Who Participates in Income Shifting? *

Transcription

1 Who Participates in Income Shifting? * Annette Alstadsæter University of Oslo annette.alstadsater@medisin.uio.no Martin Jacob WHU Otto Beisheim School of Management martin.jacob@whu.edu This version: September 2012 ABSTRACT This paper analyzes the sources of heterogeneity in tax-minimizing strategies across individuals. We summarize insights obtained from the literature into three conditions affecting participation in tax minimization: namely, incentive, access, and awareness. Using a rich Swedish administrative panel data set on corporations, individuals, and active owners of closely held corporations, we analyze the sources of heterogeneity in participation in observable income shifting at the time of the 2006 Swedish tax reform. This reform increased both the incentive and opportunities to shift income from labor to the capital income tax base for the owner managers of corporations. We find that both tax incentives and awareness matter for the decision to access income-shifting opportunities and participation in income shifting itself. We also find that the after-tax income of individuals participating in income shifting increased following the change in the income composition. Keywords: Income shifting, Tax avoidance, Income taxation Dividend taxation JEL Classification: H24, H25, H31, H32, D30 * We are grateful to Marcus Jacob, Jan Södersten, Altin Vejsiu, and seminar participants at WHU Otto Beisheim School of Management and the Workshop on Current Research in Taxation in Muenster for helpful comments and suggestions.

2 1. Introduction Recent empirical studies suggest there is substantial heterogeneity in the uptake of incentives to participate in tax-minimizing strategies across individuals (Jones, 2012; Kleven et al., 2011). Yet the question remains as to why not all individuals participate in tax minimization. In this study, we summarize existing insights obtained from the literature into three conditions required for a taxpayer to participate in tax minimization, namely, incentive, access, and awareness. To start with, the taxpayer must have an incentive to participate in tax minimization. That is, the tax benefit from tax avoidance needs to outweigh its costs. Next, the taxpayer needs access to tax-minimizing strategies. For example, employees generally have less opportunity to shift income across tax bases than entrepreneurs do. Finally, the taxpayer must have an awareness of the tax code, of the incentive to minimize taxes, and of the opportunities available to do so. Taxpayers can potentially conduct tax minimization in myriad ways, including through either legal tax avoidance or illegal tax evasion. The main challenge for empirical studies on participation in tax minimization is the lack of convincing individual income tax data that allows both the control of demographic characteristics combined with establishing the link between individual and corporate information. By definition, tax evasion is difficult to observe, and this sets a particular data challenge for empirical studies. As argued by Slemrod and Yitzhaki (2002), there is often a blurred distinction between legal tax avoidance and illegal tax evasion, because of either ambiguous laws or a lack of focus on particular types of transactions by tax administrations. We thus infer that insights into the sources of heterogeneity involved in participation in legal tax avoidance are also relevant for tax evasion. In the current analysis, we focus on legal tax avoidance. Income shifting is the legal process of relabeling existing income or assets in order to reduce total tax payments. 1 In this paper, we employ a rich administrative panel data set from Sweden to analyze the sources of heterogeneity involved in participation in observable income shifting at the time of the 2006 Swedish tax reform. This reform reduced dividend taxes for some but not all taxpayers. With a progressive tax on labor income, a proportional 1 The main types of income shifting are across time, tax bases, and taxpayers (Stiglitz, 1985). Income shifting across time can involve the deferral of the realization of capital gains (Ivkovic et al., 2005; Jacob, 2011) or the timing of dividend payouts (Chetty and Saez, 2005; Jacob and Jacob, 2012). Income shifting across tax bases involves the reclassification of labor income as capital income through the choice of organizational form and wage dividend payout (De Mooij and Nicodème, 2008; Thoresen and Alstadsæter, 2010). Income shifting across taxpayers includes the intrafamily transfer of income and assets (Stephens and Ward-Batts, 2004). See Alstadsæter and Jacob (2012a) for discussion of the concept of income shifting and a review of the related literature. 1

