Department of Economics Uppsala University Thesis, Economics C
Authors: Amanda Crabo and Alexander Källestål Tutor: Niklas Bengtsson
Spring Semester 2016
IS CORRUPTION
INHERENTLY BAD?
The effect of corruption on the Palma ratio:
A cross-regional study of Brazil’s federal states
Abstract: This thesis analyzes if there, given the size of the informal sector, is an effect of corruption on income inequality, here defined as the Palma ratio. Estimations are done with a fixed effects ordinary least squares regression using panel data for 19 federal states of Brazil over every other year between 2006-2014.
The results provide evidence that corruption increases income inequality when the informal sector is smaller than 37.97%, but decreases inequality when the informal sector exceeds 55.34%. The findings are robust to several sensitivity checks. The gained insight of the relationship between corruption and income inequality using a microeconomic perspective is the main academic contribution of this thesis.
Keywords: Corruption; Income inequality; Palma; Informal sector; Brazil
This thesis is the end product of three amazing years. It has been three years of painstakingly long days of studying, wonderful nights, and everlasting memories. The thesis has been achieved through hard work, but it could not have been done without some key ingredients. Firstly, we want to give a massive shout-out to our tutor Niklas Bengtsson. Thank you for giving us the confidence to stand our ground and to make this thesis something we are proud of. You have been the mama bird we needed; you kicked us out of the nest so we could spread our wings. Secondly, we want to thank all of our friends who have acted as personal therapists when we needed support or a creative boost. Thank you for allowing us to make a quick comment about the thesis in the middle of the preparty, and for forcing us stop two minutes later. Lastly, but most importantly, we would like to thank each other. For respecting each other’s weaknesses and amplifying each other’s strengths, for supporting the other when times were rough, motivating each other, and for still being there for each other – stronger than ever.
Puss,
Crabo and Källe
Figure 1. The federal states of Brazil. The states used in this thesis are marked according to Table 2, Chapter 3.2. Source:
Wikimedia (2005). Modified by authors.
TABLE OF CONTENTS
1. Introduction ... 5
2. Theoretical framework ... 7
2.1. Corruption ... 8
2.2. Inequality ... 9
2.3. The informal sector ... 10
2.4. Literature review ... 11
3. Empirical framework ... 14
3.1. Empirical specification ... 14
3.2. Data ... 15
4. Results ... 18
5. Discussion ... 21
6. Conclusion ... 24
7. References ... 26
Appendix 1: The Corruption variable ... 29
Appendix 2: Variable definitions ... 31
Appendix 3: Correlation matrix ... 32
Appendix 4: Distribution of Informality ... 33
Appendix 5: Robustness tests ... 34
1. INTRODUCTION
“So distribution should undo excess, and each man have enough.”
–William Shakespeare, King Lear, act 4, scene 1
Is corruption inherently bad? Multiple researchers conclude that its nature does in fact have corrosive effects on everything from the smallest microeconomic perspective to a global macro perspective (Shleifer and Vishny, 1993; Mauro, 1995; Gupta et al., 2002; Svensson, 2005; Gyimah-Brempong and Munoz de Camacho, 2006). However, no consensus has yet been met and opposition to the notion remains. Since the benchmarking article Economic development through bureaucratic corruption by Nathaniel Leff (1964), there has been multiple evidence that corruption, albeit inherently bad, under certain circumstances has positive economic effects.
Leff (1964) states that corruption can act as a lubricant for inefficient governmental processes, speeding up, and in some cases even creating, growth. As Ravallion and Chen (1997) argue, an increase in economic growth leads to a reduction in overall poverty, thus affecting the income distribution1. If corruption affects an economy’s growth, does that in turn affect income inequality? Even more interesting, does corruption have a direct effect on income inequality? As a development of previous studies on growth, this relationship has started to gain ground in the field of economics.
Early research on the matter shows how corruption increases inequality both directly and indirectly (Gupta et al., 2002). Interestingly, the effect seems to be deviant in developing countries (Chong and Calderón, 2000), especially in Latin America, compared to the rest of the world (Li et al., 2000; Gyimah-Brempong and Munoz de Camacho, 2006). In recent years, empirical findings have shown how the presence of a large informal sector2 transforms the effect of corruption on inequality (Andres and Ramlogan-Dobson, 2011;
Dobson and Ramlogan-Dobson, 2010; 2012a; 2012b). The singularity of Latin America is agreeably a valid argument for further studying of the relationship between corruption and income inequality on the continent. However, the continent experiences big intra-regional variances in terms of macro- and microeconomic indicators. To exemplify, the countries ranked 21nd and 158th on the Corruption Perceptions Index 2015 are both found in Latin America (Transparency International, 2015).
