Corporate Taxes and Firm Behavior - Evidence from South Africa

Transcription

1 Corporate Taxes and Firm Behavior - Evidence from South Africa Collen Lediga Nadine Riedel Kristina Strohmaier University of Bochum University of Bochum University of Bochum CESifo Munich, DIW Berlin University of Hohenheim Oxford University CBT PRELIMINARY AND INCOMPLETE - PLEASE DO NOT QUOTE Abstract The aim of this paper is to empirically investigate firm behavior in response to corporate taxation and administrative deterrence instruments (tax payer auditing) in a low enforcement environment. Using the population of corporate tax returns in South Africa for , we in a first step analyze bunching in the distribution of taxable income at kinks and notches in the tax schedule and estimate the elasticity of corporate taxable income with respect to statutory corporation tax rates. The analysis yields large elasticity estimates suggesting that corporate taxpayers react sensitively to tax incentives. Detailed information on tax payer audit results, tax deductions and asset investment moreover allows us to distinguish real and evasion responses to taxation. In a second step, we use audit data to study the response of tax payers to deterrence instruments and analyse a large-scale administrative intervention that aimed at bringing non-registered tax payers into the tax net. JEL Classification: H2, H7 Keywords: Corporate taxation, developing countries, bunching 1

2 1. Introduction Recent years have seen a rising academic and political interest in the design of tax systems and administration structures in developing economies and emerging markets. Starting point is the observation that many countries in the developing world fail to raise significant amounts of tax (relative to their GDP), thus hampering the provision of public goods and services and preventing independence from foreign assistance (see e.g. Mascagni et al. (2014) and IMF (2011)). In a recent paper, Besley and Persson (2013) state that tax lies at the heart of state development and that the central question is how governments in developing countries can go from raising about 10% of GDP in taxes to raising around 40%. We contribute to this debate by studying the impact of corporate tax rates and tax payer audits on firms income reporting, drawing on the population of corporate tax returns for South Africa and rich information on tax payer audits. This allows us to approximate the excess burden of corporate taxation and to study the e ectiveness of tax payer audits in lowering firms evasion activities. Studying corporate taxation in developing economies is important for several reasons: firstly, corporate taxes are amajorrevenuesourcefordevelopingandemergingeconomies,accountingforabout 17% of total tax paid on average, compared to only around 10% in OECD countries (see e.g. Gordon and Li (2009), Peter et al. (2010) and IMF (2011)). At the same time, the welfare losses related to corporate taxation may be high, especially in the developing world. Compared to the personal income tax, corporate behavioral responses are manifold, e.g. including the scale of production combined with the demand for labor or the choice of the legal form Devereux et al. (2014)). In the low-enforcement environment of developing countries and emerging markets, corporate evasion responses to taxation may moreover be pronounced. While there have been some attempts to estimate the welfare consequences of corporate taxation for countries in the developed world, the empirical literature is scarce for developing economies and emerging markets. 1 Similarly, the literature studying the e ectiveness of deterrence instruments (i.e. tax payer audits and fines) in limiting evasion activities is largely restricted to developed economies (see e.g. Slemrod et al. (2001), Kleven et al. (2011)). Studies for developing countries and emerging economies are, in turn, scarce at best (see e.g. Castro and Scartascini (2015) for an exception). Our empirical analysis relies on confidential data for the universe of corporate tax returns in South Africa between 2008 and 2015 provided by the South African Revenue Service 1 For the developed world see e.g., Devereux et al. (2014), Dwenger and Steiner (2012). Some evidence for firms responsivness to taxation in developing countries are Kleven and Waseem (2013) and Best et al. (2015). 2

3 (SARS). In the first part of the paper, we study firm responses to statutory corporate tax rates, exploiting non-linearities in the corporate tax schedule which exhibits both kinks and notches in all tax years and allows estimating firms responses to corporate taxation using the bunching approach introduced by Saez (2010) for kink points and extended to notches by Kleven and Waseem (2013). The idea behind the bunching theory is that an increase in the marginal tax rate at a certain threshold induces firms near this kink to adjust corporate taxable income in order to stay at the threshold. In this paper, we follow Chetty et al. (2011) and estimate the counterfactual distribution in the absence of the kink by local polynomial regression to obtain the excess mass at the kink. For all years, we find clear and significant bunching behavior at all thresholds. For the pooled sample from 2010 to 2015, our elasticity estimates range between 0.66 and 0.92 for the first kink in the tax schedule where the marginal tax rate jumps from a zero-rate to 7% in 2010 and 2011 and from a zero-rate to 10% in 2012 to 2015 respectively. For the second and third kink in the tax schedule where the marginal tax rate increases by 14 and 7 percentage points respectively, we find somewhat smaller elasticities between 0.10 and The elasticities at the latter two thresholds are comparable to those found in the literature for the developed world. Devereux et al. (2014) e.g. report an elasticity of corporate taxable income to corporate tax rates UK of 0.17 at maximum for the UK. For the lower threshold, the estimates are much higher than prior findings for developed countries and support recent evidence suggesting strong tax-responses of corporate tax payers in developing and emerging countries (see e.g. Waseem (2015)). Using rich tax return data on gross income, deductions, corporate asset investment and audit activity moreover allows us to study how tax payers bunch at the kink points in the tax schedule. This analysis yields no evidence for quantitatively significant real responses of tax payers. Finally, we draw on detailed audit information to assess the impact of auditing strategies on tax payer behaviour. Empirical identification draws on the fact that a fraction of audit cases were randomly selected prior to We moreover assess the sensitivity of our findings to using propensity-score-matching di erence in di erence estimates that compare the future development of taxable income reporting of similar audited and nonaudited tax payers. This strategy furthermore allows us to determine spillover e ects of audits on other tax-payers close to the audited firm (e.g. operating in the same industry and region). Finally, while standard tax-payer audits and fines are instruments to deter tax evasion of registered tax payers, SARS made a major e ort to address non-registration by comparing the SARS register with the South African trade register, identifying thousands of firms registered with the trade-register but not with the tax register. Besides learning about the structure of these firms that used to operate in the shadow economy, the data allows us to study the short-run revenue consequences of this 3

