Austerity Plans and Tax Evasion : Theory and Evidence from Greece

1 Austerity Plans and Tax Evasion : Theory and Evidence from Greece Francesco Pappadà Yanos Zylberberg January 16, 2014 Abstract The austerity ...

Author:  Melina Wilkerson
0 downloads 0 Views 719KB Size

Austerity Plans and Tax Evasion : Theory and Evidence from Greece

Francesco Pappadà

Yanos Zylberberg

January 16, 2014

Abstract The austerity plans implemented in Greece in 2010 have yielded lower than expected increases in tax receipts. We argue that this has been the result of the arbitrage that rms face when choosing to declare their activity. A tax hike has a direct eect on the degree of tax evasion, and an indirect one through credit markets. A tax increase tightens the credit constraints of rms and depresses even further their incentives to be transparent. Using a dataset of about 30'000 Greek rms per year over the period 2002-2011, we provide evidence that rms adjust their declared protability, and this adjustment depends on the tax burden and their need for credit. We then calibrate our model and show that leakages due to tax evasion are quite high : a 21% increase in tax rates only delivers a 7% increase in tax receipts. The response of transparency generates an additional investment slack which is the result of a contracting demand for credit by small and medium size rms induced by tax evasion.

JEL Classication Codes: E44, O17, H26. Key words: tax evasion, austerity plans, credit frictions. ∗

University of Lausanne. Part of this research was conducted by Francesco Pappadà during his visiting at

the University of California, Berkeley. E-mail: [email protected]

CREI (Universitat Pompeu Fabra), Ramon Trias Fargas, 25-27 08005 Barcelona.



[email protected] The research leading to these results has received funding from the European Research Council under the European Union's Seventh Framework Programme (FP7/2007-2013) / ERC Grant agreement nº 241114.


1. Introduction Following the sovereign crisis in late 2009, Southern European countries (Greece in particular but also Portugal, Italy and Spain) plunged into a deep recession and a severe political crisis. In all those countries, the response to the sovereign debt crisis consisted in large scal adjustments in order to reduce immediate decits and ultimately, get further from the threatening debt ceiling. These adjustments were accompanied by a strong economic contraction and were insucient to consolidate the primary balance compared to what was expected. Political crises in Greece, Italy or Spain emerged from the discrepancy between the popular sentiment that austerity was dampening the economic slack and the sequence of even more stringent policies adopted by the governments. Since these austerity plans were the key condition for having access to bail-out programs of international nancial institutions1 , people had the feeling that austerity was intended as a punishment from outsiders rather than a cure. We provide in this paper a very simple argument for the failure of austerity plans, particularly when they rely heavily on tax increases rather than cuts in government spendings2 : in presence of imperfect tax enforcement, the decision to declare activity results from an arbitrage between improved access to credit and a lower tax burden. We provide evidence that Greek rms adjust their declared protability, and this adjustment depends on the tax burden and their need for credit. An austerity plan distorts this arbitrage through (i) an increased tax burden and (ii) lower gains from transparency. The behavioral response to higher tax rates is twofold. First, for a given level of transparency, a higher tax burden reduces future pledgeable cash ows and tightens credit constraints (taxes are senior to debt). Second, access to credit markets is less protable, thus rms have less incentives to be transparent.3 To understand the decomposition of these behavioral responses, consider the following accounting exercise. Let τ denote the tax rate paid by rms on the reported value added γv , where γ is the reported share of value added

v . Suppose that the government wants to generate a scal surplus through an increase of value 1 This was the case for the two nancial packages delivered to Greece in 2010 and 2012. 2 For many reasons (some of them political), over-indebted countries among the GIIPS like Portugal, Greece or Italy implemented scal reforms based mainly on tax increases.

3 In addition, the marginal tax revenue generated by a marginal tax increase is low when the declared tax base is low. This mechanical consequence of tax evasion induces government of such countries to climb even further the Laer curve to extract a surplus, exposing themselves to large behavioral responses.


added tax rate (VAT) and ultimately tax revenue dT R. The impact of this scal policy is :

dT R = γvdτ + γτ dv + τ vdγ | {z } | {z } | {z } dM



We argue that the behavioral response, composed of the standard dB and our transparency component dE alleviates most of the mechanical response dM . These estimations are in line with the observed discrepancies between the targeted and actual tax revenues on rms collected by the Greek authorities during this period.4 For instance, Greece planned a scal adjustment of 6 points of GDP in 2010 (from 15.4 in 2009 to 9.4), decomposed into expenditure cuts (2.9 points of GDP) and an increase in tax revenue (3.1 points of GDP). Greek authorities increased VAT accordingly (from 9 to 11 percent for the basic rate and from 19 to 23 percent for the high rate) but only collected a surplus of 1.5 points of GDP. We build a very stylized model with heterogeneous credit-constrained rms and a passive government implementing an exogenous VAT tax shock. In order to account for the entrepreneur's trade-o between credit and tax burden, we assume that the choice of transparency, i.e. the proportion of declared plants, determines both the tax receipts and the cash ows that can be pledged to investors. In our model, a tax increase will have two eects. First, some small rms will not nd it protable anymore to be transparent and get access to credit. The reason is that there is a modern technology that necessitates a xed investment. When credit constraints tighten, small rms cannot borrow enough and make this xed investment protable. Their response is to hide their activity completely. Second, medium-size rms will still nd protable to have access to credit but they show less than before. The aggregate implication of our model is that the transparency of the economy decreases adding to the direct recessionary eect of higher taxes. We then calibrate the model using a dataset (balance sheets) of 30'000 Greek rms and show how costly it is for the government to levy VAT taxes as a function of institutional parameters, such as the protection of lenders and the tax monitoring. We provide a measure of the performance of austerity plans through their direct and indirect eects : (i) direct losses from poor enforcement (the internal revenue service is unable to collect tax receipts), (ii) indirect 4 The Greek prime minister Lucas Papademos declared in an interview to

Il Sole 24 Ore

on March 30th

2012 that the ght against tax evasion has yielded limited results partly because of the greater than previously forecast contraction of the economy. We argue that it was due to the greater than forecast contraction of the




eects through the distortion induced by taxes and the incentives of informality when nancial development is low. We show that the impact of austerity plans can be related to the interaction between fundamentals of the economy  the protection of lenders or tax monitoring  and the distribution of rms' size. To be more precise, the major channel through which a tax hike aects the economy is through small-medium rms becoming more informal. The amplitude of this eect depends on the share of activity generated by those marginal rms. Southern European countries are economies in which those eects are large: rather weak institutions are coupled with a large fraction of small-medium rms at the margin of formality. In the United States, nancial development and tax monitoring are more developed and rms ta the margin of informality would be much smaller. In developing countries, tax enforcement is poor but the distribution of rms is bimodal with few very large rms and a multitude of very small businesses. In both cases, we would expect our behavioral response to be lower. A critical mechanism of our model is the possibility for rms to strategically adjust the extent to which they declare their activity. We nd empirical support for this mechanism in our data. We show an empirical regularity for rms subject to high tax pressure: Their protability (ratio sales/total costs) jumps immediately before having access to credit. We do not observe this excess protability before credit in sectors with low tax pressure (with the lowest VAT rate, and very exports-oriented). We interpret this observation as indirect evidence that rms strategically modify their transparency, i.e. the size of their declared activity, depending on their needs for external nancing. The stylized facts on the correlation between credit access and tax evasion come from a very dierent strategy than the literature. We use the fact that access to credit is preceded by exceptional peaks in rm's protability, particularly in sectors with a high tax pressure. Doing so, we dier from Kleven et al. (2011) and Cai and Liu (2009), who estimate tax evasion using two reporting sources of income and the concordance between them. While most of the literature focuses on personal income tax,5 we rather focus on corporate tax evasion. This entails one major dierence: corporate tax evasion crucially aects the extent to which rms borrow on nancial markets. In that respect, we relate to Artavanis et al. (2012), which oers estimates of tax evasion based on bank's perceptions of true income in Greece. Their idea is that banks anticipate how reported income from borrowers maps to their real 5 See Andreoni et al. (1998) and Slemrod and Yitzhaki (2002) for a review.


