Public economics for development

DRAFT

This is a draft version of a conference paper submitted for presentation at UNU-WIDER’s conference, held in Maputo on 5-6 July 2017. This is not a formal publication of UNU-WIDER and may refl ect work-in-progress.

Public economics for development5-6 July 2017 | Maputo, Mozambique

THIS DRAFT IS NOT TO BE CITED, QUOTED OR ATTRIBUTED WITHOUT PERMISSION FROM AUTHOR(S).

WIDER Development Conference

Tax Audits as ScarecrowsEvidence from a Large-Scale Field Experiment∗

Marcelo BergoloIECON-UDELAR

Rodrigo CeniIECON-UDELAR

Guillermo CrucesCEDLAS-UNLP

Matias GiaccobassoIECON-UDELAR

Ricardo Perez-Truglia†

University of California, Los Angeles

This Draft: May 2017

Abstract

According to the workhorse model of Allingham and Sandmo (1972), firms evade taxes byoptimally trading-off between the lower tax burden and the expected penalties from audits.However, there is still no consensus about whether firms react to audits and whether they reactin a rational way. We conducted a large-scale field experiment with over 20,000 Uruguayanfirms that collectively pay over $200 million dollars in taxes per year. We provided firmswith exogenous but non-deceptive signals about the probability of being audited and thepenalty rates for tax evasion. We measure the effect of this information on their subsequentperceptions about audits, measured with survey data, as well as on the actual taxes paid.We provide evidence that firm’s reaction to audits is subject to substantial optimization andinformation frictions.

JEL Classification: tax, evasion, audits, penalties, frictions.Keywords: C93, H26, K34, K42, Z13.

∗We thank the Tax Registry of Uruguay for their collaboration. We thank Gustavo Gonzalez for his support,without which this research would not have been possible. We thank Joel Slemrod for their valuable feedback, aswell as that of seminar participants at the 2017 RIDGE Public Economics Conference and the 2017 Zurich Centerfor Economic Development Conference. This project benefited from funding by CEF, CEDLAS and IDRC.

†Corresponding author: [email protected]. University of California, Los Angeles, AndersonSchool of Management, Office C515, 110 Westwood Plaza, Los Angeles CA 90403.

1 Introduction

The workhorse model of tax evasion has been the Allingham and Sandmo (1972) model.This is an application of the model of rational crime (Becker, 1968), in which selfish and risk-averse taxpayers choose whether to conceal some income from the tax authority by comparingthe benefits (i.e., the lower tax burden) and costs (i.e., the penalties to be paid if caught).Even though there is a consensus that audits and penalties have some positive effects on taxcompliance, there is no consensus yet on whether firms react to audits in the optimal waypredicted by Allingham and Sandmo (1972). In this paper, we present novel evidence basedon a field experiment.

We study the context of compliance with Value-Added-Tax (VAT) by firms where, unlikein other sources of income such as wage income, there is no third-party reporting. As aresult, tax agencies do not know how much a firm is evading until they conduct a tax audit–and even then, they can only detect some of the evasion. In this context, the threat of beingaudited plays a key role in the enforcement of taxes. However, since tax audits are expensive,in any given year only a small share of firms are audited and –to make matters worse– thelaw prescribes low penalty rates.

In collaboration with Uruguay’s Internal Revenue Services (IRS), we conducted a fieldexperiment with a sample of over 20,000 small and medium firms from Uruguay. The IRS senta letter to the owners of all these firms, and we randomly assigned the information containedin that letter. The baseline letter included some brief generic information about taxes that isoften included in the communications with firms. The main treatment arm, audit-statistics,was identical to the baseline letter, except that it added a message listing the the probabilityof being audited and the penalty rates for tax evasion, based on tax administration statistics.

Comparing the subsequent tax compliance between individuals assigned to baseline andaudit-statistics letters, we can measure the effect of these statistics. Moreover, the audit-statistics treatment arm included sub-treatments that generated exogenous yet non-deceptivevariation in information about audit and penalty rates. As explained in the audit-statisticsmessage, we computed the average audit probability and penalty rates included in the mes-sage using a small sample of firms randomly selected from the universe of firms similar tothe firm of the recipient of the letter. Due to sampling variation, this generated exogenousvariation in the audit probability shown to the recipient (from 2% to 25%, in 1% increments)as well as the penalty rate shown to the recipient (from 15% to 66%), for a total of about 950distinct combinations. This exogenous variation allows us to measure the elasticities betweentax compliance and the audit and penalty rates.

In a complementary experiment, we used a separate sample of firms that had been pre-

2

selected by the IRS to be considered for audits. We randomly divided these firms in twohalves, one which would be audited with a 25% probability, and the other half which wouldbe audited with a 50% probability. The audit-threat letter included a message informingthese firms about the probability to which they had been assigned. This exogenous variationallows to estimate the elasticity between tax compliance and the audit probability, thusproviding an alternative estimate to the audit-statistics arm. Unfortunately, because of legalconsiderations, it was not possible to randomize the penalty rates to which these firms wouldbe subject to.

In another treatment arm, we explore another aspect of the auditing process. The audit-endogeneity letter was identical to the baseline letter, except that it added a message explain-ing that evading taxes increases the probability of being audited. According to Allingham andSandmo (1972), if individuals were unaware of it, informing individuals about this endogene-ity should increase tax compliance. In the last treatment arm, we provide a non-pecuniarybenchmark for the audit messages. The public-goods letter was identical to the baseline let-ter, only that it added a message listing all the public goods that could be provided if firmsevaded 10% less. According to the theory of a moral cost of non-compliance, adding thismessage could increase tax compliance (Cowell and Gordon, 1988).1

We use the administrative records from the IRS to estimate the effects of the information-provision experiment on tax compliance. Additionally, we invited a group of firms from theexperimental sample to answer an anonymous survey nine months after we sent the letters.In this survey, we asked firms about their perceptions of audit probabilities and penaltyrates, and we can match those responses to the experimental treatment arms that theywere assigned to nine months earlier. As a result, we can directly measure the effect of theinformation about audits included in the letters on the subsequent beliefs of these firms.

We find that, potentially consistent with Allingham and Sandmo (1972), adding mes-sages related to audits increases tax compliance: adding a paragraph with statistics aboutthe probability of being audited and the penalty rates increases tax compliance by about6.3%, and adding a paragraph that informs firms about the endogeneity of the audit proba-bility increases tax compliance by about 7.4%. These effects are not only highly statisticallysignificant, but also economically substantial: using the estimated average evasion rate of26% from Gomez-Sabaini and Jimenez (2012), these effects would amount to a reduction inthe evasion rate of 24% and 28%, respectively. Furthermore, these effects are robust to anumber of specifications and outcome variables. In comparison, the message about public

1There is a related literature on the social determinants of tax compliance. Just to mention some examples,Blumenthal et al. (2001) and Fellner, Sausgruber and Traxler (2013) show that moral suasion messages donot increase compliance, while Perez-Truglia and Troiano (2015) show that social shaming can sometimes beeffective.

3

goods did not have a statistically significant and robust effect on tax compliance.However, the fact that adding information about audits to a letter increases tax compli-

ance does not imply that firms are making an optimal cost-benefit calculation as in Allinghamand Sandmo (1972). On the one hand, it is possible that firms are rationally learning fromthe information about audits and changing their behavior because they are re-optimizingunder the new beliefs. On the other hand, firms may be reacting to the information becauseit makes the cost of evading more salient, even without any changes in beliefs.

The results from the audit-statistics treatment arm favors the salience channel. Amongfirms who were sent the audit-statistics letter, those who by chance receive higher signals ofthe audit or penalty rates do not pay significantly higher taxes. Indeed, the average elastic-ities of tax compliance with respect to audit and penalty rates are close to zero, preciselyestimated and statistically insignificant from the elasticities predicted by various calibrationsof Allingham and Sandmo (1972). The results are similar if we use instead firms assigned tothe audit-threat letter, which randomized the actual probability of audits.

The survey data also favors the salience channel. We find that the audit-statistics letterreduced the perceived probability of being audited. Thus, if individuals were reacting tothe audit-statistics letter because of re-optimization, they should have reduced, rather thanincreased, their tax compliance. We also find suggestive evidence that individuals reacted tothe audit-endogeneity letter through the salience channel rather than through re-optimization:since most firms were already aware of its content, the audit-endogeneity message did nothave a significant effect on the belief that audits are endogenous.

Our survey also shows that firms, on average, tend to over-estimate the probability ofbeing audited by a factor of two, but have unbiased beliefs about penalty rates. This findingof over-estimation of audit probabilities is consistent with prior survey evidence (Harris andAssociates, 1988; Erard and Feinstein, 1994; Scholz and Pinney, 1995). However, the priorevidence was based on individuals with wage income, for which the misperception of auditprobabilities is inconsequential due to third-party reporting. Our evidence suggest that thesemisperceptions persist even when the financial stakes of misperceiving audits can be quiteconsequential.

Our findings suggest that firms may comply with taxes because of the threat of beingaudited, but not in an optimal manner as in Allingham and Sandmo (1972). Our evidencesuggests two relevant sources of frictions. First, the fact that tax compliance is elastic withrespect to the salience of audit statistics but inelastic with respect to the audit probabilityand penalties is suggestive of significant optimization frictions. Second, the fact that firmshave large dispersion and biases in beliefs about audits, and those misperceptions largelypersist even after they are provided with accurate information, is suggestive of information

4

frictions.Our findings contribute to the debate about the determinants of tax compliance. Some

take the failure of the Allingham and Sandmo (1972) model to predict evasion rates as anindication that tax compliance does not largely depend on tax enforcement, but on otherfactors such as tax morale (Luttmer and Singhal, 2014). Our evidence suggests that taxenforcement indeed matters, but it also indicates that firms do not necessarily react to it asin the rational optimization process in Allingham and Sandmo (1972): our results highlightthe presence of optimization and information frictions. These results on the role of saliencefor tax evasion are consistent with the abundant evidence on the importance of salience fortax avoidance (Chetty, Looney and Kroft, 2009).

This paper belongs to various strands of literature. First, it belongs to a growing literaturethat uses field experiments to study the decision of individuals to pay taxes. In a seminalcontribution, Slemrod, Blumenthal and Christian (2001) showed that, for a sample of U.S.self-employed individuals, individuals who were randomly assigned to receive a letter fromthe Internal Revenue Services with an enforcement message reported higher income in theirtax returns. Similar messages about tax enforcement have been shown to have positive effectson compliance in a variety of contexts: wage income taxes in Denmark (Kleven et al., 2011),individual public-TV fees in Austria (Fellner, Sausgruber and Traxler, 2013), firm VAT taxesin Chile (Pomeranz, 2015), individual municipal taxes in Argentina (Castro and Scartascini,2015), and an individual church tax in Germany (Dwenger et al., 2016). We contribute tothis literature by disentangling the precise mechanism through which the threat of auditsaffects tax compliance.

Our subject pool is most directly related to the study by Pomeranz (2015). She showsthat, compared to firms who received a placebo letter, firms increased the VAT paymentswhen they received a letter mentioning the possibility of being audited. Pomeranz (2015)then uses that exogenous variation in payments to measure how the higher tax payments spillover to other firms in the value added chain. In our paper, instead, our focus is understandingthe mechanisms though which the letters sent by Pomeranz (2015) affected the tax paymentsof the recipients in the first place.

From the perspective of experimental design, our work is most directly related to Klevenet al. (2011) and Dwenger et al. (2016). In one treatment arm, Kleven et al. (2011) showsthat randomizing employed individuals to a higher audit probability (100% instead of 50%)increases their tax compliance by an amount that is statistically significant but economicallyvery small. However, their findings do not constitute evidence against Allingham and Sandmo(1972), because they conduct these experiments with individuals that should rationally notcare about the audit probability: i.e., individuals receiving wage income for which evasion

5

is almost always automatically detected without the need of audits, through third-partyreporting. In one treatment arm, Dwenger et al. (2016) show that announcing differentprobabilities of audits does not have a statistically significant effect on compliance with asmall local church tax in Germany – however, their statistical power is not enough to rejecteconomically significant effects.

This paper also belongs to a literature that tries to evaluate the fit of the Allingham andSandmo (1972) model. The evidence based on calibration exercises suggests that Allinghamand Sandmo (1972) would predict substantially lower tax compliance than what the compli-ance observed in the United States (e.g., Alm, McClelland, and Schulze, 1992). Other studiesrely on regression analysis based on observational data. For example, Beron, Tauchen andWitte (1988) find that the probability of being audited has an economically small effect oncompliance. Dhami and al-Nowaihi (2004) suggested the addition of behavioral features, suchas stigma cost and prospect theory, to improve the fir of the Allingham and Sandmo (1972)model. And a final group of studies is based on laboratory experiments. Most notably, Alm,Jackson and McKee (1992) show that, in a laboratory setting, taxpayer reporting increaseswith audit and penalty rates, but the magnitudes of these reactions are smaller than thosepredicted by Allingham and Sandmo (1972).