3 tax on capital income, and a top tax wedge between labor and capital income of 27 percentage points, the Swedish dual-income tax system provides high-income individuals with a strong incentive to reclassify labor income as capital income. However, because of the progressivity of labor income taxes, there is substantial cross-sectional variation in these tax incentives. Moreover, most ordinary wage earners have limited access to strategies for relabeling labor income as capital income. However, active owners in closely held corporations (CHCs) can determine how much to pay themselves as wages and how much as dividends. 2 The 2006 tax reform effectively reduced the tax rate of dividends received by active owners of CHCs, and at the same time, substantially increased the amount of income that could be distributed as low-taxed dividends that is, the dividend allowance. This increased both the incentive and opportunity to substitute dividend income for labor income and establish a CHC to obtain access to this income-shifting strategy (Alstadsæter and Jacob, 2012b). We use this tax incentive for active owners of CHCs to identify participation in income shifting. Our panel data set for contains the tax returns of all corporations in Sweden, along with the income tax returns and demographic information on a representative panel of 6.6% of the Swedish population. We identify all active owners of CHCs using a unique link between corporate tax returns and individual shareholder information from their income tax returns. This enables us to observe income shifting through a CHC and to identify sources of heterogeneity along the dimensions of incentive, access, and awareness. We define an individual as an income shifter if he actively participates in: (i) a shell corporation, (ii) a lowturnover firm, or (iii) a holding corporation. The design of these three types of CHCs is to reduce tax payments by either transferring income, (i) and (ii), or assets, (iii), from the individual to the corporate level. Following the 2006 tax reform, we find that the uptake of holding, shell, and low-turnover corporations increased substantially. For example, we can classify more than a fifth of all newly founded CHCs as holding companies. As a result, more than 25% of all new firms established after 2006 meet our income-shifting criteria. We contribute to the literature on tax avoidance by identifying the sources of heterogeneity in participation in observable income shifting. Our unique data set enables us to identify factors characterizing the individuals participating in income shifting. We find evidence of substantial heterogeneity in participation in income shifting across taxpayers. Overall, our results generally point to the relevance of incentives, access to income shifting, and awareness in income shifting decisions aimed at tax avoidance. Using a two-stage selection model to 2 We classify a corporation as closely held if four or fewer shareholders own at least 50% of the shares. We count multiple family members as a single shareholder. 2

4 control for selection as a CHC owner manager, we find that incentives matter in participation in income shifting. For instance, high-income individuals are more likely to participate in this kind of income shifting. Further, the greater the difference between the individual s marginal tax on labor and capital income, the more likely the individual is to be an active owner of a CHC designed to access income-shifting opportunities, that is, of types (i), (ii), and (iii) above. We confirm these effects in a separate analysis using a regression discontinuity design. Importantly, the economic effect we identify is substantial. Crossing the state tax threshold, i.e. being subject to an additional income tax of 20 percentage points, increases the likelihood of owning a shell corporation by 11.8 percentage points, or 186% of the unconditional mean. In addition, we find that awareness matters. For example, having a higher education, and particularly, a business degree, has a substantial positive effect on the likelihood of participating in income shifting. Indeed, having a business degree increases the likelihood of an individual participating in a holding corporation by 5.6 percentage points, or about 90% of the unconditional mean. We also employ a more exogenous measure of tax awareness in that individuals born in Sweden generally have fewer language difficulties in understanding the tax code. Consistent with this notion, we find that Swedish-born individuals are significantly more likely to participate in income shifting. Finally, we are able to observe the outcome of income shifting. We find that participation in income shifting through a specially designed CHC increases the after-tax income of active owners in a holding or low-turnover CHC. However, these owner managers did not experience higher before-tax income growth than other CHC owners. In fact, we find indications of income substitution in that owner managers in a holding CHC appear to reclassify ordinary dividends as tax-favored dividends from a CHC. Moreover, it appears that individuals use low-turnover CHCs to reclassify labor income as capital income. Therefore, even if before-tax income is unchanged for these owner managers, relabeling of income increases their after-tax income. The resulting heterogeneity in the take-up of tax avoidance and income shifting across individuals reduces both vertical and horizontal equity. That is, high-income individuals are more likely to participate in income shifting (a reduction in vertical equity) whereas taxpayers with access and higher degrees of awareness are more likely to participate in income shifting (a reduction in horizontal equity). The remainder of the paper is organized as follows. In Section 2, we discuss the relevant literature and provide a theoretical background for our analysis. In Section 3, we describe the Swedish tax system, the tax changes of 2006, and the resulting incentives. In Section 4, we 3

5 describe the data and variables and provide some summary statistics. In Section 5, we present the empirical results. Section 6 concludes. 2. Related literature and heterogeneity in participation in tax minimization Recent empirical research has documented heterogeneity in participation in tax minimization. The question is why do not all taxpayers participate in tax evasion or avoidance? First, in order to participate in tax-minimizing strategies, the taxpayer must have an incentive to do so. Second, the taxpayer also needs to have access to tax-minimizing strategies. Finally, the taxpayer requires awareness of the tax code as well as the incentives and income-shifting opportunities. In the present paper, we simultaneously test the effects of incentive, access, and awareness on participation in income shifting across individuals. We now discuss each of these conditions in detail Incentive for tax minimization As discussed by Allingham and Sandmo (1972) in their tax evasion model, taxpayers chooses the share of income they report to the tax authorities in such a way to maximize expected utility. Even though designed for illegal tax evasion, we can also use this model to describe the incentives for participating in legal tax avoidance. The benefit of tax minimization is the expected reduction in tax payments and thus an increase in disposable income. In turn, the costs of tax minimization can be monetary outlays to obtain access to the tax-minimizing strategy, e.g. fees for tax consultants, cost of time spent pursuing taxminimizing strategies, and potential social stigma and other nonpecuniary costs such as tax morale (Andreoni et al., 1998; Fortin et al., 2007). If the tax-minimizing strategy is illegal, the costs also include any expected penalty, which depends on the perceived audit probability and the expected penalty. The individual s preferences then determine the relative valuation of the benefits and the costs. Unfortunately, it is generally difficult, if not impossible, to measure these preferences using archival data. Because of a general lack of suitable data, few studies analyze the effects of incentives on participation in tax minimization and the reasons for the observed heterogeneity across individuals. There is, however, a sizeable and growing literature on income shifting through the choice of organizational form, whereby high-income individuals change the form of their businesses to reduce their total tax burden (Edmark and Gordon, 2012; Gordon and Mackie- Mason, 1994; Romanov, 2006; Thoresen and Alstadsæter, 2010). A substantial literature evaluates the responses of taxable income to tax changes or changing incentives, as surveyed by Saez et al. (2012). Kopczuk (2012) documents the increases in reported income in 4