income. Occupations characterised by high tax evasion are those which are oered large loans relatively to their reported income. Our exercise is very dierent in nature but builds on the intuition that the activity declared to banks is closely tied to reports made to tax authorities. The fact that reported activity also inuences access to nance has received some theoretical support from Straub (2005); Desai et al. (2007); Ellul et al. (2012). Firms face a trade-o when choosing their transparency: they can avoid taxes at the expense of access to credit. More generally, the literature has long established that rms can adjust the extent to which they declare their activity. In Cai and Liu (2009), the more competitive is the environment, the more reported prots dier from their imputed counterparts. Our exercise also departs from these studies since we provide a macro-estimate of the response of tax evasion to a change in tax pressure (austerity plans). Our modeling of a dual technology world with a modern and a traditional technology relates to studies of shadow economies.6 We depart from Rauch (1991) and Straub (2005) as we allow rms to adjust their degree of informality. In our setup, rms can decide to operate in the traditional sector, in which case access to credit is not needed. Accordingly, they operate as if they were completely informal. However, rms can also operate in the formal economy without being fully transparent. In this paper, we also estimate the degree to which rms switch from the formal to the informal sector following a tax hike. Lemieux et al. (1994) provides such estimates for individual labor supply in the shadow economy. Finally, to our knowledge, this project is the rst one which models the macroeconomic cost of an austerity plan in the presence of tax evasion. Nonetheless, the response of an economy to a tax shock has been extensively studied. Among others, Romer and Romer (2010), Ilzetzki et al. (2010), Favero et al. (2011), Auerbach and Gorodnichenko (2010), Alesina and Ardagna (2009) have tried to estimate a scal multiplier, some articles focusing on the identication of dierences across countries, some other on how these multipliers might vary depending on the type of scal shock considered. The paper is organized as follows: in section 2, we present the stylized facts on tax evasion that motivate the theoretical framework. In section 3, we introduce a model of transparency choice and credit access, where we detail the arbitrage faced by rms when declaring their activity. In section 4, we calibrate our model using the empirical evidence from Hellastat; we conduct 6 See Enste and Schneider (2000); La Porta and Shleifer (2008) for a review.


numerical simulations to assess the impact of tax evasion and credit market frictions on the eectiveness of austerity plans. We then discuss some of the predictions of our model for the aggregate economy and for the distribution of credit among Greek rms. Finally, section 5 discusses some extensions and section 6 briey concludes.

2. Tax evasion and credit access In this section, we discuss the trade-o that is faced by rms when deciding to evade taxes. This trade-o is the building-block of our theoretical argument. We rst describe corporate tax evasion and extract an indicator of protability (the ratio of prot to sales) that is related to rm's transparency. Then we provide evidence that access to credit is preceded by abnormally high values for this indicator: just before contacting lenders, rms declare more of their activity, which generates a sudden peak of observed protability. Since we only observe this empirical regularity for rms subject to heavy tax burden, we take this observation as indirect evidence that rms in sectors with high tax pressure adjust their degree of transparency depending on their nancial needs.

A. Corporate tax evasion In general, media outlets focus on personal tax evasion, revealing for instance the existence of hidden Swiss bank accounts, or corporate tax avoidance, e.g. rms legally avoiding taxes by settling in a scal paradise. In this paper, we rather concentrate our focus on corporate tax evasion, which encompasses all illegal methods that reduce the corporate tax burden. Generally, corporate taxes consist in (i) a prot tax, and (ii) a VAT. In Greece, the corporate income tax (prot tax) is a at rate on net operating income (sales net of total costs of production).7 The VAT is a traditional tax on value added, and exported goods are thus not taxed. The VAT rate depends on the category of the produced good. The benchmark rate is 23% and concern most of the nal goods. There exists a reduced rate of 13% that applies to fresh food and medicines. 7 Over the period 2004-2011, the tax rate has decreased from a 32% in 2004 to 29% in 2006, and then 25% in 2007. From 2010 onward, a decrease of 1% per year is planned to reach 20%. Capital gains are taxed as regular income but there is an additional withholding tax of 10% on corporate dividend that applied starting from 2009.


Cultural goods and hotel accommodation benet from a discount rate of 6.5%. Insurance, educational, legal and medical services are exempt from VAT.8 There are two main frauds that are used by rms to evade taxes:

ˆ rms conceal or under-report sales. Reporting only part of their activity or, in the extreme case, avoiding any formal registration allow rms to escape both the prot tax and the VAT. In Greece, most of the self-employed (lawyers, doctors, plumbers, electricians...) and small businesses (street shops, restaurants...) that would be subject to registration do not comply despite an increasingly aggressive policy from tax authorities. In the same vein, it is possible to report some category 1 goods that are subject to high VAT rates (23%) to discounted categories (2 and 3, respectively 13 and 6.5%).

ˆ rms can also inate their operating costs, which reduce the income on which the prot tax is deducted. Typically, such outcome is achieved by over-reporting payments of intermediate goods; overstating wages is, for instance, a simple way to articially increase costs. In both cases, tax evasion is associated to low ratios sales/total costs. We refer henceforth to this ratio as the rm's


How can tax authorities identify anomalies in this observed

quantity? Sudden drops in rm's protability or permanently low ratios sales/costs without bankruptcy point to potential frauds. We build on this observation and ask the question of what would occur if a rm suddenly needs to declare its activity. We would then expect a sudden jump in this rm's indicator of protability. Why would rms need to be transparent? Misreporting sales and operating costs of production may induce diculties in the capacity of rms to raise funds and borrow. Articially weak rm fundamentals increase the borrowing costs and reduce the availability of external funds. Consequently, reporting a large part of its activity is a requirement for access to credit. In the following subsection, we exploit balance sheet data from Greek rms and investigate how anomalies in corporate protability immediately precede credit access. 8 In addition, some areas in Greece, essentially the islands, are subject to a specic tax regime with lower rates for each category.


B. Anomalies in protability and credit access We present in this section our empirical strategy. Contrasting with Kleven et al. (2011) and Cai and Liu (2009) for instance, we cannot use the discrepancies between two sources of reporting income in order to identify tax evasion. We only observe accounting reports and cannot rely on any auditing information. Accordingly, we cannot fully ensure that anomalies in such reports are reporting anomalies, including tax evasion, or that they also reect real changes in rm's activity. In order to investigate the link between transparency and credit access, we rely on rm-level balance sheets data from Hellastat.9

This dataset consists in comprehensive balance sheet

information of Greek rms over the period 2001-2011. Firms have to publish their balance sheets whenever two of the following three criteria are fullled : (i) Turnover: 3 million, (ii) Total Assets: 1.5 million, (iii) Average sta: 50 people. We therefore observe the universe of registered rms above these thresholds in Greece. We also observe smaller rms that publish their accounts on a voluntary basis. We are aware that the nature of data is such that we miss the tax evasion decision of very small rms and self-employed. However, it is very dicult to collect data on these small businesses because they simply do not appear in business registers. Although we do not observe fully informal rms, our data include rms that are mostly self-nanced and operating in sectors plagued by tax evasion. These rms publish their accounts but adjust their transparency depending on their nancial needs, the monitoring pressure and the tax environment. After cleaning the data for missing observations, we are left with more than 25'000 rms per year. The dataset is an unbalanced panel and we cannot assess the status of entrant/exiting rms.