The paper is organized as follows. Section 2 discusses the relevant hypotheses and theexperimental design used to test them. Section 3 presents the data sources and discussesimplementation of the field experiment. Sections 4 and 5 present the results. The finalsection concludes.

2 Hypotheses and Experimental Design

2.1 Baseline Letter

Our experiment consisted of a mailing campaign from the IRS, which included a numberof treatment and sub-treatment arms. Rather than comparing individuals who received aletter to individuals who did not receive a letter, we always compare individuals who receivedreceived letters, but with subtle variations in the content of those letters. As a result, weeliminate any concerns about the many possible mechanical effects of just receiving a letter,such as a reminder to pay taxes.

These letters consisted of a single sheet of paper with the name of the recipient in theheader, the official letterhead of the Internal Revenue Services (IRS), and the hand signatureof the General Director of the IRS. These letters were folded and introduced in a envelopesealed with the official identification of the IRS, and delivered by certified mail, which guar-

6

antees that the letters are delivered directly to the recipient, who needs to certify the receiptby providing a signature.

The first type of letter is the baseline letter. A sample of this letter type is providedin Appendix A.1. This baseline letter contains some information that the IRS includesroutinely in their communications with the firms about the goals and responsibilities of thetax authority. The content of the letter explained that the individual was randomly selectedto receive this information and that the purpose of the letter was only to inform and therewas no need to reply or present any documentation to the IRS offices. The other letter typesare identical information to this baseline letter, only that including an additional paragraphin between the first and second paragraphs of the baseline letter. This additional paragraphis always in a larger font size and boldface font type.

2.2 Audit-Statistics Letter

The goal of this letter is to generate exogenous variation in the firms perceptions about auditprobabilities and penalty rates. We cannot assign different individuals to different penaltyrates, because of legal constraints: in Uruguay, as in most of the world, individuals cannotbe punished differently for the same crime. To circumvent this limitation, we designed atreatment arm in which we create exogenous variation in the perceptions about penalties, ina non-deceptive way, by exploiting sampling variation in statistics about audits.

In this letter type, we provide firms with statistics about the audit and penalty rates.According to the Allingham and Sandmo (1972) model, we would expect firms to be interestedin this information, because it would help us make more optimal evasion decisions, potentiallyincreasing their bottom-line by a significant amount. Furthermore, this information seemsparticularly valuable in the context of limited information about audits: while it is easy tofind online information about some variables that are potentially important for the decisionmaking of firms, such as the inflation rate or the exchange rate, it is almost impossible tofind online any information about audit probabilities and audit penalties.

Appendix A.2 presents a sample of the audit-statistics letter type. This letter type adds tothe baseline letter a paragraph providing information to the taxpayers of the audit probability(p) and penalty sizes (θ) among a random sample of similar firms:2

“On the basis of historical information on similar businesses, there is a likelihoodof [p%] that the tax returns you filed for this year will be audited in one ofthe coming three years. If, pursuant to that auditing, it is determined that tax

2In the message provided to the subjects we use the probability of being audited in the three coming yearsbecause, by law, tax auditors only go back three years. As a result, whatever amount a firm evaded in agiven year could be subject to a penalty if the firm is audited at least once in the following three years.

7

evasion has occurred, you will be required to pay not only the amount previouslyunpaid, but also a fee of approximately [θ%] of that amount.”

In our sample, the average value of p is 11.7%, while the average value of θ is 30.6%. Unfor-tunately, most tax agencies around the world do not publish data on the values of p and θ,making it difficult to compare the Uruguay context to other contexts. In the United States,for which there is data available, these two parameters seems to be in the same order ofmagnitude: self-employed individuals face p=11.42% and θ=20%.3

The addition of the audit-statistics message to the baseline letter may have a positive ornegative effect on VAT payments, potentially through two mechanisms. On the one hand,by providing with information about audit and penalty rates we make the expected cost ofevading more salient for the firms, which should increase VAT payments. On the other hand,firms might change their beliefs about p and θ. If firms increase their perceived p (or θ)they should declare higher VAT. On the contrary, if providing this information to the firmsreduces their perceived p (or θ), firms should declare lower VAT.

To distinguish between these two mechanisms, we introduced exogenous variation in thevalues of p and θ shown to the individuals. The details about the estimation of p and θ wereshared with the subjects in the letter, as a footnote, using the following language:

“Estimates based on data from the 2011-2013 period for a group of firms withsimilar characteristics in terms of, for instance, total invoicing. The likelihoodof being audited was calculated as a percentage of audited firms in a randomsub-sample of firms. The rate of the fee was estimated as an average of a randomsub-sample of audits.”

More specifically, we divided the firms in quintiles of total sales revenues, and then for eachfirm we randomly draw a sample of 100 firms from their same quintile (i.e., “similar firms”)and compute the average p and θ. This randomization strategy lead us to 940 differentcombinations of p and θ. Note that the information provided to the recipients was non-deceptive, because the way in which we estimated p and θ is unbiased and is consistent withthe explanation given in the footnote.

Figure 1 shows a histogram of the values of p and θ included in these letters. The valuesof p range from 2% to 25%, with an average of roughly 11.7%. The values of θ range from

3First, there is an annual 2.1% probability of being audited, according to the ratio of returns examinedcorresponding to business returns without income tax credit with a reported income between 25,000 USD and200,000 USD (Table 9a of (IRS 2014)). Each audit covers between 3 and 6 years in the past, which impliesthat the the probability that the current year’s tax declaration will be eventually audited ranges from 5.88%and 11.42%. Second, IRS usually impose a basic penalty of θ=20%, although the penalties can be higher inmost severe cases.

8

15% to 66%, with an average of about 30.6%. A small share of this variation in p and θ

(13.6% and 0.9%, respectively) comes from the fact that firms belong to different referencegroups (i.e., different quintiles of sales revenues), and the rest of the variation comes purelyfrom sampling variation.4

The goal is to compare the tax payments among individuals who were assigned by chancehigher values of p and higher values of θ. If individuals do not react to p or θ, that wouldsuggest that salience was the main mechanism through which firms reacted to the audit-statistics message. If, instead, the values of p and θ affect tax compliance, that wouldsuggest that individuals learned from the information provided and re-optimized based theirposterior beliefs – indeed, we can compare the magnitude of the reactions to the predictionsof a calibration of the Allingham and Sandmo (1972) model.

As a second strategy for disentangling between salience and learning mechanisms, weconducted a survey of individuals who were sent letters, to assess whether they had incorpo-rated the information provided to them in the letter. To study the effects of the informationcontained in the audit-statistics letter type, we included the following two questions to thesurvey:

Perceived Audit Probability: “In your opinion, what is the likelihood that thetax returns filed by a company like yours be audited at least in one of the nextthree years (from 0% to 100%)?”

Perceived Penalty Rate: “Let’s imagine that a company like yours is audited andthat tax evasion is detected. What, in your opinion, is the penalty (in %) asdetermined by law that the firm must pay in addition to the originally unpaidamount? For example, a fee of X% means that, for each $100 not paid, the firmwould have to pay those original $100 plus $X in penalties.”

2.3 Audit-Threat Letter

To complement the evidence from the audit-statistics sub-treatments, we included an alter-native way of randomizing perceptions about audit probabilities. We devised a treatmentarm called audit-threat letter which randomly assigned firms to groups with different prob-abilities to be audited with a certain probability in the following year. A sample of theaudit-threat letter is presented in the appendix A.5. These audit-threat letters were identicalto the baseline letter, except for the additional paragraph:

4We estimate the part of variation that emerges because of being in different sales groups by regressingeach parameter on the quintiles of sales revenues. Regressing p over VAT sales quintile results in R2 = 0.136while regressing θ over the same variables results in R2 = 0.009.

9

“We would like to inform you that the business you represent is one of a groupof firms pre-selected for auditing in 2016. A [X%] of the firms in that group willthen be randomly selected for auditing.”

There are two randomly assigned probabilities of audit: X=25% and X=50%. If theseletters affected the perceived audit probability, instead of just making audits more salient,firms randomly assigned to the 50% probability should report higher taxes than firms assignedto the 25% group.

For the random assignment of the audit-threat type, we were restricted to a separatesample of high-risk firms, which the IRS had pre-selected to be considered for audits. Thus,because they use different samples of subjects, the firms assigned to audit-threat letters cannotbe compared to firms assigned to the baseline letter.

2.4 Audit-Endogeneity Letter

Most tax agencies, including the one from Uruguay, do not make audit probabilities uncon-ditionally the same for all firms. Instead, they try to assign higher audit probabilities tofirms that have a higher probability of evading. As a result, evading more taxes will in-crease the probability of being audited. In addition to the case where the audit probabilityis exogenous, Allingham and Sandmo (1972), Andreoni et al. (1998), Yitzhaki (1987) andSlemrod and Yitzhaki (2002) introduce variations of the model where the audit probabilitiesare endogenous. According to these models, if unsuspecting firms were to receive news aboutthe endogeneity of audits, they should revise their tax evasion decisions reducing the amountof tax evaded. Indeed, Konrad et al. (2016) shows suggestive evidence of this mechanism inthe context of a lab experiment.5

To explore this hypothesis, we designed the audit-endogeneity letter type. We asked thetax agency to split a sample of firms in those suspected of evading at above-average ratesand those suspected of evading at below-average rates. We then computed the difference in2011-2013 audit rates between the two groups. The audit rates were approximately twice ashigh for the group of firms with above-average perceived evasion. We used this informationto create the audit-endogeneity letter type. This letter, of which a sample can be foundin Appendix A.3, is identical to the baseline letter, except for the addition of the followingparagraph:

“The IRS uses data on thousands of taxpayers to detect firms that may be evadingtaxes; most of its audits are aimed at those firms. Evading taxes, then, doubles

5They show that if laboratory taxpayers face a situation where a more suspicious attitude toward a taxofficer increases the probability of being audited, tax compliance is increased by 80%.

10

your chances of being audited.”

We will measure if individuals who received this additional paragraph in the letter pay higheror lower taxes. Like in the audit-statistics letter, two mechanisms could be at play with thisaudit-endogeneity letter. On the one hand, the information about endogeneity of audits maymake the endogeneity or the audits more salient, which should increase tax compliance. Onthe other hand, individuals may learn about the degree of endogeneity of audits and re-optimize their tax compliance under the new beliefs. If, on average, recipients revised theirbeliefs about the degree of endogeneity of audits upwards (downwards), this mechanismshould increase (decrease) their tax compliance.

Like in the case of audit-statistics letter, we conducted a survey of these firms, to assesswhether the information provided in the letter had an effect on beliefs. For that, we askedthem the following question:

Perceived Audit Endogeneity: “In your opinion, if a firm that evades taxes doublesthe amount it is evading, what is the effect on its likelihood of being audited? Itwould increase significantly; It would increase slightly; It would not change; Itwould diminish slightly; It would diminish significantly.”

2.5 Public-Goods Letter

In line with previous studies (see e.g., Blumenthal et al. 2001, Fellner et al. 2013, Dwengeret al., 2014, Pomeranz, 2015), we wanted to have a benchmark intervention based on non-pecuniary incentives. We asked various managers from the IRS about which non-pecuniarymessage would be most effective at increasing compliance. They all suggested messages aboutthe cost of evasion in terms of public good provision, in the spirit of the model of Cowelland Gordon (1988).6 This message lists a series of services that could be provided by thegovernment if all individuals evaded 10% less (Appendix A.4). To create this message, wecombined estimates from studies conducted by a number of governmental agencies.7

The public-goods letter is identical to the baseline letter, with the exception of the additionof the following paragraph:

“If those who currently evade their tax obligations were evade 10% less, theadditional revenue collected would enable all of the following: to supply 42,000

6This message is also related to the lab experiment from Alm, McClelland, and Schulze (1992), wherethey present evidence that one of the reasons why people decide to pay taxes is their valuation of the publicgoods provided with the tax revenues.

7These agencies were: Administracion Nacional de Educacion Publica (ANEP), CEIBAL, Ministerio deSalud Publica (MSP), Ministerio del Interior (MI), Ministerio de Vivienda, Ordenamiento Territorial y MedioAmbiente (MVOTMA).

11

portable computers to school children; to build 4 high schools, 9 elementaryschools, and 2 technical schools; to acquire 80 patrol cars and to hire 500 policeofficers; to add 87,000 hours of medical attention by doctors at public hospitals;to hire 660 teachers; to build 1,000 public housing units (50m2 per unit). Therewould be resources left over to reduce the fiscal burden. The tax behavior of eachof us has direct effects on the lives of us all.”