4 intervention and to assess the income reporting of these newly-registered tax payers. The rest of the paper is structured as follows: Section 2 presents the institutional setting and the data. In Section 3, we discuss the bunching approach and present estimates for bunching behaviour at kink points as well as some analysis studying how tax payers. Section 4 presents the analysis of deterrence on tax payer behaviour. Section 5 concludes. 2. Institutional Setting and Data Income taxes are levied by the national government of South Africa under the Income Tax Act 58 of The tax system is residence-based and the current headline company tax rate is 28 %. 2 The South African tax system knows two types of Small and Medium Enterprises (SMEs) which benefit from special tax treatment. So called Micro Businesses face special tax dispensation in the form of a turnover tax. This turnover tax has a graduated tax rate structure with a maximum marginal rate of 6%. To be eligible for this income tax, gross income (sales / turnover plus other income) for the year of assessment must not exceed an amount of R1 million and total assets (current and noncurrent) must not exceed R5 million. The limitation with being a Micro Business is the restriction from registering as a VAT vendor and a company must elect in the beginning of an assessment year to be classified as a micro business. The tax system further classifies Small Business Corporations (SBC) that were introduced in An SBC benefits from progressive income tax rates instead of a fixed rate of 28%. Until 2013, the tax schedule for SBC contained three tax brackets generating two convex kinks. There was a zero marginal tax rate in the first bracket that increased to 10% until 2011 and changed from 2012 onwards to 7% at a threshold value of about 55, At a threshold value of 300, 000 the jump in the marginal tax rate amounts to 18 (21 from 2012 onwards) percentage points. In 2014, a fourth bracket was introduced which implied that the jump at the second threshold was lowered. The tax schedule for the most recent year of our data is shown in Figure 1. SBCs further benefit from immediate write-o of all plant or machinery used in a process of manufacture and are eligible for an accelerated write-o of depreciable assets. To qualify as an SBC, a company must not have elected to be classified as a Micro Business for the year of tax assessment and meet specific criteria. These include among others a gross income (sales / turnover plus other income) not exceeding R20 million (R14 million prior to 2013 tax year) and limited shareholding. 2 Certain sectors of the economy such as mining and farming have di erent e ective tax rates due to sector-specific tax dispensations and deductions. Further tax dispensations have been a orded to the small and medium enterprises. 3 Note that especially the lower threshold gradually increased over time. 4

5 30% Figure 1: Tax Schedule in % MTR 10% 70, , , 000 Taxable Income Notes: The figure shows the marginal tax rates for the year At each threshold, denoted by the dashed lines, the marginal tax rate jumps up. Our analysis is based on two datasets from the South African Revenue Service (SARS). We use administrative tax return data on the population of South African companies for the period We are left with 4.3 million observations for around 1.2 million firms. To study the di erent adjustment marigins, we exploit all tax variables which are included and have to be submitted on the annual tax return that is o cially used to determine their taxable income. The information available on the tax return includes all kind of firm s balance sheet items including total assets, directors remuneration, donations, travel expenses etc. In addition, we exploit a unique dataset from SARS that provides us with information on the universe of (partly random) audits for the same time period. This dataset entails variables on the kind of audit, the final monetary audit result, penalty and interest payments. The main advantages of our dataset are that it provides us with real tax liability of firms unlike financial statements that do not reflect deductions that were allowed. Thus we are able to shed light on the nature of the behavioral response. 3. Descriptives Figure 2a depicts the distribution of taxable income (for all years between 2009 and 2014). The distribution has a striking spike at zero taxable income. Along the same lines, Figure 2b depicts the fraction of firms which report positive, negative and zero taxable income. 5

