Essays on the Effects of Taxationon Firms

Aliisa Koivisto

Doctoral dissertation, to be presented for public discussion

with the permission of the Faculty of Social Sciences of the

University of Helsinki, in Lecture hall, Economicum, on the

24th of September, 2021 at 13 o’clock.

Faculty of Social SciencesUniversity of Helsinki

©Aliisa Koivisto

Helsinki Graduate School of EconomicsSupervisors: Jukka Pirttila and Markus JanttiPre-examiners: Hakan Selin and Tuukka Saarimaa

The Faculty of Social Sciences uses the Urkund system (plagiarismrecognition) to examine all doctoral dissertations.

Valtiotieteellisen tiedekunnan julkaisuja – Publications of theFaculty of Social Sciences 194/2021

ISBN 978-951-51-7019-4 (print)ISBN 978-951-51-7020-0 (pdf)ISSN 2343-273X (print)ISSN 2343-2748 (pdf)

UnigrafiaHelsinki, September 2021

Abstract

This doctoral dissertation is a collection of an introductory chapter and three essays

on the field of public finance. In this dissertation, I look into three different tax

policies aimed at spurring business activity. Using rich Finnish microdata and state-

of-the-art econometric tools, I study how firms respond to a dividend tax, corporate

tax, and a household tax credit.

In the first essay, I study how business owners of privately held corporations re-

spond to dividend taxes. I use administrative data on all privately held Finnish

corporations and their main owners in 2006–2016 together with tax schedule dis-

continuities and changes in the schedule as variation. The dividend tax schedule in

Finland includes deduction thresholds, effectively creating clearly lower marginal tax

rates for certain amounts of dividend income in comparison to labour income. These

thresholds create exceptionally large incentives for firm owners to respond by e.g. ad-

justing their income or changing their investment choices. By using bunching method

developed by Saez (2010), I find exceptionally clear dividend payment responses to

tax rates, with elasticities ranging from 0.5 to 3.6 in different thresholds. However,

this elasticity parameter does not compare to the structural costs of taxation as it

captures tax planning and other channels that affect dividend pay-out. I examine

the potential mechanisms driving the bunching at the thresholds using changes in the

dividend tax thresholds. I find no statistically significant responses in investment or

output. Further descriptive analysis on the asset structures of the firms suggests that

most of the payment response may be due to inter-temporal income-smoothing, as

the balance sheets reveal firms at the tax thresholds accumulating financial assets in

the firm.

The second essay is about the impact of corporate taxes on small firms. It is co-

authored with Jarkko Harju and Tuomas Matikka. We look at how small firms and

their investment and production choices respond to a 4.5 percentage-point reduction

in the corporate tax rate in 2014 in Finland. This corporate tax cut was combined

with a dividend tax increase that left the effective shareholder-level tax rate mostly

unchanged. Thus, this exceptional tax cut allows us to focus solely on the effects of

firm level tax and empirically analyze the differential incentives of taxes set at the

firm level in comparison to owner level taxes. Using detailed administrative data and

difference-in-differences method, we find no significant investment responses in the

stock of productive capital after the tax cut. However, we observe an increase in sales

1

and input usage of the treated firms, implying a higher growth rate after the tax cut.

Dividing the corporations between two groups with passive and active owners, based

on the ownership type, reveals that this positive impact on sales is fully driven by

entrepreneurs who actively work and manage their firms. As this tax cut is effectively

a cut in the tax on retained earnings, it suggests that, for small firms, owner effort

and role plays an important part in how this cash injection within the firm is spent.

The third essay is about the household tax credit (HTC), and it is coauthored

with Jarkko Harju and Tuomas Kosonen. HTC is a tax credit for consumers using

household services with the aim of increasing employment in the service sector and

curbing tax evasion. We use reforms in the HTC system together with data from

Finland and Sweden to study how the HTC reaches these aims. In addition, we

explore the distributive consequences of HTC. We use two empirical settings to study

the causal impacts of the credit. First, we compare household service industries

between Sweden and Finland. We use the adoption of the current HTC system for

cleaning services in Sweden in July 2007 as variation, and Finland, which already had

the HTC system in place, as a control group. We find no increase in the reported

value of sales among cleaning firms in Sweden relative to the Finnish firms after the

introduction of HTC for cleaning services in Sweden. In our second setting, we study

the renovation industry in Finland and use other similar industries as a domestic

control group. Finland increased the amount of maximum tax credit from 1150 to

3000 euros for the renovation industry in 2009, and we compare firms operating in the

renovation industry with a matched control group before and after 2009. We do not

find any response in sales of renovation services after the increase in maximum HTC

relative to the control group, suggesting negligible demand elasticity with respect to

the size of HTC. Finally, a descriptive analysis with the administrative data shows

that a relatively large share of individuals claiming HTC make costly mistakes in their

reports to the tax authority. This shows as a large excess mass of taxpayers bunching

in a ”wrong” threshold, without any other reason than their misunderstanding of the

claiming system. Our descriptive analysis also shows that higher income households

use HTC to a much greater extent than lower income households, and very poor

households do not utilize HTC almost at all.

2

TABLE OF CONTENTS

Abstract 1

Table of Contents 3

Acknowledgements 5

Chapter 1 – Introduction 7

Summaries of the Essays 14

Chapter 2 – Dividend Tax Thresholds and Extreme Bunching 23

Introduction 23

Institutions and Data 29

Dividend Payment Responses 33

Mechanisms 40

Conclusion 49

References 51

Appendix 53

Chapter 3 – The Effects of Corporate Taxes on Small Firms 61

Introduction 61

Finnish Business Tax System and the Reform of 2014 66

Expected Impacts of the Reform 67

Data, Methods and Identification 71

Results 74

References 84

Figures 87

Tables 95

A Appendix: Additional Figures and Tables 104

B Appendix: Business Tax System in Finland 115

3

Chapter 4 – Does Household Tax Credit Increase Employment? 119

Introduction 119

Institutions 124

Empirical Approach 128

Descriptive Analysis 133

The Impact of the HTC on Consumption 140

Conclusions 147

References 150

Figures 152

Tables 174

Appendix 181

4

Acknowledgements

In writing this doctoral thesis, I have received enormous support and help from various

individuals and institutions. I wish to express my gratitude to all of those who have

helped and supported me along the way.

First, I would like to thank my supervisors Jukka Pirttila and Markus Jantti who

have encouraged and supported me through this project. Their insightful guidance

has greatly improved my thesis and motivated me to finishing it. It has been a

privilege to receive your guidance.

I am extremely grateful for Jarkko Harju and Tuomas Matikka who have had a

indispensable role in guiding me through my PhD studies. Our path started already

during my master studies when I was trainee at VATT working with you and you

encouraged me to start PhD studies. I have learned massively from working with

Jarkko and Tuomas and I hope our collaboration continues. In addition, I am thankful

for Jarkko Harju, Tuomas Matikka and Tuomas Kosonen for being such great and

competent coauthors, I have learned a lot from you.

I feel lucky to have worked at VATT for a significant share of my PhD studies, it

has been a wonderful environment to work on my research. I am also very thankful

to Seppo Kari, my supervisor at VATT, who have encouraged me in this process, and

to all other colleagues at VATT for many great conversations and for creating such

an stimulating working environment.

I wish to thank the two pre-examinators of my thesis Hakan Selin and Tuukka

Saarimaa for their helpful comments and thoughts of my thesis and I am especially

grateful for Hakan Selin for agreeing to act as my opponent.

Finishing the PhD studies would not have been possible without my fellow grad-

uate students. Aino Kalmbach, I am grateful for the fun moments, encouragement

and your friendship, it has really carried me through these studies. I would also like

to express my gratitude for Maria Jouste, Annika Nivala and other fellow graduate

students for creating such a supportive and fun environment to pursue a PhD.

I am thankful for Nordic Tax Research Council, Suomen Arvopaperimarkkinoiden

Edistamissaatio and Yrjo Jahnsson Foundation for providing me financial support on

this journey. This support enabled a visit to University of California Berkeley. I

am very grateful for Markus Jantti for organizing the visit and for professor Alan

Auerbach for hosting me. I would like to thank the whole department for interesting

lectures and discussions and my friends Laurence Wainwright, Eileen Wehmann and

5

Grant Thompson for making me feel like home.

Writing a PhD would not have been possible without the love and support of

my family and friends. Especially, I want to express my deepest gratitude for my

parents Hilve and Seppo for believing in me always. I want to thank my sister Laura

for being an inspiration and a mentor for me. Saara, Saara, Kira, Aisha, Johanna,

Jonna, Anni, Karo and other friends, thank you for your friendship and all the fun

moments, you have kept my life in balance!

Finally, I want to thank Atte for the endless support you have given me and for

always being there for me, I could not wish for a better spouse. Our daughter Hilda,

thank you for giving me a strict deadline to finish this dissertation.

6

Chapter 1

Introduction

How important are taxes in determining firm behavior? This doctoral dissertation

is a collection of three essays looking into firm behavior in various decision margins

and how taxation affects these choices. Private firms constitute a key economic agent

that is considered in charge of production of most goods in standard economic models.

Moreover, it has been argued that firms foster innovation growth1; hence, firms play

a fundamental part in economic growth. While the theoretical literature regarding

firms’ investment decisions and output choices is comprehensive, with multiple ap-

plications and production functions, the theory and the scarce empirical evidence on

the effects of taxation on them are still relatively dissonant. The academic literature

on how one should think about firms’ output decisions in relation to taxation is still

somewhat confused over a number of questions, including: what is the incidence of

corporate tax, how much does dividend tax distort output, how can employment be

increased by tax policy? In this dissertation, I aim to better understand the drivers

in firms’ investment decisions and output choices in relation to taxation by looking

empirically into three different tax policies aimed at spurring business activity. Using

rich Finnish microdata and state-of-the-art econometric tools, I study how firms re-

spond to a dividend tax, corporate tax, and a household tax credit. This dissertation

focuses on businesses and empirically studying how these three tax policies affect firm

outcomes.

Governments raise taxes to provide public goods and for redistributive purposes.

Some tax tools also have additional goals, such as incorporating externalities2 or

1See e.g. literature on endogenous technological change including Romer (1990) and Aghion andHowitt (1992).

2Externalities mean public costs or benefits of the consumption that do not fall only upon theconsumer, e.g. pollution and vaccination. Ways to incorporate externalities include cigarette taxesand fuel taxes.

7

boosting investment and employment. The primary goal of corporate income tax-

ation is effectively to raise revenue in a manner that is considered equitable and

that causes as little efficiency cost through distortions as possible. Corporate income

taxation includes taxes set on profit at the firm level –corporate tax– and taxes on

distributed profits, such as dividend tax. In addition, for some corporate forms, such

as partnerships and sole proprietors, the profits are taxed directly at the owner level.

There are several reasons to tax corporate income. Governments aim to raise

revenue to fund public services, and many of these services are also used by firms3.

Taxing corporate profit is relatively simple and it broadens the tax base. Taxation

only at the owner level (e.g. dividend taxation) is easily avoided by retaining profits

in the firm through various ownership structures4 or other tax avoidance means,

such as income shifting. If corporate tax differs notably from higher owner-level

taxes, high income individuals may avoid owner-level taxes by retaining earning in

the firm, which reduces capital mobility. In addition, large differences between labor

and business income create incentives for income shifting between tax bases, so that

e.g. a high-income professional would aim to shift their income to be taxed as capital

income. Thus, corporate income taxes are an important part of the tax base. The

revenue from corporate tax alone in 2019 was 6,015 billion euros, constituting 6 % of

the total tax revenue in Finland and that does not even include owner level taxation

or taxes set on consumption. Corporate income taxes also provide an additional

channel to tax the top incomes. Corporate owners are often well represented at the

upper end of the income and wealth distribution, and low effective taxes at the high

end are sometimes considered unequitable (Piketty and Zucman, 2014).

Many public policies, funded with taxation, are considered important by the broad

public, while taxation itself may cause some resentment. In addition to resentment,

taxation can induce real economic costs through various distortions, creating chal-

lenges for planning good tax policies and the tax system as a whole. An economy uses

its resources efficiently when the marginal products of different activities are equal.

This requires that taxes do not distort the choice between investment and other pro-

duction factors (inputs). In practice, corporate income taxation hardly ever reaches

this target of neutrality. Corporate income taxation can distort business activity, es-

pecially through investment choices and owner effort. This is why corporate income

3E.g. Transport networks, education for (future) employees.4E.g. profit shifting to tax havens (Gravelle, 2009).

8

taxation often enters policy debates when governments aim to spur investment and

growth. Accordingly, provoked by international tax competition, corporate income

taxes have been decreasing across developed countries since the 1980s (Heinemann

et al. 2010 and Comission 2019).

Theory literature discusses four main ways through which corporate income taxes

may affect investment. First, corporate income taxes increase the cost of capital,

which sets the minimum requirement for the marginal revenue of the investment.

For an investment to be profitable, its marginal revenue should at least equal to

its marginal cost. Corporate income taxes increase the marginal cost of capital and

may therefore reduce investment (Harberger 1962; Hall and Jorgenson 1967). Sec-

ond, business taxation often distorts the relative marginal costs of different sources

of capital, affecting the funding balance between debt and equity. Increased debt

financing may e.g. affect sensitivity to fiscal cycles. Firm-level corporate taxes also

reduce the amount of earnings retained in the firm, which restrains the sources of

funding for new investment5 (Mirrlees et al., 2011). Third, corporate income taxes

may distort the decisions between different investment items by e.g. incentivising

investment in assets with relatively high depreciation in tax law compared to the real

depreciation. The incentive rises from the lower present value of future depreciation

compare to depreciating the asset now, the lower present value being due to inflation

and the discount rate (Mirrlees et al. 2011). Fourth, corporate income taxes also

affect international tax competition and the location of firms and capital. Devereux

and Griffith (1998) argue that while the effective marginal tax rate affects the size

of the investment conditional on location choice, the discrete investment choice on

where to locate is primarily affected by the average effective tax rate.

Corporate tax aims at taxing the owners, however, corporate taxation also affects

employees, subcontractors and clients. Lower capital investment reduces worker pro-

ductivity, potentially affecting wages. Therefore, the tax incidence does not fall only

on the owner; employees are likely to bear some share of the burden through lower

wages (Bradford 1978; Kotlikoff and Summers 1987). This is especially relevant in an

open economy, where the level of rate of return for capital investment is fixed. More-

over, firms may be able to shift some of the burden to the prices, i.e on to customers,

especially if the markets are not perfectly competitive (Auerbach and Hines, 2001).

5However, this distortion is alleviated by the fact that investment may be funded also with newequity or debt.

9

Owner-level taxes, such as dividend tax on top of corporate tax on profit, create

so-called ”double-taxation” of business income. This can again affect the allocation of

investment within an economy. Therefore, many countries apply different corporate

tax deduction policies on owner-level taxation. However, the so-called new view in

dividend tax literature states that, because dividend tax is not paid on retained

earnings, it does not affect investment choices (Auerbach, 1979). The argument is

that if the tax rate on dividend income remains constant, then the dividend tax

reduced on the net cost to the shareholder is exactly the same as the rate at which

the eventual return is taxed. These two effects cancel out to leave the required rate of

return unaffected, and hence the effective marginal owner-level tax rate equal to zero.

However, the argument does not hold when the investment is funded with new equity,

so the cost of capital for new equity e.g. at the stage of establishment, also bears

the dividend tax (Sinn, 1991). Therefore, dividend tax is less likely to have an effect

on established firms, but can potentially distort choices at the extensive margin, i.e.

starting a business. Firm owners may avoid dividend taxation by retaining earnings

in the firm and this may lead to capitalization of dividend tax into higher share

values (Auerbach, 1979). Chetty and Saez (2010) note that retaining profits may also

amplify principal-agent conflicts of interest by leaving more cash under the control of

managerial choices and disincentivizing the close monitoring of the managers. This

may lead to unproductive investments using retained earnings.

The negative effects of corporate income taxation have induced governments to

carry out a range of tax policies aimed at reducing these distortions. These include

policies such as dual-income tax systems that tax capital income and wages separately,

lower corporate taxes6 and various investment tax credits and tax depreciation sched-

ules. However, complicated policies often bring up tax planning responses7. These

responses include behavioral responses related to the timing of economic transactions

and accounting and financial responses. According to Slemrod (1992), these responses

often exceed the real impacts in output when studying the effects of tax changes or

reforms. Often, in simple theoretical models only a real response in output is possible,

but tax systems induce more than that. Taxes create incentives to change the tim-

ing of transactions8, restructure financial claims, misreport income, change the legal

6Most developed countries have reduced their corporate taxes since the 1980’s (Heinemann et al.2010 and Comission 2019).

7Or unintended behavioral responses rising from e.g. present bias, inattention, or inertia.8Slemrod (1992); Kreiner et al. (2014).

10

form of business organization9, shift income across tax bases10 and numerous other

behavioral responses. While tax planning is considered to affect income distribution

more and efficiency less, tax planning responses are likely to affect how policies reach

their distributional and tax revenue goals, which affects the welfare implications of

the tax policies.

The active theoretical discussion around corporate income taxes has, in recent

years, been complemented with empirical literature enabled by evolved empirical

methods, data access, and computational power. These advanced empirical studies

use quasi-experimental settings to estimate parameters, such as elasticities and tax

incidences, left unsolved in the theory literature. These measures for various effects of

business taxes include e.g. elasticity of taxable income, elasticity of investment with

respect to corporate taxation, and elasticity of wage with respect to corporate tax

rate, that are then used to calculate the tax incidence. Elasticity is a measure that

quantifies how much an outcome is expected to change in percentages as a response

to a percentage change in some parameter such as tax.

Recent empirical studies on the investment effects of corporate income taxes use

variation in tax parameters related to depreciation and deduction rates to study

the effects on investment. House and Shapiro (2008), Zwick and Mahon (2017) and

Maffini et al. (2019) use changes in depreciation regulation in the US and UK (the

latter study) and find large investment elasticities with respect to the net of corporate

tax rate of around 7. Ohrn (2018) uses variation in deduction rules in the US and

find an investment elasticity with respect to net corporate of tax rate of 6.5. All these

studies apply a difference-in-difference set-up with panel data.

A recent paper by Fuest et al. (2018) studies corporate taxes’ incidence on labor.

They use tax changes in municipal level corporate taxes to study the incidence of

corporate taxes on wages. In their event-study designs together with difference-in-

differences method they find that workers bear on average 51% of the corporate tax

burden. However, they also find heterogeneity implying that low-skilled, young and

female employees bear on average a larger share of the tax burden than highly-skilled

and male employees. Their findings also suggest that labor market institutions and

profit-shifting opportunities have an impact on the incidence. The incidence found

in Fuest et al. (2018) is very also similar to that in Arulampalam et al. (2012).

9Gordon and Mackie-Mason (1994).10Harju and Matikka (2016).

11

The discussion regarding high incomes of business owners has been empirically

addressed by Smith et al. (2019), who use sudden deaths of working-age top earning

business owners in the US as an event study type of variation to study the owner’s

role in firm performance. They approximate that among top earners (0.1%) three-

quarters of owner-level profits are returns on human capital and only one quarter on

capital.

The effects of dividend taxes have been studied empirically by Alstadsæter et al.

(2017) and Yagan (2015), among others. Alstadsæter et al. (2017) use triple-difference

and a dividend tax cut in Sweden to study whether dividend taxes affect corporate

investment. Quite in line with the new view in theoretical literature, they find no

effect on corporate investment form on aggregate level. However, they find that

the cut affected the allocation of investment so that cash-constrained corporations

increased their investment relative to the cash-rich. This is in line with the agency

model of Chetty and Saez (2010), which predicted that a dividend tax cut would

improve capital allocation by releasing assets from cash-rich firms as dividends –

leading to less investment in cash-rich firms and increasing investment in cash-poor

firms, by improving access to new equity through lowering the cost of capital. Again,

in line with the new view, Yagan (2015) studies the dividend tax cut in the US in 2003

and shows that despite notable effects on dividend payouts, there was no increase in

investment as a response to the cut.

One way to estimate the elasticities with respect to tax rate is to use bunching

at tax thresholds to estimate responsiveness with respect to the variation created by

the threshold. Bastani and Selin (2014) and Chetty et al. (2011) find that business

owners do respond to these tax rate discontinuities clearly, suggesting an elasticity

of 0.07 and 0.1-0.2 respectively11. However, these estimates do not reflect the labor

supply elasticity of the business owners as both estimates include income-shifting,

which appears to cause a notable share of the excess mass. In Bastani and Selin

(2014), if various deduction channels that self-employed persons can use to adjust

their taxable income in the current year are considered, the excess mass estimate

falls close to zero. In the same Danish set-up of Chetty et al. (2011), Le Maire and

Schjerning (2013) demonstrate that the self-employed adjust their retained earnings

and profit distributions inter-temporarily to avoid the highest tax brackets.

Firms play an inseparable role in economic growth, employment, innovation etc.

11However, they do not find bunching among wage earners.

12

(E.g. Decker et al. 2014). Therefore, the tax policies studied in this dissertation

are often brought up in policy debates as governments look for ways to promote

employment as well as business activity. While the theories of these tax policies are

quite well developed, there is still a need to know much more, especially on how well

the theories match the empirical evidence.

The policies studied in this dissertation – dividend tax adjustments, corporate

tax cuts and the household tax credit (HTC) – aim to spur economic activity in dif-

ferent ways. Essentially, corporate income tax cuts aim for economic growth. The

cuts in corporate tax could potentially increase investment, which would increase

worker productivity and even employment. Similarly, dividend tax cuts usually aim

to attract investment and promote business activity. While the mechanism through

which dividend and corporate tax cuts should boost investment and employment fol-

low quite straightforwardly from standard economic theory, the household tax credit

has a very different starting point for increasing employment. HTC does not aim at

productivity growth, but simply at raising employment through increasing demand

for labor-intensive goods by altering relative prices with a tax credit for consumers.

If the HTC then leads to increased consumption of services, it could have a positive

effect on the economy by increasing employment12. However, with a low supply elas-

ticity, HTC may also just pass on to prices; with a low demand elasticity, it might

merely reward those who consume HTC services irrespective of the credit without

affecting the amount consumed.

Modern data and empirical tools have triggered a causal revolution in empiri-

cal public finance. I use modern econometric tools with rich administrative data to

study firm responses to tax policies. In all essays, I use a quasi-experimental setting

for causal inference: this means I need changes in tax policies and relevant com-

parison groups, both enabled by the Finnish setting. In this dissertation, I utilize

difference-in-differences13 methods together with visual evidence on pre-trends and

other descriptive features. Moreover, I use a bunching method14, in addition to re-

sponses derived from the difference-in-differences setup, to estimate elasticities in the

first essay. I improve the empirical precision with instrumental variables to find ex-

ogenous variation in the first essay, a weighting method by DiNardo et al. (1996) to

12Additionally, HTC aims to reduce tax evasion. If the potentially increased revenue is then betterspent by the public sector, this would also have a positive impact on the economy.

13Angrist and Pischke (2009) provides an introduction to the method.14The method originally follows Saez (2010) and Kleven (2016) provides an useful review of it.

13

improve the matching of treatment and control group in the second essay, and I apply

the CEM-matching algorithm (Blackwell et al., 2009) in the third essay.

The next subsection provides summaries on the three essays.

Summaries of the Essays

Dividend tax thresholds and extreme bunching

In the second chapter, I study how business owners respond to dividend taxes. I use

administrative data on all privately held Finnish corporations and their main owners

in 2006–2016 together with tax schedule discontinuities and changes in the schedule

as variation. The Finnish dividend tax schedule provides exceptionally large incen-

tives for firms to respond. The dividend tax schedule in Finland includes deduction

thresholds, effectively creating clearly lower marginal tax rates for certain amounts

of dividend income in comparison to labour income. The dividend tax rate jumps

notably at a threshold that is set first at 9 and then at 8 percent return on net assets.

The reasoning for this threshold is that the government wants to curb income shift-

ing: the lower and linear tax rate is only targeted at some normal return (8 %) on

invested capital, whereas a return above this rate is considered either as return on la-

bor or economic rent, and hence taxed according to the progressive earned income tax

schedule. Moreover, there is a monetary threshold for dividends exempted from most

of the capital income tax to alleviate the double taxation of corporate profits. These

discontinuities create strong incentives for firms/owners to respond by e.g. adjusting

their income or changing their investment choices. For example, the marginal tax

rate on dividends (including corporate taxes) jumped from 28% to 40.5% at 90,000

euros between 2006 and 2011. The discontinuities have also changed several times

over the past decade, creating additional variation.

I use bunching method developed by Saez (2010) to measure the responsiveness

to these dividend tax schedule discontinuities. I find exceptionally clear dividend

payment responses to tax rates, with elasticities ranging from 0.5 at the monetary

thresholds to 3.6 at the net asset thresholds. This implies that a 1 percent increase

in the net of dividend tax rate increases taxable dividend income by 0.5-3.6 percent,

which is a very large response. However, the elasticity parameter obtained using the

bunching method does not compare to the structural costs of taxation as it captures

tax planning and other channels that affect dividend pay-out.

14

I examine the potential mechanisms driving the bunching at the thresholds using

changes in the dividend tax thresholds. Moving the dividend tax threshold brings

new firms into the range of the higher marginal tax rate. I use similar size firms with

different ownership shares as treatment and control groups. I find no statistically

significant responses in investment or output. Further descriptive analysis on the

asset structures of the firms suggests that most of the payment response may be due

to inter-temporal income-smoothing, as the balance sheets reveal firms at the tax

thresholds accumulating financial assets in the firm. In other words, firm owners

avoid the higher dividend tax brackets by retaining earnings in the firm, which is in

line with the new view and the capitalization of the dividend tax (Auerbach, 1979).

Retaining profits has several tax benefits. In addition to avoiding the higher tax

bracket, the retained earnings increase the firms’ value through increased net assets,

and therefore, allows for a higher amount of dividend to be distributed in the lower

capital income tax bracket in the future, as the tax schedule depends on the net

assets of the firm. Also, some forms of capital income are taxed more lightly when

received by a firm, so saving through a firm is lucrative. This is likely to additionally

boost the capitalization of dividend taxes into share values. Finally, by studying the

income composition of firm owners around the time of tax changes, I observe that

firm owners engage in income-shifting across wage income and dividends to minimize

their tax burden, although the gross income received from the firm did not change.

The effects of corporate taxes on small firms

The third chapter is an essay about the impact of corporate taxes on small firms,

co-authored with Jarkko Harju and Tuomas Matikka. In this study, we look at how

small firms and their investment and production choices respond to a corporate tax

cut. There is little evidence on the impact of decreasing corporate tax rates on firms

and especially on small firms, even though these have been described as playing a key

role in spurring economic growth and employment (Decker et al., 2014). We study

a 4.5 percentage-point reduction in the corporate tax rate in 2014 in Finland. This

corporate tax cut was combined with a dividend tax increase that left the effective

shareholder-level tax rate mostly unchanged. As discussed earlier in this introduc-

tion, the owner and firm-level taxes induce somewhat different incentives. Thus, this

exceptional tax cut allows us to focus solely on the effects of firm level tax and empir-

ically analyze the differential incentives of taxes set at the firm level in comparison to

15

owner level taxes. The relatively large tax cut together with detailed administrative

data covering all Finnish businesses enables us to analyze the effects on multiple firm

outcomes.

We use a difference-in-differences method with a similar-sized and comparable

partnerships as a control group. Partnerships are taxed directly at the owner-level

and they do not face a change in their taxation. Our analysis focuses on small firms

with annual sales below 2.5 million euros. In addition, to ensure the comparability

of corporations and the control group, we apply weights based on industry-size cate-

gories to avoid differential trends among industries or sizes affecting the results. Our

empirical approach is validated with parallel pre-trends among the treatment and the

control group. We study the effect of the tax cut on investments relative to existing

capital assets, which is the main outcome variable in earlier literature studying the in-

vestment effects of corporate taxation15. We find no significant investment responses

in the stock of productive capital after the tax cut.

Next, we examine the impact of the tax cut on other outcomes reflecting the

business activity of firms, including sales, labor costs, input use, value added and

firm entry. While we find no investment effects, there is an increase in sales and

input usage of the treated firms, implying a higher growth rate after the tax cut.

Dividing the corporations between two groups with passive and active owners, based

on the ownership type, reveals that this positive impact on sales is fully driven by

entrepreneurs who actively work and manage their firms. As this tax cut is effectively

a cut in the tax on retained earnings, it suggests that, for small firms, owner effort

and role plays an important part in how this cash injection within the firm is spent.

This evidence extends the result by Chetty and Saez (2010), who argue that there

is a conflict of interest between managers and shareholders regarding which may

lead to inefficient allocation of retained earnings, further leading to differentiating

investment and dividend pay-out choices. Our results highlight that investment is

not the only relevant decision margin when studying the effects of corporate taxation

and that, especially among closely held small firms, the owner’s efforts may affect the

response to a tax change. Interestingly, we find no statistically significant differences

in responses for more or less cash-constrained firms and no effects on firm entry,

suggesting that these channels are not driving our main findings.

15See e.g. House and Shapiro (2008), Zwick and Mahon (2017), Ohrn (2018) and Maffini et al.(2019)

16

Does household tax credit increase employment?

In the fourth chapter, I focus on household tax credit (HTC), and the chapter is

coauthored with Jarkko Harju and Tuomas Kosonen. HTC is a tax credit for con-

sumers using household services with the aim of increasing employment in the service

sector and curbing tax evasion. HTC aims to increase employment via increased

consumption, but there are a few prerequisites. First, the consumer needs to per-

ceive the HTC as a price decrease to respond to. This is not evident, as discounting,

mental budgets, and uncertainty regarding the system may affect the perceived value

of the credit. Then, the lower prices in turn need to induce greater demand for the

services, implying a positive demand elasticity with respect to prices. If the demand

for services increases due to the tax credit, then the next condition is to have supply

effects. Effectively, if supply is not elastic and the amount of services produced does

not increase, it leads to an increase in before HTC consumer prices, with no effect

on consumption or employment in the sector. The second aim of HTC, cutting down

on tax evasion, is based on customer reporting. The HTC incentivizes customers

to require receipts for their payments so that they can claim tax credits. To claim

the tax credit, the taxpayer is required to report the transaction to the tax author-

ity. This may increase the tax compliance of firms, for example through the fear of

cross-checking by the tax administrators leading to an audit.

We use reforms in the HTC system together with data from Finland and Sweden

to study how the HTC reaches its goals of increasing demand and curbing tax evasion.

In addition, we explore the distributive consequences of HTC. We use data on firm-

level monthly value added tax reports, annual income tax filings and individual-level

reports of the use of HTC obtained from the Tax Authorities in Finland and Sweden.

We focus on understanding the causal impacts of the HTC policies on consump-

tion of household services in Finland and Sweden. For that purpose, we utilize two

empirical settings to do this. First, we compare household service industries between

Sweden and Finland. The countries share similar cultures, apply similar income tax

rules, and most importantly, apply similar institutions regarding to HTC. We use

the adoption of the current HTC system for cleaning services in Sweden in July 2007

as variation, and Finland, which already had the HTC system in place, as a control

group. In addition, we study the responses to a change in the Swedish HTC system in

July 2009, in which the HTC system was reformed so that firms claimed the HTC on

behalf of consumers, making the tax credit’s effect on prices faced by consumers more

17

immediate. We use these sources of variation to study the effects of HTC tax credit

on the cleaning industry. We find no increase in the reported value of sales among

cleaning firms in Sweden relative to the Finnish firms after the introduction of HTC

for cleaning services in Sweden. The main identification assumption of our empirical

approach is that the value of sales among cleaning service firms follow each other

across countries. We find that the pre-reform trends are very similar, validating our

empirical approach, and mitigating the potential concern that the two groups would

not be comparable. Around 2007 there were no other concurring changes that could

have affected the results.

In our second setting, we study the renovation industry with a particular focus

on Finland and use other similar industries as a domestic control group. Finland

increased the amount of maximum tax credit from 1150 to 3000 euros for the reno-

vation industry in 2009, and our empirical strategy is to compare firms operating in

the renovation industry with our matched control group before and after 2009. We

use a domestic control group by matching Finnish firms from similar industries that

seem to follow similar economic trends prior to the reform. For the matching we

use a CEM algorithm that also weights the control group observations to match the

treatment group even more closely. The control sectors include e.g. car repair and car

retail. We do not find evidence of any response in sales of renovation services after

the increase in maximum HTC relative to the control group, suggesting negligible

demand elasticity with respect to the size of HTC, similarly as for the cleaning sector

results.

Finally, a descriptive analysis with the administrative data shows that a relatively

large share of individuals claiming HTC make costly mistakes in their reports to the

tax authority. This shows as a large excess mass of taxpayers bunching in a ”wrong”

threshold, without any other reason than their misunderstanding of the claiming

system. Our descriptive analysis also shows that the HTC is highly regressive – higher

income households use HTC to a much greater extent than lower income households,

and very poor households do not utilize HTC almost at all. In addition, as the HTC

is tax credit from income taxes, those who do not pay enough taxes cannot utilize

the full amount of HTC.

18

References

Philippe Aghion and Peter Howitt. A model of growth through creative destruction.

Econometrica, 60(2):323–351, 1992. ISSN 00129682, 14680262.

Annette Alstadsæter, Martin Jacob, and Roni Michaely. Do dividend taxes affect

corporate investment? Journal of Public Economics, 151(C):74–83, 2017.

Joshua D. Angrist and Jorn-Steffen Pischke. Mostly Harmless Econometrics: An

Empiricist’s Companion. Number 8769 in Economics Books. Princeton University

Press, 2009.

Wiji Arulampalam, Michael P. Devereux, and Giorgia Maffini. The direct incidence

of corporate income tax on wages. European Economic Review, 56(6):1038 – 1054,

2012. ISSN 0014-2921. doi: https://doi.org/10.1016/j.euroecorev.2012.03.003.

Alan J. Auerbach. Wealth maximization and the cost of capital. Quarterly Journal

of Economics, pages 433–446, 1979.

Alan J. Auerbach and James R. Jr. Hines. Perfect Taxation with Imperfect Compe-

tition. NBER Working Papers 8138, National Bureau of Economic Research, Inc,

February 2001.

Spencer Bastani and Hakan Selin. Bunching and non-bunching at kink points of

the swedish tax schedule. Journal of Public Economics, 109:36 – 49, 2014. ISSN

00472727.

Matthew Blackwell, Stefano Iacus, Gary King, and Giuseppe Porro. Cem: Coars-

ened exact matching in stata. The Stata Journal, 9(4):524–546, 2009. doi:

10.1177/1536867X0900900402.

David Bradford. Factor prices may be constant but factor returns are not. Economics

Letters, 1(3):199–203, 1978.

Raj Chetty and Emmanuel Saez. Dividend and corporate taxation in an agency model

of the firm. American Economic Journal: Economic Policy, 2(3):1–31, August 2010.

doi: 10.1257/pol.2.3.1.

19

Raj Chetty, John N. Friedman, Tore Olsen, and Luigi Pistaferri. Adjustment costs,

firm responses, and micro vs. macro labor supply elasticities: Evidence from dan-

ish tax records. Quarterly Journal of Economics, 126(2):749 – 804, 2011. ISSN

00335533.

European Comission. Taxation Trends in the European Union – Data for the EU

Member States, Iceland and Norway. Publication Office of the European Union,

2019. ISBN 978-92-76-00659-6.

Ryan Decker, John Haltiwanger, Ron Jarmin, and Javier Miranda. The role of en-

trepreneurship in us job creation and economic dynamism. Journal of Economic

Perspectives, 28(3):3–24, September 2014. doi: 10.1257/jep.28.3.3.

Michael P. Devereux and Rachel Griffith. Taxes and the location of production:

evidence from a panel of us multinationals. Journal of Public Economics, 68(3):335

– 367, 1998. ISSN 0047-2727. doi: https://doi.org/10.1016/S0047-2727(98)00014-0.

John DiNardo, Nicole M. Fortin, and Thomas Lemieux. Labor market institutions and

the distribution of wages, 1973-1992: A semiparametric approach. Econometrica,

64(5):1001–1044, 1996. ISSN 00129682, 14680262.

Clemens Fuest, Andreas Peichl, and Sebastian Siegloch. Do higher corporate taxes

reduce wages? micro evidence from germany. American Economic Review, 108(2):

393–418, February 2018. doi: 10.1257/aer.20130570.

Roger Gordon and Jeffrey Mackie-Mason. Tax distortions to the choice of organiza-

tional form. Journal of Public Economics, 55(2):279–306, 1994.

Jane G. Gravelle. Tax Havens: International Tax Avoidance and Evasion. National

Tax Journal, 62(4):727–753, December 2009.

Robert E. Hall and Dale W. Jorgenson. Tax policy and investment behavior. The

American Economic Review, 57(3):391–414, 1967. ISSN 00028282.

Arnold Harberger. The incidence of the corporation income tax. Journal of Political

Economy, 70, 1962.