We can test whether, relative to the baseline letter, adding this paragraph about the socialcost affects tax compliance. Again, one possible channel is that this message makes the moralcost of evasion more salient, after which individuals should evade less. The other channel isthat individuals, on average, could have revised their beliefs about the social cost of evasionupwards (downwards), which should increase (decrease) their tax compliance.

3 Data Sources and Implementation of the Field Ex-periment

3.1 Institutional Context

Uruguay is a South American country with an annual GDP per capita of about 15,000 USDin 2015. Tax revenues in 2015 were about 19% of GDP and, as usual in many other countries,VAT represents the largest source of tax revenue in Uruguay, accounting for roughly 50%of the total tax revenues.8 As usual, firms are required to remit VAT payments during theproduction chain.9 The standard VAT rate is 22%, except for a small number of specificproducts corresponding to a basic basket of foodstuff that are taxed at a 10% rate or evenexempt.

The tax morale in Uruguay is believed to be among the highest in Latin America, andpossibly comparable to some developed nations.10 According to estimates from Gomez-Sabaini and Jimenez (2012), VAT evasion rate in Uruguay was around 26% in 2008, which

8Estimates are based on own calculations with data from the Central Bank of Uruguay and the InternalRevenue Service. The other sources of tax reveneues are personal income tax, corporate taxation, and somespecific taxes to consumption, business and wealth.

9Firms may credit VAT paid on input costs (i.e. imports and purchases from their suppliers) against thetotal sales of goods and services to their costumers (i.e. “tax debit”). They pay VAT to the IRS only on theexcess of the total “tax debit” over the tax credit. In case of tax credit exceed debit tax, the excess may becarried over the future tax credit. In theory, the VAT should have the same implications than a retail salestax, although in practice they are believed to differ in some ways (Slemrod, 2008).

10For instance, according to survey data from the 2010-2013 wave of the World Values Survey, 77.2% ofrespondents from Uruguay say that evading taxes is “Never Justifiable,” while this proportion is 68.2% amongall other Latin American countries (population-weighted) and 70.9% for the United States.

12

was the third lowest evasion rate among the nine Latin American countries included in thestudy, and roughly comparable to a 22% evasion rate in Italy as of 2006 (Gomez-Sabaini andMoran, 2014).11

3.2 Subject Pool and Randomization

Our experiment was conducted in collaboration with the IRS. As of May 2015, there were120,125 firms registered with the IRS of Uruguay. A subsample of 4,597 firms pre-selected bythe IRS was put aside for the audit-threat sample, which we call the secondary experimentalsample. Of the remaining firms, we selected a subsample to form the main experimentalsample.

To form this main sample, we excluded some firms by request of the IRS. For instance,we excluded firms with special regime for VAT payments, which are either very small or verylarge firms. To target firms that were very likely to be active, we kept firms that had madeVAT payments in at least three different months during the previous 12-month period andwith a total value added of at least $1,000 – for the sake of simplicity, all amounts shownin this paper are indexed by inflation and are converted in U.S. dollars using the nominalexchange rate from August 2015.12

We wanted the letters to be delivered to the owners of the firms. In some cases, ownersprovide the address of their accountants instead of their own addresses. Since the IRS hasdata on the addresses of all registered accountants, we dropped all of these cases from thesample. We were also concerned that in very large firms the effect of the information sentto the owner may be substantially diluted. For that reason, we excluded large firms, with avalue added above $100,000 during the previous 12 months.

After doing all the exclusions, we were left with 20,471 firms for the main experimentalsample. All these firms were randomly assigned to receive one the four letter types, with thefollowing distribution: 62.5% of the firms were assigned to our main treatment arm, the audit-statistics letter; the other three letter types (baseline, audit-endogeneity, and public-goods),were assigned 12.5% of the sample each. After removing the roughly 18.5% of the lettersthat were returned by the postal agency, the final distribution of letter types was: 10,272 toaudit-statistics; 2,064 to baseline, 2,039 to audit-endogeneity, and 2,017 to public-goods (totalN = 16,392).13

11A more recent study suggests that the VAT evasion rate in Uruguay dropped down to 13% as of 2012(DGI, 2013).

12The sample selection was conducted in May 2015, so this 12-month period spans from April 2014 toMarch 2015.

13The randomization to letter types was stratified by the quintiles of the distribution of value added overthe 12 months previous to the randomization.

13

The 4,597 firms in the secondary sample were assigned to receive the audit-threat letter.Half of them were assigned to the audit probability 25% and the other half to the auditprobability 50%. After excluding the 12% of letters returned by the postal office, we have2,015 firms in the 25%-probability group and 2,033 firms in the 50%-probability group (totalN = 4,048).

Columns (1) through (4) of Table 1 compares the pre-treatment balance of characteristicsbetween the different letter types in the main experimental sample, with characteristics suchas VAT paid prior to the experiment, the age of the firm and the number of employees.Additionally, for each characteristic, column (5) shows the p-value of the test of the nullhypothesis that the averages are the same across all four letter types. As expected, thedifferences across letter types are economically and statistically insignificant. Columns (6)through (8) of Table 1 present a similar balance test, but for the secondary sample used forthe threat letters. Again, the characteristics are balanced across firms who received the 25%threat letter and firms who received the 50% threat letter.

Table 2 provides some descriptive statistics for the firms in our experimental sample.Column (2) of Table 2 corresponds to all firms in the main experimental sample. On average,these firms had paid $1,890 in VAT over the past 3 months (implying a value added of about$8,600), they had been registered with the IRS for 15.3 years, they had 4.8 employees, 14%of them had been audited at least once over the previous three years, and 22% belonged tothe retail sector.

Column (1) of Table 2 corresponds to the universe of registered firms. As intended, com-pared to the universe of firms, the experimental sample includes smaller firms, both in termsof number of employees and level of VAT payments. Last, columns (3) of Table 2 providesstatistics about the secondary experimental sample (i.e., for the audit-threat treatment arm).Even though there are statistically significant differences in characteristics, these character-istics are nonetheless roughly comparable in magnitude. The big difference between thesetwo samples is that the audit rates are 7 percentage points higher in the audit-threat sample,which is mechanical because these high-risk firms tended to be high-risk in the past too, andthus were targeted to be audited more frequently in the past.

3.3 Timing of the Mailing Experiment and Outcomes of Interest

The letters were sent to Uruguay Postal Office on August 21 2015. The vast majority ofthe letters were delivered in the month of September, and therefore we define August as thelast month of the pre-treatment period and October as the first month of the post-treatmentperiod.

14

The main outcome of interest in our study consists of the total VAT amount by taxpayersin the 12 months subsequent to receiving the letter.14 Furthermore, we can break down theVAT payments depending on its timing. In addition to the month in which the paymentis transferred to the IRS, we also observe the month for which the payment was intendedfor. For instance, someone could make a payment in December 2015 for value added in thatsame month, or can send a payment for value added in any previous month. Since firmstypically make VAT payments on a month by month basis, they normally make payments forthe current and previous month, which we call “concurrent payments” (73,5% of the firmsin July, 2015). However, firms occasionally make payments for two or more months in thepast, which we denominate “retroactive payments” (3.9% of the firms in July, 2015).

As an additional outcome of interest, we obtained data from the IRS on the other maintaxes paid by the firms: corporate income taxes and net worth taxes. These payments aremost often made monthly, and jointly with the VAT, they comprise more than 96% of thetotal tax burden of firms.

Table 3 shows some descriptive statistics about the distribution of payments for firmswho received the baseline letter type. On average, the total amount of VAT paid in the12 months of pre-treatment is about $7,700, while the amount for the corresponding post-treatment period is approximately $6,500. This negative trend in VATs can be explainedby the fact that this is a sample of smaller firms and thus they have a high turnover rate.There is a wide dispersion in the size of post-treatment VAT payments, ranging from a 10thpercentile of $400 to a 90th percentile of $16,550.

3.4 Eliciting Beliefs: Survey Implementation

In addition to having information about the mailing address of the owners of these firms, theIRS also has email addresses for some of these firms. We sent an online survey to a sampleof firms to whom we had sent the experimental letters. This was done under the umbrella ofthe Survey on the Costs of Tax Compliance for Small and Medium-Sized Businesses, whichhad been previously administered with the support of the Inter-American Center of TaxAdministrations and the United Nations.

The email invitations, of which a sample is attached in Appendix A.6, were sent on May2016, roughly nine months after the letters were sent. We sent invitations to all the firms inthe main experimental sample with a valid email address.15 It was very important for the

14This variable includes VAT payments and also VAT withholding made by third party agents.15The taxpayers are not obligated to register an email address with the IRS, and for that reason the

majority of taxpayers do not register one. We did not include email address repeated in the full sampleaddresses more than three times. These latter group are most likely accountants, and the IRS did not wantto overburden them with so many emails.

15

IRS and for us that the survey was anonymous, so that the respondents would be honestin their responses. We partnered with local and international universities to guarantee theanonymity of the survey. However, at the same time, we wanted to measure the effects ofletter types on survey responses. To achieve that goal, we embedded a code to the surveylink that identified which treatment of the experiment each firm was assigned to: i.e., whichof the four letter types and, within the audit-statistics treatment, which combination of pand θ. Since these treatment codes were far from uniquely identifying any single firm, thisdid not compromise the anonymity of the survey in any significant way.

The last column of Table 2 shows the average characteristics among the 3,845 firms whowere invited to the survey. The firms invited to the survey were very similar in characteristicsto those in the entire experimental sample (shown in column (2)).

We want to measure the beliefs of the owners, because they were the ones who receivedour letters. One limitation of email the data is that the IRS was not certain that the emailaddress belonged to the owner of the firm. To deal with this limitation, at the very beginningof the survey we included a question asking the survey respondent to self-identify as one ofthe following four types: owner, internal accountant, external accountant, manager or otheremployee.

The survey included our module plus additional seven modules requested by the IRS forgoals that are unrelated to this survey, aimed primarily to assess the tax reporting burdenfor the taxpayers in terms of their time, money and patience. As described in Appendix A.7,our own module included two questions about the beliefs that that we wanted to influencethrough the information contained in audit-statistics letter (perceived audit probability andperceived penalty rate) and one question about the belief related to the audit-endogeneityletter (perceived audit endogeneity).

Of the 3,845 firms that we invited to participate the survey, we received 2,331 responses,which entails a response rate of 60.6%. Of these 3,845 firms, 45% self-identified as owner(45%), 4.4% as internal accountant, 5.4% as external accountant, 1.9% as manager, 4.5% asother employee, while the remaining 38.9% did not respond to this question. For the baselineresults, we use only responses from individuals who self-identified as owners, although theresults are similar under alternative specifications (See Appendix C).

An important aspect of the survey was that, by request of the IRS, none of the questionswere mandatory. Among respondents self-identified as owners, there is non-missing data forthe three key questions between 23.9% and 29.3% of the time, which is comparable to theaverage response rate among all questions in the survey (25.0%).

16

4 Results: Effects of Messages about Audits

4.1 Baseline Results

In this section, we estimate the effects of adding each of the three messages (audit-statistics,audit-endogeneity and public-goods) to the baseline letter.

Figure 2 shows the effect of these three messages over time, by comparing the VATpayments of the firms assigned to the baseline letter to the VAT payments of firms assigned toeach of the other three treatment types.16 This Figure shows the raw comparisons, withoutincluding any control variables. These effects are based on a Poisson regression, so thecoefficients can be interpreted directly as semi-elasticities.

Figure 2.a shows that the addition of the audit-statistics message had an statisticallyand economically significant effect on VAT payments. For instance, the point coefficientcorresponding to the third post-treatment quarter implies that the addition of the paragraphaudit-statistics increased by 8.7% the amount of VAT paid 7-9 months after the letters weremailed (p-value= 0.012). The effects are similar in magnitude across all the post-treatmentquarters, suggesting that the effects of the audit-statistics message were stable and persistent.The coefficients on the pre-treatment VAT amounts, corresponding to the falsification test,are economically and statistically insignificant, as expected.

Figure 2.b shows that the effects of adding the message audit-endogeneity was similarto the addition of audit-statistics, both in terms of statistical significance and economicmagnitude. In turn, Figure 2.c presents mixed evidence about the effects of the public-goodsmessage. On the one hand, there are large post-treatment differences in VAT payments. Onthe other hand, the event-study graph shows that a great deal of those differences existed inthe pre-treatment period. Thus, we need to control for pre-treatment outcomes in order toget a fairer estimate of the effects of this message, which we do in the regression analysis.

