Journal of Comparative Economics 41 (2013) 1092–1105

Contents lists available at SciVerse ScienceDirect

Journal of Comparative Economics

journal homepage: www.elsevier .com/ locate / jce

Shift from gross profit taxation to distributed profit taxation:Are there effects on firms?

0147-5967/$ - see front matter � 2013 Association for Comparative Economic Studies Published by Elsevier Inc. All rights reserved.http://dx.doi.org/10.1016/j.jce.2013.01.011

⇑ Corresponding author.E-mail addresses: Jaan.Masso@mtk.ut.ee (J. Masso), Jaanika.Merikyll@eestipank.ee (J. Meriküll), priit.vahter@ut.ee (P. Vahter).

Jaan Masso a, Jaanika Meriküll a,b, Priit Vahter a,⇑a Faculty of Economics and Business Administration, University of Tartu, Narva mnt 4, 51009 Tartu, Estoniab Eesti Pank (Bank of Estonia), Estonia pst 13, 15095 Tallinn, Estonia

a r t i c l e i n f o

Article history:Received 3 May 2012Revised 30 January 2013Available online 13 February 2013

JEL classification:H25H32D22O16

Keywords:Capital structureComparative economic developmentCorporate income taxInvestmentsLiquidity

a b s t r a c t

Masso, Jaan, Meriküll, Jaanika, and Vahter, Priit—Shift from gross profit taxation to dis-tributed profit taxation: Are there effects on firms?

This paper investigates the consequences of the corporate tax reform in Estonia in 2000.This unique reform nullified the taxation of retained earnings and maintained corporateincome tax only on distributed profits. We investigate the outcome of the reform by com-paring the performance of the affected firms in Estonia with that of firms from Latvia andLithuania, the two other Baltic countries. We use firm-level financial data and the differ-ence in differences approach for our analysis. The results are consistent with an increasein holdings of liquid assets and lower use of debt financing after the reform. A positive rela-tionship of the reform with post-reform investment and productivity has also been found.The results point to a stronger effect on smaller firms. Journal of Comparative Economics 41(4) (2013) 1092–1105. Faculty of Economics and Business Administration, University ofTartu, Narva mnt 4, 51009 Tartu, Estonia; Eesti Pank (Bank of Estonia), Estonia pst 13,15095 Tallinn, Estonia.� 2013 Association for Comparative Economic Studies Published by Elsevier Inc. All rights

reserved.

1. Introduction

Corporate income tax rates and their linkages with economic performance have received persistent attention in both aca-demic literature and policy debates. International tax competition has reduced the taxation of capital in recent decades(Devereux et al., 2002). Falling statutory corporate income tax rates have been coupled with a widening of the tax base.Several studies have endeavoured to use these reforms to study the effect of taxes on international profit shifting (see,e.g., the meta-analysis by De Mooij and Ederveen, 2008), debt shifting (e.g., Egger et al., 2010), and investment andproductivity (Vartia, 2008; Schwellnus and Arnold, 2008).

In this paper we estimate how the Estonian corporate income tax reform in 2000 is associated with changes in the capitalstructure, liquidity, investment, and productivity of firms. The reform of 2000 introduced a system that was unique in theworld, as the reform meant that firms’ profits are taxed only if they are distributed to shareholders in the form of dividends,

J. Masso et al. / Journal of Comparative Economics 41 (2013) 1092–1105 1093

while retained earnings are untaxed.1 Unlike the previous system, taxation is postponed until the moment of profit distribu-tion.2 This marks a difference from the trend in most countries because the reform narrowed the tax base and left the tax rate ata relatively high level – 26%.3 As the law was adopted on 15 December 1999 and came into force immediately from 1 January2000 we can argue that there was no anticipation effect because the period of notice was so short and the stimuli for changes inthe behaviour of firms emerged in 2000, not earlier.

The government expected the reform to promote investment, create new jobs and promote entrepreneurship (accord-ing to a survey of policy-makers, see Tartu Ülikool and Praxis, 2010). However, related literature shows that several addi-tional consequences are possible, and we aim to evaluate these in our study. In addition to cross-country studies (e.g.,Schwellnus and Arnold, 2008), empirical investigations have also looked at the impact of tax reforms (e.g., Kari et al.,2009 on Finland) on firms. The advantage of our study is the focus on the consequences of a big change in tax rates asa result of the reform: the statutory tax rate on retained earnings dropped from 26% to 0% due to the reform and the aver-age implicit tax rate fell from the average 1996–1999 level of 10% to 5% in 2000–2006 (see European Commission, 2010for the statistics on implicit tax rates). One earlier tax reform that was quite similar was a reform introduced in Chile in1984 that sharply reduced the taxation of retained earnings, from 46% down to 10% for public companies (Hsieh and Par-ker, 2007). Following the reasoning of Hsieh and Parker (2007), we would expect the reform to have an especially strongeffect in an economy characterised by financially constrained firms whose investments are heavily dependent on the avail-ability of internal funding from cash flows, as was also the case in Estonia in the period under consideration (Mickiewiczet al., 2004; Masso, 2002).

Although the non-taxation of undistributed profits has so far only been introduced in Macedonia in 2008 in addition toEstonia, it has still attracted a lot of attention from researchers. Funke (2002) and Funke and Strulik (2006) found, using atheoretical dynamic general equilibrium model of economic growth, that although the tax reform of 2000 led to higher cap-ital accumulation and per capita GDP, welfare may have decreased due to the short-term reduction in private consumption.Masso and Meriküll (2011) using a similar approach in a discrete-time setting also found that the reform increased equityfinance and reduced debt finance. The theoretical modelling by Azacis and Gillman (2010) showed that the welfare of societywould have increased more if the taxation of capital and labour had been more balanced. A report from the OECD (2009)argued that the Estonian system of tax exemption for retained earnings may reduce the economy’s ability to restructure,as it may motivate firms to keep their funds in current business instead of investing them in new growing areas. However,surveys of Estonian firms do not indicate that this is a noticeable problem (Tartu Ülikool and Praxis, 2010). Other interviewswith financial managers (Sander, 2003; Sander and Trumm, 2006) have indicated that corporate income tax plays only amodest role in the investment decisions of Estonia’s companies, but that it is more important for profit distributiondecisions.

Based on an empirical analysis of Estonian firm-level data without a comparison with a control group, Hazak (2009) foundthat the tax reform increased the share of retained earnings in total assets by 4.7% points, decreased the share of liabilities intotal assets by 12.2% points and increased liquid assets as cash and equivalents to total assets by 5.6% points. Our resultsindicate that Hazak’s (2009) before and after analysis without a control group significantly overestimates the reform effecton firm liquidity and liabilities, and significantly under-estimates the effect on retained earnings.

We estimate the outcomes of the Estonian tax reform with a difference in differences (DIDs) analysis using the firms ofthe two other Baltic countries, Latvia and Lithuania, as the control group for Estonian firms. These three countries have verysimilar historical backgrounds as they all regained independence in 1991 after the dissolution of the Soviet Union and joinedthe EU in 2004; they have similar institutions, and highly correlated business cycles. This means that firms from Latvia andLithuania could constitute an appropriate control group for Estonia’s firms. In addition to the DID analysis, we also use thepropensity score matching to test the robustness of the results.

We construct our firm-level panel database from three sources: the international firm-level database Amadeus, the Esto-nian Commercial Register and the Latvian Commercial Register. We calculate the indicators from the firms’ annual reports.While the literature has concentrated more on the effect of corporate income taxation on investment and productivity, wealso look into changes in capital structure and liquidity. The effect on capital structure or dividend payments could be ex-pected to appear relatively quickly after the reform, while the effects on investments and productivity may take longer tomaterialise. Hence, the analysis is based on two estimation periods, 1996–2003 and 1996–2008.

The paper is organised as follows: the next section presents the institutional developments in the treatment and controlgroup during the analysis period; Section 3 presents the data from all three countries; Section 4 describes the methodology;Section 5, the results of the DID analysis, robustness test of the matching analysis and placebo treatment; and the last sectionpresents a summary.

