Long-Term Effects of Temporary Corporate Income Tax. Cuts on Investment and Profits: Evidence from Vietnam

Transcription

1 Long-Term Effects of Temporary Corporate Income Tax Cuts on Investment and Profits: Evidence from Vietnam Anh Pham March 7, 2018 Abstract Using a quasi-experimental design and panel data from 2004 to 2014, I estimate how temporary 30% corporate income tax cuts affected firm investment, employment, profits, and tax revenue during the Great Recession in Vietnam. I find that investment increased about 1.5 times during the policy year and did not decrease after its termination. Employment was not affected. These findings are consistent with creditconstrained models. Reported profits of eligible foreign-owned firms doubled in the policy year and in the two years after the policy ended, but dissipated afterward. Tax payments by foreign-owned firms increased, while those by domestic firms decreased. JEL Classification Codes: H25, O12 Keywords: Corporate Tax; Profit-shifting; Investment; Development Address: Schar School of Public Policy and Government, George Mason University, 3351 Fairfax Drive, Arlington, Virginia 22201, USA. apham16@gmu.edu 1

2 1 Introduction Governments often use a temporary corporate income tax cut 1 and deferral policy 2 to encourage investment, despite a potential decrease in tax revenue. For example, during the Great Recession, the Netherlands temporarily reduced the rate for the first EUR 200,000 of profit by 13% in 2009 and Singapore decreased the rate by 5% in Taiwan offered a three-month deferral of corporate income tax in 2008 for firms that met certain criteria. Vietnam implemented a combination of a corporate income tax cut and tax deferral for the last quarter of 2008 and all of Yet, to my knowledge, no formal study has evaluated the effects of a temporary corporate income tax cut, tax deferral, or the combination of the two policies. Instead, existing studies either examine permanent changes in corporate taxes, changes in the personal income tax rate, temporary acceleration in investment depreciation, temporary durable goods subsidies, or dividend tax cuts. This paper examines the long-term effects i.e., up to five year of a temporary corporate income tax reduction and deferral policy on firm investment, employment, reported profits, and tax revenue. Vietnam provides an excellent opportunity for a causal study because of the policy design and data availability. Vietnam first enacted a 30% temporary corporate income tax cut and a nine-month tax deferral for the last quarter of 2008 and the entirety of Only firms with 300 or fewer long-term employees or ten billion VND (equivalent to $500,0000) in assets or less were eligible for the policies. In addition, Vietnam has an annual firm panel data from 2004 to The eligibility requirement and a long panel data allow me to compare firms just below and just above the 300-employee eligibility threshold several years before the policies were implemented and several years after they ended. 4 1 In Vietnam, corporate income tax is a tax on firm profits. Firms might or might not be publicly traded. 2 A tax deferral allows firms to pay taxes later than the due date. For brevity, I sometimes refer to the combination of a corporate income tax cut and deferral as corporate income tax cuts. 3 Deloittee: Tax Responses to the Global Economic Crisis in I define the treatment and control groups in terms of the employment cutoff rather than of the asset cutoff because there are many more observations around the employment cutoff. Additionally, there are too few observations around the cutoff for a regression discontinuity design. 2

3 The identifying assumption is that firms just below and just above the 300-employee eligibility threshold would have trended similarly in the absence of the policy. Several pieces of evidence support this assumption. First, to address concerns about possible manipulation around the thresholds, I use criteria in 2007, the year before the policy was announced, to define control and treatment firms. Second, treatment and control firms were similar in industries, age, and ownership structure. Third, no concurrent policies used the 300-employee threshold. 5 Fourth, dependent variables trended similarly before the policy was implemented. Finally, empirical results are robust when I make the employment windows smaller, so that treatment and control firms had an almost identical number of employees in the pre-policy year. I begin by modeling how a temporary corporate income tax cut and a temporary tax deferral separately affect firm investment with and without credit constraints. Models with credit constraints predict that investment and/or labor would increase in a year with a temporary tax cut and deferral, and would not decrease in the years after the policy. Thus, if investment increases in the policy year and labor does not, models with credit constraints predict that the investment-labor ratio would increase in the policy year and would not decrease after the policy ends. Models without credit constraints predict that if the investment-labor ratio increases in the policy year, it would eventually decrease after the policy ends. As for reported profits, I examine two broad mechanisms: changes in factor inputs, such as capital and labor; and changes in reporting behavior. If reporting behavior changes, profit would change even if labor and capital stayed the same. I then test the predictions of each model using the case of Vietnam s tax cut and deferral policy. I use annual firm survey data in Vietnam from 2004 to All registered firms in 5 From 2010 to 2014, some variations of the policy were reimplemented, but the 300-employee threshold was not predictive of the eligibility for these later policies. Specifically, the majority of eligible and ineligible firms in 2008 and 2009 in my sample were eligible for the policies from 2010 to On the contrary, almost all eligible and ineligible firms in 2008 and 2009 were ineligible for the policies in 2013 and Thus, the differential tax policy between treatment and control firms in my sample practically ended after 2009, meaning the coefficients from 2010 on mainly reflect long-term impacts of the tax cut and deferral policies in 2008 and 2009 (rather than the effects of any policies in later years). Detailed explanations are provided in Section

4 Vietnam were required to answer some version of the survey. Firms around the 300-employee threshold are in approximately the 90 th to 95 th percentile of the firm-size distribution in Vietnam. I examine domestic and foreign-owned firms separately because Vietnamese domestic firms generally do not have subsidiaries outside Vietnam. Thus, foreign-owned firms have more ability than domestic firms to shift profits across countries. There are two main findings on investment and reported profits. I find that investment increased 1.3 to two times during the tax cut year and did not decrease even five years after the policy ended. I find null results for employment. This implies that the investment-labor ratio increased in the policy year and stayed the same after the policy ended, consistent with the credit-constrained model. Results on reported profits vary between foreign-owned and domestic firms. Reported profits of eligible foreign-owned firms doubled in the tax cut year (2009) and almost tripled in the two years after the policy ended (in 2010 and 2011). This increase has dissipated since As for domestic firms, I find no evidence of increases in profits. A tax revenue calculation exercise suggests that a foreign-owned firm that took advantage of the tax cut and deferral policy paid more taxes due to a large increase in reported profits. Domestic firms did not report more profits and, thus, paid less taxes. The empirical findings are robust across different windows of employment sizes, balanced and unbalanced panel data, different measures of investment, and different definitions of treatment and control firms. 6 Additionally, I test how much the profits of foreign-owned firms changed because of capital and labor and how much it changed because of reporting behavior. I compare the coefficients of regressions that control for capital and labor and those of regressions without such controls. The two results are almost identical. Thus, multinational profit-shifting, rather than changes in labor or capital, was most likely the main mechanism. Increases in reported profits of foreign-owned firms in the two years following the tax cut 6 As mentioned in footnote 5, the 300-employee threshold could be somewhat predictive of the policy in 2010 and To address concerns that the policies from 2010 to 2012 may affect the estimates, in Appendix 9.3, I restrict the sample to only firms that were eligible for the policies in these three years. I find that the results with and without such a restriction are similar. 4

5 and deferral suggest that multinationals cannot shift profits instantaneously across countries when tax policies change. For example, some foreign-owned firms might have been uncertain about when the policy was implemented or when it ended. In addition, it could have been because the chance of getting audited was higher for firms with unusually low profits than for those with unusually high profits. This paper complements a large existing literature on the effect of taxes on firm investment. 7 Most of the studies focus mainly on developed countries; evidence on developing countries exists but is more scarce. Studies on developed countries generally find that a decrease in user cost of capital via a decrease in taxes, such as acceleration in investment depreciation 8, increases investment (see Hassett and Glenn Hubbard (2002) for a survey of empirical works). An exception is Yagan (2015), who finds that the Bush dividend tax cut, the largest dividend tax cut in the US, did not increase investment or employment. In developing countries, Cai and Harrison (2017) find that net investment did not change as a result of a VAT reform in China. Bustos et al. (2004) find that higher corporate income taxes did not affect long-term capital stock in Chile, while Cerda and Larrain (2010) find that higher corporate income taxes decreased investment in Chile. Bustos et al. (2004) and Cerda and Larrain (2010) use time variations in corporate income taxes as a source of identification. In other words, these papers do not study long-term effects of a temporary corporate income tax change. Thus, my paper contributes to this literature by examining long-term effects of a temporary corporate income tax change in Vietnam. In addition, I use a quasi-experimental approach that exploits variations in corporate income taxes across firm sizes and over time to address concerns about endogeneity. I find that the temporary tax cut and deferral policy in Vietnam increased investment between 1.3 to two times in the policy year. Investment did not decrease even five years after the policy ended, consistent with the relaxation of credit constraints. 7 Some examples are Cummins et al. (1994), Cummins et al. (1996), survey empirical work by Hassett and Glenn Hubbard (2002), Djankov et al. (2010), Ferede and Dahlby (2012), Yagan (2015), and Zwick and Mahon (2017). 8 Some recent works are House and Shapiro (2008) and Zwick and Mahon (2017). 5