6 response to an optional new tax rule that increased incentives for many taxpayers to declare income and thus weakening the tax incentives for underreporting income. There is also some evidence on the income shifting of corporations across countries using transfer pricing (Bartelsman and Beetsma, 2003) and debt (Mintz and Smart, 2004) Access to tax minimization Access to tax-minimization strategies is a second important source of heterogeneity in participation in tax minimization. In many cases, some actions are first necessary to participate in tax avoidance, for example, setting up a holding company or becoming selfemployed. See, for example, Kleven et al. (2011), Pirttilä and Selin (2011), and Shaw et al. (2010). An owner manager of his own firm can then determine the tax-minimizing combination of compensation across wage and capital income. In contrast, employed individuals have fewer opportunities for income shifting as they need to negotiate first with his employer, for example, on the compensation of labor effort with a stock compensation plan. This reduces not only the array of income-shifting opportunities but also increases the costs associated with income shifting for employed taxpayers. Access to tax evasion, in particular, underreporting income, then depends on who reports the income to the government and who remits the taxes. See, for example, Kleven et al. (2011) and Kopczuk and Slemrod (2006). Third party reporting limits the access to underreport income for ordinary wage earners as the employer remits taxes on behalf of the employee. This removes the opportunity for ordinary wage earners to not to pay wage income taxes (Shaw et al., 2010). In countries with third party reporting of income and withholding taxes, access to tax evasion is thus restricted to taxpayers that self-report their income, such as self-employed individuals. However, even among those with the opportunity to evade, only about 40% of taxpayers actually do so (Kleven et al., 2011) Awareness of tax minimization A growing literature also emphasizes tax awareness as a source of heterogeneity in participation in tax minimization. The taxpayer s awareness of tax rules and the resulting opportunities to minimize taxes depend on the visibility of taxes, on the accessibility of information about the tax system, and on the individual s ability to process information. If there is only the reporting of income and the employer remits taxes, actual tax payments are less visible to wage earners. This may reduce awareness of how much taxes individuals actually pay and therefore of the incentive to participate in tax-minimizing strategies (Chetty et al., 2009; Finkelstein, 2009). For example, Slemrod et al. (2001) find indications of 5

7 increased tax minimization among high-income individuals following a written audit threat. The audit letter appears to raise the awareness of tax payments and, subsequently, increases participation in tax minimization. Slowness in adoption to changing income levels may also explain why not all individuals respond to tax incentives (Jones, 2012). The ability to process the information available can varies across individuals, as does the flow of information through individual networks. In evidence, Alstadsæter et al. (2012) conclude that the informal spreading of information within the family is important for taxpayer participation in taxavoidance strategies. 3. The Swedish tax system and the incentives for income shifting We next present the structure and income-shifting incentives of the Swedish tax system in detail. In particular, we focus on the taxation of active owners in CHCs Tax rules for labor and capital income and the 2006 tax reform The Swedish dual-income tax system levies a proportional tax on capital income and a progressive tax on labor income. The labor income tax comprises a municipal tax at an average rate of 31.4% and two state-level surtaxes of 20% and 5%. The resulting top marginal tax rate on wage income is 56.4% at the individual level (2008 tax rates). Table 1 provides the thresholds for the surtaxes, their development, and the rates for the period The taxation of capital income, including interest income and capital gains, is at a flat rate of 30%. As of 2006, the taxation of dividend income from unlisted corporations, i.e. widely held corporations (WHCs) is at 25%. The taxation of dividend income of active owners of CHCs is at 20% within an imputed dividend allowance, as calculated under so-called 3:12 rules. In general, the dividend allowance consists of an imputed return on equity (between about 10% and 13%) and a certain percentage of total wages to employees (10% before 2006 and either 25% or 50% after 2006). 3 The taxation of dividends to active owners of CHCs in excess of the imputed dividend allowance is as labor income. The 3:12 rules apply to active owners 4 of CHCs, where the dividend allowance for each CHC is separately calculated and increasing in the level of nominal equity and wage costs. Since 2006, there is an alternative fixed-dividend allowance per corporation (the simplification rule). In this case, the dividend allowance is independent of activity, capital, 3 4 See Chapter 3 and Appendix 1 in Alstadsæter and Jacob (2012a) for a detailed overview of the tax rules and the treatment of dividend income from CHCs at the time of the 2006 tax changes. A shareholder is active if he contributes to profit generation in the firm to a considerable extent. This is not precisely defined in the tax law, and because of an apparent lack of control from the tax authorities, it would appear that in practice the owner can define himself as active and still come under the 3:12 rules should he choose, independent of the level of activity. 6

