Estimating the elasticity of turnover from bunching: Peferential tax regimes for solo self-employed in Italy

Alosa, Francesco Working Paper Estimating the elasticity of turnover from bunching: Peferential tax regimes for solo self-employed in Italy Quaderni - Working Paper DSE, No. 1186 Provided in Cooperation with: University of Bologna, Department of Economics Suggested Citation: Alosa, Francesco (2023) : Estimating the elasticity of turnover from bunching: Peferential tax regimes for solo self-employed in Italy, Quaderni - Working Paper DSE, No. 1186, Alma Mater Studiorum - Università di Bologna, Dipartimento di Scienze Economiche (DSE), Bologna, https://doi.org/10.6092/unibo/amsacta/7294 This Version is available at: https://hdl.handle.net/10419/282308 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc/4.0/ ISSN 2282-6483 Estimating the Elasticity of Turnover from Bunching: Preferential Tax Regimes for Solo Self-employed in Italy Francesco Alosa Quaderni - Working Paper DSE N°1186 Estimating the Elasticity of Turnover from Bunching: Preferential Tax Regimes for Solo Self-employed in Italy∗ Francesco Alosa† UniBo, UCFS March 2023 Abstract Turnover is a key indicator of economic activity, but we know little about how much entrepreneurs adjust it as a response to taxation. This is because business taxation is usually based on profits, rather than turnover. This paper exploits the notch created by the eligibility cut-off of the preferential (turnover) tax regime for solo self-employed in Italy to study turnover responses to taxation. I find that solo self-employed bunch below the turnover threshold to be eligible for the preferential scheme. Effects are different in different sectors, with professionals and business intermediaries showing the largest responses. Then, I estimate the turnover tax elasticity by focusing on the (last) marginal buncher. To do so, I adapt the models of Kleven and Waseem (2013) and Harju et al. (2019) to derive a modified indifference condition that fits the institutional set-up. The baseline estimate for the turnover tax elasticity is 0.072. Keywords: turnover tax elasticity, preferential tax regimes, solo self-employed, bunching. ∗I wish to thank my supervisor, Matthew Wakefield, for his thoughtful suggestions. This project also benefited from comments of the audience during several presentations. I wish to thank Giulio Zanella for discussing this paper during the PhD Forum, Vincenzo Scrutinio for his valuable comments, as well as other faculty members and PhD colleagues from University of Bologna for their feedback. I also wish to thank Spencer Bastani, Håkan Selin, Jarkko Harju, Tuomas Matikka as well as PhD students from Uppsala and Finland participating to the Public Economics PhD course in Helsinki in June 2022. For this project, I use data from the Italian National Statistics agency (Laboratorio ADELE-ISTAT). Results and opinions expressed in the paper are my own and do not represent official statistics from the agency. †Department of Economics, University of Bologna, Italy; email: [email protected] 1 Non-Technical Summary Policy makers know that stimulating entrepreneurship is key to generate economic growth. Both in advanced and developing countries, earnings taxation for small and medium enterprises have been reformed to provide preferential regimes with lower tax rates and easier compliance procedures. In many cases, such regimes tax turnover, rather then profits. As turnover is a key indicator of economic activity, it is crucial to understand to what extent it responds to tax incentives. In this paper, I investigate to what extent solo self-employed adjust sales turnover due to incentives of the tax system. I consider the Italian tax system because tax liabilities for solo self-employed in preferential regimes depend on the level of turnover: if turnover is below a certain threshold, solo self-employed can access a preferential regime. I investigate responses to this type of discontinuity in the tax schedule considering the two main preferential schemes introduced in Italy between 2012 and 2019. Then, I use responses to the turnover tax scheme threshold in 2019 to estimate the turnover tax elasticity. To carry out the analysis, I use administrative data from the Italian national statistics agency, ISTAT, on all self-employed operating in Italy between 2012 and 2019, with key information on sales turnover, costs, profits, and the sector of the economy in which the individual works. This paper delivers two main results: •First, I find that solo self-employed bunch below the turnover threshold, set by the tax code, to qualify for a preferential tax scheme. There are heterogeneous responses across sectors: Professionals and Business intermediaries are the most responsive groups, with estimated turnover reduction up to e19,000 and e15,000 respectively. I investigate responses across sectors and find that difference in the strength of responses are in line with financial incentives. Sectors where larger shares of taxpayers would benefit from being in the preferential regime do show the largest responses. •Second, using