6 This paper also extends to the literature on credit constraints in developing countries, which has focused almost exclusively on households and micro-enterprises. Evidence using micro-level data to examine the effects of credit constraints on relatively large firms in developing countries is rather sparse. The most relevant paper is Banerjee and Duflo (2014), which shows the existence of credit constraints for larger firms in India in priority sectors such as agriculture and agricultural processing, transportation, and small-scale industry. The results in this paper include all sectors and relatively large firms and are consistent with those of Banerjee and Duflo (2014). This paper also relates to the literature on the effect of taxes on reporting behavior and multinational profit-shifting. 9 The results are consistent with the general findings that multinationals report more profits in countries with lower tax rates. By examining a temporary tax cut in a long panel dataset, I am able to study firms reporting behavior several years before the policy was implemented and several years after it ended. My findings suggest that multinationals cannot always shift profits instantaneously across countries when tax policies change, a finding that has not been shown in the literature. The uncertainty about the timing of a policy or the fear of getting audited may lead to a wrong ex-post decision about where firms report profits, even though ex-ante, it could have been a right decision. The paper proceeds as follows. Section 2 introduces the tax policy in Vietnam. Section 3 presents the dataset. Theoretical models are in Section 4. Section 5 presents the identification strategy. I then show empirical results on capital, labor, reported profits, and tax revenue in section 6. Section 7 describes what we can learn from the case of Vietnam. I conclude the paper in Section 8. Formal derivations of investment and tax reporting behavior models are presented in the Appendix. 9 Some studies on tax-reporting behavior in the US are Clotfelter (1983) and Feinstein (1991). Studies in developing countries include Fisman and Wei (2004) in China, Kopczuk et al. (2012) in Poland, and Gorodnichenko et al. (2009) in Russia. Some studies on multinational income shifting are Swenson (2001), Clausing (2003), Huizinga and Laeven (2008), Weichenrieder (2009), Cristea and Nguyen (2016), Davies et al. (2017), and Liu et al. (2017). 6

7 2 The tax-cut policy Vietnam applies a flat corporate income tax that has been decreasing over time. The tax rate was 32% before 2004, 28% from 2004 to 2008, 25% from 2009 to 2013, and has been 22% from 2014 to the present. The majority of firms choose a fiscal year that is the same as the calendar year. Firms effectively pay corporate income tax on an annual basis, which is due three months after the fiscal year ends. 10 For example, annual corporate income tax forms and payments for 2008 were due on March 31, A temporary 30% corporate income tax cut and the nine-month deferral policy were first introduced in December 2008 and applied to the last quarter of 2008 and throughout Eligible firms were those with no more than an average of 300 long-term employees in 2008 or no more than ten billion VND (or $500,000) in initial assets. Long-term employees are those whose contracts last for more than three months. In addition to the 300-employee threshold, the nine-month tax deferral program also applied to firms in some specific sectors, regardless of their size. 11 Approximately 20-25% of firms that were ineligible according to the 300-employee threshold were eligible for the tax deferral under the sector-specific rule. The tax rate of eligible firms was reduced from 28% to 19.6% in the last quarter of 2008 and was further reduced from 25% to 17.5% in Some variations in the combined tax policy for small and medium-sized firms continued in later years, despite the fact that the government planned to discontinue it after The tax deferral was available for three months in 2010 and in 2011 for one year. The 30% tax-cut policy was reintroduced in 2011 and In 2013 and 2014, the corporate income tax rate of small and medium-sized firms was 20%, which represented a 20% and a 9% tax cut, respectively. 10 On paper, firms file and pay corporate income tax quarterly. However, most firms severely underestimate quarterly corporate income taxes and pay the majority of the income tax owed after the fiscal year ends, so that they can hold on to cash longer. In recent years, the government fined firms that paid quarterly income taxes less than 20% of the due amount. 11 These sectors are agriculture, aquaculture, and wood processing, textile, leather, and computer compartments. 7

8 The definition of small and medium-sized firms in these later years changed drastically compared to the rules in The eligibility rule from 2010 to 2012 depended on firm sizes and industries. In general, firms in service sectors were eligible if they had fewer than 100 employees or if their assets were less than 50 billion VND. Firms in the non-service sectors were eligible if they had fewer than 300 employees or had less than 100 billion VND in assets. 12 In 2013 and 2014, firms with less than 20 billion VND in revenue were eligible in most industries. The 300-employee threshold for the policy is not predictive for eligibility of the policy after 2009, as will be explained in detail in Section 3.3 and 3.4. Thus, by comparing firms just above and just below the 300-employee threshold, I am able to examine the longterm impact of the policies on firm assets, employment, and reported profits. The timing of these policies varied: the 2010 policy was introduced in February 2010; the 2011 policy was introduced in August 2011; and the 2012 policy was introduced in late July The policies in 2013 and 2014 were introduced in June Table 1 summarizes the policies in different years. 3 Data 3.1 Dataset This paper uses annual enterprise surveys conducted by the Vietnamese General Statistical Office (GSO) from 2004 to All registered firms in Vietnam are required to answer the survey. The dataset has information about firm balance sheets, income statements, and some basic tax variables, such as corporate income taxes paid and value added tax liabilities. 12 In 2010, the nine-month tax deferral policy also applied to firms in specific sectors regardless of their size. These sectors were agriculture, aquaculture, wood processing, textile, leather, and computer compartments. Subsidiaries whose parent companies were not small or medium-sized businesses were not eligible for the tax cut in 2011 and Also in 2011 and 2012, firms in agriculture, aquaculture, forestry, textiles, leather shoes, electronic compartments, and public construction were always eligible, regardless of how many workers they employed. 13 The survey started in

9 The enterprise survey is rolled out on March 1 every year to ask about the previous year s information. All surveys must be returned to the GSO by July 15. As the survey is administered by the government, it is reasonable to assume that information in the enterprise dataset is relatively close to the actual numbers that firms report on their annual tax returns. 3.2 Variable Definitions and Winsorization My main measure of investment is the annual dollar change in fixed assets, scaled by lagged fixed assets. I call this variable net investment for short. Fixed assets equal the book value of all fixed assets owned by firms, net of accumulated book depreciation. A change in fixed assets equals fixed assets at the end of the year minus fixed assets at the beginning of the year. Thus, this investment measure is the new annual investment in fixed assets less fixed asset retirements and accumulated book depreciation, scaled by lagged fixed assets. 14 This measure is also used in other papers, such as Yagan (2015) and Cai and Harrison (2017). I also create other investment measures from new investments in machinery and equipment, a measure also used in Yagan (2015) and Zwick and Mahon (2017). About 38% of the observations of new investments spent on machinery and equipment are missing. I examine this investment measure as reported in the data, and I also construct another investment measure by imputing missing values of the new investment variable with 0. The general results of these two variables are qualitatively similar to each other and to the results of the regression where I use net investment. Long-term employees are employees with contracts longer than three months who receive social insurance and benefits at the end of the year. Short-term employees have contracts for three months or less. Total employees are the sum of long-term and short-term employees at the end of the year. Employee compensation equals the sum of wages and salaries paid to employees, contributions to social insurance and other compensations. Profits are before-tax profits at the end of the year. 14 In the previous version of the paper, I use the log of the upfront cost of fixed assets. 9

10 Table 1: Vietnamese Tax-cut and Deferral Policies year Policy Sector Eligibility Time Regular Announced Rate long-term employees assets Revenue (billionvnd) (billionvnd ) 2009 & 30% cut & all sectors 300 in 2008 OR initial assets: 10 December % end month deferral month deferral service 100 in 2009 OR 50 in 2009 Feburary % non-service 300 in 2009 OR 100 in % tax cut & service 100 in 2011 OR 50 in 2011 August % 1-year deferral non-service 300 in 2011 OR 100 in % tax cut service 100 in 2011 OR 50 in 2011 July % non-service 300 in 2011 OR 100 in % cut most sectors <20 in 2012 June % % cut most sectors <20 in 2013 June % Note: In 2010, the nine-month tax deferral policy also applied to firms in specific sectors regardless of their size. These sectors were agriculture, aquaculture, wood processing, textile, leather, and computer compartments. Subsidiaries whose parent companies were not small or medium-sized businesses were not eligible for the tax cut in 2011 and Also in 2011 and 2012, firms in agriculture, aquaculture, forestry, textiles, leather shoes, electronic compartments, and public construction were always eligible, regardless of how many workers they employed. The information on this table is collected from ND /ND-CP, Cong Van 8336-BTC-CST. 10