6 Figure 2: Descriptives Taxable Income - All firms Density 0 5.0e All Firms, All Years Taxable Income in Rand Share of Firms Share of Firms by Taxable Income Sign All Firms Tax Year Positive Zero Negative a) Distribution Taxable Income b) Distribution of Firms Notes: Subgraph a) shows the distribution of taxable income around zero in a +/- R10000 window for all firms and all years between 2009 and The right graph shows the distribution of firms with positve (blue), negative (green) and zero (red) taxable income for the same sample. Figure 3: Descriptives Taxable Income - (Claimed) Inactives firms Density 0 5.0e Without (Claimed) Inactive Firms, All Years Taxable Income in Rand Share of Firms Share of Firms by Taxable Income Sign Without (Claimed) Inactive Firms Tax Year Positive Zero Negative a) Distribution Taxable Income b) Distribution of Firms Notes: The left graph shows the distribution of taxable income around zero for all active firms and all years between 2009 and The right graph shows the distribution of firms with positve (blue), negative (green) and zero (red) taxable income for the same sample. Strikingly, only around one third of the tax returns have a positive taxable income, while around 35% have exactly zero taxable income and the rest reports losses. The high prevalence of losses in taxable income may partly stem from generous loss carry forward provisions which allow for infinite loss carry forwards. The spike on zero income may relate to (reported) inactivity of firms in given tax years. Indeed, if we drop dormant firms and firms which report to be inactive in a given tax year in the sense that there are missing or zero entries in major tax return fields related to sales, costs and deductions, 6

7 the fraction of firms reporting zero taxable income drops to about 8% (see Subfigure 3b). The major spike at zero taxable income in the tax distribution nevertheless remains (see Subfigure 3a). Note that the loss carry forward provisions grant tax subsidies on losses which are o set against positive taxable profit in later tax years. As the standard corporate tax rate amounts to 28% during our sample period, the value of this loss carry forward provision corresponds to a subsidy of 28% for standard firms and implies a major kink in the corporate tax schedule at zero taxable income for these entities. There are analogous - though weaker - e ects for SBC as they are subject to tax free income (up to about R70,000) and increasing marginal tax rates (in steps) thereafter (see previous section). The deduction value of losses hence depends on expected income in later periods. In line with this notion, we also see bunching (although quantitatively more moderate) at zero income for SBCs (Figure 4a and Figure 4b). Figure 4: Distribution Taxable Income - Split by Company Type/Size Density 0 5.0e-051.0e-041.5e-042.0e-042.5e-04 Without (Claimed) Inactive Firms, All Years Taxable Income in Rand Density 0 5.0e Without (Claimed) Inactive Firms, All Years Taxable Income in Rand a) Small Businesses b) Others Notes: The left graph shows the distribution of taxable income around zero in a +/- R10000 window seperately for small businesses (Graph a) and all other firms (Graph b) that claimed to be active. 4. Bunching Analysis To consider the welfare consequences of corporate taxation, we follow the literature in the spirit of Feldstein (1995) and estimate the compensated elasticity of taxable income. This central parameter captures all kinds of behavioral responses including real production shifts, tax avoidance and evasion. It is defined as the percentage change in taxable income z due to an increase in the net-of-tax rate (1 ) ofonepercent.theoretically, the introduction of a kink in the budget set induces bunching behaviour within a certain income range provided that preferences are convex and smoothly distributed in the firm population. Due to adjustment costs and optimisation frictions (e.g., exisiting 7

8 long-term contracts), a broader bunching window around the kink is observed rather than a clear spike at the threshold. Comparing the income density distribution with acounterfactualscenariowithoutkink,theexcessmassoftaxpayerscanbeusedto determine the elasticity e(z) (foradetailedexplanationseesaez(2010)orchettyetal. (2011)). The compensated elasticity of taxable income, identified locally at the threshold k, isthengivenby e(k) = b k log( ), (1) where the net-of-tax rate changes by log( )percent. Therelativeexcessmassof taxpayers at the threshold is given by b, whichistheonlyparameterthatneedstobe estimated. To estimate b, wefollowchettyetal.(2011)whoproposetodeterminethe counterfactual density by running a local polynomial regression on binned data, while excluding data bins within the bunching window [l; u]: N j = qx i=0 i Z i + ux i=l i I[Z j = i]+" j. (2) We choose the number of polynoms using the BIC criterion and use an endogenous datadriven procedure to determine the bunching window (Dekker and Strohmaier (2016)) thereby allowing the window to be asymmetric. To account for increasing tax thresholds over time when pooling the tax years, the data is rescaled relative to the threshold and aweightedaverageisusedinequation1.inthefollowing,weonlypoolyearsthatdid not experience changes in the tax rate Bunching Estimates From a policy point of view, the lower thresholds might be of special interest for two reasons. First, the marginal tax rate starts from a zero-rate in all years. It jumps to 7% in 2010 and 2011 and was increased to 10% after Second, the number of firms in South Africa is highest around this lower threshold of approximately The SBC s a orded preferential income tax rates and special deductions might hamper their growth and participation in the economy. If firms instead of using these incentives to grow remain small in order to continue benefiting from them, the structure of the tax system has to redefined. Figure 5 displays the graphical bunching results for the pooled sample splitted by the di erent tax schemes whereas the exact estimation results can be found in the Appendix. The figure shows the number of firms per income bin relative to the first tax threshold. 8

