^{CS}$|⁠: the static DD results based on the new estimator by Callaway and Sant’Anna (2021). This is one of several novel alternative estimators that address the issue of ‘negative weighting’. Across all samples A to D and regression specifications (columns (2)–(7)), the canonical TWFE DD estimator and the Callaway and Sant’Anna (2021) DD estimator yield very similar results. Notably, in the full specification for the foreign-born super-rich, the |$\beta ^{CS}$| point estimate at −0.57 is virtually identical. As we show in Online Appendix Figure B5, our dynamic DD results are also robust to other recent estimators, such as those proposed by De Chaisemartin and d’Haultfoeuille (2020) and Borusyak et al. (2024).

Third, our dataset on the super-rich contains true zeros, i.e., there are a few cantons where no (foreign-born) super-rich from our rich-list data reside. As we specify our model in logs, observations with a value of zero in the dependent variable are omitted from the estimation. A potential concern is that this sample selection might affect our estimation results. In Online Appendix Table B7, we demonstrate that this is not an issue in our empirical setting by employing a Poisson pseudo-maximum likelihood (PPML) estimation strategy capable of incorporating zeros. Whether observations with zeros are included in the PPML estimation or not has hardly any effect on the estimates. If anything, restricting the sample to strictly positive values only yields slightly lower estimates in the PPML estimation.¹⁶

Fourth, because we do not observe actual expenditure-based super-rich taxpayers, but rather infer their tax status from the place of birth, we measure our outcome of interest with error. This will render the estimates less precise, increasing SEs.

Fifth, one may be concerned about the violation of the stable unit treatment value assumption (SUTVA) that would bias our estimates upward. As we discuss below, migration to a non-treatment canton is not the main mechanism driving our results. Furthermore, due to the different sizes of the treatment and control groups, the few individuals who do move within Switzerland correspond to only 3% of the size of the pre-treatment control group. Given our large effect sizes, this small within-country migration is unlikely to affect our results. To alleviate any remaining concerns regarding SUTVA violations, in Section 3.3 below we turn to an alternative estimation approach that arises from a spatial equilibrium in a location decision model, which parametrically precludes spillover effects to non-treated cantons. Reassuringly, this alternative estimation strategy yields highly similar results. Hence, we conclude that (potential) SUTVA violations are unlikely to affect our DD analysis in a meaningful way.

3.2. Event-Study Specification

To further assess the validity of the parallel-trend assumption and to study the dynamic effects of the reform, we turn to event studies. Specifically, we estimate (versions of) the following event study model in logs using OLS:

$$\begin{eqnarray} \ln N_{c,t} = \sum \limits _{j=-4}^5 \beta ^{DD}_j \cdot \Gamma _{c,t-j} + \theta _c + \theta _t + \hat{\Theta }_g^{pre} \cdot t + \Psi X_{c,t} + \epsilon _{c,t}. \end{eqnarray}$$

(4)

As before, |$\ln N_{c,t}$| is the log number of super-rich living in canton |$c$| at time |$t$|⁠, and |$\Gamma _{c,t}$| is a treatment indicator for the removal of expenditure-based taxation. Again, |$\theta _c$| and |$\theta _t$| refer to canton and year fixed effects, respectively. Analogous to the static specification in (1), |$\hat{\Theta }_g^{pre} \cdot t$| indicates that we control for the differential pre-trends between treatment and control cantons using residuals as outcomes (see Section 3.1 for details). Vector |$ X_{c,t}$| adds the same time-varying canton controls as in the static DD analysis. The endpoints in all our event-study specifications are binned (see Schmidheiny and Siegloch, 2023). SEs are clustered at the canton level.

We show the dynamic reform effect between abolition and non-abolition cantons, |$\beta ^{DD}_j$|⁠, relative to the second year prior to the abolition of expenditure-based taxation, i.e., normalising |$\beta ^{DD}_{-2}=0$|⁠. This allows for the possibility that some super-rich already chose different locations before the repeal, since it was preceded by political discussions and popular referenda.

3.2.1. Event-study results

Figure 2 presents the estimates of our main event-study specifications for the foreign-born super-rich. The pre-treatment estimates are stable and statistically not significantly different from zero, both with and without controlling for time-varying confounders. Hence, the identifying parallel-trend assumption holds in our setting. The dynamic DD results shown in Figure 2 corroborate our previous finding: the abolition of expenditure-based taxation leads to a significant reduction in the number of foreign-born super-rich, with the effect materialising particularly in the medium run (see the discussion below). Online Appendix Figure B3 presents the analogous results for Swiss-born and all super-rich, as well as rich taxpayers. For all those samples, we estimate a relatively precise dynamic zero effect.

Fig. 2.

Cumulative Event Study—Foreign-Born Super-Rich.

Note: This figure shows the cumulative effects of removing expenditure-based taxation on the log number of foreign-born super-rich in abolition cantons (see (4)). Estimates are relative to the second year prior to the abolition (⁠|$t=-2$|⁠), allowing for anticipation effects/early responses. We control for pre-trends using de-trended residuals as the outcome variable (see the text for details). The black circles correspond to a specification of (4) that contains only year and canton fixed effects, as well as the trend correction. The grey diamonds correspond to a specification that additionally contains the full vector of time-varying cantonal controls |$ X_{c,t}$|⁠. Point estimates are reported with their corresponding 95% confidence intervals, with SEs clustered by canton. Online Appendix Figure B3 displays the analogous results for Swiss-born super-rich, all super-rich and rich taxpayers.