11 A firm s initial assets are its assets when it first registered. The data allow me to identify eligible firms through reported assets and employment. Since I do not observe a firm s assets if it registered before 2000, when the data were first available, the initial asset variable is set equal to the firm s assets in the first year that the firm appeared in the dataset. Thus, the constructed initial assets are accurate only for firms that registered after If a firm was established before 2000, its initial assets are those in I winsorize profits and investment measures at the 99 th and 95 th percentile. 15 Section 6 and Appendix 9.3 show that results are similar to different levels of winsorization. 3.3 Treatment and control groups To define compatible treatment and control groups, I borrow the idea from the regression discontinuity methodology that firms around a threshold are more likely than firms farther away from the threshold to be similar to each other. I use firms around the employment thresholds because there are more observations than around the the initial asset threshold. The tax cut and deferral policy applied to firms with no more than an average of 300 long-term employees in the last quarter of 2008 or no more than $500,000 in initial assets. When the government introduced the policy in December 2008, it used the average number of long-term employees in 2008 to calculate firm employment eligibility. It is possible that firms manipulated their number of employees to stay under 300 longterm employees in 2008 and be eligible for the policy. In addition, the 300 long-term employee threshold might have been an endogenous choice by the government. For instance, if firms just below the threshold in 2008 were affected by the financial crisis the hardest, the government might haven chosen this threshold to alleviate some consequences of the crisis for these firms. To address the possible manipulation of the 300 long-term employee threshold in 2008, I 15 Winsorizing at the ninety-fifth percentile of net investment means that for any observations with values above the ninety-fifth percentile, I assign the ninety-fifth percentile value. For any observations with values below the fifth percentile, I assign the fifth percentile value. 11

12 use the number of long-term employees in 2007, the year before the policy was announced, to define tax cut eligibility. 16 I restrict the sample to firms whose initial assets were more than $500,000 and, further, to firms that had between 250 and 350 long-term employees in The sample consists of firm unbalanced panel data from 2004 to These firms were all established in or before Thus, I do not worry about manipulation of initial assets when the policy was first introduced in December Another reason that I restrict the sample to firms born in or before 2004 is that I want to control firms initial characteristics in Appendix 9.3 shows that the results are similar when I use balanced panel data; when I use the sample that include firms born in or before 2007; or when I use the sample that does not restrict to firms with certain asset levels. In the main analyses, my treatment group consists of firms with long-term employees in 2007 and initial assets greater than $500,000. My control group consists of firms with long-term employees in 2007 and initial assets greater than $500, Table 2 summarizes eligibility defined by law and the control and treatment groups defined in this paper. I use only the employment eligibility threshold for identification. My sample is an unbalanced panel dataset of about 500 firms per year from 2004 to There are two sources of measurement error that bias the estimates downward. First, the initial assets of firms that were registered before 2000 are not precisely measured. Second, the employment level in 2007, the pre-policy year, is not identical to the actual employment level that determined eligibility. Therefore, some firms that were eligible according to the rule might be categorized as ineligible in this paper and vice versa. If the tax cut and deferral policy actually increase profits, the eligibility misclassification deflates the average 16 I do not find evidence of bunching around the eligibility thresholds. The lack of evidence could be because firms did not manipulate or because the survey data are too noisy to detect the kink. 17 Another way to create treatment and control groups is to restrict the sample to firms that had more than 300 long-term employees in 2007 and initial assets between $400,000 and $600,000 (The initial asset threshold was $500,000.) Eligible firms would have had more than 300 employees in 2007 and initial assets between $400,000 and $500,000. Ineligible firms would have had more than 300 employees in 2007 and had initial assets between $500,000 and $600,000. This method gives me, on average, 84 observations each year, while the employment eligibility threshold method gives me approximately 580 observations each year. Therefore, I use the treatment and control groups defined by the employment eligibility threshold method. 12

13 profit of eligible firms and inflates the average profit of ineligible firms. Thus, the difference between the average profit of eligible firms and that of ineligible firms is deflated. Due to measurement error, all regression coefficients in this paper are downwardly biased estimates of the intent-to-treat effects of the policy. In addition, firms with around 300 long-term employees in 2007 were in approximately the 95 th percentile. Firms whose initial assets were greater than $500,000 were in approximately the 90 th percentile in Therefore, this paper applies to relatively large firms. Table 2: Program Eligibility in 2008 and 2009 Definitions used in the paper Eligible(treatment) Ineligible(control) employees in employees in 2007 AND AND initial assets were greater than $500,000 initial assets were greater than $500,000 By Law, but not used in the paper Eligible Ineligible 300 employees in 2008 > 300 employees in 2008 OR AND initial assets were smaller than $500,000 initial assets were greater than $500, How well the 300-employee threshold in 2007 predicts eligibility for the policies in later years. Table 3 shows that in the sample defined in Section 3.3 and Table 2, the eligibility threshold in 2007 predicts eligibility 44% of the time. This figure drops to approximately 10% for eligibility in , and there is almost no predictive power for eligibility in In addition, the constant terms in Table 3 show that around 60% of the control firms in this sample were eligible for the policies in 2010, 2011, and 2012, and about 5% were eligible for the policies in 2013 and So, from 2010 to 2012 the majority of firms in the sample were eligible, while in 2013 and 2014, almost all firms in the sample were ineligible. There are two reasons for interpreting the coefficients from 2010 mainly as long-term 13

14 impacts of the tax cut and deferral policies in 2008 and 2009 (rather than as effects of the policies in later years). First, the 300-employee threshold predicts the policy much better than it predicts the policies in later years. Second, Figure 11a and Figure 15a in Appendix 9.3 show that the results using the sample of only firms that were eligible for the tax policies from 2010 to 2012 were similar to the results in the sample without the restrictions, defined in Table 2. In other words, the results are not driven by the policies in later years. Table 3: How well does the threshold in 2007 predict eligibility for the policy in later years Eligible in Eligible in Eligible in Eligible in Eligible in eligible in (0.0419) (0.0449) (0.0444) (0.0212) (0.0235) Constant (0.0328) (0.0351) (0.0349) (0.0150) (0.0183) N F r Standard errors in parentheses. p < 0.1, p < 0.05, p < Eligible and ineligible firms in this paper are defined in Table Summary Statistics Tables 4 and 5 describe types of ownerships and sectors of firms in my sample in 2006, before the policy was implemented. There are 285 firms in the treatment group and 206 firms in the control group in Table 4 shows that ownership types of eligible firms in 2006 were also similar to those of ineligible firms in the same year. Specifically, in both ineligible and eligible groups, around 25% of firms were foreign-owned, and the rest were domestic. Table 5 shows that, in 2006, firms were mostly in manufacturing (60%), construction (15%), and commerce (8%). Industry types were similar among eligible and ineligible firms. Table 6 and Figure 1 show that the treatment group and the control group were similar in the pre-policy years. As Figure 1 shows, only the log of the number of employees and 14

15 the log of the total labor costs of the control and treatment groups are statistically different from each other prior to the policy. These differences are consistent with the criterion that the groups were based on the employment requirement. Table 4: Percentage of firms in 2006 by ownership types Ownership type All Eligible Ineligible private domestic 32.63% 29.61% central SOE 22.46% 22.82% local SOE 18.95% 20.39% foreign-owned firms 25.96% 27.18% Total 100% 100% Observations Eligible and ineligible firms in this paper are defined in Table 2. Table 5: Percentage of firms in 2006 by industry types Industry type All Eligible Ineligible commerce 8.07% 8.25% communication 0.35% 0.00% construction 16.84% 15.05% electricity, gas, water 3.16% 3.88% entertainment 1.05% 1.46% finance 1.05% 1.46% health and social work 0.70% 0.00% hotels, restaurants 3.16% 2.91% manufacturing 57.19% 57.28% other services 0.35% 0.97% public admin&defense 1.05% 4.37% real estates 1.05% 0.00% sciences 1.75% 0.49% transport&storage 4.21% 3.88% Total 100% 100% Observations Eligible and ineligible firms in this paper are defined in Table 2. 15

16 Table 6: Summary statistics from 2004 to 2007 eligible ineligible Total total profit before tax (1374.1) (1489.4) (1424.3) profit/lagged fixed asset (1.486) (1.149) (1.355) fixed asset (3192.7) (3121.3) (3162.3) annual change in fixed asset (649.8) (686.3) (665.4) net investment = annual change in fixed asset/lagged fixed asset (0.321) (0.346) (0.332) total labor (261.7) (337.4) (297.9) labor cost (656.9) (806.2) (726.1) annual salary per worker (2.329) (2.048) (2.215) investment- imputed 0/lagged fixed asset (0.200) (0.217) (0.208) Observations investment- no imputed 0/lagged capital (0.294) (0.323) (0.307) Observations Firms have between permanent employees in 2007 and initial asset of no more than 10 billion VND. Total profit before tax, fixed assets, net investment, labor cost, annual salary per labor are in 2005 thousand USD. Total profit before tax and profit/lagged capital are winsorized at a 99 th percentile. Fixed assets, annual change in fixed assets, investment measures, labor cost, annual salary per worker are winsorized at a 95 th percentile. 16

17 Figure 1: Annual differences between eligible firms and ineligible firms from Eligible and ineligible firms are defined in Table 2. Figures plot point estimates of the annual differences between eligible and ineligible firms and 90% confidence intervals. Profits are winsorized at the 99 th percentile. Net investment is winsorized at the 95 th percentile. 4 Theory In this section, I present, separately, how a temporary tax cut and a tax deferral would theoretically affect firm profits and capital stock and, thus, investment. I first present the predicted effects on investment and employment with and without credit constraints. When examining capital, I assume that it is easy for tax authorities to audit fixed assets, which I use to calculate investment in the empirical section. Thus, I assume that the reported capital in the survey is similar to the real amount of capital that firms have. I present the predicted effects on reported profits due to changes in reporting behavior 17

