Tax planning and investment responses to dividend taxation

Abstract

This study explores empirically how business owners respond to dividend taxes in a range of different margins including tax planning and investment. Using administrative tax data on all privately held Finnish corporations, I find exceptionally clear dividend payment responses to tax rate discontinuities and changes. Studying the income composition of owners around tax changes reveals clear income shifting between wage and dividends with negligible effect on gross income received from the firm. Evidence on the asset composition of firms indicates that a notable part of the payment response is due to inter-temporal income-smoothing, while I observe no statistically significant real responses in output or investment. Heterogeneity analysis suggests that more experienced owners and owners with lower income have higher tax base elasticities.

Appendix

1.1 Additional details of the institutions and data

Earned income taxation in Finland includes a progressive government tax, a flat municipal income tax, and pension and social security contributions. Both government and municipal taxes include deductions for low-income individuals, making the effective tax schedule very progressive, with the lowest tax rates approximately zero and the highest around 55%, excluding the payroll tax paid by the employer. In calculating the marginal tax rate on earned income, I calculate the tax rate for one extra euro of the particular income type. I exclude the payroll tax³¹, since for most business owners the payroll contribution is not defined by the wage sum, but is based on the so-called entrepreneur’s labor income, which is largely decided by the owner³². Thus the marginal payroll tax of a business owner is generally not affected by an additional euro of gross income.

The earned income tax rate applied above the net asset threshold varies depending on the taxpayer’s income and municipality. Both the municipal and government tax schedules change nearly every year. The lowest government tax rate has been zero over the whole period and, with deductions in the municipal tax for low-income earners, the aggregate earned income tax rate has also been close to zero at the low end of the income distribution. The highest overall marginal earned income tax rate has been around 55%. The government tax rates on earned income decreased over the research period of 2006–2016, especially for low- and middle-income earners. However, municipal income tax has increased; in 2000, the average rate was 17.7%, but in 2015 it was 19.9%. The municipal income tax varies across municipalities; in 2015 it ranged from 16.5% to 22.5%. Figure 11 plots the average threshold created by the net asset threshold. As the tax rate above the threshold depends on the taxpayer’s other income, the tax rates above the threshold in the figure are calculated as an average of the marginal tax rates of firm owners in each bin.

Fig. 11

Average marginal tax rate for firms paying dividends under the monetary dividend tax threshold. Note: The tax rate above the threshold is estimated as a mean of the marginal earned income tax rates in the data. The earned income tax rate varies depending on the taxpayer’s income and municipality. Both the municipal and government tax schedules change nearly every year. The lowest government tax rate has been zero over the whole period and, with deductions in the municipal tax for low-income earners, the aggregate earned income tax rate has also been close to zero at the low end of the income distribution. The highest overall marginal earned income tax rate has been around 55%. Overall government tax rates on earned income have been decreasing over the research period of 2000–2013, especially for low- and middle-income earners. However, the municipal income tax has been increasing; in 2000, the average rate was 17.7%, but in 2013 it was 19.4%. The municipal income tax varies across municipalities; in 2015 it ranged from 16.5% to 22.5%

Figure 12 visualizes the thresholds in marginal tax rates. It is three-dimensional to reflect the fact that the dividend tax rate depends not only on the dividend amount but also on the net asset position of the firm. For an individual firm owner, the entire region is not available, and the firm’s net assets define a restriction that slices the three-dimensional dividend tax schedule. For example, a sole owner of a firm with exactly 1 million euros of net assets could locate exactly at the corner of the lowest plane. By receiving more dividends, the owner would face the earned income schedule, which is the high uneven plane in the graph. The earned income tax rates above the threshold are calculated as averages of the individual marginal tax rates of owners at the threshold.

Fig. 12

Marginal tax rate for dividends 2006–2011. Note: This graph describes the thresholds in 2006–2011, when the kink was at 90,000 EUR and the net asset threshold at 9%. Above the net asset threshold, the owner pays earned income tax on 70% of income (85% since 2014) in addition to the corporate tax. The tax rate above the net asset threshold is estimated as a mean of the actual marginal earned income tax rates in each 5000-euro dividend bin

1.2 Bunching method

I estimate the extent of excess mass and the according elasticity of taxable income using the bunching method developed by Saez (2010)³³. The elasticity of taxable income (ETI) is the ratio of a percentage change in taxable income to a percentage change in the net-of-tax income rate (one minus the tax rate). The bunching method uses the excess mass at a tax schedule discontinuity to estimate the corresponding elasticity of taxable income.

To measure excess mass, I first estimate a counterfactual distribution that describes what the dividend distribution would approximately be in the absence of the kink point³⁴. The counterfactual distributions around the monetary kink points are estimated as

$$\begin{aligned} \widehat{C_{j}^{0}}=\sum _{i=0}^{p}\beta _{i}^{0}\cdot (Z_{j})^{i} +\rho \cdot \textbf{1}\left[ \frac{Z_{j}}{r}\in \mathbb {N}\right] +\varepsilon _{j},\;Z_{j}\notin \left[ -R;R\right] , \end{aligned}$$

(2)

where $\widehat{C_{j}^{0}}$ is the estimate of the counterfactual distribution in each bin j with dividend income $Z_{j}$. $\beta _{i}^{0}$ are the regression estimates, and p denotes the degree of the polynomial. $\rho$ in the second term captures the round number fixed effect that is observed in Fig. 1. $\left[ -R;R\right]$ is the excluded range of the distribution, which denotes the area where the kink point affects the owners’ behavior. Following earlier literature (e.g. Chetty et al., 2011), this area is selected by visual observation of the data. My results and conclusions are not sensitive to the choice of $\left[ -R;R\right]$ or the order of the polynomial.

