WPS8522 Policy Research Working Paper 8522 Tax Evasion in Africa and Latin America The Role of Distortionary Infrastructures and Policies Wilfried A. Kouamé Jonathan Goyette Africa Region Office of the Chief Economist July 2018 Policy Research Working Paper 8522 Abstract This paper examines the impact of the quality of the busi- capacity of the tax agency. The model simulations for each ness environment as well as the monitoring capacity of the country in the African and Latin American sample show tax agency on firms’ tax evasion and production decisions. that the model can explain 35 percent of the variation in First, the paper uses firm-level data for 30 African and Latin tax evasion and more than 49 percent of the dispersion in American countries to show that tax evasion and distortions output per worker across the sample countries. Finally, a stemming from the business environment are positively series of counterfactual experiments shows that, at the cur- and significantly correlated, while sales not reported for tax rent level of deterrence, governments could decrease sales purposes and institutional quality are negatively and sig- not reported for tax purposes by 21 percent, by reducing nificantly correlated. Second, the paper develops a general distortions stemming from the business environment by equilibrium model where heterogeneous firms make tax half. The paper presents empirical supporting evidence evasion decisions based on their assessment of the quality consistent with testable predictions of the model. of their business environment as well as the monitoring This paper is a product of the Office of the Chief Economist, Africa Region. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/research. The authors may be contacted at wkouame@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Tax Evasion in Africa and Latin America: The Role of Distortionary Infrastructures and Policies Wilfried A. Kouamé∗ and Jonathan Goyette† Keywords: Distortions, tax evasion, business environment, infrastructure, institutional quality, Africa and Latin American. JEL codes: H2; H3; H5; O1 ∗ Corresponding author: Wilfried A. Kouamé - The World Bank and Economics Department, Université de Sherbrooke: wkouame@worldbank.org/wilfried.kouame@usherbrooke.ca. Authors are grateful to Diego Restuccia, Théophile Azomahou, Moussa Blimpo, Jean-François Rouillard, Cesar Calderon, and Punam Chulan-Pole for their comments. The paper has also benefited from presentations at Oxford University CSAE 2017 Conference, Canadian Economics Association and Public Economic Theory conferences in 2016. Earlier versions of the paper circulated under the titles "Distortions, Policy Ineffectiveness and Tax Evasion" and "Distortions, Social Infrastructures and Tax Evasion". † Economics Department, Université de Sherbrooke I Introduction A sound tax system and a healthy business environment are crucial for the welfare of an economy (Easterly and Rebelo, 1993; Aterido et al., 2011). However, many African and Latin American economies face major challenges on both fronts. Indeed, high levels of tax evasion are ubiquitous in these countries. On average, 25% of total sales are not reported for tax purpose in African and Latin American countries compared with only 7% in OECD countries.1 This shortfall has important social and economic consequences, as evaded taxes reduce a government’s ability to invest in productive infrastructures and institutional quality, and to provide public goods and services. Moreover, insufficient government resources result in an inefficient business environment as various distortions (power outages, corruption, etc.) hinder firms’ performance and their ability to create jobs (Besley and Burgess, 2004; Aterido et al., 2011; Buera et al., 2013; Goyette and Gallipoli, 2015). As such, African and Latin American economies fare poorly in terms of ease of doing business.2 The coexistence of tax evasion and inefficient business environments thus raises the ques- tion about a potential link between these variables in developing countries. This paper aims to examine how distortions stemming from the business environment and institutional qual- ity, i.e., a tax agency’s monitoring capacity, affect firms’ production and tax evasion decisions. We focus on two specific mechanisms. First, entrepreneurs, knowing that institutional quality is low, also anticipate low monitoring of taxes and have, therefore, more incentive to evade their taxes. Second, we argue that a poor business environment generates costs for firms and that this creates a wedge between firms’ potential and realized profits. As a result, firms have an incentive to under-report their sales for tax purposes in order to compensate for the losses incurred due to the distortions in their business environment. Using firm-level data from the World Bank Enterprise Surveys (WBES) as well as data from the World Governance Indicators (WGI), we provide supporting evidence consistent 1 See Table A.1 in Appendix A. 2 See Appendix A. 2 with the mechanisms described above. Tax evasion is proxied by the amount of sales not declared for tax purposes. Distortions from the business environment are calculated based on the losses in annual sales due to power outages, corruption, etc. Finally, institutional quality is based on various variables related to the perceptions of the quality of public services and the credibility of the government. We find a positive and significant relationship between losses due to distortions from the business environment and tax evasion. On the contrary, institutional quality is negatively related to firms’ tax evasion. However, these relationships are subject to omitted variable bias, potential measurement error, and reverse causality. As a result, we are not able to identify, using standard econo- metrics, the mechanisms through which distortions and institutional quality affect firms’ tax evasion behavior. Instead, we develop a general equilibrium model and use simulations to match the empirical evidence. Building on Restuccia and Rogerson (2008), the economic environment consists of (i) one representative household which maximizes his inter-temporal utility, (ii) a government which balances its budget and, (iii) heterogeneous firms which max- imize their profits and make tax evasion decisions based on their idiosyncratic productivity level as well as their idiosyncratic level of distortions stemming from the business environ- ment. We calibrate the model to the United States and treat that country as an economy with no distortions as is standard in the literature. This benchmark economy allows nor- malizing GDP per worker in both the model and the data. The model is then simulated for each of the 30 African and Latin American countries from the WBES sample, using the benchmark distribution of productivity and the idiosyncratic distribution of losses due to the distortions from the business environment of a specific country. The model explains 35 percent of the variation of tax evasion and 49 percent of the dispersion of output per worker in the data. The simulation for each region provides similar findings. As robustness checks, we calibrate the model to the Chilean economy, which exhibits the lowest combination of distortions and tax evasion in the data. Moreover, the model is simulated using an alternative measure of the deterrence probability. These additional 3 robustness checks show that the findings remain robust, and the model explains at least 20 percent of the variation of tax evasion across countries in all cases. Having established the ability of the model to replicate some relevant moments in the data related to tax evasion and output across African and Latin American countries, we conduct a set of tax neutral counterfactual experiments to examine various policy implications. First, we examine what happens in Guinea, the country with the highest level of tax evasion in the data, when distortions are reduced to the average level of the sample. Such improvements generate a drop between 18.34 and 40.16 percent in tax evasion while output per worker increases by between one- to six-fold. Second, we examine what happens when governments could reduce distortions stemming from the business environment by half. Such reduction could reduce sales not reported for tax purposes by about 21%. This paper is closely related to Restuccia and Rogerson (2008) and Bah and Fang (2015). The authors argue in the former paper that a country’s policies and institutions can create taxes or subsidies (distortions) on establishment output. These distortions reduce aggregate total factor productivity (TFP) and can explain up to 50% of the cross-country differences in output, capital accumulation, and TFP. Bah and Fang (2015) introduce distortions as an idiosyncratic tax on output in the general equilibrium model of Amaral and Quintin (2010). In addition to a distribution of productivity approximated with firms’ size, Bah and Fang (2015) use the distribution of distortions from the data and collateral constraints to explain some of the variations in output in Africa. This paper differs from Restuccia and Rogerson (2008) and Bah and Fang (2015) by focusing on the effect of distortions and monitoring capacity on firms’ tax evasion behavior. The paper highlights two mechanisms explaining the role of distortions stemming from the business environment and institutional quality on tax evasion in African and Latin American countries. Well-developed infrastructures and institutions are essential to promoting economic growth by reducing transaction cost for firms as well as for households. Firms’ performance is af- fected by the quality of infrastructures such as transport, energy, water, and sanitation as 4 those infrastructures and services are used in the production processes and delivery of goods and services (Bah and Fang, 2015). However, in developing countries, the cost of trans- portation, logistics, telecommunication, water, electricity, security, and bribes are high, and firms suffer great losses due to the poor quality of public