18 and due to real changes as a result of changes in factor inputs such as capital and labor. I derive formal derivations for the real change in profits in Appendix The empirical section will test the model predictions to analyze mechanisms that are consistent with the observed changes in capital and profits. 4.1 Temporary corporate income tax cut Investment and Employment This section presents the predicted effects on investment with and without credit constraints in response to a temporary corporate income tax cut. For a model without credit constraints, I examine a basic profit maximization model of a constant return-to-scale firm in a perfectly competitive market, similar to that in Jorgenson and Hall (1967). I introduce a temporary corporate income tax cut to the basic model to study firms investment behavior, while Jorgenson and Hall (1967) examine a permanent tax credit on investment. The model predicts that the long-term capital-labor ratio does not change. Thus, the investment-labor ratio, which equals the annual change in the capitallabor ratio, increases in the tax-cut year only if the first year s marginal benefit of such an investment exceeds its first year s depreciation allowance. If the investment-labor ratio in the policy year changes, investment in post-policy years will adjust so that the long-term capital-labor ratio stays the same. Empirically, this means that if the investment-labor ratio in the policy year increases, it will decrease in post-policy years. The formal derivations are presented in Appendix 9.1. When credit constraints are introduced, firms have extra cash on hand to invest and/or hire more workers. Thus, the long-term capital stock and/or number of employees would increase. Empirically, this means that if there is no adjustment cost, investment and/or the number of employees in the policy year will increase, and these increases would not be reversed after the policy ends. Therefore, if the investment-labor ratio in the policy year increases, it will not decrease after the policy ends. 18

19 4.1.2 Reported Profits Changes in reported profits could be due to changes in reporting behavior or to changes in factor inputs such as an increase in investment, labor, or labor productivity. Reporting behavior Firms could respond to the temporary tax cut by changing their reporting behavior. They could shift profits across years i.e., shift profits from years with high tax rates to years with low tax rates or shift profits across countries i.e., shift profits from countries with high tax rates to countries with low tax rates. If firms shift profits across years, reported profits increase in the tax-cut years. Firm profits also decrease in the year just prior to the tax cut implementation and in the years immediately following its repeal. If firms shift profits across countries, firms that have more subsidiaries in different countries respond more. Foreignowned firms are likely to respond more than domestic firms because Vietnamese domestic firms generally do not have subsidiaries outside Vietnam. Real Profits I assume that changes in real profits come from changes in factor inputs such as capital and labor. When there are no credit constraints, derivations (see Appendix 9.1.2) show that profits in the tax-cut year increase if investment changes and stays the same if investment does not change. Intuitively, this means that firms change their investment only if such investment increases profits. How profits change in the year immediately after the policy s repeal is ambiguous. It depends on how the capital stock in the policy year responds to a tax change, and how the first year s deprecation of new capital compares to the second year s deprecation. When there are credit constraints, the effects on profits in the policy year and after the policy ends are also ambiguous. It depends on whether firms would invest more, and if so, whether the marginal benefit of new capital is greater or smaller than the first year s depreciation. 19

20 4.2 Temporary Corporate Income Tax Deferral Investment and Employment We can think of a tax deferral as an interest-free loan because firms can now pay taxes at a later date. This section shows that an interest-free loan increases investment and/or employment for credit-constrained firms and does not change investment or employment for unconstrained firms. Recall that the tax deferral in Vietnam lasted for nine months in A nine-month interest-free loan is similar to the case of the subsidized loan in Banerjee and Duflo (2014) If firms are credit-constrained and have not reached their full investment potential, the interest-free loan will increase their investment. The same argument applies to employment. If firms are not credit-constrained, they will substitute the interest-free loans for market interest rate loans. The size of the substitution is equal to the maximum of the total amount of lifetime market loans and the amount of the interest-free loan. If the amount of the interest-free loan is smaller than that of the market loan, firms will borrow as much as they can using the interest-free loan. They will then borrow the remainder at the market rate loan. Thus, investment does not go up because the price of the last unit of capital still depends on the unaffected market interest rate. If the size of the interest-free loan is greater than the size of the market loan, investment goes up because the price of the last unit of capital is now lower because of the interest-free loan. In the case of Vietnam, it is implausible that a nine-month interest-free loan equal to one quarter of the profit (because the tax rate is 25%) could substitute for all the money that a firm borrows during its lifetime. Thus, in the absence of credit constraints, we would not expect the long-term capital stock to increase Reported Profits Reporting Behavior In the policy year, firms might want to report higher profits to get a larger interest-free 20

21 loan. As explained above, firms could shift profits across years or across countries. The predictions are similar to those of the tax-cut case. If firms shift profits across years, we observe high profits in the tax-cut year and low profits in the surrounding years. If firms shift profits across countries, foreign-owned firms respond more i.e. report higher profits than domestic firms. Real Profits The response would be similar to that for a tax cut if firms were credit-constrained. The effects on profits are ambiguous, depending on whether firms would invest more and, if so, whether the marginal revenue of new capital is greater or smaller than the first year s depreciation. If firms are not credit-constrained and replace market interest loans with interest-free loans, profits in the tax deferral year increase because firms now pay lower interest rates to banks. However, profits in the non-policy year do not change because firms still pay market interest rates in these years. 5 Empirical Strategy Since I have only approximately 500 firms per year around the eligibility threshold, regression discontinuity design results are noisy. Therefore, I use difference-in-differences with firm fixed effects around the employment eligibility threshold to control for variations across firms. The baseline estimation regression equation is: Y it = α i + year t + β t Eligible year t + X it + ɛ it (1) Y it is the dependent variable, such as before-tax profits, investment, the number of employees, and total employee compensations (total labor cost) of firm i in year t. I run two types of regressions: one uses 2004 as a base year, and the other uses the base years from 2004 to In regressions in which 2004 is a base year, year t are year dummies from 2005 until Years 2005 to 2007 are placebo years because they were before the tax 21

22 cut and deferral program. The policy years are 2008 and 2009, and the years after the policy ended are from 2010 to When I use the base years from 2004 to 2007 to eliminate yearly fluctuation, year t are year dummies from 2008 until The coefficients of interest are the β t s, which are the coefficients of the interactions between eligibility and the years from 2005 to 2014 for the set of regressions with 2004 as a base year. If the identification strategy is valid, then the interaction of years 2005 to 2007, the pre-policy years, and firm eligibility should not be statistically different from 0. All standard errors in the difference-in-differences regressions are clustered at the firm level. 18 X it are time-variant control variables that allow for firms that had different initial characteristics to grow differently in different years. Specifically, I control for ln(asset) i,2004 year t, ln(labor) i,2004 year t, and Y i,2004 year t. ln(asset) i,2004 year t is the interaction between the log of assets of firm i in 2004 and a dummy variable for year t. ln(labor) i,2004 year t are interactions between the log of the number of employees of firm i in 2004 and year dummies from 2005 to Y i,2004 year t is the interaction between the dependent variable in 2004 and a dummy variable for year t. These variables control for possible differential time trends by different initial firm characteristics. They also control for the fact that different-sized firms might have been affected differently by the Great Recession. 6 Results This section shows regression results for which I use 2004 as a base year. This set of figures is to show that there is no pre-trend between eligible and ineligible firms. It also shows the main regression results for which I use together as base years. 18 Bootstrap standard errors are similar to clustered standard errors. 22

23 6.1 Investment Figure 2 shows the coefficient estimates of equation 1 using firm unbalanced data with net investment as a dependent variable and 2004 as a base year. Net investment of eligible and ineligible firms did not exhibit differential trends before the policy was implemented. Specifically, the coefficients of the interaction between the placebo year 2005, 2006, and 2007 and the eligibility indicator are not statistically different from 0, reducing concerns about endogeneity. Figure 2: Yearly Coefficients of Net Investment. Base year: 2004 (a) All firms (b) Foreign-owned firms (c) Domestic firms Figure 3: Yearly Coefficients of Net Investment: i.e. Columns (1), (2), and (3) of Table 7. Base years are from 2004 to 2007 (a) All firms (b) Foreign-owned firms (c) Domestic firms Coefficients of equation 1. The dependent variable is net investment winsorized at the 95th percentile. Unbalanced panel data from Eligible and ineligible firms are defined in Table 2.The policy was implemented in the last quarter of 2008 and throughout Cluster standard errors at the firm level. Figure 3 plots the coefficients of columns 1,2, and 3 in Table 7, which show the main regression results of equation 1 with base years from 2004 to Along with Figure 2, it 23