9 In all graphs, the blue dotted line shows the observed distribution of corporate taxable income while the red dooted line plots the estimated counterfactual. In addition, the right graphs show the binned income distribution for those firms, that are not eligible for being a SCB. In both cases, the BIC criterion suggests a seventh order polynomial for the counterfactual distribution and the endogenously determined bunching window goes from approximately R to R400. For both time spans, we see clear and asymmetric bunching of firms for which the progressive tax schedule applies. While there is a clear spike in both left graphs for the SBC, the income distribution for all other firms is smooth around this threshold. It can therefore be infered that nothing else, e.g., no other tax rule, causes the clear spike in the left graphs. For the years 2010 to 2011, we estimate an excess mass of ˆb =4.76which is highly significant. Using a weighted average for the threshold value, this translates into an elasticity of ê = For the time period 2012 to 2015, the excess mass of ˆb =5.81 implies an estimated elasticity of ê =0.92. Quantitatively, a decrease in the statutory net-of-tax rate by 10% induces a reduction in taxable income by 9.2%. So two things are worth noting here. First, the large bunching estimate suggests that firms, especially at the lower end of the income distribuion, strongly respond to taxation. However, to conduct further welfare analysis, the anatomy of response is of crucial interest (and shifted to the next Section). Second, the bunching is pretty impresice and the whole mass lies to the left of the threshold. Such an asymmetric bunching window might be an indication for a high risk aversion among owners of SBCs (Devereux et al. (2014)). To see how the bunching precisely evolved over time with increasing threshold values, Figure 6 shows part of the income distribution for each year seperately. The threshold incomes are indicated by vertical lines. The graphs show that the spike in the distribution moves exactly along with the threshold over time. It is striking, that although there is aclearspikeatthethresholdinallyears,thebunchingwindowiscomparablylargeand highle asymmeric to the left. Figure 8 and Figure 9 shown in the Appendix report the analogous bunching results of the second and the third threshold. As clear from first sight, firms respond at both income levels as we see a clear spike in the observed distribution in all graphs. For the second threshold, we estimate an excess mass of around ˆb =6forthefirsttwotime spans, where the change in the net-of-tax rate of about 26%. However, in the subsequent time period, where the tax schedule was split in three tax brackets, this change in the net-of-tax rate at the kink was lowered to 16%. Interestingly, this immediatly translates into a lower behavioral response with an estimated excess mass at the second threshold for the time period 2014 to 2015 of ˆb =3.4. However in relation to the tax change and the location of the kink, the estimates elasticity is relativly stable and ranges from 0.10 to Compared to the lowest threshold, the bunching behavior is more symmetric, 9

10 Figure 5: Bunching at the First Threshold - Pooled Data Small Businesses, Distance First Threshold Others, Distance First Threshold Fitted values Fitted values Small Businesses, Distance First Threshold Others, Distance First Threshold Fitted values Fitted values Notes: The figures show the observed distribution (blue dotted line) and the estimated counterfactual (red dotted line) of corporate taxable income around the lowest threshold for the pooled years 2010 to 2011 and 2012 to The sample is split in SBCs (left graphs) and Non-SBCs (right graphs). Further estimation details can be found in the Appendix. 10

11 Figure 6: Bunching at the First Threshold - Seperately for each year Density 0 1.0e-052.0e-053.0e-054.0e-055.0e-05 First Threshold Taxable Income Density 0 1.0e-052.0e-053.0e-054.0e-055.0e-05 First Threshold Taxable Income Density 0 1.0e-052.0e-053.0e-054.0e-055.0e-05 First Threshold Taxable Income Density 0 1.0e-052.0e-053.0e-054.0e-055.0e-05 First Threshold Taxable Income Notes: The figures show the distribution of corporate taxable income seperately for the years 2012 to 2015 aound the lower threshold. 11

12 but comparably imprecise as shown in column (2) in Table 1 in the Appendix. 4 For the highest threshold value of , which was introduced in 2014, the response is a bit larger with an estimated elasticity of ê = Anatomy of the Responses We can moreover draw on the full corporate tax return information submitted by firms in South Africa, including data on gross income, deductions claimed and total assets. This may allow providing some notion on how firms bunch at the threshold, with a main focus on understanding whether bunching is achieved through the adjustment of real activities or through tax evasion behavior. One complication in studying these response margins is that we only observe data reported by firms to the tax authority, which implies that all variables may be inflicted by misreporting with the aim to reduce the corporate tax burden. Firms may e.g. bunch at the threshold by underreporting taxable income, gross income, deductions as well as total assets and thus appear similar to other non-bunchers at the threshold despite the systematic misreporting. If underreporting is easier in specific adjustment margins, e.g. underreporting of gross income and overreporting of deductions is easier relative to the underreporting of assets, we might see diminished ratios of taxable income and gross income per assets at the threshold or overreporting of deductions relative to assets when firms bunch by evading income. If firms in turn adjust their real activity in response to the tax increase at the threshold, they are expected to deviate from their optimal size. Assuming optimal firm size to be mainly driven by total factor productivity of firms, we would expect to see that bunchers at the threshold manage to achieve income levels close to the threshold level at lower input costs and with less production input. Consequently, real activity responses are expected to be associated with higher levels of total factor productivity at the threshold, i.e. higher reported taxable income per assets and employees. In line with this notion, Tables 1 to 3 present results of estimation models where we regress the profitability of firms, measured as taxable income over total assets, on taxable income and an indicator for firms reporting income in the bunching window (in a given year). To mitigate the e ect of outliers and reduce measurement error, observations with profitability rates below the 5th and above the 95th percentile are dropped from the analysis. Table 1 assesses the first kink in the tax schedule. Specification (1) regresses profitability on the bunching indicator and a full set of year fixed e ects. Specification (2) adds a full set of two-digit industry controls and region controls to the set of regressors. 4 Note that tha data was binned according to the rule-of-thumb of Freedman-Diaconis. As we have more data at the lower end of the corporate income distribution, the bins are smaller there. Thus, the clear asymmetric bunching to the left in Figure 6 might look too extreme here. Table 1 in the Appendix gives a more detailed description of the bunching window. 12