Jarkko Harju and Tuomas Matikka. The elasticity of taxable income and income-

shifting: what is “real” and what is not? International Tax and Public Finance,

pages 1–30, 2016. ISSN 1573-6970. doi: 10.1007/s10797-016-9393-4.

20

Friedrich Heinemann, Michael Overesch, and Johannes Rincke. Rate-cutting tax

reforms and corporate tax competition in europe. Economics and Politics, 22(3):

498–518, 2010.

Christopher House and Matthew Shapiro. Temporary investment tax incentives: The-

ory with evidence from bonus depreciation. American Economic Review, 98(3):

737–68, 2008.

Henrik Jacobsen Kleven. Bunching. The Annual Review of Economics, 8, 2016.

Laurence Kotlikoff and Lawrence Summers. Tax incidence. In A. J. Auerbach and

M. Feldstein, editors, Handbook of Public Economics, volume 2, chapter 16, pages

1043–1092. Elsevier, 1 edition, 1987.

Claus Thustrup Kreiner, Søren Leth-Petersen, and Peer Ebbesen Skov. Year-end tax

planning of top management: Evidence from high-frequency payroll data. American

Economic Review, 104(5):154–58, May 2014. doi: 10.1257/aer.104.5.154.

Daniel Le Maire and Bertel Schjerning. Tax bunching, income shifting and self-

employment. Journal of Public Economics, 107(1):1–18, 2013.

Giorgia Maffini, Jing Xing, and Michael Devereux. The impact of investment in-

centives: Evidence from uk corporation tax returns. American Economic Journal:

Economic Policy, 11(3):361–89, 2019.

James Mirrlees, Stuart Adam, Tim Besley, Richard Blundell, Stephen Bond, Robert

Chote, Malcolm Gammie, Paul Johnson, Gareth Myles, and James M. Poterba.

Tax by design. Oxford University Press, 2011. ISBN 978-0-19-955374-7.

Eric Ohrn. The effect of corporate taxation on investment and financial policy: Ev-

idence from the dpad. American Economic Journal: Economic Policy, 10(2):272–

301, May 2018. doi: 10.1257/pol.20150378.

Thomas Piketty and Gabriel Zucman. Capital is Back: Wealth-Income Ratios in Rich

Countries 1700–2010 *. The Quarterly Journal of Economics, 129(3):1255–1310,

05 2014. ISSN 0033-5533. doi: 10.1093/qje/qju018.

Paul M. Romer. Endogenous technological change. Journal of Political Economy, 98

(5):S71–S102, 1990. ISSN 00223808, 1537534X.

21

Emmanuel Saez. Do taxpayers bunch at kink points?. American Economic Journal:

Economic Policy, 2(3):180 – 212, 2010. ISSN 19457731.

Hans-Werner Sinn. The vanishing harberger triangle. Journal of Public Economics,

45(3):271 – 300, 1991. ISSN 0047-2727. doi: https://doi.org/10.1016/0047-

2727(91)90029-2.

Joel Slemrod. Do taxes matter? lessons from the 1980s. American Economic Review,

82(2):250–256, 1992.

Matthew Smith, Danny Yagan, Owen Zidar, and Eric Zwick. Capitalists in the

twenty-first century. The Quarterly Journal of Economics, 134(4):1675–1745, 2019.

Danny Yagan. Capital tax reform and the real economy: The effects of the 2003

dividend tax cut. American Economic Review, 105(12):3531–63, December 2015.

doi: 10.1257/aer.20130098.

Eric Zwick and James Mahon. Tax policy and heterogeneous investment be-

havior. American Economic Review, 107(1):217–48, January 2017. doi:

10.1257/aer.20140855.

22

Chapter 2

Dividend Tax Thresholds and Extreme

Bunching

Abstract

In this paper, I study how business owners respond to dividend taxes. I use

administrative data on all privately held Finnish corporations and their main

owners in 2006–2016. By using tax schedule discontinuities and changes in the

schedule as variation, I find exceptionally clear dividend payment responses to

tax rates, implying taxable income elasticities from 0.5 up to 3.6. Descriptive

evidence on the asset composition of firms indicates that a notable part of the

payment response is due to inter-temporal income smoothing, while changes in

the tax schedule did not cause significant real responses in output or invest-

ment. Hence, dividend taxes may capitalize into share values, as earnings are

left in the firms to avoid dividend tax. In addition, studying the income com-

position of owners around tax changes reveals income-shifting between wage

and dividends with no effect on gross income received from the firm.

JEL classification codes: G20, H21, H24, H25

Keywords: Dividend taxation, dividend payments, real investment, income-

shifting, bunching

1 Introduction

Understanding the mechanisms of how business owners respond to dividend taxation

is essential in planning a good income tax scheme. While reasons of equity entail

taxing entrepreneurial income as progressively as labour income, efficiency considera-

tions may suggest the opposite. Dividend taxes reduce the return on invested capital

23

and the owner’s own work, hence decreasing incentives for new investments and ex-

erting effort. However, business owners have many channels for adjusting their tax

burden - i.e. tax planning and evasion - and several channels to fund investment, so

the distortions could also be small. In this paper, I study how Finnish firms and firm

owners respond to dividend taxation. I use discontinuities in the owner’s dividend

tax schedule as well as changes in the tax rates to empirically study the importance

of various response channels.

Dividend tax literature suggests that dividend tax does not enter the marginal

cost of investment when it is funded with retained earnings or debt (Auerbach 1979;

King 1974). The opportunity cost of investing the retained earnings is to distribute

the profits as dividend. In either case you pay the dividend tax - either now or

in the future - and therefore the dividend tax rate cancels out from the cost of

capital. Therefore, the so-called ’new view’ of the dividend tax literature suggests

that marginal investment is funded with retained earning (or debt) and dividend tax

does not enter the marginal investment decision.

Under the ’new view’, dividend payment may simply respond to inter-temporal

incentives to pay dividends, meaning that dividend taxes are avoided by retaining the

profits. As a response, taxes on dividends may be solely capitalized into share values

but not affect the decision to reinvest. Moreover, an increase in dividend payments as

a response to a dividend tax cut may simply reflect a firm’s response to inter-temporal

incentives and so it pays dividends while tax rates are relatively low1. Yagan (2015)

shows empirical evidence of a dividend payment windfall following a dividend tax cut

in US, supporting this theory.

However, a part of the literature, the ’old view’, suggests that dividend taxes affect

the investment choices of firms even if the investment is funded with retained earnings

(Auerbach, 1979). The reason may be that shareholders do not consider retained

earnings as valuable as paid-out profits due to asymmetric information (principal-

agent conflicts), as dividends signal the true value of the firm. Sinn (1991) argues that

while the new view can apply for mature firms with accumulated capital, for starting

firms in a growth phase, dividend taxes may retard growth. The reason is that the

firm raises less new equity to fund new capital, which distorts the investment choice.

Sinn (1991) considers that for a mature firm the ’new view’ holds and the dividend

tax is capitalized into share values. Moreover, when dividend taxation is non-linear,

1For example Zodrow (1991) describes the capitalization mechanism in more detail.

24

neutrality of investment with respect to dividend tax rate does not necessarily hold

as pointed out by (Kari and Laitila, 2014).

The model in Chetty and Saez (2010) predicts that dividend taxation may also

amplify principal-agent conflicts of interest, as retaining earnings leaves more cash

under the control of managerial choices, and thereby disincentivizes the close moni-

toring of the managers. This may lead to unproductive investments using retained

earnings. The variety in existing theory literature underlines the range of effects that

dividend taxation may create.

The Finnish dividend tax schedule provides exceptionally large incentives for firms

to respond. To begin with, the owners of the privately held corporations studied in

this paper can quite freely choose whether to receive income from the firm as dividend

(taxed as profit with corporate tax and at the owner level with dividend tax) or pay

wages (only progressive earned income tax on wages). The dividend tax schedule in

Finland includes deduction thresholds, effectively causing clearly lower marginal tax

rates for certain amounts of dividend income in comparison to eg. labor income. The

dividend tax rate jumps notably at a threshold that is set first at 9 percent (2006–

2013) then at 8 percent (2014–) return on net assets. Moreover, there is a monetary

threshold for dividends exempted from most of the capital income tax to alleviate the

double taxation of corporate profits. These discontinuities create strong incentives

and have changed several times over the past decade.2 Using administrative data, I

study the effects of the two different dividend tax schedule discontinuities and the

changes in them in 2006–2016.

In the first part of this paper, I study the incidence of dividend payments at the

thresholds using the bunching method, developed by Saez (2010). The idea of the

method is that discontinuities in the tax schedule create convex kink points in the

budget set of the owner. If the owners respond to a tax rate discontinuity, there

should be bunching of observations at the kink point. Indeed, that is what I observe:

I find exceptionally clear dividend responses to the dividend tax rate thresholds. The

excess mass at each threshold is from 6 to 20 times more than the estimated counter-

factual mass. Then, I use the excess mass at the threshold to estimate the elasticity of

taxable dividend income with respect to the net of marginal tax rate. I find elasticities

ranging from 0.5 at the monetary thresholds to 3.6 at the net asset thresholds. This

2For example, the marginal tax rate on dividends (including corporate taxes) jumped from 28%to 40.5% at 90,000 euros between 2006 and 2011.

25

implies that a 1 percent increase in the net of dividend tax rate increases taxable

dividend income by 0.5-3.6 percent, which is a very large response.

The elasticity of taxable income is a useful tool to capture all the responses cre-

ated by the threshold and it allows us to compare the results with earlier literature

on business owners’ responsiveness to taxation. In this study, the high elasticity pa-

rameters also highlight that business owners are well informed about the tax schedule

and find it easy to adjust accordingly. However, the elasticity parameter obtained

using the bunching method does not compare to the structural costs of taxation as it

captures tax planning and other channels that affect the dividend pay-out.3 Taxable

income in other income bases or in the future may increase as a response to a dividend

tax increase, in which case the effect on total income would be smaller. These other

potential response mechanisms driving the bunching evidence are also examined in

the latter part of this paper.

The elasticity estimates found in this paper are large compared to those in the

earlier bunching literature studying business owners’ responsiveness to income taxes.

Kreiner et al. (2014) and Kreiner et al. (2016) use the bunching method to study year-

end income-shifting in Denmark. They find that especially high-income individuals,

such as managers, shift income around the year’s end when the tax rates are to change

the next year. However, the observed elasticity (in Kreiner et al. 2016) estimated

with the bunching method is only 0.1 and entirely driven by year end income shifters.

Bastani and Selin (2014) study kinks in the Swedish income tax schedule and find

no bunching, not even a hump for wage earners, whereas the self-employed bunch

clearly. However, compared to bunching in the Finnish dividend tax schedule, the

excess mass is small, with elasticity estimates of around 0.02 for broader groups of self-

employed individuals and around 0.07 for the ”purely self-employed” (who only earn

income from the firm they own). Chetty et al. (2011) study bunching in the Danish

income tax schedule. They, too, find that business owners bunch more strongly. The

estimated (short run) elasticities are 0.01 for wage earners and 0.1-0.2 for the self-

employed. The main difference of this paper from the earlier bunching literature is

that I focus entirely on business owners and the dividend tax schedule, whereas the

earlier papers focus on wage income. Thereby, it is likely that the very large bunching

is driven by dividend-specific features.

3Kleven (2016) provides a good introduction to bunching and on how frictions and tax planninglimit the use of the bunching elasticity as a structural parameter to estimate the effects of policies.

26

In the second part of the paper, I examine the mechanisms driving bunching

at the thresholds. I study real economic effects, using changes in the dividend tax

thresholds. Moving the dividend tax threshold brings new firms into the range of the

higher marginal tax rate, but I find no statistically significant responses in investment

or output.4 While no real effects are found, the evidence presented in this study

suggests that most of the bunching is driven by tax planning. Further analysis on

the asset structures of the firms suggests that a notable part of the payment response

is due to inter-temporal income smoothing, as the balance sheet information shows

firms at the thresholds accumulating financial assets in the firm. Hence, owners avoid

the higher tax bracket by retaining earnings in the firm, which is also predicted by

the ‘new view’ as the capitalization of the dividend tax (Auerbach, 1979). Retaining

profits has several tax benefits. In addition to avoiding the higher tax bracket, the

retained earnings increase the firm’s value by increasing its net assets, and therefore

allows for a higher amount of dividend to be distributed with the lower capital income

tax in the future as the tax schedule depends on the net assets of the firm. Also, some

forms of capital income are taxed more lightly when received by a firm, so saving by

investing through a firm is lucrative. This is likely to further boost the capitalization

of dividend taxes into share values.

Finally, by studying the income composition of firm owners around the time of

tax changes, I observe that owners engage in income-shifting across wage income and

dividends to minimize their tax burden. Income shifting across income tax bases

means adjusting income suitably to minimize the total tax burden.

There is some existing empirical literature on how firms respond to dividend tax-

ation. Chetty and Saez (2005) study the US dividend tax cut of 2003 and show that

dividend payments responded massively to the tax cut. The effect was especially

high in firms with strong shareholders or owner-executives. Yagan (2015) extends

that research by showing that despite the notable effect on dividend payments, there

was no increase in investment. Alstadsæter et al. (2015) find no increase in aggregate

investment in response to a dividend tax cut in Sweden. However, they find that

dividend taxes distort the allocation of investment, so that as a response to the tax

cut the investment of cash-constrained firms increased relative to cash-rich firms, in

line with the principal-agent conflicts predicted by the model in Chetty and Saez

(2010). This is explained by higher dividend payouts in cash-rich firms and better

4However, these results cannot rule out global effects affecting the whole distribution of the firms,as I study these responses locally.

27

access to external equity in cash-constrained firms. Harju and Matikka (2016) show

that business owners in Finland actively shift income between tax bases, specifically

wage and dividends, and Le Maire and Schjerning (2013) shed light on the business

owners’ ability to use retained and withdrawn earnings to adjust their taxation. Con-

sidering this income smoothing, Le Maire and Schjerning (2013) extend the bunching

method to extract the real elasticity from the bunching evidence of business owners

on business income tax thresholds.

This study explores empirically how theories of corporate income tax, specifically

dividend taxation, match the empirical evidence. This paper lends support to the

modest investment elasticity of dividend taxes predicted by the ’new view’ and shown

in Yagan (2015) and in (Alstadsæter et al., 2015). In addition, this paper shows that

dividend taxes capitalize into share values as predicted by the ’new view’. In other

words, as a progressive dividend tax schedule creates incentives for income smoothing,

it leads to the accumulation of assets in the firm. The only previous empirical paper

covering this topic, albeit from a different angle, is that by Le Maire and Schjerning

(2013).

This paper contributes to several areas of public finance. First, it contributes

to the bunching literature studying local effects around tax rate thresholds. I show

sizeable responses to the dividend tax schedule and I provide detailed information

on the mechanisms driving the bunching. Second, this paper contributes to the

literature on dividend taxation by showing that a dividend tax alteration is a weak

tool for incentivizing real economic activity and investment, and mainly affects the

retained earnings in line with the ’new view’. Thus, changes in dividend tax schedule

seem to have mainly distributional effects. Third, this paper extends the literature on

income-shifting by firm owners by showing how actively Finnish business owners shift

income both in time and across income bases to avoid ending up above a marginal

tax rate threshold.

The rest of the paper is organized as follows. Section 2 outlines the institutions

and the data. In section 3, I present the payment responses to dividend taxation using

the bunching method and estimate the corresponding elasticity. Section 4 discusses

what the payment responses imply, covering real responses, income-smoothing and

income-shifting. Section 5 concludes.

28

2 Institutions and Data

2.1 Institutions

Table 1: Dividend tax schedule in Finland

A. Dividend tax thresholds

Net assetYears Kink threshold

2006–2011 90,000e 9 %2012–2013 60,000e 9 %2014–2016 150,000e 8 %

B. Owner level tax burden around the tax thresholds

Effective marginal tax rate

Below net Above netYears asset threshold asset threshold

2006–2011 26% 26–∼55%2012–2013 Below 24.5% 24.5–∼55%2014–2016 kink 26–26.8% 20–∼55%

2006–2011 Above 40.5% 26–∼55%2012–2013 kink 40.36% 24.5–∼55%2014–2016 40.4–43.12% 20–∼55%

The earned income tax rate varies depending on the taxpayer’s income and municipality. Boththe municipal and government tax schedules change nearly every year. The lowest government taxrate has been zero over the whole period and, with deductions in the municipal tax for low incomeearners, the aggregate earned income tax rate has also been close to zero at the low end of theincome distribution. The highest overall marginal earned income tax rate has been circa 55 %.Overall government tax rates on earned income have been decreasing over the research period of2000-2013, especially for low and middle income earners. However, municipal income tax has beenincreasing; in 2000, the average rate was 17.7 %, but in 2013 it was 19.4 %. The municipal incometax varies across municipalities; in 2015 it ranged from 16.5 % to 22.5 %.

There are two income tax schedules in Finland. Personal capital income, such

as capital gains and rental income, is taxed at a nearly flat capital tax rate. Other

income, such as wage and social benefits, is taxed with a progressive earned income

tax rate schedule. The ∼30 % capital income tax is lower than the highest marginal

tax rates on earned income, ∼50 %, aiming to boost capital mobility and to respond

to international tax competition. Owners of privately held firms can quite freely

choose whether to receive their income as wages or dividends, or leave income in the

29

firm as retained earnings.5

To prevent extensive income shifting, the dividend tax rate for privately held

corporations depends on the level of net assets of the firm: only the amount of

distributed dividends below a predetermined rate of return on the firm’s net assets,

8% since 2014, are taxed at the lower capital income tax rate. Moreover, below the

net asset threshold, part of the capital income tax is deducted in order to reduce

the double taxation of distributed profits, since the overall tax burden of distributed

dividends includes both the flat corporate tax rate (20% from 2014 onward) and

personal dividend taxes. In 2006–2011, dividends below both the net asset threshold

and the monetary threshold were taxed at an effective tax rate of 26 %. Dividend

payments above the net asset threshold are taxed at the progressive earned income

tax rate. However, the tax is applied only to 75 % of the excess dividends, to reduce

double taxation. The earned income share of the dividends is added to the other

earned income of the owner when calculating the effective tax rate.

Earned income taxation in Finland includes a progressive government tax, a flat

municipal income tax, and pension and social security contributions. Both govern-

ment 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 % when excluding the payroll tax paid by the employer6.

In calculating the marginal tax rate for earned income, I calculate the tax rate for one

extra euro of the particular income type. I exclude the payroll tax7, since for most

business owners, the payroll contribution is not defined by wage sum, but is based

on so-called entrepreneur’s labor income, which is largely decided by the owner8.

Thus, the marginal payroll tax is generally not affected by an additional euro to gross

income.

5The Finnish dividend tax system varies depending on the organizational form of the company.In this study, I focus on privately held corporations that are limited companies owned by a singleperson or a group of individuals. The privately held corporation is the most common corporate formin Finland covering nearly half of all firms.

6More details of the earned income tax schedule are described in the Appendix.7Employer’s social contributions (tyonantajan sairausvakuutusmaksu, tyoelakevakuutusmaksu,

tyottomyysvakuutusmaksu, ryhmahenkivakuutusmaksu) and employee’s social contributions(tyoelakevakuutusmaksu, tyottomyysvakuutusmaksu, vakuutetun sairausvakuutusmaksu).

8This so called YEL-system, where the entrepreneur sets the labor income level, applies to allself-employed persons who are taxed according to the self employed person’s pension act, implyingbusiness owners who, alone or together with family members, own at least 50 percent of their firmor hold a leading position in the firm and own over 30 percent of the company’s shares. These arethe majority of the owners studied in this essay.

30

Table 1 collects the features of the dividend tax schedules in use 2006–2016. Panel

A compiles the thresholds in the tax schedule and panel B displays the effective tax

rates around each threshold during each period. This complex system creates a

challenging tax minimization puzzle for the owner. There have been three monetary

thresholds, at 90,000, 60,000, and 150,000 euros, and two net asset thresholds, 9%

and 8%, during the period 2006-2016. The first column in Table 1 B features the

marginal tax rates below and above the monetary kink, for dividends below the

net asset threshold. For example, from 2006 to 2011, the effective tax rate below the

monetary threshold was 26 % as capital tax was fully exempted, and above the 90,000-

euro-kink, the effective tax rate rate was 40.5 %9. The marginal tax rate above the net

asset threshold in the second column depends on the owner’s other personal income,

as dividends above this threshold face the progressive earned income tax schedule,

with the highest rates around 50%. Figure 1 visualizes the thresholds in marginal

tax rates. For an individual firm owner, the whole region is not available, but the

firm’s net assets define a restriction, which slices the three-dimensional dividend tax

schedule. For example 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.

The thresholds described in the tax schedule and the amendments to them create

variation that enables me to study the effects of dividend taxes. I study bunching

caused by both the monetary and the net asset threshold, to estimate the dividend

tax elasticities. Then, I use the changes in the tax schedule to study the mechanisms

driving the elasticity.

2.2 Data

I use firm- and owner-level tax filing data that cover all privately held Finnish corpo-

rations. The data cover the years 2006–2016 and three different schedules in use. The

data are obtained annually from the Finnish tax administration and are maintained

by VATT, the Institute for Economic Research. Annual firm data are matched with

data on the main owners of the company and combined into a panel. The data include

90.26+(1-0.26)*0.7*0.3. Above the monetary threshold the capital tax rate has been applied to85 % of the excess dividend since 2014, and before 2014 to 70 %.

31

Figure 1: Marginal tax rate for dividends 2006–2011

Dividend per net assets

90k

9 %

Tax %

Dividend

26 %

Note: This graph describes the thresholds in 2006-2011, when the kink was at 90,000 euros andnet asset threshold at 9 percent. Above the net asset threshold, the owner pays earned income taxfor 70 % of the income (85% since 2014) in addition to the corporate tax. The tax rate above thenet asset threshold is estimated as a mean of the actual marginal earned income tax rates in each5000-euro-dividend bin.

information on dividends and wages paid to the owner, turnover, net assets, and new

investment by the firm. The detailed owner-level tax data allow me to calculate the

marginal tax rates for various forms of income. Tables 2 and 3 summarize the key

variables in the data, with Table 2 describing the pooled data covering all years in

the panel and Table 3 describing yearly summary statistics for the years 2006, 2011

and 2016. Turnover refers to annual sales of the firm, profit is the taxable income,

net assets refer to the book value of assets after depreciation and investment refers to

additions to assets, such as newly installed fixed capital. The owner level dividends

and wage refer to those received from the corporation studied, i.e. if the owner re-

ceives wages or dividends from other firms, those are not included in this value. The

data include more than 600,000 observations during the research period and 113,835

distinct firms.10

Figure 2 shows the dividend payment distributions during the three dividend tax

10The owner can postpone cashing in the dividends from the firm. Thus, the dividend tax ispaid according to the tax rate of the year when the dividend is cashed, not based on the year ofdistribution of dividend. Therefore, some of the owners have several dividend observations from thesame company and year. As a solution, dividend observations from an owner-company pair in asingle year have been aggregated.

32

Table 2: Summary statistics of the data 2006-2016

Firm levelmean sd p50

Turnover 1074031 8470531 210749

Profit 99678 4566064 15125

Net Assets 639844 8057283 119400

Profit/Net assets 0.22 80.96 0.16

Investment 54562 672584 1773

Owner levelmean sd p50

Dividends 25568 138318 8500

Wage 22931 28290 15660

Observations 641558

Note: This table provides the summary statistics for the whole pooled panel data covering years2006–2016. Turnover refers to annual sales, profit is the taxable income of the firm, net assets referto book value of assets after depreciation and investment refers to additions to depreciating assets,such as newly installed fixed capital. Dividends and wage are the main owner’s income from thefirm. Each firm has only one main owner in the data. The owner with highest share of stock isconsidered the main owner.

schedules studied. The main interest of this paper lies in the highest spikes, which are

driven by the thresholds. The figure also shows clear round number bunching, sug-

gesting that the dividend payout choice is not random and there is some behavioural

aspect to it. In the following section, I describe how to use this bunching evidence to

estimate the elasticity of taxable income, while taking into account the round number

effects.

3 Dividend Payment Responses

3.1 Bunching method

I estimate the extent of excess mass and the according elasticity of taxable income

with the bunching method, developed by Saez (2010)11. 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 higher the elas-

11Kleven (2016) provides an excellent review of the method and its indications.

33

Table 3: Summary statistics of the data 2006, 2011 and 2016

Firm level2006 2011 2016

mean sd p50 mean sd p50 mean sd p50

Turnover 1125410 8381004 241015 992964 7120131 202573 1156016 11536122 201493

Profit 104884 584402 21222 79296 450167 14025 170311 14030253 15114

Net Assets 501911 4255174 104207 560925 4969151 112843 849292 14312194 147004

Profit/Net assets 0.31 1.24 0.23 0.25 0.88 0.16 0.94 163.05 0.13

Investment 54236 329184 2897 47634 269958 1732 58389 639241 1300

Owner level2006 2011 2016

mean sd p50 mean sd p50 mean sd p50

Dividends 21769 75496 7624 27752 136389 8500 28943 284546 9642

Wage 18887 22926 13500 23055 27778 16414 26792 32143 19360

Observations 49101 59947 62589

Note: This table provides the summary statistics for the data in 2006, 2011 and 2016. Turnoverrefers to annual sales, profit is the taxable income of the firm, net assets refer to book value of assetsafter depreciation and investment refers to additions to depreciating assets, such as newly installedfixed capital. Dividends and wage are the main owner’s income from the firm. Each firm has onlyone main owner in the data. The owner with highest share of stock is considered the main owner.

34

Figure 2: Dividend payment distributions during the three tax schedules (nominal)

0.0

2.0

4.0

6.0

8.1

Frac

tion

of fi

rms

40000 60000 80000 100000 120000 140000 160000Dividend bin

2006−2011 2012−2013 2014−2016

Note: This figure plots the distribution of dividend payments to the main owners during threedividend tax schedules. The vertical line shows the fractions of firms in each 1000-euro dividend bin.In addition to round number bunching during each schedule, there is a clear spike at the prevailingmonetary threshold. In 2006–2011, the monetary threshold was at 90,000 euros, in 2012–2013, thethreshold was at 60,000 euros and in 2014–2016 it was at 150,000 euros.

ticity, the more strongly taxable income responds to a change in the tax rate. The

bunching method uses the excess mass at a tax schedule discontinuity to estimate the

corresponding elasticity of taxable income.

Intuitively, the bunching method works as follows. The owner of a company

withdraws dividend income from the firm until the marginal disutility of the payment

equals the marginal utility of it. Marginal disutility can be thought of as the cost

of the owner’s effort or the pre-tax return to invested capital, captured by the gross

dividends paid to the owner. Marginal utility is the after-tax dividend income. The

owners should bunch at the convex kink points of the budget set if they respond

to these tax rate discontinuities. Then, the dividend tax elasticity is recovered by

relating the excess mass in the dividend distribution to the change in the dividend

tax rate at the kink point.

To measure excess mass, I first estimate a counter-factual distribution that de-

scribes what the dividend distribution would approximately be in the absence of the

kink point. The counter-factual distribution is estimated using a seventh-order poly-

nomial excluding observations near the kink. First, I will explain how to estimate the

35

counter-factual distribution around the monetary threshold, and then how I estimate

the counter-factual distribution around the the net asset threshold.

Counter-factual distribution around monetary kink

The counter-factual distributions around the monetary kink points are estimated as

C0j =

p∑i=0

β0i · (Zj)

i + ρ · 1[Zj

r∈ N

]+ εj, Zj /∈ [−R;R] , (1)

where C0j is the estimate of the counter-factual distribution in each bin j with div-

idend income Zj. β0i are the regression estimates, and p denotes the degree of the

polynomial. ρ in the second term captures the round number fixed effect that is

observed in figure 2. [−R;R] is the excluded range of the distribution, which denotes

the area where the kink point affects the behavior of the owners. 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 [−R;R] nor the

order of the polynomial.

Counter-factual distribution around net asset threshold

I estimate the counter-factual distribution around the net asset threshold following

Equation 2.

C0j =

p∑i=0

β0i · (Zj)

i +

∑Rj=−R Cj

2A+ 1+ εj, Zj /∈ [−R;R] , j ∈ [−A;A] (2)

The basic principle is the same as in Equation 1. 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 counter-factual distribution match that of the realized distribution.

Thus, 2A+ 1 is the number of bins in the region [−A;A] of the distribution studied.

For this dividends per net assets distribution, there is no need to consider round

number bunching.

36

Excess mass and elasticity estimate

The sum of the excess observations in the bunching range is

R∑j=−R

Bj =R∑

j=−R

(Cj − C0

j

). (3)

The estimate of excess bunching b is then the estimated excess mass around the kink

relative to the average density of the counter-factual dividend distribution between

−R and R

b =

∑Rj=−R Bj∑R

j=−R C0j /(2R + 1)

. (4)

Finally, the excess bunching can be turned into an elasticity estimate. The elasticities

at the kink points are estimated as

εD =dD

d(1− τ)

1− τ

D=

b

D∗ · log(

(1−τD)(1−τD−�τD)

) . (5)

D denotes dividend income, τ the dividend income tax rate that jumps at a kink

point D∗ from τD to τD + �τD. 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.

3.2 Bunching evidence

Figure 3 provides the results estimated with the tax schedule in place in 2006–2011,

2012–2013, and 2014–2016, and with the pooled data covering all firm-year observa-

tions of the period in question. The first plot on the left depicts the evidence for 2006–

2011, when the threshold was at 90,000 euros. The horizontal axis is the dividend

amount relative to the e90,000 kink. The frequency of firms in each 1000-euro-bin is

37

Figure 3: Bunching at the monetary threshold

010

0020

0030

0040

00Fr

eque

ncy

−25 −15 −5 5 15 25Distance from the kink (1000 euros)

Excess mass: 8.34 (1.192), Elasticity: .425 (.061)90k kink 2006−2011

010

0020

0030

00Fr

eque

ncy

−25 −15 −5 5 15 25Distance from the kink (1000 euros)

Excess mass: 6.703 (.939), Elasticity: .474 (.066)60k kink 2012−2013

050

010

0015

00Fr

eque

ncy

−25 −15 −5 5 15 25Distance from the kink (1000 euros)

Excess mass: 19.598 (2.108), Elasticity: .536 (.058)150k kink 2014−2016

Observed Counterfactual

Note: These graphs plot the actual distribution of observations, represented by the solid line, andthe counter-factual distribution, represented by the dashed line, in 1000-euro-bins around the 90,000-euro-threshold in 2006–2011, the 60,000-euro-threshold in 2012–2013 and the 150,000-euro-thresholdin 2014–2016. The vertical solid lines show the bunching region. The estimated excess mass and thecorresponding elasticity estimate are reported above each graph together with the standard errors.

Figure 4: Bunching at the net asset threshold

050

000

1000

0015

0000

2000

00Fr

eque

ncy

−.08 −.06 −.04 −.02 0 .02 .04 .06 .08Distance to 0.09 threshold

Excess mass: 16.022 (.502), Elasticity: 3.603 (.006)9 % kink 2006−2013

020

000

4000

060

000

8000

0Fr

eque

ncy

−.08 −.06 −.04 −.02 0 .02 .04 .06 .08Distance to 0.08 threshold

Excess mass: 14.301 (.751), Elasticity: 3.6 (.023)8 % kink 2014−2016

Observed Counterfactual

These graphs plot the bunching mass at the 9 % net asset threshold in 2006–2013 and at the 8 %net asset threshold in 2014–2016. The elasticities are first estimated for each buncher individuallybased on their respective tax rates around the kink using the aggregate excess mass. The finalelasticity reported above the graph is a mean of all the individual elasticities. The capital incometax rate below and above are chosen using only dividend income, that is, the higher capital incometax brackets in later years are only used when taxable dividends below net asset threshold exceedthe monetary limit (eg. 2015–2016: 30,000e).

38

shown on the vertical axis. The solid line in the figure is the actual observed dividend

distribution in the region. The dashed line is the estimated counter-factual distri-

bution, which takes into account bunching at round numbers and excludes the area

near the kink. The vertical lines around the kink show the bunching range [−R;R]

that is used to estimate the excess mass and elasticity. As predicted, a substantial

excess mass takes place at the tax kink, the excess mass is more than eight times the

counter-factual, and the corresponding elasticity is 0.43. Bunching at the later mon-

etary discontinuities at 60,000 euros in 2012–2013 and at 150,000 euros in 2014–2016

is as large. The corresponding elasticities are 0.47 and 0.54 respectively.

Figure 4 shows the bunching results at the net asset thresholds. The horizontal

axis is now the dividend amount relative to the firm’s net assets. The estimated

elasticity of taxable dividend income is 3.6 in both estimations. The elasticity esti-

mate reported is the mean elasticity of individual elasticities estimated using personal

tax rates and the excess mass. Even though the excess mass at the threshold does

not differ massively in comparison to the monetary kinks, the elasticity estimate is

clearly larger. Mathematically, this is due to the lower tax difference for many of the

taxpayers in comparison to the tax rate difference at the monetary kink. The high

elasticity is likely to represent the additional incentives for inter-temporal income

shifting created by the threshold. Even though the owner cannot affect the marginal

tax rates around the threshold, it can affect the position of the threshold in euros.

That is, retaining earnings in the firm increases the net assets of the firm, thereby

allowing for a larger amount of dividends to be distributed in the future at a lower tax

rate12. I will discuss this further in sub-section 4.2. In addition to the inter-temporal

income shifting, the owner can engage in income shifting between wage and dividends

and other tax planning or even evasion. Income shifting is covered in sub-section 4.3.

Considering both of these tax planning channels captured in the bunching response,

inter-temporal and tax base income shifting, the elasticity estimates based on only

one income base should not be used as structural elasticity estimates that capture

the real economic effects. Table 4 collects all the elasticity estimates.

12The marginal tax rate in the earned income tax bracket could also be less clear for the ownerdue to the complexity of the earned income tax schedule.

39

Table 4: Elasticity estimates

ElasticityYears Threshold estimate

2006–2011 90,000e 0.425 (0.061)2012–2013 60,000e 0.474 (0.066)2014–2016 150,000e 0.536 (0.058)2006–2013 9% of net assets 3.603 (0.006)2014–2016 8% of net assets 3.6 (0.023)

Note: This table collects together all elasticities estimated with different thresholds and dataperiods.

4 Mechanisms

4.1 Real effects

Dividend taxes reduce the return on invested capital and the owner’s own work effort,

hence decreasing incentives for new investments and exertion. An ongoing debate in

the dividend tax literature is how strongly dividend taxation distorts investment.

The ’new view’ suggests that there is no significant effect on corporate investment

(Auerbach 1979 and King 1974). The ’old view’ in the theoretical literature, starting

with Feldstein (1970) and Poterba and Summers (1985), is that dividend taxation

causes substantial real responses through the cost of corporate investment. Could

the real economic effects claimed by the old view be driving some of the bunching

responses? This is difficult to examine directly. However, I can use the changes in the

dividend tax parameters to see how those facing a higher or lower marginal dividend

tax rate respond to the tax changes. Dividend payouts are an endogenous choice, so I

cannot use the dividend payments, as such, to study the real effects. Instead, I utilize

an instrumental variable and difference-in-differences set-up to study the effects of

dividend tax changes on real outcomes, namely investment and output.

As the dividend tax rate depends on the net assets of the firm, the main owners

of firms with equal net assets are taxed on dividends only, depending on the owner’s

share of the net assets. Therefore, I use a sub-sample of the panel data based on the

net assets of the firm, and the net asset share of the main owner as an instrument.

The intuition is that moving the dividend tax threshold brings new firms into the

range of the higher marginal tax rate. If the marginal tax rate distorts investment,

40

Table 5: Summary statistics of the restricted sample (year 2011)

Treated Controlmean sd p50 mean sd p50

Turnover 1541055 2556638 802944.5 1897639 2383413 1249627

Net Assets 846457.5 99281.63 842540.5 829472.9 97023.8 813990

Dividends (main owner) 80094.25 50810.17 71601.25 45592.34 35839.96 37524

Investment 76105.05 221195.7 9774.46 80362.66 161791.7 17920.2

Labor costs 329180.9 471996.3 173881.7 511846.9 587571.6 346716.9

Variable costs 1130710 2181940 430710.5 1242189 1981976 648814.7

Employees 10.69 27.59 4 14.84 24.60 8

Observations (2011) 1038 (651*) 1394Observations (total) 8478 10059

* 651 is the number of firms paying dividend in the affected region

Note: This table provides the summary statistics for the Difference-in-difference data in 2011.Turnover refers to annual sales, net assets refer to book value of assets after depreciation, divi-dends are the main owner’s dividend income from the firm, and investment refers to additions todepreciating assets, such as newly installed fixed capital. Labor costs include wage and payrolltaxes paid by the firm and variable costs other input costs such as material and intermediate goods,number of workers include number of employees in the firm in 2011. Each firm has only one mainowner in the data. The owner with highest share of stock is considered the main owner.

there should be some response in the real outcomes of these new firms.