8 and employment in the firm. 5 The individual active owner s share of the dividend allowance then depends on his share of ownership in the firm, with any unused dividend allowances carried forward with interest to the following year. [INSERT TABLE 1 ABOUT HERE] The 2006 reform of the 3:12 rules introduced changes across three main dimensions. First, the tax rate on dividends from CHCs received by active owners within the dividend allowance fell from 30% to 20%. Second, the rules for calculating the dividend allowance were relaxed, and the dividend allowance substantially increased, particularly for corporations at some levels of labor costs. Finally, there was the introduction of an alternative simplification rule for calculating the dividend allowance. This simplified compliance increased the dividend allowance substantially for corporations with low wage costs. Around 80% of active owners under the 3:12 rules choose the simplification rule to calculate the dividend allowance after In sum, the 2006 tax reform reduced the tax rates applicable to dividends and increased the amount of dividends eligible for a reduced tax rate Incentives for participation in income shifting As of 2006, the top tax differential between labor income and dividend income from CHCs is about 27 percentage points. This provides individuals that are subject to the state tax with strong tax-minimizing incentives by reclassifying labor income as capital income. While most ordinary wage earners have no opportunity to reclassify labor income as capital income, active owners in corporations with concentrated ownership can determine how much to pay themselves as wages and how much as dividends. With the 2006 tax reform, the amount of labor income classifiable as dividend income under the 3:12 rules substantially increased, with Alstadsæter and Jacob (2012b) documenting the increased wage dividend substitution by active owners of CHCs after Hence, active owners of CHC appear to capitalize on at least some of the tax incentives in the Swedish tax code. Table 2 summarizes the resulting total tax rates on different types of income from CHCs (including corporate tax and social security contributions) before and after the reform. [INSERT TABLE 2 ABOUT HERE] 5 During our sample period, the dividend allowance from the simple rule has gradually increased from 64,950 Swedish krona (SEK) in 2006 to SEK 120,000 in

9 However, the results in Alstadsæter and Jacob (2012b) are based on all CHC owners and do not specifically analyze the heterogeneity in tax incentives across individuals. For example, wage earners subject to the state income tax may even have an incentive to establish closely held firms simply to provide the opportunity to shift any additional income from labor income to capital income. We specifically define three types of firms designed for incomeshifting purposes. In our analysis, we are interested in the individual characteristics of individuals of owners of the following income-shifting vehicles: Holding Corporation: A corporation established for the purpose of owning assets and shares in other corporations. The main source of income and turnover arises from financial income (dividends and profit distributions from other corporations) from affiliated companies. With a holding company, individuals can postpone dividend and capital gains taxation until the holding company distributes funds to its owner(s). This is particularly important for minority shareholdings in unlisted corporations where the minority owner has less control over dividend payout policy. Holding corporations also allow taxpayers to reduce the tax rates on dividends from WHCs. Starting in 2004 dividends and capital gains are tax exempt if they are realized from unlisted shareholdings. 6 Shell Corporation: A corporation with no significant transactions or assets whose purpose is to serve as a vehicle to accumulate unused dividend allowances for future utilization. That is, individuals can reclassify future labor income as capital income. As corporations can carry forward unused dividend allowances with interest, a shell corporation operates like an option on the reduced dividend tax rate of 20% in future years. Low-Turnover Corporation: This is similar to a shell corporation, except the turnover requirement is relaxed. This can be, for example, a consulting firm set up as a side business to reclassify marginal labor income as capital income. For the remainder of the paper, we focus on these types of CHCs, particularly, why some individuals and not others establish them. 4. Data and summary statistics 4.1. Data sources For our analysis, we use the Firm Register and Individual Database (FRIDA) provided by Statistics Sweden. This data set is a combination of three main data sources: corporate tax statements, income tax statements, and the so-called K10-Form. The owners of CHCs file the 6 For shares in listed firms, dividends and capital gains are only tax exempt if the company holds 10% or more of the voting rights or if the shares are held for organizational purposes. 8