infrastructures and services (Eifert et al., 2006). The latter increases transaction costs and makes firms less competitive and productive than their international counterparts (Bah and Fang, 2015). Similarly, inefficient institutions in developing countries create barriers to opportunities and increase costs and risks for microenterprises as well as multinationals (World Bank, 2005; Botero et al., 2004). Inefficient institutions and policies limit market access and increase the size of the unoffi- cial economy (Botero et al., 2004; López de Silanes et al., 2002). Also, recent literature on policy distortions unanimously demonstrates that ineffective public policies lower aggregate total factor productivity (TFP) and explain an important share of TFP dispersion across countries (Hseih and Klenow, 2009; Restuccia and Rogerson, 2013; Wu, 2018). The bene- fits of improving the institutional quality are not limited to developing countries as Prado (2011) shows on a sample of OECD countries that policies reducing regulation costs have a significant positive impact on the supply of both private and publicly produced goods, and effectively reduce the size of the informal sector. We show in this paper that distortions stemming from the business environment and institutional quality affect firms’ tax evasion decisions and production. The remainder of the paper is structured as follows. Section II describes the data and examines the relationship between distortions and firms’ tax evasion empirically. Section III presents the theoretical model. In section IV, we calibrate the model using the United States as a benchmark economy; we then describe our quantitative analysis and the results of the counterfactual experiments. Section V assesses the sensitivity of the findings. The concluding remarks and policy implications are discussed in section VI. 5 II Empirical Evidence This section provides empirical evidence consistent with the testable predictions of the gen- eral equilibrium model in the next section. We first provide descriptive evidence before using a multilevel mixed model to examine the impact of distortionary infrastructures and policies on firms’ tax evasion. The multilevel mixed model allows dealing with the hierarchic structure of the data and takes into account the potential dependence between firms of a given coun- try. Moreover, this methodology allows including both microeconomic and macroeconomic variables while accounting for country fixed effects. II.1 Data and descriptive statistics We use the World Bank Enterprise Surveys (WBES) data which are a collection of a firm- level surveys of a representative sample (random stratified sampling) of firms mainly in developing countries. Questionnaires cover a wide range of business environment topics like infrastructure, performance measures, crime, corruption, competition, access to finance. The surveys are conducted within a framework of common guidelines in the design and implementation. A module of identical questions included in all questionnaires is used for assembling the data set, which allows cross-country comparisons. The analysis focuses on African and Latin American countries having at least 200 establishments as well as tax evasion and distortions data.3 The distributions of firms used in the structural model below are representative at the country level. The sample comprises 19,490 firms in 30 African and Latin American countries during the period 2002-2006.4 The appendix provides a list of 3 The sample also excludes countries that were involved in armed conflicts over the period of the survey. We do so to ensure that the measure of institutional quality, as well as the costs stemming from the business environment, are not tainted by armed conflicts. All conclusions of the paper remain the same without these restrictions and using countries with at least 100 establishments. 4 The dataset employed in this paper does not include informal firms due to data issues. However, we acknowledge that informality is pervasive in developing countries and might be connected with low domestic resource mobilization. As discussed by Besley and Persson (2014), the informal sector is inherently hard to tax because transactions are not recorded, and incomes from informal firms are difficult to measure. Moreover, informality is a source of misallocation (D’Erasmo and Boedo, 2012) affecting both productivity and tax collection (Ordóñez, 2014). Consequently, we expect in this paper that the estimated effects of distortionary infrastructures and policies provide only lower bounds for tax evasion in African and Latin 6 countries. Tax evasion is captured by the proportion of total sales not reported for tax purposes.5 Distortions are measured as the sum of the losses (in percentage of total sales) due to power outage or surges from the public grid, insufficient water supply, unavailable main line tele- phone service, transport failures, crime (loss due to theft, robbery, vandalism or arson) and gifts or informal payment to public officials to "get things done".6 Institutional quality is proxied using a measure of the effectiveness of public policies from the World Governance Indicators (WGI). More particularly, institutional quality captures the perceptions of the quality of public services, the quality of the civil service and the degree of its independence from political pressures, the quality of policy formulation and implementation, and the credibility of the government’s commitment to such policies. We use the percentile distribution of these public policies effectiveness variables. In the regressions below, we also account for a set of firm individual characteristics. Firms’ access to finance is measured as the share of working capital financed by commercial banks. Regulatory burden is measured by the percentage of time the senior management spends dealing with requirements imposed by government regulation. We include the percentage of the firm owned by foreign interests and the government, firm’s age captured by three cate- gorical variables: young (1-5 years old), mature (6-15 years old), and older (more than 15 years old), and the percentage of the establishment’s sales exported. Finally, we account for the size of the firms using four dummy variables capturing firms’ size categories. Microen- terprises have fewer than 10 permanent employees. Small and medium have between 11 and American countries, as the paper does not account for the distortions generated by informal firms and their impacts on tax collection. 5 The question is: "Recognizing the difficulties many enterprises face in fully complying with taxes and regulations, what percentage of total sales would you estimate the typical establishment in your area of activity reports for tax purposes?" 6 The questions related to the components of distortionary infrastructures are stated as follows. (i) "Please estimate the losses (as a percentage of total sales) of theft, robbery, vandalism or arson against your estab- lishment in the last year." (ii) "What percentage of your total sales value was lost last year due to power outages, insufficient water supply, unavailable mainline telephone service, and transport failures?" (iii) "We have heard that establishments are sometimes required to make gifts or informal payments to public officials to "get things done" about customs, taxes, licenses, regulations, services, etc. On average, what percentage of annual sales value would such expenses cost a typical firm like yours?" 7 50 or between 51 and 200 permanent employees, respectively. Firms with more than 200 permanent employees are classified as large firms. Older firms and larger firms are used as the reference categories in the regression analysis. Table 1 presents the descriptive statistics. On average, about 23.84 percent of total sales are not reported for tax purposes (tax evasion) and 4.69 percent of annual sales are lost due to the poor quality of infrastructures, crime, and informal payment to public officials (dis- tortionary infrastructures). Approximately, 75.77 percent of establishments’ working capital is financed by commercial banks. On average 12.16 percent of senior management’s time is spent dealing with requirements imposed by government regulation each week (taxes, cus- toms, labor regulations, licensing and registration), and only 9.80 percent of establishments’ total sales are exported. Firms are on average 20 years old. Finally, the sample contains 27.2 percent of microenterprises, 43.4 percent of small firms, 20 percent of medium firms and 9.4 percent of large firms. Figure 1 plots the correlation between tax evasion and distortionary infrastructures across countries. Each circle represents one country. As it can be seen, there is a positive correlation between tax evasion and the losses stemming from the business environment. Conversely, Figure 2 presents the evidence of a negative relationship between institutional quality and tax evasion. This evidence supports that high distortionary infrastructures and policies are correlated with the proportions of sales not reported for tax purposes. Table 1. Summary statistics Variable Mean Std. Dev. Min. Max. Tax evasion 23.84 32.53 0 100 Distortionary infrastructures 4.74 10.53 0 100 Access to finance 75.77 31.79 0 100 Management time 12.16 18.21 0 100 Foreign share 10.33 28.70 0 100 Government share 0.52 6.55 0 100 Sales exported 11.76 24.02 0 100 Age 19.95 17.88 0 196 Notes. All variables are expressed in %, except age. 8 GIN 60 SWZ Tax Evasion (% of total sales) AGO BWA MRT TZA UGA NIC 40 PAN BRA CRI ECU GTM NAM MEX 20 BOL PRY SLV RWA ARG COL URY HND CHL KEN PER ZAF MDG MAR 0 0 5 10 15 20 Distortions (% of total sales) Figure 1. Distortionary infrastructures and Tax evasion GIN 60 SWZ Tax Evasion (% of total sales) AGO MRT TZA BWA UGA NIC 40 PAN BRA CRI ECU GTM NAM MEX 20 PRY BOL RWASLV ARG COL HND URY KEN CHL PER ZAF MDG MAR 0 0 .2 .4 .6 .8 Policy effectiveness Figure 2. Distortionary policies and Tax evasion 9 II.2 Methodology: Multilevel mixed model This section examines the role of distortionary