In the regression analysis, given the effects seem to be quite stable during the 12 months ofpost-treatment period, we pool the outcomes over that entire period to maximize statisticalpower. Additionally, we include as control variables the twelve previous monthly payments.This can address the potential lack of balance in treatment assignment, as discussed abovefor the public goods treatment. Additionally, given the persistence of VAT payments, theuse of pre-treatment controls often reduces the variance of the error term and thus resultsin gains in statistical power (McKenzie, 2012). In the interest of transparency, for eachcoefficient related to post-treatment effects, we will also present a falsification test based onpre-treatment effects.

16In order to prevent that the results are affected by typos and outliers we top code all outcomes in the99.99 fractile.

17

Table 4 presents the baseline regression results. The first column displays the resultsof our baseline specification: the effects of each of the three treatments (audit-statistics inPanel A, audit-endogeneity in Panel B, and public-goods in Panel C) when compared withthe outcomes for firms who received the baseline letter. In the post-treatment coefficients,the dependent variable is the effect on the total amount of VAT paid along the 12 monthspost letter delivery; in the pre-treatment coefficients, the dependent variable is the (placebo)effect on the total amount of VAT paid in the 12 months prior to letter delivery.

The first column from Table 4 corresponds to the average effects for the entire sample. Thepost-treatment coefficient of audit-statistics (panel a.) indicates firms receiving the additionalparagraph about audit-statistics paid, on average, 6.3% more VAT in the 12 months afterthe intervention, with this effect being statistically significant (p-value=0.013).

This effect is not only highly statistically significant, but also economically substantial:using the estimated average evasion rate of 26% from Gomez-Sabaini and Jimenez (2012),this effect would amount to a reduction in the evasion rate of 24% (= 6.3%

26% ). The (placebo)effect on pre-treatment outcomes is close to zero (-0.8%), statistically insignificant and evenmore precisely estimated than the corresponding post-treatment effect: the s.e. on the pre-treatment coefficient is 0.021, which is 16% smaller than the corresponding 0.025 for thepost-treatment coefficient.

The effects of our audit-statistics message is not directly comparable to the effects of theaudit message from Pomeranz (2015), because of differences in the content of the messagesand also because we study firms from different countries and with different characteristics.With that caveat in mind, Table 4 from Pomeranz (2015) indicates that her deterrence letterled to an increase in VAT payments of 7.6%, which is similar in magnitude and statisticallyindistinguishable from the 6.3% effect of our audit-statistics message.

Panel b. from Table 4 shows that that adding the audit-endogeneity message increasedsubsequent VAT payments by 7.4% (p-value=0.021). This effect is similar in magnitude,and statistically indistinguishable from, the 6.3% effect of our audit-statistics message. Inother words, the addition of either of the two message about audits had a similar effecton VAT payments. In comparison, there is weaker evidence that the public-goods messageaffected VAT payments: panel b. from Table 4 shows that the addition of public-goods had aneffect on subsequent VAT payments that was smaller in magnitude (4.3%) and statisticallyinsignificant.

18

4.2 Robustness Checks

To assesses the robustness of these results, Table 5 presents alternative estimates based ondifferent specifications. The first two columns present estimates of the treatment effects basedonly on the extensive margin of VAT payments: i.e. the outcome is 1 if the firm made atleast one payment in the post-treatment period, and 0 otherwise. Column (1) presents resultsfrom a linear probability model, while column (2) present Probit estimates. There is no muchvariation in the extensive margin: 96% of firms in the sample made positive payments in thepost-treatment period. This is a direct byproduct of the selection of the subject pool: weexcluded all firms who did not make at least three payments in the 12 months before thetreatment assignment. The effects of the three different messages on the extensive marginare close to zero and statistically insignificant.

The specifications in columns (3), (4) and (5) of Table 5 use the amount of VAT paymentsas the dependent variable. Column (3) corresponds to our baseline Poisson specification. Inturn, column (4) and (5) present alternative models: Tobit and OLS, respectively. ThePoisson model has the notable advantage that it deals naturally with bunching of paymentsat exactly zero, while still allowing for the effects to be proportional. The OLS specification,instead, does not deal with the bunching at zero and does not allow for the effects on amountsto be proportional. The Tobit specification is more appropriate than OLS in the sense that ittakes into account the censored nature of the data at zero, but does not allow for the effectsto be proportional.

The results from columns (3), (4) and (5) of Table 5 are identical in terms of signsand statistical significance of the coefficients, indicating that the results are not qualitativelysensitive to the choice between the three specifications. If anything, the effects are statisticallymore significant when using the OLS and Tobit models. Even though the results from thePoisson, OLS and Tobit models are not directly comparable in terms of magnitudes, theyseem roughly consistent. For example, the Poisson model indicates an effect of audit-statisticsof 6.3%. The Tobit model suggests an effect of audit statistics of $480. Since the averageoutcome is $6,465, this Tobit coefficient amounts to an effect of roughly 7.4%, which is inthe same order of magnitude than the Poisson model.

4.3 Effects by Type and Timing of Tax Payments

If evaders become more concerned with being audited, they should not only change theirreports about the future months, but they should also be expected to revise retroactivelytheir payments for the previous months of the year. To explore this hypothesis further, thefirst two columns of Table 6 splits the effects on “concurrent” and “retroactive” payments.

19

For reference, column (3) shows the effects on total VAT payments, which are identical tothe baseline results from column (1) of Table 4).

Consistent with firms becoming more afraid of being audited, the effects of audit-statisticsand audit-endogeneity were relatively stronger on retroactive payments. On the other hand,the public-goods message does not have a statistically significant effect neither on retroactivepayments nor current payments, although this result has to be taken with care due to thestatistical precision.

One potential concern is that the increases in VAT payments crowded out payments ofother taxes, which could potentially make the net effect on tax revenues to be zero. Onthe other hand, firms that are more concerned with audits may want to increase their VATpayments as well as all other tax payments that could be found to be under-reported inaudits. Indeed, even though the VAT tax comprises the majority of tax payments made bythese firms, there was nothing in the letter sent by the IRS that specifically mentioned theVAT or any other taxes.

Columns (4) through (6) of Table 6 shows the effects of the messages by type of taxes:VAT, other taxes and total (i.e., VAT + other). The evidence suggests that, far from crowdingout other tax payments, the audit-statistics and audit-endogeneity messages had positiveeffect on non-VAT revenues. Indeed, the effects on payments of other taxes are economicallyand statistically as significant as the effect on VAT payments. In contrast, the effect ofthe public-goods message on other tax payments is close to zero (0.1%) and statisticallyinsignificant. Also, the effect of the public-goods message on total tax payments was small(1.8%) and statistically insignificant.

5 Results: Causal Mechanisms

5.1 Evidence from the Audit-Statistics and Audit-Threat Sub-Treatments

In this section, we disentangle between the rational and the salience channels. According tothe Allingham and Sandmo (1972) model, the VAT payments should be increasing in boththe audit probabilities and the penalty rates. Indeed, we can calibrate this model to get somequantitative predictions. The details of this calibration are discussed in Appendix B. Usinga number of alternative specifications, the predicted elasticities with respect to p and θ arealways well above 1.

We start by comparing the behavior of firms within the main treatment arm, audit-statistics, who were randomly assigned to different signals about the audit probabilities and

20

penalty rates. To estimate these elasticities with our experimental data, we use coefficientsfrom a Poisson regression with audit probabilities and penalty rates expressed from 0 to 1.Therefore, the resulting coefficients can be interpreted directly as elasticities. That is, acoefficient on the audit probability with a value of 1 would imply that a 1 percentage pointincrease in the audit probability increases the VAT payments by 1%. Similarly, a coefficienton the penalty rate with a value of 1 would imply that a 1 percentage point increase in thepenalty rates would increase VAT payments by 1%.

Table 7.a shows the estimated elasticities identified from the audit-statistics sub-treatments.The estimated elasticities are close to zero and statistically insignificant. In other words, firmswho were shown higher signals on the audit probabilities and penalty rates did not have sig-nificantly higher VAT payments. The elasticity with respect to the audit probability is 0.030(s.e. 0.236), while the elasticity with respect to the penalty rate is -0.118 (s.e. 0.115).

Given that we devoted a large fraction of our subject pool to this treatment arm, theseelasticities are quite precisely estimated. We can reject the null hypothesis that each ofthese elasticities is equal to the elasticities predicted by the calibrated model (well above 1).Furthermore, we can rule out even smaller effects: the 90% confidence interval for the auditprobability excludes elasticities above 0.418, and the 90% confidence interval for the penaltyrate excludes elasticities above 0.071.

As complementary evidence, we can use the secondary experimental sample, which gener-ated more direct variation in the audit probabilities by randomly assigning firms to the 50%audit-threat letter or the 25% audit-threat letter. Table 7.b shows the results of this treat-ment arm. As in the audit-statistics treatment, the audit probability in the audit-threat lettertakes the values 0.25 and 0.5, which means that the coefficient from the Poisson regressioncan be interpreted as an elasticity, just like above.

The estimated elasticity is small in magnitude (0.376) and borderline statistically signif-icant at the 10% level. We can reject the null hypothesis that this elasticity is equal to theelasticity predicted by the calibrated model (well above 1). Furthermore, the pre-treatment(falsification) coefficient is also borderline statistically significant at the 10% level, indicatingthat the positive coefficient on the post-treatment elasticity is probably spurious.

To assess the robustness of these findings, Figure 3 estimates the effects of the audit-statistics and audit-threat sub-treatments in a less parametric way. Figure 3.a breaks downthe effect of the audit-statistics message by decile of p, and by decile of θ. The zero elasticitydoes not seem to be driven by any particular decile.

Figure 3.b provide the event-study analysis of the audit-statistics message, but brokendown by the above- and below-median values of p; and Figure 3.b presents the equivalentanalysis for θ. Again, the evidence suggests similar and statistically indistinguishable effects

21

of these subgroups of the audit-statistics messages. Last, Figure 3.c provides the event-studyanalysis of the difference between the 25% and 50% conditions of audit-threat. Again, thereis no systematic differences in the evolution of outcomes across these two groups.

A natural candidate to explain why individuals are inelastic to audit probabilities could beprospect theory (Kahneman and Tversky, 1979). According to this framework, the decisionmaking process is based not only on probabilities but also on decision weights that reflectthe impact on the overall value of the prospect. However, prospect theory predicts thatindividuals tend to under-weight small probabilities. As a result, prospect theory wouldpredict that, far from being inelastic, individuals should exaggerate the difference between,say, a signal of p =2% and a signal of p =25%.

5.2 Evidence from the Survey Data

In this section, we disentangle between the rational and the salience channels by measuringthe effects of the audit-statistics and audit-endogeneity messages on beliefs, as measured withsurvey data nine months after the delivery of these messages.

Figure 4.a and 4.b show the perceptions about audit probabilities and penalty rates, re-spectively. The shaded bars show the distribution of perceptions for individuals who receivedthe baseline letter (i.e., perceptions). The red curves corresponds to the distribution of signalssent to the firms in the audit-statistics letters (i.e., reality).

The comparison between the shaded bars and the red curve from Figure 4.a suggeststhat, on average, firms substantially over-estimate the probability of being audited: whileour statistics indicate a probability of roughly 11.7%, the mean perception in the baselinegroup is 37.6% (p-value of the difference is <0.01). The comparison between the shaded barsand the red curve from Figure 4.b suggests that, on average, firms are right about the penaltyrates: the average penalty is 30.7%, while the mean of the perceived penalty is 31.0% in thebaseline group.

A potential explanation for the positive bias in the perceived audit probability is givenby the availability heuristic bias (Kahneman and Tversky 1974). According to that model,individuals judge the probability of an event by how easy it is to recall instances of the event.Even though audits are rare, the fact that they are impactful and salient in the media maytrick firms into judging them more probable than they actually are. Indeed, there is evidencethat individuals over-estimate the probabilities of a wide range of rare events (Lichtensteinet al., 1978; Kahneman, Slovic and Tversky, 1982).

The systematic bias in beliefs in the baseline group comprises evidence against the rational

22

mechanisms and in favor of the salience mechanism: i.e., if the audit-statistics informationeliminated some of this systematic bias, it would bring the perceived audit probability downand thus it should decrease rather than increase tax compliance.

To test this hypothesis more directly, the shallow bars in Figure 4.a and 4.b denote thedistribution of perceptions for firms who received the audit-statistics letter. By comparingthese to the shaded bars we can assess how the audit-statistics message affected perceptions.Figure 4.a shows that, if anything, the audit-statistics message reduced the perceived proba-bility of being audited, from an average of 37.6% to an average of 35.4% – even though thisdifference is not statistically significant, Table C.2 shows that the difference becomes signif-icant once we increase the sample size by pooling subjects in the baseline and public-goodsgroups.17

In turn, Figure 4.b shows that the audit-statistics message had a very small effect onthe perceived penalty rate, from an average of 29.6% to an average of 30.2%. Given thatthe audit-statistics reduced the average perceived audit probability and did not affect theaverage perceived penalty rate, the rational mechanism would suggest that the audit-statisticsmessage should have reduced VAT payments. This prediction is at odds with the observedpositive effect of the audit-statistics message, thus making salience a more likely explanation.