Our empirical strategy relies on the following intuition: abnormal variations in rm protability that precede the access to credit might reveal a transparency choice of the rm. One might argue that it is not very surprising that rms behave dierently just before contacting lenders; they could have experienced an idiosyncratic productivity shock for instance. Our ndings are a bit more subtle. We show that only rms subject to high tax rates behave dierently immediately before the loan. Our methodology can be considered as a dierence-in-dierence, 9 We thank the research director of the Foundation for Economic and Industrial Research (IOBE), Aggelos Tsakanikas, and Evaggelia Valavanioti for giving us access to these data.


comparing treated groups (high VAT) to non-treated groups (low VAT) in treatment periods (just before a loan) against non-treatment periods (the other periods). First, we construct

protability Pi,t

of rm i in period t as the ratio of sales to operating costs.

Second, for each rm, we identify the year of the largest growth of loans over the entire period and we dene a dummy We then regress

credit access Ci,t


equal to 1 in this specic year.10

in period t on lags and forwards of

credit access,

and control for

rm µi , industry×year ηind,t xed eects. This specication allows us to extract the evolution of


around the access to loans, cleaned of rm-specic heterogeneity and cleaned of

the industry-specic evolution.

Pi,t =

2 X

πτ Ci,t−τ + ηind,t + µi + εi,t

τ =−2

Letting Ti denote the period at which rm i gets access to credit, i.e. Ci,Ti = 1, then the coecient π0 is the gap between expected rm protability in Ti and its protability over the period. πτ is the gap between expected rm protability in Ti + τ and its protability over the period.11 Figure 1 displays the coecients πτ with their 95% condence interval for the full sample of rms. Two periods before having access to credit, rm protability is very close to its average level. One period before the loan, protability jumps .01 above its average, then drops below the average in the period contemporaneous to credit access, and nally reverts to the mean one period after. Our main point is to argue that this empirical regularity should not be recorded for rms having little reason to conceal activity. How does the previous shape change along the dierent tax regimes? In gures 2 and 3, we reproduce the exercise for 2 decompositions into subsamples 10 We also consider alternative denitions of credit access without any dierence for our results: the year in which the rm switches its loans from 0 to a positive amount, the year of the largest growth in loans over the entire period, the year of the largest growth in leverage, the year of the largest growth rate of loans over the entire period, the year of the largest growth rate of leverage over the entire period, and the year when loans have increased by at least 25%.

11 π

τ is the expected protability conditional on credit being granted in period

πτ = E[Pi,t |Ci,t−τ = 1] − µi As a conclusion,


is the dierence between


and the loan period.


t − τ,


Ci,t−τ = 1.

of rms facing dierent tax pressure. In gure 2, we divide the sample into non-tradable vs tradable sectors. Non-tradable sectors are dened as sectors where rms do not export.12 In tradable sectors, rms are less concerned by VAT on their produced goods, because VAT on exported goods is reimbursed.

In gure 3, we focus on rms in sectors with a lower rate

(category 2 and 3) against rms in category 1. In all subsamples, π−2 , π0 , π1 , π2 are the same. First, protability coincides with its average two periods before the loan and after the loan. Second, rm protability contemporaneous to the loan is always below average. The unique date in which the protability dynamics diers across subsamples is one period before credit access. For rms subject to high tax pressure (high rate or non-tradable), rm protability is above its average. For rms subject to low tax pressure (low rate or tradable), rm protability is close to its average. Consequently, apart from the contemporaneous drop in protability (common to all rms), rms subject to low VAT do not exhibit any signicant deviation from the average. Only rms subject to high tax pressure exhibit excess protability immediately before being granted credit. We interpret the previous observation as evidence of a transparency margin. Firms face a trade-o between paying taxes and having access to credit. Declaring a larger fraction of its activity increases observable rm protability and access to credit at the expense of higher VAT payments. Naturally, when tax pressure is low (low rate or tradable), this trade-o is not relevant and rms declare activity more frequently: they do not need external incentives such as credit.

3. A model of rm transparency and access to credit We build on the previous empirical regularity and develop a theoretical model that captures the trade-o between paying taxes and being granted access to credit. There are three crucial ingredients in our model. First, we allow rms to choose the extent to which they declare their activity.

Second, since access to external nancing is conditional to the existence of

pledgeable capital, concealing activity reduces the capacity to levy funds. Third, we introduce two technologies, one linear (the traditional technology), and the modern technology that is more productive but requires an operating xed cost. We do so in order to capture that very 12 See table in the appendix.


small rms are not able to levy sucient funds for investment in the modern technology to be protable. Accordingly, they prefer to operate in the informal sector with the traditional technology and without external nancing. Importantly, we model rms' decisions for a given endowment: we ignore the dynamics that explain rm size distribution.

A. Environment The economy is composed of a mass of homogenous rms producing competitively a unique nal good that will be the numeraire. Firms are endowed with ω . Let G(.) denote the cumulative distribution of those endowments. Each rm is organized in a unit mass of homogeneous plants. The plants or establishments are homogenous in the sense that entrepreneurs cannot use a dierent technology or a dierent investment across their plants. We assume however that entrepreneurs can choose the fraction of plants whose value added is concealed.

Each plant is either fully declared or informal.

Let γ denote the fraction of declared plants (thereafter

transparency ).

Firms produce using

capital as unique factor, which they borrow from a competitive nancial intermediary sector. The entrepreneurs have access to two technologies: a traditional one and a modern one. The access to the modern technology is conditional on paying a xed innovation cost c. She then produces with a Cobb-Douglas function: y = Ak α . The traditional technology is available to all entrepreneurs. We assume that its returns are linear and equal to the international interest rate r . Once production has taken place, rms pay a tax rate τ on the reported value added, i.e. the value added generated in the declared establishments. Taxes are collected by a tax authority, which has access to an audit technology. The tax authority detects an informal plant with probability z(ω).13 In case of auditing, rms pay the tax θτ on the concealed value added. The punishment for being detected consists in the payment of an extra tax θ ≥ 1, which is set by the government. In order to get rid of idiosyncratic risk due to the random monitoring process, we assume that each establishment can be monitored with a random probability z by the tax police. The punishment implied by tax enforcement is then deterministic: a proportion of activities z(ω) is always audited. The total amount of taxes paid by rms is equal to the taxes on declared value added τ γv , and the amount z(ω)θτ (1 − γ)v 13 We assume here that endowment is observable: tax authorities may have imperfect signals on the rm's size.


paid to tax authorities after controls. How do rms nance their investments? We assume that they borrow from international markets at a xed interest rate r . However, there exists a nancial friction arising from the imperfect pledgeability of rms' cash ows: creditors can only seize a fraction of entrepreneur's endowment in transparent plants. Taxes are junior to this recovery process. In the end, entrepreneurs can only pledge a fraction λ of the endowment stocked in their declared establishments. They reimburse as long as the amount that they need to pay is lower than the capital that can be seized by lenders.

λγω ≥ (1 + r)(k − ω) The timing of actions is as follows. Entrepreneurs rst decide on their level of transparency, which is going to jointly determine how much value added can be pledged to lenders and how much will be taxed by the government. They borrow capital k at the international interest rate subject to their pledgeability constraint. Then, they decide to use the modern technology or the traditional one, they produce, reimburse their creditors and pay taxes, or nes if any. We have not specied yet whether rms could become lenders.

We assume that credit is

fully transparent and taxed at the same rate τ , such that rms always prefer to invest in the traditional technology, rather than lending.