1 Hereafter we refer to income as gross income. The gross income is income before taxes. The income after taxes consists of two major parts: first, the partdistributed to owners as dividends; and second, the part reinvested to the company or retained to the company as cash and equivalents. The second part ishereafter referred to as retained earnings.

2 To be more precise, until 2000 firms paying dividends paid income tax of 26/74 of the dividends paid out, but this tax could be deducted from the taxes onprofits (a method similar to the imputation system). Since 2000, firms need to pay taxes only on dividends, expenses not related to commercial activities andhidden profit payouts.

3 The tax rate was reduced later on: to 24% in 2005, to 23% in 2006, to 22% in 2007 and to 21% in 2008. Section 2 discusses these changes in more detail.

Liabilities (i.e. debt and other liabilities except owner’s equity) to assets

0.3

0.35

0.4

0.45

0.5

0.55

0.6

0.65

0.7

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007

Cash to assets

0.02

0.04

0.06

0.08

0.1

0.12

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007

Adjusted top statutory tax rate on corporate income in %

0

10

20

30

40

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008

Implicit tax rates in % - Corporate income

0

10

20

30

40

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008

Investments per worker in real terms (th. EEK)

10

20

30

40

50

60

70

80

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007

Estonia Latvia Lithuania

Turnover per worker in real terms (th. EEK)

200

300

400

500

600

700

800

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007

Fig. 1. Tax rates, capital structure, liquid assets, investments and productivity in Estonia, Latvia and Lithuania for 1996–2007. Source: Tax rates fromEuropean Commission, 2010. Liabilities to assets and cash to assets originate from the websites on business statistics from the statistical offices in eachcountry. Investments and productivity are from Eurostat’s structural business statistics, deflated using the GDP deflator from Eurostat national accounts.Note: 1EEK=0.064EUR.

1094 J. Masso et al. / Journal of Comparative Economics 41 (2013) 1092–1105

2. Background of the reform and design of the control group

The simultaneous tax change for all Estonian firms makes it difficult to identify the effect of the reform on firm perfor-mance, as we have no control group within the country to control for developments that would have taken place anywaywithout the reform. We therefore use a control group of Latvian and Lithuanian firms for this purpose. One possible criticismof our approach is that in addition to the corporate tax reform there might have been other developments in Estonia andother Baltic countries around the year 2000 that might have had the same effects on the firms’ financial indicators. This sec-tion provides a general discussion of developments in these countries and also addresses these potential concerns.

Latvia and Lithuania are in many respects similar to Estonia, as they have had similar historical backgrounds, similar insti-tutional frameworks, monetary systems based on a currency board, similar trade patterns, highly correlated business cycles,close geographical proximity and economies of a similar size. All three were significantly affected by the Russian economiccrisis in 1999, had rapid growth in 2001–2007 and were severely hit by the global financial crisis in 2008–2009. The tax sys-tems have also been similar in all three Baltic countries. They were among the first countries to introduce flat taxes on capitaland labour in the mid 1990s and have the lowest tax burden on capital among EU27 countries. At the same time, the taxationof labour and indirect taxes are relatively high (European Commission, 2010).

The Baltic States have been considered rapid reformers, having already introduced most of the institutions of a marketeconomy and carried through most of the privatisation process by the mid 1990s. Most EBRD transition indexes—regardingprivatisation, enterprise restructuring, price liberalisation, banking reform, infrastructure—were rather stable, if not un-changed, in all the Baltic States after the period 1995–1997. Among any institutional changes affecting the financial perfor-mance of firms, the corporate tax reform was the single most important event in 2000. The reform was a stand-aloneinitiative aimed at supporting the development of entrepreneurship, increasing investments and promoting job creation(Tartu Ülikool, Praxis 2010).

Between 1996 and 2003, apart from significant changes in corporate income taxation, the tax system was rather stable inEstonia. This included a stable income tax rate, social tax rate and value added tax rate. Estonia’s rather simple tax system(mostly due to the reform in 1994), with a limited number of exceptions, makes it unlikely that changes in some other lawswould affect the analysis of the tax reform in 2000. Probably the largest additional change in Estonia’s tax system was theintroduction of the unemployment insurance system in 2002. A similar unemployment insurance system was introduced inLatvia in 2000 and in Lithuania in 2005. The contributions to unemployment insurance are relatively small, thus these areless likely to induce significant changes in the relative cost of capital and labour, and firm employment decisions. The samecan be said about the introduction of the 3-pillar old-age pension systems in the Baltic countries (e.g. in Estonia in 2002),which did not have any significant effect on the cost of labour.

Fig. 1 presents Estonian, Latvian and Lithuanian tax rates, firms’ aggregate liabilities (i.e. debt and other liabilities exceptowner’s equity) to assets and cash and equivalents to assets ratios from the national statistics, and investment and produc-tivity from Eurostat. The statutory tax rates have been reduced in all three countries, which is in line with international tax-ation trends. The implicit tax rates are fairly U-shaped, which somewhat captures the effect of a business cycle but also theeffect of changes in the tax base. For example, it is clearly visible that the 2000 tax base reform in Estonia led to a halving ofimplicit tax rates, while implicit tax rates started to increase after 2001, and have almost reached their pre-reform level in

J. Masso et al. / Journal of Comparative Economics 41 (2013) 1092–1105 1095

spite of reductions in the statutory tax rate. Compared to the adjustments in the flat corporate income tax rate, the tax basehas been relatively stable since 2000 in Estonia and Latvia (European Commission, 2007, 2010). Lithuania’s corporate incometax rate has been adjusted the most during the period of analysis, and they have also significantly broadened the base forcapital taxes (European Commission, 2007, 2010). The dynamics of the Lithuanian implicit tax rate on corporate incomeis the closest one to Estonia – while Lithuania has lowered the tax rate and widened the tax base, Estonia has kept thetax rate relatively high and narrowed the base.

Aggregated figures reveal that the introduction of Estonia’s corporate tax reform in 2000 coincided with an increase in thecash to assets ratio and a drop in the liabilities to assets ratio. For comparison, Hsieh and Parker (2006) found similarly thatthe liabilities to assets ratio fell after the Chilean tax reform in 1984–1986. Investments and labour productivity wereincreasing in Estonia throughout the period and it is difficult to detect any difference in the trends before and after thetax reform. Regarding the similarity of trends in these indicators across countries and before the tax reform, all the indicatorsexcept the liabilities to assets ratio, showed similar trends. Considering that we use a DID estimation in our paper, the sim-ilarity of the trends before the reform is an important feature that supports the choice of our methodological approach.

It can be concluded that Latvian and Lithuanian firms, on average, can be considered to be a reasonably close match forEstonian firms, potentially providing valuable evidence on how Estonia’s firms could have performed if they had not faced adistributed profit taxation system.

3. Data

We construct a firm-level panel dataset of the Baltic States using information from three data sources. Data for the treat-ment group of Estonian firms are extracted from the Estonian Commercial Register for the period until 2000 and from theAmadeus data from Bureau van Dijk for periods after that. The data on the control group of Latvian and Lithuanian firmsoriginate from the Amadeus database of Bureau van Dijk (the 2000 and 2009 updates of Amadeus), and are complementedby data from the Latvian Commercial Register. The Amadeus data were complemented by information from the CommercialRegister of Latvia, as the Amadeus data update for 2000 contains many missing observations for the variables needed for ouranalysis and covers a limited number of companies from the Baltic countries.

The dataset covers a random sample of manufacturing companies and business services companies. The construction sec-tor, public services, agriculture, mining and the energy sector are excluded. The main period of analysis is 1996–2003. The1996–2003 sample covers more than 40% of Estonian firms, almost 10% of Latvian firms and 3% of Lithuanian firms in theindustries studied (see Table 1 in Masso, Meriküll and Vahter, 2011 for more details on the sample).

