The elasticity of taxable income and income-shifting: what is “real” and what is not?

In the working paper version, we also estimate local tax responsiveness of Finnish business owners using the bunching method (Harju and Matikka 2015). We find that business owners bunch very actively at the dividend income tax rate kink point but not at the wage tax rate kink points. This evidence shows that the owners respond actively to dividend tax rates. However, studying excess bunching does not enable us to separate the income-shifting component from the overall response and thus does not provide us with applicable estimates for welfare analysis. Nevertheless, given the ample possibilities to shift income, it is very probable that part of the observed excess bunching for dividends is due to income-shifting between dividends and wages.

Other previous papers also consider tax avoidance and income-shifting within the ETI framework, e.g., Saez (2004) and Chetty (2009).

Costs of income-shifting include, for example, the opportunity cost of time or payments to tax consultants. Income-shifting costs can also depend on the number of owners in the firm, as it might be more burdensome to alter the composition of personal total income when the firm has multiple owners. In addition, tax-motivated income-shifting can be considered socially less acceptable.

Note that assuming \(\alpha =0\) and \(\phi _{i}(\alpha )=0\) or \(\tau _{W}=\tau _{D}\) we are back to the standard ETI model with a single tax base (see, e.g., Gruber and Saez 2002). Then, in the absence of income effects, individual taxable income supply is a function of \((1-\tau _{W})\), and the elasticity of taxable income can be written as

where \(e_{z_{W}}\) is the average ETI. The intuition behind the standard ETI model is that individuals increase \(z_{W}\) until its marginal cost equals the tax rate, and the overall inefficiency of the income tax can be summarized with ETI no matter how \(z_{W}\) is adjusted (Feldstein 1999).

A similar model can also be written for dividend income. For the sake of brevity, we only present the wage income model.

Harju and Kosonen (2013) study tax responsiveness of turnover among the owners of unincorporated firms in Finland. They find small real responses for this group.

In the ETI model of wage earners, investments are generally considered to include investments in human capital, such as education choices and other career considerations. With business owners, it is reasonable to include physical firm-level investments as well.

The same analysis can be carried out for dividends, but for the sake of brevity we only show the equations for wage income.

The fact that income-shifting decreases the welfare loss when the shifted income is also taxed is shown and discussed previously in, e.g., Saez et al. (2012).

In 2007, there are only 185 owners (out of 52,045) in our data for which the Finnish Tax Administration considered to have veiled distribution of profits. However, we do not have exact information on how actively Tax Administration audits this type of tax evasion. Nevertheless, the firms in our estimation sample are relatively small, and thus, the owner typically works in the firm and contributes significantly to the overall output of the firm.

As a whole, the Finnish income tax system follows the principle of individual taxation. The income of a spouse or other family members does not affect the marginal income tax rate of an individual. However, some tax deductions and social security benefits depend on the total income of the household.

Net assets are calculated by subtracting liabilities from assets. Assets include items such as buildings, machinery, equipment, sales receivables and cash holdings. Importantly, assets do not include liabilities. Therefore, net assets can be zero or negative for a given firm with low assets or large liabilities. For dividend tax purposes, negative net assets are treated as zero net assets. The net assets of the firm are calculated using the asset and debt values in the year before. The net asset share of an owner is calculated based on the ownership share of the firm. Also, there are some individual adjustments to the net assets. For example, if the owner or her family members live in a dwelling which is owned by the firm, the value of this dwelling is not included in net assets.

The effective MTR in this case is calculated as 26 % + 0.7*(1-0.26)*28 % = 40.5 %.

There were 416 municipalities in Finland in 2007. Each democratically elected municipal council decides on the municipal tax rate on an annual basis. Municipalities can choose their tax rates freely. However, certain legislative municipal-level duties need to be financed mainly by municipal taxes (e.g., basic health care and primary education).

We predict firm net assets after the reform for each owner using exogenous pre-reform characteristics in 2000–2003. We use the same exogenous individual- and firm-level variables as in the baseline ETI regression. These variables include, for example, owner-level age, age squared, gender, and firm-level turnover, total assets and industry and location dummies. The R-squared statistic for the net assets prediction using OLS is 0.73.

All F-test statistics in columns (1–4) are large, which implies that the instruments are strong. The difference between the number of observations in columns (1–2) and (3–4) is due to the fact that some of the owners do not withdraw wage income from the firm. As a robustness check, we also estimate the wage income model including owners with zero wages (see column (8) of Table 7 in the Appendix). Including these owners does not affect the results.

As in Weber (2014), columns (5–6) do not include income splines as controls. However, including income splines does not significantly affect the results.

It is worth noting that the size of the income component might also affect the interpretation of the estimates. As the underlying tax rate variation is the same as before, broader income components have smaller elasticities if the absolute behavioral response is the same for different income components. Therefore, it is presumable to receive smaller point estimates for the turnover of the firm than for different types of income withdrawn from the firm, such as wages and dividends.

As an additional robustness check, we add 10-piece splines of firm-level income and asset variables in order to more rigorously control for the possibility that changes in individual income and firm-level characteristics are connected. This might be a concern because firm net assets, which also reflect the size of the firm, affect changes in the marginal tax rates on dividends. However, adding firm-level splines does not significantly affect the results. Nevertheless, adding additional splines increases precision.

In general, the owners who do not pay any dividends and are thus not included in our data might respond differently to tax incentives than the owners who pay dividends. Therefore, our results might not fully reflect the average responses of all Finnish business owners. For example, it could be that owners not paying any dividends are less active in income-shifting.