the appropriate theoretical framework, I estimate the turnover tax elasticity, that expresses the proportional change in turnover due to a 1 percentage point change in average turnover tax rate. The baseline estimate for the turnover tax elasticity is 0.072. Self-employed do adjust turnover when financial incentives make it convenient to do so. These findings suggest that policy makers should be careful when setting up preferential regimes that apply to taxpayers below a certain turnover threshold. As turnover is strictly related to output, such regimes might have a negative impact on economic activity, if the observed response is real. On the other hand, if the observed responses were explained by evasion, this policy could lead to reduced tax collection. 2 1 Introduction Stimulating entrepreneurship is key to generate economic growth. In several developing and advanced economies, policy makers attempt to foster business activity by setting up preferential tax regimes for small-medium enterprises (SMEs). The idea behind this policy is that simpler tax regimes with a lower tax burden would attract entrepreneurs that, in absence of this policy, would have either not produced or remained in the informal sector. These simplified schemes often feature some form of taxation of gross reported revenues, e.g. turnover taxation, as opposed to the standard profit-based tax regimes for businesses and corporations. The seminal paper by Diamond and Mirrlees (1971) advises against turnover taxation as it violates production efficiency. However, policy makers often deviate from this theoretical benchmark as turnover taxation makes compliance easier for small businesses and is more difficult to evade. Moreover, when there is no complete tax enforcement and evasion is possible, Best et al. (2015) argue that production-inefficient tax regimes might actually enhance welfare as efficiency losses are more than outweighed by higher revenue efficiency due to increasing compliance. As turnover taxation receives more attention as a policy tool, its effects on behaviour are worth-exploring. Although turnover is a key indicator of economic activity, we still know little about how much firms actually adjust it as a response to taxation. This paper fills this gap. I investigate to what extent solo self-employed adjust sales turnover due to incentives of the tax system.1I study responses to the notches created by the eligibility cut-offs of the preferential tax regimes for solo self-employed in Italy. Then, I focus on the notch created by the turnover tax regime threshold to estimate the turnover tax elasticity. Since turnover is strictly related to output, after accounting for prices, analysing such responses is extremely important for both academic research and policy-makers. The Italian tax system provides a suitable framework to address this question as tax liabilities for solo self-employed in preferential regimes depend on the level of turnover. If turnover is below a certain threshold, Italian solo self-employed can opt out of the ordinary tax regime and choose to be taxed at a preferential rate. In addition to tax advantages, the preferential regimes also have simplified compliance procedures.2Con- versely, if turnover is above the cut-off, higher average tax rates apply as the ordinary tax regime remains the only option. This creates a series of notches in the tax schedule for solo self-employed at different levels of turnover, depending on the sector and year being considered. I show that there is bunching below some of the statutory turnover-limits of the preferential regimes, as some solo self-employed choose to adjust their revenues and/or limit growth in sales to access tax advantages. Then, I exploit the notch created by the eligibility cut-off of the preferential turnover tax regime for solo self-employed to estimate the turnover tax elasticity. 1Solo self-employed are self-employed individuals who work without collaborators or employees. The share of solo self-employed in self-employment is increasing in many OECD countries (Boeri et al. 2020). 2Similar regimes have been implemented in several developing countries (Best et al. 2015). 3 I use administrative data from ISTAT on all self-employed operating in Italy between 2012 and 2019.3In this period, self-employed could choose between the ordinary tax regime and (at least) one scheme with potentially preferential tax-rates and simplified compliance procedures. The ordinary tax regime is moderately progressive and includes personal income-tax, social security contributions and VAT. Then, various preferential tax regimes have been introduced, exempting self-employed from VAT, and replacing the progressive personal income tax schedule with a proportional levy on taxable income (“a flat tax”). The turnover tax regime is one example of the preferential schemes being introduced in Italy, with the tax base being its distinguishing feature. While the ordinary regime taxes profits, the turnover regime defines the tax base as a sector-specific share of turnover, resulting in different tax incentives across sectors. The