We also included a question about the perceived endogeneity of audits, shown in Figure4.c. The distribution of perceptions in the baseline letter suggests that firms were alreadyaware of this endogeneity. In line with the previous results, relative to the baseline group,there is no statistically significant differences in the distribution of perceptions for the audit-endogeneity group (p-value of 0.61).18

6 Conclusions

Even though firms are believed to incorporate the threat of being audited into their taxevasion decisions, there is no consensus as to whether they do it in the optimal fashionsuggested by the model of Allingham and Sandmo (1972). To study this, we conducteda large-scale field experiment with firms that collectively pay over $200 million dollars intaxes per year, in combination with survey and administrative data. On the one hand,

17Each panel in Figure 4 reports the results from an Epps–Singleton (ES) two-sample test using the em-pirical characteristic function, which is a version of the Kolmogorov–Smirnov test of equality of distributionsthat is valid for discrete data (Goerg and Kaiser, 2009). According to this test, the audit-statistics messagedid not have a statistically significant effect on the distribution of perceptions about audit probabilities andpenalties (p−values of 0.25 and 0.46, respectively).

18In a scale from 1 to 5, where 1 is more evasion increase significantly the likelihood of being audited, and5 means more evasion diminish significantly the likelihood of being audited, the average belief was 1.48 thebaseline group and 1.14 for the audit-endogeneity group.

23

and consistent with the results from Pomeranz (2015), we show that providing informationrelated to audits does increase VAT payments of the firms. However, our evidence suggeststhat this information affects tax evasion decisions primarily through making audits moresalient to the taxpayer, rather than by making the taxpayers revise their beliefs about auditsand subsequently re-optimize their behavior.

Our findings contribute to the debate about the determinants of tax compliance. We findthat firms react to the threat of audits, but not necessarily in the optimal way predicted byAllingham and Sandmo (1972): tax compliance is elastic with respect to the salience of audits,but inelastic with respect to the audit probability and the level of penalties. This suggeststhat firms in our experiment operate in the context of optimization frictions. Moreover,we document large dispersion and biases in beliefs about audits, which persist even afterwe provide accurate information to the firms, which suggests that there are also substantialinformation frictions. While the failure of the Allingham and Sandmo (1972) model topredict evasion rates may be interpreted as an indication that tax compliance depends onfactors such as tax morale (Luttmer and Singhal, 2014) rather than on tax enforcement, ourevidence suggests that tax enforcement indeed matters, but in a context of optimization andinformation frictions rather than in the context of rational learning and perfectly informedand optimizing firms.

Additionally, our findings have some specific policy implications. First, holding fixedthe actual detection rate, tax agencies may benefit from making audits and other detectionmechanisms more salient to the taxpayers.19 Indeed, there is evidence that some governmentsare already doing this. For instance, Blank and Levin (2010) find that the governmentissued a disproportionately large number of tax enforcement press releases during the weeksimmediately prior to Tax Day in order to influence taxpayers’ perceptions while they arepreparing to file their annual tax returns.20

19For a practical discussion on how to implement this type of policy, see for example Morse (2009). Fur-thermore, salience can be used to improve compliance with other laws: for instance, Dur and Vollaard (2016)show experimental evidence that the use of a salience intervention can significantly reduces illegal garbagedisposal.

20Additionally, our findings suggest that increasing transparency about the probability of being auditedmay be detrimental to the goals of tax agencies, by decreasing the average perceptions of the audit probability.Indeed, to the extent that there is little to no information on the Internet about various aspects of the auditingprocess such as audit probabilities and penalty rates, tax agencies may be aware of this already.

24

References[1] Allingham, M. G. and Sandmo, A. (1972). “Income Tax Evasion: A Theoretical Analysis.”

Journal of Public Economics, 1: 323-338.

[2] Alm, J., McClelland, G. H. and Schulze, W. (1992). “Why do people pay taxes?” Journal ofPublic Economics, 48(1): 21-38.

[3] Alm, J., Jackson, B., and McKee, M. (1992). “Estimating the Determinants of Taxpayer Com-pliance with Experimental Data. Economic Development and Cultural Change.”, National TaxJournal 45(1): 107-114.

[4] Andreoni, J, Erard, B. and Feinstein, J. (1998) "Tax Compliance", Journal of Economic Liter-ature, 32(2): 818–860.

[5] Barberis, N. (2013) "The Psychology of Tail Events: Progress and Challenges", AmericanEconomic Review, 103(3), 611-616

[6] Burns, Z., Chiu, A., and Wu, G. (2010). "Overweighting of small probabilities", Wiley Ency-clopedia of Operations Research and Management Science.

[7] Becker, G.S. (1968), “Crime and Punishment: An Economic Approach,” Journal of PoliticalEconomy, 76(2): 169-217.

[8] Beron, K. J., Tauchen, H. V., and Witte, A. D. (1988). “A Structural Equation Model for TaxCompliance and Auditing”, NBER Working Paper No. 2556.

[9] Blank, J. D. and Levin, D. Z. (2010) “When Is Tax Enforcement Publicized?” Virginia TaxReview, 30; NYU Law and Economics Research Paper No. 10-12.

[10] Blumenthal, M.; C. Christian and J. Slemrod (2001), “Do Normative Appeals Affect TaxCompliance? Evidence From a Controlled Experiment in Minnesota,” National Tax Journal,54(1): 125-138.

[11] Castro, L., and Scartascini, C. (2015). “Tax Compliance and Enforcement in the PampasEvidence From a Field Experiment.” Journal of Economic Behavior & Organization, 116, 65-82.

[12] Chetty, R. (2015), “Behavioral Economics and Public Policy: A Pragmatic Perspective,” TheAmerican Economic Review, 105(5): 1-33.

[13] Chetty, R, Looney, A, and Kroft, K.. (2009). “Salience and Taxation: Theory and Evidence.”The American Economic Review 99 (4): 1145–77.

[14] Cowell, F. A. and Gordon, J. P. F. (1988) “Unwillingness to pay, Journal of Public Economics”,Journal of Public Economics, 36(3):305-321.

25

[15] Dirección General Impositiva (DGI) (2013) “Estimating VAT Evasion Through ConsumptionMethod”, Technical Report, Economic Advisory Office.

[16] Dhami S. and al-Nowaihi A. (2007), “Why Do People Pay Taxes? Prospect theory versusexpected utility theory”, Journal of Economic Behavior and Organization, 64(1),171-192

[17] Dur, R, and Vollaard, B. (2016) “Salience of Law Enforcement: A Field Experiment” TinbergenInstitute Discussion Paper 2017-007/VII.

[18] Dwenger, N.; H. Kleven; I. Rasul and J. Rincke (2016), “Extrinsic and Intrinsic Motivationsfor Tax Compliance: Evidence from a Field Experiment in Germany,” American EconomicJournal: Economic Policy 8(3), 203-232.

[19] Erard, B. and Feinstein J. S. (1994). “The Role of Moral Sentiment and Audit Perceptions inTax Compliance”. Public Finance 49 (Supplement), 70 – 89.

[20] Fellner, G.; Sausgruber, R. and Traxler, C. (2013), “Testing Enforcement Strategies in theField: Threat, Moral Appeal and Social Information,” Journal of the European EconomicAssociation 11 (3): 634–660.

[21] Goerg, S. and Kaiser, J. (2009). “Nonparametric testing of distributions—the Epps–Singletontwo-sample test using the empirical characteristic function,” Stata Journal, 9(3):454-465.

[22] Gomez-Sabaini, J. C. and Jimenez, J. P. (2012). “Tax structure and tax evasion in LatinAmerica.” Macroeconomics of Development Series, 118, ECLAC.

[23] Gomez-Sabaini, J.C. and Moran, D. (2014). “Tax policy in Latin America Assessment andguidelines for a second generation of reforms.” Macroeconomics of Development Series, 133,ECLAC.

[24] Harris, L. and Associates, Inc. (1988), “1987 taxpayer opinion survey,” Internal Revenue ServiceDocument 7292, Washington, DC.

[25] Internal Revenue Service Data Book, 2014 (2015), Publication 55B, Washington, DC.

[26] Kahneman, D., and Tversky, A. (1972), “Subjective Probability: A Judgment of Representa-tiveness.” In “The concept of Probability in Psychological Experiments”, edited by: Carl-AxelS. Staël Von Holstein, pp. 25-48, Springer Netherlands.

[27] Kahneman, D., and Tversky, A. (1979). "Prospect theory: An analysis of decision under risk",Econometrica, 47(2), 263-291.

[28] Kahneman, D.; Slovic, P.; and Tversky, A. (1982), “Judgments under uncertainty.”

26

[29] Kleven, H. J.; Knudsen, M. B.; Kreiner, T. ; Pedersen, S.; and Saez, E. (2011), “Unwillingor Unable to Cheat? Evidence from a Randomized Tax Audit Experiment in Denmark,”Econometrica, 79 (3): 651-692.

[30] Konrad, K. A., Lohse, T., and Qari, S. (2016). “Compliance With Endogenous Audit Proba-bilities.” Scandinavian Journal of Economics, forthcoming.

[31] Lichtenstein, S., Slovic, P., Fischhoff, B., Layman, M., and Combs, B. (1978). "Judged fre-quency of lethal events", Journal of experimental psychology: Human learning and memory,4(6), 551.

[32] Luttmer, E. F. P. and Singhal, M. (2014). “Tax Morale.” Journal of Economic Perspectives,28(4):149–168.

[33] McKenzie, D. (2012) “Beyond baseline and follow-up: The case for more T in experiments”,Journal of Development Economics, 99(2): 210-221,

[34] Morse, S. C., (2009) “Using Salience and Influence to Narrow the Tax Gap.” Loyola UniversityChicago Law Journal, 40:483.

[35] Naritomi, J, (2016), “Consumers as Tax Auditors.”, London School of Economics, Workingpaper

[36] Perez-Truglia; R. and Troiano, U. (2015) "Shaming Tax Delinquents: Theory and Evidencefrom a Field Experiment in the United States", NBER Working Paper No. 21264

[37] Pomeranz, D. (2015), “No Taxation Without Information: Deterrence and Self-Enforcementin the Value Added Tax,” The American Economic Review, 105(8), 2539-2569.

[38] Scholz, J. T. and Pinney N. (1995). “Duty, Fear, and Tax Compliance: The heuristic basis ofcitizenship behavior”. American Journal of Political Science 39 (2), 490–512.

[39] Slemrod, J.; Blumenthal, M. and Christian C. (2001), “Taxpayer Response to an IncreasedProbability of Audit: Evidence from a Controlled Experiment in Minnesota,” Journal of PublicEconomics, 79 (3):455-483.

[40] Slemrod, J. and Yitzhaki, S. (2002), “Chapter 22 - Tax Avoidance, Evasion, and Administra-tion,” in “Handbook of Public Economics” edited by: Auerbach, A. J. and Feldstein, M., Vol.3:1423-1470, Elsevier.

[41] Slemrod, J. (2008). “Does It Matter WhoWrites the Check to the Government? The Economicsof Tax Remittance.” National Tax Journal 61 (2): 251–75.

[42] Yitzhaki, S. (1987), “On the Excess Burden of Tax Evasion,” Public Finance Review, 15(2):123-137.

27

Table 1: Experimental Balance Between Treatment GroupsMain Sample Secondary Sample

AuditStatistics

PublicGoods

AuditEndogeneity Baseline p-value test

AuditThreat (25%)

AuditThreat (50%) p-value test

% paid VAT taxes (3 months pre-mailing) 0.92 0.94 0.93 0.93 0.18 0.90 0.89 0.54(0.00) (0.01) (0.01) (0.01) (0.01) (0.01)

Amount of VAT paid (3 months pre-mailing) 1.87 1.96 1.93 1.91 0.56 1.74 1.75 0.95(0.03) (0.07) (0.07) (0.06) (0.10) (0.09)

Years registered in tax agency 15.34 14.75 15.70 15.01 0.27 19.45 19.42 0.94(0.17) (0.22) (0.54) (0.22) (0.28) (0.29)

% audited between 2011-2015 0.14 0.14 0.12 0.14 0.17 0.21 0.20 0.69(0.00) (0.01) (0.01) (0.01) (0.01) (0.01)

Number of employees 4.81 4.66 4.88 5.09 0.96 4.83 4.88 0.80(0.26) (0.54) (0.57) (0.64) (0.13) (0.12)

% retail trade sector 0.22 0.22 0.21 0.23 0.78 0.33 0.32 0.40(0.00) (0.01) (0.01) (0.01) (0.01) (0.01)

% Agricultural, forest and others 0.03 0.03 0.03 0.03 0.77 0.03 0.04 0.12(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)

% construction sector 0.03 0.03 0.03 0.03 0.85 0.03 0.03 0.33(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)

% other sector 0.73 0.73 0.73 0.72 0.95 0.61 0.62 0.54(0.00) (0.01) (0.01) (0.01) (0.01) (0.01)

Observations 10,272 2,017 2,039 2,064 2,015 2,033

Notes: Averages for different pre-treatment firm-level characteristics for treatment groups by type of sample. The main sample includes all firmsselected as it is described in section 3.2 while the secondary sample includes the high risk firms selected by the IRS. Standard errors in parenthesis.The last column of each sample reports the p-value of a test in which the null hypothesis is that the mean is equal for all the treatment groups.Data on VAT amount and firm characteristics from different administrative tax records including monthly payments, annual tax returns and auditingregisters.