24 Table 7: The effects of the tax policies on net investment and # of employees. Base years are from Coefficients of columns (1),(2), and (3) are plotted in Figure 3. Coefficients of columns (4),(5), (6) are plotted in Figure 5 investment investment investment labor labor labor all domestic foreign all domestic foreign eligible& (policy year) (0.0364) (0.0479) (0.0479) (0.0450) (0.0535) (0.0706) eligible& (policy year) (0.0331) (0.0444) (0.0427) (0.0597) (0.0677) (0.117) eligible& (policy ended) (0.0362) (0.0508) (0.0424) (0.0569) (0.0683) (0.0955) eligible& (policy ended) (0.0344) (0.0486) (0.0424) (0.0819) (0.102) (0.103) eligible& (policy ended) (0.0358) (0.0485) (0.0419) (0.0734) (0.0878) (0.119) eligible& (policy ended) (0.0305) (0.0378) (0.0442) (0.0874) (0.106) (0.124) eligible& (policy ended) (0.0308) (0.0418) (0.0393) (0.0895) (0.102) (0.163) Constant ( ) ( ) ( ) (0.0188) (0.0222) (0.0306) N F r Dependent variable: net investment winsorized at the 95 th percentile and log of the number of employees. Standard errors in parentheses. Method: difference-in-differences approach with firm fixed effects regressions in a unbalanced panel data. The policy was implemented in the last quarter of 2008 and the whole year of Base years are from 2004 to Treatment years: from 2008 to Observations are firms of different ownership structure. Eligible firms had no more than 300 long-term employees in 2007 and its initial asset was > 500,000 USD. Ineligible firms had more than 300 long-term employees in 2007 and its initial asset was > 500,000 USD. Eligible & year2007 is the interaction between the eligiblity indicator and year Eligible & year2008, Eligible & year2009, etc. have similar interpretations. Cluster standard errors at the firms level. Control variables include interactions between log labor, fixed asset, and dependent variable, revenue, positive profit indicator in 2004 and year dummies from 2008 to p < 0.1, p < 0.05, p <

25 shows that net investment increased in the policy year and did not decrease after the policy ended. Additionally, the results of net investment were driven preliminary by domestic firms, not by foreign-owned firms. As Column 1 of Table 7 and Figure 3a show, net investment increased by 0.11 in 2009, the policy year, and did not decrease after This ratio in years before the policy was 0.08, as shown in Table 6 or in the constant term of column 1, Table 7. Thus, net investment increased about twice as much in the policy year and did not decrease after the policy ended. As will be shown in Section 6.2, employment did not increase. Thus, the investment-labor ratio increased in the policy year and did not reverse after the policy ended. These results suggest that firms investment-labor ratio increased in the policy year, and did not decrease after the policy s repeal. Thus, firms were credit-constrained and used the extra cash on hand to invest. In Appendix 9.3, I show that the results are also similar when I use different measures of investment (Figure 9) or winsorize net investment at the 99th percentile (Figure 10) instead of the 95th percentile. In addition, Figure 11 shows that results are similar when I run the analyses only on firms eligible for the policy from 2010 to 2012; when I do not use asset requirements to define eligibility; when I also include firms born from in the sample; when I use the balanced panel dataset from ; or when I use firms with positive profits in Figure 12 shows that the results are robust to different employment windows. The magnitude of investment is reasonable if firms used all extra profits saved on taxes to invest. The average profits before tax per every dollar of fixed assets is $0.442 (Table 6). When the corporate income tax rate was reduced from 25 percentage points to 17.5 percentage points, for every dollar of annual fixed assets, firms would save, on average, $ = $ Column 1 of Table 7 shows that the point estimate of net investment in 2009 is $0.11 per every dollar of fixed assets, and the 95% confidence interval is (0.05, 0.18). The average $ saved on taxes is very close to the (0.05,0.18) range. Additionally, Figure 9 shows that $ is in the the 95% confidence intervals of (0.0015, 25

26 0.082) and (0.0062,0.154), the confidence intervals when I use investment measures that are constructed from new investment in machinery and equipment. 6.2 Labor Figure 4 shows that the employment numbers of eligible and ineligible firms did not experience differential trends before the policies were implemented. Specifically, the coefficients of the interaction between the placebo year 2005, 2006, and 2007 and the eligibility indicator are not statistically different from 0, reducing concerns about endogeneity. As shown in Figure 5 and columns 4,5, and 6 of Table 7 where base years are from , I do not find evidence of an overall change in employment among all firms and across firm types. Figure 4: Yearly DID Coefficients of # of Employees. Base year: 2004 (a) All firms (b) Foreign-owned firms (c) Domestic firms 26

27 Figure 5: Yearly Coefficients of # of Employees: i.e. Columns (4), (5), and (6) of Table 7. Base years are from 2004 to (a) All firms (b) Foreign-owned firms (c) Domestic firms Coefficients of equation 1. The dependent variable is the log of the number of employees. Unbalanced panel data from Eligible and ineligible firms are defined in Table 2.The policy was implemented in the last quarter of 2008 and the whole year of Cluster standard errors at the firm level. 6.3 Profits Figure 6 shows the coefficient estimates of equation 1 using firm unbalanced data and 2004 as a base year. Profits of eligible and ineligible firms do not exhibit differential trends before the implementation of the policy. Specifically, the coefficients of the interaction between the placebo year 2005, 2006, and 2007 and the eligibility indicator are not statistically different from 0, reducing concerns about endogeneity. Figure 6: Yearly Coefficients of Profits. Base year: 2004 (a) All firms (b) Foreign-owned firms (c) Domestic firms 27

28 Figure 7: Yearly Coefficients of Profits. Base years are from (a) All firms (b) Foreign-owned firms (c) Domestic firms Figure 8: Yearly Coefficients of Profits Controlling for Labor and Employment. Base years are from (a) All firms (b) Foreign-owned firms (c) Domestic firms Coefficients of equation 1. The dependent variable is profits before tax, winsorized at the 99 th percentile. Unbalanced panel data from Eligible and ineligible firms are defined in Table 2.The policy was implemented in the last quarter of 2008 and throughout Cluster standard errors at the firm level. The results for reported profits depend on the type of firm ownership. Columns 5 and 6 of Table 8, Figure 7, and Figure 8 show that there is an increase in reported profits for foreign-owned firms. Eligible foreign-owned firms reported around $600,000 more in 2009 and $1 million more in 2010 and 2011 compared to ineligible foreign-owned firms. These figures are approximately 1.5 times larger than the average profit of eligible and ineligible firms from 2004 to By 2012, there is no longer evidence for an increase in the reported profits among eligible foreign-owned firms. In Appendix 9.3, I show that the results are similar when I winsorize profits at the 95 th percentile (Figure 14). In addition, Figure 15 shows that results are similar when I run 28

29 Table 8: The effect of the tax policies on profits. Base years are from The coefficients are also graphed in Figure 7 and Figure 8. all all domestic domestic foreign foreign eligible& (policy year) (117.6) (115.5) (108.2) (106.1) (272.3) (264.4) eligible& (policy year) (113.2) (111.8) (96.91) (94.63) (329.1) (326.0) eligible& (policy ended) (118.7) (116.7) (118.4) (116.6) (294.1) (289.1) eligible& (policy ended) (136.9) (133.7) (124.8) (121.5) (340.3) (331.3) eligible& (policy ended) (129.7) (125.3) (126.1) (118.6) (332.7) (329.4) eligible& (policy ended) (129.9) (124.5) (110.8) (102.6) (381.6) (375.6) eligible& (policy ended) (130.2) (127.4) (111.3) (109.1) (364.7) (359.1) Constant (30.04) (475.8) (25.53) (539.0) (88.64) (1079.0) N F r Control A yes yes yes yes yes yes Control B no yes no yes no yes Standard errors in parentheses. Unbalanced panel data from Method: difference-in-differences approach with firm fixed effects regressions. The dependent variable is the profit before tax in 2005 currency and converted to dollars. Profits are winsorized a 99 th percentile. Eligible and ineligible firms are defined in Table 2.The policy was implemented in the last quarter of 2008 and the whole year of Base years are from 2004 to Treatment years: from 2008 to Eligible & year2008& is the interaction between the eligibility indicator and year Eligible & year2009, Eligible & year2010, etc. have similar interpretations. Cluster standard errors at the firm level. Control variables A: interactions between log labor, fixed asset, and dependent variable, revenue, positive profit indicator in 2004 and year dummies from 2008 to Control variables B: control variables A and log fixed asset, log number of employees, and log labor cost. p < 0.1, p < 0.05, p <