10 latter for each of their corporations and this enables us to link the corporate tax data and individual tax returns and vice versa. Figure 1 illustrates the linking procedure. [INSERT FIGURE 1 ABOUT HERE] The first data set comprises a full sample of all corporations in Sweden during the period (referred to as the Corporate Tax Data). The corporate data are a panel data set based on corporate tax returns filed using the so-called obligatory INK 2 form filed by all listed and unlisted corporations. 7 The tax returns include information on tax balance sheet items and the profit and loss statement. The raw data contain 442,772 firms and 2,904,123 firm-year observations during the period 2000 to This is a sample of all Swedish firms filing a corporate tax return. We make several sample adjustments. To start with, we drop observations where a corporation filed two or more tax statements in a single year. We further exclude firm-years where information on dividends, share capital, and capital increases/decreases is missing, observations with negative total assets, nominal capital, sales, turnover, or scheduled depreciation, and where there is information missing from the organizational form. Finally, we exclude central banks, funds, foundations, and governmentowned entities. This provides a final sample of 426,949 firms and 2,736,747 firm-year observations. Table 3 provides details on the three different types of income-shifting firms identified in this sample using the firm characteristics. We then link this information to the K10-Form data. The second pillar of our data set is the K10-Form data. Each active owner of a CHC must file this form for each of his CHCs. We identify CHCs using the successful link from the corporate tax to the K10-Form data. For 259,830 firms, we have unique identifiers from the corporate tax data, which we can successfully link to the K10-Form data, thereby identifying some 61% of all Swedish corporations as closely held. In the final step, we link information from the corporate data our income-shifting measures and the K10-Form data the identification of CHC owners to the representative panel of individuals. The third and main data source is a representative panel data set of income tax returns of about 6.6% of the Swedish population. The raw data set contains 6,118,061 observations and 702,077 individuals from the original representative sample. For the representative panel, we have 452,809 successful links to the K10-Form data. That is, about 2.8% of the population is an active owner manager in a CHC. We exclude a number of observations in two steps. First, 7 See, for example, for a corporate income tax declaration for 2010 (retrieved October 20, 2011). 9

11 we exclude observations with missing information on age, gender, or marital status. We also censor observations outside the 0.01 and the percentiles of the income distribution to remove any outliers. Second, we only include observations for individuals who are between 18 and 70 years of age as we wish to focus on those who actively participate in the labor market. This yields our final sample of 2,684,706 observations and 438,294 individuals Variable definitions Dependent variables We specify three specific measures of income-shifting behavior as our dependent variables. Owing to the unique data structure, we are able to observe directly the behavior and purpose of each corporation. We then link the status of the firm, for example, if it is inactive, to the specific individual. The dummy variable Shell Firm Owner takes a value of one if there is no activity in the CHC, otherwise zero. The main purpose of a firm is to accumulate dividend allowances to reclassify future labor income as capital income. The dummy variable Low-Turnover Firm takes a value of one if there is some activity in the CHC (for example, as a side business), otherwise zero. Together, these measures capture the incentive to run a corporation in addition to regular employment. Lastly, the dummy variable Holding Corporation takes a value of one if the CHC of the active owner is a holding company, otherwise zero. Table 3 provides definitions of the three types of firms. [INSERT TABLE 3 ABOUT HERE] As shown in Table 4, descriptive statistics indicate that about 15% of all individuals who actively participate in a CHC also actively participate in a firm regarded more as a shell, lowturnover, or holding corporation. That is, more than a seventh of Swedish owner managers participate to at least some extent in income shifting. As the tax rules became more beneficial for income-shifting purposes in 2006 following the reform and afterwards, we would expect an increase in the uptake of shell or holding companies. Panel A in Figure 2 depicts the percentage of CHCs designed for income-shifting purposes according to our definitions. Panel B in Table 4 focuses on newly registered CHCs. [INSERT FIGURE 2 ABOUT HERE] As expected, the percentage of holding companies and shell corporations increases steadily after 2006 (see Panel B, Table 4). When focusing on newly founded CHCs, the trend is even 10

12 stronger. The percentage of new firms that are either shell corporations or low-turnover firms respectively increases to about 5% and more than 10% following the reform. In 2006, we can categorize more than 20% of newly founded firms as holding companies. In sum, about a third of all newly founded CHCs in 2006 are some kind of income-shifting vehicle. We are also interested in the relation between income and the take-up of income shifting. We sort individuals into income bins of SEK 10,000 and compute the percentage of individuals participating in holding and shell corporations for each. Figure 3 plots the percentage of individuals actively participating in a holding company (the black line) or a shell corporation (the gray line) at different levels of income. We find that the percentage of individuals with a holding or a shell corporation is increasing in income up to a level of about SEK 800,000, slightly decreasing thereafter, and increasing again above income of SEK 1 million. [INSERT FIGURE 3 ABOUT HERE] Incentives for tax minimization Clearly, we expect substantial cross-sectional variation in tax incentives. As long as an individual s income is below the state tax threshold, he is only subject to the municipality tax. In this case, the tax wedge between dividends and labor income is negative and labor income is the least-taxed income source. If a taxpayer is, however, subject to the first level of the state of 20%, the incentives change, and the tax wedge is positive. 8 Income in excess of the state tax threshold is least taxed if it is earned as capital income. We thus operationalize the tax incentive effect using a simple dummy variable, State Tax Threshold t 1, that takes a value equal to one if the individual was subject to the state tax of 20% in the previous year and zero otherwise. As shown in Table 4, about 22% of the individual observations in our sample were subject to state tax in the previous year Access to tax minimization In order to participate in income shifting under the particular tax rules we address, the individuals need to be an active owner of a CHC. However, this induces certain costs. For example, running a company is associated with accounting and reporting requirements that generally increase complexity and the costs of income-shifting mechanisms. We thus expect that there is a self-selection process associated with being an active owner. If we find 8 The second stage of the state tax an increase in the marginal tax burden of 5% does not change the incentive. The thresholds for the first and second stage of the state tax vary over time. See Table AI.2 in Alstadsæter and Jacob (2012a) for an overview of the time-varying characteristics of these thresholds. 11