The negative location choice effect of the foreign-born super-rich in response to the abolition of expenditure-based taxation does not materialise immediately after the reform. Online Appendix Table B8 shows that only 21% of the treated individuals in our sample had moved to an untreated canton five years after the repeal, and we find that none of them emigrated.¹⁷ This is in line with the notion that moving takes time and is costly, even (or maybe especially) for the super-rich, who have high standards for their residence beyond tax implications and have high preference, for example, for luxurious housing surrounded by natural amenities like mountain or lake views (see, for example, Young et al., 2016).

In addition to the small outmigration response, new arrivals or entrants in the rich list are also more likely to choose to live in non-treated cantons that still offer a preferential tax treatment to foreigners (see Online Appendix Table B9). Given the marginal distributions, Swiss-born are slightly more likely to live in a treatment canton than could be expected under statistical independence—although the differences are not large enough to be considered statistically significant.

3.2.2. Robustness of event-study results

Again, our results are robust to various alterations in the estimation. First, the estimated effects on the foreign-born super-rich are not conditional on normalising our dynamic DD estimates relative to the second year prior to abolition (i.e., |$\beta ^{DD}_{-2}=0$|⁠). Online Appendix Figure B4 illustrates the analogous results when we apply the standard practice of normalising the estimates relative to the year before the reform (i.e., |$\beta ^{DD}_{-1}=0$|⁠), and highlights that our findings remain largely unchanged.

Second, our event-study results on the foreign-born super-rich are robust to using other modern dynamic DD estimators instead of the canonical TWFE estimator. Online Appendix Figure B5 shows corresponding event-study results based on the DD estimators developed by De Chaisemartin and d’Haultfoeuille (2020) and Borusyak et al. (2024), respectively.

Third, as an additional empirical exercise, we use the fact that only the foreign-born, but not the Swiss-born super-rich were affected by the repeal of the expenditure-based taxation in a triple difference-differences (DDD) estimation strategy. The DDD estimates (reported in Online Appendix Figure B6) are highly consistent with our main estimates for the foreign-born super-rich (see Figure 2). This is not surprising given that we estimate a virtually null effect for the Swiss-born super-rich (see panel (a) of Online Appendix Figure B3).

Fourth, despite using the pre-trend specification suggested by Goodman-Bacon (2019), pre-trends are not estimated very precisely. It would in principle be possible that the large confidence bands mask a diverging trend between treatment and control cantons. We use the honest DD approach suggested by Rambachan and Roth (2023) as a sensitivity analysis. These honest DD confidence intervals account for the fact that there is estimation error in both the treatment effect estimates and our estimates of the pre-trends. The latter have point estimates around zero in our setting, but are surrounded by large confidence intervals. According to this honest DD analysis, our results remain statistically significant under the assumption that the post-treatment violations of parallel trends are no larger than 50% of the maximum pre-treatment violation of parallel trends (see Online Appendix Figure B7).

Lastly, another potential concern might be that abolition cantons lowered top tax rates on income and wealth to compensate for the tax increases faced by formerly expenditure-based taxpayers. In this case, our estimates would be a lower bound of the effect of the abolition. However, we find no effects on top average tax rates on income and wealth (see the event study in Online Appendix Figure B8).

3.3. Spatial Equilibrium Identification Approach

In this section, we turn to a second, alternative estimation approach, arising from spatial equilibrium in a location choice model, first proposed by Moretti and Wilson (2017).¹⁸ Following the adaptation by Agrawal and Foremny (2019), we compare the stock of super-rich across all canton pairs, and estimate how these relationships are affected by the unilateral repeal of expenditure-based taxation by some cantons. The key idea is to compute for each year the log ratio of the stock of the super-rich for each canton pair.¹⁹ These canton-pair ratios then serve as the dependent variable in the model. A special feature of this empirical approach is that, by using canton pairs, the number of observations is significantly larger than in the DD analysis in Section 3.1 and also allows us to include a more restrictive set of fixed effects. We estimate the pairwise model

$$\begin{eqnarray} \ln \bigg (\frac{N_{d,t}}{N_{o,t}}\bigg ) = \beta ^{SR} \cdot \Gamma _{do,t} + \theta _d + \theta _o + \theta _t + \hat{\Theta }^{pre}_{do} \cdot t + \Psi X_{do,t} + \epsilon _{do,t}, \end{eqnarray}$$

(5)

where |$\ln (N_{d,t} / N_{o,t})$| is the log ratio of the super-rich across canton pairs, with |$d$| denoting the destination and |$o$| the origin canton. Note that we do not actually observe flows, but only stocks, so there is effectively no origin or destination. We nevertheless stick to the notation of Agrawal and Foremny (2019). This phrasing is helpful for discussing the empirical setup, as we do not have to refer to some arbitrary reference canton. This notation uniquely captures all canton-pair combinations. Because of how we define the right-hand side variables, it does not matter whether a canton enters the model in the numerator as a destination, or in the denominator as an origin canton. As such, |$\Gamma _{do,t}\equiv \Gamma _{d,t} - \Gamma _{o,t}$| is an indicator variable that equals 1 if destination canton |$d$| does not offer expenditure-based taxation in year |$t$|⁠, but canton |$o$| does. Conversely, |$\Gamma _{do,t}$| equals |$-1$| if destination canton |$d$| still provides expenditure-based taxation, but canton |$o$| does not. And third, |$\Gamma _{do,t}$| equals 0 if either both cantons |$d$| and |$o$| offer expenditure-based taxation in year |$t$| or neither of them does. Hence, our empirical model imposes full symmetry on the estimated effects.²⁰

