How Do Small Taxpayers Respond to Tax Simplification

How Do Small Taxpayers Respond to Tax Simpliﬁcation: Evidence from the Indonesian Turnover Tax Reform

Himawan Saputro∗ November 17, 2020

Abstract This paper studies the firms’ behavioral responses to tax simplification in Indonesia by exploiting the implementation of turnover tax in 2013 and the expansion of value added tax (VAT) registration threshold in 2014 as well as the fact that both turnover tax eligibility and VAT registration thresholds are at the same value. I find that firms intentionally bunch just above the threshold in order to avoid turnover tax regime and retain compliance benefits from income tax avoidance of around 11.5 percent of value added. This response persists after the expansion of VAT registration threshold. The results suggest that replacing income tax with turnover tax minimizes exploitation of tax loopholes by firms. I also document that, in 2014, some firms started to bunch just below the threshold in order to avoid VAT burden as well as compliance costs of about 8.9 percent of value added. The results translate into tax elasticity of around 0.651 when only taxes on income are considered and 0.879 when both taxes on income and consumption are included, indicating that firms are more responsive to VAT relative to income taxes.

∗Martin School of Public Policy and Administration, University of Kentucky, 405 Patterson Oﬃce Tower, Lexington, KY 40506 (email: [email protected]). I am grateful to my advisors David Agrawal, William Hoyt, Rajeev Darolia and Michael Reed. 1 Introduction

A large share of small firms in the developing economies with lack of administrative capacity to maintain proper recordkeeping and estimate accurate tax bases, hampers governments in observing true tax bases and levying taxes on them (Gordon and Li, 2009; Besley and Persson, 2014). The literature on compliance costs estimation suggests that small firms regressively spend more costs to comply with tax law relative to large firms (Tran-Nam et al., 2000). Presumed tax bases such as turnover, value of assets, and deemed profit are reason- able alternatives when the preferred tax bases such as income and value added are difficult for governments to measure, verify, and monitor (Tanzi and de Jantscher, 1986; Slemrod and Yitzhaki, 1994). This efficacy of presumed tax bases to simplify tax collection and reduce compliance costs has attracted governments to adopt various types of presumptive taxes (Bird et al., 2003). Recent empirical works suggest that presumptive taxes encourage small firms to enter the tax system (Fajnzylber et al., 2011; Bruhn and Loeprick, 2016). However, few studies examine whether moving from a complex tax system such as income tax to a simple presumptive tax reduces compliance costs and how small firms respond to the tax simplification. This paper exploits the implementation of the 2013 Indonesian turnover tax to replace the ordinary income tax as well as the 2014 VAT registration threshold expansion and use the population of Indonesian corporate tax returns from 2007 to 2017 to study the impact of tax simplification on compliance costs and how taxpayers respond to it. The turnover tax allows taxpayers with previous year turnover of up to 4.8 billion IDR to be taxed at a final rate of 1 percent on the turnover instead of taxed at 25 percent on profit.1 The turnover tax was designed to simplify tax system and reduce compliance costs for small firms as well as minimizing governments monitoring costs. The scheme dismisses small firms from the complexity of profit calculation in order to pay taxes. Despite the simplified tax regime for small firms, the tax laws that mandates all firms to maintain a proper book keeping and report financial information such as balance sheet and profit and loss statements in tax return have not been repealed. These requirements imply that small firms eligible for turnover tax regime potentially still bear similar compliance costs. Moreover, the average income effective tax rates around the threshold is around 0.85 percent of turnover or slightly below the turnover tax rate of 1 percent. Firms below the threshold would also loss an opportunity to obtain benefits from exploiting income tax loopholes. In 2014, the government further simplify taxation of small firms by increasing the VAT registration threshold from 600 million

1In 2013, the exchange rate for 1 IDR is around 10,000 USD. Thus, 4.8 billion IDR is around 480,000 USD.

1 to 4.8 billion IDR of annual turnover. This expansion reduces small firms compliance costs yet removes firms’ rights to claim VAT on inputs. The impact of these policies on small firms’ behavior would be determined by the change in tax incentives, compliance costs, and compliance benefits. I use a bunching approach developed by Saez (2010) and Chetty et al. (2011) and further extended by Kleven and Waseem (2013) to investigate the firms responses to the turnover tax and VAT thresholds. The approach uses excess mass near the threshold that represents discontinuity in incentives to obtain behavioral responses and estimate structural parameters. The turnover tax implementation in 2013 and VAT threshold expansion in 2014 produce several incentive discontinuities at the 4.8 billion IDR threshold, namely income tax rates, VAT rates, and compliance costs which determine whether small firms obtain benefits from the tax simplification or not. The compliance costs reduction from the turnover tax would be matched against compliance benefits from the income tax system where firms are allowed to exploit loopholes in tax law and save taxes (Johnston, 1963). I use a framework discussed in Calonico et al. (2014) to estimate compliance benefits from income taxes in the form of tax avoidance as well as tax evasion responses in the form of misreporting direct inputs. I find that the turnover tax creates an excess mass just above the threshold in 2013 because due to jump in the average tax liability to turnover ratio from around 8.5 percent to 1 percent and reduction in compliance benefits from income tax system of around 6.4 percent of annual turnover or 11.5 percent of value added. The excess mass above the threshold preserves when the VAT threshold expanded in 2014 for firms who decided to stay in the income tax regime. These firms obtain 12.6 percent higher compliance benefits relative to firms who selected to adopt the turnover tax regime. Moreover, firms in the turnover tax regime bunch just below the threshold, suggesting that the VAT rate and compliance costs are larger than the benefit of staying in the income tax regime. The results indicate that tax simplification should consider not only compliance costs but also average tax rates and compliance benefits. I also find that the VAT registration threshold expansion in 2014 is responded with tax evasion in the form of excessive under reporting of direct inputs. After the expansion, firms just below the notch in the turnover tax regime have 16,33 percent lower di direct input to turnover ratio relative to those just above the notch. In addition, firms just below the notch in the income tax regime have 5.27 percent lower direct input to turnover ratio compared to their counterparts just above the notch. Moreover, I find no discontinuity in tax liability to turnover ratio around the threshold in both regimes. The results suggest that VAT system induces evasion responses around the threshold. The results further indicates that firms may obtain greater benefits from under reporting direct input in VAT system than in income tax

2 system. Finally, I find that accounting for compliance benefits from book-tax adjustment of around 0.015 of the reported value added reduces tax elasticity from 0.725 to 0.651 when government enacted the turnover tax in 2013. I also find that the tax elasticity increases to 0.879 when the government expanded the VAT registration threshold in 2014. The tax elasticity is larger (1.183) if compliance costs, tax avoidance, and tax evasion are excluded from the estimation. The results indicate that VAT system induces larger behavioral responses from taxpayers relative to income tax system. This paper contributes to the policy discussion on taxation of small businesses (En- gelschalk, 2005). Many governments, particularly in developing countries, implement presumptive taxes to simplify tax regime for small businesses, curtail compliance costs, and limit monitoring costs. The paper introduces compliance benefits as an important factor in designing tax simplification, in addition to compliance costs. The results complements literature that finds that compliance costs reduction is the main driver of taxpayers responses to presumptive taxes (Fajnzylber et al., 2011; Bruhn and Loeprick, 2016). Moreover, the paper contributes to the literature on the limited tax capacity of emerging countries which partly driven by the large share of small businesses in the economies (Gordon and Li, 2009; Besley and Persson, 2014). The results also contribute to the literature on tax compliance costs measurement (Slem- rod and Blumenthal, 1996; Tran-Nam et al., 2000) by providing evidence on how compliance benefits from book-tax allignment can be estimated and netted out from compliance costs. The body of works in compliance costs posits that the compliance costs should be netted from compliance benefits such as managerial benefits, cash flow, deduction benefits, and tax planning or tax avoidance activities. However, many compliance costs measurement works disregard tax planning and avoidance benefits because reliable estimates on these costs are extremely difficult to find using the typical surveyed-based study. This paper also extends the relatively limited empirical evidence on corporate tax avoidance (Desai and Dharmapala, 2006, 2009). Finally, the paper contributes to the literature on bunching methodology (Kleven, 2016) by highlighting the benefits from tax avoidance as an additional driver for taxpayers behavioral responses along with other incentives such as tax rates (Saez, 2010; Chetty et al., 2011; Kleven and Waseem, 2013) and compliance costs (Harju et al., 2019). Moreover, the paper extends the bunching literature by examining and decomposing taxpayers behavioral responses under multiple notches.

3 2 Setting and Data

2.1 Taxation of small ﬁrms in Indonesia

In general, the Indonesian firms are taxed at around 25 percent on profit. Moreover, the tax schedule for firms with turnover of up to 50 billion IDR is a flat rate of 25% with a 50% reduction of the tax rate to taxable income corresponding to turnover of up to 4.8 billion IDR. On the other hand, small firms in Indonesia has been subject to a couple alternatives of presumptive tax regimes. The first regime is the deemed profit tax implemented in 1986 and the second regime is the turnover tax regime enacted in 2013. Prior to the implementation of turnover tax regime in 2013, firms below the threshold can elect their preferred tax regime, i.e. deemed profit tax or corporate income tax regime. However, since 2013, most firms below the threshold must elect the turnover tax regime. Under the deemed profit tax regime, small firms calculate their income tax liabilities using a range of profit ratios determined by the government.2 While under the turnover tax, small firms are taxed at a final rate of 1% of the gross income. Thus, the turnover tax offers lower tax burden relative to the deemed profit tax. The government sets an eligibility threshold for presumptive tax regimes in terms of annual turnover. The presumptive tax threshold is typically similar to the VAT registration threshold. For example, the annual turnover threshold for the deemed profit tax and VAT registration is 600 million for the year 2010 to 2012. And starting 2014, annual turnover threshold for turnover tax and VAT registration is 4.8 billion IDR . In other words, the government has been removing small firms from the VAT system as well as offering them less complicated presumptive tax regime. Despite this tax simplification, the laws that requires firms to maintain book keeping, estimate profit, and report financial information to tax authority have not been repealed. This provision allows me to observe firms financial reporting near the threshold and study how they respond to tax simplification. The general provision and procedures of taxation law mandates all Indonesian firms to maintain a book keeping practice.3 The book keeping practice includes consistent recording financial information such as gross income, cost of good sold, profit, expenses, assets, liabilities, and capitals as well as compiling income statement and balance sheet at the end of fiscal year. The financial information is then reported in the corporate tax return as starting

2There are around 1,435 industry categories that their proﬁt ratios are determined by the government. For example, the proﬁt ratio for wholesaler, retailer, and event organizer are around 20%, 25%, and 50%. These ratios are then multiplied by the turnover in order to get the taxable income and eventually the income tax liability. 3Article 28 of Law number 6 year 1983 as already amended several times and the latest by Law number 28 year 2007 regarding General Provision and Procedures of Taxation.