In 2012, the monetary threshold for a higher marginal dividend tax rate was

reduced from 90,000 euros to 60,000. I restrict the data to firms with net assets

between 666,666.67–1,000,000e in 2011, just before the tax change. Hence, the data

are a balanced panel based on the 2011 net asset position. The main owners of

firms of this size face the dividend tax increase only if the owner’s share of the net

assets is high enough. Thus, in the first stage the treated are the firms with owners

whose ownership share of the firm’s net assets exceeds 666,666.67e, which implies

that the maximum capital income dividend is between 60,000–90,000 euros. Thus,

they faced a marginal dividend tax increase of 14.36 percentage points. The control

group are firms whose main owner’s net assets share is below 666,666.67e. For them

the maximum capital income dividend was already below 60,000 euros. Thus, their

marginal dividend tax rate decreased by 1.5 percentage points.

The net asset position, which enables a firm to pay dividends in the lowest tax

bracket, does not imply that the firm pays the maximum capital income dividend for

the owner. However, more than 60 % of the first stage treated firms did pay dividends

between 60,000–90,000e, making this a suitable instrument (table 5).

41

Table 6: Difference-in-differences results of the 2012 tax change

Turnover (log) Variable costs (log) Investment (log) Investment per lagged capital

α2(Treat× Post) -0.015 0.003 -0.100 -0.101

(0.041) (0.050) (0.185) (0.276)

Firm fixed effects X X X X

Year fixed effects X X X X

Constant 13.828*** 12.733*** 6.807*** 0.656***

(0.020) (0.025) (0.109) (0.164)

r2 0.016 0.005 0.019 0.001

N 16857 15367 18537 13376

* p < 0.05, ** p < 0.01, *** p < 0.001

Note: This table reports the regression results estimated following Equation 6. The dependentvariables in the specifications are logarithmic transformation of annual turnover (sales), logarithmictransformation of annual variable costs, logarithmic transformation of annual investment (additionsto depreciating capital) and annual investment relative to the capital of previous year.

Table 5 describes the data in the treatment and control groups. In terms of

turnover, net assets (by definition), investment, and variable costs (spending on in-

puts, such as material or intermediate goods), both groups are quite similar. The

labour costs and number of employees differ between the groups, which is to some

extent expected, since the number of owners differs across the groups by definition,

as groups are based on the ownership share.13 There are approximately 1000 yearly

observations in both groups. The main threat for the set up would be if the number

of owners responded to the tax changes. This could happen if, for example, the owner

were to split the firm to be partially owned by his/her spouse. However, the data

show that there is no observable response in the number of owners.

Figure 5 shows a reduced form event study of annual turnover, net assets and

investment. The plots are regressed year fixed effects with 2011 as a base year and firm

fixed effects included. The upper panels represent business activity and input usage

showing that the trends were similar before the tax change, and there is no notable

response to the dividend tax increase in 2012 in the treated group. The lower panels

show logarithmic investment and investment per lagged fixed assets, indicating that

there is no statistically significant response in investment to the increase in marginal

tax rate.

13For firms of this size, it is common that the owner also works in the firm.

42

Figure 5: Outcomes in treatment and control group relative to year 2011−.

3−.

2−.

10

.1.2

.3di

ff to

yea

r 201

1

2006 2008 2010 2012 2014

Turnover (log)

−.3−.

2−.

10

.1.2

.3di

ff to

yea

r 201

1

2006 2008 2010 2012 2014

Variable costs (log)

−2−1

01

2di

ff to

yea

r 201

1

2006 2008 2010 2012 2014

Investment (log)

−1−.

50

.51

1.5

diff

to y

ear 2

011

2006 2008 2010 2012 2014

Investment per lagged fixed assets

Treated Counterfactual

Note: Real values used, inflation from Statistics Finland. Treatment group: Firm’s net assets666,666.667–1,000,000e in 2011 and main owner’s share of net assets > 666,666.667. Controlgroup: Firm’s net assets 666,666.667–1,000,000e in 2011 and main owner’s share of net assets <666,666.667.

To estimate the reduced form difference-in-differences results for the tax change

of 2012, I estimate the equation

Yit = α1 + α2(Treat× Post) + β′iFEi + λtY eart + εi, (6)

where Yit is the outcome variable, α1 is a constant, Treat is a binary variable with

value 1 for firms facing a tax increase, Post is a binary variable with value 1 for

firms after the tax change, hence α2 measures the effect. β′iFEi is a matrix capturing

firm fixed effects and λtY eart is a matrix capturing year fixed effects. εi is the error

term, standard errors are clustered at the firm level. All estimates of α2 for different

outcome variables are reported on the first line of table 6 and were close to zero and

statistically insignificant. The results confirm the visual evidence that there is no

statistically significant response to the tax change among the firms facing the tax

43

increase. As the reduced form estimates do not show statistically significant effects,

neither do the IV estimates (not reported).

In the Appendix, I perform the same analysis for the 2014 tax change. The

tax change reduced the dividend tax for firm owners paying 60,000–150,000 euros

of dividends and dividend under the net asset threshold. In addition, the corporate

tax rate was reduced. As the corporate tax cut affected all firms, I can use firms

of the same size but smaller ownership share of the main owner as a control group.

The net asset limit for the firm sample is from 750,000 to 1,875,000. Firms whose

ownership share of the net assets was 750,000 or more, are the treated firms and those

whose ownership share was under 750,000 act as a control group. Summary statistics

and number of observations are reported in table A2 in the appendix. Figure A6

in the appendix shows the event study for the tax change and table A3 reports the

difference-in-differences results. Again, the results suggest that there is no significant

response to the reduction in the dividend tax among the treated firms.

The results suggest that the main mechanisms to respond to dividend tax adjust-

ments are through channels other than investment effects. However, the set-up only

studies local effects of changes in the current marginal tax rate. Therefore, I cannot

rule out global effects caused by changes in average tax rates or indirect effects, e.g.

through the future tax burden. Thus, in the next section, I discuss the other channels

for the bunching responses, namely income shifting across time and across tax bases.

4.2 Intertemporal income-smoothing and net asset accumu-

lation

Figures 6 and 7 plot the persistence rates at the 9-percent and at the 90,000-euro

thresholds respectively. The figures show that the extensive bunching is created by

the same owners year after year. The share of firms in the bunching region that

also located in the same euro- or net asset bin 1-4 years earlier is exceptionally large

compared to the surrounding bins. At the 9-percent net asset threshold, the share of

firms bunching for a second year in a row is almost 60 % and four years after it is

still approximately 30 percent. At the 90,000-euro kink the rate is above 50 % in the

first year and 20% after four years. As the threshold relocates, a large share of the

previous bunchers follow the threshold: the share of movers is described in table A1.

Owners of privately held corporations do not need to adjust their profit to bunch

44

Figure 6: Persistence at 9 % in 2006-2013 – 1-percent-bins around the threshold

0.2

.4.6

Pers

iste

nce

rate

.02 .04 .06 .08 .1 .12 .14 .16

one year

0.2

.4.6

Pers

iste

nce

rate

.02 .04 .06 .08 .1 .12 .14 .16

two years0

.2.4

.6Pe

rsis

tenc

e ra

te

.02 .04 .06 .08 .1 .12 .14 .16

three years

0.2

.4.6

Pers

iste

nce

rate

.02 .04 .06 .08 .1 .12 .14 .16

four years

Persistence Quadratic fit without bunchers

Note: The graph plots the share of same firms locating in the same 1-percent-bins around the9-percent threshold for 1 to 4 years after.

at the tax threshold, but they can adjust owner-level taxable income using retained

and withdrawn earnings to shift income across years. By smoothing income with

retained earnings, tax filers can hold their marginal tax rates constant. Hence, there

is likely to be bunching even if taxes have no effect on real outcomes (Le Maire and

Schjerning, 2013). The incentives for inter-temporal income shifting cause bunching

mass to also accumulate from below as the owners also have incentives to spread the

payments to be paid in advance.

Retaining wealth has three advantages. First, as mentioned, using retained and

withdrawn earnings allows the owner to avoid higher marginal tax rates. Second,

savings and return on savings face a lower tax when received by a firm than at

the owner level.14 Thus, if the owner, in any case, wishes to save some share of

the income, then for tax purposes it may be desirable to keep some of those funds

incorporated. Third, by retaining earnings, the owner increases the net assets of

the firm, thereby allowing for higher amounts of capital income dividends (lower

tax bracket) to be distributed in the future. Then again, there are also arguments

14E.g. dividends received by a firm a primarily tax free.

45

Figure 7: Persistence at 90,000 euros in 2006–2011, in 1000-euro-bins0

.1.2

.3.4

.5Pe

rsis

tenc

e ra

te

75000 80000 85000 90000 95000 100000 105000

one year

0.1

.2.3

.4.5

Pers

iste

nce

rate

75000 80000 85000 90000 95000 100000 105000

two years

0.1

.2.3

.4.5

Pers

iste

nce

rate

75000 80000 85000 90000 95000 100000 105000

three years

0.1

.2.3

.4.5

Pers

iste

nce

rate

75000 80000 85000 90000 95000 100000 105000

four years

Persistence Quadratic fit without bunchers

Note: The graph plots the share of same firms locating in the same 1000-euro-bins around the 90kthreshold for 1 to 4 years after.

against leaving wealth in firm, such as controlling risk. The elasticity estimates at

the monetary thresholds are lower than the estimates at the net asset threshold, so

the incentives for firm owners to bunch seem to be higher at the net asset threshold.

There are two potential reasons for this. First, the incentive to increase the firm’s

net assets by retaining earnings may be stronger at the net asset threshold. Second,

the earned income tax schedule is a lot more complex than the capital income tax

schedule, so the marginal tax rate above the net asset threshold may be less clear for

the owner.

Figure 8 shows the firm’s turnover, net assets, fixed capital (property and ma-

chinery), and financial assets on average across the dividend distribution of the main

owners (in 5000-euro-bins). The upper left panel shows the average annual turnover

in each dividend bin. The higher the dividend, the higher the turnover, indicating

a positive linear relationship. This linear relationship does not hold in the second

panel, which shows the average net assets in each bin. When the monetary threshold

was at 90,000 euros, firms whose owners bunch at the dividend threshold have more

net assets on average than firms in the surrounding dividend bins. However, there is

46

Figure 8: Average firm outcomes in 5000-euro-dividend bins

5000

0015

0000

025

0000

0

30000 50000 70000 90000 110000 130000Dividends

Turnover

5000

0015

0000

025

0000

0

30000 50000 70000 90000 110000 130000Dividends

Net assets10

0000

2000

0030

0000

4000

00

30000 50000 70000 90000 110000 130000Dividends

Machinery and property

2000

0070

0000

1200

000

1700

000

30000 50000 70000 90000 110000 130000Dividends

Financial assets

2010 20112012 2013

Note: The figure shows mean firm outcomes in 5000-euro-dividend bins (for the main owner). Inother words, the horizontal line shows the dividend received by the main owner and the vertical linethe outcome at hand in euros. In years 2010 and 2011 the monetary threshold was at 90,000 euros,whereas in 2012-2013 it was in 60,000. This figure shows that bunchers have on average higher netassets and especially financial assets, as there is no bunching in the fixed capital (machinery andproperty).

no similar bunching in reported machinery and property, whereas it does appear in

financial assets. Moreover, when the threshold moves to 60,000 euros in 2012, the net

asset and financial asset bunching moves along with the threshold. This suggests that

firm owners bunching at the thresholds indeed retain earnings in the firm and may

even use the firm to store savings (as financial assets). As an additional detail, figure

A3 in the appendix shows that firms in the financial industry bunch at the threshold

more actively than other industries.

Figure 9 shows that privately held corporations in Finland have accumulated

wealth in the firm. To ensure that this descriptive evidence is not just driven by the

increasing number of firms or economic growth, I relate this information to aggregate

turnover. Even in relation to turnover there is still a substantial growth in the assets

of the firms. However, this does not appear as higher net investment or dividend

47

payouts.15

Figure 9: Aggregate net asset accumulation, profits and retained earnings

.2.4

.6.8

010

000

2000

030

000

4000

050

000

milli

ons

2000 2005 2010 2015

Net assets Financial assetsNet assets / turnover Financial as. / turnover

Note: The area plots in this depict the aggregate net assets and financial assets of all firms in thedata. In addition, the lines plot them both in relation to aggregate sales. The trends show, thatdespite the economic turbulence in past decades, the assets of the firms has been steadily increasingsince the adoption of the current dividend tax schedule in 2005.

4.3 Income-shifting between tax bases

Income-shifting between wage and dividends allows firm owners to minimize their

income tax burden. Figure 10 shows the owner’s total income from the firm in 50

income quantiles and how the income splits between wage and dividends on average

in each income quantile. The horizontal axis describes the average dividends in each

income quantile, and the vertical axis the average wage earned by the owner within

a particular income quantile. The gray isoquant lines indicate the total income level

so that, for example, the 44th quantile received approximately 70,000 euros from the

firm.

In 2010–2011, the monetary threshold in the dividend tax schedule was 90,000

euros. In 2012, the threshold moved down to 60,000. Income affected by the tax

15More in figure A4 in the appendix.

48

Figure 10: Income-shifting between wage and dividends

42 43

44

4546

47

48

49

4243 44

45 46

47

48

49

020

000

4000

060

000

8000

0

0 20000 40000 60000 80000 100000 120000dividends

2010−2011 2012−2013 Isoquants

wag

e

Total income split to wages and dividendsIncome share response to the reform of 2012

Note: This figure plots the income shifting between wage and dividends as a response to the taxchange on 2012, which increased the taxation for dividends higher than 60,000 euros. For the figure,the main owners’ wage and dividends from the firm have been counted together as total income.Then, the owners have been divided to 50 income quantiles (2-percentiles). Finally, for each quantilean average wage and dividends are calculated. The horizontal line shows the average dividends andthe vertical line the average wage in each bin. The isoquant lines show the total income from thefirm. The figure shows that as a response to the tax change the owners started paying more wageand cut down dividends.

change, that is, income above 60,000 euros, clearly shifts towards more wages in

comparison to dividends. The position of the quantiles in relation to the isoquant

lines reveals that, despite the tax increase and the ensuing reduction in dividend

income, the inflation-adjusted income stays the same in the affected quantiles. It is

just the division into wage and dividend that changes. There is no similar pattern

when there is no tax change, as shown in figure A5 in the appendix.

5 Conclusion

I find strong bunching in all five dividend tax discontinuities in place since 2005 in

Finland. The observed bunching gives large elasticity estimates, but I show that

49

this is not driven by real responses e.g. in investment or effort. On the contrary,

the evidence suggests that a large share of the bunching evidence captures income

shifting in time and across income bases.

Owners engage in inter-temporal income-shifting by adjusting their income using

retained earnings. Retaining earnings enables firm owners to avoid the higher tax

brackets and it is further stimulated by the possibility in the Finnish tax system to

reduce future tax burden by accumulating net assets in the firm. Leaving wealth

in the firm also has some tax benefits, mainly on the return to savings. Thus, firm

owners accumulate net assets in the firm in order to optimize returns on savings and

the future tax burden of dividends. Moreover, I find clear income shifting between

wage and dividends. I study two tax changes, both of which were followed by an

adjustment in the income composition of the owner but with no reduction in their

total income. Similar adjustments between wages and dividends were not observed

when there was no tax change.

These results lend support to the ’new view’ that the marginal dividend tax rate

is not the key parameter that affects marginal investment. In other words, dividend

tax adjustments are a weak tool for incentivizing real economic activity. However,

the results do not rule out larger differences in average dividend tax rates having

effects on investment, as this paper focuses only on marginal tax rates. This paper

shows that bunching in a single income base captures a lot more than just structural

elasticity (real responses). The extensive income-shifting responses underline that

static estimates of business tax revenue effects of tax changes are likely to go wrong.

This is because business owners can adjust their income by shifting income and by

retaining earnings. Finally, large differences between corporate and dividend taxes

are likely to cause locking effects, as savings may be left to accumulate in firms

and reduce the tax burden of both dividends and return savings. Thus, the results

highlight that large differences between income bases are likely to create behavioural

responses, causing distributional effects at least. Essentially, smaller tax differences

between wage and dividends, but also between taxes paid by firms and individuals,

affect these distributional distortions.

50

References

Annette Alstadsæter, Martin Jacob, and Roni Michaely. Do dividend taxes affect

corporate investment? Journal of Public Economics, pages –, 2015. ISSN 0047-

2727. doi: http://dx.doi.org/10.1016/j.jpubeco.2015.05.001.

Alan J. Auerbach. Wealth maximization and the cost of capital. Quarterly Journal

of Economics, pages 433–446, 1979.

Spencer Bastani and Hakan Selin. Bunching and non-bunching at kink points of

the swedish tax schedule. Journal of Public Economics, 109:36 – 49, 2014. ISSN

00472727.

Raj Chetty and Emmanuel Saez. Dividend taxes and corporate behavior: Evidence

from the 2003 dividend tax cut. The Quarterly Journal of Economics, 120(3):

791–833, 2005. doi: 10.1093/qje/120.3.791.

Raj Chetty and Emmanuel Saez. Dividend and corporate taxation in an agency model

of the firm. American Economic Journal: Economic Policy, 2(3):1–31, August 2010.

doi: 10.1257/pol.2.3.1.

Raj Chetty, John N. Friedman, Tore Olsen, and Luigi Pistaferri. Adjustment costs,

firm responses, and micro vs. macro labor supply elasticities: Evidence from dan-

ish tax records. Quarterly Journal of Economics, 126(2):749 – 804, 2011. ISSN

00335533.

M. S. Feldstein. Corporate taxation and dividend behaviour. The Review of Economic

Studies, 37(1):57–72, 1970. doi: 10.2307/2296498.

Jarkko Harju and Tuomas Matikka. The elasticity of taxable income and income-

shifting: what is “real” and what is not? International Tax and Public Finance,

pages 1–30, 2016. ISSN 1573-6970. doi: 10.1007/s10797-016-9393-4.

Seppo Kari and Jussi Laitila. Non-linear dividend tax and dynamics of the firm.

Working Papers 41, Government Institute for Economic Research Finland (VATT),

December 2014.

Mervyn A. King. Taxation and the cost of capital. The Review of Economic Studies,

41(1):21–35, 1974. doi: 10.2307/2296396.

51

Henrik Jacobsen Kleven. Bunching. The Annual Review of Economics, 8, 2016.

Claus Thustrup Kreiner, Søren Leth-Petersen, and Peer Ebbesen Skov. Year-end tax

planning of top management: Evidence from high-frequency payroll data. American

Economic Review, 104(5):154–58, May 2014. doi: 10.1257/aer.104.5.154.

Claus Thustrup Kreiner, Soren Leth-Petersen, and Peer Ebbesen Skov. Tax reforms

and intertemporal shifting of wage income: Evidence from danish monthly payroll

records. American Economic Journal: Economic Policy, 8(3):233–57, August 2016.

doi: 10.1257/pol.20140233.

Daniel Le Maire and Bertel Schjerning. Tax bunching, income shifting and self-

employment. Journal of Public Economics, 107(1):1–18, 2013.

James M. Poterba and Lawrence H. Summers. The Economic Effects of Dividend

Taxation. Recent Advances in Corporate Finance, edited by Edward Altman and

Marti Subrahmanyam, pages pp. 377–380, 1985.

Emmanuel Saez. Do taxpayers bunch at kink points?. American Economic Journal:

Economic Policy, 2(3):180 – 212, 2010. ISSN 19457731.

Hans-Werner Sinn. The vanishing harberger triangle. Journal of Public Economics,

45(3):271 – 300, 1991. ISSN 0047-2727. doi: https://doi.org/10.1016/0047-

2727(91)90029-2.

Danny Yagan. Capital tax reform and the real economy: The effects of the 2003

dividend tax cut. American Economic Review, 105(12):3531–63, December 2015.

doi: 10.1257/aer.20130098.

George R. Zodrow. On the ’Traditional’ and ’New’ Views of Dividend Taxation.

National Tax Journal, 44(4):497–509, December 1991.

52

Appendix

Additional information on the method

Figure A1: Net dividends relative to corporate income 2006-2011 and hypotheticalindifference curves

Gross dividends e

Net dividends e

30k 60k 90k 120k 150k 180k

30k

60k

90k

120k

150k

Figure A1 describes the bunching method graphically. The solid line is the relation

between corporate profit (gross dividend) needed to finance a certain amount of net

dividend. Gross profit reflects the cost in effort and capital required to produce the

profit. At e90,000, the marginal tax rate increases, creating a kink in the budget

set. Assume that the taxpayers maximize their utility relative to the kinked budget

set and that the black curve is a hypothetical indifference curve of an individual, who

decides to raise a dividend of exactly e90,000 despite the kink in the budget set. The

dashed black line is the budget set in the absence of the kink. The dashed gray curve

is the optimal dividend of another individual in that setting. After the introduction

of the kink, this individual with a dashed gray indifference curve, who earlier located

at around e100,000, moves to a lower indifference curve (solid gray) and locates at

e90,000 together with the individual with the black indifference curve. As the owner

with the black curve does not move anywhere, there are now more individuals than

in the case of a linear budget set. I use this bunching induced by the kink to estimate

the elasticity of taxable income.

53

Additional information on the institution regarding the earned income tax

schedule

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 circa 55 %. The government tax rates on earned income decreased over the

research period of 2006-2016, especially for low and middle income earners. However,

the 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 A2 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.

Additional information on the bunching responses

Figure A3 shows how various industries are represented in each bin around the 90,000-

euro 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 A1 shows the proportion of firms that move together with the threshold.

When the monetary threshold moved from 90,000 euros to 60,000 euros, 47% of the

preceding excess mass firms followed the threshold. At the 150,000 euro threshold,

one-third of the observations had previously paid exactly 60,000 euros of dividends,

which was the preceding threshold. At the 8% net asset threshold 70% of firms had

previously bunched at the 9% threshold.

Figure A5 shows the income composition in two consecutive years when there was

no tax change. There is now no change in the income composition of the owners. The

figure acts as a robustness check that the shift observed in Figure 10 was driven by

the tax change.

Figure A4 shows the accumulation of aggregated assets in privately held corpora-

54

Figure A2: Average marginal tax rate for firms paying dividends under the monetarydividend tax threshold

.1.1

5.2

.25

.3.3

5.4

.45

.5M

argi

nal t

ax ra

te

0 .05 .08 .09 .15 .2Dividends per net assets

MTR 2012−2013 MTR 2014−2015

Marginal dividend tax rate

Note: The tax rate above the threshold is estimated as a mean of the marginal earned incometax rates in the data. The earned income tax rate varies depending on the taxpayer’s income andmunicipality. Both the municipal and government tax schedules change nearly every year. Thelowest government tax rate has been zero during the whole period and, with deductions in themunicipal tax for low income earners, also the aggregate earned income tax rate has been close tozero in the low end of the income distribution. The highest overall marginal earned income taxrate has been circa 55 %. Overall government tax rates on earned income have been decreasingduring 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 itwas 19.4 %. The municipal income tax varies across municipalities; in 2015 it ranged from 16.5 %to 22.5 %.

tions in 2000–2016. Net assets consists of the retained earnings, financial assets, and

additions to depreciating capital. The figure shows also the evolution of aggregated

profits, dividends and net investment to depreciating capital. The figure shows a clear

increase in the 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.

Additional results of the real responses – tax change in 2014

Table A2 describes the data I use to study the effects of the second tax change in

2014. The tax change reduced the tax rate for dividends between 60,000–150,000

euros for owners with net asset shares between 750,000–1,875,000 euros. Figure A6

55

Figure A3: Industry shares among of 90k bunchers 2006-2011

0.0

2.0

4.0

6

60000 90000 120000dividend bin

Mining

.1.2

.3

60000 90000 120000dividend bin

Manufacturing

0.2

60000 90000 120000dividend bin

Construction

.1.2

.3.4

60000 90000 120000dividend bin

Wholesale and retail

0.0

5.1

60000 90000 120000dividend bin

Transport and storage

0.0

5

60000 90000 120000dividend bin

Hotels and restaurants

0.0

5.1

60000 90000 120000dividend bin

Information and communication

.05

.1.1

5.2

60000 90000 120000dividend bin

Finance and insurance

0.0

5.1

60000 90000 120000dividend bin

Real estate

.05

.1.1

5.2

60000 90000 120000dividend bin

Law, accounting, consulting

0.0

2.0

4.0

6

60000 90000 120000dividend bin

Support services

0.0

5.1

60000 90000 120000dividend bin

Health and social services

Industry’s share of the bin Industry’s share of the data

Note: The figure plots the shares of each industry in bincs around the 90,000-euro threshold. Thehorizontal axis shows the dividend amount and the vertical axis the share of the industry in eachbin. The dashed horizontal line denotes the average share of the industry in the data. According tothe figure, the financial sector seems to be over-represented at the kink.

Table A1: Percentage share of firm owners relocating together with the kink

Movers as a share of Movers as a share ofTax change Year bunchers before tax change bunchers after tax change

90k → 60k 2011/2012 46.72% 24.52%60k → 150k 2013/2014 8.12% 35.45%9pr → 8pr 2013/2014 60.33% 70.40%

Note: This table reports the share of observations in the bunching region following a thresholdchange that in previous years bunched at the preceding threshold range. The share is reported asa proportion of the bunchers at the preceding threshold as well as as a proportion of the bunchersafter the tax change.

plots the development for turnover, variable costs, and investment among treated and

control firms estimated with firm fixed effect regression with binary variables for each

year. Table A2 reports the corresponding difference-in-difference results. There is no

statistically significant response to the tax cut. The diff-in-diff results in table A3

56

Figure A4: Aggregate net asset accumulation, profits and retained earnings

040

0080

0012

000

1600

020

000

milli

ons

1000

020

000

3000

040

000

5000

0ne

t ass

ets,

milli

ons

2000 2005 2010 2015

Net assets ProfitsDividends Net investment

Note: This figure shows the accumulation of aggregate net assets among the privately held firmsstudied in this paper in gray. The blue line shows the annual aggregate profits, the red, dashed line,the annual aggregate dividend for the main owner and the dashed green line the annual aggregatenet investment.

confirm the result of no effect. These results are the first stage of the instrumental

variable set-up, but if there is no effect in the first stage, there is no effect for the

second stage either.

57

Figure A5: Income-shifting between wage and dividends

4243 44

45

4647

48

49

42 43

44

4546

47

48

49

020

000

4000

060

000

8000

0

0 20000 40000 60000 80000 100000 120000dividends

2008−2010 2011−2012 Isoquants

wag

e

Total income split to wages and dividendsIncome shares before reform

Note: This figure plots the income composition between wage and dividends in 2008–2009 and in2010–2011. For the figure, the main owners’ wage and dividends from the firm have been countedtogether as total income. Then, the owners have been divided to 50 income quantiles (2-percentiles).Finally, for each quantile an average wage and dividends are calculated. The horizontal line showsthe average dividends and the vertical line the average wage in each bin. The isoquant lines showthe total income from the firm. The figure shows that when there was no tax change the owners’income composition stayed quite the same. This figure acts as a robustness check for Figure 10 inthe main text.

58

Table A2: Summary statistics of the restricted sample (year 2013)

Treated Controlmean sd p50 mean sd p50

Turnover 2307248 1.51e+07 967603.8 2444987 3595202 1485020

Net Assets 1226032 325761.6 1170015 1089996 268768.1 1024905

Dividends (main owner) 85368.86 108177.2 69000 50472.34 39405.6 44000

Investment 101979.3 279940.9 16909.12 92262.13 215896.7 19639.17

Labor costs 451738.7 621883.8 240626.4 650701.4 763233.8 427757.2

Variable costs 1872973 1.67e+07 586906.2 1655490 3145893 765121.7

Employees 13.20741 27.33048 4 16.53348 24.26132 9

Observations (2013) 2027 (1568*) 2240Observations (total) 10437 10619

* 1568 is the number of firms paying dividend in the affected region

Note: These are the descriptive statistics of the firms used in the diff-in-diff set-up of figure A6studying the tax cut in 2014. This table provides the summary statistics for the Difference-in-difference data in 2013. Turnover refers to annual sales, net assets refer to book value of assetsafter depreciation, dividends are the main owner’s dividend income from the firm, and investmentrefers to additions to depreciating assets, such as newly installed fixed capital. Labor costs includewage and payroll taxes paid by the firm and variable costs other input costs such as material andintermediate goods, number of workers include number of employees in the firm in 2013. Each firmhas only one main owner in the data. The owner with highest share of stock is considered the mainowner.

Table A3: Difference-in-difference results of the 2014 tax change

Turnover (log) Variable costs (log) Investment (log) Investment per lagged capital

α2(Treat× Post) -0.030 -0.013 -0.031 0.295

0.026 0.032 0.146 0.247

Firm fixed effects X X X X

Year fixed effects X X X X

Constant 13.986*** 13.043*** 6.003*** 0.541***

0.010 0.013 0.069 0.097

r2 0.017 0.006 0.009 0.000

N 18294.000 16787.000 21056.000 17293.000

* p < 0.05, ** p < 0.01, *** p < 0.001

Note: This table reports the regression results estimated following Equation 6 for the tax cutin 2014. The dependent variables in the specifications are logarithmic transformation of annualturnover (sales), logarithmic transformation of annual variable costs, logarithmic transformation ofannual investment (additions to depreciating capital), and annual investment relative to the capitalof previous year.

59

Figure A6: Outcomes in treatment and control group relative to year 2013

−.3−.

2−.

10

.1.2

.3di

ff to

yea

r 201

3

2012 2013 2014 2015 2016

Turnover (log)−.

3−.

2−.

10

.1.2

.3di

ff to

yea

r 201

3

2012 2013 2014 2015 2016

Variable costs (log)

−1.5

−1−.

50

.51

diff

to y

ear 2

013

2012 2013 2014 2015 2016

Investment (log)

−1−.

50

.51

1.5

diff

to y

ear 2

013

2012 2013 2014 2015 2016

Investment per lagged fixed assets

Treated Counterfactual

Note: Coefficients from a firm-fixed effect regression of log annual turnover on year relative to 2013.Variables are used in real values with inflation from Statistics Finland. Treatment group: Firm’s netassets 750,000–1,875,000e in 2013 and main owner’s share of net assets > 750,000. Control group:Firm’s net assets 750,000–1,875,000e in 2011 and main owner’s share of net assets < 750,000.

60

Chapter 3

The Effects of Corporate Taxes on Small

Firms∗

Aliisa Koivisto1,2, Jarkko Harju1,3, and Tuomas Matikka1

1VATT Institute for Economic Research

2University of Helsinki3Tampere University

Abstract

We study the impact of corporate taxes on firm-level investments and business

activity by exploiting a 4.5 percentage-point corporate tax rate cut in Finland

in 2014. Using detailed administrative data and a differences-in-differences

method comparing small corporations (tax rate cut) to similar partnerships

(no change in taxes), we find no significant investment responses. However,

we observe an increase in annual sales and variable costs. These effects are

driven by firms where the main owner actively works in the firm, suggesting

that firm-level tax incentives have a larger effect on business activity among

small firms with closely connected owner-managers.

JEL classification codes: G31, G38, H21, H25

Keywords: corporate taxation; investments; business activity; small firms

1 Introduction

Over the last decade, many developed countries have reduced their corporate tax rates

in order to accelerate firm-level investments and economic activity. For example, in

∗An earlier version of this paper is published in VATT Working Papers series, 129, March 2020.

61

2017, the US cut its corporate tax rate from 35% to 21%. In addition, various other

investment stimuli have been introduced in many countries, such as more favorable

deduction and depreciation rules. These reforms have reduced the cost of capital and

relaxed the financial constraints on firms, creating incentives for new investments and

increased business activity. This development has also prompted researchers to study

the effects of these reforms using quasi-experimental methods and administrative

data, focusing mostly on investment responses and the incidence of corporate taxes

(Yagan 2015; Bond and Xing 2015; Suarez Serrato and Zidar 2016; Zwick and Mahon

2017; Ohrn 2018; Fuest et al. 2018; Maffini et al. 2019; Liu and Mao 2019; Ohrn

2019).

Despite this recent surge in quasi-experimental evidence, many key questions are

still unanswered or understudied. There is only scarce evidence for the impacts of

corporate taxes, a central parameter in policy debate, on firm-level investments and

growth. Many earlier papers focus on analyzing investment incentives that are tar-

geted at specific types of firms or industries, such as bonus depreciations and special

tax deductions. Consequently, there is a lack of knowledge on how changes in incen-

tives that apply to the universe of the firm population, such as a corporate tax cut,

affect investments and economic activity. Moreover, evidence on the impact of finan-

cial incentives among small firms is very limited, even though it has been argued that

small and growing firms play a key role in spurring economic growth and employment

(see e.g. Decker et al. 2014).

We contribute to the literature by providing credible evidence for the effects of

a corporate tax rate cut on small firms. We study a considerable 4.5 percentage-

point reduction in the corporate tax rate from 24.5% to 20% in 2014 in Finland.

Together with high-quality administrative data covering all Finnish businesses, this

reform enables us to analyze the effect of the corporate tax rate on a range of key

firm-level variables. Moreover, the corporate tax cut in Finland was combined with

a dividend tax increase that mostly eliminated the impact of the corporate tax cut

on the shareholder-level effective dividend tax rate, allowing us to focus solely on the

effects of firm-level corporate taxes.1 Also, this setup enables us to offer evidence

for how an actively-debated shift from corporate-level to owner-level taxes affects

firm-level decisions (see e.g. Grubert and Altshuler 2016 and Devereux 2019).

1In general, firm-level taxes are considered to be more relevant to new investment and growth.The dividend tax rate does not affect the marginal cost of capital when investment is funded withretained earnings or debt [Auerbach, 1979].

62

Our analysis focuses on small corporations with annual sales below 2.5 million

euros. We restrict the sample because partnership firms that faced no changes in

taxes offer a representative comparison group only for relatively small corporations.

We use a differences-in-differences method utilizing similar-sized partnership firms

operating in similar industries as a control group, allowing for credible and transparent

identification of the impact of the reform on small firms. Furthermore, we follow a

weighting estimation procedure used by Yagan [2015] and Zwick and Mahon [2017] to

ensure the comparability of the outcomes between these organizational forms. In our

empirical analysis, we show that the development of our outcome variables follows

parallel trends for the treatment (corporations) and control (partnerships) groups

prior to the corporate tax cut, ensuring the validity of our empirical approach.

First, we study the effect of the reform on investments relative to existing capital

assets, a main outcome of interest explored in both the theoretical and the empirical

corporate tax literature. We find no significant average investment responses in the

stock of productive capital after the reform, nor in the level of investments or the

number of firms with new investments (extensive margin). However, we find a small

positive investment effect for younger firms (under 10 years) compared to older firms,

suggesting that the investment decisions of younger firms can be more sensitive to

corporate taxes.

In addition, we examine the impact of the reform on firm-level business activity

measures that are rarely studied in the earlier literature, including sales, labor costs,

input use, value added, and firm entry. A corporate tax cut creates a mechanical

cash injection for the firm through increased net-of-tax retained earnings that could

positively impact business activity beyond just physical capital investments. Such

additional cash resources can be particularly relevant for smaller firms that might

often face liquidity constraints and have limited opportunities to acquire other types

of funding, but still have business opportunities to utilize. Also, small firms are often

managed by their main owners, who also work in the firm and are closely connected

to firm decision making, implying that the incentives and effort of the owner can be

affected through changes in corporate-level tax incentives.

We find a significant increase in sales and inputs for the treated corporations,

implying a higher firm growth rate after the corporate tax rate cut. These results

suggest that the overall business activity of small firms increased after the reform

even though investments in the stock of productive capital did not. We find that the

observed sales and input responses are clearly driven by firms owned by entrepreneurs

63

who actively work and manage their own firms, compared to firms with more passive

owners. This indicates that the role and effort of the main owner of the firm are

essential factors in explaining how small firms respond to changes in financial incen-

tives, providing new evidence for the mechanisms behind firm responses to corporate

taxes. Moreover, we find no significant differences in responses for more or less cash-

constrained firms in our sample and no effects on firm entry, suggesting that these

channels do not consistently explain our main findings.

Our study contributes to the existing literature in many ways. First, we contribute

to the literature studying the effects of financial incentives on investments. Most of

the recent quasi-experimental literature exploits targeted variation in investment in-

centives, such as bonus depreciations or special tax deductions, aimed at boosting

investments in particular sectors or groups of firms. This branch of literature com-

monly estimates distinctively large investment responses. For example, Ohrn [2018]

uses changes in deduction regulation in the US and finds a large investment response

and an implied elasticity of investments of 6.5. House and Shapiro [2008] and Zwick

and Mahon [2017] use changes in depreciation rules in the US and find significant

investment responses and very large investment elasticities, 7.7 and 7.2, respectively.

Maffini et al. [2019] find similar results in the UK.