B. Choice of tranparency We now derive the choice of entrepreneurs. For the sake of clarity, we display the analytical expressions with z(ω) = 0 (no tax auditing from the government). In Appendix B, we derive the expressions for a more general auditing schedule z(ω). We also propose a simple model with endogenous auditing that generates auditing probabilities increasing with observed rm size. Few preliminary remarks help us derive the behavior of entrepreneurs. First, there exists a threshold below which endowment is too low for investment in the modern technology to be sucient and cover the xed costs. Second, there exists another threshold above which the marginal investment is lower than returns on the traditional technology. Consequently, very rich and the very poor entrepreneurs are not willing to use external nance. Under which condition do entrepreneurs invest in the modern technology? As long as the marginal returns on the modern technology are higher than the marginal returns on the tradi12

tional one, entrepreneurs would gain in borrowing. However, there is an additional cost: rms need to declare their activity and pay a tax on their declared production. The minimal level of investment k upon which entrepreneurs start to use their innovation is the wealth for which entrepreneurs are indierent between operating in the traditional sector and using their innovation, given that they need to be fully transparent to do so. k is the smallest solution to the equation

A(1 − τ )k α − c = rk Let ω denote the minimum wealth that allows to levy k , i.e. λω = (1 + r)(k − ω). Any entrepreneur with ω ≥ ω could do better by being fully transparent with the modern technology than operating in the informal sector with the traditional one. Under which condition do entrepreneurs invest in the modern technology without any recourse to external nance? Absent credit frictions, entrepreneurs that have access to the modern technology would conceal all their establishments and borrow such as to maximize Ak α −c−Rk . 1

Let k ∗ = (Aα/r) 1−α denote the unconstrained solution. Rich entrepreneurs with endowment ω above ω = k ∗ , are unconstrained and thus set transparency equal to 0. They invest ω in the modern technology and the residual ω − ω in the traditional one. Those two quantities (ω, ω) allow us to isolate two zones in which the entrepreneur decision is simple. For ω < ω , even a full transparency would not allow the entrepreneur to generate any surplus from borrowing. Accordingly, small rms are better o concealing their activity and renege on any loan. For ω ≥ ω , entrepreneurs are able to levy the optimal capital without relying on external creditors. Consequently, they conceal all their establishments.14 Finally, rms whose endowments are between ω and ω invest in the modern technology and their program can be written as follows:

max {(1 − τ γ)Ak α − c − r(k − ω)} γ,k

subject to

λγω ≥ (1 + r)(k − ω) ˆ the solution to this program. Investment kˆ is the result of a trade-o between benetDene k ting from the high returns in the modern technology, and the cost that it represents in terms of transparency. In order to borrow an additional unit, which generates Aαk α−1 , the rm needs 14 Naturally, this result hinges on the hypothesis that

z(ω) = 0.


to declare part of its activity and pay taxes (second term in the square brackets below). The dierence between the gain and the cost should be equal to the price r of borrowing.

Aαk Let γ ˆ=

ˆ (1+r)(k−ω) λω


(1 + r)τ 1− λ

 1+αk −1 =r α ω


denote the associated transparency.15 The production of rms, drawn in the

second panel of gure 5, is:

   rω if ω < ω   ˆ α − c if ω ≤ ω < ω y= A[min{ (λ+1+r)ω , k}] 1+r     A¯ ω α + r(ω − ω ¯) − c if ω ≥ ω Transparency follows a non-monotonic relationship with endowment. Before threshold ω , it is equal to 0. It jumps to 1 on the threshold, with entrepreneurs obliged to show all plants such as to make the investment in the modern technology protable. Entrepreneurs borrow such as to bridge the gap between their wealth and the optimal investment. Immediately after ω , they need to bridge a very large gap, such that they declare everything. As their endowment grows, the gap becomes smaller and smaller, implying a lower transparency. It then reaches 0 for rms that do not require external nancing. Finally, remark that rms are rationed if their net worth is not sucient to reach k ∗ : investing up to k ∗ requires to make an eort in terms of transparency. Accordingly, the marginal returns on the modern technology are lower than r even for an investment equal to k ∗ . Consequently, entrepreneurs with ω < k ∗ will not borrow entirely up to k ∗ . We turn now to the properties of this allocation.

C. Predictions How do aggregate investment and aggregate tranparency depend on the fundamentals of the economy? First, for rms investing in the modern technology, a simple analysis of equation (T)

ˆ increases in capital pledgeability λ and decreases in taxes shows us that optimal investment k 15 It could be that the solution to this equation implies that transparency is greater than 1. In this case,

  k = min{ (λ+1+r)ω , k} ˆ 1+r  γ = min{1, γˆ }


τ , for any initial endowment. This observation comes from the fact that, if β denotes then

(1+r)τ , λ

k + α(k − ω) ∂ kˆ =− <0 ∂β α(1 − α)(1 + β) ωk + (1 + α)αβ

The intuition is that more tightening credit constraints (lower levels of λ) increase the cost of capital, and force rms to declare more in order to borrow a given amount. The leverage of rms is lower, and so is level of transparency. The impact of an increase in tax rates is qualitatively similar to a decrease in λ. In parallel, there is also a change in the threshold ω when nancial development decreases (or taxes increase): less rms decide to operate in the formal sector and only invest their endowment in the traditional technology. Building on this result, following an increase in nancial development, some rms nd it profitable to get access to the modern technology at the expense of tax evasion (extensive Firms that were already transparent can now borrow more (intensive

margin ).

margin ).

The size of the

aggregate eect depends on the distribution of rms. To be more precise, it depends on the number of rms around the threshold ω and the number of rms between ω and ω . Those quantities determine the weights of the extensive margin eect and the intensive margin. What happens after an increase of taxes?

On the extensive margin, an increase in taxes

induces some transparent rms in the modern sector to hide all their activity and operate in the traditional sector. On the intensive margin, rms see the investment in the modern sector less protable than before, and reduce their investment together with their transparency. We provide below the decomposition between the dierent eects. The aggregate production

Y in the economy is : Z Y =


Z rωdG(ω) +


ω ¯

Z h i α ˆ Ak(ω) − c dG(ω) +


[A¯ ω α + r(ω − ω ¯ ) − c] dG(ω)


The aggregate production of rms that do not access the credit market is

R∞ ω

Rω 0

rωdG(ω) +

rωdG(ω). Since these rms choose to conceal all their production (γ =0), the government

does not manage to levy any taxes on them. The aggregate tax base is therefore :

Z Yτ =


ˆ α dG(ω) γˆ (ω)Ak(ω)


Following an innitesimal change in taxes dτ , changes in aggregate tax base dYτ can be decomposed as follows:

dYτ = dYτext + dYτint,k + dYτint,γ 15

where dYτext is the extensive margin eect

dYτext = −

∂ω ˆ α g(ω)dτ γˆ (ω)Ak(ω) ∂τ

(D ext)

and dYτint,k and dYτint,γ are both intensive margin eects, dYτint,k being the direct eect (decrease in investment) and dYτint,γ the indirect response of transparency to the decrease in the required investment.

 α ˆ  dY int,k = R ω γˆ (ω)A ∂ k(ω) dG(ω)dτ τ ∂τ ω R  dY int,γ = ω ∂ˆγ (ω) Ak(ω) ˆ α dG(ω)dτ τ ∂τ ω

(D int)

This decomposition echoes our discussion in the introduction except that the transparency response consists in two components dYτext and dYτint,γ , while dYτint,k is the traditional behavioral response described in the literature: taxes decrease the marginal returns on investments. The three eects are all negative following an increase in τ . The advantage of our model is to isolate them cleanly and provide a simple accounting framework that can be used for counterfactual experiments.16 In the following lines, we calibrate the model to the Greek case and illustrate, in this specic calibration, the quantitative importance of each margin.

4. The impact of austerity plans on Greek economy We analyze in this section the Greek Austerity Plan implemented in 2010-2011. We rst give some gures for the crisis and its aftermath. We then study the crisis through the lens of our model: we calibrate our model using Greek data in 2010 and provide some numerical estimates for the response of the underground economy to the austerity plan. We nally discuss some additional insights on the distributional implication of the austerity plans given by our model and discuss their empirical support.