4 point of the estimation of taxable income and income tax liability. The net profit will be adjusted to obtain taxable income by accounting for differences in treatment and timing of accruing income and expenses between firms’ accounting practices and the income tax law (book-tax adjustments). The adjustment is called positive fiscal adjustment if the adjustment increases the taxable profit and is called negative fiscal adjustment if the adjustment decreases the taxable profit. Business profit will be equal to taxable profit if firm has no fiscal adjustment. Positive fiscal adjustments includes expenses incurred for the personal benefit of shareholders, in kind benefit or remuneration to employees, excessive compensation paid to shareholder, deferred expenses, and positive excess in commercial depreciation or amortization relative to fiscal depreciation or amortization. Negative fiscal adjustment includes negative excess in commercial depreciation or amortization relative to fiscal depreciation or amortization and deferred income. Naturally, taxpayers would minimize net fiscal adjustment in order to get lower income taxes. I exploit the variation of net fiscal adjustment around the turnover tax eligibility and VAT registration thresholds to test whether compliance benefits from tax avoidance drive firms’ behavioral responses to tax simplification.

2.1.1 The 2013 Turnover Tax Reform

In 2013, Indonesian government enacted a turnover tax scheme for taxpayers with previous year turnover of up to approximately 4.8 billion IDR.4 Under this scheme, eligible taxpayers are taxed at a final rate of 1 percent on the turnover in lieu of 25 percent on firms’ profit. The level of previous annual turnover would dictate the type of regime that firms should follow for the current year. If a firm earned more than 4.8 billion in the current year, they have to pay monthly income tax installments tax starting January next year. And vice versa, if firms earned less than the threshold, they have to pay monthly turnover taxes starting January next year. Moreover, firms in certain industries specifically regulated by other special tax regimes, such as construction and real estate, are not eligible for the turnover tax scheme. Thus, I exclude construction and real estate firms in the estimation of behavioral responses because they may not affected by the turnover tax reform. Figure 1 shows that the average corporate income tax to turnover ratio prior (2007-2017) to the reform is less than the mandated turnover tax rate of around 1%. The average tax ratio between the old and new small firms thresholds is around 0.85% to 0.9%. In other words, the reform raises the average effective tax burden for small firms. Thus, all else equal, small firms might prefer income tax regime relative to turnover tax regime.

4Government Regulation Number 46 year 2013 regarding taxation on business income of certain individual and corporate taxpayer.

5 Figure 1: Average Corporate Income Tax in Indonesia, 2007-2012 .012

Old small business New small business threshold threshold .011 .01 .009 Average income tax to turnover ratio .008 0 0.6 2 4 4.8 6 Turnover in billion IDR

Notes: The figure shows the scatter plot of average corporate income tax to turnover ratio in 100 equally sized bins before (2007-2012) the turnover tax reform. The left vertical line represents the old (2003) VAT registration and small firms threshold at 0.6 billion IDR and the right vertical line represents the new (2013) small firms threshold and the new (2014) VAT registration threshold. The horizontal-dashed line denotes the turnover tax rate of 1% applicable for taxpayer with turnover of up to 4.8 billion IDR.

The reform simplifies the regime for taxes on income yet does not remove the main costs of complying with income tax laws, e.g. book keeping practice and financial reporting. The income tax compliance costs in Indonesia is substantial. Susila and Pope (2012) find that the cost of complying to income tax regime for Indonesian firms, on average, is around 28% of total compliance costs or approximately 3.14% of turnover. On the other hand, the income taxes permit firms to exploit and obtain benefits from tax loopholes. This condition may discourage firms near the threshold from entering the turnover tax regime.

2.1.2 The 2014 Value Added Tax Reform

Following the turnover tax reform, in 2014, the government further simplify taxation for small firms by increasing the VAT registration threshold from 0.6 billion IDR to 4.8 billion IDR, equating to the turnover tax eligibility threshold. Firms with turnover below the VAT threshold are freed from VAT compliance costs of about 3 percent of turnover for firms with turnover above 3 billion IDR and around 11.2 percent of turnover for firms with turnover of up to 3 billion IDR (Susila and Pope, 2012). However, firms with turnover under the threshold are not allowed to claim input VAT.

6 Figure 2: Excess Mass: Single vs Multiple Thresholds

Panel A: Turnover Tax Threshold Panel B: Turnover Tax and VAT Thresholds 2000 1500 1500 1000 1000 Number of firms Number of firms 500 500

−2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Turnover in million IDR Turnover in million IDR

2012 2013 2012 2014

Notes: The ﬁgure shows the observed annual turnover distribution before (2012) and after (2013) the turnover tax reform as well as after (2014) the VAT reform. Panel A compares the 2012 turnover distribution (solid- triangle line) and the 2013 turnover distribution (solid-dotted line). Panel B compares the 2012 turnover distribution (solid-triangle line) with the 2014 turnover distribution (solid-dotted line).

Firms responses to the two reforms are well captured by Figure 2. Panel A shows that the turnover tax reform induces taxpayers to bunch just above the threshold. The 2012 distribution of turnover is a counterfactual for such responses. This type of bunching is known as bunching at tax notch where the incentives change discontinuously. However, Panel B shows that, after the 2014 VAT registration threshold increase, the shape of bunching changes into kink type of bunching where the incentives change marginally. This paper examine the mechanisms behind taxpayers responses to both turnover tax and VAT reforms. Speciﬁcally, the study looks at various incentives discontinuities at the threshold including tax rates, compliance costs, as well as compliance beneﬁts.

2.2 Data

The empirical analysis is based on Indonesian tax administrative data set that contains full population of corporate tax returns from 2007 to 2017. The data includes detailed information on revenue and expenses, fiscal adjustments, tax payments, tax credits, industry classification and regions. The data contains composition of expenses which can be used to find the determinants of small firms responses to the tax scheme. The information on industry classification can be exploited to estimate heterogeneity of responses among the sectors. I restrict the data to cover only firms and not individuals because they account for around 80 percent of total turnover tax collection. In addition, the number of firms around

7 the turnover tax threshold is larger than that of individual taxpayers and so the distribution of individual taxpayers are relatively more noisy.

Table 1: Summary Statistics

Before reforms (2012) After reforms (2014) Below Above Below Above Variables threshold threshold threshold threshold Turnover (billion IDR) 0.485 106.08 0.426 95.591 (0.864) (947.919) (0.786) (948.196) Input (billion IDR) 0.319 78.679 0.276 70.65 (0.745) (722.932) (0.848) (696.442) Profit (billion IDR) 0.098 10.457 0.091 9.062 (10.927) (249.355) (11.123) (250.406) Fiscal adjustments (billion IDR) 0.025 3.181 0.104 3.774 (7.028) (70.078) (5.403) (91.926) Income tax liabilities (million IDR) 6.297 1,879.215 2.865 1,607.941 (256.032) (38,344.33) (193.892) (53,560.04) Turnover tax liabilities (million IDR) n/a n/a 1.926 3.237 (n/a) (n/a) (5.879) (16.814) Number of firms Turnover > 0 242,279 66,278 268,829 84,571 Turnover = 0 148,846 - 193,225 - Direct input to turnover ratio All firms 0.363 0.739 0.334 0.729 (4.181) (0.334) (7.216) (0.35) Firms with non-zero turnover 0.587 0.739 0.575 0.729 (5.301) (0.334) (9.453) (0.35) Fiscal adjustments to turnover ratio All firms 0.0566 0.0544 0.188 0.087 (3.325) (1.356) (5.177) (0.773) Firms with non-zero turnover 0.0913 0.0544 0.324 0.087 (4.226) (1.356) (6.78) (0.773) Notes: Statistics reported are means with standard deviations in parentheses in the year before (2012) and after (2014) the turnover tax and VAT reforms. All monetary variables are measured in constant 2010 IDR.

Table 1 displays the summary statistics of the variables before (2012) and after (2014) the turnover tax and VAT reforms. It can be seen that firms below the threshold have considerably smaller average annual turnover, input, profit, fiscal adjustments and income tax liabilities relative to those above the threshold both before and after the reforms. Moreover, except fiscal adjustments, all monetary variables are lower in 2014 than in 2012. Firms below threshold have considerably larger increase in average fiscal adjustments (319%) relative to that of firm above the threshold (19%). This meaningful change implies that firms below threshold declare higher additional taxable income after the reforms. Given that almost

8 all firms below threshold are not subject to income tax, their declaration of higher taxable income would not affect their tax liabilities. The table also shows that firms with no turnover account more than one third of firms below the threshold in 2012 and 2014. This makes the averages for firms below threshold look much smaller relative to those of firms above the threshold where they all have non-zero turnover. The financial ratios, namely direct input and fiscal adjustments to turnover tax ratios are presented into two categories, such as all firms and firms with non-zero turnover. The results demonstrate that the size of firms with zero turnover affects the averages of financial ratios.

3 Theoretical Framework

3.1 Framework set up

I set up the framework based off the model developed by Kleven and Waseem (2013). I also consider VAT compliance costs as one of the drivers of bunching responses as in Harju et al. (2019). However, in my framework I account for income tax compliance benefits as an additional mechanism behind the bunching responses. In addition, my framework handles multiple notches (downward and upward) that occur at the same threshold. These features are excluded from the models developed in the previous bunching studies. The turnover tax and VAT thresholds cause several discontinuous increases in average costs of taxation such as tax rates (income tax and VAT), compliance costs (income tax and VAT reporting requirements), and compliance benefits (tax avoidance activities). Thus, the thresholds produce a notch in firms’ budget set. Previous studies suggest that this notch can affect behavior which can be observed as an excess mass of firms located in the tax preferred side near the threshold in the turnover distribution. Assume that a group of many firms producing a homogeneous good and selling their products to consumers. Consider perfectly elastic demand and the good’s price is normalized to one. A firm is type a, capturing its productivity. Production of value added, v = (1 − α)y, 0 ≤ α < 1, needs intermediate goods as inputs. Let α be the amount of inputs to produce one unit of revenue y. Consider firm’s disutility from producing output

a (1 − α)y 1+1/e φ(y) = (1) 1 + 1/e a where e is the elasticity of tax base with respect to the net-of-tax rate. The ﬁrm maximizes quasi-linear utility u(y) = c − φ(y), where c = (1 − α)y − τ(y, α) −

9 t(y, α) − θ(y, α) − η(y, α), τ(y, α) represents accrued VAT payment, t(y, α) denotes accrued income tax payments, θ(y, α) denotes VAT net compliance costs, and η(y, α) indicates income tax net compliance beneﬁts. The quasi-linear, isoelastic utility function before the implementation of turnover tax is as follow

1 u(y) = − α y − ty − θ(1 − α)y − ηy − φ(y) 1 + τ τ t η = 1 − − − θ − (1 − α)y − φ(y) (2) (1 + τ)(1 − α) (1 − α) (1 − α)

The framework aims at analyzing incentives created by a turnover tax and VAT registration thresholds commonly applied to small firms. As in previous studies (Liu and Lockwood, 2016; Harju et al., 2019), the assumptions above are realistic in this context. In my analysis, the firms affected by the turnover tax and VAT thresholds are firms with relatively large annual turnover. These firms are expected to be affected by the thresholds. Appendix B shows that the 2014 VAT registration threshold creates larger incentives than the 2003 VAT registration theshold because the later imposed to firms with much smaller annual turnover. These suggest that firm’s ability significantly dictate output and thus justifying the utility maximization model. On the other hand, I could set up a profit maximization model with various degree of productivity and a particular production function as in Waseem (2018) and would produce similar results as the model in used by this paper. I further assume that the firm is unable to pass VAT burden to consumers by increasing the price. This assumption is build on both theoratical literature (Keen and Mintz, 2004; Kanbur and Keen, 2014) and empirical evidence (Carbonnier, 2007; Benedek et al., 2015; Kosonen, 2015; Benzarti et al., 2017; Harju et al., 2018) studying the incidence of VAT. Likewise, I assume that the firm cannot pass the compliance costs and benefits to consumers through price increase.