13 Table 1: Response Profitability Threshold 1 (1) (2) (3) (4) (5) (6) Buncher (0.003) (0.003) (0.003) (0.003) (0.017) (0.017) Taxable Income (0.000) (0.000) (0.000) (0.000) Ln Taxable Income 0.431*** 0.459*** (0.091) (0.090) Sector FE Yes Yes Yes Region FE Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Firms All All Pos. Prof. Pos. Prof. All All Observations 23,431 23,431 18,598 18,598 23,431 23,431 R-squared Notes: This table shows the regression results for the first threshold where we regress firms profitability on a dummy indicating if the firm reports a taxable income within the bunching window. The regression is run locally around the threshold. All firms reporting an income within R10,000 around the threshold are included. In both specifications, we do not find evidence for significant jumps in profitability. This finding is robust to restricting the set of firms used in the analysis to those with positive profits in the previous year to avoid our results to be confounded by loss-carryforward provisions (cf. Columns (3) and (4)). Finally, Specifications (5) and (6) assess the sensitivity of the findings to functional form assumptions and employs a log-log-model instead of the level-specifications. This, again, yields no positive jump in reported profitability at the threshold, suggesting that real responses are not the main driver of bunching behaviour. This notion is confirmed when rerunning the analysis for the second and third kink in the tax schedule (cf. Tables 2 and 3). Note that reassuringly, we also find no significant response in profitability around the threshold in the group of firms that are not eligible for small business corporation tax and hence the progressive part of the South African corporate tax schedule. 13

14 Table 2: Response Profitability Threshold 2 (1) (2) (3) (4) (5) (6) Buncher * (0.003) (0.003) (0.003) (0.003) (0.011) (0.011) Taxable Income (0.000) (0.000) (0.000) (0.000) Ln Taxable Income 0.198*** 0.199*** (0.037) (0.037) Sector FE Yes Yes Yes Region FE Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Firms All All Pos. Prof. Pos. Prof. All All Observations 25,024 25,024 22,155 22,155 25,024 25,024 R-squared Notes: This table shows the regression results for the second threshold where we regress firms profitability on a dummy indicating if the firm reports a taxable income within the bunching window. The regression is run locally around the threshold. All firms reporting an income within R70,000 around the threshold are included. Table 3: Response Profitability Threshold 3 (1) (2) (3) (4) (5) (6) Buncher (0.013) (0.012) (0.013) (0.013) (0.048) (0.047) Taxable Income (0.000) (0.000) (0.000) (0.000) Ln Taxable Income (0.345) (0.340) Sector FE Yes Yes Yes Region FE Yes Yes Yes Firms All All Pos. Prof. Pos. Prof. All All Observations 1,411 1,411 1,287 1,287 1,411 1,411 R-squared Notes: This table shows the regression results for the third threshold where we regress firms profitability on a dummy indicating if the firm reports a taxable income within the bunching window. The regression is run locally around the threshold. All firms reporting an income within R50,000 around the threshold are included. 5. Analysis of Deterrence on Taxpayer Behavior To be completeted. 14

15 6. Combatting the Shadow Economy: Forced Tax Payer Registrations While classic tax payer audits may be an e ective instrument to deter tax evasion behaviour of registered tax payers, they are not suited to combat the large shadow economy in many developing countries, i.e. tax evasion by non-registration with the tax authorities. Many tax authorities in developing economies (in collaboration with development agencies) hence implemented programs designed to raise tax payer registration rates, among others including television and newspaper advertisements and registration audits, where tax auditors are sent in the field to identify and register informal businesses (e.g. Hoekstra et al. (2013) and Hodges (2013)). The impact of these e orts on tax payer behaviour has to the best of our knowledge not yet been empirically tested. One often-suggested strategy to enhance tax payer registrations in the corporate sector is to increase the interconnectivity [of the tax registry] with other government data (see e.g. Eckardt (2013)). In the following, we will assess the impact of major e orts by SARS along these lines, namely comparisons of registered corporate tax payers in the SARS register with the South African trade register. A comparison between the registers was conduced twice over the past years, first in early 2008 and second in early These e orts resulted in a large number of new tax payer registrations, amounting to about 300,000 newly registered tax payers at both points in time (see Figure 7), which substantially exceeds the number of new registrations in other months (corresponding to about 20,000 firms on average). Figure 7: New Registrations Number of Registrations New Registrations by Date Number of Late Registrations New Forced Registrations by Date Jan 2007Apr 2008 Feb 2014 Date Jan 2007Apr 2008 Feb 2014 Date a) Monthly Registrations b) Monthly Registrations, forced Notes: The left graph shows the number of new registrations between 2007 and The right graph depicts the number of late registrations in the same time periode. As the major aim of the intervention was to bring non-registered tax payers into the tax 15