main analysis of this paper looks at responses to the notches in the tax schedule generated by the preferential turnover regime. Heterogeneity across sectors is exploited to investigate whether differences in responses are in line with differences in tax incentives. This paper makes two main contributions to the literature. First, while the existing evidence of bunching largely focus on taxable income adjustments,4this paper focuses on responses to taxation of sales turnover; turnover is a specific component of taxable income for self employed and a key indicator of economic activity. I show that individual entrepreneurs in Italy adjust the level of revenues as a response to financial incentives of the tax system. Solo self-employed bunch below the turnover threshold, set by the tax code, to qualify for a preferential tax scheme. Moreover, I document extensive and intensive margin responses in the turnover distribution after two key tax reforms in 2015 and 2018. Turnover responses to taxation are studied by Harju et al. (2019) and Liu et al. (2021) in the context of VAT registration thresholds, and by Aghion et al. (2022) with regard to the preferential regimes for self-employed in France. The first two studies show that businesses bunch below the VAT registration threshold.5Harju et al. (2019) find that compliance costs due to VAT tax filing explain most of the observed bunching of small firms in Finland, so that the estimated elasticity of value added is quite low.6 Then, Liu et al. (2021) find that bunching is more likely when corporations have lower inputs-sales ratio, higher proportion of business-to-consumer sales, and lower mark-ups. Differently from Harju et al. (2019), and Liu et al. (2021), this paper investigates the responses at the threshold where there is an overall change of the taxation of solo selfemployment income including, but not limited to, VAT. This seems to be a more suitable case to study how turnover responds to tax incentives, aside from compliance costs related to the tax system. Finally, while Aghion et al. (2022) stress the importance of tax 3National Statistics Agency, Italy. 4Saez (2010) for the US, Chetty et al. (2011) for Denmark, Kleven and Waseem (2013) for Pakistan, Bastani and Selin (2014) for Sweden, Adam et al. (2021) for the UK. 5They assume the VAT incidence falls, at least partly, on entrepreneurs. 6This is motivated by the fact that VAT threshold in Finland is quite low (e8,500), so that the estimated compliance costs (e1,300) are relatively more important than the incentives generated by the VAT rate. 4 simplicity and evasion responses, our findings might be evidence that real responses play a role as bunching remains large after excluding self-employed with reported turnover being multiple of one thousand (round-number bunching). Second, building on Kleven and Waseem (2013) and Harju et al. (2019), I develop a new theoretical framework that fits the institutional set-up to get a structural estimate of the elasticity of turnover. The type of discontinuity that I exploit is a non-standard notch. In the theory of notches by Kleven and Waseem (2013), the elasticity is estimated by solving the indifference condition of the “marginal buncher” who faces the same average tax rate above the threshold as every other agent.7That is because the cutoff and the tax base are both expressed in the same terms: taxable income. In our case, exceeding the cut-off of the preferential turnover scheme (F-regime) involves a joint change of tax rate and tax liability, but also a change in the tax base. Above the turnover threshold, agents are taxed on actual profits, so that tax incentives vary across individuals with equal turnover. Hence, to solve the indifference condition, we should consider the specific tax incentive that the marginal buncher faces. Lastly, this paper relates to the policy discussion regarding the opportunity of taxing different types of income differently by setting up preferential tax regimes for certain taxpayers. Adam and Miller (2021) discuss the different tax rules applying to wageearners, self employed and business owners’ income in the UK, and argue that preferential tax regimes could create inefficiency, unfairness, complexity and revenue losses for the government. This paper shows that this might also be the case in Italy: many solo selfemployed declare revenues up to the eligibility thresholds for the preferential tax regime. If that is due to tax planning/evasion, then the preferential tax regime is eroding the tax base and therefore causing revenue losses for the Treasury. If bunching is due to self-employed limiting their growth in sales, then the tax system is also encouraging businesses to remain small, which is potentially detrimental to economic growth. The rest of the paper is organised as follows. Section 2 outlines the institutional background and the data being used. Section 3 presents the methodology and the evidence of bunching on turnover, including the sector-specific analysis. Section 4 describes the theoretical framework that is used to estimate the turnover elasticity. Section 5 provides structural and reduced-form estimates of the turnover elasticity. Section 6 concludes. 