3.2 Single threshold

I start with the model coresponds to the 2013 turnover tax notch and extend it to include VAT notch starting in 2014. In 2013, a firm pays turnover tax instead of income tax when its annual turnover less than the threshold. In 2014, a firm eligible for turnover tax regime is exempted from the VAT system. Thus, I will abstract the VAT system from the analysis for 2013 (single threshold case) and incorporate it to the analysis for 2014-2017 (multiple threshold case). I also exclude income tax compliance costs θ from the first part of the study because the costs do not change at the turnover tax threshold in 2013. The firm below the turnover tax

10 threshold is still required to report turnover, expenses, profit, fiscal adjustments, and submit the corporate tax return as well as remitting VAT and reporting VAT monthly tax returns . a (y 1+1/e I assume the disutility from producing output is φ(y) = 1+1/e a , instead of expression in equation (1) because the tax base is turnover y and not value added (1 − α)y. The utility of a firm under and above the turnover tax threshold in 2013 is  (1 − T )y − φ(y) if y ≤ y∗ u(y) = (3) (1 − t + η)y − φ(y) if y > y∗ where T is the turnover tax rate when T > t (based on the statutory value of T and the observed value of t). Solving maximization problem in Equation (3) for firms below y∗ yields y = a(1 − T )e. Consider a smooth cumulative distribution of firms with productivity levels F (a) and a y related density function f(a). The implications of this are F (y) = F (1−T )e and f(y) = y 1 ∗ f (1−T )e) (1−T )e . Therefore, absent any discontinuity at y a smooth turnover distribution exists. Using the first-order condition y = (a∗ − ∆a∗)(1 − T )e, the utility for lowest-productivity firm (type L) as the marginal buncher at y∗ − ∆y∗ is

e uL = (a∗ − ∆a∗)(1 − T )e+1 1 − e + 1

The utility for the highest-productivity ﬁrm (type H) at the notch point y∗ is

a∗ − ∆a∗ y 1+1/e uH = (1 − t − η)y − 1 + 1/e a∗ − ∆a∗

L H ∗ ∗ y∗−∆y∗ Given u = u and a − ∆a = (1−t)e , the following expression can be derived

e a∗ − ∆a∗ (1 − α)y 1+1/e (a∗ − ∆a∗)(1 − T )e+1 1 − = (1 − t − η)(1 − α)y − (4) e + 1 1 + 1/e a∗ − ∆a∗

The indiﬀerence condition can be expressed as the following

∆y∗ ∆y∗ −1/e 1 − t + η 1 − + e 1 − − (1 + e) = 0 (5) y∗ y∗ 1 − T

This formula provides a relationship between tax elasticity, the behavioral response of ∆y∗ the marginal buncher relative to the threshold y∗ , the eﬀective tax rates below (T ) and

11 above (t) the threshold, and the income tax compliance beneﬁts η. The above expression is the formula for bunching at a downward notch where average tax rate discretely falls at the threshold.

3.3 Multiple Thresholds

In 2014, Indonesian government increased the VAT registration threshold from 600 million to 4.8 billion IDR, equalizing the turnover tax threshold. Therefore, there are four notches at the same level of turnover, namely VAT rate, turnover tax rate, VAT compliance costs and income tax compliance beneﬁts. In this combined tax notches regime, the ﬁrm’s utility is  (1 − T )y − φ(y) if y ≤ y∗ u(y) = (6) τ t η ∗  1 − (1+τ)(1−α) − (1−α) − θ − (1−α) (1 − α)y − φ(y) if y > y Rearranging previous equations, I get the following expression  (1 − γB)(1 − α)y − φ(y) if y ≤ y∗ u(y) = (7) A η ∗ (1 − γ − θ + 1−α )(1 − α)y − φ(y) if y > y

B A τ T where γ = T and γ = (1+τ)(1−α) + (1−α) denote the effective tax rates below and above the threshold, γB < γA when 0 ≤ α ≤ 1. I assume that VAT compliance benefits and income tax compliance benefits are increasing functions of turnover, thus compliance costs and benefits just below the threshold is stricly less than those above the threshold. However, I cannot observe income tax compliance benefits and VAT compliance costs in the next fiscal year for taxpayer who selects turnover tax regime instead of income tax regime. The selection criteria for tax regime is the following  y ≤ y∗ if γB < γA − θ + η y = (1−α) (8) ∗ B A η y > y if γ > γ − θ + (1−α) The above selection equation describes the relationship between tax regimes, tax rates above and below the threshold, compliance costs and compliance benefits. Budget sets in Figure 3 illustrate bunching at multiple (turnover tax and VAT) thresholds. ∆γ represents the discontinuous change in average effective total tax burden when the turnover tax and VAT thresholds are surpassed where γ = T + t + τ, θ represents the compliance costs, and η represents the compliance benefits from tax avoidance. Panel A displays a condition when average costs of taxation falls at the threshold or known as an downward

12 notch. In this panel, Firm L represents the marginal buncher with turnover y∗ − ∆y in the absence of the threshold with ability a∗ − ∆a∗. This firm is indifferent between locating at y∗ or y∗ − ∆y. Thus, there will be bunching at the threshold for firms with abilities within the range [a∗ − ∆a∗]. Panel B illustrates a situation when average costs of taxation jumps at the threshold or known as a upward notch. In this panel, Firm H represents the marginal buncher with turnover y∗ + ∆y in the absence of the thresholds with ability a∗ + ∆a∗. This firm is indifferent between locating at y∗ or yB. Thus, there will be bunching at the threshold for firms with abilities within the range [a∗ + ∆a∗].

Figure 3: Behavioral Responses to Tax Notches

Panel A: Budget Set Diagram: Downward Notch Panel B: Budget Set Diagram: Upward Notch

Notes: The ﬁgure shows behavioral responses to an average tax costs fall (downward notch) and to an average tax costs jump (upward notch) in two budget set diagrams. Panel A illustrates behavioral responses to falls in total average costs of taxation such as tax rates, compliance costs, and compliance beneﬁts. Panel B illustrates behavioral responses to jumps in total average costs of taxation.

3.3.1 Downward tax notch: taxpayers in the income tax regime

L H Similar to the downward notch in turnover tax regime, I can state uD = uD where D refers to the downward tax notch analysis. Using the ﬁrst-order condition y(1 − α) = (a∗ − ∆a∗)(1 − γB − θB)e , the utility for lowest-productivity ﬁrm (type L) as the marginal buncher at y∗ − ∆y∗ is

e uL = (a∗ − ∆a∗)(1 − γB)e+1 1 − D e + 1

13 The utility for the highest-productivity ﬁrm (type H) at the notch point y∗I is

η a∗ − ∆a∗ (1 − α)y 1+1/e uH = (1 − γA − θ + )(1 − α)y − D (1 − α) 1 + 1/e a∗ − ∆a∗

L H ∗ ∗ y∗−∆y∗ Given uD = uD and a − ∆a = (1−t)e , the following expression can be derived

e η a∗ − ∆a∗ (1 − α)y 1+1/e (a∗−∆a∗)(1−γB)e+1 1− = (1−γA−θ+ )(1−α)y− e + 1 (1 − α) 1 + 1/e a∗ − ∆a∗ (9) The indiﬀerence condition can be expressed as the following

η ∆y∗ ∆y∗ −1/e 1 − γA − θ + 1 − + e 1 − − (1 + e) (1−α) = 0 (10) y∗ y∗ 1 − γB

This expression explains the relationship between tax elasticity, the response of the marginal buncher, the tax rates above and below the threshold, VAT compliance costs and income tax compliance beneﬁts in the context of downward tax notch analysis.

3.3.2 Upward tax notch: taxpayers in turnover tax regime

The utilities of marginal buncher H and the firms with lowest-productivity L locating at ∗ L H y are indifference, suggesting that uU = uU where U denotes the upward tax notch analysis. For a type-L firm, the utility is the following

a∗ + ∆a∗ y∗(1 − α)1+1/e uL = y∗(1 − α)(1 − γB) − (11) U 1 + 1/e a∗ + ∆a∗

For a type-H ﬁrm, based on the maximization of utility above y∗ in Equation (3), i.e. y∗(1 − α) = (a∗ + ∆a∗)(1 − γA − θ + η)e, the utility is the following

e uH = (a∗ + ∆a∗)(1 − γA − θ + η)e+1 1 − (12) U e + 1

L H Given uU = uU , the following expression can be derived

a∗ + ∆a∗ y∗(1 − α 1+1/e y∗(1 − α)(1 − tB) − (13) 1 + 1/e a∗ + ∆a∗ e = (a∗ + ∆a∗)(1 − γA − θ + η)e+1 1 − e + 1

14 H ∗ ∗ Using the ﬁrst order condition of uU in the absence of the threshold, i.e. (y + ∆y )(1 − α) = (a∗ + ∆a∗)(1 − γB)e and rearranging the terms, the following expression can be derived

!1+1/e !e+1 1 e 1 1 1 − γA − θ + η − − = 0 (14) ∆y∗ e + 1 ∆y∗ e + 1 1 − γB 1 + y∗ 1 + y∗ This expression explains the relationship between tax elasticity, the response of the marginal buncher, the tax rates above and below the threshold, VAT compliance costs and income tax compliance beneﬁts in the context of upward tax notch analysis.

4 Empirical Strategy

In general, bunching estimator can be defined as an estimation of the excess mass of taxpayers at the threshold by comparing the density of taxpayers bunching at the threshold with the density of taxpayers at the same turnover level in the absence of the threshold. The key procedure is to remove the influence of the threshold from the observed turnover distribution to obtain a counterfactual distribution. Thus, the identifying assumption of the causal interpretation is that the distribution of turnover would have been smooth if there were no discontinuity in average costs of taxation, in this case tax burden and net compliance costs. The eligibility turnover and VAT registration threshold set by the reforms at the turnover value y∗ persuade excess mass at the threshold for firms that fail to optimize the turnover tax or VAT registration. The bunching is motivated by the productivity parameter a. It will create an excess mass by firms that would have reported a turnover between y∗ and y∗ +4y∗ in the absence of the threshold as the following integration equation:

Z y∗+4y∗ B(y∗) = g(y)d(y) ' g(y) 4 y∗ (15) y∗ where B(y∗) is the excess mass and g(y) is the counterfactual density of the turnover distribution absent the threshold. I estimate the bunching responses caused by the turnover tax and VAT thresholds by comparing the observed excess mass in the turnover distribution above (below) the threshold to the counterfactual density that would occur in the absence of the discontinuity at the thresholds. This bunching estimate consists of average tax rate and net compliance costs. Following Chetty et al. (2011), Kleven and Waseem (2013), to estimate the counterfactual density, I ﬁt a ﬂexible polynomial function to the observed turnover distribution, excluding a range around the threshold.