Our parameter of interest is |$\beta ^{SR}$|⁠. Its interpretation is as follows: removing expenditure-based taxation in canton |$d$| while holding the policy fixed in canton |$o$| makes super-rich people less likely to live in canton |$d$| and more likely to live in canton |$o$|⁠, as in canton |$o$| the effective tax on income and wealth will be lower. The result is a decrease in the stock of super-rich in canton |$d$| relative to canton |$o$|⁠. If the canton of origin |$o$| abolishes expenditure-based taxation, but canton |$d$| does not, the interpretation is vice versa. If either both cantons |$d$| and |$o$| or neither of them abolish expenditure-based taxation, there is no policy change to differentially affect the stock of the super-rich. By defining |$\Gamma _{do,t}=0$| for these canton pairs, these comparisons do not contribute to the estimation of |$\beta ^{SR}$|⁠. By putting more parametric structure on the problem than in the DD analysis, which may still suffer from minor SUTVA violations, we explicitly allow certain spillovers to other cantons: tax policy in canton |$d$| may affect the number of super-rich in canton |$o$|⁠, by increasing the number of super-rich in canton |$o$| and thereby reducing the stock ratio |${N_{d,t}}/{N_{o,t}}$|⁠.

The destination and origin fixed effects, |$\theta _d$| and |$\theta _o$|⁠, capture amenities and other unobserved time-invariant characteristics in the destination and origin cantons. Parameter |$\theta _t$| captures year fixed effects. Here, |$\hat{\Theta }^{pre}_{do}$| controls for different pre-trends between treated and control canton pairs (for details, see also Section 3.1). Vector |$ X_{do,t}$| adds the same control variables as employed in the DD analysis, but specified as log ratios for each of the canton pairs. We follow Agrawal and Foremny (2019) and employ three-way clustered SEs at the canton-pair, origin-year and destination-year levels.

Identification relies on the assumption that, in the absence of policy differences (i.e., no changes in |$\Gamma _{do,t}$|⁠), and given the set of fixed effects, canton-pair-specific time trends accounted for and control variables, the canton-pair stocks of super-rich taxpayers remain constant over time. Any canton-pair-specific unobservable factor correlated with both the elimination of expenditure-based taxation and the location choice of the super-rich between a canton pair may jeopardise our identification strategy. Introducing destination and origin fixed effects separately is a conservative estimation procedure that likely captures much of the unobserved variation across canton pairs. Moreover, the event-study analysis above provided evidence that the location decisions of the foreign-born super-rich do not precede, but follow tax reforms.

3.3.1. Stock-ratio results

Table 2 presents the estimation results of (5) for the years 1999 to 2020. As before, the results are presented separately for foreign-born super-rich (panel A) and other samples (panels B to D). Column (1) shows the estimates with only destination, origin and year fixed effects. Again, we find a negative response of approximately 24%–29% across all sub-samples |$(\beta ^{SR} \in [-0.28, -0.34])$|⁠. As in the DD analysis, however, the results for the Swiss-born super-rich, all super-rich and rich taxpayer samples (panels B–D) are entirely driven by the differential pre-trend between the treated and non-treated cantons. Once we control for such pre-trends, the effect disappears or, in the case of the Swiss-born, reverses sign. This estimate, however, does not remain significantly positive once we add further time-varying controls (see columns (5)–(6)).

Table 2.

Stock-Ratio Estimation across Canton Pairs.

Model	(1)	(2)	(3)	(4)	(5)	(6)
Panel A: foreign-born super-rich, 1999–2020
|$\beta ^{SR}$|	−0.34***	−0.56***	−0.56***	−0.57***	−0.54***	−0.54***
	(0.09)	(0.09)	(0.10)	(0.10)	(0.10)	(0.10)
No. of observations	3,671	3,671	3,671	3,671	3,671	3,671
No. of canton pairs	228	228	228	228	228	228
Panel B: Swiss-born super-rich, 1999–2020
|$\beta ^{SR}$|	−0.29***	0.11***	0.12***	0.10***	−0.02	−0.01
	(0.04)	(0.04)	(0.04)	(0.04)	(0.04)	(0.04)
No. of observations	4,719	4,719	4,719	4,719	4,719	4,719
No. of canton pairs	296	296	296	296	296	296
Panel C: all super-rich, 1999–2020
|$\beta ^{SR}$|	−0.28***	0.03	0.02	0.01	−0.03	−0.01
	(0.04)	(0.04)	(0.04)	(0.04)	(0.04)	(0.04)
No. of observations	5,573	5,573	5,573	5,573	5,573	5,573
No. of canton pairs	299	299	299	299	299	299
Panel D: rich taxpayers, 2003–20
|$\beta ^{SR}$|	−0.33***	−0.03	−0.02	−0.05	−0.06*	−0.13***
	(0.04)	(0.03)	(0.03)	(0.03)	(0.03)	(0.03)
No. of observations	5,850	5,850	5,850	5,850	5,850	5,850
No. of canton pairs	325	325	325	325	325	325
Controls	(1)	(2)	(3)	(4)	(5)	(6)
Destination fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Origin fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Canton-treatment linear trend	No	Yes	Yes	Yes	Yes	Yes
Top average wealth tax rates	No	No	Yes	Yes	Yes	Yes
Top average income tax rates	No	No	No	Yes	Yes	Yes
Inheritance tax rates	No	No	No	No	Yes	Yes
Share of foreigners	No	No	No	No	No	Yes

Note: This table shows the estimation result of the model presented in (5). Panel A uses the number of foreign-born super-rich in our BILANZ data set as the dependent variable. More detailed results for this sub-sample, including estimation coefficients on the control variables, are presented in Online Appendix Table B10. Analogously, panel B employs the number of Swiss-born super-rich. Again, for more detailed results, see Online Appendix Table B11. Panel C utilises the full sample of super-rich. Panel D employs an alternative sample, namely, the number of rich taxpayers (i.e., taxpayers with net wealth greater than 10 million CHF). In columns (2) to (6), we control for pre-trends using de-trended residuals as the outcome variable (see the text for details). SEs allow for three-way clustering (canton-pair, origin-year, destination-year levels) and are shown in parentheses beneath the estimates. |$^\ast\ p\lt .10$|⁠, |$^{\ast \ast \ast }\ p\lt .01$|⁠.