2 Institutional Background and Data 2.1 Tax regimes for solo self-employed In Italy, self-employed have two options for income taxation: i) the ordinary tax regime; ii) one of the existing preferential tax regimes. The first includes the progressive personal income tax schedule (see table 1), social security contributions (see table 2), and VAT. 7The marginal buncher is the individual who is just indifferent between bunching and not bunching. 5 Sellers charge VAT on their sales, remit it to the tax authorities every three months, and claim back the VAT paid on inputs of production. The standard VAT rate was 21% in 2012-2013, 22% from 2014 onwards, and it applies to most goods and services.8 In the 2010’s, two preferential tax schemes – alternative to the ordinary regime – were introduced, allowing solo self-employed with turnover below a certain threshold to have tax advantages and simplified compliance procedures.9They provide lower income tax rates and/or a different tax base on which reduced rates apply. Moreover, a distinctive feature of these schemes is the exemption from VAT, meaning that the turnover cut-off to access these schemes coincides with the VAT registration threshold. Next, I describe two such schemes, alternative to the ordinary regime described above: 1) M- regime (2012-2015); 2) F-regime (from 2015 onwards). Table 1: Ordinary regime: Income tax rates 2012-2019 Personal Income Tax Rates Starting Basic Middle Higher Top Thresholds (e) 0 15,000 28,000 55,000 75,000 Tax rates 23% 27% 38% 41% 43% Table 2: Social security contributions Contributions Thresholds Year Category Variable Fixed Basic Middle Top 2012 Commerce 21.40% e3,200 e14,930 e44,204 e96,149 2013 Commerce 21.84% e3,360 e15,357 e45,531 e99,034 2014 Commerce 22.3% e3,460 e15,516 e46,031 e100,123 2015 Commerce 22.7% e3,540 e15,548 e46,123 e100,324 2016 Commerce 23.2% e3,610 e15,548 e46,123 e100,324 2017 Commerce 23.6% e3,680 e15,548 e46,123 e100,324 2018 Commerce 24% e3,790 e15,710 e46,630 e101,427 2019 Commerce 24% e3,830 e15,878 e47,143 e102,543 Notes: Commerce includes wholesale, retail trade and other self-employed. The contribution rate for Commerce applies between the basic and middle threshold, and then rises by 1 p.p. for any profit between the middle and top threshold. No contributions are due on profits exceeding the top threshold. Different contribution rates apply for members of professional associations. 8Italy has two reduced VAT rates: 4% for food and agricultural products; 10% for energy and gas used by households. 9These include an exemption from filing VAT reports and bookkeeping for income tax purposes. However, entrepreneurs must keep all documents they receive and produce for their transactions. 6 Table 3: Tax credits: 2012-2019 Tax credits Type Brackets (e) Amount (e) Self-employment 0 −55,000 55,000−TP 50,200 ×1,104 one child <(≥) 3 y.o h1−TP 95,000i×1220 (950)a two children <(≥) 3 y.o h1−TP 110,000i×1220 (950)a Non-working spouse 0 −15,000 800−(110×TP) 15,000 15,001 −40,000 690 40,001 −80,000 (80,000−TP) 40,000 ×690 Other share of expenditure Notes: Other includes the 19% tax credit for expenditures in healthcare services, gyms, university fees. This rises to 20% for high-efficiency refrigerators, 36% for renovations and 55% for energy-saving devices. Tax credits reduce the amount owed to the tax authorities. aIn 2012, the per-child amount was lower: e800 (e900) for <(≥)3 y.o. child. TP: Taxable Profits =Profits −social security contributions Preferential profit tax scheme: M-regime Between 2012 and 2015, solo self-employed with turnover below e30,000 could opt out of the ordinary regime and choose the M-regime. This scheme exempts entrepreneurs from VAT registration, annual VAT declaration to the tax authority, as well as record-keeping on clients, suppliers, purchases and payments. Then, the progressive PIT schedule is replaced by a proportional 5% tax rate on profits. Access to this scheme is limited to new businesses (no more than five years old) or until entrepreneurs are 35 years old. While the scheme was abolished in 2015, people already in and satisfying its requirements could keep it. Although the M-regime has no tax credits, the lower statutory profit tax rate, compared to the ordinary regime, is enough to make this scheme advantageous for most taxpayers. Hence, it is safe to assume that any taxpayer meeting the entry criteria would be better off in the M-regime. As the turnover threshold of e30,000 is not related to any other tax policy in 2012-2014, any excess mass of taxpayers below that threshold can be safely explained by the tax incentive of this scheme. Preferential turnover tax scheme: F-regime With its introduction in 2015, the F-regime replaces the M-regime as the main preferential tax scheme for solo self-employed. Table 4 shows the sector-specific turnover cut-offs that solo self-employed needed not to exceed if they want to choose this regime. The largest group of taxpayers – including lawyers, doctors, professors, architects and other professionals – faces the e30,000 threshold in 2016-2017. Then, the cut-offs were equalised to e65,000 across sectors in 2018 (autumn budget). 