15 I group ﬁrms into turnover bins of 100 million IDR and estimate a counterfactual density

by regressing the following equation and excluding the region around the threshold [yB, yA] from the regression

q yA X l X cj = βi(yj) + θi1(yj = i) + εj (16)

l=0 i=yB where cj is the count of ﬁrms in turnover bin j, yj is the level of turnover in bin j, q is

the order of the polynomial, and θi is dummy for the excluded region. The range (yB, yA) speciﬁes turnover bins near the threshold where bunching exists and thus excluded from the regression. The excess bunching mass start from lower bound of the excluded turnover

region, yB, to the upper bound of the excluded region yA. εj is the error term of the turnover density equation.

The counterfactual distribution estimate is the predicted bin countsc ˆj from (2) excluding

the dummies in the omitted region [yB, yA]. The excess bunching is deﬁned as the diﬀerence between the observed and the predicted bin counts over the region near the turnover and

VAT thresholds. I estimate yA iteratively so that the region below counterfactual density (B) is equal to the region above the observed density (A) such as the following:

∗ y yA ˆ X ˆ X B = (c ˆj − cj) and A = (cj − cˆj) (17) ∗ i=yB i=y

I raise repeatedly the value of yA until the excess mass (A) equals the missing mass (B). I then estimate the relative excess bunching ˆb by dividing the excess mass from the excluded region by the average excess mass in the same region

∗ Py Pi=yA (c ˆj − cj) ∗ (cj − cˆj) ˆ ∗ i=yB ˆ ∗ y b(y ) = ∗ or b(y ) = (18) Py cˆj Pi=yA cˆj ∗ i=yB Nk y Nj

∗ ∗ 5 where Nk is the number of bins within [yB, y ] and Nj is the number of bins within [y , yA].

5In initial bunching papers, e.g. Chetty et al. (2011), the ˆb estimator is deﬁned as the ratio of excess mass over the height of the counterfactual density at the threshold.

16 5 Results

5.1 Bunching at Single Threshold

Figure 4 displays the turnover distributions for all ﬁrms around the turnover tax threshold before (2012) and after (2013) the implementation of turnover tax scheme. Panel A shows that there is no observable bunching at the threshold in 2012, suggesting that any behavioral response to the threshold after the implementation of the turnover tax can be solely attributed to the reform. Moreover, Panel B shows that excess bunching at the threshold emerges and is statistically signiﬁcant (1.173 (0.359)). This excess bunching indicates a type of bunching at downward notch where average tax rate discretely falls at the threshold.6

Figure 4: Behavioral Responses to Single Threshold

Panel A: All Firms in 2012 Panel B: All Firms in 2013

Excess bunching (b) = 0.063 (0.118) Excess bunching (b) = 1.173 (0.359) 1200 1200 1000 1000 800 800 600 600 Number of firms Number of firms 400 400 200 200 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold in million IDR Distance to the threshold in million IDR

Observed Counterfactual Observed Counterfactual

Notes: The ﬁgure shows the observed distribution (solid-dotted line) and the estimated counterfactual (solid-dashed line) of annual turnover in bins of 100 million IDR within a range of -2000 million IDR and 2000 million IDR. The threshold is represented by a vertical solid line. The counterfactual is a ﬁfth-order polynomial estimated as in equation (16). The shaded area is the excess mass in the upper part of the excluded region. Excess bunching b is excess bunching in the omitted region near the threshold relative to the average counterfactual count in this region.

This bunching responses are driven by the jump in tax incentives as well as compliance benefits from tax avoidance. The 1 percent turnover tax rate is relatively higher than the average income tax to turnover ratio around the threshold which is around 0.85 percent (See Figure 1). Moreover, Figure 5 provides suggested evidence that compliance benefits from exploitation of income tax loopholes may also cause the responses. The figure shows

6Table A1 in Appendix A shows that the excess bunching estimates are relatively similar under seveth- order polynomial counterfactual but overestimated when third-order polynomial is used to construct counterfactual distribution.

17 that firms above the threshold have significantly less fiscal adjustments than those below the threshold and thus the former would pay lower average income tax burden relative to the latter.

Figure 5: Fiscal Adjustments at Single Threshold

Panel A: Fiscal Adjustments in 2012 Panel B: Fiscal Adjustments in 2013

Difference at the threshold d = −0.001 (0.009) Difference at the threshold d = −0.068 (0.019) Number of observations: left = 16,890 / right = 9,209 Number of observations: left = 15,931 / right = 11,133 .25 .25 .2 .2 .15 .15 .1 .1 .05 .05 0 0 Average fiscal adjustment to turnover ratio Average fiscal adjustment to turnover ratio −.05 −.05 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR)

Notes: The figure shows the plot of average fiscal adjustments to turnover ratio around the turnover tax eligibility threshold. Panel A and B show fiscal adjustments before (2012) and after (2013) the policy implementation. The vertical solid line is the normalized threshold. The number of bins used to the left of the threshold is 50 and to the right of the threshold is also 50. The solid lines in the left and the right of the threshold represent the conditional expectation function for firms below and above the threshold. Difference at the threshold d is the size of discontinuity of the dependent variable at the normalized threshold. The fitted lines are estimated using approach presented in Calonico et al. (2014).

Panel B shows that the difference in fiscal adjustments at the threshold is economically and statistically significant (-0.068(0.019)). In other words, on average, firms in the income tax regime have 6.8 percent more downward adjustments on their taxable income than those in the turnover tax regime. This difference creates incentive for firms to be eligible for income tax and obtain opportunities to conduct tax planning, exploiting loopholes in the tax law in order to minimize their tax burdens. I use this evidence as well as observed marginal buncher to independently solve equation (5) and estimate tax elasticity under single threshold case.

5.2 Bunching at Multiple Thresholds

Figure 6 presents bunching around the threshold after (2014-2017) the expansion of VAT registration threshold. Despite the discontinuity of tax rates and compliance costs at the threshold, panel A shows that the threshold induces a kink regime instead of a notch regime. However, the turnover tax and VAT threshold expansion theoreritically introduce no marginal change in both tax incentives or compliance costs around the threshold. In

18 fact, when taxpayers are grouped by tax regimes, i.e. income tax and turnover tax, the kink regime would disappear and two notches appear as demonstrated in panel B and C.7

Figure 6: Behavioral Reponses to Multiple Thresholds

Panel A: All Firms 2014-2017 Panel B: Firms in the Income Tax Regime

Excess bunching (b) = 0.535 (0.331) Excess bunching (b) = 0.818 (0.246) 8000 4000 6000 3000 4000 2000 Number of firms Number of firms 2000 1000 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold in million IDR Distance to the threshold in million IDR

Observed Counterfactual Observed Counterfactual

Panel C: Firms in the Turnover Tax Regime Panel D: Firms in the Mixed Regime

Excess bunching (b) = 0.864 (0.401) Excess bunching (b) = 0.769 (0.455) 4000 4000 3000 3000 2000 2000 Number of firms Number of firms 1000 1000 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold in million IDR Distance to the threshold in million IDR

Observed Counterfactual Observed Counterfactual

Notes: The ﬁgure shows the observed distribution (solid-dotted line) and the estimated counterfactual (solid-dashed line) of annual turnover after the turnover tax reform and VAT threshold expansion (2014- 2017) in bins of 100 million IDR within a range of -2000 million IDR and 2000 million IDR. The threshold is represented by a vertical solid line. The counterfactual is a ﬁfth-order polynomial estimated as in equation (16). The shaded area is the excess mass in the upper part of the excluded region. Excess bunching b is excess bunching in the omitted region near the threshold relative to the average counterfactual count in this region.

Panel B shows the distribution of turnover for ﬁrms with previous year turnover more than the threshold and thus subject to income tax regime. It presents a clear evidence of a downward tax notch for ﬁrms subject to income tax regime. This notch occurred in 2013 and

7Figure A1 and A4 in Appendix A present observed bunching by year in each tax regime. Moreover, Figure A3 plots estimates for excess bunching after the reforms for each tax regime. The ﬁgures demonstrates that excess bunching estimates in the income tax regime has been slightly decreasing while those of the turnover tax regime has been consistently increasing.

19 was induced by the turnover tax reform (see Figure 2 and 4) and continue to exist even after the VAT registration threshold was increased to the same level of turnover tax threshold. The number of firms bunching near the tax notch is around 1,787 and the excess mass is meaningful and statistically significant (0.818 (0.246)). Panel C present the distribution of turnover for firms with previous turnover of up to 4.8 billion IDR and thus are required to be in the turnover tax regime. It shows an evidence of an upward tax notch for firms subject to turnover tax regime. The notch occurred after the government increase the VAT registration threshold in 2014. The firms in the bunching area are those with last year turnover less than or equal to the threshold. The number of buncher near the tax notch is around 1,824 and the excess bunching is around 0.864. Panel D shows a downward notch for taxpayers subject to both income tax and turnover tax regimes (mixed regime). Nevertheless, the excess bunching is smaller than that of income tax regime and statistically less significant. Similarly, this downward notch might be driven by discontinous jump in compliance benefits (fall in fiscal adjustments). The compliance benefits from income tax regime in the context of multiple thresholds are discussed specifically in the next sub-section.

5.3 Compliance Beneﬁts from Income Tax Regime

The above findings imply that complying to income tax law provides benefits in terms of tax planning opportunities and saving taxes. This incentive discourages firms near the eligibility threshold to move from complex income tax regime to simple turnover tax regime. Figure 7 shows average fiscal adjustments to turnover tax ratio declared by all firms as well as each group of firms in various tax regimes. The ratio represents the rate where firms declare their additional taxable income due to various reasons, such as non-deductible expenses, larger actual depreciation expenses, larger actual amortization expenses and so on. Although information on fiscal adjustments is not used by firms in the turnover tax regime to calculate their tax liabilities, the government still requires them to file corporate tax return and to declare not only their turnover but also inputs, expenses, profits, and fiscal adjustments. This policy might create incentive for these firms to declare the true fiscal adjustments because it would not affect their tax liabilities in the current fiscal year. Panel A shows that the difference in fiscal adjustments to turnover ratio at the threshold for all firms is meaningful and statistically significant (-0.126 (0.008)). In other words, on average, firms in the income tax regime have 12.6 percent lower income tax burden relative to those in the turnover taxes had they still taxed under income tax regime. When the firms are divided based on the tax regime they selected, firms in the income tax regime (Panel B)

20 have significantly lower level of average fiscal adjustments relative to those in the turnover tax regime (Panel C). The level of average fiscal adjustments declared by firms in the income tax regime is about less than 15 percent while those in the turnover tax regime report around 35 percent in fiscal adjustments. I use this evidence and the observed marginal buncher to simultaneously solve equation (10) and (14) and estimate tax elasticity and VAT compliance costs under the multiple thresholds case.

Figure 7: Fiscal Adjustments at Multiple Thresholds

Panel A: All Firms 2014-2017 Panel B: Firms in the Income Tax Regime

Difference at the threshold d = −0.126 (0.008) Difference at the threshold d = −0.052 (0.008) Number of observations: left = 96,789 / right = 55885 Number of observations: left = 40,226 / right = 40,056 .45 .45 .4 .4 .35 .35 .3 .3 .25 .25 .2 .2 .15 .15 .1 .1 .05 .05 Average fiscal adjustment to turnover ratio Average fiscal adjustment to turnover ratio 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR) Panel C: Firms in the Turnover Tax Regime Panel D: Firms in the Mixed Regime

Difference at the threshold d = 0.002 (0.017) Difference at the threshold d = −0.064 (0.018) Number of observations: left = 52,114 / right = 11,426 Number of observations: left = 3,624 / right = 3,785 .45 .45 .4 .4 .35 .35 .3 .3 .25 .25 .2 .2 .15 .15 .1 .1 .05 .05 Average fiscal adjustment to turnover ratio Average fiscal adjustment to turnover ratio 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR)

Notes: The figure shows the plot of average fiscal adjustment to turnover ratio for all firms (Panel A) as well as for each regime (Panel B, C, and D) near the threshold after (2014-2017) the turnover tax reform and VAT threshold expansion. The vertical solid line is the normalized threshold. The number of bins used to the left of the threshold is 50 and to the right of the threshold is also 50. The solid lines in the left and the right of the threshold represent the conditional expectation function for firms below and above the threshold. Difference at the threshold d is the size of discontinuity of the dependent variable at the normalized threshold.