In contrast, the results for the foreign-born super-rich in panel A increase with the inclusion of |$\hat{\Theta }^{pre}_{do}$| (columns (2)–(6)), and remain highly robust to the addition of further controls (columns (3)–(6)). Consistent with our DD analysis, we again find that the abolition of expenditure-based taxation caused a reduction in the stock of the super-rich by 42%–43% compared to the cantons without such a reform |$(\beta ^{SR} \in [-0.54, -0.57])$|⁠. These results are throughout highly statistically significant at the 1% level across specifications in columns (2)–(6) (based on three-way clustered SEs).

3.4. Backing Out the Implied Mobility Elasticity

So far, we have shown that the abolition of expenditure-based taxation has led to a drop in the stock of super-rich foreigners of around 43%. In this section, we perform a back-of-the-envelope calculation to obtain the implied wealth and capital income tax mobility elasticities, respectively.

A limitation of the Swiss setting is that the stipulated tax rates themselves did not change. What differs between regular taxation and expenditure-based taxation is the definition of the tax base. Unfortunately, we do not know the difference between the synthetic expenditure tax base and the true income and wealth tax base of the eligible taxpayers.²¹ What follows is, therefore, an approximation of the effective tax rates under ordinary and expenditure-based taxation for super-rich foreigners.

3.4.1. Wealth tax elasticity

We define the mobility elasticity, |$\eta ^{FB}_{1-\tau }$|⁠, as the percentage change in the number of foreign-born super-rich divided by the relative change in the total net-of-tax rate on wealth resulting from the abolition of expenditure-based taxation. Using our estimate from (1), we can write

$$\begin{eqnarray} \eta ^{FB}_{1-\tau } = \frac{e^{(\beta ^{DD})}-1}{[(1-\tau ^{\text{ord}}) / (1-\tau ^{\text{exp}}) - 1]} . \end{eqnarray}$$

(6)

According to our preferred estimates, the percentage change in the number of foreign-born super-rich is −43% (⁠|$\beta ^{DD} = -0.56$|⁠). Since expenditure-based taxation results in a reduced base instead of an explicitly reduced rate, and because we do not observe actual tax burdens, we rely on a series of simplifying assumptions to come up with estimates of |$\tau ^{\text{ord}}$| and |$\tau ^{\text{exp}}$|⁠.

First, let |$\tau ^{\text{exp}} = T^{\text{exp}} / W$| denote the effective wealth tax rate faced by a super-rich subject to expenditure-based taxation, where |$\tau ^{\text{exp}}$| is just the ratio of the total tax bill under expenditure-based taxation, |$T^{\text{exp}}$|⁠, to actual net wealth |$W$|⁠.²² Similarly, we define |$\tau ^{\text{ord}} = T^{\text{ord}} / W$| as the effective wealth tax rate under ordinary taxation. Here, the total tax bill |$T^{\text{ord}}$| is given by |$ T^{\text{ord}} = W \cdot \tau ^w + W \cdot r^{tax} \cdot \tau ^{inc} ,$| where |$W \cdot \tau ^w$| is the wealth tax burden, and |$r^{tax}$| is the average taxable return on net wealth (i.e., excluding tax-free capital gains) such that |$W \cdot r^{tax} \cdot \tau ^{inc}$| denotes the tax burden on taxable capital income.²³

3.4.2. Ordinary taxation

We use the example of the Canton of Zurich in 2008 (i.e., the year before the vote on the abolition) to provide plausible values of the different tax rates for four model types of super-rich taxpayers from our rich-list data: the poorest (Min.), median (Med.), average (Avg.) and richest (Max.) foreign-born super-rich in 2008. Row 1 of Table 3 shows their net wealth in 2008. We assume that their capital income (excluding tax-free capital gains) amounts to 2% of their net wealth (i.e., |$r^{tax}=0.02$|⁠).²⁴ To determine the total ordinary tax rate on wealth, |$\tau ^{\text{ord}}$|⁠, reported in Table 3, we employ the tax calculator of the Federal Tax Administration (FTA).²⁵

Table 3.

Implied Tax Mobility Elasticities—An Approximation.