30 the analyses only on firms eligible for the policy from 2010 to 2012; when I do not use asset requirements to define eligibility, when I also include firms born from in the sample; and when I use the balanced panel dataset from Figure 16 in Appendix 9.3 shows that these results are consistent when I vary the window of number of employees. Two broad mechanisms that explain the results in reported profits are changes in factor inputs and changes in reporting behavior. I show that the increase in reported profits among foreign-owned firms did not come from changes in factor inputs by demonstrating that the results are robust to including controls for capital and labor. Factor inputs are firm fixed assets, the number of employees, and total labor cost. Assuming that workers get paid their marginal product, total labor cost, conditional on the number of workers, is a measure of labor productivity. Results are reported in column 6 of Table 8 and in Figure 8. The magnitude and the significance level of the eligibility coefficients do not change when I include factor input variables (comparing column 5 of Table 8 with column 6 of Table 8, and comparing Figure 7 with Figure 8). This finding implies that the changes in profits were not caused by firms increasing real factor inputs, such as the number of workers, labor productivity, or capital. With a temporary tax cut and deferral policy, foreign-owned firms could save money by shifting profits from the non-tax-cut years to the tax-cut years. Firms could shift profits from 2008 and 2010 to 2009 because 2009 had the lowest tax rates of these three years. Such behavior predicts negative coefficients in 2008 and 2010 and a positive coefficient in However, the coefficients of foreign-owned firms in 2008 and 2010 are not negative and statistically significant. Therefore, I do not find evidence that the increase in reported profits from 2009 to 2011 among foreign-owned firms came from profit shifting across years. Multinational businesses could shift profits from countries with high tax rates to countries with low tax rates. I find an increase in reported profits among foreign-owned firms, but not among domestic firms (columns 5 and 6 versus columns 3 and 4 of Table 8). This evidence supports the theory that multinationals shifted profits across countries to take advantage of 30

31 the tax cut and deferral program. Profits continued to remain high for eligible foreign-owned firms in 2010 and 2011, suggesting that multinationals could not instantaneously change their reported profits when tax policies changed. Foreign-owned firms might have been uncertain about when the policies started and ended. Additionally, perhaps the changes in behavior leading to the increase in reported profits in the tax-cut year were costly to undo in the years following the policy s repeal. For example, foreign-owned firms might have reported a higher output price to increase profits in the tax-cut year. It might have been hard to report a very low output price for the same item in the years after the tax cut was repealed because the chance of getting audited is higher for firms with unusually low profits than for those with unusually high profits. Thus, firms would report more in 2010 and 2011 if they reported more in The results for 2009 are smaller than those for 2010 and 2011, suggesting that it may have taken firms time to learn about the tax policies and plan their taxes according. A model of this intuition is presented in Appendix 9.2. I find no evidence that, from 2012 to 2014, three to five years after the policy was implemented, eligible foreign-owned firms reported higher profits than ineligible foreign-owned firms. I find no evidence for increases in reported profits among domestic firms as a result of the tax cut and deferral policy, as columns 3 and 4 of Table 8 indicate. On the contrary, some domestic firms may have reported less profits after the policy. Point estimates in 2008 are similar to those in 2011 and 2012 with much larger standard errors. It could be that some domestic firms shifted profits from 2008 to 2009, but there were not enough firms shifting to provide statistically significant results. The point estimates in 2010, 2011 and 2012 are negative, and those in 2012 are statistically significant. The firms eligible for the 2009 policy were no longer eligible in 2011 and 2012 and may have felt entitled to receive the benefits of the policy again. As a result, they might have reported less profits to compensate for paying a higher tax rate. 31

32 6.3.1 Tax Revenue This section shows that foreign-owned firms that took advantage of the policy paid more in taxes as a result of the higher profits they reported. I start by calculating the present discounted value of the tax payments that the government received as a result of the 2009 tax cut and deferral policy. Assume that t was the normal tax rate of foreign-owned firms, and 0.7t was the reduced rate. A nine-month tax deferral lowered the present value of their tax payment due to the later payment. With an average nominal interest rate of 9%, 19 the average nominal interest rate over the nine-month period was 9%/12*9=6.75%= The present value of the tax amount that eligible firms owed the government as a result of the tax cut and deferral policy was (0.7t profit)/( ) 0.65 t profit. Thus, the combined tax reduction from both policies was a [1 (0.65t profit/t profit)] = a 35% decrease in the corporate income tax rate. Looking at reported profits alone does not provide a definitive story about tax revenues. First, tax revenues are affected not only by profits, but also by tax rates. It is unclear whether or not a firm would have had to pay more in taxes if it reported a higher profit in a low-tax year. In addition, if a firm reported $100,000 more in profits and faced a corporate tax rate of 17.5%, it is not necessarily true that the firm would have had to pay $17,500 more in taxes. This is because firms with negative or 0 profit did not need to pay the corporate income tax. Consider a firm that made a loss of $99,900 last year and a profit of $100 this year. The firm s reported profit increases by $100,000. However, its tax liability goes from 0 to $17.5, not to $17,500. Or if a firm made a loss of $101,000 last year and a loss of only $1000 this year, the firm s reported profit also increases by $100,000, but its tax liability stays at zero. I create a taxable profit variable that replaces all negative profits with 0. In other words, taxable profit=max(0, reported profit). If t is the tax rate that a foreign-owned firm is

33 supposed to pay, estimated tax payments would then equal t taxable prof its for ineligible foreign-owned firms. Eligible foreign-owned firms would pay 0.65 t taxable prof its, reflecting the combined effect of the 30% tax cut and the nine-month tax deferral. 20 Let a be the profits that a firm would report if it received no tax cut, and b the extra profits that a firm would report if the firm received the tax cut. The total taxes to be paid if the firm did not receive the tax cut is t a. The total taxes to be paid if the firm received a 30% tax cut and a deferral is 0.65ta tb. Tax revenue would increase if t a > 0.65ta tb. Thus, b/a > Because the coefficient of foreign-owned firms in 2010 is larger and more precisely estimated than those in 2009, I now calculate a and b of foreign-owned firms in Column 3 in Table 9 controls for firm fixed effects, year dummies, and interaction terms between the eligibility indicator, year dummies, and differential time trend of firms with different initial sizes in Thus, the constant coefficient of about $1 million in column 3 of Table 9 represents the average profits of eligible and ineligible foreign-owned firms from 2004 to The coefficient of year dummy 2010 is statistically insignificant. Therefore, the average profit of ineligible firms in 2010 was also around $1 million. In other words, a foreign-owned firm would report $1 million in 2010 if it did not get the tax cut and deferral i.e., a =$1 million. Because of the tax cut and deferral policy, foreign-owned firms reported about $788,600 more than they would have in the absence of the policy. Thus, b = $788, 600 in I have 788,600 1,000,000 (1.09) > Note that the factor 1.09 accounts for the average inflation rate from 2009 to Thus, eligible foreign-owned firms paid more taxes than their ineligible counterparts in Similar results hold for Column 2 of Table 9 shows that the taxable profits of domestic firms did not increase. Thus, as a result of the policy, the tax revenues of domestic firms decreased. 20 The tax liability variable is available, but a lot of observations are missing. Ten percent of these missing observations also reported negative profits before tax. Approximately 15% of firms with positive profits did not report their tax liability data. 33

34 Table 9: The effect of the tax cut on taxable profits. all domestic foreign eligible& (100.6) (99.95) (221.7) eligible& (102.5) (93.67) (282.0) eligible& (111.0) (113.1) (268.3) eligible& (122.7) (113.7) (309.3) eligible& (113.6) (109.7) (293.3) eligible& (116.9) (100.0) (348.6) eligible& (113.6) (96.03) (324.6) year (572.3) (618.9) (1353.3) year (831.1) (958.1) (1782.0) year (876.8) (1188.5) (1384.5) year (843.6) (993.5) (1631.7) year (741.7) (975.1) (1294.8) year (826.4) (1011.2) (1515.4) year (760.6) (879.1) (1693.6) Constant (26.98) (23.55) (78.61) N Standard errors in parentheses. Unbalanced panel data from Method: difference-in-differences approach with firm fixed effects regressions. The dependent variable is the constructed taxable profits in 2005 currency and converted to dollars, which equals profits if the profits> 0, and 0 otherwise. I winsorize profits at a 99 th percentile to avoid outliers. Eligible and ineligible firms are defined in Table 2. Base years: 2004 to Treatment years: 2008 to Eligible &2008& is the interaction between the eligibility indicator and year Eligible &2009, etc. have similar interpretations. Cluster standard errors at the firm level. Control variables include interactions between log labor, fixed asset, revenue, profits, and 2009-positive profits indicators in 2004 and year dummies from 2008 to p < 0.1, p < 0.05, p <

35 7 Policy Implication What do we learn from the case study of Vietnam? From a budgeting perspective, the Vietnamese government faced a trade-off between gaining tax revenues from foreign-owned firms and losing them from domestic firms. From an economic stimulus perspective, firms invested more because they had more cash on hand. How applicable is the evidence from Vietnam to other countries? During the Great Recession, the policy in Vietnam was similar to those of other countries in terms of duration,but much more generous in terms of magnitude. For example, the Netherlands temporarily reduced the rate for the first EUR 200,000 of profit by 13% in 2009 and Singapore decreased the rate by 5% in Taiwan offered a three-month deferral of corporate income tax in 2008 for firms that met certain criteria. In the UK, eligible firms could defer corporate and other tax payments. In comparison, Vietnam offered a 30% tax cut and a nine-month deferral policy in 2009 and the last quarter of Despite the government s large influence on the economy, the Vietnamese business environment shares many similarities with that of other countries, especially developing countries. Credit constraints are thought to deter growth. In fact, the temporary tax cut and deferral policy in Vietnam and other stimulus policies in different countries during the Great Recession were implemented to relax firm liquidity constraints. 21 Additionally, foreign businesses have been increasingly important to the development and direction of the Vietnamese economy. As a result, Vietnam also faces multinational profit shifting, one of the central topics in corporate taxation in developed countries Deloittee: Tax Responses to the global economic crisis in The G20-OECD project on base erosion and profit shifting (BEPS) aims to strengthen the international corporate tax system by limiting illegal multinational profit-shifting. 35