A. The drastic austerity plan in Greece 2010-2011 In this paper, we analyze one channel through which austerity plans may prove inecient as a way to reduce government decits and we think of Greece as the perfect guinea pig. We explain here why this crisis is a good benchmark. 16 See Appendix A for the details of output decomposition.


During the beginning of the 2000, Greece experienced a credit boom fostered by the integration to the Euro zone. At this time, there were already some concerns about (i) the exibility of labor markets and (ii) the high indebtedness. Both concerns were attenuated by the globally positive perspectives on output growth. In the aftermath of the global crisis of 2008, those concerns materialized: the spreads peaked and Greece was forced to restructure its debt. A troika (European Commission, European Central Bank and International Monetary Fund) took over and imposed some conditions to the Greek government for them to roll-over the Greek debt17 under some conditions. The government had to reduce decits through the adoption of severe austerity plans. Since then, Greece has experienced a series of such plans. The process has been more dicult than expected because of constant mismatches between the forecasts and the actual outcomes of each reform. In short, expected tax receipts were always over-estimated either by the government or by independent sources (research departments of Greek banks). This over-estimation reected both optimistic estimations as regards the drop in GDP and inelastic estimates of the tax base (once accounted for the economic slack). In reality, the Greek economy responded to the tax hikes by concealing more of its activity to the government. As an example on the amplitude of the misalignment, between 2009 and 2010 the Bank of Greece (together with the Greek authorities) estimated that the increase in tax revenues should be around 15.5%, of which only 7.4% was realized. This shortfall was compensated by additional last-minute expenditures cuts: −9.5% instead of −5.3%. The same misalignment has been repeated the year later in Greece. Those readjustments point to behavioral responses as being large. The measures to rebalance the government account had very strong contracting eects. In 2010, Greece has experienced a GDP contraction of 4.5% explained by the fall of private consumption (contributing for −3.3%), the reduction of government consumption (−1.3%), a fall of investment (−3.1%, gross capital formation), partially compensated by a rebalancing of the external account. In our model, this contraction can be related to a reduction of leverage for rms, and a general tightening of credit constraints, both triggered by higher taxes and lower transparency. In the following subsection, we analyze how our model predicts such responses, once calibrated using our database on Greek balance sheets. 17 Cyprus, Ireland and Portugal also rescheduled their debt under the control of this troika.


B. Calibration Our model is an accounting tool, which allows us to match quite precise moments of the Greek economy. Naturally, these degrees of freedom are obtained at the expense of some others: we consider the size distribution of rms as exogenous, so do we for the policy of tax authorities. In our view, rm's size is not as responsive as investment or transparency. Similarly, we shut down the possibility for technology or tax monitoring, i.e. fundamentals of the economy, to evolve during the period 2009-2013. Our calibration strategy is the following. We calibrate our model using the rms' balance sheet information provided by our dataset. We observe a subsample of rms in Greece that represent a very high share of Greek activity (more than 80%).18 We give in the following lines the predictions of our model on this subsample of rms. In order to clarify how such results can map to aggregate predictions, we need to make assumptions on the rest of the economy that we do not observe. We make such assumptions and interpret our results at the aggregate level in the following section. We choose our underlying parameters such as to match important features of those rms, i.e. the leverage, the output and tax receipts. First, we set the elasticity sales/assets for range of mid-size Greek rms α equal to 0.8 in line with our estimates (see gure 8). We consider rms with sales above 0.1M Euros and estimate the elasticity of sales with respect to their size. It is well-known that such estimations suer from endogeneity bias that we cannot fully alleviate. However, both cross-rms and within-rm across-time estimates give similar results  respectively 0.8 and 0.81. Figure 8 shows the t of the relationship. Second, we use our dataset to measure the average tax rate paid by rms. We use the sector classication used in the analysis of the protability of rms to measure the average VAT tax rate paid by rms. In our dataset, about 69.4% of rms produce goods in the high VAT regime (19%), whereas 12.4% of rms are subject to the middle VAT regime (9%) and the remaining 18.2% of rms is either subject to the high regime or exempted (4%).19 This provides us an 18 Our notion of aggregate variable may be a bit dierent than the standard ones, as we only aggregate using our sample of rms, excluding de facto very small rms, whose contributions to total tax revenues are quite small and inelastic  they are informal irrespectively of credit conditions.

19 In our database, over the period, we observe 60'662 rms under the low VAT regime, 41'238 rms under the middle VAT regime and 231'114 rms under the high VAT regime.


aggregate tax rate of 0.167. Finally, we set the interest rate r to 4%. Third, sanctions θz(ω) are set such that the rst rm for which θz(ω) = 1, i.e. the smallest rm which nds it more convenient to be fully transparent than partially informal, has a turnover equal to 9M Euros. This is the level of activity above which all Greek rms are subject to have an external certication of their account. We interpret this threshold as the level of activity for which tax authorities may audit rms with ease, inducing rms to be fully transparent. Table 1 reports the benchmark calibration. Given parameters α, τ , r and the function θz(ω), we are left with parameters A, c, and λ. These three parameters are chosen such to (i) minimize the distance between the theoretical leverage and the empirical leverage for rms with assets between 0 and 50M Euros (cf. gure 6), (ii) minimize the distance between the theoretical and the empirical output for rms with assets between 0 and 50M euro, and (iii) match the observed aggregate output of rms with assets between 0.5 and 100M Euros in our dataset. We discretize the set of assets w and solve the optimal transparency decision of rms at each level of endowment. We then weight each variable by the observed density and compute aggregate quantities. At the initial equilibrium, we nd a level of aggregate transparency, dened as the ratio between the aggregate tax base and aggregate output, equal to 0.98. This is substantially higher than what is typically estimated in the literature (the shadow economy in Greece would be at least around 20%): it comes from the fact that we underestimate the inuence of small rms in our analysis. Those informal rms are not in our sample and they typically do not respond to changes in tax conditions  they form an inelastic informal sector. Accounting for these rms boils down to adding a xed informal sector, which would mechanically reduce our estimates for aggregate transparency. In table 2, we show some targets that we want to match despite our calibration not being directly tied to those objectives. A rst important feature is for the aggregate theoretical output to match the aggregate empirical output. As shown in the rst line, our estimate is slightly lower than in the data, a discrepancy that arises mainly from very small rms: we under-estimate the contribution of rms below 2 Million Euros of endowment. The third and fourth lines give a measure for the discrepancy between our theoretical distributions of leverage and output and the empirical distributions. We compute the sum of squares of dierences 19

Table 1: Benchmark calibration Parameter



Returns to scale



Value added tax rate



Risk-free interest rate



Credit constraints



Fixed cost



Productivity factor



Table 2: Targets


Model Data

Output rms [0.5, 100] M



Full transparency threshold





Distance distribution theory-data Leverage (mean=.3)

between the empirical and theoretical series for output and leverage weighted by the densities of rms for each size (between 0.5 and 50 Millions of endowment). The result can be interpreted as a standard deviation of the theoretical series relatively to the empirical one. Both standard deviations are non-negligible, and are essentially explained by the discontinuous jumps that our model generates between informality and formality. Firms suddenly produce a much higher output at the cost of a larger dependence on external nance. In the data, such jumps are not observed. Nonetheless, as shown in the second line, the size above which rms are completely transparent in our model is very close to the threshold above which rms are audited in the data. Finally, we cannot match the overall receipts from auditing, but we do not see it as a failure of our model. Both in the data and in our model, sanctions are very low. Consequently, they only act as a threat and whether we capture them well or not is visible on our levels of transparency rather than on the actual receipts due to tax monitoring. 20

C. Measuring the behavioral response after a tax hike Using our benchmark calibration, we analyze the eect of changes in the tax rate on our economy. The objective of our numerical simulations is to replicate the Greek austerity plans and analyze how the transparency response could explain the observed misalignments between predicted tax receipts and actual tax receipts. To this purpose, we set the same tax rates as the government and estimate our predicted tax receipts.