infrastructure and policies on firms’ tax eva- sion using a multilevel mixed model. This model takes into account the fact that firms in a given country share similar contextual characteristics such as institutional environment, macroeconomic framework, and policies which affect their tax evasion behavior. Standard estimation methods ignore such clustering effects which generate biased standard errors. As firms in a given country may not be independent, standard errors in standard estimation methods may be underestimated. The model allows the intercept to vary across countries (Hox et al., 2010). Moreover, this methodology allows to include both microeconomic and macroeconomic variables as well as country fixed effects. The approach considers a two-level model where the highest level is the country, and the lowest level is the firms such that: Level 1: T ax_Evasionic = α0c + βDic + ηXic + γIQc + δc + µs + ηt + ic , ic ∼ N (0, σ 2 ) (1) Level 2: α0c = α00 + ϑc , ϑc ∼ N (0, σ 2 ), ϑc ⊥ ic (2) Combining the two previous equations, the model can be written as follows: T ax_Evasion_ic = α00 + βDic + γIQc + ηXic + δc + µs + ηt + ϑc + ic (3) where, ϑc + ic is the error term of the model with ϑc the country-specific error term and ic the firm-level error term. T ax_Evasion_ic refers to the proportion of sales not reported for tax purposes of the firm i in the country c. Dic denotes distortions, IQc institutional quality, and Xic defines firm individual characteristics. All these variables are the same as defined previously. Finally, δc , µs , and ηt refer to country, sector and year fixed effects respectively. These fixed effects control for potentially important omitted variables at the country, sector and year level while accounting for differences in demand conditions, productive structure and culture of opportunistic behaviors. 10 II.3 Results Table 2 reports the findings of the estimation of the equation (3). As it can be seen from the first row, distortions are positively related to firms’ tax evasion. These coefficients are statistically significant at the 1 percent level and suggest that a 1 percent loss in total sales due to distortionary infrastructures increases firms’ tax evasion from 0.11 to 0.14 percent. Conversely, sales not reported for tax purposes decrease with institutional quality (last row). A one-unit increase in the index of the public policies effectiveness decreases firms’ tax evasion from 62.79 to 70.71 percentage points. Moreover, as it can be seen in columns (3) to (8) having a higher working capital financed by commercial banks decreases firms’ tax evasion, suggesting that access to finance reduces the likelihood that an establishment under-reports its total sales for tax purposes. Financially constrained establishments seem to use tax evasion as an alternative source of financing or as a way to survive in a highly distorted environment. Columns (7) and (8) shows that larger and medium establishments are less likely to underreport sales for tax purposes (large firms are the omitted category).7 Finally, foreign ownership status is negatively associated with firms’ tax evasion. The coefficients are statistically significant at the 1 and 5 percent levels. All estimates include country, sector and year fixed effects. Although these results corroborate similar findings in the literature, one should treat them with caution as we do not tackle potential endogeneity issues such as omitted variable bias, potential measurement error, and reverse causality. Moreover, the empirical approach limits the possibility to identify the mechanisms through which the losses stemming from the poor business environment and public policies ineffectiveness affect firms’ tax evasion. That is why we develop a theoretical model in the next section, which allows examining two mechanisms relating distortions, institutional quality, and tax evasion behavior. 7 The findings remain the same capturing the size and age of the firms by continuous variables. 11 Table 2. Impacts of distortionary infrastructures and policies on firm’s tax evasion (1) (2) (3) (4) (5) (6) (7) (8) Dependent variable: Tax evasion Distortions 0.137*** 0.132*** 0.127*** 0.122*** 0.122*** 0.121*** 0.109** 0.107** (0.0406) (0.0445) (0.0448) (0.0455) (0.0455) (0.0457) (0.0457) (0.0451) Management Time -0.0193 -0.0198 -0.0201 -0.0201 -0.0190 -0.0151 -0.0146 (0.0240) (0.0244) (0.0243) (0.0243) (0.0246) (0.0237) (0.0237) Access to finance -0.0230* -0.0235* -0.0236* -0.0230* -0.0226* -0.0221* (0.0128) (0.0128) (0.0128) (0.0129) (0.0125) (0.0124) Foreign share -0.0457*** -0.0459*** -0.0410*** -0.0267** -0.0279** (0.0140) (0.0140) (0.0130) (0.0121) (0.0120) Government share -0.0339 -0.0340 -0.00391 -0.00209 (0.0374) (0.0371) (0.0353) (0.0352) Sales exported -0.0300* -0.00792 -0.00915 (0.0170) (0.0157) (0.0161) Age -0.0384 -0.0199 (0.0259) (0.0193) Microenterprise 8.402*** 7.917*** (1.623) (1.586) 12 Small 4.369*** 4.103*** (1.466) (1.414) Medium 1.351 1.246 (1.167) (1.145) Young 2.219 (2.044) Mature 0.696 (0.868) Institutional Q. -62.79*** -70.71*** -69.15*** -68.11*** -68.03*** -69.04*** -66.46*** -67.08*** (16.34) (14.38) (14.91) (15.63) (15.69) (16.19) (16.53) (16.30) Observations 19,490 17,250 17,100 17,089 17,089 17,068 17,068 17,068 Notes. The table presents the estimates of the effects of distortionary infrastructures and policies on firms’ tax evasion using a multilevel mixed effects model. All regressions include country, sector, and year fixed effects. Robust standard errors clustered at the country-sector level in parentheses. ***, **, * denote significance at the 1, 5, and 10 percent level. III The model In this section, we develop a general equilibrium model to analyze potential mechanisms linking distortions and institutional quality to firms’ tax evasion decision. We identify two mechanisms that can explain the relationship between these variables. First, entrepreneurs, knowing that institutional quality is low, also anticipate low monitoring of their taxes and have, therefore, more incentives to evade their taxes. The second mechanism is when dis- tortionary infrastructures and policies generate losses for firms by creating a wedge between firms’ potential and realized profits. We assume that firms compensate for some of these losses by evading their taxes. The economic environment consists of (i) firms which maximize their profits and are het- erogeneous along two dimensions: productivity and the level of distortions they face due to unsound business infrastructures and policies; (ii) one representative household which maximizes its inter-temporal utility, and (iii) a government which balances its budget. III.1 Representative household Consumers are aggregated through a representative household with preferences described by the following utility function: ∞ β t u(ct ) t=0 where ct is consumption at date t and β ∈ (0, 1) is the discount factor. The household is endowed with one unit of productive time in each period and K0 > 0 units of the capital stock at date 0. We assume that u(ct ) satisfies the usual Inada conditions. The representative household maximizes its lifetime utility subject to the following budget constraint: ∞ ∞ (ct + Kt+1 + (1 − δ )Kt ) = (rt Kt + wt Nt + Πt ) t=0 t=0 13 where, wt and rt are, respectively, the rental price of labor lt and capital Kt at t, and Nt the total labor supply to the market. We assume that the representative consumer does not value leisure, i.e. Nt = 1. Finally, Πt is total profits from the operations of all firms. III.2 Firms Entrepreneurs are risk-neutral and produce a homogenous consumption good. Each estab- lishment i makes production decisions to maximize profits based on an idiosyncratic produc- tivity level zi , which is constant over time and varies across establishments. Values of the parameter zi , are drawn from a probability density function of g (z ). The production function F (zi , A, ki , li ) takes as input capital ki and labor li . We include a variable A capturing insti- tutional quality. This variable is the same for all the establishments in the same country and captures the quality of public policies formulation and implementation as well as the quality of public services. The production function is assumed to exhibit decreasing returns to scale in both capital and labor, and to satisfy the usual Inada conditions: η γ F (zi , A, ki , li ) = zi Aki li , 0 < η + γ < 1. There is a proportional tax τ on establishments’ total sales, which is assumed to have two components, τ c and τid . The first component τ c represents a tax collected by the government to finance institutional quality, which can be assimilated public goods and services. The second component τid refers to the fraction of output which is lost due to distortionary infrastructures. To summarize τ = τid + τ c . We assume that establishments differ in their idiosyncratic levels of productivity and distortions. In each period an establishment reports a proportion (1 − ϑi ) of its total sales for tax purposes based on its idiosyncratic level of distortions. Hence, each establishment evades a proportion ϑi of its total sales. We assume there is a probability p to be detected for tax evasion. In this case, the evading firm must pay a fine. The expected fine is assumed to be a convex 14 function of the establishment’s tax evasion so that the marginal cost of tax evasion is positive and increasing in tax evasion: B = pϑθ i F (zi , A, ki , li ), with θ > 2. The expected profit for an establishment with productivity zi is given by: π (zi , τi ) = (1 − τi )(1 − ϑi )F (zi , A, ki , li ) + ϑi F (zi , A, ki , li ) −wli − rki − pϑθ i F (zi , A, ki , li ) −cf (4) T B where cf is a fixed cost of operation for an incumbent establishment, p is the probability of detection, w and r are, respectively, the rental prices of labor and capital. T is the amount of total sales evaded. We assume that this amount is beneficial for the establishment (private benefit) but is a dead loss at the country level. By rearranging equation (4) we have π (zi , τi ) = [1 − (1 − ϑi )τi − pϑθ i ]F (zi , A, ki , li ) − wli − rki − cf (5) For the sake of simplicity, we abandon the index i; however, it is understood that z , k ,l, τ , ϑ are different for each establishment in what follows. III.3 Government The government provides public goods and services with institutional quality A by balancing its budget in each period: zmax zmax τ c (1 − ϑi )F (zi , A, ki , li )d(τ, z ) + pϑθ i F (zi , A, ki , li )d(τ, z ) ≥ C (A, p) 1 1 The government revenues appear on the left hand side (LHS) of this equation. The first component is the revenues from proportional taxation on establishments’ total sales. The second component is the revenues from detection activities. The right hand side (RHS) is the cost associated with the provision of institutional quality A and detection activities p which is assumed to be a convex cost function in both A and p. 15 III.4 Equilibrium We consider the steady-state competitive equilibrium of the model in which the decision problems are described as follows. III.4.1 Consumer’s problem Using the first order conditions, we find a solution to the consumer’s problem with the rental price of capital r and the consumption c being constant, that is: 1 r= − (1 − δ ) (6) β III.4.2 Incumbent establishment’s problem Solving for the first order conditions, the optimal tax evasion and factor demands are: 1 θ− 1 ϑ∗ = [ τ] 1 (7) θp 1 − θ 1 θ− 1 1 γ γ α 1−γ k ∗ = [zA(1 − (1 + ( τ ) 1 )τ )] 1−α−γ [ ] 1−α−γ [ ] 1−α−γ (8) θ pθ w r 1 − θ 1 θ− 1 1 γ 1−α α α l∗ = [zA(1 − (1 + ( τ ) 1 )τ )] 1−α−γ [ ] 1−α−γ [ ] 1−α−γ (9) θ pθ w r Given that the establishment-level productivity and tax rate are constant over time, the discounted present value of an incumbent establishment is given by : π (z, τ ) V (z, τ ) = (10) (1 − ρ) 16 1−λ where ρ = 1+r−δ is the discount rate for the establishment and λ the probability of death which is assumed to be constant. Substituting ρ in equation (10) we have (1 − δ ) + r V (z, τ ) = π (z, τ ) (11) (λ − δ ) + r III.4.3 Entry Establishments’ entry decision is made on the basis of the distribution over potential draws ¯(z, τ ) denote the optimal entry decision with the convention that the for the pair (z, τ ). Let x ¯(z, τ ) = 1. The actual discounted value of establishment enters and remains in operation if x a potential entrant Ve is given by: Ve (z, τ ) = x(z, τ )V (z, τ )d(z, τ ) − ce ] max [¯ ¯∈[0,1] x (z,τ ) where, ce is the entry cost paid by a new establishment, and g (z, τ ) is the probability density function. In an equilibrium with entry, the free entry condition is fulfilled for Ve (z, τ ) = 0. In the steady state and according to equations (6) and (11), V (z, τ ) is determined only by the endogenous variable w. Hence, there is a unique value of the wage rate w for which Ve = 0 as in Restuccia and Rogerson (2008). III.5 Invariant distribution of establishment Let E and µ(z, τ ) denote, respectively, the mass of entrants and the distribution of firms in ¯(z, τ ), the next period t. Given the decision rule for production of entering establishments x period distribution of firms µ over the pair (z, τ ) satisfies the following condition: µ (z, τ ) = (1 − λ)µ(z, τ ) + x ¯(z, τ )d(z, τ )E where (1 − λ)µ(z, τ ) refers to the mass of incumbent establishments that have survived, ¯(z, τ )d(z, τ )E represents the mass of entering establishments that enter and remain in and x 17 ˆ(z, τ ) is characterized by a operation. The unique invariant distribution of establishment µ constant distribution of µ over time and a death rate bounded away from 0. This invariant distribution of establishments is given by: ¯(z, τ ) x ˆ(z, τ ) = E µ d(z, τ ) λ III.5.1 Labor market clearing Given values for w and r, the steady state of this model is characterized by the functions ¯(z, τ ), ¯ ¯(z, τ ), k ϑ ¯(z, τ ) and the associated invariant distribution of establishments l(z, τ ), x ˆ(z, τ ). Using these functions, the aggregate labor demand and the steady-state equilibrium µ level of entry are given by: N (r, w) = E ¯ µ(z, τ ) l(z, τ )ˆ (z,τ ) N (r, w) E= ¯ µ(z, τ ) (z,τ ) l (z, τ )ˆ III.5.2 Definition of a competitive equilibrium The steady-state competitive equilibrium with entry is a set of prices {w∗ , r∗ }, a set of decision rules {k ∗ , l∗ , ϑ∗ ; c∗ , K ∗ }, a distribution of establishments µ(z, τ ), value functions π (z, τ ), V (z, τ ), Ve (z, τ ), and a mass of entry E such that : 1. Given prices (w, r) and preferences, the pair {c∗ , K ∗ } maximizes the consumer lifetime utility; 2. Given prices (w, r), the functions π (z, τ ), V (z, τ ), and Ve (z, τ ) solve incumbent and entering establishment’s problems, with {k ∗ , l∗ , ϑ∗ } the optimal policy functions; 3. The free-entry and invariant distribution conditions are satisfied i.e. Ve = 0, 18 ¯(z, τ ) x µ(z, τ ) = E d(z, τ ), ∀ z, τ λ 4. Market clearing conditions are satisfied : c + δK = ¯ ¯ [f (z, A, k, l) − cf ]µ(z, τ ) − ce E (z,τ ) K= ¯(z, τ )µ(z, τ ) k (z,τ ) 1= ¯ l(z, τ )µ(z, τ ) (z,τ ) III.6 Theoretical predictions From the optimal tax evasion equation (7), we can draw two predictions: ∂ϑ∗ Prediction 1: ∂τ d > 0, ∀ τ ∈ ]0, 1]. An increase in distortions stemming from busi- ness environment increases firms’ tax evasion. ∂ϑ∗ Prediction 2: ∂p < 0, ∀ p ∈ ]0, 1]. A higher probability of detection reduces firms’ tax evasion. 19 IV Quantitative model In this section, we calibrate the model using the United States economy and then simulate it on a sample of 30 African and Latin American countries. IV.1 Calibration The model is calibrated on the United States, which is considered as an economy with no distortions as is standard in the literature. Several of the parameter values are assigned following the literature. The period in the model corresponds to one year in the data. Preferences. The yearly interest rate is targeted to 4% as in Restuccia and Rogerson (2008) 1 and Bah and Fang (2015). This implies a value of β = (1+0.04) = 0.96. Technology. We follow the literature by using 0.85 as the returns to scale of the production function.8 Parameter values η and γ are attributed to match capital and labor shares of 1 2 income, respectively, 3 and 3 of the returns to scale of the production function. We choose δ so that the investment to output ratio is equal to 20%, implying δ = 0.08 as in Restuccia and Rogerson (2008). The range of employment across establishments in the data determines the range of establishment-level productivity. According to the United States data from the Census Bureau,9 the number of employees at the establishment level ranges from 1 to 10,000. Hence, the minimum and maximum levels of productivity are chosen to obtain the range of employment, as in the data. Normalizing the lowest firm-level productivity to 1, the highest level of productivity is chosen to obtain the maximum number of employees of 10,000, as in the data. We approximate the distribution of establishment-level productivity with 100 grid points. We choose a log-spaced grid so that the invariant distribution of establishment size across employment level matches the data. Descriptive statistics about firms highlight one important stylized fact. Establishments with fewer than 20 employees represent 86.1 percent of all the establishments. However, these establishments account only for 24.9 percent of the 8 See for instance Restuccia and Rogerson (2008), Atkeson and Kehoe (2005), and Pavcnik (2002). 9 The data can be downloaded from the website of the US Census Bureau: http://www.census.gov/ econ/susb/data/susb2007.html in the table "U.S. & states, totals". 20 employment. Distortions We use the distortions from WBES in the same way as we did in the empir- ical section as a proxy of distortionary infrastructures. As mentioned previously, we sum up the percentage of sales lost due to poor infrastructure and services (losses due to power outage or surges from public grid, insufficient water supply, unavailable mainline telephone service, transport failures); crime (losses due to theft, robbery, vandalism or arson against the establishment); and corruption (informal payments to "get things done"). For a given country, we compute losses for each establishment and use the distribution of distortionary infrastructures across establishments in the simulation. Institutional Quality We calibrate institutional quality A using the measure of public policies effectiveness from the Worldwide Governance Indicators (WGI) as described above. Recall that this variable is defined as perceptions of the quality of public services, the quality of the civil service and the degree of its independence from political pressures, the quality of policy formulation and implementation, and the credibility of the government’s commitment to such policies. We use the percentile distribution of policy effectiveness variable ranged from 0 to 1.10 In the calibration, we approximate the detection probability using this in- dex of public policies effectiveness. We argue that effective government in formulating and implementing public policies will also be effective in detecting illegal activities such as tax evasion. In the sensitivity analysis, we relax this assumption using an alternative measure of detection probability and show that the predictions of the paper remain the same and are not driven by this assumption. In the model, the cost of tax evasion for an establishment is assumed to be a convex function, with a convexity parameter θ > 2. This assumption guarantees that the marginal cost of tax evasion is positive and increasing with the share of sales not reported for tax purposes. The convexity parameter θ is calibrated to 3. Sensitivity analysis shows that the findings are 10 To the best of our knowledge, the proportional tax rate on sales is not publicly available for each country in the sample. Calibrating institutional quality using the data allows us minimizing errors arising from a not accurate calibration of the proportional tax rate on sales τ c for each country. The proportional tax rate on sales τ c is therefore normalized to 0 in the quantitative model. 