28

Table 2: Comparison of Observables across Universe of Firms, Experimental Samples and Survey Sample

Experimental Sample

All firms Main SecondaryInvited tothe survey

% paid VAT taxes (3 months pre-mailing) 0.78 0.93 0.89 0.93(0.42) (0.26) (0.31) (0.26)

Amount of VAT paid (3 months pre-mailing) 3.72 1.89 1.74 1.89(11.55) (2.83) (4.25) (2.98)

Years registered in tax agency 14.21 15.26 19.44 14.46(14.85) (17.16) (12.84) (10.08)

% audited between 2011-2015 0.08 0.14 0.21 0.11(0.28) (0.34) (0.40) (0.31)

Number of employees 12.65 4.84 4.89 6.43(302.98) (26.60) (5.76) (53.77)

% retail trade sector 0.13 0.22 0.33 0.15(0.34) (0.41) (0.47) (0.36)

% Agricultural, forest and others 0.03 0.03 0.03 0.02(0.17) (0.17) (0.18) (0.15)

% construction sector 0.03 0.03 0.03 0.03(0.17) (0.17) (0.17) (0.18)

% other sector 0.84 0.73 0.61 0.80(0.37) (0.45) (0.49) (0.40)

N 120,125 16,392 4,048 3,845

Notes: Column (1) includes all firms that remited at least one payment in 2014 or 2015. Column (2) includes the subset of firms selected for theexperimental sample acording to the criteria described in section 3.2 and is restricted to firms between 1,000 USD and 100,000 USD of added value.We also exclude firms that were tagged by the IRS as accountancy firms and those that did not report to receive the letter. Column (3) represents agroup of high risk firms that were selected from a special sample defined by the IRS and received the audit-threat letter. Column (4) corresponds tofirms with declared and valid e-mail and therefore selected to participate in the on-line survey conducted after the experiment. All data is based ondifferent administrative tax records including monthly payments, annual tax returns and auditing registers.

29

Table 3: Descriptive Statistics about Tax Payments

Mean SD 10th 25th 50th 75th 90th

VAT AmountsPost-treatment 6.47 7.77 0.44 1.30 3.74 8.48 16.55Pre-Treatment 7.77 8.07 0.96 1.99 4.86 10.94 19.73

Retroactive VAT AmountsPost-treatment 0.30 1.40 0.00 0.00 0.00 0.00 0.62Pre-Treatment 0.40 1.76 0.00 0.00 0.00 0.00 0.81

Other Taxes AmountsPost-Treatment 3.30 5.43 0.00 0.95 1.81 3.52 7.42Pre-Treatment 4.07 8.57 0.04 1.43 2.14 4.37 8.72

Notes: In this table we include only firms that received the baseline letter (N=2,064 ). The pre-treatment period rangesfrom September 1, 2014 to August 31, 2015 and the post-treatment period ranges from October 1, 2015 to September 30,2016. Data on payments from administrative data records. VAT amounts corresponds to VAT payments and withholdings.Retroactive VAT amount corresponds to two months or more retroactive VAT payments and withholding, e.g. VATpayments made in March 2016 corresponding to September 2015. Other taxes amount includes other tax payments includingcorporate tax, wealth tax and other specific taxes to bussiness activity.

30

Table 4: Average Effects of Audit-Statistics, Audit-Endogeneity and Public-Goods Messages

By Sector

All Services Goods

a. Audit - Statistics (N= 10,272) vs Baseline (N= 2,064)

Post-Treatment 0.063** 0.092** 0.034(0.025) (0.037) (0.031)

Pre-Treatment -0.008 0.020 -0.024(0.021) (0.030) (0.028)

b. Audit - Endogeneity (N= 2,039) vs Baseline (N= 2,064)

Post-Treatment 0.074** 0.094** 0.075(0.032) (0.043) (0.048)

Pre-Treatment -0.006 0.045 -0.031(0.027) (0.034) (0.037)

c. Public - Goods (N= 2,017) vs Baseline (N= 2,064)

Post-Treatment 0.043 0.025 0.075*(0.030) (0.041) (0.039)

Pre-Treatment -0.004 0.055 -0.042(0.026) (0.036) (0.035)

Notes: * significant at the 10% level, ** at the 5% level, *** at the 1% level. Robust standard error. In each. The resultsare based on a Poisson regression, whose coefficients can be interpreted directly as semi-elasticities. Panel a. comparesthe audit-statistics message with the baseline letter while panels b. and c. replicates the analysis for audit-endogeneityand public goods messages. In the first row of each panel we use the amount in dollars of VAT payments after receivingthe letter as the dependent variable. The second row shows a falsification test in which we estimate the same regressionbut using the amount contributed before receiving the mailing (pre-treatment). All regresions are estimated with a set ofmonthly controls that correspond to the year prior the outcome, i.e. in the post treatment outcome we include montlhyVAT payment controls from September, 2014 to August, 2015 and in pre-treatment outcome we include the same variablesfor the September, 2013 - August 2014 period. We also restrict the analysis to those firms that effectively received theletter. First column shows the result for the total number of firms that received each letter. Columns (2) and (3) show theeffects by industry code grouped by goods firms and services firms.

31

Table 5: Average Effects of Audit-Statistics, Audit-Endogeneity and Public-Goods Messages: AlternativeSpecifications

Prob. Making PositiveVAT Payments

VATPayments

OLS Probit Poisson OLS Tobit

a. Audit - Statistics (N= 10,272) vs Baseline (N= 2,064)

Post-Treatment -0.001 -0.019 0.063** 0.493*** 0.480***(0.004) (0.066) (0.025) (0.140) (0.147)

Pre-Treatment -0.008 0.047 0.049(0.021) (0.128) (0.128)

b. Audit - Endogeneity (N= 2,039) vs Baseline (N= 2,064)

Post-Treatment 0.001 -0.006 0.074** 0.591*** 0.592***(0.006) (0.085) (0.032) (0.189) (0.195)

Pre-Treatment -0.006 0.057 0.060(0.027) (0.169) (0.169)

c. Public - Goods (N= 2,017) vs. Baseline (N= 2,064)

Post-Treatment 0.006 0.045 0.043 0.333* 0.357**(0.006) (0.087) (0.030) (0.171) (0.177)

Pre-Treatment -0.004 0.059 0.065(0.026) (0.162) (0.162)

Baseline Mean 0.943 6.465

Notes: * significant at the 10% level, ** at the 5% level, *** at the 1% level. Robust standard error. Panel a. comparesthe audit-statistics message with the baseline letter while panels b. and c. replicates the analysis for audit-endogeneityand public goods messages. In the first row of each panel we use post treatment information about the correspondingoutcome while the second row shows a falsification test in which we estimate the same regression but using pre-treatmentinformation. All regressions are estimated with a set of monthly controls that correspond to the year prior the outcome,i.e. in the post treatment outcome we include montlhy VAT payment controls from September, 2014 to August, 2015 andin pre-treatment outcome we include the same variables for the September, 2013 - August 2014 period. We also restrict theanalysis to those firms that effectively received the letter. . Columns (1), (2) and (3) show the effect of treatment by type ofpayments. Columns (1) and (2) show the treatment effect on the extensive margin using two alternative strategies. Column(1) shows the treatment effect on the probability of making at least one VAT payment in the post treatment period usinga OLS model. Column (2) replicates the same analysis using a probit model. Columns (3), (4) and (5) present differentestimation strategies for the intensive margin, i.e. the total amount of VAT paid. In column (4) we present the results ofa Poisson estimation, while column (5) uses an OLS regression and column (6) depicts the Tobit estimation results. Thelast row of the table shows the placebo mean for each group of outcome.

32

Table 6: Average Effects of Audit-Statistics, Audit-Endogeneity and Public-Goods Messages: VAT Timingand Other Taxes

VAT,By Payment Timing

VATvs. Other Taxes

Retroactive ConcurrentRetroactive +Concurrent VAT Other

VAT +Other

a. Audit - Statitstics (N= 10,272) vs Baseline (N= 2,064)

Post-Treatment 0.381*** 0.044* 0.063** 0.063** 0.077** 0.052**(0.131) (0.026) (0.025) (0.025) (0.037) (0.026)

Pre-Treatment -0.132 -0.005 -0.008 -0.008 0.016 0.037(0.105) (0.022) (0.021) (0.021) (0.045) (0.026)

b. Audit - Endogeneity (N= 2,039) vs Baseline (N= 2,064)

Post-Treatment 0.314** 0.061* 0.074** 0.074** 0.084** 0.079***(0.139) (0.033) (0.032) (0.032) (0.041) (0.031)

Pre-Treatment 0.008 -0.003 -0.006 -0.006 0.006 0.027(0.125) (0.027) (0.027) (0.027) (0.056) (0.031)

c. Public - Goods (N= 2,017) vs Baseline (N= 2,064)

Post-Treatment 0.195 0.032 0.043 0.043 0.001 0.018(0.134) (0.031) (0.030) (0.030) (0.048) (0.029)

Pre-Treatment -0.192 0.002 -0.004 -0.004 -0.047 -0.004(0.124) (0.026) (0.026) (0.026) (0.046) (0.027)

Notes: * significant at the 10% level, ** at the 5% level, *** at the 1% level. Robust standard error. In each. The resultsare based on a Poisson regression, whose coefficients can be interpreted directly as semi-elasticities. Panel a. comparesthe audit-statistics message with the baseline letter while panels b. and c. replicates the analysis for audit-endogeneityand public goods messages. In the first row of each panel we use the amount in dollars of VAT payments after receivingthe letter as the dependent variable. The second row shows a falsification test in which we estimate the same regressionbut using the amount contributed before receiving the mailing (pre-treatment). All regresions are estimated with a set ofmonthly controls that correspond to the year prior the outcome, i.e. in the post treatment outcome we include monthlyVAT payment controls from September, 2014 to August, 2015 and in pre-treatment outcome we include the same variablesfor the September, 2013 - August 2014 period. We also restrict the analysis to those firms that effectively received theletter. . Columns (1), (2) and (3) show the effect of treatment by type of payments. Columns (1) and (2) show separatelythe effect of treatment on current and retroactive VAT payments, while column (3) shows the overall effect. Columns (4),(5) and (6) present the results by type of tax. While column (4) depicts the effect of the treatment in post experiment VATpayments, column (5) shows the effect on the rest of taxes considered and column (6) shows the effect on the total amountof taxes.

33

Table 7: Effects of Audit-Statistics and Audit-Threat Sub-Treatments

By Sector

All Services Goods

a. Audit - Statistics Letters (N= 10,272)

Audit Probability (%)

Post-Treatment 0.030 -0.020 0.040(0.236) (0.373) (0.288)

Pre-Treatment 0.025 0.143 -0.096(0.115) (0.150) (0.173)

Penalty Size (%)

Post-Treatment -0.118 0.211 -0.316**(0.115) (0.143) (0.160)

Pre-Treatment -0.001 -0.192** 0.180(0.088) (0.081) (0.148)

b. Audit - Threat Letters (N= 4,048)

Audit Probability(%)

Post-Treatment 0.376* 0.766* 0.213(0.210) (0.424) (0.234)

Pre-Treatment -0.342* -0.694** -0.315(0.178) (0.303) (0.194)

Notes: * significant at the 10% level, ** at the 5% level, *** at the 1% level. Robust standard error. Panel a. shows theeffect of providing different information regarding p and in the audit-statistics message. Panel b. compares the differentaudit-threat messages, i.e. the 50% threat of audit vs. the 25% threat of audit. Rows (1) and (3) of panel a. presentthe effect of informing an additional percentage point of p and θ respectively on the post treatment VAT payments. Rows(2) and (4) shows a falsification test in which we estimate the same regression but using pre treatment information. Allregresions are estimated with a set of monthly controls that correspond to the year prior the outcome, i.e. in the posttreatment outcome we include montlhy VAT payment controls from September, 2014 to August, 2015 and in pre-treatmentoutcome we include the same variables for the September, 2013 - August 2014 period. We also restrict the analysis to thosefirms that effectively received the letter. Row (1) in panel b. represents the post treatment effect of receiving the letter of50% threat relative to receive the 25% letter. Row (2) of panel b. replicates the estimates for the pre treatment outcomes.First column shows the result for the total number of firms that received each letter. Columns (2) and (3) show the effectsby industry code grouped by goods firms and services firms.