16 net, we will, in the following, focus on firms which were newly registered in the course of the trade register intervention and were labelled to be late registered. This applies to around 220,000 firms in 2014 and around 170,000 firms in Note that in South Africa all firms are by law required to register with the tax authority for corporate tax purposes, irrespective of their size or income. When SARS, in the wake of the trade register comparison, automatically registered non-compliant firms, they were requested to submit tax returns for all tax years from their date of incorporation onwards and for subsequent years to come. Table 4 depicts the number of tax returns that were submitted by tax payers that were compulsorily registered by SARS in the wake of the trade register comparisons in 2008 and Note that our tax return data only comprises tax returns between 2009 and 2014, i.e. we cannot identify returns submitted by tax payers in early or later years. The table shows that only 12.5% and 9.3% of the tax payers registered submitted a tax return in at least one tax year between 2009 and In total, the number of tax returns rose by around 82, 000 and 53, 000 related to the reform. 6 We then turn to a closer analysis of compulsorily registered firms that submitted tax returns. As described above, the distribution of taxable income is characterised by a large fraction of firms reporting zero or negative taxable income. To determine whether the newly registered tax payers actually do pay taxes, we hence in a first step assess their propensity to report zero profits (Table 5, Columns (1) and (2)) and zero profits or losses (Table 5, Columns (3) and (4)). The findings suggest that the firms which were compulsorily registered with SARS in the wake of the trade register comparisons have a higher probability to report zero and negative taxable income than other firms. The propensity to report zero income is raised by around 29.6 percentage points and the propensity to report zero or negative income by 18.8 percentage points (cf. Columns (1) and (3) respectively), which is large relative to the baseline propensity to report zero taxable income and zero or negative income of 44.3% and 75.1% respectively. Table 6 determines whether zero taxable income reporting among the newly registered tax payers stems from (reported) inactivity or whether the firms are active but nevertheless report taxable income of zero. It shows that a large fraction of the compulsorily registered entities are dormant in the respective tax year. The picture looks even stronger when inactivity is measured by no or zero entries in major tax return fields related to sales, costs, assets and tax deductions. 5 Note that the number of late registrations amounts to about 7000 firms in other months on average. 6 One caveat has to be noted. SARS only provided us with a current version of the company register which allows us to identify firms that were compulsorily registered with SARS in 2008 and 2014 respectively and were still in the register in 2015, i.e. any firm that was identified as non-compliant through the trade register comparison but deregistered in the meantime won t be identified as a 16

17 Table 4: Returns Submitted by Compulsorily Registered Tax Payers Tax Year Comparison 2008 Comparison , ,032 1, ,805 3, ,482 10, ,181 20, ,842 16,293 Total Number of Returns 82,112 52,946 Tax Payers Subm. at leat 1 Return 20,921 21,383 (% of Total) 12.5% 9.3% Tax Payers Subm. No Return 146, ,325 (% of Total) 87.5% 90.7% Total Number of Tax Payers 167, ,708 Notes: This table presents the number of tax returns that were submitted by firms that were forced to register in the course of the trade register comparisons of 2008 and Table 5: Propensity to Report Zero Income and Losses (1) (2) (3) (4) Taxable Income of Zero Zero Zero or Neg. Zero or Neg. Forced 0.296*** 0.286*** 0.188*** 0.144*** (0.003) (0.002) (0.003) (0.002) Year FE Yes Yes Yes Yes Est. Model Logit OLS Logit OLS Observations 4,225,704 4,225,704 4,225,704 4,225,704 R-squared Notes: This table presents the regression results for the propensity to report zero profits or zero or losses depending on forced registration. Table 7 moreover determines whether firms which were compulsorily-registered in the course of the trade register comparisons have a higher propensity to report zero taxable income (Specifications (1) and (2)) and zero income or losses (Specifications (3) and (4)), disregarding dormant and inactive firms which are dropped from the sample. Again, we find that compulsorily registered firms (indicated by the variable Forced ) that report respective case in our data. 17

18 Table 6: Nonactivity Among Compulsorily Registered Tax Payers Comparison 2008 Comparison 2014 Dormant Inactive Dormant Inactive (8.4%) 12,798 (64.7%) 290 (53.3%) 516 (94.9%) (10.4%) 10,198 (59.9%) 670 (53.4%) 1181 (94.1%) (13.5%) 8367 (56.5%) 2114 (55.9%) 3537 (93.6%) (21.9%) 6577 (52.7%) 5919 (56.1%) 9912 (93.9%) (31.6%) 4960 (48.7%) 11,830 (57.6%) 18,751 (91.4%) (27.2%) 3457 (44.1%) 7945 (48.7%) 13,657 (83.8%) Notes: This table reports the number of newly registered firms that reported to be dormant and the number of firms that reported to be inactive (measured by no or zero entries in sales, cost of sales, assets and tax deductions). to be active have a higher propensity to have taxable income of zero or weakly negative income. Precisely, the propensity to report zero income is by 1.1 percentage point larger than that of other firms and the propensity to report zero income or losses exceeds that of other firms by around 5.5 percentage points. Again, these e ects are substantive given the baseline propensities to report zero income and zero income or losses of 7.9% and 58.4% respectively. Table 7: Bunching on Zero (Active Firms) (1) (2) (3) (4) Taxable Income of Zero Zero Zero or Neg. Zero or Neg. Forced 0.011*** 0.011*** 0.055*** 0.054*** (0.002) (0.002) (0.004) (0.004) Year Fixed E ects Yes Yes Yes Yes Est. Model Logit OLS Logit OLS Observations 2,527,808 2,527,808 2,527,808 2,527,808 R-squared Notes: This table shows the regression results for the propensity to bunch at zero and and/or report losses depending on forced registration. Finally, we focus on the group of newly registered firms with positive taxable income and assess whether they are more tax-sensitive than other entities, measured by their propensity to bunch at the kink points in the tax distribution if they are classified as SBCs. The respective results are presented in Tables 8 to 10. Table 8 assesses whether the newly registered firms have a higher propensity than other entities to bunch at any threshold in the corporate income tax schedule for SBCs, Table 9 determines the 18