However, only a few previous papers have studied the investment responses of uni-

versal business tax reforms using firm-level data. A notable exception is Yagan [2015],

who studies the massive 2003 dividend tax cut in the US and finds a large response in

dividend payments but no effect on new investments, indicating a very small elasticity

of investments with respect to dividend tax rates. According to our knowledge, our

paper is the first to use firm-level data and quasi-experimental variation to study the

investment effects of a universal change in the corporate tax rate.

In contrast to many recent empirical studies, we find a small and statistically

insignificant average investment response and an elasticity estimate below 0.5. Our

results also allow us to statistically rule out elasticity of investments larger than 1.3

with respect to the net-of-corporate tax rate. Therefore, our results illustrate that

responses to a universal cut in a corporate tax rate for smaller firms combined with

a dividend tax increase can be very different and lead to different policy implications

compared to the impacts of more targeted changes in investment incentives for larger

firms, as analyzed in the previous literature.

One explanation for the negligible average investment effect, according to the

so-called old view of dividend taxes, could be that firms finance their investments

64

with new equity installments. In that case, the simultaneously increased dividend

tax would mitigate incentives for investing (see e.g. Feldstein 1970 and Poterba and

Summers 1985). Nevertheless, we find no investment effects for older firms that are

argued to rely more on retained earnings or debt when funding their investments

(Auerbach 1979), suggesting that small responsiveness of investments to corporate

taxes among smaller firms is more likely to explain our findings. However, we high-

light that physical capital investments are not the only relevant firm-level margin to

consider, as our results show that improved financial incentives increased the overall

business activity of small firms.

Furthermore, we make a clear contribution beyond the field of public economics

as we study relatively small firms, which, it has been argued, play a key role in gen-

eral economic development and employment growth [Decker et al., 2014]. Therefore,

the responsiveness of these firms is highly relevant but still relatively understudied,

although there have been some recent exceptions (see e.g. Benzarti et al. 2020). Also,

we increase the current knowledge of how relaxing firm-level financial constraints af-

fects firm behavior (see e.g. Rauh 2006) by showing that an increase in available cash

reserves at the firm level accelerates the business activity of small firms.

Moreover, we study the responses separately for firms with more active and pas-

sive main owners in order to better understand the role of the owner behind firm-level

decisions. According to our results, the role of the owner-manager is crucial in explain-

ing how small firms respond to financial incentives, as firms with closely connected

owner-managers respond much more actively to reduced corporate taxes compared

to firms with more passive owners. This analysis contributes to the literature aiming

to explain how ownership and management structures affect firm-level decisions and

responses to business taxes (see e.g. Chetty and Saez 2010 and Jensen and Meckling

1976). Using our detailed data, including information on ownership shares and the

role of the main owner in the firm, we provide new empirical evidence to this dis-

cussion, in which earlier empirical studies are scarce, particularly for smaller firms.

Our evidence also relates to the findings of a recent paper by Smith et al. [2019], who

highlight the importance of active owners to firm performance in the US.

This paper proceeds as follows. In Section 2 we describe the business tax system

in Finland, and in Section 3 we present our testable hypotheses. In Section 4, we

provide the details of the data and discuss our methods. Section 5 presents and

discusses the results and Section 6 concludes.

65

2 Finnish Business Tax System and the Reform of

2014

In this paper we study the effects of a large corporate tax rate cut in Finland. The

tax rate on corporate profits was reduced by 4.5 percentage points from 24.5% to

20% from January 2014 onward. The Finnish government justified the tax cut on

the grounds of its potential positive effects on firm investment and growth. This

reform continues the downward trend in corporate tax rates in Finland and many

other developed countries. For example, corporate taxes were cut in Germany in

2008, the UK in 2008, 2011, 2012 and 2013, Sweden in 2009 and 2013, Canada in

2008-2012, and the US in 2017. In Finland, the 2014 tax rate cut was preceded by

a smaller 1.5 percentage-point tax cut at the beginning of 2012. We examine both

of these changes, but focus mainly on the latter and larger tax cut. Also, at the

time of the corporate tax rate cut in 2014, the dividend tax rate was increased for

many owners of small corporations, which mitigated the impact of the corporate tax

cut on shareholder-level effective dividend taxes. Next, we discuss the main features

of the Finnish business tax system and the changes in tax rates we exploit in the

empirical analysis we introduce below. A more detailed description of the tax system

and recent changes in the taxation of corporations and partnership firms in Finland

is presented in Appendix B.

The 4.5 percentage-point corporate tax rate cut in 2014 affected all public and

privately held corporations, but other organizational forms were unaffected by the

reform. Therefore, we use partnership firms as a control group for small corporations

in our analysis. The privately held corporation is the most common organizational

form in Finland, representing nearly half of all Finnish firms. Privately held corpora-

tions are separately tax-liable, meaning that their profits are taxed at the firm level

according to the corporate tax rate. Owners of privately held corporations pay an

additional tax on the income withdrawn from the firm. In contrast, partnerships are

pass-through entities, meaning that their profits are taxed directly at the owner level.

Therefore, the corporate tax rate does not affect them.

An owner of a privately held corporation can withdraw income from the firm either

as wages or dividends. In Finland, wage and capital income are taxed at different tax

rate schedules. The wage tax schedule is progressive, with top marginal tax rates of

approximately 55%. The dividend income tax system is rather complicated, including

one tax rate kink determined by firm-level net assets and another kink based on the

66

euro amount of the dividends withdrawn from the firm. In contrast, the profits of

partnership firms are taxed directly at the owner level based on a predetermined tax

schedule. However, the owners of partnership firms also have a net asset threshold

in their tax schedule, which divides taxable profits into capital income and wage

income components based on firm-level net assets. Profits that fall under the net

asset threshold are taxed according to the capital income tax schedule, and any

income above the threshold is taxed as the owner’s wage income.

At the time of the corporate tax rate cut in 2014, the dividend tax rate was

increased for many owners of small corporations. Therefore, at the owner level, the

impact of the reduction in the corporate tax rate was offset by this dividend tax

increase, which typically increased the effective dividend tax rate by 1.5 percentage

points. Thus, as we will argue in greater detail below, the corporate tax cut mainly

affected incentives for investment funded from retained earnings or debt, while the

cost of capital for new equity (share issues) remained mostly unchanged or faced

a small increase. However, for the owners of larger firms with large firm-level net

assets, the effective dividend tax rate was typically reduced within the 2014 reform

(see Appendix B for more details). Nevertheless, as we restrict our sample to small

firms, this does not concern our empirical analysis.

Furthermore, there were no changes in the taxation of partnership firms at the

time of the 2014 reform. Therefore, partnership firms constitute a suitable comparison

group for small privately held corporations. Also, there were no other significant

changes in business taxation at the time of the corporate tax cut. For example,

depreciation rules and the corporate tax base remained unchanged. Furthermore, the

depreciation rules are similar for both organizational forms, and investment costs are

depreciated over their lifetime following similar category-specific regulations for both

corporations and partnerships.

3 Expected Impacts of the Reform

3.1 Investments

Next, we discuss the changes in incentives created by the reform of 2014 and present

our hypotheses on its impact on investments and business activity of small firms. A

corporate tax cut can affect investment decisions by decreasing the cost of capital and

increasing the amount of retained earnings available for new investment. As discussed

67

above, the corporate tax cut in 2014 was executed together with an adjustment in

dividend tax rates such that the effective owner-level taxation was largely unaffected

for the owners of small corporations. Therefore, the expected impact of the reform

on investment incentives depends on how small firms finance their investments.

In brief, the increased dividend tax mitigates the incentives to increase investments

if they are financed by new equity, as then the effective dividend tax rate, including

the corporate tax, defines the rate of return on new investment. In other words, in

this case the dividend tax burden is included in the cost of capital. According to

the so-called old view of dividend taxes, owner-level (effective) dividend taxes are

most relevant for investment choices (Harberger 1962; Feldstein 1970; Poterba and

Summers 1985), and thus under this hypothesis the 2014 reform would not induce a

significant increase in investment incentives.

In contrast, it has been argued that dividend taxes are less relevant to investments

when new investment is financed by retained earnings or debt. Then the dividend

tax rate does not enter the marginal cost of capital that determines the level of

investment, whereas the corporate tax does [Auerbach, 1979]. For example, when

retained earnings are used for investment instead of profit distribution, the dividend

tax is reduced on the net cost to the shareholder at the same rate at which the

eventual return is taxed. Thus, these two effects cancel each other out to leave the

required rate of return unaffected by the dividend tax rate. According to the so-called

new view in the dividend tax literature, at the margin, new investment is financed by

retained earnings or debt, and therefore, the dividend tax does not affect investment

choices (Auerbach 1979; Bradford 1981).2

Overall, there is very limited earlier evidence on the investment behavior of small

firms and their sources of investment funding (new equity, retained earnings or debt,

or a combination of them all). Therefore, we have no clear theoretical hypothesis on

the expected impact of the reform on investment incentives. Thus, the aim of our

empirically estimated effects on investments is to reveal whether a change in firm-level

financial incentives accompanied by an owner-level dividend tax increase affected the

investment decisions (on average), given the available funding sources of small firms.

2This argument has been further supported by, for example, the evidence presented in Yagan[2015] showing that firm-level investments in the US did not respond to the large dividend tax cut of2003. Thus, recent empirical studies on investment effects have been more focused on corporate-leveltaxes. However, a recent working paper by Moon [2020] shows that the capital gains tax has aneffect on investments of large firms in South Korea, supporting the old view of the dividend taxliterature.

68

However, we conduct a suggestive test for the role of funding sources by studying

the investment responses separately for older and younger firms in our sample. It

has been argued that young and newly established firms are more prone to raise new

equity instead of using debt or retained earnings to finance their new investments,

compared to more mature firms (Auerbach 1979; Sinn 1991). Therefore, under this

hypothesis, older firms should be more prone to respond to the 2014 reform if firm-

level financial incentives drive investment decisions. Our detailed data enabled us to

define the age of the firm and thus analyze how firm age affects investment responses.

Finally, the 2014 reform allows us to empirically study a case where the tax burden

is shifted from corporate taxes toward the personal income taxes of the owner (see e.g.

Grubert and Altshuler 2016). In principle, such a reform could provide a cost-effective

avenue for improving investments and growth if reduced firm-level financial incentives

are indeed the drivers of investment decisions, and the increased owner-level taxes do

not distort investment decisions at a similar magnitude. In other words, the tax

revenue losses from the reduced corporate tax rate could be alleviated by an increase

in owner-level taxes while simultaneously boosting investments through improved

firm-level financial incentives. Thus, we are able to provide a novel empirical test for

how far this hypothesis holds for small firms, and whether investments can really be

increased by shifting the tax burden from the corporate to the owner level.

3.2 Business Activity

In addition to investments, the cut in the corporate tax rate can affect the overall

business activity of small firms in other ways. Given the key role played by small and

growing firms in spurring economic development and employment (see e.g. Decker

et al. 2014), it is highly relevant to study margins other than investments to better

understand the responsiveness of these firms. Our setup enables us to provide new

evidence on the impact of firm-level financial incentives on the overall business activity

of small firms, including sales, input use and labor costs. Also, we provide novel

evidence on how the ownership structure and the role of the main owner in the firm

affect firm responsiveness to relaxed financial constraints at the firm-level.

Even though the dividend tax increase mostly offsets the incentives created by the

corporate tax cut at the owner-level, the corporate tax cut still creates a mechanical

cash injection for the firm through increased net-of-tax retained earnings. Such addi-

tional financial resources can be particularly important for smaller firms. They often

69

face liquidity constraints and have limited opportunities to acquire other types of

funding such as new equity and debt, but might still have new business opportunities

to utilize. Thus, new available cash resources after the corporate tax rate cut could

have a significant impact on the potential of small firms to expand their business

activity.

Consequently, while limited access to debt might reduce the observed responsive-

ness of larger physical capital investments to the 2014 reform, it could still boost other

types of business activity in small firms. Therefore, the overall business activity of

small firms could be affected by the reform even without a direct and instant impact

on investments such as plants or machinery that typically require large resources.

Our hypothesis is that if such channels are relevant, we should observe an increase in

firm-level sales and variable costs after the 2014 reform.

Moreover, the ownership and management structures of the firms could affect how

they respond to changes in financial incentives. In particular, smaller firms are often

managed and organized by their main owners, and thus the role of the owner can be

particularly relevant for them. It has been argued that active owner-managers are

likely to be more able and eager to affect firm-level decisions and business activity

compared to more passive owners or investors. For example, Chetty and Saez [2010]

build an agency model of the firm, motivated by the observation that the dividend

responses to the dividend tax cut of 2003 in the US were driven by firms with active

share owners in executive positions. Also, recent empirical evidence by Smith et al.

[2019] supports the significance of the role played by the owner in the firm, as they

find that active owners contribute greatly to firm performance in the US.

Also, owners who are more closely connected to their firms are presumably more

able to utilize the firm and its cash reserves as a source of private spending (see e.g.

La Porta et al. 2000 and Bennedsen and Wolfenzon 2000) and engage in dynamic tax-

planning procedures in order to minimize long-run tax payments on income withdrawn

from the firm (see e.g. le Maire and Schjerning 2013). Therefore, improved firm

performance and growth have a more direct impact on the available personal resources

of more closely connected owners. Hence, even when the dividend tax is increased,

the cut in the corporate tax rate can increase the effort and labor input of an active

main owner through these channels, potentially leading to greater business activity

of the firm after the reform.

As our study concerns relatively small firms with, on average, only six employees

in our baseline sample, the role of the main owner is likely to have a particularly

70

significant impact. Therefore, if the management and ownership structures matter to

how firms respond to the reform, we expect to observe an increase in overall business

activity in our sample (on average). Furthermore, using our detailed data, including

information on ownership shares and the role of the main owner in the firm, we can

directly test the significance of the ownership and management structures behind our

results. These results also increase our current knowledge of the impact of active main

owners behind firm-level decisions where earlier empirical evidence is still scarce.

4 Data, Methods and Identification

We use tax record data from the Finnish Tax Administration, including all Finnish

privately held corporations and partnerships in 2008–2016. These data enable us

to access a wide range of firm-level outcomes such as investments, sales, labor costs

and other input categories. In addition, we link firm owners to their firms using

unique identifiers. The owner-level data include information on, for example, income

withdrawn from the firm and the ownership share of the firm. We restrict our sample

to firms that are owned by individuals, and exclude firms that are owned by other

firms, institutional investors, hedge funds, etc. We use these data in an unbalanced

panel form.

We use a differences-in-differences method to study the effects of the reform. The

main outcomes we study are investment per lagged capital, sales, variable costs, labor

costs and value added of the firm. Our baseline estimable equation is of the following

form:

Yit = α1 + α2(Treati × Postt) + βi + λt + εit (1)

where Y is the outcome variable of interest, i refers to a firm and t is time. The

treatment group (Treat) is privately held corporations and the control group is part-

nership firms. Post refers to the period after the corporate tax rate cut (from 2014

onward). We include firm and year-fixed effects (βi, λt) in the baseline estimation,

and ε is the error term. Standard errors are clustered at the firm level.

The main identifying assumption in our differences-in-differences approach is the

parallel trends assumption. This means that in the absence of the 2014 reform,

the development of firm outcomes would have been similar among corporations and

partnerships. To ensure that the firms are comparable between the organizational

forms, we apply two modifications: we restrict our sample to small firms and use a

71

re-weighting strategy.

First, our baseline sample in each year includes firms with previous year’s annual

sales between 100,000 and approximately 2,500,000 euros and net assets below 750,000

euros. We restrict the sample due to the much longer tail of the size distribution of

corporations compared to partnerships, stemming from the fact that the majority of

large firms in Finland are corporations. Therefore, partnership firms offer a represen-

tative comparison group only for relatively small corporations. Furthermore, we drop

the smallest firms from the sample in order to focus on full-time businesses and more

established firms. To be precise, the upper sales limit is based on the 99th percentile

of the sales of partnership firms (2,503,624 euros). The net asset limit is set to ensure

that firms in our sample are affected by the owner-level taxes in a similar way, i.e.

we do not include the owners of corporations with large net assets (above 750,000

euros) whose effective dividend tax rate was reduced in 2014, as mentioned above in

Section 2 and discussed in greater detail in Appendix B. Table 1 presents the number

of observations dropped as a result of implementing these restrictions.3

Table 2 presents the summary statistics of our main outcome variables for the

restricted sample in 2011. Investments refer to the purchase price of all newly installed

capital assets. These include, for example, investments in machinery, equipment,

buildings and research and development.4 On average, annual investments relative

to the existing capital stock in the year before, a variable that has been widely used

to study firm-level investment responses in the literature (see e.g. Yagan 2015), were

0.23 for corporations and 0.20 for partnerships. In addition, as shown in Figure 4,

roughly 60% of corporations and 50% of partnerships had positive investments in

2011, and these shares have remained relatively constant throughout the time period

we study. The majority of investments for both corporations (81% of all investments)

and partnerships (86%) are concentrated in fixed assets such as machinery, equipment

and hardware.

Furthermore, in Table 2, labor costs include gross wages and other related costs

such as mandatory pension contributions (excluding individual-level income taxes).

Variable costs refer to all costs other than fixed costs and labor costs, including

3The exact sample restrictions do not affect our results in a meaningful manner, as is visible fromAppendix A Table 14. In the table we vary the sample restrictions by reducing and increasing theupper and lower sales limits of the sample by 20%, but these modifications do not significantly affectthe estimates of interest compared to our baseline specification introduced above.

4We winsorize the investment per lagged capital variable at a 95% percentile to avoid the potentialimpact of extreme outliers.

72

intermediate goods, materials and services used in production. Value added is defined

as sales minus variable costs. Table 3 presents the detailed definition of each variable

we use in our analysis.

Table 2 shows that while the organizational forms are relatively similar to each

other on average, the partnership firms in the restricted sample are still slightly

smaller. In addition, both corporations and partnerships operate in similar indus-

tries, as shown in Figure 1, but there are some differences in the shares of firms in

each industry category between the organizational forms. Therefore, in our baseline

regression specification, we follow Yagan [2015] and Zwick and Mahon [2017] and use

a non-parametric re-weighting strategy based on DiNardo et al. [1996] to control for

any size or industry-specific shocks. First, we assign each observation to one of the

ten industry categories presented in Figure 1. Then we divide each industry cate-

gory into four size groups based on the previous years’ annual sales. This creates

40 different industry-size bins for both organizational forms for each year. Following

the approach in Yagan [2015], the sum of sales in each bin is weighted to match the

base-year group of privately held corporations in 2011. Intuitively, the weight factor

is higher than one if the sum of sales of the firms in a group is lower than in the base

group, and vice versa.

In order to evaluate the validity of our empirical approach, we examine the evolu-

tion of our main outcome variables prior to the 2014 reform for both organizational

forms. As shown and discussed below in Section 5, the trends follow each other rea-

sonably well for all the main outcomes that we consider. The fact that the variables

follow parallel trends prior to the reform mitigates the concern that small firms would

not be comparable between the organizational forms. In addition, we control for the

potential impact of the smaller corporate tax cut in 2012 by adding a separate control

for this for the treated corporations, where we interact the treatment status with the

years right after this reform (Treati × Y ear2012,2013).

Finally, the corporate tax rate cut may have increased incentives for existing

partnership firms to change their organizational form to corporations, if the potential

gains from the 2014 reform in terms of changing the organization form exceed the

administrative costs of this change. This would be a concern for our empirical set-

ting if such changes in the organizational form were prevalent. However, Figure 9 in

Appendix A shows that this is not the case. We observe only 0.2–0.3% of partnership

firms changing their organizational form from partnership to privately held corpora-

tion each year, and more importantly, there is no change in this share at the time of

73

the reform. Therefore, we believe that this is not an issue for our empirical analysis.

The firms that changed their organizational form are included in our sample, and in

the regression we include an indicator variable denoting whether a firm has changed

its organizational form. However, dropping firms that changed their organizational

form does not affect our results in any significant manner.

5 Results

5.1 Investments

Figure 2 shows the development of investments per lagged capital in 2008–2016 for

corporation and partnership firms using the re-weighting strategy described in Sec-

tion 4. First, the figure illustrates that there is no difference in the development of

investments between corporations and partnerships before the tax rate cut in 2014,

demonstrating that the pre-reform trends are parallel between the groups. After the

reform, the development of investments remains parallel between the firm groups, in-

dicating no significant investment response to the corporate tax cut for corporations.

In addition, the figure shows no response to the smaller corporate tax rate cut imple-

mented in 2012, two years prior to the larger reform. Figure 3 shows the associated

differences-in-differences estimates in 2008–2016 relative to year 2011, which support

the visual observations from Figure 2.

Table 4 quantifies the estimates using various regression specifications following

the estimation strategy presented in Section 4. Column (1) shows the estimates with

firm and year-fixed effects and controlling for the 2012 reform, and column (2) when

also using the re-weighting approach. The differences-in-differences estimates are

small, 0.8% and 2.7% relative to mean investments per lagged capital, and neither

of the estimates is statistically significant. Therefore, the regression results confirm

the graphical evidence presented above. Our implied estimates for the investment

elasticity with respect to the net-of-corporate-tax rate presented in Table 4 are 0.13

and 0.46, and statistically insignificant in both specifications. Also, our results allow

us to statistically rule out larger than 1.3 elasticity of investments with respect to the

net-of-corporate tax rate. Therefore, following the small average investment responses

presented above, we find a small estimate for the responsiveness of investment with

respect to the corporate net-of-tax rate.5

5In addition, we find no significant effects of the reform on various investment categories such as

74

In addition, the corporate tax rate cut might affect the number of firms making

investments. Figure 4 presents the share of firms with positive new investments

in both firm groups in 2008–2016, illustrating the extensive margin of investment

decisions. The figure shows no significant change in the share of corporations investing

after 2014. In addition, we have a small and insignificant differences-in-differences

estimate for corporations with positive investments relative to partnerships, further

suggesting no significant extensive margin responses to the 2014 reform.

Our findings on the small responsiveness of investments are in contrast to a large

body of earlier quasi-experimental literature that estimates distinctively large in-

vestment elasticities with respect to the cost of capital, ranging from 6 to 7 (see e.g.

Maffini et al. 2019, Ohrn 2018, Zwick and Mahon 2017, and House and Shapiro 2008).

Therefore, our results illustrate that the average responses of small firms to a cut in

a corporate tax rate can be very different and lead to different policy implications

compared to the findings in the previous quasi-experimental literature. However, an

important feature of our setup in contrast to many of the earlier studies is that the

dividend tax rate was increased at the same time as the firm-level tax cut, indicat-

ing that the owner-level incentives to increase investments were mitigated (at least

partly). As discussed above in Section 3, this implies that if the investments of small

firms are financed primarily by new equity, the observed small investment response

can potentially be explained by a negligible change in the (effective) investment in-

centives of the owners.

Nevertheless, as discussed in Section 3, the simultaneous increase in the dividend

tax rate is irrelevant in the light of investment incentives if investments are financed

by retained earnings or debt (at the margin). This assertion is further supported

by a recent finding by Yagan [2015], who studies the massive 2003 dividend tax cut

in the US and detects no effect on new investments from this reform. Therefore,

another feasible explanation for our finding is that the investments of small firms are

in general less responsive to changes in firm-level financial incentives compared to the

larger firms that are typically targeted in the reforms analyzed in the earlier literature.

Importantly, this argument is supported by our results in the next section, where we

observe that corporations significantly increased their sales and input use after the

reform, which is inconsistent with the hypotheses following the old view that the

2014 reform would have no impact on incentives. Also, significant responses on other

outcomes than investments after 2014 imply that the negligible average investment

plants, machinery or hardware, but the estimates for the smaller categories are imprecisely measured.

75

effect cannot be easily explained by unawareness or inattention to the 2014 reform.

We provide further suggestive evidence to support the above assertions by dividing

our sample by the age of the firm. As discussed in Section 3, it has been argued that

young and newly established firms are more prone to raising new equity instead of

using debt or retained earnings to finance their new investments (see e.g. Auerbach

1979). Therefore, as we are studying relatively small firms, it could be that young

firms relying mostly on new equity are driving the small average response obtained.

However, we find no evidence that the investment response would be larger for older

firms when we split our sample by the age of the firm in Table 5. In contrast, younger

firms (under 10 years) seem to be more responsive in terms of new investments after

the tax rate cut compared to older firms (over 10 years), but the estimate for young

firms is still rather small (0.03) and statistically insignificant.6 Therefore, the divided

sample result based on the age of the firm does not support the role of the old view

in explaining our results.

However, the results above illustrate that new investments by younger firms might

be more responsive to relaxed financial constraints at the firm-level. Thus, these re-

sults tentatively suggest that reduced corporate tax rates can help to boost the in-

vestments of younger firms that might have more profitable investment opportunities

available but more limited resources to finance them. This result is also in line with

Benzarti et al. [2020], who show that younger firms are more responsive to financial

incentives compared to older firms. We study the impacts of other potential mecha-

nisms, such as cash constraints and the role of the main owner in the firm, in Section

5.3.

One further potential feature that could explain our more moderate investment

estimate compared to the earlier quasi-experimental studies is that in our setup all

firms face a cut in the corporate tax rate regardless of their investment behavior,

whereas the special tax deductions and accelerated depreciations utilized in the pre-

vious studies cited above are typically granted only for new investments that fall into

the specific program categories, and the tax reductions inflicted by these programs

are materialized only when and if a firm invests. By affecting the relative price of

investment as well as the relative price between different investment projects, such

targeted stimulus policies are potentially more likely to spark larger observed invest-

ment elasticities compared to a universal cut in the corporate tax rate. Hence, the

6We define the age of the firm based on the year 2013 such that firms older than 10 years wereobserved in the tax record data in 2002 or earlier.

76

implications of these various types of reforms (general corporate tax rate cut vs. spe-

cial tax reductions) could be different even though their theoretical effects on the cost

of capital might be similar.

Finally, our setup provides novel evidence of the investment effects of a reform

including a shift of the tax burden from the corporate to the owner level (see e.g.

Grubert and Altshuler 2016 and Devereux 2019). In principle, such a reform could

provide a way to increase investments in a cost-effective manner if reduced firm-level

financial incentives are the key factors explaining the investment decisions of firms,

and if the increased owner-level taxes do not distort investment decisions with a

similar magnitude. However, our evidence here shows that such a reform does not

appear to increase investment, at least among smaller firms.

To sum up, we find that the corporate tax rate cut did not cause a significant

increase in investments among small firms, at least in a context of a simultaneous

increase in the dividend tax rate of the owner. Nevertheless, as we only focus on

relatively small firms, we of course cannot rule out the possibility that larger firms

not included in our analysis that might, for example, have less limited access to debt

financing, could have increased their investments because of the tax cut.

5.2 Business Activity

Next, we focus on other outcomes that reflect the overall business activity of firms.

Figure 5 shows the development of sales, variable costs, labor costs and value added

(sales minus variable costs) for corporations and partnerships using the re-weighting

strategy. First, the figure confirms that the comparison between partnerships and

corporations is feasible, as the trends in the outcomes follow each other closely before

the reform. The only small deviation from this pattern is in 2009, when the evolution

of the outcomes for corporations were slightly behind those of partnership firms.7

The firm-level sales of corporations increased significantly relative to partnerships

right after the tax cut in 2014. Also, we find a similar increase in variable costs, and

a small increase in labor costs. The observed increase in sales, therefore, suggests

that the tax cut boosted overall firm production, as there were no other significant

7However, this minor deviation does not significantly affect the interpretation of our results.Table 15 in Appendix A shows that the differences-in-differences estimates for our main outcomesare very similar when we exclude the years 2008 and 2009 from the sample and focus on comparingthe three years prior to 2014 to the three years after the tax cut. This implies that the minordeviation in 2009 from the overall trend does not drive our results. Also, there is no deviation in thetrend of investments in Figure 2, illustrating that the groups do not systematically differ in 2009.

77

simultaneous shocks in the Finnish economy that could be clearly associated with the

increase in the business activity of small corporations relative to similar-sized part-

nership firms operating in similar industries. Moreover, an increase in sales together

with a simultaneous increase in variable costs indicates that the activity increase is

likely to represent a real response instead of a potential reporting response related to

tax evasion. Furthermore, we observe no responses in sales or other variables to the

smaller corporate tax rate cut implemented in 2012, as the average effects are clearly

visible only after the larger tax cut in 2014. In addition, both sales and variable costs

increased gradually over time after the reform instead of a sharp and discontinuous

jump right after 2014, indicating that the impact of the tax cut on business activ-

ity materialized gradually. Figure 6 presents the associated differences-in-differences

estimates in 2008–2016 relative to 2011, which confirm the visual observations.

Table 6 presents the differences-in-differences estimates for all four outcomes.

These results confirm the visual observations above: there is a statistically signifi-

cant increase in sales and variable costs after the reform. In terms of magnitudes, the

sales response represents a 3.2% increase relative to the mean sales of corporations,

and a 6.1% increase in variable costs. Furthermore, the point estimates for both labor

costs and value added are positive but not statistically significant.8 Overall, clear re-

sponses in other economic activity measures than capital investments illustrate that

the corporate tax cut has spurred the growth of small corporations.

Our results above show that small corporations responded to the corporate tax

rate cut by increasing their overall production, but not by increasing their investments

in physical capital. An increase in sales and inputs without an increase in investment

may seem unintuitive given that there was no significant reduction in the owner-level

tax burden because of the increased dividend tax rate. However, the corporate tax

cut still creates a mechanical cash injection into the firm through increased net-of-

tax retained earnings. Such additional resources can be particularly important for

smaller firms that might have limited opportunities to acquire other types of funding

such as new equity and debt, but can still have new business opportunities to utilize.

Therefore, the sales and inputs of small firms could still be affected by the reform even

8Our measure for labor costs includes wages paid to the main owner. The point estimate forlabor costs excluding the owner’s wage is positive and remains insignificant, but is approximatelyhalf the size of the estimate including all wages paid by the firm. Also, in Table 16 in Appendix A,we winsorize our outcome variables from both ends of the distribution in order to avoid the potentialeffects of outliers. Overall, the results using these specifications are similar but the estimates aresmaller compared to those in our baseline analysis.

78

without an effect on physical capital investments that typically require much larger

resources. Below we discuss in greater detail the mechanisms that might further

explain these results.

5.3 Heterogeneity and Potential Mechanisms

First, as discussed above in Section 3, improved firm-level financial incentives might

be more relevant to firms with more closely connected owners. It has been argued that

these owner-managers are more actively involved in firm-level decisions and business

activity compared to more passive owners or investors (see e.g. Chetty and Saez

2010), suggesting that they would also respond more strongly to improved firm-level

financial incentives. Also, closely connected owners are presumably more able to

utilize their firm and its cash reserves as a source of private spending (La Porta

et al. 2000; Bennedsen and Wolfenzon 2000) and engage in dynamic tax-planning

procedures in order to minimize long-run tax payments (le Maire and Schjerning

2013). These suggest that a cut in the corporate tax rate can also affect the labor

input and effort of active owner-managers. As our sample includes only relatively

small firms with, on average, only six employees, the role of the main owner is likely

to have a particularly significant impact on our findings.

We test the significance of the role of the owner in greater detail by studying the

impact of the reform separately for firms with more active and passive main owners.

In our data, we observe whether the owner of a firm is an active manager-worker or

a more passive investor. Active owner-managers include those who, by themselves

or together with family members, own at least 30% of the corporation and hold

an active leading position in their firm, such as CEO or chairman of the board.

This classification is based on mandatory pension insurance regulations in Finland,

as these active owner-managers are insured under different regulations compared to

other corporate owners. Therefore, we have a reliable register-based indicator for

classifying entrepreneurs into more active and more passive owners.9

Figure 7 plots the difference-in-difference estimates for investments, sales, variable

costs, and labor costs for corporations with active owner-managers in comparison to

partnerships, and the same outcomes comparing corporations with more passive own-

ers to partnerships. The figure shows that corporations with active owner-managers

9See Benzarti et al. [2020] for more details on pension insurance regulations and evidence of howentrepreneurs in Finland respond to changes in insurance contribution rules.

79

are the group that increased their business activity after the reform, as we find a sig-

nificant increase in sales and variable costs for them. In contrast, there is no response

from corporations with more passive owners. However, we find no statistically signifi-

cant impact on investments for either of the groups, indicating that the small observed

average response in investments does not mask a larger response for firms with more

active owners. Tables 8 and 9 present the associated regression estimates using the

full set of controls for all of the outcomes, which confirm the visual observations from

Figure 7.

These results suggest that the observed responses to the tax cut are closely linked

to the role of the owner in the firm, and show that passive and active owners consti-

tute an important division in distinguishing responses to firm-level incentives. This

evidence contributes to the scarce empirical literature on the role of the main owner

behind firm-level decisions, and is in line with the recent paper by Smith et al. [2019]

highlighting the role of the main owner behind firm decisions and performance in the

US.10

Furthermore, as we focus on small firms, a notable share of them may face cash

constraints. Retained earnings largely contribute to the firm’s cash reserves, and

after the tax cut the net-of-tax retained earnings of firms increased mechanically

even absent any behavioral responses. Therefore, as a result of such a cash injection,

liquidity constraints may be loosened and corporations could now have more liquid

funds to boost their business activity. Correspondingly, this would show up as an

increased output by the firm in our sample.

Next, we study whether more cash-constrained firms drive the observed average

response. We define a firm-level cash constraint measure based on the average share

of liquid financial assets (cash, cash at bank, stock holdings, and other liquid funds)

relative to the total assets of the firm in 2010–2011, and divide our sample into

firms below and above the median (0.26) of this variable. Figure 10 in Appendix A

presents the results using this division, showing that there are no significant differences

in responses by cash-to-assets ratios. Tables 10 and 11 in Appendix A present the

associated differences-in–differences estimates using the full set of controls, which

confirm the visual observations from the figure. Therefore, according to these results,

firm-level liquidity constraints alone appear not to explain our findings of increased

10Similarly to our results on investments and the role of the main owner, a recent working paperby Moon [2020] finds no statistically significant difference between the investment responses of firmswith different ownership structures to changes in capital gains taxes in South Korea.

80

business activity after the tax cut.

However, liquidity constraints might affect the responses differently for firms with

active vs. passive owners if cash-constrained active owners are more eager to use the

extra cash to boost business activity. Figure 11 in Appendix A presents the results

for firms with active owners separately for more and less cash-constrained firms us-

ing similar sample division rules as above. The figure shows that the responses for

active owner-managers are significant regardless of cash constraints, but the impact

of the tax cut is greater for firms with smaller cash-to-assets ratios. This suggests

that the additional cash injection after the tax rate cut spurred the business activity

of more cash-constrained firms, but only when the firm is actively managed by the

owner. Also, the figure shows that cash-constraint firms with active owner-managers

already appeared to increase their sales and variable costs in 2012 when the corpo-

rate tax rate was first reduced by 1.5 percentage points. Therefore, these results

again highlight the key role played by active owners behind the responses of small

firms to financial incentives. Tables 12 and 13 in Appendix A present the associated

differences-differences estimates using the full set of controls, which confirm the visual

observations.

All of the specifications above included firm fixed effects, concealing a potential

entry effect of new firms. In order to ensure that we are not disregarding any potential

effects on new business creation, we also examine the entry decisions of firms. Figure

8 plots the share of new corporations and partnerships relative to existing firms, and

Table 7 tabulates the number of all firms and the share of new firms in 2008–2016.

Overall, over the last decade, the privately held corporation has been a more popular

organizational form for new firms than partnerships in Finland, but the share of

new firms has been decreasing in both groups over time.11 There appears to be no

significant change in this longer-run trend that could be clearly associated with the

2014 corporate tax cut. Therefore, we conclude that the corporate tax cut did not

accelerate new business creation, although this evidence is mostly descriptive.12

Finally, one additional behavioral margin that could be relevant to firm owners

is income-shifting between tax bases. Such responses are well documented in the

11Firm entry rates have also been declining over the last decade in many other developed countriesincluding the US, as documented and discussed in e.g. Decker et al. [2016a].

12In addition, the corporate tax rate cut could affect the exit of firms if, for example, larger firmsbought out smaller firms to a larger extent after the tax cut than before it. However, Table 7 showsthat there is no clear reduction in the number of existing firms after 2014, suggesting that this typeof effect does not appear to be significant.

81

recent empirical literature studying the effects of income taxes for business owners

and entrepreneurs (see e.g. Harju and Matikka 2016 for Finland and le Maire and

Schjerning 2013 for Denmark). However, as discussed above in Section 2, tax incen-

tives for withdrawing dividend (or wage) income from firms did not change much in

the 2014 reform for the owners of small corporations. To illustrate that the reform

did not create any changes in income composition, Figure 12 in Appendix A plots

the average share of dividends of total income withdrawn from the firm (wages +

dividends) for the main owners of corporations in our baseline sample. Evidently, the

share of dividend income has been relatively constant before and after the reform,

implying no income-shifting responses, as expected.13

6 Discussion

In this paper, we provide novel evidence for how small, privately held corporations

responded to a sizeable 4.5 percentage-point corporate tax rate cut implemented

together with a dividend tax increase at the owner level. We find no significant

average impact of the tax cut on investments per lagged capital when comparing

small corporations to otherwise similar partnership firms that were unaffected by

the tax cut. This suggests that reducing the corporate tax rate is not an effective

measure to spur capital investments among small firms, at least when combined with

a simultaneous dividend tax increase. However, we find a clear increase in sales and

input usage after the tax cut, illustrating that the owners of small firms are responsive

to corporate taxes and that small firms may also respond to firm-level tax cuts on

other relevant margins than physical investments. We find that the observed effects

are clearly driven by firms that are owned by active owner-managers rather than more

passive investors. This suggests that the role played by the owner in the firm is an

important factor in explaining how small firms respond to financial incentives.