34

Table 8: Effects of Audit-Statistics and Audit-Threat Sub-Treatments: Alternative Specifications

Prob. Making PositiveVAT Payments

VATPayments

OLS Probit Poisson OLS Tobit

a. Audit - Statistics (N= 10,272)

Audit Probability (%)

Post-Treatment 0.086* 0.915* 0.030 -0.441 0.141(0.045) (0.512) (0.236) (1.617) (1.728)

Pre-Treatment 0.025 0.183 0.212(0.115) (0.943) (0.945)

Penalty Size (%)

Post-Treatment 0.012 0.214 -0.118 -0.362 -0.319(0.023) (0.267) (0.115) (0.856) (0.896)

Pre-Treatment -0.001 -0.123 -0.133(0.088) (0.785) (0.785)

b. Audit - Threat Letters (N= 4,048)

Post-Treatment -0.013 -0.103 0.376* 1.325 1.253(0.030) (0.258) (0.210) (0.900) (0.940)

Pre-Treatment -0.342* -0.958 -0.954(0.178) (0.695) (0.699)

Baseline Mean 0.943 6.465

Notes: * significant at the 10% level, ** at the 5% level, *** at the 1% level. Robust standard error. Panel a. shows theeffect of providing different information regarding p and in the audit-statistics message. Panel b. compares the differentaudit-threat messages, i.e. the 50% threat of audit vs. the 25% threat of audit. Rows (1) and (3) of panel a. presentthe effect of informing an additional percentage point of p and θ respectively on the post treatment VAT payments. Rows(2) and (4) shows a falsification test in which we estimate the same regression but using pre treatment information. Allregresions are estimated with a set of monthly controls that correspond to the year prior the outcome, i.e. in the posttreatment outcome we include montlhy VAT payment controls from September, 2014 to August, 2015 and in pre-treatmentoutcome we include the same variables for the September, 2013 - August 2014 period. We also restrict the analysis tothose firms that effectively received the letter. Row (1) in panel b. represents the post treatment effect of receiving theletter of 50% threat relative to receive the 25% letter. Row (2) of panel b. replicates the estimates for the pre treatmentoutcomes. Column (1) shows the treatment effect on the probability of making at least one VAT payment in the posttreatment period using a OLS model. Column (2) replicates the same analysis using a probit model. Columns (3), (4)and (5) present different estimation strategies for the intensive margin, i.e. the total amount of VAT paid. In column (4)we present the results of a Poisson estimation, while column (5) uses an OLS regression and column (6) depicts the Tobitestimation results. The last row of the table shows the placebo mean for each group of outcome.

35

Table 9: Effects of Audit-Statistics and Audit-Threat Sub-Treatments: VAT Timing and Other Taxes

VAT,By Payment Timing

VATvs. Other Taxes

Retroactive ConcurrentRetroactive +Concurrent VAT Other

VAT +Other

a. Audit - Statitstics (N= 10,272)

Audit Probability(%)

Post-Treatment -2.136** 0.162 0.030 0.030 0.127 0.094(1.040) (0.246) (0.236) (0.236) (0.440) (0.266)

Pre-Treatment -1.984** 0.123 0.025 0.025 0.192 0.111(0.806) (0.121) (0.115) (0.115) (0.388) (0.169)

Penalty Size (%)

Post-Treatment 1.044 -0.191* -0.118 -0.118 -0.291 -0.179(0.742) (0.112) (0.115) (0.115) (0.180) (0.112)

Pre-Treatment 0.032 -0.001 -0.001 -0.001 -0.376** -0.136(0.470) (0.093) (0.088) (0.088) (0.168) (0.086)

b. Audit - Threat Letters (N=4,048)

Post-Treatment 0.826 0.331 0.376* 0.376* 0.178 0.301*(0.944) (0.216) (0.210) (0.210) (0.185) (0.170)

Pre-Treatment 0.013 -0.289 -0.342* -0.342* -0.211 -0.306**(0.647) (0.182) (0.178) (0.178) (0.163) (0.142)

Notes: significant at the 10% level, ** at the 5% level, *** at the 1% level. Robust standard error. Panel a. shows theeffect of providing different information regarding p and in the audit-statistics message. Panel b. compares the differentaudit-threat messages, i.e. the 50% threat of audit vs. the 25% threat of audit. Rows (1) and (3) of panel a. presentthe effect of informing an additional percentage point of p and θ respectively on the post treatment VAT payments. Rows(2) and (4) shows a falsification test in which we estimate the same regression but using pre treatment information. Allregresions are estimated with a set of monthly controls that correspond to the year prior the outcome, i.e. in the posttreatment outcome we include montlhy VAT payment controls from September, 2014 to August, 2015 and in pre-treatmentoutcome we include the same variables for the September, 2013 - August 2014 period. We also restrict the analysis to thosefirms that effectively received the letter. Row (1) in panel b. represents the post treatment effect of receiving the letter of50% threat relative to receive the 25% letter. Row (2) of panel b. replicates the estimates for the pre treatment outcomes.. Columns (1), (2) and (3) show the effect of treatment by type of payments. Columns (1) and (2) show separately theeffect of treatment on current and retroactive VAT payments, while column (3) shows the overall effect. Columns (4), (5)and (6) present the results by type of tax. While column (4) depicts the effect of the treatment in post experiment VATpayments, column (5) shows the effect on the rest of taxes considered and column (6) shows the effect on the total amountof taxes.

36

Figure 1: Distribution of Statistics Shown in Audit-Statistics Letter Typea. Audit Probability (p) b. Penalty rate (θ)

05

1015

2025

Per

cent

0 5 10 15 20 25Probability of being audited (%)

010

2030

40P

erce

nt

0 20 40 60Average Penalty Rate (%)

Notes: The data corresponds to the distribution of the information provided in the audit-statistics letter (N=10,272). Panela. shows the distribution of the probability of being audited (p ). Panel b represents the distribution of the average penaltyrate (θ ). p and θ arise from the following proceedure: 1) divide the firms in total sales revenues quintiles, 2) randomlydraw a sample of 100 similar firms (i.e. from the same quintile) and 3) computing the average p and θ from tan sample.This mechanism lead to almost 950 different combinations of the two parameters.

37

Figure 2: Average Effects of Audit-Statistics, Audit-Endogeneity and Public-Goods Messagesa. Audit-Statistics vs. Baseline b. Audit-Endogeneity vs. Baseline

−0.

10.

00.

10.

2

Tre

atm

ent E

ffect

(p.

p)

−2 −1 +1 +2 +3 +4

Quarter

Point Estimate 90% CI

−0.

10.

00.

10.

2

Tre

atm

ent E

ffect

(p.

p)

−2 −1 +1 +2 +3 +4

Quarter

Point Estimate 90% CI

c. Public-Goods vs. Baseline Letter

−0.

10.

00.

10.

2

Tre

atm

ent E

ffect

(p.

p)

−2 −1 +1 +2 +3 +4

Quarter

Point Estimate 90% CI

Notes: This figure plots the trimestral effect of each treatment arm in comparison to the baseline letter . Panel a. (N=12,336)shows the effect of the audit-statistics message on the quarterly total amount of VAT payments, while panel b. (N=4,103)represents the effect of the audit-endogeneity message and panel c. (N=4,081) depicts the effect of the public goods messagein the same variable. Each point (red circle) in the plot represents the estimate of the effect of treatment on VAT paymentsfor a specific trimester from two trimesters before treatment, and up to 4 trimesters after receiving the letter. Regressionsdo not include monthly pre-treatment controls. The estimates correspond to Poisson regressions . Confidence intervals,represented by red lines, are computed with heteroskedastic-robust standard errors.

38

Figure 3: Effects of Audit-Statistics and Audit-Threat Sub-Treatmentsa. Effect of audit-statistics vs. baseline, by deciles of p and θ b. Effect of audit-statistics vs. baseline, by level of p

0.0

5.1

.15

Aud

. Sta

tistic

Effe

ct (

%)

0 4 8 12 16 20Probability of being audited (%)

coef_p Fitted values

0.0

5.1

.15

Aud

. Sta

tistic

Effe

ct (

%)

15 20 25 30 35 40 45 50Average penalty rate(%)

coef_M Fitted values

−0.

10.

00.

10.

2

Tre

atm

ent E

ffect

(p.

p)

−2 −1 +1 +2 +3 +4

Quarter

Above Below

c. Effect of audit-statistics vs. baseline, by level of θ d. Effect of audit-threat, p = 0.50 vs. p = 0.25

−0.

10.

00.

10.

2

Tre

atm

ent E

ffect

(p.

p)

−2 −1 +1 +2 +3 +4

Quarter

Above Below

−1.

0−

0.5

0.0

0.5

1.0

Tre

atm

ent E

ffect

(p.

p)

−2 −1 +1 +2 +3 +4

Quarter

Point Estimate 90% CI

Notes: Panel a. and b. plots the trimestral effect of each parameter (pandθ) in the audit-statistics letter. Panel c. plots thetrimestral effect of the 50% letter in comparison to the 25% letter. Panel a. (N=10,272) shows the effect of the p reportedin the audit-statistics message on the quarterly total amount of VAT payments, while panel reports the same estimationsfor θ. Panel c. (N=4,048) represents the effect of the audit-threat message (50% vs 25%). Each point in the plot. representsthe estimate of the effect of treatment on VAT payments for a specific trimester from two trimesters before treatment, andup to 4 trimesters after receiving the letter. Red points in panels a. and b. represent the effect of those who receivedthe audit-statistics letter with a reported porθ above the median. Green points represent the same effect but for thosepandθ below the median. Regressions do not include monthly pre-treatment controls. The estimates correspond to Poissonregressions . Confidence intervals, represented by red and green lines, are computed with heteroskedastic-robust standarderrors.

39

Figure 4: Distribution of Beliefs, by Treatment Groupa. Audit Probability - p b. Penalty rate - θ

Note: ES test p−value: 0.25

010

2030

40P

erce

nt

0 20 40 60 80 100Probability of being audited (%)

Perceived − BaselinePerceived − Aud. StatisticsShown

Note: ES test p−value: 0.46

010

2030

40P

erce

nt

0 20 40 60 80 100Average Penalty Rate

Perceived − BaselinePerceived − Aud. StatisticsShown

c. Endogeneity

Note: ES test p−value: 0.61

020

4060

80P

erce

nt

Incr

ease

a lo

t

Incr

ease

little

No ch

ange

s

Reduc

e a

little

Reduc

e a

lot

Perceived BaselinePerceived − Aud. Endogeneity

Notes: Histograms are based on owners responses to the post treatment survey . Perceived - Baseline (N=61) refers torespondents who received the baseline letter during the experimental stage, while Perceived - Aud. Statistics (N=341) andPerceived Endogeneity (N=76) refers to respondents who received audit-statistics and audit-endogeneity letters respectively.These answer correspond to the questions Q2, Q4 and Q6 of the survey questionnaire (appendix A.7). In panel a. thex-axis represents the probability of being audited; in panel b. it represents the average penalty rate while in panel c. itrepresents the different categories allowed in the relationship between evasion and probability of being audited question.Perceived Baseline refers to the histogram of the owners of firms that received the baseline letter. Perceived Aud. Statisticsand Perceived Endogeneity refers to the histograms of firms that effectively received audit-statistics and audit-endogeneityletters. Shown refers to the distribution of the information in the audit-statistics letter informed to the firms treated withthe audit-statistics message.

40

Online Appendix: For Online Publication Only

A Replication Material: Letters and Survey

This appendix shows samples of the five letter types: baseline letter (A.1), audit-statistics letter (A.2),audit-endogeneity letter (A.3), public goods letter (A.4) and audit-threat letter (A.5). Additionally, Ap-pendix A.6 shows a sample of the invitation email sent by the IRS to complete the online survey, andAppendix A.7 shows the module from that survey that we included with the questions that are mostrelevant for our experiment.

i

A.1 Sample Letter: Baseline Letter

Montevideo, August 20th 2015 Mr../Ms. Taxpayer: The DGI has the authority to perform inspections (see Art. 68 of the tax code) and routine audits of taxpayers on the basis of crosschecks and assessment of data compiled to detect oversights and inconsistency on tax returns as well as pending tax debts. The aim of the DGI, and the primary challenge it faces, is to ensure the collection of revenue to sustain life in society. Additionally, its task is to generate a framework of fair and transparent competition where the failure of some to meet their obligations does not have an unfavorable impact on honest taxpayers. In order to meet these goals, inspections are performed in a routine fashion. Your micro, small, or medium-sized business has been randomly selected to receive this information. It is solely for your information and its receipt does not require you to present any documentation to the DGI offices. We ask you to comply with your tax obligations for the sake of the country we all want, a more and more developed Uruguay with greater and greater social cohesion. Sincerely,

Collection and Controls Division

Internal Revenues Services

ii

A.2 Sample Letter: Audit-Statistics Letter

Montevideo, August 20th 2015 Mr../Ms. Taxpayer: The DGI has the authority to perform inspections (see Art. 68 of the tax code) and routine audits of taxpayers on the basis of crosschecks and assessment of data compiled to detect oversights and inconsistency on tax returns as well as pending tax debts.