I estimate the counterfactual distribution around the net asset threshold following Eq. (3).

$$\begin{aligned} \widehat{C_{j}^{0}}=\sum _{i=0}^{p}\beta _{i}^{0}\cdot (Z_{j})^{i} +\frac{\sum _{j=-R}^{R}C_{j}}{2A+1}+\varepsilon _{j},\;Z_{j} \notin \left[ -R;R\right] ,\;j\in [-A;A] \end{aligned}$$

(3)

The basic principle is the same as in Eq. (2). Given the very strong bunching, the second term is used to spread the bunchers to the surrounding region to make the sum of firms in the counterfactual distribution match that of the realized distribution. For this distribution, there is no need to consider round number bunching. In estimating the counterfactual distribution, I include both the region below and above the threshold, as the system is likely to induce early payments of dividends, as discussed in Kari and Laitila (2015).

The sum of the excess observations in the bunching range is

$$\begin{aligned} \sum _{j=-R}^{R}\widehat{B}_j=\sum _{j=-R}^{R}\left( C_{j} -\widehat{C_{j}^{0}}\right) . \end{aligned}$$

(4)

The estimate of excess bunching $\hat{b}$ is then the estimated excess mass around the kink relative to the average density of the counterfactual dividend distribution between $-R$ and R

$$\begin{aligned} \hat{b}=\frac{\sum _{j=-R}^{R}\widehat{B}_{j}}{\sum _{j=-R}^{R} \widehat{C_{j}^{0}}/(2R+1)}. \end{aligned}$$

(5)

Finally, the excess bunching can be turned into an elasticity estimate. The elasticities at the kink points are estimated as

$$\begin{aligned} \varepsilon _{D}=\frac{dD}{d(1-\tau )}\frac{1-\tau }{D} =\frac{\hat{b}}{D^{*}\cdot \log \left( \frac{(1-\tau _{D})}{(1-\tau _{D}-\triangle \tau _{D})}\right) }. \end{aligned}$$

(6)

D denotes dividend income, $\tau$ the dividend income tax rate that jumps at a kink point $D^{*}$ from $\tau _{D}$ to $\tau _{D}+\triangle \tau _{D}$. When estimating the elasticities at the net asset thresholds, I specify the marginal tax rate above the threshold for each firm owner individually. Then I use the aggregate bunching response to estimate the elasticity for each owner and report the mean elasticity.

Following earlier literature, I use the bootstrap method to construct standard errors (see Kleven (2016) for a review). In the bootstrap method, I sample the residuals from the regression a large number of times (300), with replacement, and estimate an elasticity for each draw. Using these elasticities, I calculate a standard error for the original elasticity estimate.

1.3 Additional empirical details

Table 15 gives the summary statistics for the full data. Altogether, the data consist of 641,558 observations across 11 years.

Table 15 Summary statistics of the data 2006–2016

Fig. 13

Industry shares among 90k bunchers 2006–2011. Note: The figure plots the shares of each industry in bins around the 90,000 EUR threshold. The horizontal axis shows the dividend amount and the vertical axis the share of the industry in each bin. The dashed horizontal line denotes the average share of the industry in the data. According to the figure, the financial sector seems to be over-represented at the kink

Figure 13 shows how various industries are represented in each bin around the 90,000 EUR threshold. The horizontal dashed line shows the industry’s average share in the data. The figure shows that the finance industry is overrepresented among the firms bunching at the monetary threshold.

Table 16 shows the proportion of firms that move together with the threshold. When the monetary threshold moved from 90,000 EUR to 60,000 EUR, 47% of the preceding excess mass firms followed the threshold. At the 150,000 EUR threshold, one third of the observations had previously paid exactly 60,000 EUR of dividends, which was the preceding threshold. At the 8% net asset threshold, 70% of firms had previously bunched at the 9% threshold.

Figure 15 shows the income composition in two consecutive years when there was no tax change. There is now no change in the owners’ income composition. The figure acts as a robustness check that the shift observed in Fig. 4 was driven by the tax change.

Figure 14 shows the accumulation of aggregated assets in privately held corporations in 2000–2016. Net assets consists of retained earnings, financial assets, and additions to depreciating capital. The figure also shows the evolution of aggregated profits, dividends and net investment in depreciating capital. The figure shows a clear increase in accumulated assets, starting especially after the introduction of the current dividend tax system in 2005. There is no increase in aggregate investment, so this is not likely to solely explain the accumulation of assets.

Table 16 Percentage share of firm owners relocating together with the kink

Fig. 14

Aggregate net asset accumulation, profits and retained earnings. Note: This figure shows the accumulation of aggregate net assets among the privately held firms studied in this paper in gray. The blue line shows annual aggregate profits, the red dashed line the main owner’s annual aggregate dividends and the dashed green line annual aggregate net investment

Fig. 15

Income shifting between wages and dividends. Note: This figure plots the income composition between wages and dividends in 2008–2009 and in 2010–2011. For the figure, the main owners’ wages and dividends from the firm are counted together as total income. Then the owners are divided into 50 income quantiles (2-percentiles). Finally, for each quantile average wages and dividends are calculated. The horizontal line shows the average dividends and the vertical line the average wage in each bin. The isoquant lines show total income from the firm. The figure shows that when there was no tax change the owners’ income composition stayed more or less the same. This figure acts as a placebo check for Fig. 4 in the main text