1 Income distribution is expressed interchangeably with income inequality and inequality throughout the thesis, all referring to the distribution of income in the formal economy unless explicitly stated otherwise.
2 The informal sector is throughout the thesis also expressed as the informal economy or the unofficial economy.
A country, which in itself encapsulates large regional disparities, is Brazil. The vast country has an area of twice the size of the European Union3 and its economy is the 8th largest on the globe in terms of gross domestic product (GDP) (CIA, 2016). Even though progress has been made, Brazil still had a Gini coefficient of 52.9 in 2013 (The World Bank, 2016). The Palma ratio of Brazil is 3.8, meaning that the top 10 percent of the income distribution has 2.8 times greater income than the bottom 40 percent. To put this in perspective, Sweden and the US have Palma ratios of 0.9 and 2.0 respectively (UNDP, 2016)4. Furthermore, political turmoil linked to corruption is highly present in Brazil. The state-owned oil company Petrobras has been exposed as the key player in a long-going multi-billion dollar corruption scandal. The price of construction contracts has been illicitly boosted with excessive revenue, which then has been funneled into the pockets of high-ranked officials (The International Institute for Strategic Studies, 2015). Considering the sheer financial and geographical size of the country (CIA, 2016), the ripple effect of such events can be expected to be global.
It is with this backdrop the thesis presents itself. The aim of this thesis is to bring a piece to the puzzle that explains the relationship between corruption and inequality. The economic and political setting of Brazil renders it as a qualified choice in where the relationship between the two can be studied. In line with recent research on the uniqueness of Latin America, the thesis also considers the informal economy. Therefore, this thesis’ research question is:
Given the size of the informal sector, is there an effect of corruption on income inequality in Brazil?
As a macroeconomic perspective has been traditionally preferred, a micro perspective was deemed as more rewarding for this thesis. Additionally, by performing a national cross-regional study, a closer proximity to individuals’ experiences can possibly be attained, which is perceived by the authors as more tangible and relevant for policy reasons.
It is to be highlighted that inequality throughout the thesis refers to income inequality in the formal sector, since the inequality data solely describes the official economy. Therefore, any estimated effect of corruption needs to take place in the formal economy to be visual in the results.
No explicit model is being used for the thesis, as no model seemed to capture the intended purpose of the thesis. Instead, the approach of the thesis is to perform a regression based on results from previous empirical
3 Calculations are based on European Union data (2016) and CIA (2016).
4 Data refers to the most recent year available during 2005-2013.
studies. Given that framework, a time and entity fixed effects ordinary least squares regression is run using income inequality, defined as the state’s Palma ratio, as regressand, and corruption as the main regressor. The configuration of the two variables is motivated by the thesis’ aspired microeconomic aspect.
The empirical results show that corruption with 95% confidence has an effect on income inequality, given most levels of informality. When the informal sector employs less than 38% of the total workforce, increased corruption leads to increased inequality between the top 10 and bottom 40 groups of the income distribution. However, if the informal sector exceeds 55% of the total workforce corruption instead favors the poor relative the rich, and increased corruption leads to decreased inequality. Based on the point estimates, the turning point is around 45% informality. Hence, corruption is not affecting income inequality in a consistent manner.
The findings are in line with multiple studies and can be considered a contribution to existing literature stemming from Leff (1964). Our estimates are mainly in line with the research series from Dobson, Ramlogan-Dobson (2010; 2012a; 2012b) and Andres and Ramlogan-Dobson (2011), but the thesis contributes to the academic field in two principal ways: through the usage of microeconomic variables rather than macroeconomic, and through the focus on Brazil. Even when doing so, the results indicate similar patterns as previous literature.
The thesis will continue in the following manner: 2) a theoretical framework with key concept definitions and a literature review elaborating on the relationship between corruption and inequality, 3) the empirical framework consisting of a methodological walkthrough of the used regression and the data, 4) the results, and lastly, 5/6) Discussion and Conclusion.
2. THEORETICAL FRAMEWORK
Initially, it is worth mentioning that this thesis was not based on a particular theoretical model due to the fact that no model sufficiently relevant to both the research question and the essence of the thesis was found5. Therefore, it was concluded that the thesis would rather use empirical findings on the relationship
5 One model closely related to this thesis is developed by Okumu and Forgues-Puccio (2014). The theoretical model attempts to identify how corruption mitigates income inequality in the presence of a large informal sector. However, this model focuses on how credit constraints, arisen due to corruption, leads to households choosing informal entrepreneurship over formal employment. Hence, this model has a much narrower perspective of the effect of corruption on inequality than this thesis has. Furthermore, the model only attempts to explain the mechanisms behind the empirical findings on how corruption under certain conditions decreases income inequality. The mechanism
between corruption, income inequality, and informality from Latin America as base for the empirical specification.