We update the VAT rates according to the austerity measures implemented in 2010. The low VAT rate increased to 5.5%, the middle VAT rate to 11% and the high VAT rate to 23%. The repartition is quite invariant with rm size such that the average tax rate increases to 18.2% for our sample. We then measure the increase in aggregate tax receipts, and compute the change in the aggregate transparency. We also decompose the drop in the aggregate tax base in the extensive and intensive margin (with respect to transparency and capital) as suggested by the theoretical decompositions in equations (D ext) and (D int). We nally dene a measure of the scal multiplier associated to the austerity plan. Contrary to usual measures of scal multipliers that are related to government expenditures, our scal multiplier reports the change in total output for an increase of 1 unit in tax receipts (∆Y /∆T R). The results are reported in the second column of table 4. Following the increase in the tax rate of 21.4%, the model predicts a drop in the tax base of 11.7% explained by a decrease of transparency (−8.9%) and output (−3.1%). Given the amplitude of both responses (essentially the transparency adjustment), tax receipts only increase marginally (+7.4%). Interestingly, we can see that most of the drop in tax receipts is concentrated in mid-size rms that either drop o the formal economy or adjust their transparency downward (gure 9 shows the theoretical eect). Overall, this simple exercise points to a large inuence of the transparency channel, and this channel is sucient in itself to explain the failure of tax hikes. The drop in tranparency may explain also the large ndrop in output observed in Greece after the implementation of the austerity plans in 2010.


Table 3: The impact of austerity measures Variable

Austerity Plans

tax rate


tax receipts


tax base






Estimate of scal multiplier


Note : The gures in the top panel refer to the percentage change in each variable after the implementation of the austerity plans. The estimate of scal multiplier is dened as the change in total output for an increase of 1 unit in tax receipts.

Table 4: The drop in tax base at the extensive and intensive margin Extensive margin

Intensive margin k

Intensive margin γ




Note : The extensive margin, intensive margin


and intensive margin


refer to the shares of each margin in

the drop of the tax base after the implementation of the austerity plans.

D. Aggregate predictions One motivation behind our study is to reconcile the small increase in tax receipts collected by the Greek government with the large increase in taxes. Our thought experiment in the previous section (a tax hike similar to the real VAT increase in Greece) predicts a very high behavioral response: taxes increases by 21.4%, but the aggregate tax base decreases by 11.7% in response. As a consequence, tax revenues only increases slightly, much less than in reality. The dierence between the gures discussed in preamble and our model-based estimates may be explained by the absence of many concurrent factors in our model. For instance, we abstract from changes in other tax regimes, from changes in the functioning of labor markets or from heterogeneous eects across sectors. The model-based behavioral response is composed of two elements, the standard behavioral response with a decrease in the real activity, and the decrease in the extent to which the


activity is declared. We estimate the second element to be the largest: 3.1% is lost through actual GDP contraction, and 8.9% through evasion. The evasion eect is large; it does more than bridging the gap between the loss in tax receipts and the loss in output. In other words, evasion, in our model, probably over-reacts compared to the data. The aggregate initial level of transparency, however, is in line with estimates of the literature (see Schneider et al. (2010) for instance). More generally, under which conditions should we expect a large response to tax hikes? Our theoretical analysis shows that the impact of such experiments depends on the number of rms at the margin between informality and formality, i.e. the number of rms that are currently relying on external nance but are close to being indierent with full informality. The number of such rms is determined by (a) the threshold at which rms are indierent between informality and access to credit, (b) the density of rms around this threshold. In Greece, for instance, nancial development is not very high, which implies that a large range of small-medium rms are quite indierent. We nd a very large response because there are many of those rms. In contrast, in the United States, nancial development is higher, which implies that the indierent rm would be very small. The impact of an austerity plan would depend on the weight of such rms in the economy, arguably small. This simple analysis points to the distribution of rm size as a crucial, and so far under-studied, factor behind the success of an austerity plan. The next section, by specically analyzing the distributional evolution of rm's leverage, provides additional support for this statement.

E. Distributional implications Our model of tax evasion makes two key predictions that are not visible in aggregate quantities: (i) credit shifts towards larger rms, and (ii) some rms exit the credit markets. Before discussing the stylized facts, let us describe the behavior of credit over the period. Aggregate data on credit in Greece tell us that there has been a credit boom before the sovereign debt crisis of 2009. As shown in gure 12, our data conrm this pattern: the average amount of loans has increased steadily between 2003 and 2008, whereas in the aftermath of the Greek sovereign debt crisis in 2009, there has been a global decrease of bank loans. However, the analysis of micro-data allows us to go more into details and show a dierent response between large rms and small rms. In gure 10, we report the coecients of a panel estimation where 23

we regress the ratio of bank loans over total assets on year dummies.20 Figure 10 clearly shows that the drop in the ratio of bank loans over total assets is much more pronounced for small rms (those with total assets below 10M euro) than large rms.21 This observation leads to the rst stylized fact:

Stylized fact 1 (Access to credit and rm size) : There has been a shift of credit from small rms to medium-large rms during the crisis (see gure 13). This shift is qualitatively similar to our theoretical predictions (see gure 4). The aggregate drop in credit results from dierential eects across the distribution of rms. As predicted by the model, the drop in credit comes essentially from small rms renouncing to contract credit. Figure 13 presents more explicit evidence. This gure reports the average leverage as a function of size in 2011 and 2007. In 2011 small rms with total assets ranging from 1 to 10 M Euros had a leverage substantially lower than the one the same rms had in 2007. However, medium rms in 2011 are more leveraged than their counterparts in 2007. We interpret this shift in the distribution of leverage as an indicator that the credit crunch was demand- and small rms- driven. Figure 13 is computed on the cross-section of rms, but is sensibly similar when computed excluding rms present only in 2007 or 2011. This shift is very close to the predictions of our theoretical model. This shift may indicate that the credit drop is explained by the extensive margin, i.e. a fraction of rms renouncing to credit. A simple analysis of access to credit conrms this intuition.

Stylized fact 2 (Access to credit and the extensive margin) : The number of new rms that have access to credit market decreased after the beginning of the crisis. The rst column of table 5 reports for each year in our panel the percent share of rms that shifted their loans from 0 to a positive amount, as suggested by our denition of loanaccess. 20 Figure 10 is computed cleaning for rms xed eects. As such, the decrease of leverage does not account for any composition eect. Since we control for industry/year xed eects, we interpret the gure as the evolution of leverage for one rm over the period. The discrepancies between this gure and the aggregate numbers (these dierences are small) come from composition eects (entry-exit) over the period.

21 In contrast, the patterns of net income before taxes for small and large rms are more similar. As shown in gure 11, small rms on average have lower net income before taxes than large rms, and the drop after the beginning of the recession is slightly bigger for large rms.


We interpret this measure as an indicator of credit access at the extensive margin. While being quite stable at 3.5-4 percent over the period 2003-2009, the fraction of new rms having access to credit falls to around 2.5 percent in 2010 and 2011. Our alternative indicators of credit access also show that after the credit boom that reached its peak in 2007-2008, both the share of rms having their larger growth in the level of loans (loanaccess1 ) and the share of rms having the larger growth in the level of leverage (loanaccess2 ) fall to their lower levels over the entire period. Table 5: Loan access by year. Year








































In conclusion, the predictions of our model on the distributional impact of tax hikes are empirically veried, at least qualitatively.