Moreover, Panel B shows discontinuity in average ﬁscal adjustments to turnover ratio of around -5.2 percent at the threshold, suggesting that ﬁrms that move from income tax

21 to turnover tax in the next year have slightly lower book-tax adjustments than those who stay in the income tax regime. On the other hand, Panel C demonstrates that firms that move from turnover tax to income tax regime have similar book-tax adjustments relative to those who stay in the turnover tax regime. The results imply that firms with relatively large positive fiscal adjustments and thus higher taxable income prefer to stay in the turnover tax regime.

5.4 Mechanism: Real vs Evasion Responses

In this part, I investigate the possibility of evasion responses induced by the discontinuity of tax and non-tax incentives at the threshold. Under the single threshold (turnover tax regime) in 2013, given that firms bunch above the threshold (see Figure 4), the evasion response, such as under reporting turnover as well as direct inputs would be less likely since firms prefer higher turnover relative to the threshold in order to avoid the turnover tax regime. The evasion response might occur when there is a continuous direct input ratio and a downward discontinuity in income tax ratio at the threshold. In this case, firms increase turnover while also disproportionately overreporting inputs and thus decrease tax ratio. On the other hand, a real response would be indicated by a continuity in direct input as well as in income tax ratio. In other words, firms just above the threshold proportionately raise both turnover as well as direct input, keeping income tax to turnover ratio continue at the threshold. Under the multiple thresholds (turnover tax and VAT regimes) from 2014 to 2017, the analysis is conditioned on the types of regimes because each regime has different bunching responses (see Figure 6). Moreover, I add some indications that signal responses to VAT system where firms tends to avoid passing the threshold through misreporting inputs as discussed in Liu and Lockwood (2016). Assuming that under reporting direct inputs provides greater benefit in VAT system than in income tax, the evasion response to VAT notch is signalled by an upward discontinuity in direct input ratio and a downward discontinuity in tax ratio at the threshold. In this situation, firms just below the notch under report direct input as they reduce the reported turnover in order to avoid VAT registration, but consequently pay higher income taxes to firms just above the notch.

22 Figure 8: Direct Input and Tax Liability at Single Threshold

Panel A: Direct Input Ratio in 2012 Panel B: Direct Input Ratio in 2013

Difference at the threshold d = 0.012 (0.018) Difference at the threshold d = 0.012 (0.019) Number of observations: left = 16,896 / right = 9,209 Number of observations: left = 15,939 / right = 11,137 .8 .8 .75 .75 .7 .7 .65 .65 .6 .6 Average direct input to turnover ratio Average direct input to turnover ratio .55 .55 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR) Panel C: Tax Ratio in 2012 Panel D: Tax Ratio in 2013

Difference at the threshold d = −0.0002 (0.0006) Difference at the threshold d = −0.001 (0.001) Number of observations: left = 16,896 / right = 9,209 Number of observations: left = 15,939 / right = 11,137 .015 .015 .01 .01 .005 .005 Average income tax to turnover ratio Average income tax to turnover ratio 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR)

Notes: The figure plots the average fiscal adjustment to turnover ratio (Panel A and B) as well as tax liability to turnover ratio (Panel C and D) near the threshold before (2012) and after (2013) the turnover tax reform. The vertical solid line is the normalized threshold. The number of bins used to the left of the threshold is 50 and to the right of the threshold is also 50. The solid lines in the left and the right of the threshold represent the conditional expectation function for firms below and above the threshold. Difference at the threshold d is the size of discontinuity of the dependent variable at the normalized threshold.

Figure 8 shows no indication of evasion esponses to the turnover tax policy near the threshold when the reform is enacted in 2013. Panel A and B show that the average direct input ratios before (2012) and right after (2013) the reform do not discontinue at the threshold. Panel C and D demonstrate that the average income tax to turnover ratio before (2012) the turnover tax implementation are distributed smoothly near the threshold. The results suggest that firms bunch just above the threshold by proportionately increase turnover and direct input, keeping the income tax burden unchanged. This behavior might reflect real response since firms increases their outputs and correctly report their inputs to stay in the income tax regime.

23 Figure 9: Direct Input and Tax Liability at Multiple Thresholds 2014-2017

Panel A: Direct Input in Income Tax Regime Panel B: Direct Input in Turnover Tax Regime

Difference at the threshold d = 0.071 (0.011) Difference at the threshold d = 0.098 (0.011) Number of observations: left = 41,129 / right = 40,515 Number of observations: left = 55,769 / right = 12,003 .8 .8 .75 .75 .7 .7 .65 .65 .6 .6 Average direct input to turnover ratio Average direct input to turnover ratio .55 .55 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR) Panel C: Tax Ratio in Income Tax Regime Panel D: Tax Ratio in Turnover Tax Regime

Difference at the threshold d = −0.001 (0.000) Difference at the threshold d = −0.000 (0.000) Number of observations: left = 41,121 / right = 40,512 Number of observations: left = 55,770 / right = 12,006 .02 .02 .015 .015 .01 .01 .005 .005 Average income tax to turnover ratio Average turnover tax to ratio 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR)

Notes: The figure plots the average fiscal adjustment to turnover ratio (Panel A and B) as well as average tax liability to turnover ratio (Panel C and D) near the threshold after (2014-2017) the turnover tax reform and the VAT threshold expansion. The vertical solid line is the normalized threshold. The number of bins used to the left of the threshold is 50 and to the right of the threshold is also 50. The solid lines in the left and the right of the threshold represent the conditional expectation function for firms below and above the threshold. Difference at the threshold d is the size of discontinuity of the dependent variable at the normalized threshold.

The VAT threshold expansion in 2014 introduces new discontinuities at the 4.8 billion IDR threshold such as VAT rates and compliance costs which complicate the estimation of taxpayers responses to the turnover tax around the notch. Figure 9 shows the emergence of discountinuities in direct input ratio after the expansion of VAT threshold in 2014. Panel A and B show that ﬁrms just below the nocth that will stay in the turnover tax regime in the following year have around 7.1 percent to 9.8 percent less direct input to turnover ratio relative to those just above the notch that will stay in the next year income tax regime. These discontinuities can be thought to be mainly driven by the expansion of VAT threshold

24 in 2014 because there is no discontinuity of average direct input to turnover ratio in 2012 and 2013 (see Figure 8). Moreover, panel C shows that firms in the income tax regime just below the threshold have similar average income tax to turnover ratio relative to firms just above the notch. Panel D shows no discontinuity of tax ratio because firms in the turnover tax regime are taxed at a single final tax rate on turnover. The results suggest that firms have excessively under reported direct inputs to avoid VAT registration. This excessive inputs misreporting by firms can be explained by the fact that firms do not have to face the consequence of having higher income tax from under expensing inputs when they already in the turnover tax regime. This response is also consistent with the assumption that firms obtain significantly larger benefits from under reporting direct input in the VAT system than the additional income tax burden from under expensing inputs.8

5.5 Estimates for Compliance Costs, Beneﬁts and Tax Elasticity

I estimate the VAT compliance costs, income tax compliance benefits and tax elasticities using models outlined in Section 3 and behavioral responses estimated in Section 5. I independently solve equation (5) in the case of single threshold and simultaneously solve equations 10 and 14 in the case of multiple thresholds to recover the estimates using observed A B ∆y∗ value of α, γ , γ , and the estimates of the marginal buncher y∗ . Table 2 presents the estimates for compliance costs, compliance benefits and tax elasticity with respect to reported value added. The table shows that the compliance benefits from the income tax system obtained by firms near the notch is around 11.5 percent of the reported value added. The tax elasticity in the single notch in 2014 is relatively large (0.724) when compliance costs are excluded in the estimation (η = 0). However, the tax elasitic- ity decreases to 0.651 when I consider compliance benefits as additional driver of buncing responses. The results suggest that compliance benefits motivate firms around the notch under the single threshold condition to bunch above the notch and avoid the turnover tax regime. I also find that the compliance costs and compliance benefits are about 8.9 and 14.8 percent of the value added respectively. With direct input ratio (α) of about 0.7, the compliance costs is around 12.7 percent of turnover. This estimate is similar to the previous

8The real response, instead of evasion response, is documented around the previous (2003) VAT threshold. Figure B2 in Appendix B shows that the distributions of direct input ratio from year 2007 to 2012 relatively continue before and after the notch. This trend indicates that firms that bunch just below the notch proportionally under report both turnover and direct input to avoid VAT registration (see Figure B4). In contrast, firms near the 2014 VAT threshold disproportionately under report direct input. The results suggest that VAT threshold might induce real responses from smaller-sized firms but create evasion responses from larger-sized firms.

25 Table 2: Compliance Beneﬁts and Tax Elasticity Estimates

Compliance Compliance Tax elasticity (e) costs (θ) benefits (η) θ > 0 and η > 0 θ = 0 and η = 0 Single tax notch (2013) - 0.115 0.651 0.725 (0.061) (0.0105) (0.0179) Multiple tax notches (2014-2017) 0.089 0.148 0.879 1.183 (0.000) (0.035) (0.000) (0.0657) Notes: the table presents the estimates for tax elasticity and compliance benefits. To obtain the estimates, A B ∆y∗ I solve conditions in equations (5), (10), and (14) by using values for α, γ , γ , and y∗ . estimates in Susila and Pope (2012) that find the VAT compliance costs are 3 percent of turnover for firms with turnover above 3 billion IDR and 11.2 percent of turnover for firm with turnover of up to 3 billion IDR. Moreover, the compliance benefits estimate from the income tax under multiple threshold is slightly larger than that of the single threshold. This discrepancy is driven by movement of firms with smaller compliance benefits that bunch above the threshold in 2013 from above to below the threshold after the VAT registration is expanded and discontinuity in VAT compliance costs emerge in 2014. Finally, I find the tax elasticity under multiple thresholds is about 0.879 and is larger than that of the single threshold. It may suggest that firms are more responsive to the VAT system relative to the income tax. Moreover, the tax elasticity increases to 1.183 if the compliance costs and compliance benefits are absent in the estimation. The results suggest that firms respond not only to change in tax incentives but also to change in administrative incentives.