The descriptive statistics of our constructed dataset are presented in Table 1. All the monetary variables are shown inthousands of Estonian kroons (EEK, 1EEK=0.064EUR) and deflated using the GDP deflator at the one-digit NACE level. Allthree countries maintained a fixed exchange rate over the period studied.4 We remove outliers with extreme values forthe key variables from the dataset by retracting observations with a liabilities to assets or cash to assets below 0 or over 1,and retained earnings to total assets observations below �1 or over 1; we further exclude observations in the lowest 5% andthe highest 5% of values for the investment rate and the lowest 1% and the highest 1% of values for the productivity growth rate.The investments variable was cleaned of outliers so extensively because the investment data were rather noisy. This exclusionof the lowest or highest values for investment rate is not sizable in terms of the total sum of investments excluded. For example,exclusion of the highest 5% of values for the investment rate excludes only 1.37% of total non-negative investments in the sam-ple, indicating that rather smaller firms are removed. Exclusion of the lowest 5% of values for the investment rate excludes two-thirds of total negative investments in the sample. The results from other outlier detecting schemes are also investigated anddiscussed in the results section.

4. Methodology

This paper uses a difference in differences (DID) analysis and matching methods to estimate the relationship of the taxreform of 2000 in Estonia with post-reform firm-level financial indicators and performance. The DID approach has been usedas a baseline approach and the matching method to provide a robustness test. The DID approach has been used to study theeffects of corporate income tax reforms by, for example, Kari et al. (2008, 2009) and Sivadasan and Slemrod (2008).5

As discussed in Section 2, we construct our analysis on the basis that the treatment group consists of Estonian firms andthe control group of Latvian and Lithuanian firms. We estimate the following standard version of the DID regression modelwith firm fixed effects:

4 Theconvert

5 See

Yit ¼ ai þ cðCountryEE � Post2000Þ þXk

i¼1

bkXk þ st � Sectorj þ st � Sizes þ eit ; ð1Þ

re have been small adjustments in the bilateral exchange rates of the three Baltic countries. The financial data of Latvian and Lithuanian firms areed into thousands of EEK using the yearly average exchange rate.e.g. Blundell and Costa Dias (2000) for a discussion of the methodologies for evaluating the impact of a policy reform.

Table 1Definition of the variables and descriptive statistics, annual data, 1996–2008. Source: Amadeus, Estonian Commercial Register and Latvian Commercial Registerfirm-level data.

Variable name Definition Average Std. dev. Min. Max. No. of obs.

Liabilities to total assets Liabilities to total assets (assets = liabilities + equity) 0.49 0.31 0 1 321 160Cash stock Cash and equivalents to total assets 0.22 0.28 0 1 377 093Retained earnings Ratio of retained earnings and reserves to total assets 0.24 0.35 –1 1 369 306Investment rate1 Ratio of investments to the stock of tangible fixed assets2 0.195 0.39 –0.75 1 98 216Productivity growth3 Growth rate of TFP4 0.183 0.722 –0.811 6.658 181 561Firm size Logarithm of number of employees 1.59 1.29 0 9.916 555 094

1 Indicates that observations in the lowest 5% and the highest 5% of values over 1996–2003 or 1996–2008 are excluded from the analysis.2 Investment calculation is based on tangible fixed assets and does not take into account depreciation, as the data on the depreciation of control group

firms was not available.3 Indicates that observations in the lowest 1% and the highest 1% of values over 1996–2003 or 1996–2008 are excluded from the analysis.4 Total Factor Productivity (TFP) is based on a separate OLS estimation of Cobb-Douglas production functions for 2-digit level NACE Rev.1.1 industries. We

employ operational revenue instead of value added in these estimations as value added can be calculated only for a small number of firms in the controlgroup.

1096 J. Masso et al. / Journal of Comparative Economics 41 (2013) 1092–1105

where t indicates time, t = 1996, . . . , 2008 (the main period of analysis is 1996–2003, the alternative period of analysis is1996–2008); i indicates firms and Yit is the dependent variable (either liabilities to assets ratio, cash to assets ratio, retainedearnings to assets ratio, investment rate or productivity growth); and ai captures the firm-specific fixed effect. CountryEE is adummy variable with a value 1 for Estonian and 0 for Latvian and Lithuanian companies. Post2000 is a dummy variable witha value 0 before the reform in 2000 and 1 after the year of the tax reform in 2000. The coefficient c captures the treatmenteffect and provides an estimate for the effect of the reform (given the assumptions of the DID approach). eit is an error termwith conventional properties. The firm-level fixed effects take into account all time-invariant state, sector and firm-level fac-tors (such as managerial quality) that are constant during the period studied.

We have limited the list of other control variables to ensure a sufficiently large number of observations in the regressions.The sample size was a concern especially in the pre-reform period. Xk denotes other control variables (number of employees,statutory tax rates,6 interaction of Post2000 with pre-reform firm average TFP and liabilities to assets ratio7) and st denotes thetime dummies in Eq. (1). We allow each industry j, j = 1–39, and each firm size group s, s = 1–4, to have individual time trends.7

This is motivated first by possibly different business cycle and technological change across industries and size groups. Second, asbrought out in the data section, the sample coverage differs over time and over countries, which means that industry or sizespecific shocks could mimic the treatment effect unless we control for these group-specific trends. The sample covers relativelymore small firms from Estonia and the coverage per country varies over time. (For example, smaller firms enter the Lithuaniansample more at the end of the studied period. The four size dummies used in these interaction terms show whether the firmfalls into a particular employment quartile within a 2-digit NACE Rev. 1.1 industry.

Interaction terms of Post2000 with pre-reform firm average TFP and liabilities to assets ratio were included to control forfirm-specific differences before the reform, which could be related to post-reform trends in dependent variables.

5. Results

5.1. Tax reform, capital structure and liquidity

This section provides the results of analysis over the main sample period studied, 1996–2003. We have also investigatedthe outcome of the reform over the longer period 1996–2008. The main results were, however, similar for both periods. Seethe estimations for the longer sample period in the working paper version of this article (in Masso et al., 2011).

The following Tables 2–4 report the results of the regression analysis for the relationship between tax reform and capitalstructure, liquidity (cash and equivalent to total assets) and retained earnings of firms. The results from the regression anal-ysis confirm the general trends shown in Section 2. The DID regressions have been estimated separately for the manufactur-ing and services sectors, and for small and large firms.

The relationship between the tax reform and the ratio of liabilities to total assets of the firm is statistically and econom-ically significant during the period 2000–2003. The share of liabilities in total assets fell in Estonia after the income taxreform by an average of 5.7% points. These estimated effects were 1% point larger among manufacturing than services com-panies, and almost twice as high as the average among small companies (see Table 2). The fall in the use of debt financing in

6 We have also estimated specifications controlling for implicit tax rates, see the results in Table 15 of Masso et al., 2011. Nevertheless, the reform left thestatutory tax rate on distributed profits unchanged even though the zero tax rate on retained earnings substantially lowered the overall tax burden ofcompanies in Estonia. Not controlling for the overall reduction in the tax load may lead us to estimate the effect of a tax load reduction rather than the changein taxation scheme from gross profit taxation to distributed profits taxation. However, the estimations controlling for implicit tax rates showed qualitatively thesame results and we can conclude that the impact of the reform presented below was not a simple effect of the reduction in the tax load but the shift in thetaxation scheme.

7 We thank an anonymous referee for drawing our attention to these points.

J. Masso et al. / Journal of Comparative Economics 41 (2013) 1092–1105 1097

Estonia is even more remarkable given that firms in all the Baltic States had improved access to credit under conditions ofhigh growth in aggregate credit supply, falling interest rates and heavy competition between commercial banks (see Brixiovaet al., 2009; Sirtaine and Skamnelos, 2007).

Liquidity, or liquid assets as cash and equivalent to total assets, has increased by an average of 2% points during the post-reform period (compared to Latvian and Lithuanian firms), and this result has again been two times larger for small firmswith less than 10 employees (Table 3). In the literature, the level of liquidity has been used to test for the existence of financ-ing or liquidity constraints, as a high correlation between investments and liquidity might indicate that firms face difficultiesin raising funding from outside the firm by issuing debt or equity. The improved access to external finance in the periodunder consideration supports the hypothesis that the increased liquidity is due to income tax reform. On the other hand,higher liquidity could also be associated with lower operating efficiency among firms; for example, because a firm is not able

Table 2The relationship between income tax reform and the post-reform ratio of liabilities to total assets. Source: firm-level data from the Amadeus database, EstonianCommercial Register and Latvian Commercial Register.