5. Discussion and extensions In our model, we focus on the transparency decision of rms, and credit demand. There are additional mechanisms at play that we have not discussed so far. These mechanisms pertain to the role of the nancial intermediary sector (credit supply) and the government. One crucial element that we do not explore is that the austerity plans were a response to a debt overhang, and thus to a high default risk. One such situation has implications on the functioning of credit markets. The domestic banking sector usually owns a large share of sovereign bonds. 25

A negative shock on the value of those bonds - a debt overhang - lowers the value of bank's assets and limits their capacity to lend. This situation leads to lower transparency because fewer rms are granted access to credit. We ignore this channel because we do not think that it would change our conclusions in the specic Greek case: the injection of capital in undercapitalized banks exactly oset the depreciation of collateral held by domestic banks. The undercapitalization of Greek banks was rapidly tackled with large injections of capital ensured through the Hellenic Financial Stability Fund (HFSF). This policy was successful at saving banks from liquidation but not at revitalizing credit (the purpose was only to stabilize bank's collateral). One could argue, however, that an additional injection of liquidity into the nancial sector could counteract the incentives of rms to be less transparent. In our model, one such policy may be insucient at fostering rms' credit demand, because the decrease in the cost of external nance (the interest rate) needs to be very large to oset the increase in tax burden. One solution could be to target the credit access of small to middle size rms. Taking the default risk and debt overhang as exogenous does not allow us to model a mechanism frequently evoked in the public debate. If the austerity measures deliver a lower than expected scal adjustment, the markets may not believe in the capacity of the country to implement its scal adjustment and the risk premia on the sovereign bonds may rise again, fostering the rst default shock. Since the nancial sector is exposed to sovereign debt default, there could be a further valuation loss for the banking sector leading to a larger credit crunch and more tax evasion from the rms' side.

6. Conclusion What have we learnt in this paper?

When rms adjust the degree to which they declare

their activity, an increase in taxes is diluted through the usual contraction of output, but also through a lower aggregate transparency. Since transparency guarantees a better access to credit market, its decrease aggravates the contraction by forcing rms out of credit markets. The amplitude of the transparency response depends upon the number of rms at the margin between formality and informality. The behavior of those rms is very sensitive to changes in the trade-o credit/tax evasion. In Greece, rms at the margin are quite large and very numerous.


Quantitatively, we can explain the gap between the expected tax receipts and the realized ones, only with this transparency channel. Following an increase in VAT of around 3-4 points, the Greek government expected an increase in tax receipts only slightly lower due to output contraction. In our quantitative framework (and in reality), the increase in tax receipts was, at least, twice lower than with a xed level of tax evasion. One important contribution of the present paper is to calibrate our model on a subsample of rms that represents the universe of medium and large rms and a large subsample of smallmedium rms. In order to clarify why we expect those rms to adjust their transparency, we also provide some evidence that the protability of the in-sample rms exhibit abnormal protability levels immediately before getting access to credit. Another indirect support for our analysis is that we replicate closely the evolution of the leverage of rms as a function of their size. In particular, we expect credit to ow from smaller to larger rms and we observe such pattern in the data. Naturally, even if we observe most of the Greek production, we cannot observe very small rms that are expected to constitute most of the informal sector. In order to compensate for this caveat and provide some aggregate predictions, we would need to infer the behaviors of unobserved rms. The policy implications of our analysis are not obvious. We show that austerity plans in an economy with low tax enforcement and low nancial development are very likely to be diluted. Improving these institutions would help but is a dicult task: it is desirable even in the absence of austerity plans, and periods of economic turbulences may not be times in which structural reforms are simple to implement. One immediate implication of our model is that the impact of a tax increase essentially depends on the number of rms (and their size) that are almost indierent between being formal or informal. This insight could help policy makers choose the timing or the type of tax reforms which reduce this margin as much as possible.


References Alesina, Alberto F. and Silvia Ardagna (2009), Large changes in scal policy: Taxes versus spending, Working Paper 15438, National Bureau of Economic Research. Andreoni, James, Brian Erard and Jonathan Feinstein (1998), `Tax compliance',

Journal of

Economic Literature 36(2), 818860. Artavanis, Nikolaos T., Adair Morse and Margarita Tsoutsoura (2012), Tax evasion across industries: Soft credit evidence from greece, Technical Report 12-25. Auerbach, Alan J. and Yuriy Gorodnichenko (2010), Measuring the output responses to scal policy, Working Paper 16311, National Bureau of Economic Research. Bartelsman, Eric J. and Roel M.W.J. Beetsma (2003), `Why pay more? corporate tax avoidance through transfer pricing in oecd countries',

Journal of Public Economics 87(9-10),


2252. Cai, Hongbin and Qiao Liu (2009), `Competition and corporate tax avoidance: Evidence from chinese industrial rms*',

The Economic Journal 119(537), 764795.

Clotfelter, Charles T. (1983), `Tax evasion and tax rates: An analysis of individual returns',

The Review of Economics and Statistics 65(3), pp. 363373. Desai, Mihir A., Alexander Dyck and Luigi Zingales (2007), `Theft and taxes',

Journal of

Financial Economics 84(3), 591623. Ellul, Andrew, Tullio Jappelli, Marco Pagano and Fausto Panunzi (2012), Transparency, tax pressure and access to nance, Technical report. Enste, Dominik H. and Friedrich Schneider (2000), `Shadow economies: Size, causes, and consequences',

Journal of Economic Literature 38(1), 77114.

Favero, Carlo, Francesco Giavazzi and Jacopo Perego (2011), Country heterogeneity and the international evidence on the eects of scal policy, Working Paper 17272, National Bureau of Economic Research.


Ilzetzki, Ethan, Enrique G. Mendoza and Carlos A. Végh (2010), How big (small?) are scal multipliers?, Working Paper 16479, National Bureau of Economic Research. Kleven, Henrik Jacobsen, Martin B. Knudsen, Claus Thustrup Kreiner, SÃ ren Pedersen and Emmanuel Saez (2011), `Unwilling or unable to cheat? evidence from a tax audit experiment in denmark',

Econometrica 79(3).

La Porta, Rafael and Andrei Shleifer (2008), `The unocial economy and economic development',

Brookings Papers on Economic Activity 39(2 (Fall)), 275363.

Lemieux, Thomas, Bernard Fortin and Pierre Fréchette (1994), `The eect of taxes on labor supply in the underground economy',

The American Economic Review 84(1), pp. 231254.

Rauch, James E. (1991), `Modelling the informal sector formally',

Journal of Development

Economics 35(1), 3347. Romer, Christina D. and David H. Romer (2010), `The macroeconomic eects of tax changes: Estimates based on a new measure of scal shocks',

American Economic Review 100(3), 763

801. Schneider, Friedrich, Andreas Buehn and Claudio Montenegro (2010), `New estimates for the shadow economies all over the world',

International Economic Journal 24(4), 443461.

Slemrod, Joel (1985), `An empirical test for tax evasion',

The Review of Economics and Statis-

tics 67(2), pp. 232238. Slemrod, Joel and Shlomo Yitzhaki (2002), Tax avoidance, evasion, and administration, Vol. 3 of

Handbook of Public Economics, Elsevier, pp. 1423  1470.

Straub, Stéphane (2005), `Informal sector: The credit market channel',

Economics 78(2), 299  321.


Journal of Development

Figures Protability Figure 1: Firm protability πτ around loan access.


Figure 2: Firm protability πτ around loan access, non-tradable vs tradable.

(a) Non-tradable sectors

(b) Tradable sectors

Source: Hellastat, 2001-2011.

Figure 3: Firm protability πτ around loan access, high vs low VAT.

(a) High VAT rate

(b) Low VAT rate

Source: Hellastat, 2001-2011.


Theoretical predictions Figure 4: Leverage, output and transparency : increase in τ

Leverage, output and transparency for the benchmark calibration (solid line) and the austerity plans simulation (dashed line).