	Foreign-born super-rich type in 2008
	Min.	Med.	Avg.	Max.
Net wealth (million CHF)	150	1,100	1,975	35,500
Taxable capital income (⁠|$r^{tax}=2\%$|⁠)	3.0	22.0	39.5	710.0
Ordinary taxation canton Zurich
|$\tau ^{\text{ord}}$|	1.4935%	1.5097%	1.5106%	1.5115%
Expenditure-based taxation
Scenario A: |$T^{{\text{exp}}}_{\text{mean}}$| (million CHF)	0.115531	0.115531	0.115531	0.115531
Scenario A: |$\tau ^{\text{exp}}_{\text{mean}}$|	0.0770%	0.0105%	0.0058%	0.0003%
Scenario B: |$T^{{\text{exp}}}_{\text{P}}$| (million CHF)	0.246995	0.462300	1.095248	23.963874
Scenario B: |$\tau ^{\text{exp}}_{\text{P}}$|	0.1647%	0.0420%	0.0555%	0.0675%
Implied tax mobility elasticities for the stock of the super-rich
Scenario A: mean tax (⁠|$T^{{\text{exp}}}_{\text{mean}}$|⁠)
Change in net-of-tax rate	−1.42%	−1.50%	−1.50%	−1.51%
|$[(1-\tau ^{\text{ord}}) / (1-\tau ^{\text{exp}}_{\text{mean}}) - 1]$|
Wealth tax elasticity |$\eta ^{FB}_{1-\tau }$|	30.25	28.60	28.49	28.37
	(8.66)	(8.19)	(8.16)	(8.12)
Capital income tax elasticity |$\eta ^{\textit{FB}}_{1-\lambda }$|	1.42	1.36	1.36	1.35
	(0.41)	(0.39)	(0.39)	(0.39)
Scenario B: Pareto-based tax (⁠|$T^{{\text{exp}}}_{\text{P}}$|⁠)
Change in net-of-tax rate	−1.33%	−1.47%	−1.46%	−1.44%
|$[(1-\tau ^{\text{ord}}) / (1-\tau ^{\text{exp}}_{\text{P}}) - 1]$|
Wealth tax elasticity |$\eta ^{\textit{FB}}_{1-\tau }$|	32.22	29.20	29.45	29.67
	(9.22)	(8.36)	(8.43)	(8.50)
Capital income tax elasticity |$\eta ^{\textit{FB}}_{1-\lambda }$|	1.48	1.38	1.39	1.39
	(0.42)	(0.39)	(0.40)	(0.40)

Note: This table presents mobility elasticity estimates implied by our estimate of the drop of super-rich foreigners in repeal cantons under two different scenarios and for four types of super-rich from our rich-list data set. Specifically, we consider the poorest (Min.), median (Med.), average (Avg.) and richest (Max.) foreign-born in 2008, as indicated in the respective columns. For each of these four types of super-rich, we consider two different scenarios: scenario A assumes that each of them pays the average expenditure-based tax, |$T^{\text{exp}}_{\text{mean}}$|⁠. In scenario B, their tax payments, |$T^{\text{exp}}_{\text{P}}$|⁠, follow a Pareto distribution (see Online Appendix C.1 for details). Using these tax rates, we compute the implied wealth tax and capital income tax elasticity, respectively, as described in (6)–(8). SEs are shown in parentheses beneath the elasticity estimates.

To obtain possible values of |$T^{\text{exp}}$| allowing us to estimate the effective total tax rate on wealth under expenditure-based taxation, |$\tau ^{\text{exp}} = T^{\text{exp}}/W$|⁠, we next consider two different scenarios.

Expenditure-based taxation—scenario A: mean tax

In scenario A, we use simple averages for |$T^{\text{exp}}$| from a statistic published in the past by the Conference of Cantonal Directors of Finance.²⁶ In 2008, the reported average expenditure-based tax bill, |$T^{\text{exp}}_{\text{mean}}$|⁠, was 115,531 Swiss francs. Given that the effective tax burden can be largely independent of actual income and wealth, in this scenario we assume that each of the four types of super-rich could be paying this average expenditure-based tax. In Table 3, we therefore divide the average expenditure-based tax, |$T^{\text{exp}}_{\text{mean}}$|⁠, by the respective total net wealth of each of our four super-rich sample cases to determine their hypothetical tax rates under expenditure-based taxation, |$\tau ^{\text{exp}}_{\text{mean}}$|⁠. Consistent with the design of expenditure-based taxation, the effective tax rates decrease significantly with wealth. With |$\tau ^{\text{exp}}_{\text{mean}}$|⁠, we are finally able to estimate |$\eta ^{FB}_{1-\tau }$| using (6) for the four types of foreign-born super-rich under scenario A. As shown in the lower half of Table 3, our causal estimate of a 43% drop in the stock of super-rich taxpayers implies a substantial elasticity of the stock of super-rich taxpayers with respect to the total net-of-tax rate on wealth. Across all four types, the estimates of |$\eta ^{FB}_{1-\tau }$| range between 28.37 and 30.25.

Expenditure-based taxation—scenario B: Pareto-based tax

In scenario B, we adopt an alternative approach to estimate the taxes paid by the 150 richest expenditure-based taxpayers. Instead of simply relying on average amounts reported in aggregated statistics, we assume that these tax payments follow a Pareto distribution (the approach is outlined in full detail in Online Appendix C.1). With this approach, we obtain a different value |$T^{\text{exp}}_{\text{P}}$| for each of the four types of super-rich. For instance, for the Min. super-rich, |$T^{\text{exp}}_{\text{P,150}} = 0.246995$| million Swiss francs corresponds to the tax payment of the 150th richest taxpayer in the Pareto distribution, while for the richest taxpayer, |$T^{\text{exp}}_{\text{P,1}}$| equals 23.963874 million Swiss francs. By dividing |$T^{\text{exp}}_{\text{P}}$| by the respective total net wealth, we obtain alternative estimates for the tax rates under expenditure-based taxation, |$\tau ^{\text{exp}}_{\text{P}}$|⁠. As a result, the effective tax rates are considerably more stable across the wealth gradient. Using the tax rates |$\tau ^{\text{exp}}_{\text{P}}$| and (6), we again estimate |$\eta ^{FB}_{1-\tau }$|⁠. In scenario B, our estimates range between 29.20 and 32.22, and are thus highly similar to our estimates reported in scenario A.