13 evidence for such self-selection, the distribution of access to income shifting is not equal across individuals. As discussed in Section 5.2, we employ a two-stage estimation approach to test this expectation Awareness of tax minimization Our set of demographic controls includes the age of the taxpayer, gender, marital status, the type and level of education, 9 and whether the individual was born in Sweden. We use three different proxies to indicate a stronger awareness of the tax code and tax avoidance opportunities: namely, having higher education in either business (Business Education) or law (Law Education), and being born in Sweden (Born in Sweden). 10 After controlling for education and income, the country of birth is a suitably exogenous proxy for the knowledge hurdle required for understanding the tax code. We expect native speakers, as proxied by birth of the individual in Sweden, to be better able to process changes in the Swedish tax codes than foreign-born individuals. As shown in Table 4, about 85% of the observations are for individuals born in Sweden. Around 11% have a tertiary-level business education and around 1% have a tertiary level law education Outcomes of tax minimization We are interested in the income composition of owner mangers in CHCs. To address this, we include variables indicating the amount of dividends received from the CHC(s) (Dividend from CHC) and an indicator variable equal to one if the individual received dividends from any CHC (Div_CHC Received?). We further include information on dividends in excess of the dividend allowance (Excess Dividend CHC and Excess_Div_CHC Received?). Finally, we include labor income from the CHC (Labor Income CHC) and a dummy equal to one if the individual received a salary from a CHC (Wage Received CHC?), otherwise zero. Table 4 presents summary statistics for our final sample. Panel A in Table 4 summarizes the income variables for the full sample. Panel B in Table 4 presents statistics on the demographic characteristics of education. Lastly, in Panel C of Table 4, we focus on owner managers in CHCs and their income variables We define tertiary education as a four-year university or college degree. Unfortunately, we only have information on whether the individual was born in Sweden. A better measure to capture awareness would be distance, e.g. geographical or cultural, of the country of birth to Sweden. This biases against our variable as we treat all foreign-born individuals similarly. For example, Norwegians or Danes can more easily understand the Swedish language (and concomitantly, the Swedish tax code) than individuals from more distant countries. Furthermore, our measure would be more accurate if we could control for the year of immigration. For instance, if an individual immigrated to Sweden very early in life, they should face very little problem in reading and understanding the tax rules. Both of these concerns qualify our findings, at least in part. 12

14 As shown, the individuals in our sample have an average income over three years of SEK 236,066 (in 2006, USD 1 = SEK 7.38), of which most is labor income. Business income and capital income are clearly not important for the average taxpayer. For example, only 1.7% of the population generates business income in excess of SEK 150,000. In contrast, 53% of the population receives dividend income. The average taxpayer is 44.5 years of age and there are slightly more men than women in our sample (50.54%). A tertiary business degree is the dominant type of education among the higher degrees. Panel C shows that the average income of an owner manager excluding CHC income is SEK 199,103. Adding labor and dividend income from the CHCs, the overall income of owner managers becomes higher than that of the average individual in the sample. About 61% of owner managers pay themselves wages. Owner managers declare dividends from a CHC in 26% of individual tax returns, but excess dividends are declared in only 3.5%. [INSERT TABLE 4 ABOUT HERE] 5. Empirical evidence on income shifting 5.1. Univariate results for uptake of income shifting To evaluate differences between active owners of CHCs and ordinary wage earners (who typically have less access to income-shifting strategies), we compare CHC owners to the remaining population. This provides a first indication of the characteristics of owner managers with access to income shifting through CHCs. We then compare the owners of shell, low-turnover, and holding corporations. This allows us to discuss the differences between income shifters, owner managers, and employed individuals. To ensure a meaningful comparison across groups, we exclude income shifters from the group of CHC owners. Panel C in Table 5 follows Panel C in Table 4 with the exception that we specify average income over three years, including income from the CHC (Average Income 3y CHC). [INSERT TABLE 5 ABOUT HERE] We are first interested in the differences between owner managers and the population as a whole. As shown by subtracting column (2) from column (1) in Table 5, there are substantial differences between these groups with respect to age, education, income, and income composition. When including income from the CHC, individuals with a CHC have a larger overall income (SEK 398,532 versus SEK 191,652), higher labor, capital, and business income, and are more likely to be subject to the state tax (+26 percentage points). We also 13