19 Table 8: Forced Registrations and Bunching (Any Threshold) (1) (2) (3) (4) (5) Forced 0.013*** 0.012*** 0.015*** 0.013*** 0.013*** (0.002) (0.002) (0.002) (0.002) (0.003) Year Fixed E ects Yes Yes Yes Yes Yes Sector Fixed E ects No Yes No Yes Yes Area Fixed E ects No No No No Yes Estimation Method Logit Logit OLS OLS OLS Observations 746, , , , ,987 R-squared Notes: This table shows the regression results for the propensity to bunch at any threshold depending on forced registration. Table 9: Forced Registrations and Bunching (Lower Threshold) (1) (2) (3) (4) (5) Forced 0.014*** 0.013*** 0.017*** 0.015*** 0.015*** (0.002) (0.002) (0.003) (0.003) (0.003) Year Fixed E ects Yes Yes Yes Yes Yes Sector Fixed E ects No Yes No Yes Yes Area Fixed E ects No No No No Yes Estimation Method Logit Logit OLS OLS OLS Observations 746, , , , ,987 R-squared Notes: This table shows the regression results for the propensity to bunch at the lower threshold depending on forced registration. propensity to bunch at the lower threshold of the tax schedule (where the marginal tax rate increase from 0 to 7%). Table 10 reestimates the models in Table 9 restricting the sample to firms close to the lower threshold, precisely firms which report taxable income within the bounds of +/- 20,000 Rand around the threshold. All specifications suggest that newly registered firms have a higher propensity to bunch relative to other tax payers. The specifications in Tables 8 and 9 suggest that newly registered firms have a propensity around 1.3 and1.5 percentagepointshighertobunchatthethreshold(relativetothe baseline propensity of 4.4% and 3.4% respectively). In relative terms the e ect declines if the sample is restricted to firms with taxable income close to the threshold, suggesting that the propensity of newly registered firms to bunch at the threshold is 2.2 percentage points or 8.1% higher than that of other firms (baseline propensity to bunch: 27.0%). Finally, Table 11 - along the lines with the previous findings - suggests that, conditional 19

20 Table 10: Forced Registrations and Bunching (Lower Thresh., Restricted) (1) (2) (3) (4) (5) Forced 0.025** 0.022* 0.025* 0.023* 0.022* (0.013) (0.013) (0.013) (0.013) (0.013) Year Fixed E ects Yes Yes Yes Yes Yes Sector Fixed E ects No Yes No Yes Yes Area Fixed E ects No No No No Yes Estimation Method Logit Logit OLS OLS OLS Observations 92,889 87,189 92,889 87,190 87,190 R-squared Notes: This table shows the regression results for the propensity to bunch at the lower threshold. The sample is restricted to comprise all firms within +/- R20,000 around the threshold. Table 11: Tax Payable No Obs. Mean Std. Dev. Min Max Tax Payable No Forced Registries 4,090, , e e+09 Forced ,109 13, , e+07 Forced , , ,803,655 Tax Payable, Binary No Forced Registries 4,090, Forced , Forced , Notes: This table reports descriptive statisitics of the variable Tay Payable (total and binary) for the subgroup of firms that registered voluntarily, for those who were forced to register in 2008 and in 2014 respectivly. on submitting a tax return, newly registered firms are substantially less likely than other firms to have a positive tax burden (propensities of 2.2% and 10.6% vs. 21.6%). Analogously, average tax payments per newly registered firm and return are equally substantially smaller than that of other firms. 20

21 7. Concluding Remarks In this paper we estimate the firms responsiveness to taxation in a low enforcement environment. We use the population of tax return records in South Africa to exploit clustering of firms a kinks and notches in the tax schedule for Small and Medium Entreprises. We find clear evidence of bunching behavior at all kinks of the tax schedule. However, the response highly di ers across income levels. The response is highest at the lower end of the corporate income distribution with an elasticity estimate of nearly one. The di erence in these estimates is consistent with the literature on tax responsiveness in the developed world even though the estimates are somewhat larger as expected in a lower compliance environment. First profitability estimates on the nature of responses do not suggest that firms respond in real activities to stay artificially small. The next steps will be to shed further light on the margins of adjustments and to exploit the return data on Micro Businesses to assess the e ectiveness and potential revenue gains of di erent tax systems. We will also draw on detailed audit information to assess the impact of auditing strategies on tax payer behaviour and to quantify spillover e ects of audits. 21