3.4.3. Transforming wealth to income tax elasticities

Previous work has estimated mobility elasticities with respect to the net-of-tax rate on (capital) income. To allow for a better comparison with previous estimates, we follow Kopczuk (2019), Brülhart et al. (2022) and Agrawal et al. (2024), and convert our estimate of |$\eta ^{FB}_{1-\tau }$| into an equivalent mobility elasticity with respect to the net-of-income-tax rate |$\eta ^{FB}_{{1-\lambda }}$| using the formula

$$\begin{eqnarray} \eta ^{FB}_{{1-\lambda }} = \eta ^{FB}_{1-\tau } \cdot \frac{d \ln (1-\tau )}{d \ln (1-\lambda )} , \end{eqnarray}$$

(7)

where |$\lambda$| is the revenue-equivalent capital income tax rate to a wealth tax rate |$\tau$|⁠, defined as

$$\begin{eqnarray} \lambda = \frac{1+\rho }{\rho } \cdot \tau \end{eqnarray}$$

(8)

with |$\rho$| denoting the rate of return on wealth (including capital gains). We assume that |$\rho =$| 5%, in which case a 1% wealth tax rate results in a revenue-equivalent capital income tax rate of 21% |$ =({1.05}/{0.05} ) \cdot 1\%$|⁠. Our lower-bound estimate for |$\eta ^{FB}_{1-\tau } = 28.37$| then implies a mobility elasticity with respect to the capital income net-of-tax rate of 1.35. In Table 3, we further report mobility elasticities with respect to income taxes for the other types and under both scenarios. All estimates imply an elasticity with respect to the net-of-income-tax rate of |$\eta ^{FB}_{1-\lambda }$| ranging from 1.35 to 1.48.

3.4.4. Robustness

We address the robustness of our elasticity estimates presented in Table 3 with regard to two further assumptions. First, we address the role of the choice of the ordinary tax rate. In our main specification, we use the prevailing tax rate in the municipality of Zurich as the default under ordinary taxation. In Online Appendix C.2, we relax this assumption and recompute our elasticity estimates for the lowest, median and highest tax municipalities in the canton of Zurich. As illustrated in Online Appendix Table C1, the implied mobility elasticities are largely unaffected by using tax rates from different municipalities.

Second, we assess how sensitive the transformation of our estimates from |$\eta ^{FB}_{1-\tau }$| to |$\eta ^{FB}_{1-\lambda }$| is to the choice of the return rate |$\rho$|⁠, using (7) and (8). As shown in Online Appendix Figure B9, lower values of |$\rho$| result in lower estimates of |$\eta ^{FB}_{1-\lambda }$|⁠, and vice versa. While the estimates for |$\eta ^{FB}_{1-\lambda }$| are somewhat sensitive to the choice of |$\rho$|⁠, they remain within the range documented in the literature for any plausible rates of return.

3.4.5. Comparison with existing elasticity estimates

As Kleven et al. (2020) pointed out, the specific setting influences the observed migration elasticity. Overall, our mobility estimates lie within the upper range of what previous studies have found in a variety of settings (Online Appendix Figure B10 plots a series of existing estimates along with our own). This reflects our focus on super-rich foreigners, who are inherently more mobile, in a setting that exploits within-country variation in tax burdens. Furthermore, the individuals we study are, by law, not allowed to work in Switzerland and are, therefore, less attached to the country than (domestic) workers.

Elasticity estimates are not only context dependent; they often differ in their definition between studies, further complicating a comparison across studies. Our estimates are similar to some of the previous estimates of within-country mobility of top earners (e.g., Moretti and Wilson, 2017; Martínez, 2022), but almost four times larger than what Agrawal et al. (2024) found for responses to the Spanish wealth tax. Furthermore, our estimated elasticities are considerably larger than in Jakobsen et al. (2024), who studied responses of international migration to wealth taxation. Brülhart et al. (2022) also reported smaller estimates with respect to the wealth tax in the Swiss context, but their analysis refers to all Swiss taxpayers, not just highly mobile super-rich foreigners. In addition, the outcome in Brülhart et al. (2022) is taxable wealth, not taxpayers. In particular, for expenditure-based taxpayers, the two elasticities are going to differ: while these super-rich individuals are likely very sensitive to taxation, the wealth tax base observed for them in tax data is heavily underestimated, such that even if individuals respond strongly, the taxable wealth would seem to respond less.

4. Conclusion

The preferential tax treatment Switzerland offers to super-rich foreigners eligible for expenditure-based taxation presumably contributes to the large share of foreigners at the very top of the Swiss wealth distribution. While we cannot quantify the pull effect of this policy at the international level, we provide first-time evidence of how sensitive super-rich foreigners are to this policy when it comes to their choice of where to reside within Switzerland. We exploit the abolition of expenditure-based taxation in some cantons, using two alternative identification strategies. Both approaches suggest that location choices of the super-rich are sensitive to taxation: the abolition of these preferential tax treatments reduces the stock of the super-rich in a canton by approximately 40%–45%.

The novel rich-list data we use limit the statistical power for the analysis of the mechanisms. Yet descriptive evidence suggests that while some super-rich are indeed pushed out and move to other cantons, another part of the effect comes from new arrivals who choose to move to those cantons still offering the preferential tax regime.