7 Fig. 3 (a) plots the turnover distribution of 2016-2017 over the one of 2013-2014 (used as counterfactual) and shows the excess mass below the main threshold at e30,000 is larger than the missing mass above it. Fig. 3 (b) plots the 2019’s distribution with that of 2016-2017, used as counterfactual: the 2018’s reform that raised the eligibility threshold of the regime to e65K gave solo self-employed an incentive to move up along the turnover distribution. Sector-specific evidence is presented in Appendix 7.4. The next figures focus on the e65,000 cut-off of the preferential turnover regime in 2019. Figure 4 shows the distributions of turnover of the different sectors. Figure 5 reports the values of the excess bunching coefficient bfor all sectors. It is defined as the ratio between the excess mass of taxpayers to the left of the threshold and the value of the counterfactual distribution at the threshold, and serves as a measure of how strong bunching is. We can see there are heterogeneous responses across sectors, with Professionals showing the highest bunching coefficient. Self employed in different sectors have different incentives to bunch for two reasons: i) some sectors are on average more profitable than others, and the self-employed in higher value added industries have a larger tax burden in the ordinary regime than lower value added ones, conditional on turnover; ii) the taxable share of turnover, which is the tax base in the preferential regime, is sector-specific. The incentive to bunch will therefore depend on the gap between actual profitability, which determines the tax burden in the ordinary regime, and the notional profits in the turnover regime. The higher actual profits for an individual, compared to the notional profits in the preferential regime, the higher the chances of bunching at the turnover threshold. Hence, the theoretical prediction is that bunching should be stronger in those sectors in which actual profits tend to be consistently higher than notional profits, as there would be more people that would potentially benefit from a lower tax base in the preferential turnover regime. To find whether this is actually the case, we compare the bunching coefficient of the different sectors with the difference between actual profit and notional profits for the median agent in the profit distribution. This theoretical prediction is supported by the data: there is a positive relationship between the extent of bunching and the difference in tax bases across regimes for the median profitability level. We observe more bunching in those sectors in which larger shares of taxpayers would have a larger tax base in the ordinary regime. Since tax incentives are different in different sectors, the standard bunching coefficient alone cannot be used to compare responses across sectors. Without accounting for different tax incentives at the threshold, it is not possible to disentangle large responses per se from large responses due to large tax-burden-gaps across regimes. One way to tackle this issue is to normalise the standard bunching coefficient by the tax gap between the ordinary and preferential regime. For that, we need to know the whole tax incentive that the agent faces, including VAT incidence if VAT is not neutral for the entrepreneur. 14 4 Theory 4.1 A frictionless model without extensive margin responses This section describes the theoretical framework that will be used to estimate the elasticity of turnover with respect to net-of-turnover-tax rate. Building on Kleven and Waseem (2013) and Harju et al. (2019), I develop a model describing agent’s behaviour around the turnover threshold of the preferential turnover tax (F) regime, in line with the evidence provided in section 3. Below (above) the threshold, agents are taxed on turnover (profits). This creates a non-standard notch in the tax schedule with a change of tax rate, tax base and tax liabilities. Following the bunching literature (Kleven 2016), preferences are represented by a quasi-linear utility function (exp. 2), with turnover yand consumption C. Turnover y generates disutility ϕ(y, n), that is increasing in turnover, but decreasing in the agent’s ability n. The elasticity of turnover with respect to net-of-tax rate is denoted by e. The production costs of generating turnover yare given by ci=fi+vi(y), with fixed and variable costs, fiand vi(y)respectively, that can be heterogeneous across agents. Each agent-type iis therefore identified by ability and costs: θi={ni, ci}. Ability ngoverns willingness wo work and therefore where in the turnover distribution an agent will be. Then, individual production costs cidetermine where in the profit distribution an agent is located, conditional on generating a certain level of turnover. U=C−ϕ(y, n),(3) ϕ(y, n) = n 1 + 1 ey n1+1 e.