6 Conclusion

Governments implement simple tax regime to small businesses in order to reduce compliance costs which particularly expensive for them. This type of policy often requires a certain eligibility threshold. This threshold creates discontinuity in tax incentives. Previous works show that tax rates and compliance costs are essential driver in the behavioral responses to the discontinuity. However, few works has been focused on moving from income tax system to turnover tax system where discontinuity occurs not only in tax rates and compliance costs but also in benefits from tax avoidance. This paper studies the behavioral responses to the introduction of turnover tax system. I find that around the eligibility threshold, some firms prefers to stay in income tax system due

26 to benefits from tax avoidance. Moreover, after the VAT registration threshold expansion, some other firms decide to move to turnover tax regime because they do not get larger benefit from tax avoidance relative to VAT burden and related compliance costs. I also find some evasion responses from firms just below the threshold by over reporting direct input after the VAT registration threshold expansion. I document that tax elasticity is larger when compliance costs and benefits are excluded, suggesting the importance of non-tax incentive drivers of behavioral responses. I further show that tax elasticity increases due to inclusion of VAT incentive, indicating that firms are more responsive to VAT than to income taxes. The findings suggest that income tax system does not only levy taxes and compliance costs but also offers private benefits from tax avoidance. However, the results also indicate inefficiency because firms have deviated their resources from productive activities toward toward tax planning activities. The findings consistent with the literature that turnover tax reduce tax avoidance and improves efficiency. Finally, the findings imply that governments should consider both tax and non-tax incentives in formulating tax policy, in particular for small businesses.

27 References

Benedek, M. D., De Mooij, R. A., and Wingender, M. P. (2015). Estimating VAT pass through. Number 15-214. International Monetary Fund.

Benzarti, Y., Carloni, D., Harju, J., and Kosonen, T. (2017). What goes up may not come down: asymmetric incidence of value-added taxes. Technical report, National Bureau of Economic Research.

Besley, T. and Persson, T. (2014). Why do developing countries tax so little? Journal of Economic Perspectives, 28(4):99–120.

Bird, R., Wallace, S., et al. (2003). Is it really so hard to tax the hard-to-tax? the context and role of presumptive taxes. International Tax Program Papers, (0307).

Bruhn, M. and Loeprick, J. (2016). Small business tax policy and informality: evidence from georgia. International Tax and Public Finance, 23(5):834–853.

Calonico, S., Cattaneo, M. D., and Titiunik, R. (2014). Robust nonparametric conﬁdence intervals for regression-discontinuity designs. Econometrica, 82(6):2295–2326.

Carbonnier, C. (2007). Who pays sales taxes? evidence from french vat reforms, 1987–1999. Journal of Public Economics, 91(5-6):1219–1229.

Chetty, R., Friedman, J. N., Olsen, T., and Pistaferri, L. (2011). Adjustment costs, ﬁrm responses, and micro vs. macro labor supply elasticities: Evidence from danish tax records. The quarterly journal of economics, 126(2):749–804.

Desai, M. A. and Dharmapala, D. (2006). Corporate tax avoidance and high-powered incentives. Journal of Financial Economics, 79(1):145–179.

Desai, M. A. and Dharmapala, D. (2009). Corporate tax avoidance and ﬁrm value. The review of Economics and Statistics, 91(3):537–546.

Engelschalk, M. (2005). Small business taxation in transition countries. Tax notes international, 40(6):523.

Fajnzylber, P., Maloney, W. F., and Montes-Rojas, G. V. (2011). Does formality improve micro-ﬁrm performance? evidence from the brazilian simples program. Journal of Devel- opment Economics, 94(2):262–276.

28 Gordon, R. and Li, W. (2009). Tax structures in developing countries: Many puzzles and a possible explanation. Journal of public Economics, 93(7-8):855–866.

Harju, J., Kosonen, T., and Skans, O. N. (2018). Firm types, price-setting strategies, and consumption-tax incidence. Journal of Public Economics, 165:48–72.

Harju, J., Matikka, T., and Rauhanen, T. (2019). Compliance costs vs. tax incentives: Why do entrepreneurs respond to size-based regulations? Journal of Public Economics, 173:139–164.

Johnston, K. S. (1963). Corporations’ federal income tax compliance costs: a study of small, medium-size, and large corporations. Number 110. Bureau of Business Research, College of Commerce and Administration, Ohio . . . .

Kanbur, R. and Keen, M. (2014). Thresholds, informality, and partitions of compliance. International Tax and Public Finance, 21(4):536–559.

Keen, M. and Mintz, J. (2004). The optimal threshold for a value-added tax. Journal of Public Economics, 88(3-4):559–576.

Kleven, H. J. (2016). Bunching. Annual Review of Economics, 8:435–464.

Kleven, H. J. and Waseem, M. (2013). Using notches to uncover optimization frictions and structural elasticities: Theory and evidence from pakistan. The Quarterly Journal of Economics, 128(2):669–723.

Kosonen, T. (2015). More and cheaper haircuts after vat cut? on the eﬃciency and incidence of service sector consumption taxes. Journal of Public Economics, 131:87–100.

Liu, L. and Lockwood, B. (2016). Vat notches, voluntary registration and bunching: Theory and uk evidence.

Saez, E. (2010). Do taxpayers bunch at kink points? American economic Journal: economic policy, 2(3):180–212.

Slemrod, J. and Yitzhaki, S. (1994). Analyzing the standard deduction as a presumptive tax. International Tax and Public Finance, 1(1):25–34.

Slemrod, J. B. and Blumenthal, M. (1996). The income tax compliance cost of big business. Public ﬁnance quarterly, 24(4):411–438.

Susila, B. and Pope, J. (2012). The magnitude and the features of tax compliance costs of large companies in indonesia. In Australian Tax Forum, volume 27, pages 719–772.

29 Tanzi, V. and de Jantscher, M. C. (1986). The use of presumptive income in modern tax systems.

Tran-Nam, B., Evans, C., Walpole, M., and Ritchie, K. (2000). Tax compliance costs: Research methodology and empirical evidence from australia. National Tax Journal, pages 229–252.

Waseem, M. (2018). Taxes, informality and income shifting: Evidence from a recent pakistani tax reform. Journal of public economics, 157:41–77.

30 Appendix A

Figure A1: Bunching in the Income Tax Regime by Year

Panel A: 2014 Panel B: 2015

Excess bunching (b) = 1.092 (0.227) Excess bunching (b) = 1.201 (0.174) 1000 1000 800 800 600 600 400 400 Number of firms Number of firms 200 200 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold in million IDR Distance to the threshold in million IDR

Observed Counterfactual Observed Counterfactual

Panel C: 2016 Panel D: 2017

Excess bunching (b) = 0.781 (0.283) Excess bunching (b) = 0.608 (0.405) 1000 1000 800 800 600 600 400 400 Number of firms Number of firms 200 200 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold in million IDR Distance to the threshold in million IDR

Observed Counterfactual Observed Counterfactual

Notes: The ﬁgure shows the observed distribution (solid-dotted line) and the estimated counterfactual (solid- dashed line) of annual turnover after the VAT registration threshold expansion (2014-2017) in bins of 100 million IDR within a range of -2000 million IDR and 2000 million IDR. The threshold is represented by a vertical solid line. The counterfactual is a ﬁfth-order polynomial estimated as in equation (16). The shaded area is the excess mass in the upper part of the excluded region. Excess bunching b is excess bunching in the omitted region near the threshold relative to the average counterfactual count in this region.

31 Figure A2: Bunching in the Turnover Tax Regime by Year

Panel A: 2014 Panel B: 2015

Excess bunching (b) = 0.635 (0.199) Excess bunching (b) = 0.82 (0.374) 1000 1000 800 800 600 600 400 400 Number of firms Number of firms 200 200 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold in million IDR Distance to the threshold in million IDR

Observed Counterfactual Observed Counterfactual

Panel C: 2016 Panel D: 2017

Excess bunching (b) = 1.238 (0.594) Excess bunching (b) = 1.707 (0.616) 1000 1000 800 800 600 600 400 400 Number of firms Number of firms 200 200 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold in million IDR Distance to the threshold in million IDR

Observed Counterfactual Observed Counterfactual

Notes: The ﬁgure shows the observed distribution (solid-dotted line) and the estimated counterfactual (solid- dashed line) of annual turnover after the turnover tax reform (2014-2017) in bins of 100 million IDR within a range of -2000 million IDR and 2000 million IDR. The threshold is represented by a vertical solid line. The counterfactual is a ﬁfth-order polynomial estimated as in equation (16). The shaded area is the excess mass in the upper part of the excluded region. Excess bunching b is excess bunching in the omitted region near the threshold relative to the average counterfactual count in this region.

32 Figure A3: Excess Bunching 2014-2017

Panel A: Income Tax Regime Panel B: Turnover Tax Regime 3 3 2 2 1 1 Excess bunching Excess bunching 0 0 −1 −1 2013 2014 2015 2016 2017 2013 2014 2015 2016 2017 Fiscal year Fiscal year

Notes: The ﬁgure displays the yearly excess bunching with 95 percent conﬁdence intervals from 2013 to 2017. Panel A shows that the excess bunching in the income tax regime due to turnover tax threshold was stable from 2013 to 2015 but has been trending down afterward. Panel B shows that the excess bunching in the turnover tax regime due to VAT threshold has been increasing since 2014.

33 Table A1: Excess Bunching by Counterfactual’s Degree of Polynomial

Degree of Polynomial 3rd 5th 7th A. Single threshold (2013) 1.370 1.173 1.163 (0.354) (0.359) (0.356) B. Multiple thresholds (2014-2017) B.1 Income tax regime 2014 1.434 1.092 1.081 (0.491) (0.227) (0.224) 2015 1.198 1.201 1.167 (0.405) (0.174) (0.169) 2016 1.213 0.781 0.751 (0.377) (0.283) (0.277) 2017 0.736 0.608 0.577 (0.422) (0.405) (0.396) B.2 Turnover tax regime 2014 1.011 0.635 0.573 (0.406) (0.199) (0.197) 2015 1.049 0.82 0.846 (0.414) (0.374) (0.318) 2016 1.519 1.238 1.272 (0.585) (0.594) (0.470) 2017 1.626 1.707 1.405 (0.626) (0.616) (0.583) Notes: The table displays the average excess mass b(y∗) estimated in equation (18) by counterfactual’s degree of polynomial in each ﬁscal year after the reforms. In general, the estimates shrink as the degree of polynomials go up, yet converge at the 7th degree polynomial.

34 Figure A4: Fiscal Adjustments at Multiple Thresholds by Year

Panel A: All Firms 2014 Panel B: All Firms 2015

Difference at the threshold d = −0.109 (0.017) Difference at the threshold d = −0.139 (0.014) Number of observations: left = 18,566 / right = 12,283 Number of observations: left = 22,609 / right = 14,115 .35 .35 .3 .3 .25 .25 .2 .2 .15 .15 .1 .1 .05 .05 Average fiscal adjustment to turnover ratio Average fiscal adjustment to turnover ratio 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR) Panel C: All Firms 2016 Panel D: All Firms 2017

Difference at the threshold d = −0.129 (0.013) Difference at the threshold d = −0.104 (0.012) Number of observations: left = 26,464 / right = 14,688 Number of observations: left = 29,150 / right = 14,799 .35 .35 .3 .3 .25 .25 .2 .2 .15 .15 .1 .1 .05 .05 Average fiscal adjustment to turnover ratio Average fiscal adjustment to turnover ratio 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR)

Notes: The figure shows the plot of average fiscal adjustment to turnover ratio near the threshold after the turnover tax reform and VAT registration threshold expansion (2014-2017). The vertical solid line is the normalized threshold. The number of bins used to the left of the threshold is 50 and to the right of the threshold is also 50. The solid lines in the left and the right of the threshold represent the conditional expectation function for firms below and above the threshold. Difference at the threshold d is the size of discontinuity of the dependent variable at the normalized threshold.