36 8 Conclusions This paper evaluates the impact of a temporary 30% corporate income tax cut and a ninemonth corporate income tax deferral in Vietnam from the end of 2008 throughout First, I find that investment increased, mostly due to relaxation of credit constraints. I find null results for employment. Second, eligible foreign-owned firms reported a large increase in profits in the policy year and in the two years after the policy ended. Multinational profitshifting was the most likely mechanism. The persistence of high reported profits in later years implies that multinationals cannot instantly change their reporting behavior when tax polices change. Tax payments by foreign-owned firms that took advantage of the tax cut and deferral program increased due to higher reported profits. Multinational profit-shifting has been one of the central topics in corporate taxation in developed countries. 23 Despite a heavy reliance on corporate income tax for government revenues, in developing countries, multinational profit-shifting has only recently attracted the attention of policy makers, and academic research on this issue is rare. 24 What capacity do developing countries have to assess profits that multinational corporations earn in their jurisdictions? What is the appropriate penalty for misreporting profit? How much should a developing country tax multinational corporations? These are important questions that must be addressed to help developing countries retain profits earned by foreign-owned corporations in their jurisdictions. References Banerjee, Abhijit V. and Esther Duflo, Do Firms Want to Borrow More? Testing Credit Constraints Using a Directed Lending Program, Review of Economic Studies, 23 The new G20-OECD project on base erosion and profit shifting (BEPS) aims to strengthen the international corporate tax system by limiting illegal multinational profit-shifting. 24 See Crivelli et al. (2015). Also in 2012, Vietnam, for the first time convicted some foreign-owned firms for undue shifting abroad of profits attributable to operations in Vietnam. After 2012, the Vietnam government created a new office to monitor asset profit-shifting of multinational corporations. 36

37 2014, 81 (2), Bustos, Alvaro, Eduardo M. R. A. Engel, and Alexander Galetovic, Could higher taxes increase the long-run demand for capital? Theory and evidence for Chile, Journal of Development Economics, April 2004, 73 (2), Cai, Jing and Ann Harrison, Industrial Policy in China: Some Unintended Consequences?, Cerda, Rodrigo A. and Felipe Larrain, Corporate taxes and the demand for labor and capital in developing countries, Small Business Economics, February 2010, 34 (2), Clausing, Kimberly A, Tax-motivated transfer pricing and US intrafirm trade prices, Journal of Public Economics, September 2003, 87 (9), Clotfelter, Charles T., Tax Evasion and Tax Rates: An Analysis of Individual Returns, The Review of Economics and Statistics, 1983, 65 (3), Cremer, Helmuth and Firouz Gahvari, Tax evasion and optimal commodity taxation, Journal of Public Economics, 1993, 50 (2), Cristea, Anca D. and Daniel X. Nguyen, Transfer Pricing by Multinational Firms: New Evidence from Foreign Firm Ownerships, American Economic Journal: Economic Policy, August 2016, 8 (3), Crivelli, Ernesto, Ruud de Mooij, and Michael Keen, Base Erosion, Profit Shifting and Developing Countries, IMF Working Paper 15/118, International Monetary Fund May Cummins, Jason G., Kevin A. Hassett, and R. Glenn Hubbard, A Reconsideration of Investment Behavior Using Tax Reforms as Natural Experiments, Brookings Papers on Economic Activity, 1994, 25 (2),

38 ,, and, Tax reforms and investment: A cross-country comparison, Journal of Public Economics, October 1996, 62 (1), Davies, Ronald B., Julien Martin, Mathieu Parenti, and Farid Toubal, Knocking on Tax Haven s Door: Multinational Firms and Transfer Pricing, The Review of Economics and Statistics, April Djankov, Simeon, Tim Ganser, Caralee McLiesh, Rita Ramalho, and Andrei Shleifer, The Effect of Corporate Taxes on Investment and Entrepreneurship, American Economic Journal: Macroeconomics, July 2010, 2 (3), Feinstein, Jonathan S., An Econometric Analysis of Income Tax Evasion and its Detection, RAND Journal of Economics, 1991, 22 (1), Ferede, Ergete and Bev Dahlby, THE IMPACT OF TAX CUTS ON ECONOMIC GROWTH: EVIDENCE FROM THE CANADIAN PROVINCES, National Tax Journal, 2012, 65 (3), Fisman, Raymond and ShangJin Wei, Tax Rates and Tax Evasion: Evidence from Missing Imports in China, Journal of Political Economy, April 2004, 112 (2), Gorodnichenko, Yuriy, Jorge MartinezVazquez, and KlaraSabirianova Peter, Myth and Reality of Flat Tax Reform: Micro Estimates of Tax Evasion Response and Welfare Effects in Russia, Journal of Political Economy, June 2009, 117 (3), Hassett, Kevin A. and R. Glenn Hubbard, Tax Policy and Business Investment**RGH acknowledges financial support from Harvard Business School and the American Enterprise Institute., Handbook of Public Economics, January 2002, 3, House, Christopher L. and Matthew D. Shapiro, Temporary Investment Tax Incentives: Theory with Evidence from Bonus Depreciation, American Economic Review, June 2008, 98 (3),

39 Huizinga, Harry and Luc Laeven, International profit shifting within multinationals: A multi-country perspective, Journal of Public Economics, June 2008, 92 (5), Jorgenson, Dale and R. E. Hall, Tax Policy and Investment Behavior, American Economic Review, 1967, 57 (3), Kopczuk, Wojciech, John Friedman, Harvey Galper, Marco Manacorda, Emmanuel Saez, Dan Silverman, and Joel Slemrod, The Polish business flat tax and its effect on reported incomes: a Pareto improving tax reform? Liu, Li, Tim Schmidt-Eisenlohr, and Dongxian Guo, International Transfer Pricing and Tax Avoidance: Evidence from Linked Trade-Tax Statistics in the UK, SSRN Scholarly Paper ID , Social Science Research Network, Rochester, NY August Swenson, Deborah, Tax Reforms and Evidence of Transfer Pricing, National Tax Journal, 2001, 54 (n. 1), Weichenrieder, Alfons J., Profit shifting in the EU: evidence from Germany, International Tax and Public Finance, June 2009, 16 (3), Yagan, Danny, Capital Tax Reform and the Real Economy: The Effects of the 2003 Dividend Tax Cut, American Economic Review, December 2015, 105 (12), Zwick, Eric and James Mahon, Tax Policy and Heterogeneous Investment Behavior, American Economic Review, January 2017, 107 (1),

40 9 Appendices 9.1 Model Without Credit Constraints Effects on Capital Assume that firms have constant return-to-scale production functions. Assume that the market is perfectly competitive. The relationship between investment I s and capital K s in period s is as follows: I s = K s (1 δ)k s 1 with δ the capital replacement rate. K 0 = I 0 K 1 = I 1 + (1 δ)i 0 K 2 = I 2 + (1 δ)k 1 = I 2 + (1 δ)i 1 + (1 δ) 2 I 0 K 3 = I 3 + (1 δ)k 2 = I 3 + (1 δ)i 2 + (1 δ) 2 I 1 + (1 δ) 3 I 0... In period t, a firm pays taxes on its tax accounting profits, which equal to its revenues minus labor costs minus all investment depreciation from period 0 until the period t. Thus, its tax accounting before tax profit in period t is π t = p t f(k t, L t ) w t L t t 0 q s D t s I s e r(t s) ds (2) f(k t, L t ): production function with capital K t and L t δ: economic depreciation of capital r: discount rate D t s : capital depreciation rate from tax accounting standard at time t s q s : price of a unit of investmentl I s in period s p t : output price at period t. Firms maximize their lifetime profits. A firm chooses a sequence of investments I 0, I 1, I 2,... from period 0 to the end of time. max I0,I 1,I 2,...E(π) = 0 (p t f(k t, L t )e rt wl t q t I t ) }{{} before tax economic profits t τ t (p t f(k t, L t ) w t L t q s D t s I s e r(t s) ds) dt 0 }{{} amount of tax paid in period t from equation 2 τ t : corporate income tax rate at time t FOC w.r.t I s and differentiating K s, K s+1,... w.r.t I s gives s p t f Kt e (δ+r)(t s) (1 τ t )dt + s τ t q s D t s e r(t s) dt = q s (3) Equation 3 implies that a firm chooses an investment I s at time s until its marginal revenue product equals its upfront cost minus its depreciation until the end of time. Assume that the firm s production function is Cobb-Douglas i.e., K α L 1 α. Thus, f K is a function of k = K/L. Therefore, we can write f K = α(k/l) α 1 = αk α 1, which is a decreasing function in k. Thus, f Ksk s < 0. I am interested in how a temporary change in the corporate income tax rate affects capital-labor ratio k. Assume that the corporate income tax rate τ t temporarily changes in 40