15 observe substantial differences in the demographic characteristics of these groups. All other things being equal, active CHC owners are slightly older (+4.4 years), more likely to be married (+20 percentage points), are more likely to hold a university degree (+ 2.4 percentage points), and have an increased likelihood of holding a business degree. For the most part, women are less likely to be business owners, with less than a quarter of CHC owners being female. The simplest way of assessing the characteristics that influence the likelihood of participating in income shifting is to compare individuals participating in income-shifting firms to those who do not participate in any of these firms. Table 5 provides this comparison for CHC owners. Subtracting column (3) from column (2) provides the differences between other CHC owners and CHC owners with a shell corporation. As shown, individuals participating in income shifting have higher incomes on average (SEK +230,590), are more likely to be subject to the state tax, and generate more dividends. In line with the inactivity argument, individuals with a shell corporation receive substantially lower wages from CHCs (SEK 129,754) and generate more income outside their CHC. These individuals are also more likely to hold a business degree. The differences are lower when evaluating lowturnover firm owners, obtained by subtracting (2) from column (4) in Table 5. Finally, we are interested in the difference between holding company and the average CHC owner. With this, we identify even greater differences in overall income. Individuals with holding corporation(s), on average, have a total income in excess of SEK 770,000. About 74% of all holding firm owners are subject to the state tax and they typically receive higher capital income. The majority of this is from CHCs. Once again, we observe that holding firm owners a more likely to have a higher education with a strong focus on business education. All together, the univariate statistics indicate that there are not only differences in individual characteristics between firm owners and the population as a whole, but also differences in income-shifting activity within the group of firm owners. These are our first indications of the sources of heterogeneity in the uptake of income shifting with respect to awareness and incentive Multivariate analysis of income shifting We next turn to a regression analysis taking advantage of the entire cross-sectional variation found in the data. In this, we are most interested in the likelihood of owning a shell, low-turnover, or holding corporation. The differences already found between firm owners and the remaining population indicates that we need to control for the potential selection bias of 14

16 being an active CHC owner. This relates to the dimension of access to income shifting in that only the owners of CHCs can fully exploit the advantages inherent in the 3:12 rules. We thus run the following two-stage regression model., = +, + γ 150, + + +, (1) h, = +, + ζ, + + +,, =1 (2) In the first stage, we estimate if an individual owns a CHC using a probit model (Equation 1). The dependent variable Owner is a dummy variable taking a value of one if the individual actively participates in a CHC and zero otherwise. The individual-level variables, denoted by the vector,, include: (1) average income over the past two years excluding income from CHC, and (2) a dummy variable if the individual was subject to the state tax in the preceding year, zero otherwise. The vector also includes (3) demographic characteristics for being born in Sweden, age, gender, and marital status, (4) dummy variables for tertiary education and a business, law or information technology (IT) degree, and (5) a dummy variable if the taxpayer resides in a city. We also include age squared to control for nonlinear relations with age and being a firm-owner and income shifter, respectively. As an exogenous variable in the criterion function, we include a dummy variable Business150 that takes a value of one if the individual realized income of more than SEK 150,000 in the previous year from self-employment and business activity, otherwise zero: this proxies the incentive of these individuals to incorporate their business. We also include county-fixed effects ( ) and year-fixed effects ( ) to control for differences across time and counties. In the second stage, we specify our three measures of income shifting and tax avoidance activity as dependent variables. These three variables capture different dimensions of income shifting. First, we set our indicator variable Shifter to one if the CHC owner owns (at least) one CHC which has no economic activity. Second, we set our indicator variable Shifter to one if the firm the CHC owner owns (at least) is considered a low-turnover firm. Third, we set Shifter to one if at least one of the individual s CHCs is a holding company. We include the vector,, the inverse Mills ratio to control for sample selection and access to income shifting, county-fixed effects ( ) and year-fixed effects ( ) as explanatory variables. In both models, our statistical inference is based on heteroscedasticity robust standard errors clustered at the individual level. [INSERT TABLE 6 ABOUT HERE] 15

17 Table 6 provides the regression results. The results for the criterion function (Equation 1) show that being born in Sweden, being subject to the state tax in the preceding year, marital status, a university degree, and having substantial business income increases the likelihood of being an active CHC owner. In contrast, the probability of being a CHC owner is lower if the individual is female or if the taxpayer lives in a large city. Age has a positive but decreasing effect on the likelihood of being a firm owner. We are, of course, mainly interested in the individual characteristics that affect income-shifting behavior, as shown in columns (2) to (4). We find that being subject to the state tax positively affects shifting behavior. This is consistent with our expectations. All other things being equal, highly taxed individuals have a greater incentive to shift income into the CHC in that they benefit from the tax wedge between higher-taxed labor income and lower-taxed dividend income. The effect we find is substantial, with the likelihood of being a holding firm owner increasing by over 11 percentage points (over 170% of the unconditional mean) if the individual is subject to the state tax. Likewise, the likelihood of being a shell corporation owner increases by 1.4 percentage points (or 84% of the unconditional mean). We are next interested in the factors relating to the awareness of income shifting. We designate the variables Born in Sweden, Tertiary Education, and the types of education as our main variables of interest. Our results suggest that having a degree in business or economics significantly increases the likelihood of participating in income shifting. However, we cannot interpret our results as indicating a causal link between education and income shifting because our measures capture many dimensions related to awareness. For example, an individual with a business degree may not only have greater knowledge about tax rules through education, but also highly individual preferences for business matters, at least compared with graduates in other disciplines. We also find some evidence that individuals born in Sweden are more likely to participate in income shifting. After controlling for education and income, the country of birth is a good proxy for hurdles in understanding the tax code. We thus cautiously conclude that awareness of the tax rules appears to be one economically significant factor explaining part of the heterogeneity in participation in income shifting across individuals. In addition, we find that the propensity to shift income is greater for older individuals, but the effect decreases with age. We also observe consistent effects for the Female variable. Once again, the effects are economically substantial in that the likelihood of owning a holding company increases by 8.89 percentage points (over 100% of the unconditional mean) if the individual is male. Our general education variable is significant in only two of three 16