Back-of-the-envelope estimations suggest that the implied elasticity of the stock of super-rich taxpayers with respect to the total net-of-tax rate on wealth ranges between 28.4 and 32.2. The more commonly reported equivalent elasticity with respect to the net-of-income-tax rate lies between 1.4 and 1.5. While they seem plausible also in the context of other studies, these elasticities remain approximative. This is due to the lack of access to tax data, which would, not only contain the full universe of expenditure-based taxpayers (that still amounts to over 4,500 after the cantonal repeals), but would give insights into the difference in effective tax rates under regular versus expenditure-based taxation. Therefore, one of the policy conclusions we draw from this paper is that micro-level tax data would promote evidence-based policymaking. This is particularly important in a country where citizens regularly vote on specific policies, including tax policies. We hope that our paper serves as an impetus for better data access to study these topics more thoroughly in the future, including the revenue effects of such tax changes.

Additional Supporting Information may be found in the online version of this article:

Online Appendix

Replication Package

Notes

We thank the editor, Steffen Huck, and four anonymous referees for comments. The paper also benefited from comments from David Agrawal, Konstantinos Angelopoulos, Marius Brülhart, Reto Föllmi, Roland Hodler, Matthieu Leimgruber, Heiner Mikosch, Marko Köthenbürger, Dominik Sachs, Michael Siegenthaler, David Torun and Gabriel Zucman, as well as participants at numerous conferences and seminars. Alexandra Borsigova, Christian Gschwendt and Johannes Waschk provided valuable research assistance. We are grateful to Raphaël Parchet (2019) for sharing his data on wealth and income tax rates. A previous working paper titled ‘Behavioral Responses to Special Tax Regimes for the Super-Rich: Insights from Swiss Rich Lists’ contains the analyses described here, as well as a detailed description of the rich-list data set and extensive descriptive analyses of the super-rich and their wealth (see EU Tax Observatory Working Paper No. 12, February 2023). The latter analysis can now be found in a separate paper titled ‘Using Rich Lists to Study the Super-Rich and Top Wealth Inequality: Insights from Switzerland’ (see Baselgia and Martínez, 2024). Both authors greatly appreciate financial support through the Swiss National Science Foundation, grant no. 176458 ‘The Influence of Taxation on Wealth and Income Inequality’.

Footnotes

1

The Federal Income Tax defines the following lower bounds (whichever is highest is applied): 400,000 CHF, or seven times the annual rental value of the home, or the sum of gross incomes from capital and IP earned in Switzerland and incomes earned abroad for which the taxpayer claims full or partial relief from foreign taxes. Online Appendix Table B2 provides a detailed overview of all requirements across all cantons and when they were implemented.

2

On average, they paid CHF 180,162 (approximately 176,000 US dollars in 2018) in taxes, with the highest tax bill amounting to CHF 11,967,953 (approximately 11,705,000 US dollars in 2018). Source: FDK (Konferenz der kantonalen Finanzdirektorinnen und Finanzdirektoren) https://www.fdk-cdf.ch/themen/steuerpolitik/aufwandbesteuerung. The last available year is 2018, as cantons have discontinued the publication of this statistic.

3

Blankart and Margraf (2011) studied voting behaviour in the proposed repeal of expenditure-based taxation in the canton of Zurich in 2010.

4

For the Federal Income Tax, the scheme is in principle still available, but in practice there are hardly any taxpayers claiming expenditure-based taxation at the federal level if they live in a canton that abolished expenditure-based taxation.

5

With the exception of pension funds, which are tax exempt.

6

Hänni (2021) provided a detailed description of the Swiss tax system, including recent changes and developments. The vast literature on Swiss tax competition at work includes, inter alia, Feld and Kirchgässner (2001), Feld and Reulier (2005), Schmidheiny (2006), Brülhart and Parchet (2014), Luthi and Schmidheiny (2014), Eugster and Parchet (2019), Parchet (2019), Brülhart et al. (2022).

7

In a companion paper (Baselgia and Martínez, 2024), we discuss the implications of this preferential tax treatment for studying inequality in Switzerland using tax data.

8

Anecdotal evidence suggests that the undervaluation can become extremely large in certain cases: when the richest Swiss-based billionaire, IKEA founder Ingvar Kamprad, left the country in 2013, it became public that he was not even among the top fifteen taxpayers in his longtime tax domicile of Epalinges (a village of less than 10,000 inhabitants), because he was taxed according to his expenditures. See https://www.nzz.ch/schweiz/minus-ein-pauschalbesteuerter-1.18106985.

9

Blankart and Margraf (2011) studied voting behaviour in the proposed abolition of expenditure-based taxation in the canton of Zurich in 2010.

10

The following episode illustrates the reluctance to shed light on expenditure-based taxation in practice: in 2012, in the run-up to the vote in 2014 to abolish lump-sum taxation in the canton of Bern, Bernese National Councillor Margret Kiener Nellen requested insight into the tax returns of expenditure-based taxpayers from the cantonal finance department. A five-year legal dispute followed, which was only settled by a Federal Court ruling in 2017, granting her access. In the meantime, however, the Bernese Tax Act has been tightened and access has become even more difficult (Der Bund, 2017, Federal Court Ruling 31.08.2017, BGer 1C_447/2016, 1C_448/2016, 1C_449/2016).

11

Since we are working with only a small sample of taxpayers, in some smaller cantons we do not observe any super-rich for some years (see Online Appendix Figure B2). By estimating (1), these observations are dropped from the model, since the logarithm of zero is undefined. We address this sample selection issue by using a Poisson pseudo-maximum likelihood (PPML) estimation strategy as a robustness exercise; see Online Appendix Table B7 for further details. For a detailed discussion of PPML estimation, see the seminal contribution by Santos Silva and Tenreyro (2006).