Nevertheless, more empirical and theoretical research is needed in the future to

better understand how small and growing firms and their owners respond to changes

in financial incentives, and what the important drivers of these responses might be.

Declining firm entry rates and productivity in many developed countries (see e.g.

13The income tax system for the owners of partnership firms does not include similar income-shifting incentives as they do not have the opportunity to withdraw income from their firms in theform of dividends. Therefore, the owners of partnership firms are unable to directly shift incomefrom one tax base to another, and thus we cannot plot a corresponding measure for them in thefigure.

82

Decker et al. 2016b) pose serious economic challenges that further underline the need

for additional research on the incentives affecting young and growing firms.

83

References

Alan Auerbach. Wealth maximization and the cost of capital. Quarterly Journal of

Economics, 93(3):433–446, 1979.

Morten Bennedsen and Daniel Wolfenzon. The balance of power in closely held

corporations. Journal of Financial Economics, 58(1):113–139, 2000. ISSN 0304-

405X. doi: https://doi.org/10.1016/S0304-405X(00)00068-4.

Youssef Benzarti, Jarkko Harju, and Tuomas Matikka. Does mandating social in-

surance affect entrepreurial activity? American Economic Review: Insights, 2(2):

255–268, 2020.

Stephen Bond and Jing Xing. Corporate taxation and capital accumulation: Evidence

from sectoral panel data for 14 OECD countries. Journal of Public Economics, 130:

15–31, 2015. ISSN 0047-2727. doi: https://doi.org/10.1016/j.jpubeco.2015.08.001.

David Bradford. The incidence and allocation effects of a tax on corporate distribu-

tions. Journal of Public Economics, 15:1–22, 1981.

Raj Chetty and Emmanuel Saez. Dividend and corporate taxation in an agency model

of the firm. American Economic Journal: Economic Policy, 2(3):1–31, August 2010.

doi: 10.1257/pol.2.3.1.

Ryan Decker, John Haltiwanger, Ron Jarmin, and Javier Miranda. The role of en-

trepreneurship in US job creation and economic dynamism. Journal of Economic

Perspectives, 28(3):3–24, September 2014. doi: 10.1257/jep.28.3.3.

Ryan Decker, John Haltiwanger, Ron Jarmin, and Javier Miranda. Where

has all the skewness gone? The decline in high-growth (young)

firms in the U.S. European Economic Review, 86:4–23, 2016a. doi:

https://doi.org/10.1016/j.euroecorev.2015.12.013.

Ryan Decker, John Haltiwanger, Ron Jarmin, and Javier Miranda. Declining business

dynamism: What we know and the way forward. American Economic Review, 106

(5):203–207, 2016b.

Michael Devereux. How should business profit be taxed? Some thoughts on concep-

tual developments during the lifetime of the IFS. Fiscal Studies, 40(4):591–619,

2019.

John DiNardo, Nicole M. Fortin, and Thomas Lemieux. Labor market institutions and

the distribution of wages, 1973-1992: A semiparametric approach. Econometrica,

64(5):1001–1044, 1996.

84

M. S. Feldstein. Corporate taxation and dividend behaviour. The Review of Economic

Studies, 37(1):57–72, 1970. doi: 10.2307/2296498.

Clemens Fuest, Andreas Peichl, and Sebastian Siegloch. Do higher corporate taxes

reduce wages? Micro evidence from Germany. American Economic Review, 108

(2):393–418, February 2018. doi: 10.1257/aer.20130570.

Harry Grubert and Rosanne Altshuler. Shifting the burden of taxation from the

corporate to the personal level and getting the corporate tax rate down to 15

percent. National Tax Journal, 69(3):643–676, September 2016.

Arnold Harberger. The incidence of the corporation income tax. Journal of Political

Economy, 70, 1962.

Jarkko Harju and Tuomas Matikka. The elasticity of taxable income and income-

shifting between tax bases: What is real and what is not? International Tax and

Public Finance, 23(4):640–669, 2016.

Christopher L. House and Matthew D. Shapiro. Temporary investment tax incentives:

Theory with evidence from bonus depreciation. American Economic Review, 98(3):

737–68, June 2008. doi: 10.1257/aer.98.3.737.

Micheal Jensen and William Meckling. Theory of the firm: Managerial behavior,

agency costs and ownership structure. Journal of Financial Economics, 3(4):305–

360, 1976.

Rafael La Porta, Florencio Lopez-de Silanes, Andrei Shleifer, and Robert W. Vishny.

Agency problems and dividend policies around the world. The Journal of Finance,

55(1):1–33, June 2000.

Daniel le Maire and Bertel Schjerning. Tax bunching, income shifting and self-

employment. Journal of Public Economics, 107(1):1–18, 2013.

Yongzheng Liu and Jie Mao. How do tax incentives affect investment and produc-

tivity? Firm-level evidence from China. American Economic Journal: Economic

Policy, 11(3):261–91, August 2019. doi: 10.1257/pol.20170478.

Giorgia Maffini, Jing Xing, and Michael P. Devereux. The impact of investment in-

centives: Evidence from UK corporation tax returns. American Economic Journal:

Economic Policy, 11(3):361–89, August 2019. doi: 10.1257/pol.20170254.

Terry Moon. Capital gains taxes and real corporate investment. Working Paper,

2020.

Eric Ohrn. The effect of corporate taxation on investment and financial policy: Ev-

idence from the DPAD. American Economic Journal: Economic Policy, 10(2):

272–301, May 2018. doi: 10.1257/pol.20150378.

85

Eric Ohrn. The effect of tax incentives on U.S. manufacturing: Evidence from state

accelerated depreciation policies. Journal of Public Economics, 180:104084, 2019.

ISSN 0047-2727. doi: https://doi.org/10.1016/j.jpubeco.2019.104084.

James M. Poterba and Lawrence H. Summers. The Economic Effects of Dividend

Taxation. Recent Advances in Corporate Finance, edited by Edward Altman and

Marti Subrahmanyam, pages pp. 377–380, 1985.

Joshua D. Rauh. Investment and financing constraints: Evidence from the fund-

ing of corporate pension plans. The Journal of Finance, 61(1):33–71, 2006. doi:

10.1111/j.1540-6261.2006.00829.x.

Hans-Werner Sinn. The vanishing harberger triangle. Journal of Public Eco-

nomics, 45(3):271–300, 1991. ISSN 0047-2727. doi: https://doi.org/10.1016/0047-

2727(91)90029-2.

Matthew Smith, Danny Yagan, Owen Zidar, and Eric Zwick. Capitalists in the

twenty-first century. Quarterly Journal of Economics, 134(4):1675–1745, 2019.

Juan Carlos Suarez Serrato and Owen Zidar. Who benefits from state corporate

tax cuts? A local labor markets approach with heterogeneous firms. American

Economic Review, 106(9):2582–2624, September 2016. doi: 10.1257/aer.20141702.

Danny Yagan. Capital tax reform and the real economy: The effects of the 2003

dividend tax cut. American Economic Review, 105(12):3531–63, December 2015.

doi: 10.1257/aer.20130098.

Eric Zwick and James Mahon. Tax policy and heterogeneous investment be-

havior. American Economic Review, 107(1):217–48, January 2017. doi:

10.1257/aer.20140855.

86

Figures

Figure 1: Shares of firms in different industries, corporations and partnerships0

2040

6080

100

2008 2010 2012 2014 2016

Corporations

020

4060

8010

0

2008 2010 2012 2014 2016

Partnerships

Farming, fishery, forestry Industry

Construction Wholesale and retail

Logistics Hospitality

Administration and information Finance and real estate

Education, heath, social service Art, entertainment and other

Note: Figure presents the shares of firms in different industry categories for both corporations andpartnerships in the restricted sample in 2008–2016.

87

Figure 2: The development of investment per lagged capital, 2008–2016

.15

.2.2

5.3

2008 2009 2010 2011 2012 2013 2014 2015 2016

Corporations Partnerships

Note: Figure plots the development of the ratio of investment per lagged capital assets in 2008–2016 for corporations and partnerships using the restricted sample and the re-weighting procedurepresented in Section 4 and firm-fixed effects. The first vertical line denotes the smaller corporatetax cut in 2012, and the second line the larger tax rate cut from 24.5% to 20% in 2014.

88

Figure 3: Differences-in-differences estimates for investment per lagged capital, 2008–2016

−.05

−.02

50

.025

.05

2008 2009 2010 2011 2012 2013 2014 2015 2016year

Diff−in−diff estimate CI

Note: Figure shows the differences-in-differences estimates between corporations and partnershipsand 95% confidence intervals for investments per lagged capital relative to year 2011 using thebaseline restricted sample in 2008–2016. The specification in the figure includes firm-fixed effectsand uses the re-weighting strategy presented in Section 4. The first vertical line denotes the smalltax cut in 2012, and the second line the larger tax rate cut from 24.5 to 20% in 2014.

89

Figure 4: Firms with positive investments (extensive margin), 2008–2016

.3.4

.5.6

.7.8

Shar

e of

firm

s in

vest

ing

2008 2009 2010 2011 2012 2013 2014 2015 2016

Corporations Partnerships

Extensive margin investment DiD−estimate: .011 (.006)

Note: Figure shows the shares of firms with positive annual investments for both corporations andpartnerships in 2008–2016, and a difference-in-difference estimate for corporations with positive in-vestments relative to partnerships estimated following the estimation procedure presented in Section4. The first vertical line denotes the small tax cut in 2012, and the second line the larger tax ratecut from 24.5 to 20% in 2014.

90

Figure 5: Development of sales, variable costs, labor costs and value added, 2008–2016

5000

0055

0000

6000

00

2008 2009 2010 2011 2012 2013 2014 2015 2016

Sales

2200

0024

0000

2600

00

2008 2009 2010 2011 2012 2013 2014 2015 2016

Variable costs

1400

0016

0000

1800

0020

0000

2008 2009 2010 2011 2012 2013 2014 2015 2016

Labor costs

3000

0032

0000

3400

00

2008 2009 2010 2011 2012 2013 2014 2015 2016

Value added

Corporations Partnerships

Note: Figure plots the development of sales, variable costs, labor costs and value added in 2008–2016for corporations and partnerships using the restricted sample and the re-weighting procedure withfirm-fixed effects. The first vertical line denotes the small tax cut in 2012, and the second line thelarger tax rate cut from 24.5% to 20% in 2014.

91

Figure 6: Differences-in-differences estimates for sales, variable costs, labor costs andvalue added, 2008–2016

−400

00−2

0000

020

000

4000

0

2008 2009 2010 2011 2012 2013 2014 2015 2016

Sales

−400

00−2

0000

020

000

4000

0

2008 2009 2010 2011 2012 2013 2014 2015 2016

Variable costs

−400

00−2

0000

020

000

4000

0

2008 2009 2010 2011 2012 2013 2014 2015 2016

Labor costs

−400

00−2

0000

020

000

4000

0

2008 2009 2010 2011 2012 2013 2014 2015 2016

Value added

Diff−in−diff estimate CI

Note: Figure shows the differences-in-differences estimates between corporations and partnershipsfor firm-level sales, variable costs, labor costs and value added relative to year 2011 using the baselinerestricted sample in 2008–2016. The specifications include firm fixed effects and uses the re-weightingstrategy presented in Section 4. The first vertical line denotes the small tax cut in 2012, and thesecond line the larger tax rate cut from 24.5 to 20% in 2014.

92

Figure 7: Differences-in-differences estimates for active owner-managers and otherowner types

−.06

−.04

−.02

0.0

2.0

4.0

6

2008 2009 2010 2011 2012 2013 2014 2015 2016

Investment per lagged capital

−500

00−2

5000

025

000

5000

0

2008 2009 2010 2011 2012 2013 2014 2015 2016

Sales

−500

00−2

5000

025

000

5000

0

2008 2009 2010 2011 2012 2013 2014 2015 2016

Variable costs

−500

00−2

5000

025

000

5000

0

2008 2009 2010 2011 2012 2013 2014 2015 2016

Labor costs

Active owner−managers Passive owners

Note: Figure shows the differences-in-differences estimates between corporations and partnership forinvestments per lagged capital, firm-level sales, variable costs and labor costs relative to year 2011separately for corporations with active owner-managers and other owner types using the baselinerestricted sample in 2008–2016. The regressions include firm fixed effects and use the re-weightingstrategy presented in Section 4. The first vertical line denotes the small tax cut in 2012, and thesecond line the larger tax rate cut from 24.5% to 20% in 2014.

93

Figure 8: Shares of new firms, 2008–2016

0.0

5.1

.15

.2

2008 2009 2010 2011 2012 2013 2014 2015 2016

new corporations new partnerships

Note: Figure shows the shares of new firms relative to the total number of existing firms for bothcorporations and partnerships in 2008–2016. We define a firm as a new firm based on the year whenthe firm is first observed in tax record data. The vertical line denotes the larger tax rate cut from24.5 to 20% in 2014.

94

Tables

Table 1: Sample restriction – number of observations excluded

Data restrictions 2008-2016

Total Below Above Net assets Sample

N Share N Share N Share N Share N

Corporation 537,501 100.0 138,688 25.8 130,714 24.3 25,468 4.7 242,631Partnership 303,901 100.0 161,513 53.1 21,162 7.0 7126 2.3 114,100

Note: Table presents the number and share of observations dropped from the full sample for bothorganizational forms using pooled data from 2008–2016. Table shows the total number of observa-tions for both organizational forms in the unrestricted full data, the number and relative share ofobservations dropped from below and above the previous year’s sales limit of 100,000–2,503,624 eu-ros, and when excluding firms with previous year’s net assets above 750,000 euros. The last columnshows the restricted baseline sample used in the main analysis.

95

Table 2: Summary statistics of the restricted sample, 2011

Corporations Partnershipsmean sd median mean sd median

Sales 549,886 584,125 351,281 340,902 373,850 218,716

Labor costs 173,887 203,327 110,680 78,538 110,732 43,850

Variable costs 251,447 393,944 109,749 160,717 266,671 77,598

Value added 325,327 334,899 225,957 190,498 186,642 135,852

Investment 31,510 88,951 2295 21,883 61,846 385

Investment per lagged capital 0.236 0.393 0.033 0.209 0.391 0

Observations 26,739 13,200

Note: Table shows the summary statistics in 2011 for corporations and partnerships for the restictedsample used in the main analysis.

96

Table 3: Definitions of the variables used in the analysis

Variable Definition

Sales Gross annual sales of the firm from its primary operatingactivity minus any discounts given, valued-added taxes, andother taxes based on sales volumes.

Investments Annual euro value of gross investments including all newlyinstalled capital assets, such as machinery, buildings andequipment.

Capital assets Capital assets include balance sheet information ofproductive capital such as machinery, buildings andequipment.

Labor costs Annual wages and other wage-related compensations paid bythe firm, including social insurance contributions paid onwage income but excluding income taxes.

Variable costs Annual euro value of the costs used as intermediate inputs inproduction, such as materials and services used.

Value added Annual euro value of sales minus variable costs.

Financial assets Financial assets of the firm including cash, cash at bank,stock holdings, and other liquid funds.

Age of the firm We measure the age of the firm using the first observation inour full tax record data starting from 1998. We define theage of the firm based on the year 2013 so that firms olderthan 10 years were observed in tax record data in 2002 orearlier.

Active owner-managers Active owner-managers include those owners of corporationswho, by themselves or together with family members, own atleast 30% of the firm and hold an active leading position intheir firm. This classification is based on the Finnish pensioninsurance regulations included in the data. Activeowner-managers are insured under different social insuranceregulations compared to other corporate owners.

97

Table 4: Differences-in-differences results: Investment per lagged capital

(1) (2)

Diff-in-diff estimate 0.002 0.006

(0.004) (0.006)

Firm fixed effects X X

Year fixed effects X X

Weighting X

Control for 2012 reform X X

Constant 0.302*** 0.304***

(0.005) (0.005)

R2 0.352 0.361

N 322,059 322,059

Elasticity 0.126 0.457

(0.300) (0.417)

* p < 0.05, ** p < 0.01, *** p < 0.001

Note: Table shows the differences-in-differences estimates and firm-level clustered standard errors(in parentheses) for investments per lagged capital with different estimation specifications followingthe estimation procedure presented in Section 4. Column (1) shows the results when the re-weightingstrategy is not used and when year and firm-fixed effects and the 2012 reform are controlled for.Column (2) shows the same result when the re-weighting strategy is used. In addition, the tableincludes the implied elasticity estimates of investments relative to the net-of-corporate tax rate andthe associated standard errors for both specifications. The elasticity is estimated using the effectrelative to the pre-reform mean of 0.23. None of the elasticity estimates are statistically differentfrom zero.

98

Table 5: Differences-in-differences results: Investment per lagged capital by firm age

Firms aged 0–10 Firms older than 10

Diff-in-diff estimate 0.033* 0.007

(0.013) (0.006)

Firm fixed effects X X

Year fixed effects X X

Weighting X X

Control for 2012 reform X X

Constant 0.345*** 0.281***

(0.009) (0.006)

R2 0.403 0.321

N 133,816 188,243

* p < 0.05, ** p < 0.01, *** p < 0.001

Note: Table shows the differences-in-differences estimates and firm-level clustered standard errors(in parentheses) for investments per lagged capital for firms of different ages following the estimationprocedure presented in Section 4. Column (1) shows the results for firms that are younger than orequal to 10 years old, and column (2) for firms that are older than 10 years. We define the age ofthe firm based on the year 2013 so that firms older than 10 years were observed in tax record datain 2002 or earlier.

99

Table 6: Differences-in-differences results: Business activity outcomes

Sales Var.costs Labor costs Value added

Diff-in-diff estimate 17,687** 15,381*** 4851* 4976

(6053) (4581) (2397) (3641)

Firm fixed effects X X X X

Year fixed effects X X X X

Weighting X X X X

Control for 2012 reform X X X X

Constant 595,592*** 269,532*** 181,101*** 346,865***

(4342) (2624) (1517) (2602)

R2 0.851 0.828 0.902 0.867

N 356,582 331,552 343,481 331,552

* p < 0.05, ** p < 0.01, *** p < 0.001

Note: Table shows the difference-in-differences estimates and firm-level clustered standard errors (inparentheses) for firm-level sales, variable costs, labor costs and value added following the estimationprocedure presented in Section 4. All regressions include firm and year fixed effects, the re-weightingprocedure, and controls for the 2012 reform. The labor cost variable in the table includes the wagespaid to the main owner.

100

Table 7: New firms of all firms (unrestricted data)

Corporate form

Year Old corp. Old part. New corp. New part. Total

No. Row%

No. Row%

No. Row%

No. Row%

No. Row%

2008 46656 50.9 34344 37.4 7517 8.2 3220 3.5 91737 100.02009 49052 53.1 34445 37.3 6868 7.4 2063 2.2 92428 100.02010 51295 55.4 33543 36.2 5818 6.3 2010 2.2 92666 100.02011 53865 57.0 32975 34.9 5774 6.1 1907 2.0 94521 100.02012 56011 57.8 32363 33.4 6658 6.9 1813 1.9 96845 100.02013 59348 60.1 31542 31.9 6338 6.4 1592 1.6 98820 100.02014 55642 60.8 30640 33.5 3888 4.2 1364 1.5 91534 100.02015 56102 61.4 29582 32.4 4475 4.9 1193 1.3 91352 100.02016 57443 62.8 28241 30.9 4753 5.2 1064 1.2 91501 100.0Total 485414 57.7 287675 34.2 52089 6.2 16226 1.9 841404 100.0

Note: Table shows the number and relative share of existing corporations (Old corp.) and part-nerships (Old part.) and the number and share of new corporations (New corp.) and partnerships(New part.), and the total number of firms in 2008–2016 in the unrestricted full data.

101

Table 8: Differences-in-differences results: Firms with active owner-managers

Investment Sales Var.costs Labor costs Value added

Diff-in-diff estimate 0.001 26,212*** 17,480*** 6934** 9160**

(0.006) (5500) (3750) (2242) (3548)

Firm fixed effects X X X X X

Year fixed effects X X X X X

Weighting X X X X X

Control for 2012 reform X X X X X

Constant 0.295*** 474,422*** 213984*** 106,940*** 273,074***

(0.013) (11,422) (6560) (3987) (6590)

R2 0.369 0.864 0.844 0.908 0.867

N 264,307 291,181 272,374 284,598 272,374

* p < 0.05, ** p < 0.01, *** p < 0.001

Note: Table shows the difference-in-differences estimates and firm-level clustered standard errors (inparentheses) for firm-level investment per lagged capital, sales, variable costs, labor costs and valueadded including only corporations with active owner-managers. All regressions follow the estimationprocedure presented in Section 4 and include firm and year fixed effects, the re-weighting procedure,and controls for the 2012 reform.

102

Table 9: Differences-in-differences results: Firms with other owners

Investment Sales Var.costs Labor costs Value added

Diff-in-diff estimate 0.005 -18,259 -519 -4713 -10,912*

(0.008) (9668) (7750) (3685) (5404)

Firm fixed effects X X X X X

Year fixed effects X X X X X

Weighting X X X X X

Control for 2012 reform X X X X X

Constant 0.240*** 532,979*** 251,938*** 133,537*** 298,079***

(0.018) (18,998) (9604) (8018) (12,701)

R2 0.417 0.865 0.832 0.923 0.912

N 162,055 179,502 166,772 168,961 166,772

* p < 0.05, ** p < 0.01, *** p < 0.001

Note: Table shows the difference-in-differences estimates and firm-level clustered standard errors(in parentheses) for firm-level investment per lagged capital, sales, variable costs, labor costs andvalue added including only corporations with passive owners. All regressions follow the estimationprocedure presented in Section 4 and include firm and year fixed effects, the re-weighting procedure,and controls for the 2012 reform.

103

A Appendix: Additional Figures and Tables

Figure 9: The annual share of partnership firms that changed their organizationalform to a corporation, 2008–2016

0.0

025

.005

.007

5.0

1

2008 2009 2010 2011 2012 2013 2014 2015 2016

Note: Figure shows the share of partnership firms that changed their organizational form to acorporation in 2008–2016. The first vertical line denotes the small tax cut in 2012, and the secondline the larger tax rate cut from 24.5 to 20% in 2014. The figure illustrates that this share is verysmall, around 0.25% each year, and that there is no significant change in the share at the time ofthe 2014 corporate tax cut.

104

Figure 10: Differences-in-differences estimates for firms with a cash-to-assets ratiobelow and above the median

−.06

−.04

−.02

0.0

2.0

4.0

6

2008 2009 2010 2011 2012 2013 2014 2015 2016

Investment per lagged capital

−500

00−2

5000

025

000

5000

0

2008 2009 2010 2011 2012 2013 2014 2015 2016

Sales

−500

00−2

5000

025

000

5000

0

2008 2009 2010 2011 2012 2013 2014 2015 2016

Variable costs

−500

00−2

5000

025

000

5000

0

2008 2009 2010 2011 2012 2013 2014 2015 2016

Labor costs

Cash−to−assets ratio below median Cash−to−assets ratio above median

Note: Figure shows the differences-in-differences estimates between corporations and partnershipsfor investments per lagged capital, firm-level sales, variable costs and labor costs relative to year2011 separately for firms with a cash-to-assets ratio below the median (0.26) and a cash-to-assetsratio above the median using the baseline restricted sample in 2008–2016. The specifications includefirm fixed effects and use the re-weighting strategy presented in Section 4. The first vertical linedenotes the small tax cut in 2012, and the second line the larger tax rate cut from 24.5% to 20% in2014.

105

Figure 11: Differences-in-differences estimates for firms with active owner-managersand cash-to-assets ratio above and below median

−.06

−.04

−.02

0.0

2.0

4.0

6

2008 2009 2010 2011 2012 2013 2014 2015 2016

Investment per lagged capital

−500

00−2

5000

025

000

5000

0

2008 2009 2010 2011 2012 2013 2014 2015 2016

Sales

−500

00−2

5000

025

000

5000

0

2008 2009 2010 2011 2012 2013 2014 2015 2016

Variable costs

−500

00−2

5000

025

000

5000

0

2008 2009 2010 2011 2012 2013 2014 2015 2016

Labor costs

Active owner−manager & more cash−constrained Active owner−manager & less cash−constrained

Note: Figure shows the differences-in-differences estimates between corporations and partnershipsfor investments per lagged capital, firm-level sales, variable costs and labor costs relative to year2011 for two groups of corporations with active owner-managers; those with a cash-to-assets ratioabove median and those with a cash-to-assets ratio below the median, using the baseline restrictedsample in 2008–2016. The regressions include firm fixed effects and use the re-weighting strategypresented in Section 3. The first vertical line denotes the small tax cut in 2012, and the second linethe larger tax rate cut from 24.5 % to 20% in 2014.

106

Figure 12: The average share of dividends of total income withdrawn from the firm(wages + dividends), 2008–2016

.2.3

.4.5

.6

2008 2009 2010 2011 2012 2013 2014 2015 2016

Share of dividend income

Note: Figure plots the average share of dividends of total income (dividends + wages) withdrawnfrom the firm for the owners of privately held corporations in the baseline restricted sample in 2008–2016. The first vertical line denotes the small tax cut in 2012, and the second line the larger taxrate cut from 24.5 to 20% in 2014.

107

Table 10: Differences-in-differences results: Firms with cash-to-assets ratio belowmedian

Investment Sales Var.costs Labor costs Value added

Diff-in-diff estimate 0.008 22,726** 14,986** 5950* 8533

(0.008) (8496) (5681) (2957) (5026)

Firm fixed effects X X X X X

Year fixed effects X X X X X

Weighting X X X X X

Control for 2012 reform X X X X X

Constant 0.292*** 550,668*** 260,860*** 124,344*** 295,694***

(0.018) (19,148) (11,131) (6397) (10,346)

R2 0.362 0.857 0.827 0.901 0.862

N 151,140 164,049 155,746 158,607 155,746

* p < 0.05, ** p < 0.01, *** p < 0.001

Note: Table shows the difference-in-differences estimates and firm-level clustered standard errors (inparentheses) for investment per lagged capital, firm-level sales, variable costs, labor costs and valueadded for firms with a cash-to-asset ratio below the median (0.26) in the restricted sample. Allregressions follow the estimation procedure presented in Section 4 and include firm and year fixedeffects, the re-weighting procedure, and controls for the 2012 reform. The labor cost variable in thetable includes the wages paid to the main owner.

108

Table 11: Differences-in-differences results: Firms with cash-to-assets ratio abovemedian

Investment Sales Var.costs Labor costs Value added

Diff-in-diff estimate 0.004 14,326* 12,577* 6095 3486

(0.008) (7234) (4992) (3987) (5553)

Firm fixed effects X X X X X

Year fixed effects X X X X X

Weighting X X X X X

Control for 2012 reform X X X X X

Constant 0.260*** 409,745*** 175,925*** 96,502*** 253,274***

(0.018) (14,104) (7403) (6143) (9971)

R2 0.360 0.852 0.830 0.904 0.871

N 167,880 188,362 174,835 182,987 174,835

* p < 0.05, ** p < 0.01, *** p < 0.001

Note: Table shows the difference-in-differences estimates and firm-level clustered standard errors (inparentheses) for investment per lagged capital, firm-level sales, variable costs, labor costs, and valueadded for firms with a cash-to-asset ratio above the median (0.26) in the restricted sample. Allregressions follow the estimation procedure presented in Section 4 and include firm and year fixedeffects, the re-weighting procedure, and controls for the 2012 reform. The labor cost variable in thetable includes the wages paid to the main owner.

109

Table 12: Differences-in-differences results: Firms with active owner-managers andcash-to-assets ratio below median

Investment Sales Var.costs Labor costs Value added

Diff-in-diff estimate 0.002 33,570*** 20,878*** 7954** 12,851*

(0.008) (8569) (5793) (2936) (5211)

Firm fixed effects X X X X X

Year fixed effects X X X X X

Weighting X X X X X

Control for 2012 reform X X X X X

Constant 0.309*** 549,088*** 258,954*** 120,620*** 297,083***

(0.018) (17,655) (10,228) (6054) (9828)

R2 0.369 0.870 0.840 0.910 0.862

N 123,462 133,466 127,110 130,911 127,110

* p < 0.05, ** p < 0.01, *** p < 0.001

Note: Table shows the difference-in-differences estimates and firm-level clustered standard errors(in parentheses) for investment per lagged capital, firm-level sales, variable costs, labor costs, andvalue added for corporations with active owner-managers and firms with a cash-to-assets ratio belowmedian. All regressions follow the estimation procedure presented in Section 4 and include firm andyear fixed effects, the re-weighting procedure, and controls for the 2012 reform. The labor costvariable in the table includes the wages paid to the main owner.

110

Table 13: Differences-in-differences results: Firms with active owner-managers andcash-to-assets ratio above median

Investment Sales Var.costs Labor costs Value added

Diff-in-diff estimate -0.000 18,790** 13,694** 7224* 7079

(0.008) (6865) (4696) (3515) (5059)

Firm fixed effects X X X X X

Year fixed effects X X X X X

Weighting X X X X X

Control for 2012 reform X X X X X

Constant 0.280*** 406,183*** 171,540*** 96,474*** 253,481***

(0.019) (12,952) (7324) (4825) (8256)

R2 0.369 0.863 0.847 0.907 0.872

N 139,108 155,421 144,689 152,540 144,689

* p < 0.05, ** p < 0.01, *** p < 0.001

Note: Table shows the difference-in-differences estimates and firm-level clustered standard errors (inparentheses) for investment per lagged capital, firm-level sales, variable costs and labor costs forcorporations with active owner managers and firms with a cash-to-assets ratio above median. Allregressions follow the estimation procedure presented in Section 4 and include firm and year fixedeffects, the re-weighting procedure, and controls for the 2012 reform. The labor cost variable in thetable includes the wages paid to the main owner.

111

Table 14: Differences-in-differences results: Varying the baseline sample restrictions

Sample restrictions for previous year’s annual sales

Lower limit 100,000 100,000 80,000 80,000 80,000 120,000 120,000 120,000

Upper limit 2,002,899 3,004,349 2,503,624 2,002,899 3,004,349 2,503,624 2,002,899 3,004,349

Investment per lagged capital

DID estimate 0.005 0.006 0.007 0.007 0.006 0.008 0.007 0.008

(0.006) (0.006) (0.005) (0.005) (0.005) (0.006) (0.006) (0.006)

R2 0.362 0.360 0.359 0.360 0.359 0.362 0.363 0.361

N 312,107 323,228 346,302 339,380 350,501 294,352 287,430 298,551

Sales

DID estimate 14,547** 21,952*** 17,179** 12,366* 20,368*** 20,271*** 14,801** 23,371***

(5280) (6219) (5518) (4810) (6029) (6118) (5316) (6640)

R2 0.844 0.866 0.860 0.849 0.870 0.851 0.838 0.862

N 345,091 356,850 384,377 377,047 388,806 323,854 316,524 328,283

Variable costs

DID estimate 11,033** 15,745*** 14,007*** 9829** 15,535*** 15,381*** 10,383** 16,809***

(4144) (4284) (4220) (3770) (4569) (4207) (3702) (4597)

R2 0.815 0.844 0.831 0.816 0.845 0.826 0.811 0.841

N 323,363 334,960 357,809 350,582 362,179 305,816 298,589 310,186

Labor costs

DID estimate 4820* 6602** 4970* 3912* 5935* 6117* 5475* 6958*

(2031) (2558) (2177) (1843) (2322) (2599) (2269) (2702)

R2 0.900 0.908 0.905 0.902 0.910 0.902 0.898 0.907

N 334,394 345,981 371,084 363,875 375,462 314,796 307,587 319,174

Value added

DID estimate 4820* 6602** 4970* 3912* 5935* 6117* 5475* 6958*

(2031) (2558) (2177) (1843) (2322) (2599) (2269) (2702)

R2 0.900 0.908 0.905 0.902 0.910 0.902 0.898 0.907

N 334,394 345,981 371,084 363,875 375,462 314,796 307,587 319,174

* p < 0.05, ** p < 0.01, *** p < 0.001

Note: Table shows the difference-in-differences estimates and firm-level clustered standard errors (inparentheses) for investment per lagged capital, firm-level sales, variable costs, labor costs, and valueadded varying the sample restrictions. All regressions follow the estimation procedure presentedin Section 4 and include firm and year fixed effects, the re-weighting procedure, and controls forthe 2012 reform. We vary the baseline sample restrictions by reducing and increasing the upper(2,503,624 euros) and lower (100,000 euros) sales limits of the sample by 20%.

112

Table 15: Differences-in-differences results: Outcomes estimated with years 2010–2016

Investment Sales Var.costs Labor costs Value added

Diff-in-diff estimate 0.006 23,702** 18,737** 5571* 7179

(0.007) (7779) (6881) (2578) (3797)

Firm fixed effects X X X X X

Year fixed effects X X X X X

Control for 2012 reform X X X X X

Weighting X X X X X

Constant 0.207*** 462,221*** 205,126*** 113,647*** 271,620***

(0.015) (14,867) (8185) (5121) (8543)

R2 0.392 0.876 0.848 0.920 0.891

N 248,450 275,279 256,285 265,131 256,285

* p < 0.05, ** p < 0.01, *** p < 0.001

Note: Table shows the difference-in-differences estimates and firm-level clustered standard errors (inparentheses) for firm-level sales, variable costs, labor costs, and investment without the years 2008and 2009. All regressions follow the estimation procedure presented in Section 4 and include firmand year fixed effects, the re-weighting procedure, and controls for the 2012 reform.

113

Table 16: Differences-in-differences results: Outcomes estimated with winsorized data

Sales Var.costs Labor costs Value added

Diff-in-diff estimate 15,225*** 11,026*** 8524*** 9762***

(4197) (2597) (1241) (2196)

Firm fixed effects X X X X

Year fixed effects X X X X

Control for 2012 reform X X X X

Weighting X X X X

Constant 454,076*** 204,785*** 95,434*** 254,182***

(10,721) (5587) (4033) (6470)

R2 0.888 0.886 0.915 0.900

N 356,583 331,553 343,482 331,553

* p < 0.05, ** p < 0.01, *** p < 0.001

Note: Table shows the difference-in-differences estimates and firm-level clustered standard errors (inparentheses) for firm-level sales, variable costs, and labor costs with data winsorized at 2.5% levelfrom each end. All regressions follow the estimation procedure presented in Section4 and includefirm and year fixed effects, the re-weighting procedure, and controls for the 2012 reform.

114

B Appendix: Business Tax System in Finland

In this Appendix, we describe the main features and recent changes in the taxation

of privately held (non-listed) corporations and partnership firms.

Privately held corporations

Privately held corporations are separately tax-liable, meaning that their profits are

taxed at the firm level according to the corporate tax rate. The tax rate for corporate

profits was reduced by 4.5 percentage points from 24.5% to 20% from January 2014

onward. The 2014 tax rate cut was preceded by a smaller 1.5 percentage-point tax

cut in 2012.

Owners of privately held corporations pay an additional tax on the income with-

drawn from the firm. Income withdrawn from the firm can be paid to the owner

either as wages or dividends. The wage income tax schedule is progressive, and the

tax rates vary between 0–55%. Wage income taxation contains three different parts: a

central government progressive tax schedule, proportional municipal-level taxes, and

employee’s social security contributions. Over the last decade,there have not been

any significant changes in wage taxation in Finland.

The dividend tax schedule for the owners of privately held corporations includes

many different thresholds and rules. The imputed return on the net assets of the

firm, calculated as a fixed percentage share of 8% of firm-level net assets (assets

minus debt), defines the amount of dividends that are taxed at an effective tax rate

of 26%. This rate includes both owner-level dividend taxes and corporate taxes. This

dividend income is 75% tax-free, and 25% is taxed as personal capital income at a

rate of 30%. Combined with the corporate tax of 20%, this yields an effective tax

rate of 26% (0.20+(0.8*0.25*0.30)=0.26). The rate increases to 26.8% if the annual

personal capital income of the owner exceeds 30,000 euros, since then the personal

capital income tax rate increases to 34%. Dividend income above the computational

net asset threshold is 15% tax-free, and 85% is taxed according to the progressive

wage tax schedule excluding social security contributions. Finally, dividends below

the net asset threshold but above a predetermined monetary threshold of 150,000

euros are subject to a tax rate of 40.4% (43.1% if personal capital income exceeds

30,000 euros). This dividend income is 15% tax-free and 85% is taxable as personal

capital income.