18 specifications. Finally, the significance of the inverse Mills ratio indicates the importance of controlling for sample selection and the access to income-shifting opportunities Regression discontinuity evidence for the tax incentive We next turn to a more detailed analysis of the tax incentive as there are several concerns about our tax incentive measure. First, our definition could capture an income effect (high versus low income) instead of a tax effect. Even if we include the marginal tax rate as an explanatory variable, concerns about the correlation between income and the marginal tax rate remain. Second, the regression analysis using Equation 2 captures a correlation. However, we are particularly interested in the causal link between taxes and income shifting. The marginal tax rate structure of the Swedish tax code provides many opportunities to test the take-up of income-shifting activities. Following the approach in Alstadsæter, Kopczuk, and Telle (2012), we use the kink of 20 percentage points in the marginal tax rate function around the first state tax threshold and employ a regression discontinuity design. If taxes drive the uptake of income shifting, then individuals immediately above the state tax threshold should be more likely to participate in income shifting than individuals below the threshold. We believe that a tax rate increase of 20 percentage points is sufficiently large to identify this change in behavior. We thus sort individuals according to their earned income of the preceding year (labor income and income from self-employment) relative to the state tax threshold in year t. Figure 4 plots the percentage of individuals with a shell corporation in SEK 500 (about USD 70) income bins around the state tax threshold. We use the postreform period (Panel A) and the full sample (Panel B) of CHC owners. [INSERT FIGURE 4 ABOUT HERE] Figure 4 shows that the likelihood of actively participating in a shell corporation increases to an even higher level if the individual s earned income is above the state tax threshold. The difference is statistically significant (p-value < 0.02). It also appears that there is a discontinuity in the increase around the state tax threshold. We consistently find this for both the postreform period and the full sample. In Table 7, we formalize the graphical evidence in Figure 4 and present the regression results from a linear probability model with our income-shifting measures as the dependent variable and the State Tax Threshold dummy as the explanatory variable of interest. The regression discontinuity design can, however, only be applied if the individual characteristics around the state tax threshold are similar. Table A.3 in the Appendix provides information on 17

19 the differences between individuals above and below the threshold. The data around the threshold appears to be reasonably smooth, and we are therefore confident we are able to identify a tax effect. However, there are some differences with respect to age, gender, and marital status. We thus test the robustness of our results by controlling for demographic characteristics, selection, and county as well as year effects. 11 We present the results for different income ranges around the state tax threshold for all three income-shifting measures. [INSERT TABLE 7 ABOUT HERE] Our results show that crossing the state tax threshold increases the likelihood of participating in income shifting. We find statistically significant results for two of our three variables in the specifications without controls. When including the control variables, we consistently find a statistically and economically significant effect of crossing the state tax threshold on the likelihood of being an income shifter for all of our three measures of income shifting. The effects we estimate are once again economically significant. For example, crossing the state tax threshold (using the SEK 10,000 range around the threshold) increases the likelihood of owning a holding company by 11.8 percentage points or 186% of the unconditional mean. The effects are even stronger for shell corporations and holding companies. These results are consistent with the prediction that tax incentives drive the participation of individuals in income shifting and tax avoidance. From this, we conclude that tax incentives are one key source of the heterogeneity in income shifting across individuals Income substitution and income shifting We next assess the validity of our income-shifting measures. If our measures of income shifting are meaningful, we should be able to observe changes in the income composition of individuals participating in income shifting. In particular, we are interested in the income composition of owner managers in holding companies and low-turnover corporations. 12 First, we focus on holding corporations. In the presence of income shifting, individuals transfer shares in listed and unlisted corporations to their CHC. In doing so, the owner managers of a holding corporation can postpone dividend and capital gains taxation until they decide to distribute these proceeds to the individual level. Second, individuals can reduce the dividend We also employ a matching procedure to obtain samples of individuals that are similar across the variables of gender, marital status, age, and city, but that differ in being subject to the state tax. This addresses concerns that the other covariates drive our results. We replicate Table 5 using a nearest-neighbor propensity scorematching procedure. Our results using the matched sample are similar to the results presented in Table 5. A focus on shell corporation owners is not meaningful as income substitution can be observed at the time of income realization at the corporate level. Our definition, however, rules out income realization in our sample. 18