12

Note that the |$\beta$|%-change interpretation only holds approximately and for small changes. As we estimate sizeable |$\beta$| coefficients in some specifications, we convert them into percentage changes when describing them in the main text by using the conversion |$\exp (\beta )-1$|⁠. In all tables and figures in this paper, however, we report the estimated |$\beta$| coefficient.

13

We use this additional layer in a triple-DD approach as a robustness exercise in Section 3.2 below.

14

In Section 3.2 below, we present estimation results from a triple-DD model as an alternative identification strategy.

15

Online Appendix A provides details on the data used in panel D and the source of the control variables.

16

It is critical to note that the results from Table 1 (OLS estimation) and Online Appendix Table B7 (PPML estimation) are not directly comparable, with the exception of column (1). This is due to the fact that the PPML estimation does not allow us to control for pre-trends between the treatment and control groups, as suggested by Goodman-Bacon (2019) and implemented in our DD-OLS analysis. In our DD-OLS estimations, regression residuals are used as outcome variables (see (3)), and these by default contain both positive and negative values. However, PPML cannot incorporate negative values. Therefore, the results presented in Online Appendix Table B7 should not be interpreted causally, as they are vulnerable to the pre-trend bias illustrated in Figure 1 (we do not control for canton-treatment trends at all in Online Appendix Table B7). The sole purpose of this robustness exercise presented in Online Appendix Table B7 is to show that sample selection due to the omission of zeros does not affect our estimation results.

17

Nevertheless, foreign-born individuals are twice as likely to move within Switzerland than what could be expected from the marginal distributions of movers and stayers across all treated and untreated Swiss-born and foreign-born observations in our sample. Overall, foreigners are also more likely to exit the sample than expected under statistical independence. Overall, Online Appendix Table B8 is in line with the notion that Swiss-born are likely more attached to Switzerland than foreigners. Swiss-born are also more likely to stay in their canton—especially if they live in a treatment canton. The Pearson |$\chi ^2$| test statistic for the entire contingency table (reported in Online Appendix Table B8) implies that the differences are statistically significant at the 1% level.

18

For the theoretical models, the interested reader is referred to Moretti and Wilson (2017) for a flow model, and Agrawal and Foremny (2019) for a stock version of the same model. We confine the discussion here to our modified empirical model.

19

The same as in our DD-OLS analysis in Table 1, cantons that do not host at least one super-rich are excluded from the estimation, but this does not drive our estimates. To alleviate concerns regarding sample selection due to the omission of zeros in the estimation, we perform two robustness checks. First, in Online Appendix Table B12 we provide the stock-ratio estimation results for a fixed sample, consisting of at least one foreign-born and one Swiss-born super-rich person per canton. Consequently, the estimates in panels A to C of Online Appendix Table B12 are all based on the very same sample. These estimates closely align with our main results in Table 2, suggesting that sample selection is not a primary concern. Second, our PPML robustness exercise, shown in Online Appendix Table B7 in conjunction with the DD-OLS estimation, has already established that omitting zeros hardly has any effect on the estimates in our empirical setting.

20

We start from a situation where |$\Gamma _{do,t} = 0$| for all canton pairs, since all cantons offer expenditure-based taxation. If canton |$d$| abolishes expenditure-based taxation in year |$t$|⁠, but canton |$o$| does not, then |$\Gamma _{do,t} = 1 - 0 = 1$| (or vice versa due to the symmetry imposed by |$\Gamma _{do,t} = 0 - 1 = -1$|⁠) for as long as canton |$d$| does not offer, but canton |$o$| does offer expenditure-based taxation. Variable |$\Gamma _{do,t}$| will switch back to zero only if either canton |$d$| reintroduces expenditure-based taxation (which never occurs in our setting) or if canton |$o$| removes it as well.

21

Despite our many efforts, we have unfortunately been unable to obtain either micro-level administrative data on (former) expenditure-based taxpayers or, alternatively, aggregate statistics on the number of (former) expenditure-based taxpayers who remained in or moved out of a canton, respectively, along with their average income, wealth and tax burden.

22

Expenditure-based taxpayers pay both income and wealth tax, but, for simplicity, we express their total tax burden relative to their overall net wealth as |$\tau ^{\text{exp}}$|⁠.

23

We abstract from labour income given that, for the super-rich, labour income is negligible and expenditure-based taxpayers are not allowed to have labour income earned within Switzerland. Furthermore, at the time Switzerland did not apply separate tax rates to capital and labour income.

24

This return may seem low. However, Baselgia (2024) provided evidence that returns on financial assets (excluding untaxed realised capital gains) are between 1.5% and 2% at the top of the distribution (see Figure B11 in Appendix B.3 of Baselgia, 2024). In our context, a rather low rate of return takes into account the fact that the super-rich typically have legal ways to reduce their income tax burden, in particular by sheltering part of their income in a holding company.

25

The FTA tax calculator is available online: https://swisstaxcalculator.estv.admin.ch/. We specified |$\tau ^w$| and |$\tau ^{inc}$| for a single taxpayer of Protestant faith without children in the city of Zurich in 2010 (the first year for which the tax calculator is available; however, there were no tax changes between 2008 and 2010). Our results are robust across various types of municipalities (see Online Appendix C.2 for details).

26

The evaluation by the Conference of Cantonal Finance Directors is available at https://www.fdk-cdf.ch/-/media/FDK_CDF/Dokumente/Themen/Steuerpolitik/Aufwandbesteuerung/190607_AufwBest_MM_FDK_DEF_F.pdf?rev=2d0ac274e6df497ca691d0ef21a2ed74 (only available in German).