On the basis of historical information on similar businesses, there is a likelihood of p% that the tax returns you filed for this year will be audited in one of the coming three years. If, pursuant to that auditing, it is determined that tax evasion has occurred, you will be required to pay not only the amount previously unpaid, but also a fee of approximately M% of that amount.* The aim of the DGI, and the primary challenge it faces, is to ensure the collection of revenue to sustain life in society. Additionally, its task is to generate a framework of fair and transparent competition where the failure of some to meet their obligations does not have an unfavorable impact on honest taxpayers. In order to meet these goals, inspections are performed in a routine fashion. Your micro, small, or medium-sized business has been randomly selected to receive this information. It is solely for your information and its receipt does not require you to present any documentation to the DGI offices. We ask you to comply with your tax obligations for the sake of the country we all want, a more and more developed Uruguay with greater and greater social cohesion. Sincerely,

Collection and Controls Division

Internal Revenues Services

* Estimates based on data from the 2011-2013 period for a group of firms with similar characteristics in terms of, for instance, total invoicing. The likelihood of being audited was calculated as a percentage of audited firms in a random sub-sample of firms. The rate of the fee was estimated as an average of a random sub-sample of audits.

iii

A.3 Sample Letter: Audit-Endogeneity Letter

Montevideo, August 20th 2015 Mr../Ms. Taxpayer: The DGI has the authority to perform inspections (see Art. 68 of the tax code) and routine audits of taxpayers on the basis of crosschecks and assessment of data compiled to detect oversights and inconsistency on tax returns as well as pending tax debts.

The DGI uses data on thousands of taxpayers to detect firms that may be evading taxes; most of its audits are aimed at those firms. Evading taxes, then, doubles your chances of being audited. The aim of the DGI, and the primary challenge it faces, is to ensure the collection of revenue to sustain life in society. Additionally, its task is to generate a framework of fair and transparent competition where the failure of some to meet their obligations does not have an unfavorable impact on honest taxpayers. In order to meet these goals, inspections are performed in a routine fashion. Your micro, small, or medium-sized business has been randomly selected to receive this information. It is solely for your information and its receipt does not require you to present any documentation to the DGI offices. We ask you to comply with your tax obligations for the sake of the country we all want, a more and more developed Uruguay with greater and greater social cohesion. Sincerely,

Collection and Controls Division

Internal Revenues Services

iv

A.4 Sample Letter: Public-Goods Letter

Montevideo, August 20th 2015 Mr../Ms. Taxpayer: The DGI has the authority to perform inspections (see Art. 68 of the tax code) and routine audits of taxpayers on the basis of crosschecks and assessment of data compiled to detect oversights and inconsistency on tax returns as well as pending tax debts.

If those who currently evade their tax obligations were evade 10% less, the additional revenue collected would enable all of the following: to supply 42,000 portable computers to school children; to build 4 high schools, 9 elementary schools, and 2 technical schools; to acquire 80 patrol cars and to hire 500 police officers; to add 87,000 hours of medical attention by doctors at public hospitals; to hire 660 teachers; to build 1,000 public housing units (50m2 per unit). There would be resources left over to reduce the fiscal burden. The tax behavior of each of us has direct effects on the lives of us all. The aim of the DGI, and the primary challenge it faces, is to ensure the collection of revenue to sustain life in society. Additionally, its task is to generate a framework of fair and transparent competition where the failure of some to meet their obligations does not have an unfavorable impact on honest taxpayers. In order to meet these goals, inspections are performed in a routine fashion. Your micro, small, or medium-sized business has been randomly selected to receive this information. It is solely for your information and its receipt does not require you to present any documentation to the DGI offices. We ask you to comply with your tax obligations for the sake of the country we all want, a more and more developed Uruguay with greater and greater social cohesion. Sincerely,

Collection and Controls Division

Internal Revenues Services

v

A.5 Sample Letter: Audit-Threat Letter

Montevideo, August 20th 2015 Mr../Ms. Taxpayer: The DGI has the authority to perform inspections (see Art. 68 of the tax code) and routine audits of taxpayers on the basis of crosschecks and assessment of data compiled to detect oversights and inconsistency on tax returns as well as pending tax debts.

We would like to inform you that the business you represent is one of a group of firms pre-selected for auditing in 2016. A p% of the firms in that group will then be randomly selected for auditing. The aim of the DGI, and the primary challenge it faces, is to ensure the collection of revenue to sustain life in society. Additionally, its task is to generate a framework of fair and transparent competition where the failure of some to meet their obligations does not have an unfavorable impact on honest taxpayers. In order to meet these goals, inspections are performed in a routine fashion. Your micro, small, or medium-sized business has been randomly selected to receive this information. It is solely for your information and its receipt does not require you to present any documentation to the DGI offices. We ask you to comply with your tax obligations for the sake of the country we all want, a more and more developed Uruguay with greater and greater social cohesion. Sincerely,

Collection and Controls Division Internal Revenues Services

vi

A.6 Sample Letter: Invitation to the Online Survey

Dear Taxpayer: The DGI’s strategic objectives for this period include improving taxpayer services. In 2013, the first Survey on the Costs of Tax Compliance for Small and Medium-Sized Businesses was administered with the support of the Inter-American Center of Tax Administrations (CIAT) and the United Nations (UN). The DGI, in conjunction with a group of academics, has designed a new version of the survey (for more information, visit www.dgi.gub.uy). You can give us your answers on the website where you will find instructions on how to fill out the simple questionnaire; the entire process should take no more than fifteen minutes.

Respond to survey To address these concerns, a random sample of taxpayers will receive a survey to be answered anonymously. You are one of the randomly selected taxpayers, which is why you have received this communication. We are grateful for the time and effort you dedicate to assessing this questionnaire and to responding to it as precisely as possible. Let me assure you that the survey is completely anonymous and the selection of recipients entirely random. The success of this project lies in the precision of your responses. It is on the basis of those responses and the real information they provide that the DGI will be able to hone the design, in the present and in the future, of its strategies to reduce the costs of compliance. If you have any questions about this questionnaire, please send an e-mail to [email protected]. We would like to thank you once again for your contribution to this project, which we are sure will benefit all taxpayers. Sincerely, Joaquín Serra Director of the Income Tax Department PS: If the "Respond to survey" link doesn’t open, copy the following address in your browser:https://URL.

vii

A.7 Excerpt from the Online Survey Questionnaire

Introductory Text: We would like you to respond to a survey about the costs of paying taxes. We hope you have the ten minutes that responding to the questionnaire will require. We are interested in your opinion and hope you will be frank in your responses, which are anonymous and used only for statistical purposes. We would like to thank you for your participation. Questions Included in Main Module: Q1) Have you been subject to a DGI audit (inspection or monitoring) at any point in the last three years? Yes. No. Q2)In your opinion,what is the likelihood that the tax returns filed by a company like yours be audited at least once in the next three years (from 0% to 100%)?

Q3) How sure are you of your response? Not at all sure. A little sure. Somewhat sure. Very sure. Q4)Let’s imagine that a company like yours is audited and that tax evasion is detected. What, in your opinion, is the penalty (in %) as determined by law that the firm must pay in addition to the originally unpaid amount? For example, a fee of X% means that, for each $100 not paid, the firm would have to pay those original $100 plus $X in fees.

Q5) How sure are you of your response? Not at all sure. A little sure.

viii

B A Calibration of Allingham and Sandmo (1972)

Let Y be the total value-added. Let τ be the value added tax rate (in practice, it should include the VATplus all the other taxes that the individual must pay as a consequence of declaring higher value-added,like for example income tax, benefits for formal employees, etc.). Let E be the amount to be under-declared (so τ · E is the amount evaded), p be the probability of audit and θ be the penalty (as definedin audit-statistics). The optimal evasion is given by maximizing the expected utiltiy:

maxE

1 − p

1 − σ(Y (1 − τ) + τE)1−σ + p

1 − σ(Y (1 − τ) − θτE)1−σ (B.1)

The VAT remitances elasticities with respect to the audit parametes (p and θ) depend on the set ofaudit parameters, the tax level (τ), and the CRRA utility function parameters σ

∂τ(Y − E)∂p

p

τ(Y − E) = −(1 − τ)∂x∂p

(1 + θ)(θx+ 1)2

p[τ − (1 − τ)

(x−11+θx

)] (B.2)

∂τ(Y − E)∂θ

θ

τ(Y − E) = −(1 − τ)(1 + θ)∂x

∂θ− x(x− 1)

(θx+ 1)2θ[

τ − (1 − τ)(x−11+θx

)] (B.3)

with

x =(

pθ

1 − p

)− 1σ

∂x

∂p= − 1

σ

(pθ

1 − p

)− σ+1σ θ

(1 − p)2

∂x

∂θ= − 1

σ

(pθ

1 − p

)− σ+1σ p

1 − p

We calibrate the model with two sets of parameters. In the first one, we use the actual parameters i.e.the level of audit probability p is 0.113 and the amount of fines (θ) is 30.3% of the amount evaded. Thesecond set of parameters are firm owner perception in the survey conducted after the experiment. Theseparameters in the case of audit probability is three times higher than the actual one at 0.376, and withthe same magnitude in the case of the fines at 31%. For both parameter sets in Allingham and Sandmo(1972) model, a rational individual would evade 100% of his tax duties for the range of risk aversion levelused in the literature.

In Table B.1 we show the computed elasticities respect to p and θ and the evation rate asuming aCRRA utility function with a wide range of risk aversion parameter from 1 to 35. We calibrate the modelwith the actual set of audit paremeters in Panel A, and the perceived audit parameters in Panel B.

ix

Panel A of Table B.1 shows predicted elasticities with the mean value of actual audit characteristics.We find larger elasticities for those individuals with less risk aversion, but only for σ larger than 9.6we find evasion rates lower than 1. Moreover, to find the evasion rate (26%) which is estimated forLatin America in Gómez Sabaini and Jiménez (2012) we need the risk aversion coeffcient at 35. Thisrate means, tax payers are indiferent to receive $102.05 for sure than participate in a lottery with equalprobabilities to receive $100 or $200.

Panel B of Table B.1 shows the predicted elasticities and aversion rate of the AS model calibratedwith the perceived set of audit parameters. Here, the risk aversion coefficient to find a 26% of evasionrate is also high (18.3). This risk aversion means for the same lottery the tax payers are indifferent witha sure payment fo $103.9.

x

Table B.1: Effects of Audit and Penalties Rates, Audit-Statistics and Audit-Threat Sub-TreatmentsA. Actual { p,θ} B. Perceived { p,θ}

p = 0.113, θ=0.303 and τ = 0.22 p = 0.367, θ=0.31 and τ = 0.22

σ ∂τ(Y−E)∂p

pτ(Y−E)

∂τ(Y−E)∂θ

θτ(Y−E)

EY

∂τ(Y−E)∂p

pτ(Y−E)

∂τ(Y−E)∂θ

θτ(Y−E)

EY

< 4.93 1 15 1 193.150 215.149 0.9869.63 1 2.667 2.887 0.50510 80.588 50.503 0.961 2.383 2.583 0.47415 5.361 3.272 0.624 1.182 1.271 0.31218.3 3.288 1.989 0.507 0.903 0.964 0.2620 2.738 1.650 0.462 0.784 0.839 0.23225 1.832 1.096 0.366 0.586 0.626 0.18530 1.374 0.818 0.304 0.468 0.498 0.15435 1.099 0.652 0.259 0.389 0.414 0.131

Notes: The results are based in the computation of the elasticities described in the equations B1 and B2, and the evationrate calibrated for a wide range of CRRA risk aversion parameter from 1 to 35. In Panel A, audit parameters are calibratedwith the actual ones and tax rate is calibrated with the VAT tax. In Panel B, audit parameters are calibrated with those whoare effectively peceived by the firm owners. Only for σ above to 9.63 the evation rate estimated for the actual parametersis belo to 1, and in the case of the perceived audit parameters σ should bw above of 4.93.

xi