There have been several changes in these rules and thresholds over the last decade

115

in Finland. Table 17 summarizes the thresholds and rules affecting dividend and

corporate tax rates from 2006 to 2016, and Table 18 presents the equivalent effective

dividend tax rates over time in different regimes. The computational return on net

assets was lowered from 9% to 8% in 2014. Also, the share of dividend income below

the imputed return on net assets that was taxed as capital income was increased

from zero to 25% at the same time as the corporate tax cut in 2014. These led

to an increase in the effective dividend tax rate of approximately 1.5 percentage

points for the owners of small corporations, as illustrated in the first column of Table

18. Furthermore, in 2012, the monetary threshold was first reduced from 90,000 to

60,000 euros, and then increased to 150,000 euros in 2014. The latter increase in

the threshold implied a reduction in the effective dividend tax rate for the owners of

large corporations with large firm-level net assets. Also, the share of dividends taxed

as wage income above the net asset thresholds increased from 70% to 75% in 2014,

and the share of dividends taxed as capital income above the 150,000 euro threshold

but below the net asset threshold increased from 70% to 85% in 2014. Nevertheless,

combined with the corporate tax rate cut, these changes did not significantly affect

the effective dividend tax rates presented in the second and third columns of Table

18.

Partnership firms

A partnership is a pass-through entity, meaning that its profits are taxed only at the

owner level. Owners of partnership firms also have a firm-level net asset threshold

in their income tax schedule. Profits that fall under a 20% firm net asset threshold

are taxed according to the capital income tax schedule, and any income above the

threshold is taxed as the wage income of the owner. The net assets of partnership

firms are defined as assets - debt + 30% of labor costs, while corporations follow a

simpler assets - debt definition. In 2012, the effective marginal tax rate for profits

below the net asset threshold was increased slightly from 26.6% to 28.5% (30.4%

if capital income exceeds 30,000 euros). There were no changes in the taxation of

partnership firms at the time of the corporate tax rate cut in 2014. The effective

owner-level tax rates and their recent changes for the owners of partnership firms are

described in Table 19.

116

Tab

le17:Dividendtaxthresholdsan

dcorporatean

dcapital

incometaxrates,2006–2016

Yea

rNet

asset

thresh

old

Monetary

thresh

old

Tax-exem

pted

Tax-exem

pted

Tax-exem

pted

Corp

orate

Capitalinco

me

Capitalinco

me

(NAT)

(MT)

D<

NAT

D<

NAT

&D

>MT

D>

NAT

taxrate

taxrate

taxth

resh

old

2006–2011

9%

90,000

100%

30%

30%

26%

28%

0

2012–2013

9%

60,000

100%

30%

30%

24.5%

30/32%

50,000

2014

8%

150,000

75%

15%

25%

20%

30/32%

40,000

2015

8%

150,000

75%

15%

25%

20%

30/33%

30,000

2016–2018

8%

150,000

75%

15%

25%

20%

30/34%

30,000

Notes:

Drefers

todividen

ds,

NAT

refers

tothenet

asset

threshold

andMT

refers

tothemonetary

threshold.Tax

-exem

ptedshare

den

otesthe

share

oftax-freedividen

dswithin

each

regim

e.W

hen

D>

NAT,theremainingshare

istaxed

aslaborincome,

otherwisethecapitalincometax

schedule

isused.Capitalincometaxrate

denotesthecapitalincometaxratesbelow

andabovethecapitalincometaxthreshold.

Tab

le18:Effective

dividendtaxrates,2006–2016

Year

D<

NAT

D<

NAT

D>

NAT

andD

<MT

andD

>MT

2006

-201

126%

40.5%

54.5%

2012

-201

324.5%

40.3%

53.6%

2014

26%

(26.4%

)40.4%

(41.8%

)53%

2015

26%

(26.6%

)40.4%

(42.4%

)53%

2016

-201

826%

(26.8%

)40.4%

(43.1%

)53%

Notes:

Drefers

todividen

ds,NAT

refers

tothenet

asset

threshold

andMT

refers

tothemonetary

threshold.Figuresin

bracketsreferto

effective

dividen

dtaxratesab

ovethecapitalincometaxthreshold.

117

Tab

le19:Tax

ratesof

partnership

firm

s,2006–2016

Effective

taxrate

below

Effective

rate

above

the20%

net

asset

threshold

the20%

net

asset

threshold

2006–2011

26,6%

0-∼

55%

2012–2018

28,5

(30,4%)

0-∼

55%

Note:Table

show

sthetaxratesontheprofits

ofpartnership

firm

s(taxed

attheow

ner-level)in

2006–2016.Theprogressivewageincometaxrate

isapplied

toprofits

exceed

ingtheim

puted20%

return

onfirm

net

assets.

Figuresin

bracketsreferto

effective

taxratesab

ovethecapitalincome

taxthreshold.

118

Chapter 4

Does Household Tax Credit Increase

Employment?∗

Aliisa Koivisto1,2, Jarkko Harju2,3, and Tuomas Kosonen2

1University of Helsinki

2VATT Institute of Economic Research3Tampere University

Abstract

This paper studies the effects of household tax credit (HTC). HTC is a tax

credit for consumers who use household services, with the aims of increasing

employment in the service sector and curbing tax evasion. We use data on

firm-level monthly value added tax reports, annual income tax filings, and

individual-level reports of the use of HTC obtained from the Tax Authorities

in Finland and Sweden. Our results show that at best the HTC system has very

limited effects on the consumption of services and employment in the service

sector. In addition, we do not find evidence supporting the idea that the HTC

is efficient in reducing tax evasion. Our descriptive analysis shows that the

HTC system is regressive, i.e. high-income individuals claim a large share of

the HTC.

Keywords: Household tax credit, demand, employment, consumer price, tax

evasion.

1 Introduction

Governments use various policies aiming to boost employment by increasing con-

sumption of labor-intensive services, and to increase tax revenue by reducing tax

∗Another version of this paper is published in Publications of the Government´s analysis, assess-ment and research activities 2021:1.

119

evasion. One such effort has been to use special tax treatments and subsidies to

spur the demand for services in labor-intensive industries. The EU, for example, has

allowed member countries to reduce VAT rates on labor-intensive services. Another

commonly used strategy has been to provide tax credits for consumers in sectors pro-

viding household services such as renovation and cleaning services. These so-called

household tax credits (HTC) are used in various countries such as France, Finland

and Sweden. This special tax treatment allows taxpayers to deduct a share of labor

costs against their income taxes, effectively lowering the price of services. The aim of

HTC is to boost the consumption of services and to curb tax evasion. Government

spending on these is comparable to, or even higher than, the lost tax revenue due

to reduced VAT rates in the labor-intensive sectors. However, unlike the effects of

reduced VAT (see e.g. Harju et al. 2018), little is known of the effects of HTC, espe-

cially the effects of HTC on the consumption of services, employment, or the shadow

economy.

In this study, we provide a comprehensive analysis of HTC policies in Finland and

Sweden. We describe various aspects of the usage of services and HTC, and evaluate

the impact of HTC on the consumption of household services and other outcomes.

We utilize changes in HTC details that took place in different years in Finland and

Sweden to perform a credible causal analysis. We use detailed microlevel data on

the usage of HTC and microlevel administrative data on firms providing the services

in Finland and Sweden. With the empirical design and comprehensive data, we can

provide causal evidence on the effectiveness of HTC policies.

We focus on understanding the causal impacts of the HTC policies on consumption

of household services in Finland and Sweden by utilizing two empirical settings. First,

we compare household service industries between Sweden and Finland over time.

Sweden and Finland share similar cultures, apply similar income tax rules, and most

importantly, apply similar institutions regarding HTC. The HTC details differ in ways

that allow us to examine the impacts of different institutional details on consumption.

Most importantly, the countries have changed the details of the tax credit system at

different times. Finland had the HTC system in place when Sweden adopted the

current HTC for cleaning services in July 2007. Later, in July 2009, the Swedish

HTC system was reformed so that firms would claim the HTC on behalf of consumers,

making the tax credit’s effect on the prices faced by consumers more immediate. We

use these sources of variation to study the effects of HTC tax credit in the cleaning

industry.

120

In our second setting, we study the renovation industry, with a particular focus

on Finland, and use other similar industries as a domestic control group. Finland

increased the amount of maximum tax credit for the renovation industry significantly

from 1150 to 3000 euros in 2009, and thus our empirical strategy is to compare firms

operating in the renovation industry with our matched control group before and after

2009. In both settings, we conduct the analysis with firm-level VAT data including

reported sales and input usage, which will reveal both whether consumption among

households has increased and whether tax evasion among firms has decreased.

Finally, a descriptive analysis of the administrative data shows that a relatively

large share of individuals using HTC-eligible services make mistakes in their reports

to the tax authority. This is evident as claiming a lower amount than eligible for

based on the costs of services, which makes this type of mistake costly for claimants.

In addition, as the HTC is tax credit from income taxes, those who do not pay enough

taxes cannot utilize the full amount of HTC. Our descriptive analysis also shows that

the HTC is highly regressive – higher income households use HTC to a much greater

extent than lower income households, and very poor households almost do not utilize

HTC at all.

We do not find any statistically significant or economically meaningful effect of

HTC on the consumption of household services. First, our best scenario to find such

effects is the introduction of HTC for cleaning services in Sweden. However, we find

no increase in the reported value of sales of cleaning firms in Sweden relative to the

Finnish firms used as the comparison group. The main identification assumption of

our empirical approach is that the value of sales among cleaning service firms follow

each other across countries. We find that the pre-reform trends are very similar,

validating our empirical approach and mitigating the potential concern that the two

groups would not be comparable. Around 2007 there were no other concurring effects

that could have affected the results. Our analysis of the 2009 reform that switched

the credit-claiming responsibility from customers to firms shows no clear increase

in service consumption, but the results are somewhat more mixed. In Sweden, the

number of small firms increases more rapidly than in Finland, and also the consumer

prices of cleaning services seem to increase somewhat. Since we do not observe any

significant increase in the aggregate consumption relative to Finland, we consider the

observed differential effects to be caused mainly by the Great Recession, which also

seems to have affected the two countries differently in other aspects, beginning in

2009.

121

We complete our causal analysis by studying the effects of HTC on the consump-

tion of renovation services in Finland. In this analysis, we exploit the 2009 reform that

expanded the maximum amount of HTC for renovation services. However, we cannot

use Sweden as a control group for Finnish renovation firms because the Great Re-

cession affected the renovation services in the two countries very differently. Instead,

we use a domestic control group by matching Finnish firms from similar industries

that seem to have followed similar economic trends prior to the reform. Similarly as

for the cleaning sector results, we do not find any evidence of an increase in sales of

renovation services after the reform relative to the control group.

To sum up, our results suggest insignificant consumer responses to the changes

in the HTC rules. This result is consistent with very low elasticity of demand with

respect to the HTC. Our descriptive evidence showing that most households do not

consume these services and that individuals are not very aware of the details of

HTC is also consistent with this result. The low demand elasticity suggests that

the prospects of increasing consumption through HTC regulation are likely to be

quite limited. Moreover, the results of no increase in the amount of reported sales of

services due to an expansion of HTC are consistent with no changes in tax evasion

among firms providing these services. If tax evasion had decreased, the reported sales

or input usage would have increased, which we did not find in the data.

Tax theory literature offers arguments for and against policy tools such as HTC.

Most taxes distort behavior1, yet society needs to collect revenue to pursue various

social objectives. An attempt to avoid distortions arises from a Ramsey (1927) type

labor-leisure model: proportional tax on all commodities including leisure. However,

taxing leisure is considered to be infeasible. When taxation of leisure is ruled out,

an alternative is to try to compensate for the missing tax on leisure by taxing its

complements. Nonetheless, economists and policymakers usually favor uniform com-

modity taxation. The reason is that uniform taxation is considered to be the optimal

solution under certain conditions, including weak separability of leisure and consump-

tion choices and the presence of optimal income taxation (Atkinson and Stiglitz 1976;

Deaton 1979). Moreover, uniform taxation is preferred due to its political and bu-

reaucratic simplicity (Mirrlees et al., 2011). However, uniformity of taxation requires

that factor shares are identical across activities, but clearly this is not very realistic.

Kleven (2004) argues that with varying goods share of different activities, optimal

1Only lump-sum taxes or taxes on economic rent are often considered not to distort behavior(Mirrlees et al., 2011).

122

tax rates are positively related to time intensities. In other words, if transformation

of time into utility requires goods, the social planner may imitate the inverse factor

share rules. Any type of consumption that uses little time, or even saves time, should

carry a relatively low rate of tax. Thus, goods such as childcare, cleaning, house-

keeping, cooking, dishwashing, garden care, and house and car repairs save household

time, and should be taxed leniently or even subsidized.

Our study contributes to a growing empirical literature analyzing various aspects

of consumption taxation. Studies on similar sector-specific tax incentives examine

policies reducing VAT rates for labor-intensive industries based on similar reasons

as governments aim to implement HTC policies (Kosonen 2015; Harju et al. 2018;

Benzarti and Carloni 2019; Benzarti et al. 2020). Reducing a VAT rate is intended

to lower consumer prices, which effectively is also the aim of HTC. The results of

this literature are largely similar to the results reported here. The reduced VAT rate

has an incomplete pass-through to prices, but even when prices decline somewhat

due to the reduced VAT rate, there does not seem to be any effect on increasing the

consumption of hairdressing or restaurant services. Instead, firms seem to benefit

from the reduced VAT rates. These results as well as our evidence on HTC all point

towards inelastic demand with respect to prices for these types of services.

The aim of reducing tax evasion is based on the idea that the tax credit incentivizes

customers to require receipts for their payments so that they can claim tax credits.

Claiming the tax credit requires reporting the transactions to the tax authority and

may therefore increase the compliance of firms through the fear of cross-checking by

the tax administrators leading to an audit. Studies focusing on tax evasion from

consumption taxation tend to find some tax evasion (Pomeranz 2015; Naritomi 2019;

Doerr and Necker 2020). Although these studies do not all investigate precisely the

same industries as our study, they highlight that, in the absence of withholding and

other procedures in income taxation, there likely is some tax evasion among firms.

Thus, the result that HTC is not effective in curbing tax evasion seems surprising.

The explanation for the lack of effect could instead be that households are not very

responsive to HTC policies in their consumption decisions and those consumers who

engage in buying services from tax-evading firms might not be the ones that claim

HTC. Moreover, tax evasion among cleaning services may be lower to begin with,

which could explain some of our results. While our results are limited in the effect

of HTC on tax evasion in the renovation sector, we show that the increase of the

maximum credit for renovation services from 1150 to 3000 euros did not have a

123

notable effect on tax evasion. 2

Finally, we extend the literature on HTC by providing both a comprehensive

description of the consumption and usage patterns as well as an analysis of the causal

effects that HTC has on the consumption of services. Earlier reports have shown

that the HTC system is very regressive (Hakkinen-Skans 2011; Skatteverket 2011),

hence increasing income inequality. Gronberg and Rauhanen (2015) focus on how

HTC affects pensioners. They find that the usage of HTC among elderly is hampered

in two ways. First, elderly may not be able to claim the credit due to poor health

or the difficulty of using the system. Second, low-income pensioners do not have the

income to buy the services in the first place, and they do not pay enough taxes to be

able to use the full amount of HTC. A recent report by Swedish national audit office

focuses on the cleaning and care sectors, and finds that the tax credit does not fund

itself, and meets its target of increasing employment poorly (Riksrevisionen, 2020).

Again, these observations mirror our own.

This paper proceeds as follows. In Section 2, we describe the institutions regarding

HTC in Finland and Sweden. In Section 3, we discuss the predictions of the responses

we could expect to HTC, and the data and methods we use to study the potential

effects. In Section 4, we provide descriptive evidence of the usage of the tax credit,

and the targeted industries. In Section 5, we apply our quasi-experimental set-up to

study the causal impact of HTC on the consumption of household services and hence

its effect on demand and employment in sectors. Finally, Section 6 concludes.

2 Institutions

In this section, we discuss the main institutional details of the household tax credit

in Finland and in Sweden. The household tax credit (HTC) allows individuals to

deduct part of the labor costs of household work services from the buyer’s income

tax liability (N.B. Not taxable income). The deductible household services eligible

for the tax credit include renovation, cleaning services, and care. Depending on the

institution, the buyer claims the credit by herself or the deduction is filed directly at

the time of purchase by the seller.

HTC consists of three important parameters. The first parameter is the percentage

share of labor costs deductible from income taxes. The labor costs include wage costs,

2However, our evidence is limited in drawing conclusions about lower amounts of HTC, as mostHTC consumers do not claim the maximum amount.

124

value added taxes (VAT), and payroll costs, forming the base for deductible tax credit.

The second parameter is the maximum annual amount of tax credit, which is set per

taxpayer, not per household, in both Finland and Sweden. Thereby, households with

multiple taxpayers can deduct higher expenses. Third, an own liability is excluded

from the credited labor costs. For example, if a service costs 1200 euros, of which

1000 euros consists of labor costs, a 50-percent deduction with 100-euro own liability

is 400 euros3.

Next we describe the details of HTC and changes over time for Finland and Sweden

separately. Some of the changes are important for the study as we use them to draw

causal inference.

2.1 Finland

The household tax credit was trialed in Finland as early as 1997–2000. At the time,

the experiment divided Finland into two parts: in half of the country, firms claimed

the credit directly, and in the other half, individuals claimed it by themselves. Oth-

erwise, the rules of the HTC were the same in the whole country. Due to the ad-

ministrative costs arising in the system where the seller credits the costs directly, the

government decided to adopt a nationwide system based on individuals claiming the

credit, starting in January 2001. In this system, a taxpayer claims the tax credit

by filing household service purchases with the tax authority. The tax authority then

aggregates all the purchases, calculates the HTC that the taxpayer is eligible for, and

finally deducts the credit from the taxpayer’s tax liability. To receive the tax credit,

the taxpayer either waits for the tax returns of the ongoing year, which, during our

research period, was late the next year, or then the taxpayer can reduce their with-

holding of income tax for the rest of the year and thereby benefit from the credit

slightly sooner.

In Finland, the parameters of HTC governing its generosity have changed many

times since the adoption of the nationwide system. Figure 1 lists the changes in the

parameters on a timeline. In the figure, the maximum tax credit is presented above

the timeline in the middle and the percentage-share deductible from the income taxes

is below. The deductible share of labor costs was 60% from the start and decreased

from 60% to 45% in 2012. In 2017 the deductible share increased from 45% to 50%.

The maximum HTC limit was doubled from 1150 to 2300 euros for cleaning services

3(0.5 · 1000− 100) = 400.

125

and care in 2006, but the upper limit for renovation services remained 1150 euros until

the end of 2008. In 2009, the maximum HTC increased to 3000 euros for renovation,

cleaning, and care. In January 2012, the limit was reduced from 3000 to 2000 euros.

In 2014, the maximum was increased from 2000 to 2400 euros. The own liability was

originally 85 euros, but was soon increased to 100 euros in 2002 without subsequent

changes.

Figure 1: Time-line of HTC in Finland

2001 2003 2005 2007 2009 2011 2013 2015 2017

845e;85e

900e

1150e 2300e* 3000e 2000e 2400e

40% 60% 45% 50%

Note: Above the time-line is the maximum tax credit e (own liability 100e) and below the time-lineis the %-share allowed to deduct.

2.2 Sweden

In Sweden, the parameters of HTC have been quite stable after the introduction of

the HTC system in 2007. Note that Sweden had already applied similar occasional

tax credit tools for renovation services for fiscal purposes before the current HTC

system4.

Figure 2 shows the timeline of the tax credit from 2007 to 2017 in Sweden. The

Swedish HTC system has been more stable over time compared to the Finnish system.

The most important changes for this study are the implementation of the system and

change in the claiming of the credit in Sweden. Since the adoption of the current

HTC system in July 2007 until 2009, the tax credit was applied only to cleaning and

child- and elderly-care services in Sweden. During this time the credit was claimed

by individuals. In July 2009, the HTC system was reformed, so that the claiming

responsibility was moved from the buyer to the seller of the household work. In

addition, renovation services were included in the HTC. In this so-called invoicing

model, the seller charges the HTC part to the tax authority and the tax authority

pays the credit directly to the firm based on their invoice. The buyer gets the benefit

4Sweden applied temporary HTC policies for renovation in 1993–1994, 1993–1999 and 2004–2005.It was argued that the temporary policies balanced the business cycles.

126

directly and is only responsible for knowing his/her eligibility. The tax authority

confirms the eligibility only at the end of the year when finalizing the taxation. If

it then turns out that the credit was falsely claimed, the credit is charged from the

taxpayer in income taxation. In both systems, it requires the taxpayer to have tax

liability so that she or he can benefit from the credit. In the invoice model, it means

that if the tax liability is lower than the amount of HTC credited in the bill, the tax

authority claims the HTC back similarly as tax debt.

The maximum deductible amount, own liability, and the percentage share of the

costs were all at the same level from 2007 to 2015. However, the deduction for reno-

vation services, formally applied starting in 1 July 2009, was granted for renovation

work that had been performed and paid starting December 8. In 2017, the Swedish

government decided to make the following changes to the HTC system: 1) introduce

a 25,000 kr upper limit for cleaning services and care if the buyer is under 65 years

old, 2) keep the total maximum HTC at 50,000kr (for renovation, cleaning services

and care) per individual, and 3) reduce the share of deductible renovation services

from 50% to 30%. We do not utilize these latter changes in the study due to data

limitations.

Figure 2: Time-line of HTC in Sweden

2007 2009 2011 2013 2015 2017

50,000kr;1000kr 50,000kr*;1000kr

Cleaning and care Renovation, cleaning and care

50% 30%*/50%

Self claimed Direct credit for buyer, tax credit from hiring

Note: Above the time-line is the maximum tax credit own liability in SEK and below the time-lineis the %-share allowed to deduct, 10SEK≈1e.*25,000kr upper limit for RUT-work (cleaning and care) if the buyer is under 65 year old. Maximumfor solely ROT (renovation) or for RUT+ROT is still 50,000. ROT deduction – 30 %, RUT deduction– 50 %.

Figures 3 and 4 show the aggregate amount of tax credits and the number of

individuals using the tax credit for Finland and Sweden, respectively. In general, the

trend of HTC usage is increasing in both countries. However, at the same time there

are clearly the most users and largest amount of credits in 2009–2011 in Finland,

when the Finnish system was at its most generous. Similarly, it is evident that the

cut in the Swedish HTC system decreased especially the amount of tax credit usage

127

but also somewhat decreased the number of individuals using the credit.

3 Empirical Approach

3.1 Predictions

The main aim of HTC systems both in Finland and Sweden has been to increase em-

ployment in the service sector through increased consumption of household services.

An additional argument for the HTC is its potential to decrease tax evasion because

credit claims are reported to the tax authority. Moreover, the HTC might increase

labor supply among households as it could induce households to use household ser-

vices instead of doing household work themselves, and therefore release time for paid

work. These effects are discussed in greater detail below.

Employment in the service sector. The main desired effect of the HTC is to

increase employment among labor-intensive service sectors5. HTC aims to increase

employment via increased consumption, but there are a few prerequisites. First, the

consumer needs to perceive the HTC as a price decrease to respond to. This is not

evident, as discounting, mental budgets and uncertainty regarding the system may

affect the perceived value of the credit. Then, the lower prices in turn need to induce

more demand for the services, implying a positive demand elasticity with respect

to prices. If the demand for services increases due to the tax credit, then the next

condition is to have supply effects. Effectively, if supply is not elastic and the amount

of services produced does not increase, it leads to an increase in before-HTC consumer

prices, with no effect on consumption nor employment in the sector. Thus, HTC is an

instrument that aims to lower the prices of services, and empirically we are interested

in what is the demand elasticity of services with respect to it. If we were not to

detect any increases in the consumption of household services, we would not expect

any increase in employment among service sector firms either.

Two more important factors affect the employment effects of the HTC system.

First, how easily firms can hire new staff or the availability of current staff to work

more. If these are somehow constrained, even higher demand for services does not

necessarily increase employment levels. But even if there is an increase in employment

in sectors producing HTC-eligible services, it could just shift labor from industry to

another, not affecting the level of total employment. Therefore, the effectivity of the

5See e.g. Government’s proposal proposition (140/2000).

128

tax credit also depends on where this potential extra worker input comes from: is it

replacement from other sectors or more workers from unemployment6?

Tax evasion. The HTC might be effective in reducing tax evasion as the tax

credit incentivizes customers to require receipts for their payments so that they can

claim tax credits. To claim the tax credit, the taxpayer is required to report the

transaction to the tax authority. This may increase the compliance of firms, for

example through the fear of cross-checking by the tax administrators leading to an

audit. Therefore, a rise in the HTC could imply a decrease in overall tax evasion as it

increases the incentives to declare transactions. However, there may also be opposite

effects, as firms can increase false reporting of labor costs relative to other inputs

in the client’s bill, and households may engage in false claiming in a system where

households claim the credit7.

Labor supply. The HTC system may increase the labor supply of the buyers

of these services. From the perspective of the buyer of these household services, the

opportunity cost of time changes. For example, hiring someone to clean your house

will free up time to be allocated either to leisure or work. Thus, a more generous

HTC could potentially increase the labor supply of individuals. However, this implies

that the HTC really increases the demand for these services and not just rewards

those using or producing these services irrespective of the system.

Distributional effects. As the HTC is credited on tax liability and not income,

low income individuals might not have enough tax liability for HTC to be credited

against. Therefore, it increases the relative prices of these services at the lower end of

the income distribution, making it unequal across individuals by income levels. HTC

can also be considered regressive as HTC services are mainly consumed by the rela-

tively wealthy. The majority of HTC, almost 80% in 2014, was used for renovation

services. Renovation services are usually mainly consumed by house and apartment

owners, and owning a property is associated with a relatively good economic situa-

tion. The lower end of the income distribution potentially cannot afford to use these

6A similar conditionality actually applies to demand of services as well. If the demand shiftsfrom other labor-intensive services to household services, the effect on total demand on services andhence total employment might stay unaffected.

7Or firms and consumers can collude to report more labor costs. Doerr and Necker (2020) studiedcollaborative evasion in this setting with an online experiment finding some evidence for collaborativeevasion, when the consumer signals willingness to collude. In most estimates the difference betweenevasion price and official price is not profitable for the consumer when the foregone tax credit is takeninto account However, when restricted to an anonymous market, the proposed evasion discount ishigher than the subsidy (20%).

129

services in the first place. Cleaning, while not requiring property ownership, is not

a necessity and therefore not likely to be consumed by low income individuals even

when incentivized with later received tax credits. However, if the HTC were to have

a positive effect on total employment, then this would potentially decrease inequality.

3.2 Data

Our main data come from the Finnish and Swedish Tax Administrations. We have

data for all firms from certain industries that can apply HTC. These industries are

selected based on their use of HTC with industry codes. The selected industry codes

are two-digit industry codes 41 or 43 for renovation and 5-digit codes 81210 and 81100

for cleaning. In our empirical analysis, we mainly use firm-level data of VAT reports

of firms. These data include information on the amount of VAT that firms have paid

monthly, quarterly, or annually, and similarly the amount of VAT deductible inputs.

These data are available from January 2006 until December 2014. Exploiting these

data, we can study how sales and inputs respond to changes in the HTC system. As

sales is producer prices times the quantity of product or services sold, it quantifies

changes both in prices and demand. In addition to the data on firm-level VAT reports,

we have annual-level income tax data, which we use to complement our analysis.

These data include annual labor costs, other input costs, and taxable profits. We

describe the firm-level data, the usage of the HTC, and the trends in the HTC sectors

in detail in Section 4.

On top of the firm-level data, we also exploit various other data sources. We have

individual-level data for those individuals who have claimed HTC from 2006 to 2014.

We also use industry-level data from Eurostat to show the overall development of in-

dustries across countries over time to show macro-level changes. In addition, to study

how HTC is used across income distribution, we use Income Distribution Statistics,

which is a representative sample of Finnish taxpayers maintained by Statistics Fin-

land. Finally, we have consumer price index data to show how the consumer prices

develop over time in general but also in different sectors before and after changes in

the HTC system.

3.3 Methods

To study how HTC affects the consumption of services, we apply a Difference-in-

Differences (DiD) approach. We analyze cleaning and repair services separately with

130

different designs. In analyzing cleaning services, we utilize the implementation of

HTC in Sweden as our main identification strategy. In addition, we look into the

2009 change in the claiming system in Sweden, though the set-up suffers from fiscal

cycles as well as from a simultaneous HTC change in Finland. We use Finland, where

HTC for cleaning services was already in place without changes in the policy, as a

control group representing how consumption of cleaning services within households

in Sweden would have developed in the absence of the introduction of HTC. The

identification requires in this approach that the consumption trends in the absence

of changes in the policy are parallel. These countries are broadly comparable to each

other as, for example, institutions, geography, culture, climate, and seasonal vacation

periods are similar in Sweden and Finland. We can test the parallel trends assumption

by looking at consumption trends of cleaning services before the implementation of

the policy in Sweden in 2007.

We study the responses by estimating the following equation.

yit = β0 + β1 × (Treati × Post1) + β2 × (Treati × Post2) +Mt + μi + εit (1)

where yit is the outcome of interest (in logs) for a given firm i in time t. Treat

is a binary variable with value 1 for Swedish firms in cleaning sector and zero for

Finnish firms, Post1 is a binary variable with value 1 after the introduction of HTC

for cleaning services in Sweden in January 2007, Post2 is a binary variable denoting

the introduction of the invoicing system in 2009. Mt captures the month (or year)

fixed effects, μi the firm fixed effects, and εit represents the error term. Our main

outcomes of interest are monthly-level sales and inputs from VAT reports of firms.

We also use annual-level measures such as labor costs and taxable profits to study

firm-level responses. All these outcomes are in logs.

As noted above, the main identifying assumption in the DiD method is parallel

trends in the outcome of interest between treatment and control groups before the

studied policy change. There are two potential concerns in our empirical setting.

First, different macroeconomic trends between Sweden and Finland might invalidate

the use of the DiD setting. In Section 4.3 we present evidence suggesting that macroe-

conomic conditions do not invalidate the use of DiD for the cleaning industry. Second,

firms in these particular industries could operate differently over time in the two coun-

tries. In Section 5.1, we explicitly show that the pre-reform trends for control and

131

treatment groups are parallel, suggesting again that this identification challenge does

not apply in our setting.

Second, we study the effects of a large increase in the maximum level of annual

tax credits of renovation services in January 2009 in Finland. In this analysis we

again utilize the DiD method, but here we compare firms in the Finnish renovation

sector to other Finnish firms in similar sectors unaffected by this reform. We do not

apply a cross-country DiD method as renovation sectors do not develop similarly in

Finland and Sweden due to the very different macroeconomic impact of the financial

crisis on the renovation sector at the time of the policy change and the simultaneous

HTC change in Finland.

We instead take an alternative approach and use coarsened exact matching8

(CEM) to select a control group among firms in other sectors in Finland. We first

select firms in the renovation industry to form the treatment group. These include

industries providing construction or renovation of buildings, flooring, roofing, instal-

lation work, etc. The industry codes included in the treatment group are listed in

greater detail in Table 8 in the Appendix. The control group is matched from indus-

tries that are not included in the treatment group and do not provide services that

could be used to claim HTC. Otherwise, the control industries resemble the treat-

ment group industries in terms of, for example, size and the type of business activity.

Hence, we have chosen car retail and repair and logistics as our control industries,

listed in greater detail in Table 9 in the Appendix. Then, we use a CEM matching

algorithm to select firms into the control (and treatment) groups. The idea of the

method is to use an algorithm to pick suitable control firms from the control indus-

tries based on pre-reform information. We then give them weights to match them

to the treatment group. The variables we use in the matching are pre-reform yearly

sales, sales growth, and annual labor costs right before the reform.

In the DiD analysis, we estimate equation 2 presented below, where Post is a

binary variable with the value 1 after the increase in the maximum amount of annual

credits, January 2009, and Treat is an indicator having value one for firms in the

renovation industry and zero for the control group firms.

yit = β0 + β1 × (Treat× Post) +Mt + μi + εit (2)

Outcome variables are the same as for cleaning sector; monthly-level sales and

8Blackwell et al. (2009).

132

inputs from VAT reports of firms, annual-level labor costs, and taxable profits. All

these outcomes are transformed to logarithmic values to give us relative effects and

to avoid outliers driving the results.

4 Descriptive Analysis

In this section, we first describe the overall use of HTC in Finland and Sweden over

time. We do this showing evidence of both individuals claiming deductions and firms

operating in sectors providing HTC services. Second, we describe the macroeconomic

conditions in Finland and Sweden that are essential to consider in our empirical

approach for cleaning services. Finally, we show evidence of the comparability of

firms across countries before and after the reforms in HTC rules.

4.1 Consumers

We begin the descriptive analysis by describing data on HTC claims in Finland.

Figure 3 shows the number of HTC recipients (left axis) and the sum of claimed HTC

in euros in Finland from 2001 to 2017. The number of recipients has increased over

time, with a small dip in 2012, when the government cut the level of maximum tax

credit from 3000 to 2000 euros. In 2017 over 400,000 taxpayers in Finland claimed

HTC. The sum of claimed tax credits (right axis) has also increased substantially from

the early years of HTC adoption to over 400 million euros in 2017. The aggregate

costs of the system in 2009–2011 clearly stand out from the Figure, as during those

years the maximum HTC was at its highest level, 3000 euros. Note that an increase

in the aggregate cost does not imply an increase in consumption of household services

themselves due to the credit, rather than just higher credit claims due to the more

generous system. We will analyze the impact question in the next section.

Figure 4 paints a similar picture for Sweden. Initially the number of individuals

claiming HTC is modest but increases rapidly when renovation is included in the

scheme and the system changed to the invoice system in 2009. Also, the aggregate

value of claimed HTC increases sharply, with a peak in 2015, when the amount of

claimed HTC is over 2 billion Swedish Krona. From the beginning of 2016 the Swedish

government reduced the maximum annual tax credit substantially, and this decreased

the aggregate value of tax credits sharply to less than 1.5 billion Swedish Krona.

Table 1 describes the features of individuals receiving HTC in Finland. More

133

taxpayers have received the tax credit in the course of time; hence, we are not only

interested in the characteristics of an average HTC claimant, but also how these have

changed over time. The table shows summary statistics of the HTC claimants in

2006, 2011 and 2014. Men are as likely as women to claim HTC9. However, a notable

common feature among HTC claimants is their relatively high age. The average age of

HTC claimants seems to have been increasing during the past decade. Related to age,

a noteworthy feature among the HTC claimants is the high share of pensioners. The

fact that HTC claimers are older than average taxpayers is to be expected. This is

because older taxpayers tend to have higher incomes, at least up to a point. Moreover,

a person is more likely to use renovation services when they own property, such as a

house, secondary home, or summer cottage, the likelihood of which increases by age.

Finally, much care services are likely to be consumed by individuals seeking help to

live at home at an advanced age. However, care services play a very small part among

HTC claims.

Table 1 also shows that mean and median HTC claimants have notably higher

taxable income than an average Finnish taxpayer. While the average income of a

Finnish taxpayer was 25,852 euros in 2010, the average income of an individual re-

ceiving HTC was 47,750 euros. The summary statistics also show that HTC recipients

have relatively high capital income, the average being around 10,000 euros a year.

Finally, Table 1 shows that the average HTC as well as the number of HTC

claimants have been increasing over time. Moreover, there is some persistence in the

claimant group as approximately 45% of HTC claimants also claimed HTC in the

previous year.

Figure 5 illustrates the income distribution of HTC recipients by plotting the

income distribution of HTC recipients and non-recipients according to the tax payer

income Distribution Statistics of 2014. The horizontal axis denotes the income in

1000-euro bins and the vertical axis the percent of taxpayers in each bin. The figure

tells the same story as the summary statistics in Table 1. HTC recipients, displayed

as green bars, have clearly higher income compared to non-recipients (hollow bars),

implying that the tax credit is directed to the upper end of the income distribution

Figure 6 presents how HTC claims increase with household income. The figure

shows in blue the average claimed HTC by household income simulated with the SISU-

microsimulation model and based on a large set of representative administrative data

9The fact that females are slightly more strongly represented among HTC claimants can berelated to the high average age of HTC claimants and the higher life expectancy of women.

134

of Finnish taxpayers. The blue line in the figure shows that poorer households do not

benefit from HTC at all, and the amount of HTC that households claim on average

increases steadily with household disposable income. This demonstrates quite clearly

the regressive nature of HTC. The figure also shows two reform scenarios, one where

the maximum HTC amount is increased by 1000 euros and one where the maximum is

decreased by 1000 euros. The simulations are static and do not attempt to anticipate

how the consumption of services or credit claims would change in response to changes

in policy. The simulation is based on the amount of household services individuals

report when making the credit claims. The reforms have a quite monotonic effect on

the extent to which individuals would use the HTC. The cut and the increase have an

asymmetric effect in the sense that increasing the HTC would have a smaller effect

than decreasing HTC by the same amount in the simulation, but this is likely driven

by individuals not claiming the full amount of services in the data after they have

exhausted the full amount of benefits. Thus, we would expect the actual effect to be

more symmetric than the simulation shows.