Figure 5: Leverage, output and transparency : increase in λ

Firm transparency for λ equal to 0.36 (solid line, benchmark calibration) and λ equal to 0.37 (dashed line).


Model calibration Figure 6: Empirical and theoretical leverage.

Note : Benchmark calibration. Solid line is the theoretical leverage, the dashed line is the empirical leverage for rms with assets between 0.5 and 50M euro (smoothed using a HP lter).


Figure 7: Size distribution.

Source: Hellastat, 2001-2011.

Figure 8: Empirical production function.

(a) Polynomial estimates

(b) Density

Source: Hellastat, 2002-2012. We use the whole sample of rms (approximately 30'000 rms per year).


Simulations Figure 9: Firm transparency before and after the austerity plans

Firm transparency for the benchmark calibration (solid line) and the austerity plans simulation (dashed line).


Empirical evidence on credit market Figure 10: Evolution of bank loans over total assets, 2002-2012.

(a) Small rms, total assets less than 10 M euros

(b) Large rms, total assets more than 10 M euros

Source: Hellastat, 2002-2012. We use a panel estimation on the whole sample of rms (approximately 30'000 rms per year). The values reported in the gures above are the coecients of the year dummies. We weight for the size of rms. Thus the evolution of each variable can be interpreted as its aggregate evolution. Shaded areas are 95% condence intervals.

Figure 11: Evolution of net income before taxes over total assets, 2002-2012.

(a) Small rms, total assets less than 10 M euros

(b) Large rms, total assets more than 10 M euros

Source: Hellastat, 2002-2012. We use a panel estimation on the whole sample of rms (approximately 30'000 rms per year). The values reported in the gures above are the coecients of the year dummies. We weight for the size of rms. Thus the evolution of each variable can be interpreted as its aggregate evolution. Shaded areas are 95% condence intervals.


Figure 12: Average bank loans (M euro), 2000-2011.

Source: Hellastat, 2001-2011. This graph displays the average bank loans per rm over the period 2000-2011

Figure 13: Bank loans/Total assets and Total assets.

Source: Hellastat, 2001-2011. This graph displays the distribution of total bank loans over total assets before (2007) and after (2011) the austerity plan.


Appendix A Total output and tax base decomposition ˆ, γˆ ), and the total output after We dene the total output before the austerity plan as Y (ω , k ˆ0 , γˆ 0 ). The dierence in output after the implementation of the the austerity plan as Y (ω 0 , k austerity plan may be decomposed as follows :

ˆ γˆ 0 ) + Y (ω, k, ˆ γˆ 0 ) − Y (ω, k, ˆ γˆ ) ∆Y = Y (ω 0 , kˆ0 , γˆ 0 ) − Y (ω, kˆ0 , γˆ 0 ) + Y (ω, kˆ0 , γˆ 0 ) − Y (ω, k, | {z } | {z } | {z } extensive margin

intensive margin (k)

intensive margin (γ )

A similar decomposition applies for the aggregate tax base.

B Auditing Exogenous audit schedule A weakness of the model so far is the prediction that large rms, less dependent on external nance, choose very low levels of transparency. This feature does not map to the real transparency of large rms. Shareholders or the government would not allow part of rm's activity to escape their reach. Likewise, tax authorities cannot allow large rms to hide a big fraction of their plants and renege on these revenues. For instance, according to Greek law, rms with turnover above a given threshold (9 M euro) have to be audited by external accountants. In this section, we take into account the impact of this auditing regulation on the rms' transparency decision by allowing the tax authority to target rms. Consider now that tax authorities can observe the initial endowment of rms ω at zero cost.22 Let z : ω 7→ z(ω) denote the audit schedule as a function of initial rm size. We consider it as exogenously given. Our problem is then very close to our benchmark framework. For simplicity, we assume that small rms (i.e. with ω < ω ¯ ) are never monitored. Accordingly, the threshold between small informal rms and rms who decide to invest in the modern technology remains 22 Even if tax authorities only receive a signal on size, our results would go through. The only important feature is that tax audits do not respond to the individual decision of rms (both in terms of investment and transparency).


the same. Second, entrepreneurs that are unwilling to borrow may not be entirely informal: it depends on the cost paid to tax authorities on a unit produced in a concealed rm θz(ω)τ against the cost τ paid on units produced in transparent plants. As long as θz(ω) > 1, the punishment is so high that rms are willing to declare all their plants. For θz(ω) < 1, the punishment is too low and rms hide their activity. Only for θz(ω) = 1, rms are indierent between declaring or not the marginal plant. In the general case, the program can be written as follows:

max {(1 − τ γ − (1 − γ)θz(ω)τ )Ak α − c − r(k − ω)} γ,k

subject to

λγω ≥ (1 + r)(k − ω) ˆ the solution to this program. Dene k Aαk


(1 + r)[1 − θz(ω)]τ 1 − θz(ω)τ − λ

 1+αk −1 =r α ω


In presence of tax auditing, the aggregate tax receipts include the taxes paid by rms on their transparent activity (τ times the aggregate tax base) and the fees raised by the auditing process. We can therefore dene the aggregate tax receipts as :

Z T R = τ Yτ +

ω ¯

ˆ α dG(ω) θz(ω)τ [1 − γˆ (ω)]Ak(ω)


Note that both tax receipts and equation Tz extend the formulas derived in the text for the case z = 0. In our calibration, as observed in Greece, z will be an increasing function of size

ω . We do not internalize the possible endogenous adjustment of scal authorities to changes in the decisions of rms. Our main reason for not incorporating this to our model is that we actually ignore the objective function of tax authorities. However, it is possible to understand why z is increasing in rm's size in a stylized model of tax auditing. We develop such a model below.

Endogenous auditing We propose here a simple model that determines the extent to which tax authorities inspect rms of a given (observed) size. As before, consider that tax authorities perfectly observe rm size but none of their decisions afterwards. 40

Monitoring is not perfect in the sense that the tax authority does not discover the total number of hidden plants when they choose to monitor a rm. The eective probability z(p) to discover the number of hidden plants increases in the intensity of monitoring p and decreases in the size of the hidden production. The tax authority can observe the initial endowment of rms ω at zero cost. Visiting a plant has a cost c(p) which is increasing convex in the monitoring intensity. The tax authority maximizes tax retrieval from auditing activity, taking as given the hidden production of rms:

ˆ − c(p) max z(p)θτ (1 − γˆ )f (k) p

The rst order condition gives the probability of monitoring chosen by the tax authority :

ˆ 0 (p) c0 (p) = θτ (1 − γˆ )f (k)z This equation, coupled with the rm response, describe the equilibrium investment and auditing for a given rm size.

  c0 (p) = θτ (1 − γˆ )f (k)z ˆ 0 (p) h  i  Aαkˆα−1 1 − θz(p)τ − (1+r)[1−θz(p)]τ 1+α kˆ − 1 = r λ α p ˆ − γˆ ), p). Any increase in tax potential revenue Figure 14 shows the equilibrium in the plan (k(1 (an increase in ω represented by the dashed line in graph 14) induces the governmment to monitor with higher intensity, such that rms respond by declaring a larger fraction of their plants. What happens for larger rms? On the one hand, the relative cost of inspection decreases and tax authorities become more pressing. On the other hand, rms rely less on external nance. Both eects together imply that the eect on resulting hidden investment is ambiguous, but the auditing eort is unambiguously higher.


Figure 14: Equilibrium response of transparency to the monitoring intensity

p ˆ 0 c0 (p) = θτ (1 − γˆ )f (k)z p

p∗ (T z)



Life Enjoy

" Life is not a problem to be solved but a reality to be experienced! "

Get in touch


© Copyright 2013 - 2019 DOKUMENTIS.COM - All rights reserved.