Wholesample

Manufacturing Businessservices

Number of employees

1–9 10–49 50–249 6250

CountryEE � Post2000 �0.057*** �0.051*** �0.060*** �0.111*** �0.053*** �0.061*** 0.006(0.005) (0.01) (0.006) (0.01) (0.007) (0.011) (0.021)

PreReformTFP � Post2000 �0.0001 0.001 �0.0001 �0.0001 0.001 �0.001 �0.017(0.000) (0.001) (0.000) (0.000) (0.001) (0.004) (0.011)

PreReform liabilities to totalassets � Post2000

�0.416*** �0.421*** �0.415*** �0.445*** �0.355*** �0.309*** �0.300***

(0.007) (0.015) (0.008) (0.009) (0.012) (0.022) (0.047)Log(tax rate) �0.069*** �0.042* �0.081*** �0.100** �0.070*** �0.026 �0.043

(0.012) (0.024) (0.013) (0.041) (0.018) (0.021) (0.039)Ln(employment) 0.031*** 0.029*** 0.032*** 0.033*** 0.018** 0.044*** 0.028

(0.004) (0.007) (0.004) (0.005) (0.008) (0.012) (0.019)Industry � year dummies Yes Yes Yes Yes Yes Yes YesSizea � year dummies Yes Yes Yes No No No NoNumber of obs. 48759 11972 36787 25369 15092 6310 1988Number of firms 10780 2514 8266 6754 4474 1630 434Within-group R2 0.209 0.214 0.208 0.229 0.190 0.169 0.278

Note: Data from Estonia, Latvia and Lithuania, 1996–2003. Fixed effects model, standard errors corrected for heteroscedasticity are reported in theparenthesis.a The four size dummies are based on within 2-digit NACE industry employment quartiles.* Statistical significance at the 1% level.** Statistical significance at the 5% level.*** Statistical significance at the 10% level.

Table 3The relationship between income tax reform and post-reform liquidity (the ratio of cash and equivalents to total assets). Source: authors’ calculations using datafrom the Amadeus database, Estonian Commercial Register and Latvian Commercial Register.

Whole sample Manufacturing Business services Number of employees

1–9 10–49 50–249 6250

CountryEE � Post2000 0.020*** 0.012*** 0.023*** 0.041*** 0.018*** 0.011** 0.002(0.003) (0.005) (0.003) (0.008) (0.004) (0.005) (0.009)

PreReformTFP � Post2000 �0.0001** �0.001 �0.0001** �0.0001** �0.004*** �0.002 �0.009(0.000) �0.001 (0.000) (0.000) (0.001) (0.002) (0.006)

PreReform liabilities to total assets � Post2000 0.001 0.014 �0.002 �0.0001 0.006 0.0001 0.005(0.006) (0.01) (0.007) (0.009) (0.009) (0.01) (0.018)

Log(tax rate) 0.002 �0.005 0.004 0.038 0.003 �0.004 0.002(0.006) (0.009) (0.007) (0.026) (0.009) (0.009) (0.016)

Ln(employment) �0.016*** �0.015*** �0.017*** �0.026*** �0.014*** �0.008 �0.009(0.003) (0.005) (0.003) (0.004) (0.005) (0.005) (0.007)

Industry � year dummies Yes Yes Yes Yes Yes Yes YesSizea � year dummies Yes Yes Yes No No No NoNumber of obs. 48759 11972 36787 25369 15092 6310 1988Number of firms 10780 2514 8266 6754 4474 1630 434Within-group R2 0.031 0.042 0.029 0.037 0.042 0.059 0.208

Note: Data from Estonia, Latvia and Lithuania, 1996–2003. Fixed effects model, standard errors corrected for heteroscedasticity are reported in theparenthesis.a The four size dummies are based on within 2-digit NACE industry employment quartiles.⁄Statistical significance at the 1% level.** Statistical significance at the 5% level.*** Statistical significance at the 10% level.

Table 4The relationship between income tax reform and post-reform retained earnings (the ratio of retained earnings and reserves to total assets). Source: firm-leveldata from the Amadeus database, Estonian Commercial Register and Latvian Commercial Register.

Whole sample Manufacturing Business services Number of employees

1–9 10–49 50–249 6250

CountryEE � Post2000 0.081*** 0.079*** 0.082*** 0.121*** 0.082*** 0.079*** �0.014(0.005) (0.011) (0.006) (0.012) (0.008) (0.011) (0.026)

PreReformTFP � Post2000 �0.0001 �0.003** �0.0001 �0.0001 �0.002 0.0001 0.021**

(0.000) (0.001) (0.000) (0.000) (0.002) (0.004) (0.011)PreReform liabilities to total assets � Post2000 0.229*** 0.272*** 0.218*** 0.221*** 0.251*** 0.175*** 0.142**

(0.008) (0.017) (0.01) (0.012) (0.015) (0.027) (0.056)Log(tax rate) 0.094*** 0.060** 0.110*** 0.159*** 0.104*** 0.059** 0.085**

(0.013) (0.025) (0.015) (0.04) (0.019) (0.024) (0.042)Ln(employment) 0.008* 0.012 0.007 0.027*** 0.014 �0.001 0.042

(0.004) (0.008) (0.005) (0.005) (0.009) (0.015) (0.027)Industry � year dummies Yes Yes Yes Yes Yes Yes YesSizea � year dummies Yes Yes Yes No No No NoNumber of obs. 48759 11972 36787 25369 15092 6310 1988Number of firms 10780 2514 8266 6754 4474 1630 434Within-group R2 0.206 0.226 0.201 0.228 0.191 0.140 0.272

Note: Data from Estonia, Latvia and Lithuania, 1996–2003. Fixed effects model, standard errors corrected for heteroscedasticity are reported in theparenthesis.a The four size dummies are based on within 2-digit NACE industry employment quartiles.* Statistical significance at the 1% level.** Statistical significance at the 5% level.*** Statistical significance at the 10% level.

1098 J. Masso et al. / Journal of Comparative Economics 41 (2013) 1092–1105

to profit from its investment opportunities. This could also be one argument in favour of the proposition raised by the OECD(2009) that the income tax reform of 2000 has lowered reallocation and forced capital to stay in its current area of activity.

The ratio of retained earnings and reserves to total assets has increased by 8.1% points after the reform and again thisresult is less evident among larger companies (Table 4). In particular, no major effects of the reform are noticeable in termsof liquidity and retained earnings among firms with more than 250 employees. The stronger effect on small firms’ liabilitiesto assets and cash to assets ratios could be expected, given the evidence from previous studies that small firms in Estoniawere especially subject to financing constraints (e.g., Mickiewicz et al., 2004), and so the extra cash-flow due to the taxchange should have affected their behaviour more.

The higher expected effect of the reform on small firms follows also from the differences in observed dividend distribu-tion policies of small versus large firms. The report by the University of Tartu about the tax reform (Tartu Ülikool, Praxis2010) presented the results of a web-based survey of 471 firms on the impact of corporate tax reform on firm performance.It included a question on whether the tax reform had any impact on the dividend pay-out decisions of the firm. In general,the percentage of firms answering positively to that question was around 30–40%, depending on the particular period. Theeffect was regarded as less important among larger firms (esp. those with more than 50 employees). Our firm-level financialdata includes information on dividend pay-outs only since 2008. For earlier periods it is possible to derive some impreciseestimate by comparing balance sheets year-by-year. In particular, we calculated the approximate sums of dividends as netincome minus change in retained earnings minus change in reserves. The problem of this approach is that it ignores the issu-ing of new shares based on firm’s retained earnings. We do not observe this information. Our calculations showed that whencomparing average pay-out ratios (fraction of profits paid out in the form of dividends) in 1998–1999 and 2000–2003, theaverage pay-out ratio decreased among smaller firms with up to 50 employees by about 1.5% points, while among largerfirms the pay-out ratio actually increased. Again, this may suggest why the effect of the reform might have been largerfor small firms.