35 Figure A5: Bunching in the Income Tax Regime by Industry

Panel A: Retail and Wholesale Trade Panel B: Manufacturing

Excess bunching (b) = 1.275 (0.374) Excess bunching (b) = 1.525 (0.456) 250 1400 1200 200 1000 150 800 600 100 Number of firms Number of firms 400 50 200 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold in million IDR Distance to the threshold in million IDR

Observed Counterfactual Observed Counterfactual

Panel C: Construction Panel D: Transportation

Excess bunching (b) = 0.771 (0.304) Excess bunching (b) = 1.274 (0.341) 140 400 120 300 100 80 200 60 Number of firms Number of firms 40 100 20 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold in million IDR Distance to the threshold in million IDR

Observed Counterfactual Observed Counterfactual

Panel E: Agriculture, Forestry, and Fishery Panel F: Accommodation and Food Services

Excess bunching (b) = 1.132 (0.331) Excess bunching (b) = 2.042 (0.426) 30 80 25 60 20 15 40 10 Number of firms Number of firms 20 5 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold in million IDR Distance to the threshold in million IDR

Observed Counterfactual Observed Counterfactual

Notes: The ﬁgure shows the observed distribution (solid-dotted line) and the estimated counterfactual (solid- dashed line) of annual turnover after the turnover tax reform and VAT registration threshold expansion (2014-2017) in bins of 100 million IDR within a range of -2000 million IDR and 2000 million IDR. The threshold is represented by a vertical solid line. The counterfactual is a ﬁfth-order polynomial estimated as in equation (16). The shaded area is the excess mass in the upper part of the excluded region. Excess bunching b is excess bunching in the omitted region near the threshold relative to the average counterfactual count in this region.

36 Figure A6: Bunching in the Turnover Tax Regime by Industry

Panel A: Retail and Wholesale Trade Panel B: Manufacturing

Excess bunching (b) = 0.943 (0.511) Excess bunching (b) = 1.003 (0.294) 2500 400 2000 300 1500 200 1000 Number of firms Number of firms 100 500 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold in million IDR Distance to the threshold in million IDR

Observed Counterfactual Observed Counterfactual

Panel C: Construction Panel D: Transportation

Excess bunching (b) = 0.635 (0.079) Excess bunching (b) = 1.399 (0.612) 300 200 150 200 100 Number of firms Number of firms 100 50 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold in million IDR Distance to the threshold in million IDR

Observed Counterfactual Observed Counterfactual

Panel E: Agriculture, Forestry, and Fishery Panel F: Accommodation and Food Services

Excess bunching (b) = 1.068 (0.491) Excess bunching (b) = 1.177 (0.287) 40 150 30 100 20 Number of firms Number of firms 50 10 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold in million IDR Distance to the threshold in million IDR

Observed Counterfactual Observed Counterfactual

Notes: The ﬁgure shows the observed distribution (solid-dotted line) and the estimated counterfactual (solid- dashed line) of annual turnover after the turnover tax reform and VAT registration threshold expansion (2014-2017) in bins of 100 million IDR within a range of -2000 million IDR and 2000 million IDR. The threshold is represented by a vertical solid line. The counterfactual is a ﬁfth-order polynomial estimated as in equation (16). The shaded area is the excess mass in the upper part of the excluded region. Excess bunching b is excess bunching in the omitted region near the threshold relative to the average counterfactual count in this region.

37 Figure A7: Fiscal Adjustments at Multiple Thresholds by Industry

Panel A: Retail and Wholesale Trade Panel B: Manufacturing

Difference at the threshold d = −0.117 (0.009) Difference at the threshold d = −0.149 (0.022) Number of observations: left = 55,795 / right = 32,091 Number of observations: left = 8,347 / right = 5,311 .4 .4 .35 .35 .3 .3 .25 .25 .2 .2 .15 .15 .1 .1 .05 .05 Average fiscal adjustment to turnover ratio Average fiscal adjustment to turnover ratio 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR) Panel C: Construction Panel D: Transportation

Difference at the threshold d = −0.037 (0.017) Difference at the threshold d = −0.129 (0.024) Number of observations: left = 24,125 / right = 10,330 Number of observations: left = 4,776 / right = 2,854 .4 .4 .35 .35 .3 .3 .25 .25 .2 .2 .15 .15 .1 .1 .05 .05 Average fiscal adjustment to turnover ratio Average fiscal adjustment to turnover ratio 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR) Panel E: Agriculture, Forestry, and Fishery Panel F: Accommodation and Food Services

Difference at the threshold d = −0.169 (0.079) Difference at the threshold d = −0.176 (0.044) Number of observations: left = 969 / right = 610 Number of observations: left = 2,846 / right = 1,784 .4 .4 .35 .3 .2 .25 .2 0 .15 .1 −.2 .05 Average fiscal adjustment to turnover ratio Average fiscal adjustment to turnover ratio 0 −.4 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR)

38 Figure A8: Bunching in the Income Tax Regime by Type of Business

Panel A: Corporation Panel B: Partnership

Excess bunching (b) = 1.070 (0.263) Excess bunching (b) = 1.421 (0.504) 1400 1400 1200 1200 1000 1000 800 800 600 600 Number of firms Number of firms 400 400 200 200 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold in million IDR Distance to the threshold in million IDR

Observed Counterfactual Observed Counterfactual

Notes: The ﬁgure shows the observed distribution (solid-dotted line) and the estimated counterfactual (solid- dashed line) of annual turnover after the turnover tax reform and VAT registration threshold expansion (2014-2017) in bins of 100 million IDR within a range of -2000 million IDR and 2000 million IDR. The threshold is represented by a vertical solid line. The counterfactual is a ﬁfth-order polynomial estimated as in equation (16). The shaded area is the excess mass in the upper part of the excluded region. Excess bunching b is excess bunching in the omitted region near the threshold relative to the average counterfactual count in this region.

Figure A8: Bunching in the Turnover Tax Regime by Type of Business

Panel A: Corporation Panel B: Partnership

Excess bunching (b) = 1.102 (0.494) Excess bunching (b) = 1.245 (0.494) 2500 2500 2000 2000 1500 1500 1000 1000 Number of firms Number of firms 500 500 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold in million IDR Distance to the threshold in million IDR

Observed Counterfactual Observed Counterfactual

Notes: The ﬁgure shows the observed distribution (solid-dotted line) and the estimated counterfactual (solid- dashed line) of annual turnover after the turnover tax reform and VAT registration threshold expansion (2014-2017) in bins of 100 million IDR within a range of -2000 million IDR and 2000 million IDR. The threshold is represented by a vertical solid line. The counterfactual is a ﬁfth-order polynomial estimated as in equation (16). The shaded area is the excess mass in the upper part of the excluded region. Excess bunching b is excess bunching in the omitted region near the threshold relative to the average counterfactual count in this region.

39 Figure A9: Fiscal Adjustments at Multiple Thresholds by Type of Business

Panel A: Corporation Panel B: Partnership

Difference at the threshold d = −0.137 (0.009) Difference at the threshold d = −0.125 (0.011) Number of observations: left = 54,498 / right = 35,321 Number of observations: left = 34,138 / right = 16,131 .4 .4 .35 .35 .3 .3 .25 .25 .2 .2 .15 .15 .1 .1 .05 .05 Average fiscal adjustment to turnover ratio Average fiscal adjustment to turnover ratio 0 0 −2000 −1000 0 1000 2000 −2000 −1000 0 1000 2000 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR)

40 Appendix B

I further study the impacts of VAT registration threshold by comparing the behavioral responses at the old (2003) and the new (2014) thresholds. The threshold is increased fourfold from 600 million to 4.8 billion IDR. Thus, firms around the old threshold have smaller size than the those around the new threshold and might respond the discontinuity differently. Similar to the new threshold, income tax notch occurs at the old threshold where firms with turnover of up to 600 billion IDR can select between presumptive tax (deemed-profit-tax) and income tax. The difference is that the turnover tax offers significantly smaller average tax rate relative to deemed-profit tax scheme. Figure B1 displays bunching responses at the old and the new thresholds. The excees bunching at the new VAT notch is similar that of the new threshold. This might suggest that large firms are not more responsive to the VAT notch than small firms. On the other hand, unlike the new VAT threshold, the previous threshold creates no discontinuity in direct input to turnover ratio as demonstrated in Figure B4. The results suggest that the old VAT threshold persuade real responses rather than evasion responses (misreporting inputs) as in the new VAT notch. The differences in the size of annual turnover between firms around the old and those near the new threshold might explain this heterogeneity.

Figure B1: Bunching at the VAT Registration Thresholds

Panel A: Old Threshold (2007-2012) Panel B: New Threshold (2014-2017)

Excess bunching (b) = 0.937 (0.375) Excess bunching (b) = 1.072 (0.494) 12000 4000 10000 3000 8000 6000 2000 Number of firms Number of firms 4000 1000 2000 0 0 −400 −200 0 200 400 −2000 −1000 0 1000 2000 Distance to the threshold in million IDR Distance to the threshold in million IDR

Observed Counterfactual Observed Counterfactual

Notes: The ﬁgure displays the observed distribution (solid-dotted line) and the estimated counterfactual (solid-dashed line) of annual turnover. Panel A shows the turnover distribution at the 2003 VAT registration threshold in bins of 10 million IDR within a range of -400 million IDR and 400 million IDR. Panel B shows the turnover distribution at the 2014 VAT registration threshold in bins of 100 million IDR within a range of -2000 million IDR and 2000 million IDR. The threshold is represented by a vertical solid line. The counterfactual is a ﬁfth-order polynomial estimated as in equation (16). The shaded area is the excess mass in the upper part of the excluded region. Excess bunching b is excess bunching in the omitted region near the threshold relative to the average counterfactual count in this region.

41 Figure B2: Bunching at the 2003 VAT Registration Threshold (2007-2012)

Panel A: 2007 Panel B: 2008

Excess bunching (b) = 0.508 (0.160) Excess bunching (b) = 0.394 (0.145) 3000 3000 2500 2500 2000 2000 1500 1500 Number of firms Number of firms 1000 1000 500 500 0 0 −400 −200 0 200 400 −400 −200 0 200 400 Distance to the threshold in million IDR Distance to the threshold in million IDR

Observed Counterfactual Observed Counterfactual

Panel C: 2009 Panel D: 2010

Excess bunching (b) = 0.500 (0.081) Excess bunching (b) = 0.583 (0.137) 3000 3000 2500 2500 2000 2000 1500 1500 Number of firms Number of firms 1000 1000 500 500 0 0 −400 −200 0 200 400 −400 −200 0 200 400 Distance to the threshold in million IDR Distance to the threshold in million IDR

Observed Counterfactual Observed Counterfactual

Panel E: 2011 Panel F: 2012

Excess bunching (b) = 0.583 (0.192) Excess bunching (b) = 0.744 (0.264) 3000 3000 2500 2500 2000 2000 1500 1500 Number of firms Number of firms 1000 1000 500 500 0 0 −400 −200 0 200 400 −400 −200 0 200 400 Distance to the threshold in million IDR Distance to the threshold in million IDR

Observed Counterfactual Observed Counterfactual

Notes: The ﬁgure shows the observed distribution (solid-dotted line) and the estimated counterfactual (solid-dashed line) of annual turnover before the VAT reform (2007-2012) in bins of 10 million IDR within a range of -400 million IDR and 400 million IDR. The threshold is represented by a vertical solid line. The counterfactual is a ﬁfth-order polynomial estimated as in equation (16). The shaded area is the excess mass in the upper part of the excluded region. Excess bunching b is excess bunching in the omitted region near the threshold relative to the average counterfactual count in this region.