21 robust using alternative values of θ. Table 3 summarizes the parameter values. Table 3. Benchmark calibration Parameter Value Description Source τd [0; 1] Distortionary infrastructures WBES A [0; 1] Public policies Effectiveness WGI p [0; 1] Detection probability WGI β 0.96 Real rate of return Literature η 0.283 Capital income share Literature γ 0.567 Labor income share Literature δ 0.08 Investment to output ratio Literature λ 0.1 Annual exit rate Literature θ 3 Evasion cost parameter Assumption ce 1 Entry costs Literature cf 0 Fixed costs Literature z [1; 3.98] Distribution of productivity Relative labor demand Notes. See Bah and Fang (2015), Restuccia and Rogerson (2008), Atkeson and Kehoe (2005) and Pavcnik (2002) for parameter values from the literature. 22 IV.2 Quantitative analysis The quantitative analysis consists in validating the model and conducting counterfactual ex- periments to assess the implications of changes in the distortions stemming from the business environment and the detection probability. IV.2.1 Validation of the model We simulate the model for 30 African and Latin American countries using the calibrated parameters described above. For each country, the model predicts the level of tax evasion and output per worker. Figure 3 plots GDP per worker from the model against GDP per worker from the World Bank’s World Development Indicators (WDI). The reported value of the GDP per worker is normalized by the United States levels in both the model and the data. Each circle represents one country and corresponds to the correlation of output per worker between the simulated data and the WDI data. The straight line is obtained from an OLS regression between the model and the data. As it can be seen, the predicted values of output per worker in the model are positively correlated with the data. The coefficient is statistically significant at the 1% level, and the regression coefficient is 4.58. Similarly, Figure 4 plots the level of tax evasion from the model and the data. The predicted values of tax evasion are highly correlated with the WBES data. The regression coefficient is 0.92 and is statistically significant at the 1% level. Focusing on the R-squared, the model explains 49% of the variation of the GDP per worker and 35% of the dispersion of tax evasion among African and Latin American countries. 23 .5 CHL .4 MEX ZAF .3 URY EGY BWA PAN Model NAM ARG BRA CRI SWZ COL .2 ECU SLV GTM MAR PER AGO PRY HND BOL NIC .1 MRT KEN GIN UGA MDG TZA RWA 0 0 .02 .04 .06 .08 Data R−squared=.4931813103211073 Figure 3. GDP per worker - data vs the model predictions GIN .6 SWZ AGO BWA TZA UGA MRT NIC .4 PAN Model BRA CRI GTM ECU NAM MEX BOL .2 SLV RWA PRY ARG COL URY HND EGY CHL KEN PER ZAF MDG MAR 0 .1 .2 .3 .4 .5 Data R−squared=.3505665524954171 Figure 4. Tax evasion - data vs the model predictions 24 IV.2.2 Counterfactual experiments Having established the ability of the model to replicate some of the variations of tax evasion and output per worker across African and Latin American countries, we conduct in this sec- tion three tax neutral counterfactual experiments to examine the implications of changes in the distortions stemming from the business environment and the detection probability. Tax neutral experiments guarantee that the tax revenues remain the same as in the benchmark. In those experiments, the tax revenues from the decrease in tax evasion are used as a subsidy to production, i.e., the price of the final good is now a function of this subsidy. The counterfactual experiments focus on Guinea, the country with the highest level of tax evasion in the data. We first examine what happens regarding tax evasion, ceteris paribus, when the losses stemming from the business environment in Guinea are reduced to the aver- age level observed in the sample. In particular, we replace the distribution of losses stemming from business environment across size in Guinea by the average distribution of losses across firms’ size from the sample. All the parameters of the model remain the same as previously, except the losses stemming from the business environment. In the second experiment, we explore the second mechanism by examining the impact of a change in Guinea’s deterrence probability. In particular, we increase the deterrence probability by 50 percent in Guinea. All other parameters remain the same. Finally, we combine in a third experiment both changes, i.e., a distribution of distortions similar to the average level of the sample and a detection probability 50% larger. Table 5 reports the change on key variables relative to the baseline findings on Guinea. Experiment 1. In the first scenario, we conduct a tax neutral experiment analysis and set the distribution of losses stemming from business environment to the average level by size category in the sample. As described in Table 4, the distribution of losses by size in Guinea is between 14.97 and 20.60 percent of sales, whereas in the sample the latter is between 5.18 and 7.36 percent of their sales. On average, counterfactual experiment 1 corresponds to a 25 65.25 percent drop in the losses stemming from the business environment in Guinea. Table 4. Distribution of losses in Guinea vs. the sample Establishment size Average losses Average losses (number of employees) in Guinea in the sample < 10 19.59 7.36 10 to 19 19.24 7.00 20 to 49 14.97 6.39 50 to 99 20.60 5.49 ≥ 100 17.00 5.18 The results show that a drop to the average level by size category generates a 26.72 percent drop in sales not reported for tax purposes in Guinea. Guinean firms’ sales not reported for tax purposes decline from 53.74 percent to 39.38 percent. As discussed above, losses stemming from the business environment are a proportional tax on firms’ sales and create therefore a wedge between potential and realized profits. In such environment, a drop in additional costs stemming from business environment decreases sales not reported for tax purposes as witnessed in experiment 1. Using the proceeds from increased tax revenues to subsidize production, a drop in distortionary infrastructures in Guinea to the average level of the sample generates a 1.06-fold increase in output per worker. The aggregate TFP is one of the channels through which the output per worker is stimulated. As Table 5 shows, the aggregate TFP increases slightly by 0.16 percent. Also, a drop in distortions stemming from the business environment stimulates the entry of firms in the market and the aggregate capital as both increase by 22 percent. This experiment seems to benefit more to medium and large firms, which represent firms having between 20-100 and more than 100 employees. The share of firms in both groups increases by about 8.23 and 2.12 percent respectively while the share of small firms in the distribution of establishments decreases by 0.98 percent. Moreover, this experiment gener- ates a scale effect as the average size of firms increases by 21.52 percent. This is a direct consequence of the increase in the share of medium and large firms in the distribution of establishments. Finally, this experiment implies a reallocation of the labor force towards 26 large firms. The share of total employment in large firms grows by around 3 percent, while both medium and small firms face a drop in their employment share. Experiment 2. In the second counterfactual experiment, we examine what happens in Guinea when the detection probability can be improved by 50 percent while the distribution of losses stemming from the business environment is left untouched. When the detection probability is increased from 0.0634 to 0.0951, sales not reported for tax purposes decline from 53.73 percent to 43.88 percent, corresponding to a 18.34 percent drop in firms’ tax eva- sion in Guinea. The improvement of the institutional quality raises the expected costs related to tax evasion activities as monitoring and auditing increase. Consequently, the incentive for firms to underreport their sales for tax purposes declines. In this tax neutral experiment, the output per worker is nearly 6.33-fold as large as the initial output per worker of the country.11 Compared to the baseline findings on Guinea, a 50 percent improvement in the institutional quality causes the aggregate TFP to rise by 208 percent and boosts the entry of firms by 173 percent. Similarly, the aggregate capital increases by 173 percent. These important changes in aggregate TFP and the entry of firms might explain the substantive growth in output per worker. This experiment facilitates the emergence of medium firms as their share in the distribution of firms increases by 8.71 percent while the share of small firms decreases by 0.98 percent for a constant share of large firms. The evidence is consistent with the literature on the missing middle in the size distribution of firms in developing countries, where the quality of the business environment is identified as 11 Although the change in the output per worker might appear high, they are in line with similar evidence in the literature. IMF (2003) shows for instance that improving the quality of institutions in Cameroon (-0.72) to the average level of institutions in the sample (0.13) generates almost a 5-fold increase in income per capita in Cameroon. Rodrik et al. (2004) find that one standard deviation increase in institutional quality generates a 6.4-fold difference in income per capita. Similarly, Acemoglu et al. (2001) show that improving the quality of institutions in Nigeria to the level of Chile could lead to a 7-fold increase in Nigeria’s income. Also, Hall and Jones (1999) show in a cross-country regression that difference in institutions between Niger and the United States which is only 3.78-fold is more than enough to explain the 35-fold difference in output per worker. The contribution of Hall and Jones (1999), one of the most important in the literature, highlights how a small change in institutions can generate an exponential variation in output per worker. 