On the other hand, there is only limited evidence about the different expectations about the effects of reforms on smalland large firms at or before the time of implementation. The interviews with policy-makers in 2010 revealed that while theexpectations of the impact of the reform did not differ across small and large firms, one intended consequence was that rap-idly expanding enterprises (more likely to belong to the group of smaller firms) obtained better access to debt financing (Tar-tu Ülikool, Praxis 2010). Still, during the time the reform was implemented there were no ex-ante studies on its economiceffects at firm level, and only the impact on tax revenues was analysed. The explanatory letter to the draft of the proposedlaw also did not mention small enterprises Seletuskiri, 1999.

The earlier study by Hazak (2009) on Estonia’s corporate income tax reform of 2000 defined the effect of the reform sim-ply as a shift in the average values of the studied financial indicators in Estonia since 2000 relative to their earlier values.Therefore, the approach taken in this study of comparing Estonian firms to Latvian and Lithuanian firms is more appropriateas an attempt to identify the consequences of the reform.

As predicted, our estimated magnitude of the outcome of the reform is significantly different from that of Hazak (2009),who did not use a control group in his analysis. Our estimated ‘effect’ on the liabilities to assets ratio is �5.7% points, whileHazak (2009) found the ‘effect’ of the reform to be �12.2% points. Our estimated ‘effect’ on liquidity is 2% points, while

J. Masso et al. / Journal of Comparative Economics 41 (2013) 1092–1105 1099

according to Hazak (2009) it was 5.6% points. Our estimated ‘effect’ on retained earnings to assets is 8.1% points, whileaccording to Hazak (2009) it was 4.7% points. Hence, our estimates of this effect are about half the magnitude of the effectfound by Hazak (2009) in the case of liabilities to assets ratio and liquidity indicators, and almost twice as large in the case ofretained earnings. In conclusion, the use of a control group and the DID estimation approach is important for the analysis ofthe outcomes of the reform. Simple before-after comparisons lead to significantly misleading estimates.

Our final point in this section concerns the issue of whether the lower debt financing and increased liquidity induced bythe income tax reform may have also helped Estonian companies to cope relatively better with the recent economic crisis of2008–2010. This idea is suggested by the fact that despite the equally dramatic downturn in all the Baltic economies in 2009,in which GDP dropped in real terms by 14% in Estonia, 15% in Lithuania and 18% in Latvia, the non-performing loans becamemuch less of a problem in Estonia than in the other Baltic countries. Motivated by that, we estimated the probit models offirm survival before and during the crisis; for example, the probability that a company operating in 2008 remained active in2009. We found the higher liquidity to be positively and significantly correlated with firm survival during but not before thecrises. However, given the data limitations, the significant gap in time and the other macroeconomic changes in Estonia andother countries, it would be rather difficult, if not impossible, to establish a causal link between the tax reform and firm sur-vival during the crisis.

5.2. Tax reform, investments and productivity

There are some difficulties that complicate the analysis of the effect of tax reform on investment and productivity. Invest-ments tend to be volatile across firms and within firms. Many authors do not find any statistically significant effects of cor-porate income tax reforms on investments (Kari et al., 2009). Furthermore, the quality of investment data is often low,including large variations in the variable and many outliers in firm-level databases. As investments vary widely, it is alsodifficult to implement an outlier elimination scheme based on the amplitude of the variation in investments. We controlfor outliers by excluding the highest 5% and the lowest 5% of the values of the investment rate variable.8 It should also beborne in mind that our investment variable does not include information on depreciation and is just a year-to-year changein tangible fixed assets. These limitations are made due to the extremely low data coverage for depreciation and intangible fixedassets in the Amadeus data before 2000. However, the latter should not cause large problems in estimating the effects of thereform as intangible fixed assets make up only 2.5% of total fixed assets in our data.

Table 5 presents these results on the relationship of the reform and the investment rate of firms. The results indicate astatistically significant relationship between the reform and the post-2000 investment rate, and this ‘effect’ is stronger forservices and for smaller firms. The economic size of the estimated ‘effect’ of the reform on investment is quite strong.The tax reform appears to be associated with an increase in the investment rate by 17.3% points over the 4 years sincethe reform. Considering that the average investment rate for the sample is 19.5 (see Table 1), this corresponds to an increasefrom the 58th percentile to the 69th percentile in the distribution of firms according to the investment rates. We find theassociation between the reform and post-reform investment to be similar also over the longer timespan – the investmentrate increased between 2000 and 2008 by 17.2% points (long-run results are not reported here in detail to save space andare available from the authors upon request). Hsieh and Parker (2007) also found quite a strong change in investments thatwas likely to be due to the reduction of Chilean retained earnings tax, with a tax reduction of almost 50% related to a 4.5%increase in the investments to GDP ratio in the first year after the reform.

As a next step, we investigate how the reform is associated with the post-reform total factor productivity (TFP) of Esto-nia’s firms. Due to problems related to the stationarity of the productivity data, we use productivity growth and not levels asthe dependent variable here. We note that one control variable we use here, employment growth, may be endogenous, forexample, because more productive firms could grow faster. To control to some extent for the endogeneity of employmentgrowth we lag this explanatory variable by 1 year, which somewhat reduces the number of observations. We expect thecoefficient for the post-2000 period dummy, in interaction with the treatment group indicator, to have the same sign asthe dependent variable in the case of investments because the effect on productivity should take place through changesin the investment patterns.

Table 6 presents the results of the association between the reform and firm productivity.9 The results are consistent withthe reform having a statistically significant effect on investment rate: the productivity growth rate has increased by 8.3% pointsover the 4 years since the reform. Unlike the relationship with investment (Table 5), the results on productivity imply strongereffects of the reform on productivity in the manufacturing industry. The most affected group again appears to be micro-firmswith less than 10 employees.

Putting these numbers on a relative scale, we observe an economically large effect, from the average productivity growthrate of 0.183 to 0.266, which corresponds to a move from the 67th percentile to the 73rd percentile in the productivity

8 Alternatively, we also performed the estimation excluding the lowest 1% and the highest 1% of the values of the investment rate from the sample, whichgave quantitatively similar results with somewhat larger standard errors. We also experimented with estimations excluding cases where real capital stockincreased or decreased more than two times over a single year, which again gave similar results, but with a somewhat smaller effect from the reform.

9 We find somewhat larger ’effect’ on the growth of labour productivity (turnover per employee), 13.4% points. We also experimented with different outlierexclusion schemes. Excluding these cases, where productivity increased or decreased more than two times over a year, produced largely similar results to theones presented in Table 6. However, the size of the ‘effect’ becomes somewhat smaller.

Table 5The relationship between the income tax reform and the post-reform investment rate. Source: firm-level data from the Amadeus database, Estonian CommercialRegister and Latvian Commercial Register.

Whole sample Manufacturing Business services Number of employees

1–9 10–49 50–249 6250

CountryEE � Post2000 0.173*** 0.171*** 0.195*** 0.618*** 0.175 0.159*** 0.094**

(0.018) (0.025) (0.025) (0.045) (0.155) (0.026) (0.042)PreReformTFP � Post2000 0.0004 �0.005* 0.002 �0.001 0.003 �0.003 �0.005

(0.0015) (0.003) (0.002) (0.003) (0.005) (0.011) (0.033)PreReform liabilities to total assets � Post2000 �0.020 0.001 �0.027 0.001 �0.081*** �0.047 �0.033

(0.016) (0.030) (0.017) (0.019) (0.031) (0.049) (0.107)Log(tax rate) 0.015 �0.163*** 0.114*** 0.253 0.165*** �0.081 �0.159**

(0.031) (0.043) (0.041) (0.172) (0.058) (0.054) (0.070)Employment growth rate 0.155*** 0.156*** 0.154*** 0.115*** 0.165*** 0.207*** 0.182***

(0.010) (0.017) (0.011) (0.014) (0.019) (0.029) (0.044)Industry � year dummies Yes Yes Yes Yes Yes Yes YesSizea � year dummies Yes Yes Yes No No No NoNumber of obs. 46801 11869 34932 23776 15310 5894 1821Number of firms 11630 2681 8949 7238 4544 1672 448

Note: Estonia, Latvia and Lithuania, 1996–2003. Observations with the lowest 5% and the highest 5% of the values for the investment rate for 1996–2003 areexcluded from the sample as outliers. Fixed effects model, standard errors corrected for heteroscedasticity are reported in parentheses.a The four size dummies are based on within 2-digit NACE industry employment quartiles.* Statistical significance at the 1% level.** Statistical significance at the 5% level.*** Statistical significance at the 10% level.