(4) Agents maximise utility Uby choosing how much to work, namely the level of turnover y, and face an upward notch at y∗. Below the cut-off y∗, agents have access to the preferential tax regime in which turnover is taxed proportionally at rate tB. While entrepreneurs don’t charge VAT to customers, they also cannot deduct VAT payments on inputs (citV). The effective tax on turnover in the preferential tax regime is therefore tP=tB+tV(ci/y). Above the threshold, agents are taxed on their profits Π, and a different tax schedule applies: tA(π)is the implicit average turnover tax rate (IATTR), that is the equivalent proportional tax on turnover that the agent would pay, given the actual profit tax schedule for self-employed. Moreover, if VAT is not fully passed on to selling prices, revenues are scaled down by 1 + α tVwhere αcaptures the split of the tax incidence between consumers and sellers.14 α= 0 means VAT is fully passed on to consumers, so that changes in VAT are irrelevant for the entrepreneur. The opposite 14Harju, Kosonen, and Nordström-Skans (2018) found evidence of VAT-non-neutrality among small restaurants in Finland and Sweden, as they did not alter prices after reductions in their VAT rate. 15 case is α= 1, when entrepreneurs bear the whole VAT burden. C=   y(1 −tB)−ci(1 + tV)if y≤y∗ y1−tA(Π) 1+α tV−ciif y > y∗ I make the following assumptions: 1) smooth distributions of ability (n), turnover (y) and profits (π); 2) turnover can be changed by changing output that is always demanded; 3) no optimisation frictions (frictionless model). Extensive margin responses are ruled out at this stage. Agent’s optimisation For an agent optimising to the left of the turnover cut-off (y≤y∗), the FOC is given by y∗=n[(1 −tB)−c′(y)(1 + tV)]ewhere tBis the preferential turnover tax rate and tVis VAT. Provided that variable costs do not change much for those affected by the turnover threshold of the preferential regime, the FOC simplifies to y∗=n[(1 −tB)]e.(5) Then, to the right of the cut-off, (y > y∗), utility maximisation yields the following FOC 1 1 + α tV   1−tA(πi)−c′(y) | {z } Direct effect of changing y −y t′ A(πi)∂πi ∂y  | {z } Indirect effect    =y ni1 e where ∂π ∂y = 1 −c′(y). Utility is raised by the net-of tax and net-of-marginal-costs part of additional turnover (direct effect). Then, changing turnover also varies profit π, and this affects tax liabilities, and therefore utility (indirect effect). The sign of the indirect effect depends on whether changing turnover at the margin increases or decreases profits. If profits go up (down), then the tax base is larger (smaller) and the tax liability increases (decreases), so that the indirect effect term is negative (positive). Provided that variable costs do not change much for those affected by the turnover threshold of the preferential regime, the FOC simplifies to 1 1 + α tV1−tA(πi)−t′ A(πi)y=y ni1 e .(6) Condition (6) implies that if two agents have equal turnover yat the optimum, but different profits, then the agent with higher profits πi, and therefore higher tax liability t(πi), must also have higher ability niand/or higher elasticity e. By allowing an imperfect negative correlation between ability nand individual cost function ci– therefore keeping niand cidistinct – one can account for heterogeneous elasticity as well as heterogeneous profitability across agents, conditional on a certain level of turnover. 16 Figure 6: Optimisation with multiple notched budget sets Heterogeneous Profitability and Incentives to bunch Agents with different profits (costs) have different tax incentives at the threshold. Suppose that two agents, Mand B, have equal turnover, but B has lower costs - therefore higher profits - than M. As earnings taxation above the threshold depends on profits, the implicit average turnover tax rate t(π)will differ for these two different profit-types. Figure 6 shows that agent M faces a lower average tax rate, tA(πM)< tA(πB), and therefore has larger consumption than agent B. While agent Mis just indifferent between bunching and remaining at the interior point yI(marginal buncher), agent B is going to bunch to get a higher payoff. Any other agent with turnover yI, but with profits lower than agent Mis not going to bunch at the notch point y∗. This also implies that knowing the position of the marginal buncher is not enough to isolate the relevant tax incentive. We shall return to this issue in the section devoted to the elasticity estimation. The multiplicity of budget sets above the threshold also implies that there can be multiple marginal bunchers at different levels of turnover. Hence, this notch does not create a clear dominated region. The decision to bunch will not just depend on preferences and turnover, but also on individual’s profits conditional on turnover. For the purpose of the elasticity estimation, we consider the last marginal buncher, namely the marginal buncher with the highest pre-notch turnover y, for whom we can estimate the proportional turnover response ∆y∗/y∗using bunching below the cut-off y∗. 