Figure 7 describes the distribution of declared labor costs of HTC services in

the left panel. In Finland taxpayers report the labor costs with eligibility for HTC

when filing their taxes, and the tax authority uses these to calculate the amount of

tax credit they receive. The implied distribution of the effective HTC received by

taxpayers is plotted in the right panel. During the years 2009–2011 the maximum

amount of HTC was 3000 euros. To receive the full credit the buyer would need to

record labor costs of 5166.66 euros10. Figure 7 demonstrates that the majority of

labor costs of services are relatively small, but there is also some bunching in larger

amounts. The left panel also shows that there is clear bunching exactly at the HTC

maximizing value of labour costs. This bunching is likely to occur at this point even

when the buyers do not match their consumption of services to this point, because

there is hardly any incentive to report service purchases above this value.

However, there is also clear bunching at a lower value of labor costs, which happens

to equal exactly the amount of maximum tax credit. There should be no service value

or HTC-related reason why we should expect reported labor costs to feature excess

mass at this value. The value of 3000 euros is mentioned in the guidelines describing

the HTC system, it is the maximum credit amount, but not any particular value

when reporting labor costs, to which the credit amount is linked through a formula.

Thus, bunching at this value suggests that many individuals have made a mistake and

105166.66 · 0.6− 100 ≈ 3000

135

reported the maximum credit amount when they should have reported the amount of

labor costs that entitle them to the maximum credit. This is a costly mistake: when

reporting this amount of purchased labor costs, the buyer receives only 1700 euros

in tax credit in the years 2009–2011 instead of the 3000 euros they might have been

eligible for.

In Figure 8, we compare the excess masses at each bunching region to a coun-

terfactual distribution with bunching method (see e.g. Kleven 2016). This allows us

to calculate the magnitude of the excess mass. The counterfactual density is esti-

mated by fitting a seventh-degree polynomial to the distribution when excluding the

bunching region. We find an excess mass of 1 at 3000 euros. This indicates that the

number of taxpayers, around 24,000, is twice as high as in the surrounding distribu-

tion, where around 12,000 taxpayers reported purchases of approximately 3000 euros

in 2009-2011. The right-hand graph shows that the excess mass estimate near the

maximum HTC value is almost 2.9 times the counterfactual.

Figures 9 to 12 show a similar analysis for different HTC regimes. In these other

regimes the maximum amounts of HTC are 2400 and 2000 euros. We find significant

excess mass at these amounts, to the same extent as above. Observing that the

bunching moves together with this wrong parameter further suggests that the excess

mass arises due to a mistake. Moreover, the extent of bunching has stayed quite

notable over time, suggesting that learning has not corrected the misunderstandings

of the claiming system.

Features of the HTC user data lead to some interesting observations. First, the

high average age of HTC claimants and number of pensioners limits the effectiveness

of HTC in increasing labor supply11. Moreover, the vast majority of HTC claims

regard renovation services, which seem even less likely substitutes for paid work, as

many of these require some professional knowledge not in the possession of most

taxpayers12.

Second, HTC falls on high income individuals. The mean income of an HTC

recipient is approximately double the income of an average taxpayer. The average

amount of HTC used clearly increases with household income, as shown in this section.

This means that HTC is a highly regressive tax credit, and quite expensive to the

public sector, with an annual cost of more than 400 million euros in Finland. In other

words, HTC increases income inequality unless we find in the analysis below that it

11This is in line with a finding in Riksrevisionen (2020) for Sweden.12E.g. Housing corporations require that plumbing or electrical work is done by a qualified expert.

136

tremendously increases employment among lower income workers.

Third, the distribution reveals that thousands of people each year make a mistake

and claim a lower amount than what they would presumably be eligible for. The

several thousand taxpayers observed bunching at the wrong values are only the ob-

servations we could identify as mistakes, but likely there are many more mistakes we

could not identify. This suggests that some features of the system are not fully salient

to claimants. This in turn is likely to reduce the incentives the system is supposed to

create to consume more household services.

4.2 Firms

We describe here the firm-level data for the cleaning and renovation industries in

Finland and Sweden. Our data identifies which firms provided services eligible for

the HTC and how much credit was claimed. Figure 13 shows the sum of aggregate

sales and the share of HTC claims of renovation and cleaning sectors in Finland and

Sweden. It is evident that the aggregate sales (right axis) are clearly higher in both

sectors in Sweden compared to Finland. This is expected as Sweden has a larger

economy than Finland13. There has been some increase in the aggregate sales in both

industries in Sweden, but no clear increase in Finland.

The share of HTC from total sales is given in the connected line plot in Figure

13 and the right axis shows the scale. The share of HTC is quite similar across

countries. In Sweden the share increases very rapidly in the cleaning industry after

the introduction of the HTC system in 2007. In 2012, Finland cut the maximum

credit from 3000 to 2000 euros, which clearly decreased the HTC share of total sales

in 2012 for both Finnish sectors. However, the aggregate sales in the industry stayed

around at the same level, suggesting that the drop in share follows largely from the

static change in rule rather than any behavioral effect regarding the consumption of

services.

Figure 14 shows the number of all firms in the renovation and cleaning industries

and marks the share of those firms that have provided HTC-eligible services in Finland

and Sweden. The number of renovation and cleaning firms has increased in Sweden

much more rapidly than in Finland. The number of firms providing services eligible

for HTC has not increased quite as rapidly, suggesting that the increase in the number

13Swedish krona have been first adjusted with consumer price index to January 2006 and thenadjusted with the currency rate of January 2006 to have comparable values between countries.Finnish euro-values are also in values of January 2006.

137

of firms is an economic trend that does not necessarily have anything to do with HTC.

This is especially visible in the bar plot for the renovation industry in Sweden where

there was already an increasing trend before the adoption of HTC in 2009, and no

visible break in the trend in 2009

Figures 15 and 16 show the number of firms in both HTC industries in different

size categories. As Figure 14 showed a clear increase in the number of firms in Sweden,

these Figures 15 and 16 show that the increase in the number of firms is driven by

an increase in the number of relatively small firms with sales below 1 million euros

(≈10 million krona). This is somewhat expected as Figure 13 depicts a less striking

increase in the aggregate sales. While there are more firms in both the renovation

and cleaning industries in Sweden, it seems that in Finland the share of firms that

provided services with HTC is somewhat higher among smaller firms. A possible

reason is that the system is more complex for firms in Sweden as they apply the HTC

rules. Instead, in Finland the customer does the tax filing and thus bears the costs

of reporting, so there is hardly any cost to the firm in this system. While there are

more firms with sales under 1 million euros in Sweden, there are more firms with sales

above 1 million euros in Finland.

Table 2 provides summary statistics for cleaning firms in Finland and Sweden in

2008 and 2011. The table shows the descriptive statistics for all firms in the industry

and separately for those firms that sold services that were used to claim HTC. The

values are in thousands of euros for Finland and thousands of krona for Sweden,

without accounting for inflation. These summary statistics highlight several points

brought up in the previous figures. In Sweden, the number of cleaning firms has

increased notably, while at the same time their average sales have decreased. In

Finland there is no such change. The HTC for each firm is calculated by assigning

the claimed HTC for each firm ID as the customer needs to file the firm ID in their

tax return. The average HTC per firm is around twice as high per firm in Sweden in

comparison to Finland. However, note that the tax credit was also higher in Sweden in

2008 and 2011. In both countries, the majority of cleaning firms are sole proprietors.

The corresponding summary statistics for renovation firms in Finland and Sweden

are provided in Appendix table 10.

Table 3 describes the summary statistics for Finnish renovation firms and control

industries and separately for the CEM-matched sample (unweighted) and the whole

industry group. For example, Ren – All refers to all firms in the renovation industry

and Ren – CEM to the renovation firms in the matched sample without weighting.

138

While firms in the control group are larger on average, the groups are more alike after

the matching, even without weighting.

4.3 Macro-level description

In this section we describe the macro trends in the Finnish and Swedish economies to

validate our comparison of the two countries in the causal analysis for the cleaning

industry. In Figure 17 we plot the development of log value added relative to the

last quarter of 2007. This data on the total amount of value added from the OECD

Statistics for Sweden and Finland shows that their macroeconomic trends follow each

other fairly closely before the financial crisis that affected these countries starting in

2009, but are different after that. The figure shows that the economic downturn at

the beginning of 2009 hit Finland much harder than Sweden. This event is in the

middle of our study period, and thus might invalidate our identification strategy.

This divergence in the value added is even more pronounced in the manufacturing

sectors, as is visible from Figure 18. However, when we plot the same statistics for all

services in Figure 19, there seems to be no divergence before mid-2013. Therefore, it

seems that the different development post-2008 in the national aggregates is mostly

driven by manufacturing and other industries but not by the service sector, which

includes cleaning services. Thus, the comparison across the two countries around the

2007 reform seems to be a valid strategy when the focus is on the service sector.

Note that all services include retail trade, hospitality, transportation and other large

service industries, the cleaning industry being one small part of this group. Thus,

this macro-level analysis presents the economic background of the analysis but is not

a test of the effectiveness of the HTC in the cleaning industry.

To show country-level statistics relevant to the renovation industry (not included

in manufacturing, for example), in Figure 20, we use aggregated micro data on the log

changes in total sales in renovation industries by countries over time relative to year

2007. There seems to be a clear dip in Finland due to the Great Recession starting

in 2009 that leaves a clear gap between the trends across countries. This obviously

creates challenges in identifying the effects of HTC by comparing the renovation

industry in these countries. Thus, as explained above, we use an alternative approach,

CEM matching within sectors in Finland, to study the renovation sectors around the

HTC increase in 2009.

139

5 The Impact of the HTC on Consumption

In this section we provide evidence on the impact of HTC on the value of reported

sales among firms providing cleaning or renovation services and, therefore, on the

consumption of household services and on tax evasion. We first perform the analysis

on the cleaning industry by comparing Swedish firms to Finnish firms in the cleaning

industry before and after the introduction of the HTC system in Sweden. Second,

we analyze the renovation industry by comparing Finnish renovation firms to other

Finnish firms in similar industries before and after the large increase in the maximum

amount of HTC in January 2009.

5.1 Cleaning industry

We study the effect of HTC policies on consumption using the introduction of HTC

for cleaning services in Sweden in July 2007. We also study the responses to the

introduction of the invoicing system in 2009 in Sweden that switched the claiming

responsibility from consumers to firms, but this analysis suffers slightly from the

differential trends after the Great Recession.

We begin the analysis for the cleaning industry by showing the development of

monthly sales in our firm-level data for cleaning industry by country. We do this

by regressing month indicators against monthly sales (in logs) by country with firm-

level fixed effects. The series are weighted by market shares, giving a better picture

of changes in aggregate sales and thus aggregate consumption, i.e. we weight the

estimates with the winsorized pre-reform market shares of firms in the cleaning service

industry. The unweighted series are provided in figures 34–37 in the Appendix. Using

this method, we show the development of the average value of reported sales in

logarithmic values relative to the starting period. Figure 21 shows the development

of point estimates from this regression relative to June 2007, which is scaled to zero

in the figure. Two clear observations arise from the figure: first, there seems to be

a lot of seasonality in the data within years but this is apparently the case in both

countries. Second, the development before the introduction of HTC in Sweden in

2007 is very similar across the countries. This provides evidence that the pre-trends

are well in line with each other, supporting the identification assumptions required in

the DiD strategy. Figure 22 shows a comparable graph for inputs spending, which is

used to credit against paid VAT. The seasonality seems very similar to that presented

in Figure 21 and the pre-trends are well in line across countries.

140

To make sure that the seasonality does not cover any important responses, we next

exclude the within-year variation from the analysis. To facilitate the comparison of

the two series, we eliminate some of the seasonality that is constant across years from

these series by first regressing month indicator variables (January–December) on sales

and inputs, and then plotting the firm-level residuals from these regressions. Figure

23 shows the average firm-level residuals of sales regression by country. It is evident

that the pre-trends are very similar across countries, validating our comparison for

the cleaning industry.

The figure also shows our main result, which is that there is no change in the

reported value of sales among Swedish cleaning industry firms compared to Finnish

firms right after the introduction of HTC in July 2007. We do not observe any

response in the unweighted sales series either. This means that the introduction of

HTC in Sweden did not increase the value of reported sales among cleaning firms.

This seems to be the case for two years after the introduction, as the monthly reported

sales in the two countries follow each other very closely during this period. The result

implies no changes in demand for the services, because that cannot happen without

firms increasing their sales of services. The result also implies no increase in the prices

of services, because that would require an increase in the value of sales. Moreover,

the result implies no reduction in tax evasion among firms because that would require

the reported value of sales to increase. Finding no effect suggests a very low elasticity

of demand with respect to HTC.

In 2009, the Swedish HTC system was changed to the invoicing system where

firms can deduct the HTC directly from the client’s bill and charge the credit to the

Tax authority. This makes HTC more salient, as the buyer of services eligible for

HTC observes the effect immediately in lower cleaning service prices. Analyzing the

Swedish 2009 reform with these same figures, we do not see any increase in sales in

the weighted series immediately after the introduction of the invoicing system, but

some difference between the countries from 2011 onward. There are several potential

explanations for this divergence, the perhaps most obvious being the Great Recession,

which affected Finland and Sweden differently, with worse outcomes in Finland. In

fact, in Figure 23 we see that the sales seem to develop similarly in Sweden and decline

more in Finland, consistent with this hypothesis. Other potential explanations for this

divergence are that demand for cleaning services increases before the credit consumer

prices for services increase, or the amount of tax evasion decreases. Furthermore,

note that in January 2012, Finland cut both the level of maximum credit (from 3000

141

to 2000 euros) and the share of labor costs to be credited against individual-level

income taxes (from 60 to 45%), which makes it harder to compare countries after

2011. Nevertheless, we next examine some evidence that can help us to distinguish

between these mechanisms.

Firms’ input usage is informative on the development of their transactions, because

having more transactions should translate into more inputs being used. Figure 24

shows seasonality-adjusted and weighted development for inputs similarly as before

for sales in Figure 23. The pre-trends are very similar across countries. Again, we do

not observe any response to the introduction of the HTC in Sweden, in line with the

evidence on sales. The response to the introduction of the invoicing system in 2009

seems more spurious and will be discussed further together with the DID-estimates.

Figure 25 shows the development of the log value of aggregate sales in Finland and

Sweden. Notably, in this figure the aggregate value increases more in Finland than in

Sweden after 2009. From this we interpret that if anything happens in Sweden after

2009, it seems to be a shift of economic activity from larger firms to smaller firms,

and more likely to be caused by the Great Recession and the possible fiscal stimulus

response to that than the HTC reform in 2009. This intuition is supported by the

evidence in Figure 26 in the Appendix, showing that there is a large increase and

strong post-trends among smaller firms in Sweden, but much more modest develop-

ment among larger firms.

To study how consumer prices change after 2009, we show the development of

aggregated price levels for cleaning services across countries. These measures are

collected by the Statistical Offices for the purposes of calculating the consumer price

index. The price indices for cleaning services are plotted in the right panel of Figure

27. The left panel of Figure 27 shows the development of the overall CPI index across

countries. The prices of cleaning services seem to follow each other well before the

introduction of HTC in Sweden in 2007. After the reform, the price levels in Sweden

increase somewhat compared to Finland. The indices diverge even further after the

2009 reform, suggesting that increasing service prices seems to be a relevant expla-

nation for part of the divergence in the value of reported sales. Higher prices mean

the value of sales increases without an increase in the amount of services provided.

This together with the effects of the Great Recession suggests a limited role for HTC

causing the divergence between value of sales in the two countries in 2011.

In addition, we study annual tax filing measures such as firm-level taxable an-

nual profits and labor costs. Unfortunately, our annual data is limited to the period

142

starting in 2007, so we cannot study how these outcomes responded to the introduc-

tion of the HTC. Therefore, we study how profit and labor costs changed after the

implementation of the invoicing system in Sweden, and keep in mind that the Great

Recession may have caused some of the differences across the countries. Another

important point to keep in mind is that firm types in the cleaning industry are very

heterogeneous. Therefore, profit and labor cost measures are not identical for all

firms. For sole proprietors, income from the firm is often mainly distributed as profit,

while such firms often do not report any wage costs. On the other hand, for firms

organized as corporations, reported labor costs may include the wages of the owners

themselves14, and especially for small corporations the owners’ wages may compose

a large share of the wage costs of the firm.

Figure 28 shows the development of taxable profits among cleaning firms in Swe-

den and Finland over time. The first panel shows the development of log taxable

profits relative to 2008, estimated in firm-level data with firm fixed effects and mar-

ket share weights. In this case there are already some diverging developments across

the countries even before 2009. We do observe some further divergence after 2009,

but at this point the evidence points more towards the Great Recession than to other

factors. However, note that when using logarithmic transformation for taxable profit,

negative and zero profits are not included in the analysis. Therefore, in the second

panel we plot the firm fixed-effect regression results for an indicator variable of report-

ing any positive profit over time using a similar approach as for log taxable profits.

There seems to be an increase in reporting any positive profit among Swedish firms

compared to Finnish firms. The third panel shows an aggregate level measure of the

two first panels, the sum of all taxable profits in the industry by country. The figure

shows that, at the industry level, there is a clear increase in taxable profits in Swe-

den relative to Finland after the introduction of the invoicing system and the Great

Recession. All the figures include all firms in the sector, not just firms that provide

HTC-eligible services.

Furthermore, we study the effects on firm-level annual reported labor costs. If

labor costs increase after the introduction of HTC in Sweden, it could be due to

the Great Recession or three other factors: (1) wages of employees increase, (2) the

number of working hours of all employees increase, or (3) wages of owners increase.

The first mechanism would indicate that firms are sharing some of the increased

14The owners may want to distribute some of the income as wages instead of dividends for taxpurposes.

143

profit margin with their employees. The second mechanism implies that employment

in the HTC industry increases. The third mechanism would indicate no changes in

employment or wage rates for employees, but rather follow from some business owners

preferring to pay part of the increased profitability as wages instead of capital income

due to tax purposes15. We cannot directly observe the number of working hours within

firms, and therefore it is not possible to give definite answers as to the contribution

of each channel. However, we observe the total labor costs at the firm-level, which

include all these potential explanations.

Figure 29 shows the development of annual labor costs. We find that the trends

in log labor costs relative to the year 2008 among Finnish and Swedish firms are

somewhat similar before 2009. We find an increase in labor costs after 2009 among

Swedish firms compared to Finnish cleaning industry firms. As a large fraction of firms

in the cleaning service sector are sole proprietors, many of these small businesses do

not even report labor costs as they receive their taxable income as profit. Thus, in

the second panel we plot the firm fixed effect regression results with a binary variable

denoting reporting any positive labor costs as a dependent variable. The figure shows

quite limited effects. The third panel plots the aggregate nominal labor costs in the

cleaning sectors in Sweden and in Finland. The figure suggests that while there are

positive effects at firm-level, on the aggregate level the effect of the invoicing system

on labor costs is modest, given the preceding positive trend in both countries. This

indicates that the positive effects are likely driven mostly by the negative effects of

the Great Recession in Finland and to a lesser extent some firms’ owners being able

to profit more from their firms.

Finally, we report the main regression estimates using the Difference-in-Differences

approach described in Section 3.3. Tables 4 and 5 report the regression results es-

timated following equation 1. Table 4 shows the results for sales and inputs with

three different specifications. As we do not want to include additional reforms to the

studied period, we limit the data to the end of 2011. For sales or input usage, there

is no significant response to the introduction of HTC system in Sweden in July 2007,

consistent with the graphical evidence above. For the introduction of the invoicing

system in 2009, we find only weakly statistically significant positive coefficients that

are not very robust, reflecting mainly increases from 2010 onward, which as we ana-

lyzed above, most likely reflect the Great Recession more than the effects of the 2009

15There are a lot of small firms in the cleaning industry, especially in Sweden, making the role ofthe owner particularly relevant (Fig. 15).

144

reform.

Table 5 reports the estimation results for profit and labor costs before and after

the 2009 change to the invoicing system. Due to data limitations in the annual

variables, we cannot study the first reform with the regression analysis. As the

reform took place in the middle of the year, we also exclude the reform year from

the data. The last year included in the estimation is 2011, to avoid the reform of

2012 in Finland affecting the results, and to consider the differential macro trends

in the services sectors of the countries after 2012. Both profit and labor costs show

positive coefficients that are not very precisely estimated. The point estimates are

positive but again weakly statistically significant. Most consistently, the extensive

margin responses are statistically significant, suggesting that after the 2009 reform

in Sweden firms start to report positive taxable income and labor costs compared to

the pre-reform period and Finland.

5.2 Renovation industry

In this section, we study how Finnish renovation firms respond to the increase in

the maximum amount of HTC from 1150 to 3000 euros in 2009. In Section 4.3

we discussed that, while the cleaning industry in Finland and Sweden had evolved

similarly over this period, the trends in renovation were notably different due to

the Great Recession treating the two countries differently. In addition, there was a

simultaneous increase in the maximum HTC in Finland together with the introduction

of HTC in Sweden. Therefore, we use a different approach for renovation than we

used for the cleaning industry, and construct a domestic control group for Finnish

renovation industries using firms from industries providing broadly considered similar

services to the renovation industries. The treated industries include building, flooring,

roofing, and installation work among others. We choose car retail and repair and

logistics as our control industries as they are relatively similar in terms of size and

cyclicality to the renovation industry, but these firms do not provide services that

could be used to claim HTC16. Then we use coarsened exact matching to select

firms into control and treatment groups from these industries and give them weights

according to which they are used in the subsequent analysis. The parameters we

use for CEM-matching are yearly sales, sales growth in the previous year, and labor

costs. The CEM-matching method as well as the empirical approach are discussed in

16The treated and control industries are listed in tables 8 and 9 in the Appendix.

145

greater detail in Section 3.3.

We study the effects of HTC using as variation the increase in the maximum

amount of credit for renovation services from 1150 euros to 3000 euros in January

2009. This change is the largest single increase there has been in the Finnish system.

Therefore, we argue that such a large increase is least likely to go unnoticed and

should have an impact on demand and employment if the HTC size is effective in

that regard. The deductible share of 60% for renovation services did not change in

the reform.

In Figure 30, we examine how the share of HTC claims of total sales develops

around the HTC increase between the renovation and control firms. We use infor-

mation from individual’s HTC claims and allocate these to each firm. We then sum

up these HTCs at the firm-year level and calculate the size of their share of total

firm-level annual sales. This share is plotted in Figure 30 from 2007 to 2011. The

figure shows clearly that after the increase in the maximum tax credit in 2009, the

claimed HTC share increases in the repair firms. There is no similar increase in the

control group that provides evidence supporting our empirical approach. However,

the increase in tax credit claims does not necessarily imply higher consumption of

these services, but merely that taxpayers use tax credit more than before, out of

services that they might have consumed in any case. Next, we study how the HTC

increase affects the consumption of these services by studying the firm-level reported

value of sales and inputs as we did for the cleaning industry above.

Figure 31 plots the development of CEM-weighted monthly sales of renovation

firms and the control group relative to January 2008. The figure shows the monthly

sales relative to January 2008 separately for both groups. The figure shows that the

trends before the tax credit change are parallel, providing support for our identifica-

tion strategy. After the HTC change, the trends develop very similarly. If the increase

in HTC were to have a positive effect on sales in renovation firms, it should show up

here as increased sales relative to the control group despite the falling trend in both

groups. In addition, if the change were to have a negative effect on tax evasion, there

should also be an increase in VAT reported sales in comparison to the control group.

However, we observe no increase in the sales of renovation firms: if anything, there

is a decrease in the renovation sector compared to the control group after January

2009.

In Figure 32 we plot the same regression results for material input usage. The

trends before 2009 again seem very parallel, validating our setup. There is no dif-

146

ferential increase after the change in HTC, in line with the earlier figure showing no

increase in sales in the renovation sector relative to the control group. Furthermore,

Figure 33 plots the yearly effects of profit and labor costs respectively relative to year

2008. Also, these outcomes suggest that there are no strong responses to the large

increase in the household tax credit.

Tables 6 and 7 report the difference-in-differences results for the 2009 HTC in-

crease, actually suggesting a modest decrease in sales. Figures 31–33 already sug-

gested that there is no clear increase in sales, material input usage, profit or labor

costs of firms. The evidence in section 4.2 showed that there is no notable increase in

the number of firms either. Therefore, we conclude that there is no increase in sales

or in other outcomes, implying that there is no increase in consumption of service

nor in employment in the industries. If the increase in HTC were to lead to less tax

evasion, this would also show up as an increase in sales reported in VAT filing. As

there is no increase in sales, the results suggest no significant effect on tax evasion

either. In sum, the analysis in this section shows that we were not able to find any

effect of the expansion of maximum amount of HTC on the number of services sold

among renovation firms or tax evasion from these. We need to interpret these results

with some caution because, after all, the economic conditions were not very stable

during this period, and this may have affected industries differently.

6 Conclusions

In this study we have shown the consumption patterns of household services and, with

the help of a number of reforms in HTC systems in Finland and Sweden, studied to

what extent HTC could increase the consumption of services.

The results on the impact of HTC on the consumption of services or reported

sales of firms showed very limited effects. Our strongest evidence comes from the

introduction of HTC in Sweden in 2007 for cleaning services. We use Finland as

a control group and show that the comparison is valid in terms of stable economic

conditions surrounding the reform and the two groups developing similarly over time

before the reform. Despite the sizable HTC and the fact that many individuals use it,

we find no evidence of HTC increasing the reported value of firm sales or inputs. This

result provides no support for HTC causing increased cleaning services consumption

among households, or decreased tax evasion among cleaning firms reporting their sales

to tax authority. We also analyze the change of invoicing system from consumers

147

to firms in Sweden in 2009 and an increase in the maximum amount of HTC in

renovation industry in Finland in 2009 from 1150 euros to 3000 euros. Both analyses

are slightly sensitive to the Great Recession starting to affect Finland and Sweden

broadly in precisely 2009, and creating potentially differential trends across countries

and industries that are not related to HTC. Nevertheless, the evidence we can present

is consistent with the 2007 Swedish reform showing no clear increases in the number

of services provided to household due to changes in HTC.

The consumption patterns revealed that, although the consumption of household

services has been steadily increasing, it is still the case that in any given year only

a small share of all households use cleaning, renovation, or care services. For HTC

we showed that the higher the income of the household, the more they utilize HTC

on average. The fact that HTC is a credit against income taxes may constitute a

reason why HTC does not impact service consumption in lower-income households.

Moreover, we found that many taxpayers are not aware of the HTC rules. This

was visible in the administrative data as apparent mistakes that individuals make in

claiming the HTC.

Our results create a certain tension with the view often stated in public discussion

that HTC would be very effective in creating employment through increasing the con-

sumption of household services. There is a gradual rise in the house cleaning sector

in both countries, as observed in Section 4.2. According to the evidence presented

in this paper, this seems to follow from pre-existing trends driven by e.g. increasing

income, global trends in the growing service economy, urbanization, or other mecha-

nisms. These increasing trends occur at the same time as the introduction of HTC

policies, which may be partly why individuals perceive that HTC causes increased

consumption, where it is just a question of spurious pre-existing trends.

Many previous studies that have examined policies closely related to HTC, such as

reduced VAT rate experiments for labor-intensive services, find virtually no effect on

consumption of services or employment in the industries17. This low elasticity suggests

that even if the prices decrease, it does not increase the consumption of these services

significantly. In comparison to salient changes in prices (e.g. VAT rate changes),

the impact of HTC on prices may be less salient. In the HTC system, consumers

need to first pay the price of the service and then later receive the tax credit on their

income taxes, which is not directly related to the price of service. Understanding

the impact on prices requires that consumers know the HTC rules. Our analysis

17See e.g. (Kosonen, 2015); (Harju et al., 2018) and (Benzarti et al., 2020).

148

of the administrative data on usage of services revealed mistakes in HTC claiming.

These results demonstrate that a large share of consumers are not able to calculate

the effective price after taking HTC into account, which would further dampen the

incentive effect created by the HTC, and in turn reduce any changes in consumption

patterns.

The other main aim of HTC is to reduce tax evasion. This is based on the idea that

the tax credit incentivizes customers to require receipts for their payments and report

them to tax authorities so that they can claim tax credits. Note that a reduction

in tax evasion could occur even in the absence of any real changes in consumption:

consumers would simply switch their purchasing of the services from tax evading

firms to non-evading firms that provided the necessary receipts from the transactions.

However, as our results are on the reported sales of firms, this outcome also includes

tax evasion. If tax evasion were to decrease, reported sales would increase. However,

we did not observe any increase in reported sales as a response to the introduction

of the HTC for cleaning services nor to the increase in the maximum HTC amount

for renovation. This result is particularly strong for cleaning services. We find no

indication of such effects from repair services either, but given that the analysis suffers

from a simultaneously occurring economic downturn, the result leaves room for some

negligible tax evasion reduction. This could be analyzed in greater depth by studying

the change in the percentage that is deductible as HTC.

To sum up the most important policy conclusions, HTC policies seem largely

ineffective in achieving the stated objectives of increasing employment or curbing tax

evasion. They do not seem to increase the demand for services, increase employment

in labor-intensive services, or reduce tax evasion to a significant extent. Finally, the

HTC is a relatively expensive subsidy for consumers, with an annual income tax loss

of over 400 million euros, covering 45 to 60% of the labor costs from cleaning and

repair services. In addition, the HTC system increases income inequality. We show

that the average usage of HTC policy is steadily increasing in household disposable

income, i.e. high income households receive a large share of the tax credits.

149

References

A.B. Atkinson and J.E. Stiglitz. The design of tax structure: Direct versus indirect

taxation. Journal of Public Economics, 6(1):55 – 75, 1976. ISSN 0047-2727. doi:

https://doi.org/10.1016/0047-2727(76)90041-4.

Youssef Benzarti and Dorian Carloni. Who really benefits from consumption tax

cuts? evidence from a large vat reform in france. American Economic Journal:

Economic Policy, 11(1):38–63, February 2019. doi: 10.1257/pol.20170504.

Youssef Benzarti, Dorian Carloni, Jarkko Harju, and Tuomas Kosonen. What Goes

Up May Not Come Down: Asymmetric Incidence of Value-Added Taxes. Journal

of Political Economy, 128(12), October 2020.

Matthew Blackwell, Stefano Iacus, Gary King, and Giuseppe Porro. Cem: Coars-

ened exact matching in stata. The Stata Journal, 9(4):524–546, 2009. doi:

10.1177/1536867X0900900402.

Angus Deaton. Optimally uniform commodity taxes. Economics Letters, 2(4):357–

361, 1979. ISSN 0165-1765. doi: https://doi.org/10.1016/0165-1765(79)90050-8.

Annabelle Doerr and Sarah Necker. Collaborative tax evasion in the provision of

services to consumers – a field experiment, 2020.

Sami Gronberg and Timo Rauhanen. Kotitalousvahennys pienituloisen elakelaisen

nakokulmasta. VATT Institute for Economic Reasearch, 2015.

Jarkko Harju, Tuomas Kosonen, and Oskar Nordstrom Skans. Firm types, price-

setting strategies, and consumption-tax incidence. Journal of Public Economics,

165(C):48–72, 2018. doi: 10.1016/j.jpubeco.2018.06.

Iida Hakkinen-Skans. Kotitalouspalveluiden verovahennykset Suomessa ja Ruotsissa.

VATT Institute for Economic Reasearch, 2011.

Henrik Kleven. Optimum taxation and the allocation of time. Jour-

nal of Public Economics, 88(3):545 – 557, 2004. ISSN 0047-2727. doi:

https://doi.org/10.1016/S0047-2727(02)00192-5.

Henrik Jacobsen Kleven. Bunching. Annual Review of Economics, 8(1):435–464,

2016. doi: 10.1146/annurev-economics-080315-015234.

150

Tuomas Kosonen. More and cheaper haircuts after vat cut? on the efficiency and

incidence of service sector consumption taxes. Journal of Public Economics, 131

(C):87–100, 2015.

James Mirrlees, Stuart Adam, Tim Besley, Richard Blundell, Stephen Bond, Robert

Chote, Malcolm Gammie, Paul Johnson, Gareth Myles, and James M. Poterba.

Tax by design. Oxford University Press, 2011. ISBN 978-0-19-955374-7.

Joana Naritomi. Consumers as tax auditors. American Economic Review, 109(9):

3031–72, September 2019. doi: 10.1257/aer.20160658.

Dina Pomeranz. No taxation without information: Deterrence and self-enforcement

in the value added tax. American Economic Review, 105(8):2539–69, August 2015.

doi: 10.1257/aer.20130393.

Government’s proposition. Hallituksen esitys, 140/2000. URL

https://www.finlex.fi/fi/esitykset/he/2000/20000140.

F. P. Ramsey. A contribution to the theory of taxation. The Economic Journal, 37

(145):47–61, Mar 1927.

Riksrevisionen. Rutavdraget – konsekvenser av reformen, 2020.

Skatteverket. Fordjupad analys av rot- och rutavdrag, investeringssparkonto och

delagare i famansbolag ur ett jamstalldhetsperspektiv. Skattesystemets utveckling

2006-2015 Del II, 2011.

151

7 Figures

Figure 3: Household tax credit usage in Finland

010

020

030

040

050

0M

illion

s of

eur

os

1000

0020

0000

3000

0040

0000

Rec

ipie

nts

2001 2003 2005 2007 2009 2011 2013 2015 2017

N:o recipients Aggr. cost

Note: The black line and the left scale denote the number of consumers claiming the HTC and thebars on the right vertical axis denote the aggregate sum of HTC granted. Euros are in nominalvalues. Vertical lines denote changes in the HTC system. Source: Statistics Finland.

152

Figure 4: Household tax credit usage in Sweden

050

0010

000

1500

020

000

2500

0M

illion

s of

SEK

050

0000

1000

000

1500

000

2000

000

Rec

ipie

nts

2007 2009 2011 2013 2015 2017

N:o recipients Aggr. cost

Note: The black line and the left scale denote the number of consumers claiming the HTC and thebars on the right vertical axis denote the aggregate sum of granted HTC. Krona are in nominalvalues. Vertical lines denote changes in the HTC system. Source: Statistics Sweden.

153

Figure 5: Income distribution of HTC recipients and other taxpayers

Note: This figure plots the income distribution of HTC recipients in the Income Distribution Statis-tics of 2014. Income Distribution Statistics is a representative sample of Finnish taxpayers main-tained by Statistics Finland. The vertical axis denotes the percentage of taxpayers in each 1000-euro-income-bin. The distribution is winsorized at the 1 percent level from each end. The dashedvertical lines depict the 5th, 25th, 50th, 75th, and 95th income percentiles.

154

Figure 6: Simulated amount of claimed HTC by household income and two reformscenarios

Note: This figure shows, in blue, the average claimed HTC by household income simulated withthe SISU-microsimulation model, which is based on a large representative administrative data setof Finnish taxpayers. The blue line in the figure shows that, on average, claiming increases steadilywith household disposable income. The figure also shows simulations of two reform scenarios, onewhere the maximum HTC amount is increased by 1000 euros (red) and one where the maximum isdecreased by 1000 euros (yellow). The simulations are static and do not attempt to anticipate howthe consumption of services or credit claims would change in response to changes in policy.

155

Figure 7: Distribution of HTC claims in 2009–2011

020

000

4000

060

000

Freq

uenc

y

0 1000 2000 3000 4000 5000 6000 7000

HTC eligible labor costs per buyer

050

000

1000

0015

0000

Freq

uenc

y

0 500 1000 1500 2000 2500 3000

HTC per buyer

Note: The left figure plots the distribution of HTC-eligible labor costs of the total HTC claim(including all services consumed that year) in 2009–2011, when the maximum credit was 3000 eurosand the minimum labor costs to receive that were 5166.66 euros. The distribution of the effectiveHTC is displayed on the right. In the left graph the horizontal axis depicts the amount of laborcosts in the claim and the vertical line the number of tax payers within each 100-euro-category. Theright figure plots the distribution of granted household tax credits with the size of the HTC on thehorizontal line and the number of tax payers within each 100-euro category on the vertical line.

Figure 8: Frequency of HTC claims in 2009–2011

5000

1000

015

000

2000

025

000

Freq

uenc

y

1500 2000 2500 3000 3500 4000 4500

Observed Counterfactual

Excess mass: 1.002

050

0010

000

1500

0Fr

eque

ncy

3100 3600 4100 4600 5100 5600 6100 6600 7100

Observed Counterfactual

Excess mass: 2.857

Note: The left figure depicts bunching at 3000 euros of labor costs and the right figure bunchingat 5166.66. In order to claim the full credit, the labor costs should be at least 5166.66 euros, sobunching at 3000 euros is not in any way incentivized. The excess mass estimate is calculated incomparison to the counter-factual distribution, which is estimated as a seveth degree polynomialwhen excluding the bunching regions denoted with the grey vertical lines.