1100 J. Masso et al. / Journal of Comparative Economics 41 (2013) 1092–1105

distribution of firms. Estimations based on the sample up to 2008 shows a somewhat lower coefficient for the interactionterm of Estonia’s dummy using the post-2000 period dummy (0.057) than based on the sample up to 2003.

5.3. Robustness tests

5.3.1. Propensity score matchingA central issue in the analysis of the causal effects of policy changes is how to construct a suitable proxy for the unob-

served counterfactual: how to find a suitable control group for the treatment group affected by the policy change. The controlgroup needs to be as similar as possible to the treatment group in terms of its pre-treatment characteristics and structure(see, for example, Angrist and Pischke, 2009; Caliendo and Kopeinig, 2008).

One way to try to analyse the effects of the tax reform is to use a matching approach to build a control group of firms thatis as similar as possible to the firms in Estonia in their pre-tax reform or pre-2000 characteristics. Assuming that such a con-trol group can be built, and further assuming that the relevant variables of the control and treatment groups of firms wouldhave followed the same trends over time in the absence of the tax reform, the after-tax reform differences between these twogroups could be used to estimate the effects of the tax reform. In the DID analysis, we assume that the overall sample ofLatvia’s or Lithuania’s firms provides a suitable control group. The matching analysis tests the robustness of this approach.Instead of using the whole sample of firms in Latvia and Lithuania as controls, it constructs a control group that is as similarto Estonian firms as possible in terms of the observable financial indicators and productivity during the pre-treatment year.Obviously, matching can rely on the observed pre-reform characteristics10 of firms. If there are large unobserved differences inkey determinants of financial indicators, investment and productivity between Estonia’s and Latvia’s or Lithuania’s firms thenthe estimated effects of tax reform in Estonia can be biased.

We use propensity score matching (Leuven and Sianesi, 2003; Caliendo and Kopeinig, 2008). As predictors of the firms’propensity to be affected by the tax reform we use the following data from 1999: productivity, firm size, firm age, investmentrate, the business sector (at 2-digit NACE Rev.1.1. level) of the firm, ratio of liabilities to total assets and ratio of cash to as-sets. The propensity ‘to be affected by the tax reform’ is estimated using a probit model.11 The technical application of pro-pensity score matching is implemented in Stata using the module psmatch2 (Leuven and Sianesi, 2003). We use both thenearest neighbour and kernel algorithms to implement propensity score matching. The results of the nearest neighbour andkernel matching indicate similar results, so we present here only the findings from the kernel matching.

The difference between the post-2000 indicators of the Estonian and matched Latvian or Lithuanian firms can show us theeffects of the tax reform only if the difference between the Estonian firms and the constructed control group is not significant

10 Or also those post-reform characteristics that are not affected by the reform itself (firm age, sector, etc.).11 Note that the estimation sample of the propensity score includes firms from Estonia, and also from Latvia and Lithuania. The ‘propensity of treatment’ is

calculated for firms in all three countries, both in the affected country (Estonia) and the control group (firms in Latvia, Lithuania). We use propensity scorematching with a probit model of ‘treatment’ to come up with a propensity score of belonging to the affected group, because covariate matching based on allrelevant covariates is limited in its applicability due to the high dimensional vector of controls. Rosenbaum and Rubin (1983) have suggested using balancingscores (including propensity scores) instead in such cases.

Table 6The relationship between the income tax reform and post-reform TFP growth. Source: firm-level data from the Amadeus database, Estonian CommercialRegister and Latvian Commercial Register.

Whole sample Manufacturing Business services Number of employees:

1–9 10–49 50–249 P250

CountryEE � Post2000 0.083** 0.131** 0.066 0.344*** 0.011 0.050 0.221**

(0.036) (0.052) (0.050) (0.010) (0.448) (0.044) (0.096)PreReformTFP � Post2000 0.005 0.011 0.003 �0.006 �0.002 0.020 0.240*

(0.004) (0.007) (0.005) (0.007) (0.009) (0.015) (0.127)PreReform liabilities to total assets � Post2000 0.053* 0.042 0.059 0.056 0.095** �0.054 �0.186

(0.029) (0.049) (0.036) (0.037) (0.047) (0.112) (0.198)Log(tax rate) �0.108* �0.312*** �0.014 0.206 �0.041 �0.052 �0.547***

(0.062) (0.104) (0.076) (0.243) (0.094) (0.093) (0.202)Employment growth rate 0.302*** 0.193*** 0.335*** 0.292*** 0.176*** 0.156*** 0.249*

(0.021) (0.043) (0.024) (0.032) (0.029) (0.047) (0.132)Industry � year dummies Yes Yes Yes Yes Yes Yes YesSizea � year dummies Yes Yes Yes No No No NoNumber of obs. 37134 9371 27763 18758 12227 4670 1479Number of firms 10232 2390 7842 6007 3991 1496 406CountryEE � Post2000 0.133 0.171 0.124 0.045 0.067 0.457 0.615

Note: Estonia, Latvia and Lithuania, 1996–2003. Observations with the lowest 1% and the highest 1% of values for the TFP growth rate for 1996–2003 areexcluded from the sample as outliers. Fixed effects model, standard errors corrected for heteroscedasticity are reported in parentheses.a The four size dummies are based on within 2-digit NACE industry employment quartiles.* Statistical significance at the 1% level.** Statistical significance at the 5% level.*** Statistical significance at the 10% level.

J. Masso et al. / Journal of Comparative Economics 41 (2013) 1092–1105 1101

in the pre-treatment period – 1999. Unfortunately, this is not always the case (see Appendix 2 in Masso et al., 2011). For theliquidity, investment and productivity indicators, this difference is still statistically significant even after firms have beenmatched according to their similarities in pre-treatment variables. The difference is not significant for liabilities to total as-sets ratio and retained earnings. One possible explanation for the differences between the treatment and control groupscould be that the sample of Latvian and Lithuanian firms included fewer small companies than the sample of Estoniancompanies.

The estimates of the matching analysis are given in Table 7, where the coefficient of the effect of the tax reform indicatesthe difference between the treatment group and the constructed control group for each variable in the given year. As thepropensity score matching approach was not able to produce a control group that was very similar to the Estonian firms,the results from the following table, with the exception of the capital structure variables, need to be interpreted with a de-gree of caution, as they may not show just the causal effects of the tax reform, but may also reflect pre-treatment differencesand the effects of other variables.

The sign and significance of the effects or correlations estimated using the matching approach largely confirm the previ-ous findings from the DID analysis. For the cash to assets and capital structure indicators, the magnitude of the estimatedrelationship is also quite similar in size to that found by the DID approach.

A simple conclusion from Table 7 is that the association between tax reform and the firms’ capital structure and liquidityexhibit a crucial dependence on the year. This suggests the significant role of the tax reform in Estonia. The estimated post-reform difference between the affected group and the constructed control group is larger in the services sector and whenmore time has passed since the reform. This difference between Estonia’s firms (‘treatment’ group) and the constructed con-trol group from Latvia and Lithuania seem to take longer to materialise in the manufacturing industry, as in 2000 none of theindicators in Table 7 were significantly different in Estonia from those in the constructed control group. Differences thatcould be interpreted as effects of the reform seem to appear after 2001. The result that the effect of the reform seems tobe possibly lagging further behind in manufacturing could be caused by investment sometimes needs a longer planning per-iod in the manufacturing industry (e.g. due to larger sunk investment costs, the intensity of the activities of tangible assets,etc.) than it does elsewhere.