4.2 Indifference condition To estimate the elasticity of turnover, I exploit the indifference condition for the last marginal buncher, whose utility from bunching at the threshold is given by Uy∗= (1 −tB)y∗−c(y∗)(1 + tV)−n 1 + 1 ey∗ n1+ 1 e 17 while at the best interior point, yI(and profits πI), the agent’s utility reads UyI=1−t(πI) 1 + α tVyI−c(yI)−n 1 + 1 eyI n1+1 e. With zero variable costs, UyIcan be rewritten using the FOC from (2) with c′(y) = 0. UyI=n1−tA(π) 1 + α tV1−tA(π)−yIt′ A(π) 1 + α tVe1−e 1 + e1−tA(π)−yIt′ A(π) 1−tA−c(yI). Using the agent’s FOC in absence of the threshold, y∗+∆y∗=n(1−tB)e, and assuming that c(y∗) = c(yI), we derive the indifference condition Uy∗=UyIfor the last marginal buncher: 1 1+∆y∗/y∗1−c tV (1 −tB)y∗−e e+ 1 1 1+∆y∗/y∗1+1/e −1 1 + α tV ·1−tA 1−tB1+e1−tA−yIt′ A(π) 1−tAe1−e e+ 1 ·1−tA−yIt′ A(π) 1−tA= 0.(7) where t′ A(π) = ∂tA(π) ∂π π=πI. As in Kleven and Waseem (2013), expression 7 characterises the relationship between the behavioural response of the marginal buncher ∆y∗/y∗, the average net-of-tax-rate in the two regimes 1−tA 1−tB, and the elasticity e. We also allow for imperfect VAT pass-through on selling prices, implying some of the VAT tax burden falls on entreprenuers, as in Harju, Matikka, Rauhanen (2019). However, differently from Kleven and Waseem (2013), and Harju, Matikka, Rauhanen (2019), agents face two alternative regimes that have different tax bases and tax rates around the turnover threshold. Agents are taxed on turnover below the threshold, and on profits if they are above it. Thus, expression 7 also includes the effect that changing turnover has on tax liabilities above the threshold via changes in profits. 5 Elasticity estimation Using the tax parameters and behavioural responses to the turnover regime’s 65K threshold in 2019, I estimate the elasticity by solving the indifference condition (7) for 5 sectors: Professionals, Other Activities, Business Intermediaries, Construction & Real Estate, Retail and Accommodation. This results in a system of 5 equations, one indifference condition for each sector, and 6 unknowns, given by the VAT incidence parameters αj in each group and the elasticity parameter. Hence, we need an additional restriction to solve the system of equations. In the next section, I describe how I tackle this issue. Moreover, since earnings taxation depends on profits, agents with equal turnover but different profits (costs) have different incentives to bunch. Hence, the position of 18 the (last) marginal buncher in the turnover distribution is not enough to isolate the tax rate in the ordinary regime tAto use in the indifference condition. For the elasticity estimation presented below, I consider the sector-specific taxable shares of turnover in the preferential regime as reference-profitability-levels to compute the tax burden in the ordinary regime. 5.1 Assumptions First, I impose that all VAT incidence parameters must lie between zero and one, i.e. 0≤αj≤1for any j. Then, I pick the group that has the highest estimated VAT incidence parameter among all sectors, for any value of the elasticity, and I set the VAT incidence parameter of that sector to unity. This means that in the estimates of the simultaneous parameters, we select the highest possible VAT parameters, compatible with our initial assumption. As choosing the highest VAT incidence parameters means that we consider the largest possible tax gaps between the two alternative regimes at the threshold, the resulting tax elasticity is a lower bound estimate. The group with the highest VAT incidence parameter is Retail and Accommodation. Therefore, I set αRetail = 1, meaning self-employed in this sector bear VAT entirely.15 Hence, I estimate the elasticity parameter, equal for all sectors, and the VAT incidence parameters such that 0≤αj≤1for any sector j. Lastly, for the baseline estimation, I assume that costs do not vary around the threshold, i.e. c(y∗) = c(yI). 5.2 Structural elasticity - results Table 6 presents the point estimates for the tax elasticity and VAT incidence parameters in the case of zero additional compliance costs in the ordinary regime, relative to the preferential regime (column 1 - baseline), and when the ordinary regime determines additional compliance costs of e1300, which is the estimate of Harju et al. (2019). As the largest simplifications of the preferential regime come from the VAT exemption, as in Harju et al. (2019), this estimate is also a good reference for the additional hassle costs of the ordinary regime in Italy. In table 6, column (2), the estimated elasticity is lower than in the baseline scenario as behavioural responses are now partly explained by the additional hassle costs in the ordinary regime. Finally, figure 7 reports the 95% bootstrap confidence intervals for the VAT incidence parameters. 