27 a critical factor explaining the limited number of medium-sized firms.12 The counterfactual experiment shows that improving the quality of business environment favors the distribution of firms toward medium-sized firms. Moreover, the experiment shows no scale effect. The average size of firms decreases slightly by 4.69 percent. Finally, the improvement in the institutional quality generates a reallocation of the labor force towards medium firms, which account for an 8.86 percent increase in employment share. Experiment 3. Finally, we combine both changes, i.e., a reduction in distortions by 65.25% on average and an increase in detection by 50% and observe a decrease in tax evasion by 40.16 percent, i.e., sales not reported for tax purposes declines from 53.74 percent to 32.16 percent, while the output per worker is now nearly 6.26-fold larger than its initial level. As it can be seen, the combination of a drop in losses stemming from the business environment with an improvement of the institutional quality generates a more substantive decline in sales not reported for tax purposes relative to previous experiments. In this experiment, both changes contribute to improving the overall business environment. Qualitatively the channels are similar to those described above: as distortions in infrastructures drop and in- stitutional quality increases, the aggregate TFP, the mass of firms entering on the market, and the aggregate capital increase by 185.86 and 200.22 percent respectively. Moreover, we observe a reallocation of firms towards medium and large firms and a scale effect. As Table 5 shows, the shares of medium and large firms increase 8.23 and 2.12 percent respectively. The average size of firms grows by 20.99 percent suggesting thereby a scale effect. Finally, this experiment generates a reallocation of the labor force towards large firms. The total employ- ment share of large firms increases by about 3 percent, while the employment shares of small and medium firms decrease by 16.43 and 6.53 percent respectively. 12 See for instance Sleuwaegen and Goedhuys (2002) and Tybout (2000) for extensive discussion on the missing middle in developing countries. 28 Table 5. Effects from counterfactual experiments Counterfactual experiments Variables Experiment 1 Experiment 2 Experiment 3 Tax evasion -26.72 -18.34 -40.16 Output per worker 1.06 6.33 6.26 Aggregate TFP 0.16 208.07 184.86 Aggregate capital 22.92 173.24 200.22 Mass of entry 22.92 173.24 200.22 Share of small firms -0.98 -0.98 -0.98 Share of medium firms 8.23 8.71 8.23 Share of large firms 2.12 0 2.12 Employment share - Small -16.75 - 0.09 -16.43 Employment share - Medium -6.94 8.86 -6.53 Employment share - Large 3.04 -1.04 2.95 Average size of firms 21.52 -4.69 20.99 Notes. Except for the change in output per worker that is reported in fold, all effects are reported in percentage change. Here small, medium and large firms refer to firms having between 1-19, 20-99, and 100 and more employees. In summary, the counterfactual experiments show that a drop in losses stemming from the business environment along with an improvement in detection probability in Guinea could decrease sales not reported for tax purposes by between 17.78 and 32.86 percent. The drop in tax evasion combined with the improvement in institutional quality generate between a 1.06- and 6.33-fold increase output per worker in Guinea through their effects on the aggregate TFP, the mass entering on the market, and the reallocation of firms and employment. 29 V Sensitivity analysis This section assesses the sensitivity of the findings simulating the model for each region, employing alternative detection probability and a recalibrating the model on the Chilean economy. V.1 Simulations for each region So far, the model shows the role of distortionary infrastructures and policies in explaining the variation in tax evasion and output per worker across African and Latin American countries. In this section, we simulate the model for the two groups of countries separately. The intuition of this exercise is twofold. First, we evaluate how the model fares in explaining the variation of tax evasion and GDP per worker within each region. Second, we show that compositional effects across the continents do not drive the findings of the paper. The simulation procedure is the same as previously, except that we are now focusing on each region. Figure 5 plots tax evasion and output per worker from the model against the data for African countries. As previously, each circle represents one country and the straight line is obtained from the OLS regression between the model and the data. The simulated data are highly correlated with the actual data, with a regression coefficient of around 0.88 for tax evasion and 5.58 for GDP per worker. Based on R-squared values, the model explains 31% of the variation of tax evasion and 43% of the dispersion in GDP per worker across African countries. Sim- ilarly, Figure 6 plots tax evasion and output per worker from the model against the data for Latin American countries. For this group of countries, the regression coefficient is 0.77 for tax evasion and 3.98 for GDP per worker. Using the R-squared, the model explains 23% of the dispersion of tax evasion and 71% of the variation of the output per worker in Latin American countries. All regression coefficients are statistically significant. These findings are comparable to those obtained in the overall sample. 30 .4 GIN .6 ZAF SWZ AGO .3 BWA TZA UGA MRT EGY BWA NAM .4 SWZ Model Model .2 MAR NAM AGO .2 RWA EGY .1 MRT KEN ZAF KEN MDG MAR GIN UGA MDG TZA RWA 0 0 .1 .2 .3 .4 .5 0 .01 .02 .03 .04 Data Data R−squared=.3079162075672508 R−squared=.4296512555503219 (a) Tax evasion (b) GDP per worker Figure 5. Tax evasion and GDP per worker - data vs the model predictions across African countries .5 NIC .4 PAN CHL .4 BRA VEN MEX .3 CRI Model Model GTM ECU .3 URY PAN MEX ARG BRA CRI COL BOL .2 SLV PRY .2 ARGCOL ECU SLV HND GTM PER URY CHL PRY PER HND BOL NIC .1 .1 .05 .1 .15 .2 .25 0 .02 .04 .06 .08 Data Data R−squared=.2279314032419751 R−squared=.4735910912144733 (a) Tax evasion (b) GDP per worker Figure 6. Tax evasion and GDP per worker - data vs the model predictions across Latin American countries 31 V.2 Alternative detection probability In the baseline calibration, we use the same variable to capture both institutional quality and the detection probability. As argued previously, the level of institutional quality may send an equivalent signal of the level of monitoring and auditing, i.e., the detection probability. How- ever, this assumption can raise questions as the latter may drive our findings. In this section, we check the sensitivity of the results using the perception of the control of corruption as an alternative measure of the detection probability. The latter is from the World Governance Indicators’ database and is defined as the perceptions of the extent to which public power is exercised for private gain, including both petty and grand forms of corruption, as well as "capture" of the state by elites and private interests. We use the percentile ranking of the index which is ranged between 0 and 1. The simulation procedure is the same as previously, and the model is calibrated to the United States economy. Figure 7 plots tax evasion and output per worker from the model against the data for African and Latin American countries. As previously, the model’s prediction is strongly correlated with the data. The model still explains around 20% of the variation in tax evasion and 49% of the dispersion of GDP per worker across African and Latin American countries. 32 .5 GIN .6 SWZ CHL .4 AGO MEX BWA TZA MRT ZAF UGA NIC .4 .3 URY EGY BWA PAN PAN Model Model BRA NAM ARG BRA CRI SWZ COL CRI ECU GTM .2 NAM MEX ECU SLV GTM PER BOL AGO MAR .2 SLV RWA PRY PRY ARG COL EGY URY HND KEN HND BOL NIC CHL .1 MRT PER ZAF MDG KEN MAR GIN UGA MDG TZA RWA 0 0 .1 .2 .3 .4 0 .02 .04 .06 .08 Data Data R−squared=.1977976014267405 R−squared=.4941246242786512 (a) Tax evasion (b) GDP per worker Figure 7. Tax evasion and GDP per worker - Data vs the model predictions - Alternative detection probability Counterfactual experiments. With the calibrated model, we first conduct three tax neutral counterfactual experiments on Guinea as in the quantitative analysis before exploring two additional counterfactual analyses. In the first round of experiments, we examine what happens in Guinea regarding tax evasion if the level of distortionary infrastructures and deterrence probability can be improved to the average level of the sample and by 50 percent respectively. These changes lead to a decline in sales not reported for tax purposes by 26.72 percent and 18.34 percent if the level of distortionary infrastructures and deterrence probability can be improved to the average level of the sample and by 50 percent respectively. Regarding the output per worker, it grows by 1.22- and 6.46-fold respectively. The third experiment combines both changes in the level of distortionary infrastructures and deterrence probability and leads to a 40.16 percent decline in sales not reported for tax purposes, while the output per worker grows by 6.62-fold. These counterfactual experiments provide the same conclusion as in the quantitative analysis, where an improvement in the level of distortionary 33 infrastructures and institutional quality leads to a decline in tax evasion between 18.34 and 40.16 percent. We also conduct two additional counterfactual experiments that are consistent with the previous findings. Firstly, using the current level of control of corruption as a proxy of the detection probability, we conduct a counterfactual experiment that examines what happens in Africa and Latin America if the losses stemming from the business environment are cut down by half. The deterrence probability is proxied by the index of the control of corruption, while the other parameters remain the same, except the level of distortionary infrastruc- tures. The latter is assumed to be cut down by half in each country. This counterfactual experiment shows that cutting down losses stemming from the poor business environment by half, governments could reduce firms’ tax evasion by 20.62% on average in African and Latin American countries. Firms’ tax evasion declines from 17.84% to 14.16%. Secondly, we use so far the difference in both the detection probability and the level of distor- tionary infrastructures to explain 35% of the variation of tax evasion across African and Latin American countries. Using such approach, we are unable to distinguish between the effects of deterrence policies and distortionary infrastructures on tax evasion, while the distinction can be highly relevant regarding policy implications. Alternatively, we conduct a counterfactual experiment consisting of setting the same level of detection probability for all the countries. The latter corresponds to the average level of the detection probability in the sample. We further examine how much the difference in losses stemming from the business environment can explain the variation of tax evasion across African and Latin American countries. The calibration and simulation procedures are the same as previously, except that the detection probability is the same for all the countries. Figure 8 plots tax evasion and output per worker from the model against the data for African and Latin America countries. As it can be seen, the model’s predicted value remains highly correlated with the data, and the model explains 14% of the variation of tax evasion across African and Latin America countries. Also, the model explains 50% of the variation of GDP per worker across countries. 