The difference between the investment rates of the treatment group and control group is statistically significant both be-fore and after 2000. For this reason, it is difficult to make conclusions about the causal effects of tax reform on the investmentratio using this approach. However, there is a post-2000 increase in the gap in the investment ratio between the treated andcontrol groups in manufacturing. Again, this may be due to the positive effect of tax reform. We do not find evidence sug-gesting significant effects of the tax reform on productivity with this approach, as the post-treatment difference betweenEstonian firms and the constructed control group is not statistically significant.

5.3.2. Effects in different yearsThis section seeks to test whether the results of the DID analysis in the above sections were just a coincidence, and whether

we could find similar effects across some placebo treatment groups. We take a similar approach to Angrist and Pischke (2009)

Table 7The results of propensity score matching analysis. The estimated effect of the income tax reform on the ratio of liabilities to total assets, liquidity, retainedearnings, investment rate and productivity growth. Source: firm-level data from the Amadeus database, Estonian Commercial Register and Latvian CommercialRegister.

Sector Variable The treatment effect (the average of the indicator in the treatment group minus theaverage of the indicator in the control group)

2000 2001 2002 2003

Manufacturing Liabilities/Total assets –0.044 –0.073* –0.086** –0.063(0.041) (0.039) (0.039) (0.038)

Cash/Total assets –0.006 0.044*** 0.037*** 0.049***

(0.016) (0.013) (0.014) (0.016)(Retained earnings + Reserves)/Total assets 0.046 0.062 0.122*** 0.133***

(0.049) (0.047) (0.047) (0.049)Investment rate 0.031 0.083** 0.112*** 0.060

(0.025) (0.039) (0.04) (0.039)Productivity (TFP) growth 0.048 0.057 0.091

(0.194) (0.181) (0.194)

Services Liabilities/Total assets –0.102*** –0.10*** –0.133*** –0.169***

(0.029) (0.028) (0.028) (0.027)Cash/Total assets –0.005 0.037*** 0.056*** 0.047***

(0.015) (0.013) (0.014) (0.015)(Retained earnings + Reserves)/Total assets 0.093*** 0.127*** 0.15*** 0.186***

(0.026) (0.027) (0.026) (0.026)Investment rate 0.136*** 0.101*** 0.096 –0.046

(0.032) (0.038) (0.089) (0.103)Productivity (TFP) growth 0.045 0.033 0.042

(0.044) (0.045) (0.042)

Note: Period of study, 1996–2003. Treatment group: Estonian firms. Control group: a similar subset of firms from Latvia and Lithuania. Method: propensityscore matching (kernel matching), the results of the nearest neighbour matching algorithm are similar.Standard errors of estimates are given in parentheses. The treatment effect is calculated as the post-reform difference between the values of the indicatorsof the treated and control groups.* Statistical significance at the 1% levels respectively.** Statistical significance at the 5% levels respectively.*** Statistical significance at the 10% levels respectively.

Table 8Difference in differences treatment effect in different years: effects on capital structure, liquidity, retained earnings, investment rate and productivity growth.Source: firm-level data from the Amadeus database, Estonian Commercial Register and Latvian Commercial Register.

Dependent variable Periods before and after the tax reform in 2000

CountryEE � 1998 CountryEE � 1999 CountryEE � 2000 CountryEE � 2001 CountryEE � 2002 CountryEE � 2003

Liabilities/total assets 0.008 0.020 –0.014 –0.056*** –0.086*** –0.104***

(0.014) (0.015) (0.015) (0.015) (0.016) (0.017)(Cash and bank accounts)/

total assets0.011* 0.005 0.020*** 0.031*** 0.026*** 0.025***

(0.007) (0.007) (0.007) (0.007) (0.008) (0.008)(Retained earning and

reserves)/total assets�0.008 �0.026 0.026 0.077*** 0.117*** 0.145***

(0.016) (0.018) (0.018) (0.018) (0.019) (0.020)Investment rate 0.165*** 0.263*** 0.218*** 0.296*** 0.236***

(0.024) (0.023) (0.024) (0.027) (0.030)Productivity (TFP) growth 0.053 0.194*** 0.075 0.177*** 0.146**

(0.070) (0.062) (0.062) (0.069) (0.074)

Note: Estonia, Latvia and Lithuania, 1996–2003. CountryEE � 1998 denotes interaction of country dummy for Estonia with year dummy for 1998, etc. Thecontrol variables are those from the specification in Eq. (1), only the CountryEEx time dummy coefficients are reported. 1 That the lowest 5% and the highest5% of values are excluded from the sample as outliers. 2 That the lowest 1% and the highest 1% of values are excluded from the sample as outliers. Fixedeffects model, standard errors corrected for heteroscedasticity are reported in parentheses.* Statistical significance at the 1% level.** Statistical significance at the 5% level.*** Statistical significance at the 10% level.

1102 J. Masso et al. / Journal of Comparative Economics 41 (2013) 1092–1105

in terms of DID analysis (Eq. (5.2.7) and Fig. 5.2.4 in Angrist and Pischke, 2009). Instead of the previous CountryEE � Post2000,dummy, we introduce a section of variables resulting from multiplying the year dummies (e.g. for 1999 or 2000) with Esto-nia’s country dummy. Table 8 and Fig. 2 present the results. Only the values of the estimated parameter (and its standard er-ror) for ‘‘CountryEE � Year’’ are reported, with each row of the Table 8 corresponding to one regression model.

Each cell in Table 8 reports how a particular variable for Estonia’s companies (defined at the beginning of each row in thetable) differed in a particular year (given by the column of the table) from Latvia’s and Lithuania’s counterparts, conditional

Fig. 2. Dynamics for key indicators for Estonian companies compared to companies in Latvia and Lithuania, conditional on a full set of controls (see Eq. (1)),1996–2003. Note: Marker on the figure shows the estimated ‘‘CountryEE � Year’’ coefficients reported in Table 8, line through the marker reports thecoefficient’s 95% confidence interval. Source: firm-level data from the Amadeus database, Estonian Commercial Register and Latvian Commercial Register.

J. Masso et al. / Journal of Comparative Economics 41 (2013) 1092–1105 1103

on the controls used in Eq. (1). The reported interaction dummies are mostly statistically insignificant before the reform,indicating that conditional on the controlled variables, there were only small differences between the companies in eachof the Baltic States. The only difference concerns the investment rate that was already higher in Estonia before the actualreform was launched. This could be a result of the introduction of the deduction of investments in material fixed assetsto the regions outside the area of Estonia’s capital city in 1998 and 1999. However, the gap between the investment ratein Estonia and the control group increased even further after the reform. The large values of the dummy indicating the period1 year after the launch of the reform, especially for liabilities to assets ratio and retained earnings to assets ratio, probablycomes from the sluggish reaction of firms to the reform in terms of their timing or optimisation of taxes. As most of the re-ported interaction terms become statistically significant and sizeable after 2000, it appears likely that the reform had quite astrong economic effect.

6. Conclusions

This article studied how the Estonian corporate income tax reform of 2000 is associated with changes in the capital struc-ture, liquidity, investments and productivity of firms. In order to analyse the outcomes of the reform, we used both differencein differences and propensity score matching analysis, in which Latvian and Lithuanian firms were used as the control groupfor the ‘treatment group’ of Estonian companies.

We find that after the corporate income tax reform (a shift to distributed profit taxation instead of gross profit taxation),the share of liabilities in total assets decreased in Estonia, on average, by about 5.7% points. The share of cash and equivalentsin assets, which we use as our indicator of liquidity, increased by 2% points. The share of undistributed profits and reserves intotal capital grew by 8.1% points. We can summarise that the reform appears to have affected smaller companies signifi-cantly more. This result was to be expected given the previous evidence on the importance of liquidity constraints for smallfirms, in particular in Estonia.