41 period t=s. Differentiate FOC 3 w.r.t to τ s : p s f Ks + q s D 0 + p t (1 τ s )f Ksk s k s / τ s = 0 Therefore, k s / τ s = p sf Ks q s D 0 p s (1 τ s )f Ksks Since f Ksks<0, if the marginal revenue product is higher than the first year s depreciation, the capital ratio increases as the tax rate in period s, τ s, decreases. If the marginal revenue product is smaller than the first year s depreciation, the capital labor ratio decreases as the tax rate decreases. In period s + 1, when there is no tax cut, firms choose investment I s+1. The first-order condition when choosing I s+1 is the same as when choosing I s 1. Now, the tax rate from period s + 1 returns to normal, which is the same as it was before period s. Therefore, the capital labor ratio would come back to its normal level Effects on Profits Observed tax accounting profits before tax in year t are: π t = p t f(k t, L t ) w t L t t 0 q s D t s I s ds Thus, π t = p t f(k t, L t ) w t L t t 0 q s D t s (K s (1 δ)k s 1 )ds π t / τ t = p t f Kt K t / τ t + p t f Lt L t / τ t w t L t / τ t q t D 0 K t / τ t (4) = K t τ t (p t f Kt q t D 0 )(because p t f Lt w t = 0 due to FOC of labor ) If the firm decides to invest more in the tax-cut year i.e. K t / τ t < 0 it must be that p t f Kt > q t D 0. Thus, π t / τ t < 0. If the firm decides to invest less in the tax-cut year, i.e. K t / τ t > 0, it must be that p t f Kt < q t D 0. Thus, π t / τ t < 0. In other words, if investment changes as a result of the decrease in the tax rate, tax-accounting profit before tax increases. If investment does not change when tax rate decreases because of adjustment cost, tax account profit before tax stays the same. In period t + 1, we have tax-accounting profits before tax: Thus π t+1 = p t+1 f(k t+1, L t+1 ) w t+1 L t+1 t+1 0 q s D t+1 s (K s (1 δ)k s 1 )ds K t+1 K t+1 K t π t+1 / τ t = p t+1 f Kt+1 q t+1 D 0 + (1 δ)q t D 1 τ t τ t τ t = K t+1 (p t+1 f Kt+1 q t+1 D 0 ) + K t [(1 δ)q t+1 D 0 q t D t+1 ] τ t τ t The tax-accounting profits before tax in period t + 1 is ambiguous. For example, let 41

42 us take the case in which the long-term capital stock stays the same i.e K t+1 / τ t = 0. Thus, π t+1 / τ t = Kt τ t [(1 δ)q t+1 D 0 q t D 1 ]. This implies that π t+1 / τ t depends on how the capital stock in year t responds to a change in the tax rate, and how the first year s deprecation compares to the second year s. 9.2 A two-period tax evasion model I extend the static framework of Cremer and Gahvari (1993) to construct a two-period tax evasion model. A firm in country A lives for two periods. The tax rates in these two periods are t 1 and t 2. Assume that the market is perfectly competitive and that firms are price takers. Also assume that prices are determined by the world market and are not affected by policies in country A. For simplicity, assume that per unit output price and cost are the same in the two periods. I denote them as p and c, respectively. Assume that output quantities in the two periods are x 1 and x 2. Assume that firms evade by under-reporting sales. Let h 1 and h 2 be the proportion of sales that firms hide from the government, 0 h 1, h 2 1. A tax-evading firm would report only ((1 h 1 )p c)x 1 in the first period and ((1 h 2 )p c)x 2 in the second period to the government. In the first period, the firm faces an increasing and convex cost g(h 1 ) to hide h 1 units of evasion. Thus, its total cost of hiding in the first period is pg(h 1 )x 1. Define h = h 2 h 1. In the second period, to hide h 2 units of evasion, the firm faces an increasing and convex cost g(h 2 ) and an increasing and convex adjustment cost f( h ). Also assume that f (0) = 0. Thus, its total cost of hiding h 2 proportion of sales in the second period is p(g(h 2 ) + f( h))x 2. The adjustment cost function f( h) represents the idea that it is expensive to change habits, or it is expensive to adjust the evasion rates. For instance, firms can hide evasion by engaging only in cash transactions. However, if a firm allows credit card transactions in the tax-cut year to attract more customers, it is hard for them to switch back to cash transactions at the end of the tax-cut period. Assume that the government decides to randomly audit firms in each period with the probability β. If a firm gets caught evading in one period, it has to pay τ times the amount of tax it evades, where τ > 1. It also has to pay an extra of ph 1 xτ in the first period and ph 2 xτ in the second period. Thus, a firm s expected tax payments in the two periods are (((1 h 1 )p c)x 1 + ph 1 x 1 τ)t and (((1 h 2 )p c)x 2 + ph 2 x 2 τ)t. Let δ be the time discount rate. Thus, a firm s lifetime expected profit is as follows: E(π) = (p c pg(h 1 ))x }{{} 1 [(1 h 1 )p c + h 1 βpτ]t 1 x }{{} 1 profit before tax period 1 expected tax payment in period 1 + δ (p c pg(h 2 ) pf( h))x }{{} 2 δ [(1 h 2 )p c + h 2 βpτ]t 2 x }{{} 2 profit before tax period 2 expected tax payment in period 2 (5) FOC w.r.t h 1 : [ g/ h 1 ( 1 + βτ)t 1 ]x 1 + δ( f/ h)x 2 = 0 g It follows that: h 1 x 1 δ f x h 2 = (1 βτ)p 1 t 1 x 1 g FOC w.r.t h 2 : h 2 + f = (1 βτ)pt h 2 42

43 From 9.2, I have 1 βτ > 0 since I assume that g(.) and f(.) are increasing functions. Intuitively, if βτ > 1, there would be no evasion because the cost of evasion is always higher than its benefit. Comparative statics: What happens to the evasion rates in the first period and the second period, h 1 and h 2, when the tax rate in the first period t 1 changes? Differentiating LHS and RHS of equation 9.2 w.r.t t 1 gives us: 2 g h 1 x h δ 2 f 1 t 1 h x h δ 2 f t 1 h x h = (1 βτ)px 1 t 1 Thus, h 1 t 1 ( ) 2 g x h δ 2 f 1 h x 2 2 h 2 2 f δx 2 t 1 h = (1 βτ)p 1x 2 1 (6) Differentiating LHS and RHS of equation 9.2 w.r.t t 1 gives us: 2 g h 2 2 f h f h 2 = 0 h 2 2 t 1 h 2 t 1 h 2 t 1 h 1 t 1 ( ) 2 f h }{{ 2 } A = h 2 t 1 ( ) 2 g h + 2 g 2 h 2 2 }{{} B Substituting h 2 / t 1 in equation 7 to equation 6, I get h 1 2 g 1 2 f x t 1 h δx 2 1 h δx ( 2 f ) 2 h ( 2 f + 2 g h 2 = (1 βτ)px 1 (8) It implies that h 1 t 1 h 2 2 ) [ ] 2 g x h 2 1 ( 2 f 1 h + 2 g 2 f 2 g ) + δx 2 h 2 2 = (1 βτ) px 2 h 2 h 2 2 }{{} 1 (9) }{{} D>0 C>0 Model predictions: - Since C and D > 0, h 1 t 1 > 0. Thus,the evasion rate decreases in the first period when the tax rate in the first period decreases. This implies higher profits in the tax-cut year compared to the years prior to the tax cut. -The evasion rate responds less to a temporary tax cut than to a permanent tax cut. -The evasion rate in the second period h 2 decreases as tax rate t 1 decreases. This implies higher profits in the year immediately after the tax cut compared to the years prior to the tax cut. -Since A < B, the evasion rate decreases less in period 2 than in period 1 when the tax rate in the first period t 1 decreases. This implies higher profits in a tax- cut year compared to the year immediately after the tax cut. (7) 43

44 9.3 Robustness Check Investment Figure 9: Effects on Investment: Measures are New Investments in Machinery. (a) Imputed 0 (b) No Imputed 0 Eligible and ineligible firms in this paper are defined in Table 2. Figure 10: Effects on Net Investment: Winsorized at the 99 th percentile (a) all firms (b) foreign-owned firms (c) domestic firms Eligible and ineligible firms in this paper are defined in Table 2. 44

45 Figure 11: Effects on Net Investment. Different Samples. All firms. Winsorized at the 95 th percentile (a) Eligibility in Table 2, only eligi-(bble firms from set Eligibility in Table 2, but No As- Requirement (c) Eligibility in Table 2, include(d) Eligibility in Table 2, balanced(e) Eligibility in Table 2, positive firms born in or before 2007 panel data profits in

46 Figure 12: Net investment Graph the estimates and the 90% confidence interval. Eligible and ineligible firms in this paper are defined in Table 2. Net investment is winsorized at the 95 th percentile. 46

47 Figure 13: Log number of employees (a) foreign-owned firms (b) domestic firms Graph the estimates and the 90% confidence interval. Eligible and ineligible firms are defined in Table 2. 47