34 .5 GIN .6 SWZ CHL .4 AGO MEX BWA MRT UGA TZA ZAF NIC .4 .3 URY EGY BWA PAN PAN Model Model BRA NAM ARG BRA CRI SWZ COL CRI ECU GTM .2 MEX NAM ECU SLV GTM MAR PER BOL PRY AGO .2 SLV RWA PRY ARG COL EGY URY HND KEN HND BOL NIC CHL .1 MRT PER ZAF MDG KEN MAR GIN MDG UGATZA RWA 0 0 .1 .15 .2 .25 0 .02 .04 .06 .08 Data Data R−squared=.1407139537033489 R−squared=.4995535949815552 (a) Tax evasion (b) GDP per worker Figure 8. Tax evasion and GDP per worker - data vs the model predictions using the same detection probability for all the countries V.3 Alternative calibration The analysis so far relied on the model that is calibrated to the United States’ economy. How- ever, the latter can raise criticism the parameters for the United States and those for African and Latina American economies might be different. As a sensitivity check, we recalibrate the model to the Chilean economy, the country with the lowest combination of distortions and tax evasion in our data and one of the largest sample size. The parameter value for the labor share in Chile is set to the average share of labor income in the GDP from the ILOSTAT.13 We set labor share to be 0.414 which is the average labor’s share in the GDP.14 As in the baseline calibration, we assume a decreasing return in the firm-level production function. The extent of the returns is obtained under the assumption 13 The ILOSTAT is the world leading data source on labor statistics from the International Labour Orga- nization (ILO). The data can be downloaded from the following website: http://www.ilo.org/ilostat. 14 The average labor’s share of the GDP is available for 2009, 2011, and 2013. 35 2 that the average share of labor income in the GDP corresponds to 3 of the returns to scale. The returns to scale obtained is 0.62, very close to the calibration in Gourio and Kashyap (2007) of the returns to scale for Chile.15 We set the interest rate to be 17.6 percent which is the average discount rate for African and Latin American economies from the IMF Inter- national Financial Statistics (IFS) dataset. Using this information, the parameter β is set to 0.850. We choose the depreciation rate of capital to be δ = 0.06 which is the calibrated value from Gourio and Kashyap (2007). Since there are no studies on the exit rate parameter in developing countries, we set the annual exit rate to the same value as previously, i.e., λ = 0.1. The other parameters of the model remain the same as previously. Table 6 below summarizes the parameters in the alternative calibration. Table 6. Alternative calibration Parameter Value Description Source τd [0; 1] Distortionary infrastructures WBES A [0; 1] Public policies effectiveness WGI p [0; 1] Detection probability WGI β 0.850 Real rate of return IMF International Financial Statistics (IFS) η 0.207 Capital income share ILOSTAT and Literature γ 0.414 Labor income share ILOSTAT δ 0.06 Investment to output ratio Gourio and Kashyap (2007) λ 0.1 Annual exit rate Literature θ 3 Evasion cost parameter Assumption ce 1 Entry costs Literature cf 0 Fixed costs Literature z [1; 3.98] Distribution of productivity Relative labor demand Notes. See Restuccia and Rogerson (2008) for parameter values from the literature. 15 Gourio and Kashyap (2007) calibrated the returns to scale to 0.6 in Chile. Our findings remain the same using this returns to scale and 1 2 3 and 3 as the share of capital and labor respectively. 36 Figure 9 plots tax evasion and output per worker from the model against the data for African and Latin American countries. Here, the output per worker is normalized by the output per worker of Chile. As previously, the data on GDP per worker are from the World Bank’s World Development Indicators. The reported value of the GDP per worker is nor- malized by the levels in Chile in both the model and the data. As previously, each circle represents one country and corresponds to a combination of the output per worker from both the model and the data. As it can be seen, the model’s predicted values are highly corre- lated with the data, with a regression coefficient of around 0.92 for tax evasion and 0.20 for GDP per worker. These coefficients are statistically significant at the 1 and 10 percent level respectively. Using Chile as a benchmark economy, the model explains 35% of the variation of tax evasion and around 10% of the dispersion in GDP per worker across African and Latin American countries. These findings are comparable to those obtained from the calibration to the United-States, where the model explains 35% of the variation of tax evasion and 49% of the variation of GDP per worker across African and Latin American countries. With the calibrated model on the Chile economy, we conduct three tax neutral counterfac- tual experiments using Guinea as in the quantitative analysis. Especially, we examine what happens in terms of tax evasion if both the level of distortionary infrastructures and policies can be improved to the average level of the sample and by 50 percent respectively. The approach consists of examining the effects of the change separately in experiments 1 and 2 before combining them in experiment 3. Experiments 1 and 2 lead respectively to a decline in sales not reported for tax purposes by 26.72 and 18.34 percent. In these experiments, the output per worker increases up to 13.20-fold respectively. The third experiment leads to a decrease in firms’ tax evasion by 40.16 percent, while the output per worker is nearly 5-fold as large as in the benchmark situation of the country. The counterfactual experiments are in line with the findings from the calibration to the U.S. economy. 37 GIN CHL 1 .6 SWZ MEX .8 AGO ZAF BWA TZA UGA MRT URY BWA PAN EGY NIC .4 .6 PAN NAM ARG BRA CRI Model Model SWZCOL BRA CRI GTM ECU ECU SLV .4 NAM GTM PER MEX MAR AGO BOL PRY .2 SLV RWA PRY ARG COL URY HND EGY HND BOL NIC KEN MRT CHL .2 PER ZAF MDG KEN MAR GIN UGA TZA MDG RWA 0 0 .1 .2 .3 .4 .5 0 .5 1 1.5 2 Data Data R−squared=.3505665524954171 R−squared=.1007588272129979 (a) Tax evasion (b) GDP per worker Figure 9. Tax evasion and GDP per worker - Data vs the model predictions across - Alter- native calibration 38 VI Conclusion Firms in developing countries face considerable constraints in their development due to dis- tortionary infrastructures and policies, and weak institutions. In parallel, tax evasion is widespread and pervasive in these countries, generating a substantial negative effect on do- mestic resource mobilization as well as on government’s ability to provide infrastructure and public goods. African and Latin American countries rank poorly in terms of ease of doing business, with around one-quarter of annual sales not reported for tax purposes. As a novelty, this paper examines empirically and theoretically the impacts of distortionary infrastructures and institutional quality on firms’ tax evasion in African and Latin American countries. The paper sheds light on two potential mechanisms driving firms’ tax evasion which have not been given a lot of attention in the literature. First, entrepreneurs, knowing that govern- ment effectiveness is weak, also anticipate low monitoring and have, therefore, more incentive to evade their taxes. In the second mechanism, distortions and poor institutional quality gen- erate losses for firms by creating a wedge between potential and realized profits. Tax evasion is a way for firms to counterbalance the negative effects of distortionary infrastructures and policies. This paper provides supporting evidence of the negative impact of distortionary infrastructures and policies on firms’ tax evasion. In particular, using WBES data, we show that losses stemming from the business environment are positively correlated with firms’ tax evasion. Using WGI data, we show that institutional quality is negatively related to the proportion of sales not reported for tax purposes. To understand the mechanisms through which distortionary infrastructures and policies affect firms’ tax evasion, we develop a general equilibrium model with heterogeneous firms. The model is calibrated using the United States as a benchmark economy. We simulate tax evasion and output data for firms in a sample of 30 African and Latin American countries. The simulated output per worker and tax evasion are strongly correlated with the data, as the model explains 35 percent of the variation in tax evasion and 49 percent of the dispersion in output per worker across African and Latin American countries. 