15There is empirical evidence that this is the case in Finland and Sweden: Harju, Kosonen, and Nordström-Skans (2018) found evidence of VAT-non-neutrality among small restaurants, as they did not alter prices after reductions in their VAT rate. 19 Table 6: Structural estimates (1) (2) Turnover Elasticity 0.072 0.043 VAT incidence Sector parameters (α) Retail & accommodation 1 1 Business intermediaries 0.788 0.813 Other activities 0.518 0.531 Professionals 0.339 0.366 Real estate 0.125 0.161 Compliance costs - e1300+ +Compliance costs estimate by Harju et al. (2019) Note: to obtain these estimates, I solve condition (7) by using the estimated turnover responses and the observed values for tA(π), tB, tV, t′(π), yIfor the e65K threshold of the F-regime in 2019. . Figure 7: VAT incidence parameters αPαOαRE αBI αRA −0.25 0 0.25 0.50 0.75 1 Point estimates with Bootstrap 95% CI 5.3 Structural vs Reduced-form elasticity Following Kleven (2018), adapted to our case to include VAT incidence, I compute the sector-specific reduced-form elasticities {eP,eO,eRE,eBI ,eRA}. To do so, I use the bunching estimates for the proportional change in turnover (∆y∗/y∗)for each sector as well as the VAT incidence parameter estimates from section 5.2 to compute the full tax gap ∆tbetween the ordinary and preferential regime. Figure 8 reports these estimates next to structural elasticity e, estimated as described in section 5.2. Except for the real estate sector, all reduced-form sector-specific estimates are within with the confidence interval of the structural elasticity. This suggests that the model described in section 4.1, although simple, is able to capture the key incentives at play in this setting. e≈1 2·(∆y∗/y∗)2 ∆t/(1 −tB),∆t=tA−tB+VAT incidence. (8) 20 Figure 8: Structural and reduced-form elasticity estimates e ePeOeRE eBI eRA 0 0.05 0.1 0.15 0.20 Point estimates with Bootstrap 95% CI 5.4 Discussion The evidence on turnover responses shows that some solo self-employed adjust their turnover to locate themselves below the eligibility cut-off for the preferential regime. After accounting for the tax incentives in different sectors, the largest responses come from professionals and business intermediaries. Responses could reflect changes in productive effort (labour supply), but at this stage it is not possible to exclude the hypothesis that evasion might explain part of the adjustments in turnover. Other authors (e.g. Aghion et al, 2022) have argued that the simplest evasion strategies would involve reporting turnover as a round number at, or very close to, the eligibility threshold. The facts that bunching is often quite dispersed below the threshold, and that responses remain large even after omitting observations that report turnover as a multiple of 1000, would therefore be consistent with real responses. Another issue is whether the introduction of the preferential turnover regime reduces tax revenues for the government. Answering this question would require us to know the following: i) how much do self-employed adjust turnover, i.e. how large bunching is; ii) how large is the inflow from the ordinary to the preferential regime for those taxpayers that are already below the preferential regime threshold; iii) how large are extensive margin responses. The first two channels would have a negative impact on tax revenues, while the third one would have a positive effect as new economic activity generates additional tax revenues. This paper provides evidence on the first point and partly on the third one. 21 6 Concluding Remarks This paper investigates to what extent solo self-employed adjust sales turnover due incentives of the tax system. I study the turnover responses to the notches created by the eligibility cut-offs of the preferential tax regimes for solo self-employed in Italy. I find that solo self-employed bunch below some of the turnover thresholds, set by the tax code, to qualify for a preferential tax scheme. Then, I adapt the models of Kleven and Waseem (2013) and Harju et al. (2019) to derive a modified indifference condition that fits the institutional set-up and I use it to estimate the turnover tax elasticity. To do that, I exploit the behavioural responses to the turnover threshold of the preferential (turnover) regime in 2019. The estimated elasticity is small but larger than zero. 22 7 Appendix 7.1 Derivation of the Indifference condition For the last marginal buncher M, utility from bunching at the threshold is Uy∗= (1 −tB)y∗−c(y∗)(1 + tV)−n 1 + 1 ey∗ n1+ 1 e Then, at the best interior point, yI, with profits πI, the agent’s utility reads UyI=1−t(πI) 1 + α tVyI−c(yI)−n 1 + 1 eyI n1+1 e Using the FOC, 1 1+α tV[1 −tA(π)−t′ A(π)y]=(y/n)1/e, we can rewrite UyIas UyI=n1−tA(π) 1 + α tV1−tA(π)−yIt′ A(π) 1 + α tVe1−e 1 + e1−tA(π)−yIt′ A(π) 1−tA−c(yI) We assume that variable costs are not crucial for the decision of bunching around the threshold, that is c(yI) = c(y∗)so that setting U∗ y−UyI= 0 gives (1 −tB)y∗−c(y∗)tV−n·e 1 + ey∗ n1+1 e −n1−tA(π) 1 + α tV1−tA(π)−yIt′ A(π) 1 + α tVe1−e 1 + e1−tA(π)−yIt′ A(π) 1−tA= 0 Divide all terms by n, and use the agent’s FOC in absence of the threshold, y∗+ ∆y∗= n(1 −tB)e. Finally, after pre-multiplying the condition by 1/(1 −tB)1+eand collecting terms, we can rewrite the indifference condition as 1 1+∆y∗/y∗1−c tV (1 −tB)y∗−e e+ 1 1 1+∆y∗/y∗1+1/e −1 1 + α tV ·1−tA 1−tB1+e1−tA−yIt′ A(π) 1−tAe1−e e+ 1 ·1−tA−yIt′ A(π) 1−tA= 0. where t′ A(π) = ∂tA(π) ∂π π=πIand c=c(y∗). 7.2 Appendix A - Ordinary tax regime The ordinary tax regime includes the progressive personal income tax (IRPEF) schedule, social security contributions (SSCs), and VAT. 23