22 References Besley, T. J. and T. Persson (2013): Taxation and Development, CEPR Discussion Papers 9307, C.E.P.R. Discussion Papers. Best, M. C., A. Brockmeyer, H. J. Kleven, J. Spinnewijn, and M. Waseem (2015): Production versus Revenue E ciency with Limited Tax Capacity: Theory and Evidence from Pakistan, Journal of Political Economy, 123, Castro, L. and C. Scartascini (2015): Tax compliance and enforcement in the pampas evidence from a field experiment, Journal of Economic Behavior and Organization, 116, Chetty, R., J. N. Friedman, T. Olsen, and L. Pistaferri (2011): Adjustment Costs, Firm Responses, and Micro vs. Macro Labor Supply Elasticities: Evidence from Danish Tax Records. The Quarterly Journal of Economics, 126, Dekker, V. and K. Strohmaier (2016): A Data-Driven Procedure to Determine the Bunching Window - An Application for the Netherlands, mimeo. Devereux, M. P., L. Liu, and S. Loretz (2014): The Elasticity of Corporate Taxable Income: New Evidence from UK Tax Records, American Economic Journal: Economic Policy, 6, Dwenger, N. and V. Steiner (2012): Profit Taxation and the Elasticity of the Corporate Income Tax Base: Evidence from German Corporate Tax Return Data, National Tax Journal, 65, Eckardt, U. (2013): Regional ITC workshop Taxation in ASEAN - Identifying experiences and lessons learned in six countries, Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ), Bonn. Feldstein, M. (1995): The E ect of Marginal Tax Rates on Taxable Income: A Panel Study of the 1986 Tax Reform Act, Journal of Political Economy, 103, Gordon, R. and W. Li (2009): Tax structures in developing countries: Many puzzles and a possible explanation, Journal of Public Economics, 93, Hodges, Y. (2013): Detailed Guidelines for Improved Tax Administration in Latin America and the Caribbean, Chapter 6. Taxpayer Services, USAID Leadership in Public Financial Management. 22

23 Hoekstra, W., F. G. Pereira, and J. Queijo (2013): Unlocking tax-revenue collection in rapidly growing markets, Public Sector Report, McKinsey&Company. IMF (2011): Revenue Mobilization in Developing Countries, Kleven, H. J., M. B. Knudsen, C. T. Kreiner, S. Pedersen, and E. Saez (2011): Unwilling or Unable to Cheat? Evidence From a Tax Audit Experiment in Denmark, Econometrica, 79, Kleven, H. J. and M. Waseem (2013): Using notches to uncover optimization frictions and structural elasticities: Theory and evidence from Pakistan*, The Quarterly Journal of Economics. Mascagni, G., M. Moore, and R. McCluskey (2014): Tax revenue mobilisation in developing countries: Issues and challenges, European Parliament, EXPO/B/DEVE/2013/35. Peter, K. S., S. Buttrick, and D. Duncan (2010): Global Reform of Personal Income Taxation, : Evidence from 189 Countries, National Tax Journal, 63, Saez, E. (2010): Do Taxpayers Bunch at Kink Points? American Economic Journal: Economic Policy, 2, Slemrod, J., M. Blumenthal, and C. Christian (2001): Taxpayer response to an increased probability of audit: evidence from a controlled experiment in Minnesota, Journal of Public Economics, 79, Waseem, M. (2015): Taxes, Informality and Income Shifting: Evidence from a Recent Pakistani Tax Reform, University of Manchester, mimeo. 23

24 A. Appendix Table 12: Estimation Results Bunching Analyses Bunching Window q b se e First Threshold [-11000;400] [-10600;350] Second Threshold [-10400;3500] [-9800;5000] [-10000;4200] Third Threshold [-9000;4000] Notes: The table shows the estimates of the excess mass b and the respective elasticity of corporate taxable income with respect to the statutory net-of-tax rate e. The order of polynomial is denoted by q. 24

25 Figure 8: Bunching at the Second Threshold - Pooled Data Small Businesses, Distance Second Threshold Others, Distance Second Threshold Fitted values Fitted values Small Businesses, Others, Distance Second Threshold Distance Second Threshold Fitted values Fitted values Small Businesses, Others, Distance Second Threshold Distance Second Threshold Fitted values Fitted values Notes: The figures show the observed distribution (blue dotted line) and the estimated counterfactual (red dotted line) of corporate taxable income around the second threshold for the pooled years 2010 to 2011, 2012 to 2013 and 2014 to The sample is split in SBCs (left graphs) and Non-SBCs (right graphs). Further estimation details can be found in the Appendix. 25

26 Figure 9: Bunching at the Third Threshold - Pooled Data Small Businesses, Distance Third Threshold Others, Distance Third Threshold Fitted values Fitted values Notes: The figures show the observed distribution (blue dotted line) and the estimated counterfactual (red dotted line) of corporate taxable income around the third threshold for the pooled years 2014 to The sample is split in SBCs (left graphs) and Non-SBCs (right graphs). Further estimation details can be found in the Appendix. 26