39 Having established the ability of the model to replicate some of the variations of tax eva- sion across countries, we conduct counterfactual experiments to examine the implications of changes in distortionary infrastructures and policies. The counterfactual experiments show that African and Latin American countries could reduce tax evasion by about 21 percent by cutting down the losses stemming from the business environment by half. Finally, setting the same detection probability for all the countries, the last counterfactual experiment shows that distortionary infrastructures explain 14 percent of the variation of tax evasion across African and Latin American countries. The paper highlights that distortionary infrastructures and policies can generate a vicious circle of underdevelopment in which tax evasion and distortions interact positively. Firms’ tax evasion shrinks public revenues and this, in turn, reduces the government’s capacity to curtail tax evasion, invest in productive infrastructures, and thereby reduce distortions in the business environment. Hence, improving the quality of the business environment in Africa and Latin America would generate a double-dividend. On the one hand, it would boost firms’ productivity through better public infrastructure, better access to finance, and a lighter regulatory environment. On the other hand, improving the quality of the business environment could reduce firms’ tax evasion. African and Latin American countries could generate additional domestic revenues and boost the aggregate productivity simultaneously by pushing the agenda of reforms that help to improve the quality of infrastructures and policies related to the business environment. The impacts of distortionary infrastructures and policies examined in this paper might represent lower bounds for tax evasion at the country level as the standardized dataset employed does not include informal firms. Those firms are widespread in African and Latin American countries, are inherently hard to tax, and might represent an additional source of distortion and resource misallocation. 40 References Acemoglu, D., Johnson, S., and Robinson, J. (2001). The colonial origins of comparative development: An empirical investigation. American Economic Review, 91(5):1369–1401. Amaral, P. and Quintin, E. (2010). Limited enforcement, financial intermediation, and economic development: A quantitative assessment. International Economic Review, 51(3):785–811. Aterido, R., Hallward-Driemeier, M., and Pagés, C. (2011). Big constraints to small firms’ growth? business environment and employment growth across firms. Economic Develop- ment and Cultural Change, 59(3):609–647. Atkeson, A. and Kehoe, P. J. (2005). Modeling and measuring organization capital. Journal of Political Economy, 113(5):1026–1053. Bah, E.-H. and Fang, L. (2015). Impact of the business environment on output and produc- tivity in africa. Journal of Development Economics, 114(C):159–171. Besley, T. and Burgess, R. (2004). Can labor regulation hinder economic performance? evidence from india. The Quarterly Journal of Economics, 119(1):91–134. Besley, T. and Persson, T. (2014). Why do developing countries tax so little? Journal of Economic Perspectives, 28(4):99–120. Botero, J. C., Djankov, S., La-Porta, R., de Silanes, F. L., and Shleifer, A. (2004). The regulation of labor. Quarterly Journal of Economics, 119(4):1339–1382. Buera, F. J., Moll, B., and Shin, Y. (2013). Financial frictions and the persistence of history: A quantitative exploration. Journal of Political Economy, 121(2):221–272. D’Erasmo, P. and Boedo, H. (2012). Financial structure, informality and development. Journal of Monetary Economics, 59(3):286–302. 41 Easterly, W. and Rebelo, S. (1993). Fiscal policy and economic growth. Journal of Monetary Economics, 32(3):417–458. Eifert, B., Gelb, A., and Ramachandran, V. (2006). Business environment and comparative advantage in africa: Evidence from the investment climate data. In Annual World Bank Conference on Development Economics (ABCDE) 2006: Growth and Integration, pages 195–233. The World Bank. Gourio, F. and Kashyap, A. (2007). Investment spikes: New facts and a general equilibrium exploration. Journal of Monetary Economics, 54(1):1–22. Goyette, J. and Gallipoli, G. (2015). Distortions, efficiency and the size distribution of firms. Journal of Macroeconomics, 45:202–221. Hall, R. E. and Jones, C. I. (1999). Why do some countries produce so much more output per worker than others? The Quarterly Journal of Economics, 114(1):83–116. Hox, J. J., Moerbeek, M., and van de Schoot, R. (2010). Multilevel analysis: Techniques and Applications. Routledge. Hseih, C.-T. and Klenow, P. (2009). Misallocation and manufacturing tfp in china and india. The Quarterly Journal of Economics, 124(4):1403–1448. IMF (2003). World Economic Outlook, April 2003 : Growth and Institutions. IMF. López de Silanes, F., Djankov, S., La Porta, R., and Shleifer, A. (2002). The regulation of entry. Quarterly Journal of Economics, 117(1):1–37. Ordóñez, J. (2014). Tax collection, the informal sector, and productivity. Review of Economic Dynamics, 17(2):262–286. Pavcnik, N. (2002). Trade liberalization, exit, and productivity improvements: Evidence from chilean plants. The Review of Economic Studies, 69(1):245–276. 42 Prado, M. (2011). Government policy in the formal and informal sectors. European Economic Review, 55(8):1120–1136. Restuccia, D. and Rogerson, R. (2008). Policy distortions and aggregate productivity with heterogeneous establishments. Review of Economic Dynamics, 11(4):707–720. Restuccia, D. and Rogerson, R. (2013). Misallocation and productivity. Review of Economic Dynamics, 16(1):1–10. Rodrik, D., Subramanian, A., and Trebbi, F. (2004). Institutions rule: the primacy of institutions over geography and integration in economic development. Journal of Economic Growth, 9(2):131–165. Sleuwaegen, L. and Goedhuys, M. (2002). Growth of firms in developing countries, evidence from cote d’ivoire. Journal of development Economics, 68(1):117–135. Tybout, J. (2000). Manufacturing firms in developing countries: How well do they do, and why? Journal of Economic Literature, 38(1):11–44. World Bank (2005). World Development Report 2005 : A better Investment Climate for Everyone. Oxford University Press. Wu, G. L. (2018). Capital misallocation in china: Financial frictions or policy distortions? Journal of Development Economics, 130(1):203–223. 43 Appendices A Investment climate indicators The World Bank’s Doing Business index measures aspects of regulation that enable or pre- vent private sector businesses from starting, operating and growing. These regulations are measured using 11 indicator sets: opening a business, dealing with construction permits, getting electricity, registering property, getting credit, protecting minority investors, paying taxes, trading across borders, enforcing contracts, resolving insolvency and labor market reg- ulation. The annual report provides a ranking at the country level in term of the ease of doing business. In the last Doing Business Report in 2017, the average regional rankings of African and Latin American countries were 143 and 107,16 respectively, constraining firms’ creation and the long-term performance of incumbent firms. Moreover, comparing different indicators of investment climate variables from the World Bank Enterprise Surveys (WBES) for the high-income countries of the Organisation for Eco- nomic Co-operation and Development (OECD) and African and Latin American countries, it can be seen that firms in these countries face a poor investment climate compared to their counterparts in OECD countries (cf. Table A.1 below). For example, respectively, 22.4 per- cent and 19.3 percent of firms in African and Latin American countries identify corruption as a major or severe obstacle to business, while the corresponding number is 7.2 percent in the OECD countries. Similarly, the average percentile rank of policy effectiveness is 45 in African and Latin American countries against 85 in the OECD sample. The high-income OECD countries include Germany, Greece, Ireland, Portugal, The Republic of South Korea and Spain. 16 See the 2017 Doing Business Report: http://www.doingbusiness.org/~/media/WBG/DoingBusiness/ Documents/Annual-Reports/English/DB17-Report.pdf 44 Table A.1. African and Latin American countries vs OECD counterparts AFR LAC OECD Electricity as an obstacle to business (% of firms) 17 9.3 5.2 Costs of power outages (%) 7 4 2.3 Corruption as an obstacle to business (% of firms) 22.4 19.3 7.2 Costs of corruption (% of sales) 2.7 1.9 0.27 Crime as an obstacle to business (% of firms) 14 17 6 Costs of crime (% of sales) 1.5 1 0.42 Regulation (% of time spent) 8.60 13.9 1.69 Tax evasion (%) 28.29 22.51 6.80 Institutional Quality 39 48 85 Notes. An investment climate variable is considered as an obstacle to business if a firm states that the latter is a major or a very severe obstacle to business. 45 B List of countries Table B.1. Business environment in Africa and Latin American countries Country Year Sample size Tax evasion (%) Distortions (%) Institutional Quality Angola 2006 409 50.95 7.10 0.06 Argentina 2006 913 17.21 1.76 0.55 Bolivia 2006 544 20.34 3.72 0.29 Botswana 2006 309 47.80 3.14 0.70 Brazil 2003 1506 32.70 2.82 0.61 Chile 2004 894 2.97 2.20 0.88 2006 925 13.20 2.13 0.83 Colombia 2006 918 17.10 2.47 0.53 Costa Rica 2005 284 28.49 7.04 0.59 Ecuador 2003 356 20.47 8.71 0.20 2006 603 26.24 3.31 0.17 Egypt, Arab Rep. 2004 946 16.50 7.36 0.48 El Salvador 2003 409 23.21 3.93 0.43 2006 546 18.93 3.90 0.50 Guatemala 2003 437 22.59 6.49 0.39 2006 491 26.83 4.76 0.31 Guinea 2006 213 64.25 19.70 0.06 Honduras 2003 333 31.66 5.97 0.33 2006 355 15.85 4.69 0.30 Kenya 2003 218 13.32 11.94 0.31 Madagascar 2005 271 6.41 11.07 0.42 Mauritania 2006 225 47.13 6.22 0.26 Mexico 2006 1295 23.67 2.02 0.60 Morocco 2004 833 3.91 0.49 0.56 Namibia 2006 301 24.68 2.29 0.59 Nicaragua 2003 346 32.58 9.03 0.25 2006 415 40.85 10.51 0.21 Panama 2006 545 36.99 5.50 0.57 Paraguay 2006 452 19.34 6.02 0.20 Peru 2006 607 10.53 1.00 0.32 Rwanda 2006 209 19.00 8.89 0.45 South Africa 2003 560 9.16 1.43 0.74 Swaziland 2006 290 58.13 4.04 0.21 Tanzania 2003 229 30.06 9.95 0.41 2006 416 47.03 12.53 0.43 Uganda 2006 546 46.50 13.59 0.37 Uruguay 2006 341 14.85 0.83 0.66 46