156

Figure 9: Distribution of HTC claims in 2012–2013

010

000

2000

030

000

4000

0Fr

eque

ncy

0 1000 2000 3000 4000 5000 6000 7000

HTC eligible labor costs per buyer

2000

040

000

6000

080

000

1000

00Fr

eque

ncy

0 500 1000 1500 2000 2500 3000

HTC per buyer

Note: The left figure plots the distribution of HTC-eligible labor costs of the total HTC claim(including all services consumed that year) in 2012–2013 when the maximum credit was 2000 eurosand the minimum labor costs to receive that were 4666.66 euros. The distribution of the effectiveHTC is displayed on the right. In the left graph the horizontal axis depicts the amount of labor costsin the claim and the vertical line the number of tax payers within each 100-euro category. The rightfigure plots the distribution of granted household tax credits with the size of the HTC on horizontalline and the number of tax payers within each 100-euro category on the vertical line.

Figure 10: Frequency of HTC claims in 2012–2013

5000

1000

015

000

2000

0Fr

eque

ncy

1000 1500 2000 2500 3000

Observed Counterfactual

Excess mass: .786

2000

4000

6000

8000

1000

012

000

Freq

uenc

y

2100 2600 3100 3600 4100 4600 5100 5600 6100 6600 7100

Observed Counterfactual

Excess mass: 2.396

Note: The left figure depicts bunching at 2000 euros of labor costs and the right figure bunchingat 4666.66. In order to claim the full credit the labor costs should be at least 4666.66 euros, sobunching at 2000 euros is not incentivized in any way. The excess mass estimate is calculated incomparison to the counter-factual distribution, which is estimated as a seventh degree polynomialwhen excluding the bunching regions denoted with the grey vertical lines.

157

Figure 11: Distribution of HTC claims in 2014–2016

020

000

4000

060

000

Freq

uenc

y

0 1000 2000 3000 4000 5000 6000 7000

HTC eligible labor costs per buyer

050

000

1000

0015

0000

2000

00Fr

eque

ncy

0 500 1000 1500 2000 2500 3000

HTC per buyer

Note: The left figure plots the distribution of HTC-eligible labor costs of the total HTC claim(including all services consumed that year) for 2014–2016, when the maximum credit was 2400 eurosand the minimum labor costs to receive that were 5555.55 euros. The distribution of the effectiveHTC is displayed on the right. The horizontal line depicts the amount of labor costs in the claimand the vertical line the number of tax payers within each 100-euro category. The right figure plotsthe distribution of granted household tax credits with the size of the HTC on the horizontal lineand the number of tax payers within each 100-euro category on the vertical line.

Figure 12: Frequency of HTC claims in 2014–2016

1000

015

000

2000

025

000

3000

0Fr

eque

ncy

1000 1500 2000 2500 3000 3500 4000

Observed Counterfactual

Excess mass: .626

050

0010

000

1500

020

000

Freq

uenc

y

3500 4000 4500 5000 5500 6000 6500 7000 7500

Observed Counterfactual

Excess mass: 1.918

Note: The left figure depicts bunching at 2400 euros of labor costs and the right figure bunchingat 5555.55. In order to claim the full credit, the labor costs should be at least 5555.55 euros, sobunching at 2400 euros is not incentivized in any way. The excess mass estimate is calculated incomparison to the counter-factual distribution, which is estimated as a seventh degree polynomialwhen excluding the bunching regions denoted with the grey vertical lines.

158

Figure 13: Aggregate sales within industry and the amount of HTC relative to ag-gregate sales (real values)

0.0

1.0

2.0

3.0

4H

TC s

hare

010

000

2000

030

000

4000

050

000

milli

on e

uros

2006 2008 2010 2012 2014

Renovation

0.0

1.0

2.0

3.0

4H

TC s

hare

010

0020

0030

0040

00m

illion

eur

os

2006 2008 2010 2012 2014

Cleaning

Aggregate sales Sweden Aggregate sales FinlandHTC share of sales Swe HTC share of sales Fin

Note: The bars denote the aggregate sales within industry in each country and the line plots theshare of claimed HTC relative to aggregate sales. Sales of firms and HTC are deflated to 2006price levels. Blue and red vertical lines denote changes in the Finnish and Swedish HTC systemsrespectively.

159

Figure 14: Number of firms in HTC sectors and those providing services that wereused to claim HTC

2000

4000

6000

8000

1000

012

000

2006 2008 2010 2012 2014

Cleaning0

2000

040

000

6000

080

000

2006 2008 2010 2012 2014

Renovation

SWE, all SWE, firms with HTC claimsFIN, all FIN, firms with HTC claims

Note: The dark red and blue bars plot the number of all firms in the HTC sector and the lighterbars the amount of those firms that provided services from which the customer received HTC. Blueand red vertical lines denote changes in the Finnish and Swedish HTC systems respectively.

160

Figure 15: Number of cleaning firms in size categories

0.1

.2.3

.4.5

.6Sh

are

010

0020

0030

0040

0050

0060

00

2006 2008 2010 2012 2014

Sales under 30,000

0.1

.2.3

.4.5

.6Sh

are

010

0020

0030

0040

0050

0060

00

2006 2008 2010 2012 2014

Sales 30,000 − 1 million

0.1

.2.3

.4.5

.6Sh

are

010

0020

0030

0040

0050

0060

00

2006 2008 2010 2012 2014

Sales above 1 million

Finland, all Finland, HTC Finland, shareSweden, all Sweden, HTC Sweden, share

Note: The dark red and blue bars plot the number of all cleaning firms in the size category andthe lighter bars the number of those cleaning firms that provided services from which the customerreceived HTC. The connected scatter plots show the shares of HTC firms of all firms in the sector.Sales of firms are deflated to 2006 price levels. Blue and red vertical lines denote changes in theFinnish and Swedish HTC systems respectively. Size categories set based on sales deflated to 2006price level.

161

Figure 16: Number of renovation firms in size categories

0.1

.2.3

.4.5

.6.7

Shar

e

010

000

2000

030

000

4000

050

000

2006 2008 2010 2012 2014

Sales under 30,000

0.1

.2.3

.4.5

.6.7

Shar

e

010

000

2000

030

000

4000

050

000

2006 2008 2010 2012 2014

Sales 30,000 − 1 million

0.1

.2.3

.4.5

.6.7

Shar

e

010

000

2000

030

000

4000

050

000

2006 2008 2010 2012 2014

Sales above 1 million

Finland, all Finland, HTC Finland, shareSweden, all Sweden, HTC Sweden, share

Note: The dark red and blue bars plot the number of all firms in the renovation industry in thesize category and the lighter bars the amount of those firms that provided services from which thecustomer received HTC. The connected scatter plots show the shares of HTC firms of all firms inthe sector. Sales of firms are deflated to 2006 price levels. Blue and red vertical lines denote changesin the Finnish and Swedish HTC systems respectively. Size categories set based on sales deflated to2006 price levels.

162

Figure 17: Logarithmic value added of all industries in both countries

−.4

−.2

0.2

.4

2000Q1 2003Q1 2006Q1 2009Q1 2012Q1 2015Q1 2018Q1

Finland Sweden

Note: Log value added in the economy relative to the last quarter of 2007. Blue and red verticallines denote changes in the Finnish and Swedish HTC systems respectively. The blue dashed verticallines denote changes in the household tax credit system in Finland and the red dashed vertical linesdenote changes in Sweden.

Figure 18: Logarithmic value added of manufacturing industry in both countries

−.6

−.4

−.2

0.2

2000Q1 2003Q1 2006Q1 2009Q1 2012Q1 2015Q1 2018Q1

Finland Sweden

Note: Log value added of manufacturing sector relative to the last quarter of 2007. Blue andred vertical lines denote changes in the Finnish and Swedish HTC systems respectively. The bluedashed vertical lines denote changes in the household tax credit system in Finland and the reddashed vertical lines denote changes in Sweden.

163

Figure 19: Logarithmic value added of service sector in both countries

−.5

0.5

2000Q1 2003Q1 2006Q1 2009Q1 2012Q1 2015Q1 2018Q1

Finland Sweden

Note: Log value added of service sector relative to the last quarter of 2007. Blue and red verticallines denote changes in the Finnish and Swedish HTC systems respectively. The blue dashed verticallines denote changes in the household tax credit system in Finland and the red dashed vertical linesdenote changes in Sweden.

Figure 20: Log difference of aggregate sales within renovation industry relative to2007

−.1

0.1

.2.3

2006 2008 2010 2012 2014

Finland Sweden

Note: This figure plots the log difference of aggregate sales within the renovation industry relativeto 2007. Sales of firms are deflated to 2006 price level. Blue and red vertical lines denote changes inthe Finnish and Swedish HTC systems respectively.

164

Figure 21: Cleaning – Firm-level trends in sales relative to June 2007

RUT starts in SWE

Firms can deduct directly in SWE

−.4

−.2

0.2

.4.6

log

turn

over

rela

tive

to J

une

2007

Jan2006 Jan2008 Jan2010 Jan2012 Jan2014

Finland Sweden

Note: Coefficients from a firm-fixed effect regression of log monthly turnover on month, binaryvariables relative to June 2007. Sales are adjusted to the price level of January 2006 and weightedby firm-level average annual market shares in 2006–2008. These weights are winsorized by 2% fromeach end. The blue dashed vertical lines denote changes in the household tax credit system inFinland and the red dashed vertical lines denote changes in Sweden.

165

Figure 22: Cleaning – Firm-level trends in input spending relative to June 2007

RUT starts in SWE

Firms can deduct directly in SWE

−.2

0.2

.4.6

log

sale

s re

lativ

e to

Jun

e 20

07

Jan2006 Jan2008 Jan2010 Jan2012 Jan2014

Finland Sweden

Note: Coefficients from a firm-fixed effect regression of log monthly input spending reported forVAT purposes on month, binary variables relative to June 2007. Material input spending refers toinput usage for which the firms have claimed VAT deductions. Input spending is adjusted to theprice level of January 2006 and weighted by firm-level average annual market shares in 2006–2008.These weights are winsorized by 2% from each end. The blue dashed vertical lines denote changesin the household tax credit system in Finland and the red dashed vertical lines denote changes inSweden.

166

Figure 23: Cleaning – Firm-level trends in sales relative to June 2007, seasonalityadjusted

RUT starts in SWE

Firms can deduct directly in SWE

−.2

0.2

.4.6

log

sale

s re

lativ

e to

Jun

e 20

07

Jan2006 Jan2008 Jan2010 Jan2012 Jan2014

Finland Sweden

RUT starts in SWE

Firms can deduct directly in SWE

−.2

0.2

.4.6

log

turn

over

diff

eren

ce

Jan2006 Jan2008 Jan2010 Jan2012 Jan2014

DD estimate 95% CI

Note: Coefficients from a firm-fixed effect regression of log monthly turnover on month, binaryvariables relative to June 2007. Sales are adjusted to the price level of January 2006 and weightedby firm-level average annual market shares in 2006–2008. These weights are winsorized by 2% fromeach end. The blue dashed vertical lines denote changes in the household tax credit system inFinland and the red dashed vertical lines denote changes in Sweden.

Figure 24: Cleaning – Firm-level trends in input spending relative to June 2007,seasonality adjusted

RUT starts in SWE

Firms can deduct directly in SWE

−.2

0.2

.4.6

log

sale

s re

lativ

e to

Jun

e 20

07

Jan2006 Jan2008 Jan2010 Jan2012 Jan2014

Finland Sweden

RUT starts in SWE

Firms can deduct directly in SWE

−.2

0.2

.4.6

log

turn

over

diff

eren

ce

Jan2006 Jan2008 Jan2010 Jan2012 Jan2014

DD estimate 95% CI

Note: Coefficients from a firm-fixed effect regression of log monthly material input spending onmonth, binary variables relative to June 2007. Material input spending refers to input usage forwhich the firms have claimed VAT deductions. Input spending is adjusted to the price level ofJanuary 2006 and weighted by firm-level average annual market shares in 2006–2008. These weightsare winsorized by 2% from each end.The blue dashed vertical lines denote changes in the house-holdtax credit system in Finland and the red dashed vertical lines denote changes in Sweden.

167

Figure 25: Log difference of aggregate sales within cleaning industry relative to 2007

0.1

.2.3

2006 2008 2010 2012 2014

Finland Sweden

Note: This figure plots the log difference of aggregate sales within cleaning industry relative to 2007.Sales of firms deflated to 2006 price level. Blue and red vertical lines denote changes in the Finnishand Swedish HTC systems respectively.

Figure 26: Cleaning – Firm-level trends in firms with 2008 sales below and above30,000 euros relative to June 2007, seasonality adjusted

RUT starts in SWE

Firms can deduct directly in SWE

−.2

0.2

.4.6

log

sale

s re

lativ

e to

Jun

e 20

07

Jan2006 Jan2008 Jan2010 Jan2012 Jan2014

Firms with sales under 30,000

RUT starts in SWE

Firms can deduct directly in SWE

−.2

0.2

.4.6

log

sale

s re

lativ

e to

Jun

e 20

07

Jan2006 Jan2008 Jan2010 Jan2012 Jan2014

Firms with sales above 30,000

Finland Sweden

Note: Coefficients from a firm-fixed effect regression of log monthly material input spending onmonth, binary variables relative to June 2007. Size groups are defined based on 2008 annual sales.Sales are adjusted to the price level of January 2006. The blue dashed vertical lines denote changesin the household tax credit system in Finland and the red dashed vertical lines denote changes inSweden.

168

Figure 27: General consumer price index and CPI of household maintenance

100

110

120

130

140

2005m1 2007m1 2009m1 2011m1 2013m1

General measure10

011

012

013

014

0

2005m1 2007m1 2009m1 2011m1 2013m1

Household services

Finland Sweden

Note: 2005=100. Source: Statistics Finland and Statistics Sweden.

169

Figure 28: Cleaning micro trends: Annual taxable profits

−.2

−.1

0.1

.2lo

g pr

ofits

rel

ativ

e to

200

8

2007 2008 2009 2010 2011 2012 2013

Log taxable profits

−.2

−.1

0.1

.2re

port

ing

posi

tive

prof

it

2007 2008 2009 2010 2011 2012 2013

Reporting positive profit

Finland Sweden

05

1015

2025

30S

wed

en, 1

00 m

illio

n kr

onas

0.5

11.

52

2.5

3F

inla

nd, 1

00 m

illio

n eu

ros

2007 2008 2009 2010 2011 2012 2013

Finland Sweden

Aggregate taxable profits

Note: Upper-left graph plots coefficients from a firm-fixed effect regression of log annual taxableprofits on year. Binary year variables capture the yearly profits relative to 2008. Taxable profit isadjusted to the price level of 2006 and weighted by firm-level market shares in 2006-2008. Theseweights are winsorized by 2% from each end. Blue dashed vertical lines denote changes in the taxcredit in Finland and red dashed vertical lines denote changes in Sweden. The second figure plotsthe coefficients from a firm-fixed effect regression of binary variable, with zero indicating reporting ofpositive profits that year. The third graph plots the aggregate nominal profits in cleaning sectors inSweden and in Finland. The figure suggests that the firm-level effect also appears on the aggregatelevel. As a response to the change in HTC system in Sweden in 2009, firms seem to make significantlymore profits.

170

Figure 29: Cleaning micro trends: Annual taxable labor costs−

.2−

.10

.1.2

log

labo

r co

sts

rela

tive

to 2

008

2007 2008 2009 2010 2011 2012 2013

Log labor costs

−.2

−.1

0.1

.2re

port

ing

posi

tive

labo

r co

sts

2007 2008 2009 2010 2011 2012 2013

Reporting positive labor costs

Finland Sweden

010

2030

40S

wed

en, b

illio

n kr

onas

0.2

.4.6

.81

Fin

land

, bill

ion

euro

s

2007 2008 2009 2010 2011 2012 2013

Finland Sweden

Aggregate labor costs

Note: First graph plots coefficients from a firm-fixed effect regression of log annual labor costs onyear. Binary year variables capture the yearly labor costs relative to 2008. Labor costs are adjustedto the price level of 2006 and weighted by firm-level market shares in 2006–2008. These weightsare winsorized by 2% from each end. Blue dashed vertical lines denote changes in the tax creditin Finland and red dashed vertical lines denote changes in Sweden. The second figure plots thecoefficients from a firm-fixed effect regression of binary variable, with zero indicating reporting ofpositive labor costs that year. While not reporting labor costs is common as there are many smallone-person businesses, more firms have started to report labor costs after 2009. The third graphplots the aggregate nominal labor costs in cleaning sectors in Sweden and in Finland. The figuresuggests that while at firm-level there are positive effects driven by small firms, on the aggregatelevel the effect of the reform on labor costs is modest given the preceding positive trend in bothcountries.

171

Figure 30: Renovation – HTC share of sales in treatment and control group in micro-level data

0.0

2.0

4.0

6

2007 2008 2009 2010 2011

Renovation Counterfactual

Note: This figure plots coefficients denoting the HTC share of firm’s annual sales, estimated witha CEM-weighted firm-fixed effect regression and 2008 as a baseline year. The dashed vertical linedepicts the increase in the tax credit starting in 2009. The figure shows that, after the increase inthe maximum tax credit, the claimed HTCs increase in relation to annual sales.

Figure 31: Renovation – Sales, seasonality adjusted in the micro-level data

−.2

−.1

0.1

.2lo

g sa

les

rela

tive

to J

an 2

008

Jan2007 Jan2008 Jan2009 Jan2010 Jan201

Renovation industry Control

Sales

Note: Coefficients from a CEM-weighted firm-fixed effect regression of log monthly sales on binarymonth-variables relative to January 2008. The dashed vertical line depicts the increase in the taxcredit starting in 2009. The monthly sales are reported by firms in their VAT filing.

172

Figure 32: Renovation – Input spending, seasonality adjusted in the micro-level data

−.2

−.1

0.1

.2lo

g in

puts

rel

ativ

e to

Jan

200

8

Jan2007 Jan2008 Jan2009 Jan2010 Jan201

Renovation industry Control

Inputs

Note: Coefficients from a CEM-weighted firm-fixed effect regression of log monthly inputs on binarymonth variables relative to January 2008. The dashed vertical line depicts the increase in the taxcredit starting in 2009. The monthly sales are reported by firms in their VAT filing.

Figure 33: Renovation micro trends — Annual taxable profit and labor costs

−.1

5−

.1−

.05

0.0

5.1

.15

2007 2008 2009 2010 2011

Renovation Counterfactual

Profit

−.1

5−

.1−

.05

0.0

5.1

.15

2007 2008 2009 2010 2011

Renovation Counterfactual

Labor costs

Note: Coefficients from a CEM-weighted firm-fixed effect regression of log annual profits/labor costson binary year variables. Year effects are relative to 2008. The dashed vertical line denotes thechange in the tax credit in 2009.

173

8 Tables

Table 1: Summary statistics of HTC claimers in Finland in 2006, 2010 and 2014

2006 2010 2014mean sd p50 mean sd p50 mean sd p50

Female 50.55% 51.26% 50.77%

Pensioner 35.56% 39.45% 45.43%

Age 53.55 15.77 53 55.25 16.05 55 57.88 16.14 58

Taxable income 45,914 160,083 30,620 47,750 134,565 33,954 52,375 159,556 36,373

Wage etc. 28,373 60,636 23,738 29,334 42,488 24,990 29,397 45,665 18,650

Pension 7073 13,378 0 9075 15,959 0 11,896 18,926 0

Capital income 7226 111,326 18 5598 74,058 0 7257 109,454 25

Dividends etc. 3242 50,224 10 3744 63,018 10 3825 57,163 10

Household tax credit 769 570 703 1088 996 717 905 816 625

HTC for renovation 484 463 333 856 1015 368 715 837 306

HTC for cleaning 254 556 0 197 457 0 155 369 0

HTC for care 29 231 0 28 241 0 22 192 0

Received HTC t-1 na 43.55% 45.58%

Observations 241,999 374,530 347,678

Note: This table describes the characteristics of an average HTC claimer. An individual is indicatedas being a pensioner if he/she received a pension that year. The monetary values are in nominalterms. Taxable income is the total taxable income of the taxpayer, including wages, capital income,pension, and social security. For comparison, an average (median) taxable income of a Finnishtaxpayer was 22,896 euro (18,596) in 2006, 25,852 euro (21,311) in 2010 and 28,800 euro (23,805)in 2014 (StatFin, 2020). Wages etc. are income taxed according to the earned income tax schedule,including wage, social security, and pensions. Capital income refers to income such as rental incomeand capital gains, but not dividends, which are included, together with interest on cooperativecapital, under dividends etc. Household tax credit is the average HTC from all service categoriesand the averages from each categories are listed below. The second to last row displays the share oftaxpayers receiving HTC who also received HTC the previous year.

174

Table 2: Summary Statistics of Cleaning Firms in 2008 and 2011

2008

FIN - All FIN - HTC SWE - All SWE - HTC firmsmean sd mean sd mean sd mean sd

Sales 251,066 6,152,179 374,494 8,250,249 7,894,563 122,566,511 8,698,661 130,390,236

Sales (VAT) 221,283 5,823,653 351,514 8,081,021 7,639,021 120,787,849 8,452,837 128,496,654

Inputs 62,959 1,694,584 97,103 2,350,603 2,687,781 50,500,016 2,921,155 53,711,007

Labor costs 110,059 3,107,979 173,905 4,167,780 9,798,417 163,016,191 10,898,092 172,568,277

Profit 28,557 212,003 34,615 283,723 411,005 3,049,407 421,182 3,190,059

Deducted VAT 13,851 372,808 21,363 517,133 671,945 12,625,004 730,289 13,427,752

HTC per firm 3827 17,781 7379 24,156 76,539 505,812 173,251 750,014

Limited company 19.7% 19.2% 38.2% 34.1%

Sole proprietorship 59.3% 63.4% 61.6% 65.7%

Observations 5125 2658 4466 1973

2011

FIN - All FIN - HTC SWE - All SWE - HTC firmsmean sd mean sd mean sd mean sd

Sales 294,032 7,664,358 410,950 9,634,425 5,171,012 98,714,240 4,134,546 85,345,578

Sales (VAT) 244,788 7,012,594 368,499 9,147,357 4,905,234 95,191,737 3,904,785 82,842,827

Inputs 76,830 2,482,480 114,811 3,237,848 1,581,677 35,360,268 1,218,349 29,400,497

Labor costs 129,744 3,678,880 187,527 4,624,295 7,585,304 140,074,961 6,709,386 137,213,395

Profit 31,434 323,340 36,979 401,140 325,652 2,890,889 295,535 2,744,936

Deducted VAT 17,671 570,970 26,406 744,705 395,419 8,840,067 304,587 7,350,124

HTC per firm 6105 31,876 10,394 41,057 110,422 603,420 215,624 829,707

Limited company 20.7% 21.5% 32.8% 27.4%

Sole proprietorship 58.5% 63.4% 66.9% 72.1%

Observations 5554 3262 7807 3998

Note: This table provides summary statistics for cleaning firms in Finland and Sweden in 2008 andin 2011. The values are reported in thousands and are for Finnish firms in nominal euros and forSwedish firms in nominal kronas. 10kr≈1e. Sales refer to sales reported in tax return and Sales(VAT) refers to sales reported in VAT filings. Inputs consists of material inputs such as materialand intermediate goods. Labor costs include payroll taxes and other side costs on top of wage, theincome of sole proprietors is taxed as profit, so labor costs are 0 for such firms. Profit refers totaxable profit reported in tax return and deducted VAT is the aggregate VAT deduction claims ofthe year. HTC per firm is calculated by assigning HTC claims to each firm based on the firm id,that customer has reported in one’s tax return.

175

Table 3: Summary Statistics of Finnish Renovation and Control Firms 2008

2008

Ren - All Ren - CEM Contr - All Contr - CEMmean sd mean sd mean sd mean sd

Sales 573,781 9,718,734 573,462 9,734,355 1,178,926 25,269,878 390,691 19,627,990

Sales (VAT) 500,552 8,532,272 500,918 8,545,035 711,186 9,546,311 204,054 344,499

Inputs 332,740 6,576,796 332,836 6,586,619 584,788 8,806,474 144,262 2,746,784

Labor costs 101,456 1,390,743 101,311 1,392,862 135,478 2,609,688 47,147 111,386

Profit 44,355 279,090 44,295 279,277 45,721 538,885 33,161 219,383

Deducted VAT 73,203 1,446,895 73,224 1,449,056 128,653 1,937,424 31,738 604,292

HTC per firm 2619 14,573 2621 14,577 43 682 38 467

Observations 35,573 35,466 34,970 33,158

2011

Ren - All Ren - CEM Contr - All Contr - CEMmean sd mean sd mean sd mean sd

Sales 550,484 10,187,001 633,057 11,035,656 1,085,343 21,226,480 410,996 15,888,364

Sales (VAT) 412,557 8,112,996 506,574 9,037,772 669,783 9,379,916 242,978 622,568

Inputs 318,884 6,924,618 388,898 7,688,949 533,254 7,951,531 149,151 548,542

Labor costs 97,244 1,293,214 112,442 1,420,415 140,047 2,532,293 56,537 148,650

Profit 35,375 204,571 39,267 229,932 41,918 301,580 33,488 168,323

Deducted VAT 73,343 1,592,662 89,447 1,768,458 122,649 1,828,852 34,305 126,165

HTC per firm 5993 33,086 6743 36,509 130 1648 130 1607

Observations 38,556 27,808 35,211 27,737

Note: This table provides summary statistics for renovation firms and the control industries inFinland in 2008 and in 2011. The exact industry codes included are listed in Appendix tables 8and 9. Sales refer to sales reported in tax return and Sales (VAT) refers to sales reported in VATfilings. Inputs consists of material inputs such as material and intermediate goods. Labor costsinclude payroll taxes and other side costs on top of wage. Profit refers to taxable profit reported intax return and deducted VAT is the aggregate VAT deduction claims of the year. HTC per firm iscalculated by assigning HTC claims to each firm based on the firm id, that a customer has reportedin one’s tax return.

176

Table 4: Differences-in-differences results for cleaning industry – reforms in Sweden

Sales Inputs

(1) (2) (3) (1) (2) (3)

1st reform estimate -0.006 -0.024 -0.017 0.041* 0.009 0.023

0.016 0.014 0.014 0.021 0.019 0.018

2nd reform estimate 0.109*** 0.041 0.049* 0.117*** 0.055 0.076**

0.027 0.023 0.023 0.032 0.028 0.027

Weighted X X X X X X

Firm fixed effects X X X X X X

Year fixed effects X X X X X X

Winsorized data X X X X

Seasonality adjusted X X

Constant 11.294*** 11.198*** 1.851*** 9.659*** 9.576*** 1.907***

0.018 0.016 0.015 0.022 0.020 0.019

N 464809 464809 451239 457422 457422 444408

* p < 0.05, ** p < 0.01, *** p < 0.001

Note: This table reports the difference-in-differences estimation results of the 2007 and 2009 reformsin Sweden for the cleaning industry following equation 1. All specifications are weighted by firm-levelmarket shares in 2006–2008. The weights are winsorized by 2% from each end. 1st reform refersto the introduction of the HTC for cleaning services and the second reform refers to the changeto the invoicing system in which firms claim the credit on behalf of the customers (no change inparameters). The first three columns report results for sales with different specifications and thesecond three columns report results for material input usage. The first specification includes firm andyear fixed effects, the second specification additionally uses data winsorized at the 1-percentage-pointlevel and the third specification is adjusted for monthly seasonality.

177

Table 5: Differences-in-differences results for cleaning industry: 2009 reform in Swe-den

Profit Labor Costs

(1) (2) (3) (1) (2) (3)

Ext. marg Ext. marg

2nd reform estimate 0.109 0.082 0.065*** 0.130* 0.120* 0.027***

0.076 0.065 0.015 0.054 0.051 0.006

Weighted X X X X X X

Firm fixed effects X X X X X X

Year fixed effects X X X X X X

Winsorized data X X

Constant 41.564 95.053 1.011 78.918 239.070 0.872

. 23656263.447 453562.934 55339836.413 85334128.796 .

N 18924 18924 23502 12923 12923 23502

* p < 0.05, ** p < 0.01, *** p < 0.001

Note: This table reports the difference-in-differences estimation results of 2009 reform in Swedenfor cleaning industry. All specifications are weighted by firm-level market shares in 2006–2008. Theweights are winsorized by 2% from each end. 2nd reform refers to the change to the invoicingsystem in which firms claim the credit on behalf of the customers (no change in parameters). Thefirst three columns report results for profit with different specifications and the second three columnsreport results for labor costs. The first specification includes firm and year effects and the secondspecification additionally uses data winsorized at 1-percentage-point. The third column for eachoutcome variable reports the extensive margin regression results, that is the effects on whether thefirm reported any profit or any labor costs.

178

Table 6: Difference-in-differences results for renovation industry – 2009 HTC increasein Finland

Sales Inputs

(1) (2) (3) (1) (2) (3)

DiD estimate -0.092*** -0.090*** -0.089*** -0.011 -0.012* 0.001

0.005 0.005 0.005 0.006 0.006 0.005

Weighted X X X X X X

Firm fixed effects X X X X X X

Year fixed effects X X X X X X

Winsorized data X X X X

Seasonality adjusted X X

Constant 9.079*** 9.075*** 0.060*** 7.734*** 7.725*** 0.215***

0.005 0.005 0.005 0.006 0.006 0.006

N 2914077 2914077 2914077 3114970 3114970 2851982

* p < 0.05, ** p < 0.01, *** p < 0.001

Note: This table reports the CEM-weighted difference-in-differences estimation results followingequation 2 for studying the effect of the HTC increase in 2009 on sales and inputs. The first threecolumns report the estimation results for sales and columns 4 to 6 for material input usage. Allspecifications use firm and year fixed effects as well as CEM-weights. The second specificationfor both dependent variables uses data winsorized at the 2,5-percentage-point level and the thirdspecification uses data that is additionally adjusted for monthly seasonality.

179

Table 7: Difference-in-differences results for annual profit and labor costs of renovationfirms – 2009 HTC increase in Finland

Profit Labor costs

(1) (2) (1) (2)

DiD estimate -0.030 -0.033** -0.043*** -0.043***

0.010 0.009 0.010 0.009

Weighted X X X X

Firm fixed effects X X X X

Year fixed effects X X X X

Winsorized data X X

Seasonality adjusted

Constant 9.821*** 9.829*** 10.586*** 10.583***

0.004 0.004 0.003 0.003

N 235796 235796 167287 167287

* p < 0.05, ** p < 0.01, *** p < 0.001

Note: This table reports difference-in-differences estimation results following equation 2 for studyingthe effect of the HTC increase in 2009 on taxable profit and labor costs. The first two columns reportthe estimation results for sales and columns 3 and 4 for labor costs. All specifications use firm andyear fixed effects as well as CEM weights. The second specification for both dependent variablesuses data winsorized at the 1-percentage-point level.

180

9 Appendix

Figures 34–37 show the visual evidence for the sales and input usage of firms without

the market share weights. In comparison to the main figures 21–24, here, smaller firms

receive a higher weight due to the log transformation. As in the main analysis, there

is no statistically significant response to the introduction of the HTC for cleaning

services in 2007. This supports the conclusion that neither demand nor employment

increased as a response to the HTC, at least during its first years. However, the

figures show a divergence between Sweden and Finland starting around 2010. The

evidence in Section 5.1 suggests that this divergence is driven by small firms and

some part of it is likely to be due to the different trends in the Finnish and Swedish

economies after the Great Recession.

Figure 34: Cleaning micro trends: Sales – Unweighted

RUT starts in SWE

Firms can deduct directly in SWE

−.2

0.2

.4.6

.8lo

g tu

rnov

er re

lativ

e to

Jan

200

8

Jan2006 Jan2008 Jan2010 Jan2012 Jan2014

Finland Sweden

Note: Coefficients from a firm-fixed effect regression of log monthly turnover on month, binaryvariables relative to January 2008. The blue dashed vertical lines denote changes in the householdtax credit system in Finland and the red dashed vertical lines denote changes in Sweden.

181

Table 8: Renovation industries studied in Section 5.2

Industry code Classification

31020 Manufacturing of kitchen furniture41200 Building43210 Electrical work43220 Heating, plumbing, AC43320 Carpenter43330 Floor and wall work43341 Painting43342 Glazing43910 Roofing47523 Kitchen retail47596 Locksmith’s work, security systems

Note: This table reports the 5-digit industry classifications used in analyzing the renovation sector.

Table 9: Control industries for renovation sector in Section 5.2

Industry code Classification

36 Water cleaning and distribution49 Ground transport50 Water transport51 Air transport52 Storage and transport services4511 Car and motorcycle retail45201 Motor vehicle repairs

Note: This table reports the 2, 4 and 5-digit industry classifications used as a control group whenanalyzing the renovation sector.

182

Table 10: Summary Statistics of Renovation Firms in Finland and Sweden 2008

2011

FIN - All FIN - HTC SWE - Allmean sd mean sd mean sd

Sales 560320 9117013 700798 12070470 8964286 246134779

Inputs 313544 6130713 406709 7945743 4470078 110604155

Labor costs 100509 1327300 131606 1798994 3500462 70969339

Profit 43994 265999 47094 249245 412677 2538188

Deducted VAT 68980 1348757 89476 1748063 1117520 27651039

HTC per firm 2303 13362 4732 18852 449 13139

Observations 43026 20939 48846

2011

FIN - All FIN - HTC SWE - All SWE - HTC firmsmean sd mean sd mean sd mean sd

Sales 553592 10255589 681274 12320895 7357937 208316451 7191473 192935034

Inputs 301284 6440975 431819 8729579 3919009 103918993 4244896 113620980

Labor costs 96536 1234532 122593 1592009 3014089 52771423 3123457 55915891

Profit 38174 425052 37791 221798 411400 6575191 364249 1287922

Deducted VAT 69295 1481424 99318 2007803 979752 25979748 1061224 28405245

HTC per firm 5503 32193 10253 43387 109038 461526 264056 689139

Observations 46178 24783 68764 28395

Note: This table provides summary statistics for renovation firms in Finland and Sweden in 2008and in 2011. Sales refer to sales reported in tax return and Sales (VAT) refers to sales reported inVAT filings. Inputs consists of material inputs such as material and intermediate goods. Labor costsinclude payroll taxes and other side costs on top of wage. Profit refers to taxable profit report intax return and deducted VAT in the aggregate VAT deduction claims of the year. HTC per firm iscalculated by assigning HTC claims to each firm based on the firm id, that customer has reportedin one’s tax return.

183

Figure 35: Cleaning micro trends: Inputs – Unweighted

RUT starts in SWE

Firms can deduct directly in SWE

−.2

0.2

.4.6

.8lo

g in

puts

rela

tive

to J

an 2

008

Jan2006 Jan2008 Jan2010 Jan2012 Jan2014

Finland Sweden

Note: Coefficients from a firm-fixed effect regression of log monthly input spending reported forVAT purposes on month, binary variables relative to January 2008. The blue dashed vertical linesdenote changes in the household tax credit system in Finland and the red dashed vertical linesdenote changes in Sweden.

Figure 36: Cleaning micro Diff-in-diff: Sales, seasonality adjusted, unweighted

RUT starts in SWE

Firms can deduct directly in SWE

−.2

0.2

.4.6

log

valu

e ad

ded

rela

tive

to J

an 2

008

Jan2006 Jan2008 Jan2010 Jan2012 Jan2014

Finland Sweden

RUT starts in SWE

Firms can deduct directly in SWE

−.2

0.2

.4.6

log

turn

over

diff

eren

ce

Jan2006 Jan2008 Jan2010 Jan2012 Jan2014

DD estimate 95% CI

Note: Coefficients from a firm-fixed effect regression of log monthly turnover on month, binaryvariables relative to January 2008. The blue dashed vertical lines denote changes in the householdtax credit system in Finland and the red dashed vertical lines denote changes in Sweden.

184

Figure 37: Cleaning micro Diff-in-diff: Inputs, seasonality adjusted, unweighted

RUT starts in SWE

Firms can deduct directly in SWE

−.2

0.2

.4.6

log

valu

e ad

ded

rela

tive

to J

an 2

008

Jan2006 Jan2008 Jan2010 Jan2012 Jan2014

Finland Sweden

RUT starts in SWE

Firms can deduct directly in SWE

−.2

0.2

.4.6

log

inpu

ts d

iffer

ence

Jan2006 Jan2008 Jan2010 Jan2012 Jan2014

DD estimate 95% CI

Note: Coefficients from a firm-fixed effect regression of log monthly turnover on month, binaryvariables relative to January 2008. The blue dashed vertical lines denote changes in the householdtax credit system in Finland and the red dashed vertical lines denote changes in Sweden.

185