42 Figure B3: Bunching at the 2003 VAT Registration Threshold (2013-2017)

Panel A: 2013 Panel B: 2014

Excess bunching (b) = 0.526 (0.227) Excess bunching (b) = 0.107 (0.043) 3000 3000 2500 2500 2000 2000 1500 1500 Number of firms Number of firms 1000 1000 500 500 0 0 −400 −200 0 200 400 −400 −200 0 200 400 Distance to the threshold in million IDR Distance to the threshold in million IDR

Observed Counterfactual Observed Counterfactual

Panel C: 2015 Panel D: 2016

Excess bunching (b) = 0.041 (0.128) Excess bunching (b) = 0.031 (0.095) 3000 3000 2500 2500 2000 2000 1500 1500 Number of firms Number of firms 1000 1000 500 500 0 0 −400 −200 0 200 400 −400 −200 0 200 400 Distance to the threshold in million IDR Distance to the threshold in million IDR

Observed Counterfactual Observed Counterfactual

Panel E: 2017

Excess bunching (b) = 0.093 (0.016) 3000 2500 2000 1500 Number of firms 1000 500 0 −400 −200 0 200 400 Distance to the threshold in million IDR

Observed Counterfactual

43 Figure B4: Direct Input Ratio at the 2003 VAT Registration Threshold (2007-2012)

Panel A: 2007 Panel B: 2008

Difference at the threshold d = 0.018 (0.012) Difference at the threshold d = −0.005 (0.015) Number of observations: left = 44,341 / right = 19,084 Number of observations: left = 48,886 / right = 20,682 .8 .8 .75 .75 .7 .7 .65 .65 .6 .6 .55 .55 Average direct input to turnover ratio Average direct input to turnover ratio .5 .5 −400 −200 0 200 400 −400 −200 0 200 400 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR) Panel C: 2009 Panel D: 2010

Difference at the threshold d = 0.011 (0.013) Difference at the threshold d = 0.039 (0.012) Number of observations: left = 50,275 / right = 21,515 Number of observations: left = 53,912 / right = 23,820 .8 .8 .75 .75 .7 .7 .65 .65 .6 .6 .55 .55 Average direct input to turnover ratio Average direct input to turnover ratio .5 .5 −400 −200 0 200 400 −400 −200 0 200 400 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR) Panel E: 2011 Panel F: 2012

Difference at the threshold d = 0.012 (0.012) Difference at the threshold d = 0.021 (0.011) Number of observations: left = 53,593 / right = 24,127 Number of observations: left = 59,065 / right = 27,835 .8 .8 .75 .75 .7 .7 .65 .65 .6 .6 .55 .55 Average direct input to turnover ratio Average direct input to turnover ratio .5 .5 −400 −200 0 200 400 −400 −200 0 200 400 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR)

Notes: The figure plots the average direct input to turnover ratio near the 2003 VAT threshold. The vertical solid line is the normalized threshold. The number of bins used to the left of the threshold is 50 and to the right of the threshold is also 50. The solid lines in the left and the right of the threshold represent the conditional expectation function for firms below and above the threshold. Difference at the threshold d is the size of discontinuity of the dependent variable at the normalized threshold.

44 Figure B5: Direct Input Ratio at the 2003 VAT Registration Threshold (2013-2017)

Panel A: 2013 Panel B: 2014

Difference at the threshold d = 0.027 (0.012) Difference at the threshold d = 0.005 (0.015) Number of observations: left = 56,194 / right = 28,188 Number of observations: left = 57,682 / right = 31,064 .8 .8 .75 .75 .7 .7 .65 .65 .6 .6 .55 .55 Average direct input to turnover ratio Average direct input to turnover ratio .5 .5 −400 −200 0 200 400 −400 −200 0 200 400 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR) Panel C: 2015 Panel D: 2016

Difference at the threshold d = −0.006 (0.014) Difference at the threshold d = −0.015 (0.015) Number of observations: left = 65,505 / right = 35,559 Number of observations: left = 67,391 / right = 37,288 .8 .8 .75 .75 .7 .7 .65 .65 .6 .6 .55 .55 Average direct input to turnover ratio Average direct input to turnover ratio .5 .5

−400 −200 0 200 400 −400 −200 0 200 400 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR) Panel E: 2017

Difference at the threshold d = −0.027 (0.015) Number of observations: left = 65,808 / right = 36,905 .8 .75 .7 .65 .6 .55 Average direct input to turnover ratio .5

−400 −200 0 200 400 Distance to the threshold (Million IDR)

45 Figure B6: Tax Liabilities at the 2003 VAT Registration Threshold (2007-2012)

Panel A: 2007 Panel B: 2008

Difference at the threshold d = 0.00066 (0.00058) Difference at the threshold d = −0.000009 (0.00008) Number of observations: left = 44,332 / right = 19,082 Number of observations: left = 48,878 / right = 20,679 .015 .015 .01 .01 .005 .005 Average income tax to turnover ratio Average income tax to turnover ratio 0 0 −400 −200 0 200 400 −400 −200 0 200 400 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR) Panel C: 2009 Panel D: 2010

Difference at the threshold d = 0.00072 (0.00051) Difference at the threshold d = 0.00033 (0.00046) Number of observations: left = 50,272 / right = 21,514 Number of observations: left = 53,903 / right = 23,818 .015 .015 .01 .01 .005 .005 Average income tax to turnover ratio Average income tax to turnover ratio 0 0 −400 −200 0 200 400 −400 −200 0 200 400 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR) Panel E: 2011 Panel F: 2012

Difference at the threshold d = −0.00091 (0.00054) Difference at the threshold d = −0.0004 (0.00042) Number of observations: left = 53,584 / right = 24,127 Number of observations: left = 59,057 / right = 27,832 .015 .015 .01 .01 .005 .005 Average income tax to turnover ratio Average income tax to turnover ratio 0 0 −400 −200 0 200 400 −400 −200 0 200 400 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR)

Notes: The figure plots the average tax liability to turnover ratio near the 2003 VAT threshold. The vertical solid line is the normalized threshold. The number of bins used to the left of the threshold is 50 and to the right of the threshold is also 50. The solid lines in the left and the right of the threshold represent the conditional expectation function for firms below and above the threshold. Difference at the threshold d is the size of discontinuity of the dependent variable at the normalized threshold.

46 Figure B7: Tax Liabilities at the 2003 VAT Registration Threshold (2013-2017)

Panel A: 2013 Panel B: 2014

Difference at the threshold d = −0.00022 (0.0003) Difference at the threshold d = −0.000019 (0.00027) Number of observations: left = 56,179 / right = 28,114 Number of observations: left = 57,675 / right = 31,062 .015 .015 .01 .01 .005 .005 Average income tax to turnover ratio Average income tax to turnover ratio 0 0 −400 −200 0 200 400 −400 −200 0 200 400 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR) Panel C: 2015 Panel D: 2016

Difference at the threshold d = −0.00021 (0.00027) Difference at the threshold d = 0.00054 (0.00041) Number of observations: left = 65,495 / right = 35,554 Number of observations: left = 67,383 / right = 37,288 .015 .015 .01 .01 .005 .005 Average income tax to turnover ratio Average income tax to turnover ratio 0 0 −400 −200 0 200 400 −400 −200 0 200 400 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR) Panel E: 2017

Difference at the threshold d = 0.00027 (0.00031) Number of observations: left = 65,802 / right = 36,901 .015 .01 .005 Average income tax to turnover ratio 0 −400 −200 0 200 400 Distance to the threshold (Million IDR)

47 Figure B8: Fiscal Adjustments at the 2003 VAT Registration Threshold (2007-2012)

Panel A: 2007 Panel B: 2008

Difference at the threshold d = −0.005 (0.005) Difference at the threshold d = 0.003 (0.004) Number of observations: left = 44,206 / right = 19,026 Number of observations: left = 48,698 / right = 20,599 .05 .05 .04 .04 .03 .03 .02 .01 .02 0 .01 −.01 Average fiscal adjustment to turnover ratio Average fiscal adjustment to turnover ratio −.02 0 −400 −200 0 200 400 −400 −200 0 200 400 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR) Panel C: 2009 Panel D: 2010

Difference at the threshold d = 0.002 (0.005) Difference at the threshold d = 0.009 (0.007) Number of observations: left = 50,045 / right = 21,415 Number of observations: left = 53,663 / right = 23,705 .05 .05 .04 .04 .03 .03 .02 .02 .01 .01 Average fiscal adjustment to turnover ratio Average fiscal adjustment to turnover ratio 0 0 −400 −200 0 200 400 −400 −200 0 200 400 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR) Panel E: 2011 Panel F: 2012

Difference at the threshold d = 0.005 (0.004) Difference at the threshold d = 0.003 (0.004) Number of observations: left = 53,312 / right = 24,011 Number of observations: left = 58,699 / right = 27,658 .05 .05 .04 .04 .03 .03 .02 .02 .01 .01 Average fiscal adjustment to turnover ratio Average fiscal adjustment to turnover ratio 0 0 −400 −200 0 200 400 −400 −200 0 200 400 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR)

Notes: The figure plots the average fiscal adjustment to turnover ratio near the 2003 VAT threshold. The vertical solid line is the normalized threshold. The number of bins used to the left of the threshold is 50 and to the right of the threshold is also 50. The solid lines in the left and the right of the threshold represent the conditional expectation function for firms below and above the threshold. Difference at the threshold d is the size of discontinuity of the dependent variable at the normalized threshold.

48 Figure B9: Fiscal Adjustments at the 2003 VAT Registration Threshold (2013-2017)

Panel A: 2013 Panel B: 2014

Difference at the threshold d = −0.033 (0.008) Difference at the threshold d = 0.000 (0.011) Number of observations: left = 55,332 / right = 27,764 Number of observations: left = 55,310 / right = 29,922 .16 .24 .22 .14 .2 .12 .18 .1 .16 Average fiscal adjustment to turnover ratio Average fiscal adjustment to turnover ratio .08 .14 −400 −200 0 200 400 −400 −200 0 200 400 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR) Panel C: 2015 Panel D: 2016

Difference at the threshold d = 0.007 (0.01) Difference at the threshold d = 0.019 (0.008) Number of observations: left = 62,479 / right = 34,102 Number of observations: left = 64,025 / right = 35,556 .24 .22 .22 .2 .2 .18 .18 .16 .16 .14 Average fiscal adjustment to turnover ratio Average fiscal adjustment to turnover ratio .14 −400 −200 0 200 400 −400 −200 0 200 400 Distance to the threshold (Million IDR) Distance to the threshold (Million IDR) Panel E: 2017

Difference at the threshold d = 0.000 (0.009) Number of observations: left = 62,507 / right = 35,290 .22 .2 .18 .16 .14 Average fiscal adjustment to turnover ratio

−400 −200 0 200 400 Distance to the threshold (Million IDR)