The higher accumulation of liquid assets has both positive and negative consequences. Although the improved liquiditycould be beneficial (as was shown by Hazak, 2009), the downside is the tendency for the funds to be kept in low-risk assets

1104 J. Masso et al. / Journal of Comparative Economics 41 (2013) 1092–1105

rather than invested in new machinery and equipment or R&D. On the positive side, the post-reform change in the capitalstructure may have helped Estonian companies to cope better with the global financial crisis that started in 2008. This idea isin accordance with the IMF study (Purfield and Rosenberg, 2010), showing that during the crisis the share of overdue loans inEstonia in 2010 has remained at one-third of the levels in Latvia and Lithuania, at 6% compared to almost 20%.

The reform is associated with an increase in the investment rate, by around 17% points. Total factor productivity grew inEstonia by 8% points more than it did in Latvia and Lithuania during the 4 years following the reform. This correlation withpost-reform investments and productivity was strongest among smaller companies. The finding is in accordance with thepresence of an effect of the reform on capital structure and liquidity and with a more positive effect on small firms. However,the matching analysis refers to a more ambiguous effect of the reform on investment and productivity.

We conclude that the outcome of the reform was manifested more clearly in capital structure and liquidity and lessclearly in investments and productivity. While it is difficult to quantify the reform effect and disentangle it from the effectof other simultaneous changes, especially in terms of investments and productivity, this reform may be assumed to havecontributed widely to the alleviation of credit constraints, modernisation of production technology, growth in productivityand faster catching-up with high-income economies. This all makes the experiment to shift profit taxation from gross profitsto distributed profits worth considering in other countries that are catching-up or countries that plan to reduce the tax bur-den of capital and target a change in firms’ capital structure and investments.

Acknowledgments

The research presented in the paper is largely based on the work undertaken for the project The impact of non-taxation ofundistributed profits on investments and economic development ordered by the Ministry of Finance of the Republic of Estonia.In addition to that project, authors also acknowledge financial support from the Ministry of Education and Research of theRepublic of Estonia target financed project no. SF0180037s08 and Estonian Science Foundation Grant No. 8311. We are grate-ful for comments made by the anonymous referee, Vyacheslav Dombrovsky, Indrek Seppo, Ott Toomet, Andres Võrk and theparticipants of seminars in Tartu, Tallinn and Riga. All errors are wholly the responsibility of the authors only.

References

Angrist, J., Pischke, J., 2009. Mostly harmless econometrics. An empiricist’s companion. Princeton University Press, 373p.Azacis, H., Gillman, M., 2010. Flat tax reform: the baltics 2000–2007. Journal of Macroeconomics 32 (2), 692–708.Blundell, R., Costa Dias, M., 2000. Evaluation methods for non-experimental data. Fiscal Studies 21 (4), 427–468.Brixiova, Z., Vartia, L., Worgotter, A., 2009. Capital Inflows, Household Debt and the Boom-Bust Cycle in Estonia, William Davidson Institute Working Paper

No. 965.Caliendo, M., Kopeinig, S., 2008. Some practical guidance for the implementation of propensity score matching. Journal of Economic Surveys 22, 31–72.De Mooij, R.A., Ederveen, S., 2008. Corporate Tax Elasticities. A Reader’s Guide to Empirical Findings. Oxford University Centre for Business Taxation

Working Paper No. 08/22.Devereux, M.P., Griffith, R., Klemm, A., 2002. Corporate income tax reforms and international tax competition. Econ Policy 17 (35), 449–495.Egger, P., Eggert, W., Keuschnigg, C., Winner, H., 2010. Corporate taxation, debt financing and foreign-plant ownership. Eur Econ Rev 54, 96–107.European Commission, 2007. Taxation trends in the European Union: Data for the EU Member States and Norway. Publications Office of the European

Union, Luxembourg, 2007.European Commission, 2010. Taxation trends in the European Union: Data for the EU Member States, Iceland and Norway. Publications Office of the

European Union, Luxembourg, 2010.Funke, M., 2002. Determining the taxation and investment impacts of Estonia’s 2000 income tax reform. Finnish Economic Papers 15 (2), 102–109.Funke, M., Strulik, H., 2006. Taxation, growth and welfare: dynamic effects of Estonia’s 2000 income tax act. Finnish Economic Papers 19 (1), 25–38.Hazak, A., 2009. Companies’ financial decisions under the distributed profit taxation regime of Estonia. Emerging Markets Finance & Trade 45 (4), 4–12.Hsieh, C.-T., Parker, J.A., 2007. Taxes and growth in a financially underdeveloped country: evidence from the Chilean investment boom. Economia 8 (1), 121–

160.Kari, S., Karikallio, H., Pirttilä, J., 2008. Anticipating tax changes: evidence from the finnish corporate income tax reform of 2005. Fiscal Studies 29 (2), 167–

196.Kari, S., Karikallio, H., Pirttilä, J., 2009. The Impact of Dividend Taxation on Dividends and Investment: New Evidence Based on a Natural Experiment. CESIFO

Working Paper No. 2756.Leuven, E., Sianesi, B., 2003. PSMATCH2: Stata Module to Perform Full Mahalanobis and Propensity Score Matching, Common Support Graphing, and

Covariate Imbalance Testing. Boston College Department of Economics, Statistical Software Components S432001.Masso, J., 2002. Financing constraints as determinants of investment behaviour among Estonian firms. Baltic Journal of Economics 3 (1), 8–30.Masso, J., Meriküll, J., Vahter, P., 2011. Gross Profit Taxation Versus Distributed Profit Taxation and Firm Performance: Effects of Estonia’s Corporate Income

Tax Reform. University of Tartu, Faculty of Economics and Business Administration Working Papers, No 81, 33p.Masso, J., Meriküll, J., 2011. Macroeconomic effects of zero corporate income tax on retained earnings. Baltic Journal of Economics 11 (2), 81–100.Mickiewicz, T., Bishop, K., Varblane, U., 2004. Financial constraints in investment – foreign versus domestic firms. Panel data results from Estonia, 1995–

1999. Acta Oeconomica 54 (4), 425–449.OECD, 2009. Estonia. OECD Economic Surveys. Volume 2009/3. April 2009, 151p.Purfield, C., Rosenberg, B., 2010. Adjustment Under a Currency Peg: Estonia, Latvia and Lithuania during the Global Financial Crisis 2008–2009. IMF Policy

Discussion Paper No. 10/213.Rosenbaum, P., Rubin, D., 1983. The central role of the propensity score in observational studies for causal effects. Biometrika 70, 41–50.Sander, P., 2003. Capital structure choice in Estonian companies: a survey. Management of Organizations: Systematic Research 27, 123–135.Sander, P., Trumm, K., 2006. Dividendipoliitika Eesti suurettevõtetes, Ettevõtte finantsjuhtimine: regionaalaspekt (toim. Prof V. Raudsepp, P. Sander), Tartu

Ülikooli Kirjastus, pp. 142–167.Schwellnus, C., Arnold, J., 2008. Do Corporate Taxes Reduce Productivity and Investment at the Firm Level? Cross-Country Evidence from the Amadeus

Dataset. OECD Economics Department Working Paper No. 641.Seletuskiri tulumaksuseaduse eelnõu juurde 143 SE I (Comments to the Draft of Income Tax Act 143 SE I), 1999.<http://www.riigikogu.ee/

?op=emsplain2&content_type=text/html&page=mgetdoc&itemid=991690021.Sirtaine, S., Skamnelos, I., 2007. Credit Growth in Emerging Europe: A Cause for Stability Concerns? World Bank Policy Research Working Paper No. 4281.

J. Masso et al. / Journal of Comparative Economics 41 (2013) 1092–1105 1105

Sivadasan, J., Slemrod, J., 2008. Tax law changes, income-shifting and measured wage inequality: evidence from India. Journal of Public Economics 92, 2199–2224.

Tartu Ülikool, Praxis, 2010. Ettevõtete jaotamata kasumi mittemaksustamise mõju investeeringutele ja majandusarengule, mimeo. <www.fin.ee/doc.php?106865>.

Vartia, L., 2008. How Do Taxes Affect Investment and Productivity? – Industry Level Analysis of OECD Countries. OECD Economics Department WorkingPaper No. 656.