2.1. CORRUPTION
How can one determine whether or not an act is corrupt? The definition of a corrupt act generally lies in the eye of the beholder. Therefore, to date, no universal answer has been given to the rhetorical question.
However, broad definitions have been made. A common definition of corruption is “the misuse of public office for private gain” (Svensson, 2005: p. 20), where the main focus of the term is the usage of governmental services and advantages as personal leverage. Within this definition follows two distinct components of the term: corruption can be associated with illicit activities, since bribes are used as methods to avoid repercussions, but also be associated with poor governance as corrupt acts are used to facilitate governmental processes (Svensson, 2005).
From an economic perspective, corruption is frequently viewed as a principal-agent problem between the government apparatus and a government official. As popularized by Shleifer and Vishny (1993), corruption can be viewed as an act conducted with or without theft. Corruption with theft implies that an official, when dealing with e.g. a permit, embezzles potential profit for the government. Corruption without theft does not affect the government’s revenue of a service, but works as an additional tax on the service claimed entirely by the bribe taker.
Generally, there are two schools of measuring corruption: the direct method stemming from household survey data, and the indirect method using macroeconomic indicators. According to the United Nations Office on Drugs and Crime (UNODC) (2009), the most policy relevant method is the direct method. By using survey data, specifically focusing on individuals’ experiences with corruption rather than their perceptions of the phenomenon, a higher level of tangibility can be achieved. Additionally, this facilitates the quantification of an otherwise highly subjective variable.
The used definition of corruption is purposely constructed for this thesis, namely the share of a state’s population who has been asked to pay a bribe when in contact with any of a number of institutions6. This enables a satisfactory measurement on ratio scale. In line with UNODC’s (2009) recommendations, the
behind the inconsistent nature of the effect, as found by Dobson and Ramlogan-Dobson (2010; 2012a; 2012b) and Andres and Ramlogan-Dobson (2011), is not mentioned.
6 See section 3.2. Data and Appendix 1 for further specification of the Corruption variable.
experienced corruption of respondents is the main focal point. Lastly, the definition also adheres to the general aspirations of the thesis. By focusing on individuals’ experiences rather than macroeconomic indicators corruption can be accessed at the lowest microeconomic scale possible.
2.2. INEQUALITY
The term inequality refers to something being unequally distributed among a group; income inequality is subsequently loosely defined as “an indicator of how material resources are distributed” (OECD, 2014: p.
110). Even though it might appear relatively easy to grasp, it is relatively difficult to define as an economic concept. Additionally, there are academic disparities regarding the appropriate measurement. The used financial variable is typically consumption or income (OECD, 2014), where consumption tends to be somewhat more flexible over time whereas income appears to be more rigid (Haughton and Khandker, 2009).
Multiple methods of measuring income inequality have been developed over the years. The most commonly used measurements are the Gini coefficient, Atkinson’s inequality measurement, and decile dispersion ratios.
The Gini coefficient is calculated using the Lorenz curve and is based on the difference between the cumulative income distribution and perfect equality. The Lorenz curve displays cumulative income earnings of the population and illustrates “perfect equality” when e.g. 10% of the population earns 10% of the total income. By using the distance to perfect equality, a Gini coefficient between 0 and 1 is obtained (Haughton and Khandker, 2009). The Atkinson’s inequality measurement is focused on sensitivity differences between different income groups and therefore introduces a weighting parameter. This allows taking social welfare systems into account, focusing on the redistributive effect on income of for instance governmental policies (Atkinson, 1970). As for decile dispersion ratios, the basic idea is to divide the income of two different income distribution groups by each other in order to receive an income earnings ratio; how many times more does the top share earn relative to the bottom? This is used at times when intuitive understanding is key, as decile dispersion ratios generally are easier to understand relative to the other measurements (Haughton and Khandker, 2009). One of the widely used decile dispersion ratios is the Pareto ratio, which divides the income of the top 20 percent with income of the bottom 80 percent (Pareto, 1909).
Recently, a decile dispersion ratio has gained momentum, namely the Palma ratio. This measurement focuses entirely on the top 10 relative to the bottom 40 percent, thusly excluding the middle 50 percent of the income distribution (Palma, 2011). Cobham and Sumner (2013) identified that this middle group globally has
been fairly economically stable over time, while the distributional changes have been taking place at the top and bottom groups. In return, this has implications on the conventional Gini, as it is more sensitive to changes in the middle percentages rather than at the extremes.
In this thesis, the Palma ratio for each individual state will be used as the measurement of inequality for several reasons. Firstly, it is intuitively easier to understand than other traditional measurements of inequality (Palma, 2011). Secondly, the middle 50 percent of the Brazilian income distribution have been proved to be rather constant over time (Cobham and Sumner, 2013), meaning that the Gini coefficient would underestimate inequality. Thirdly, the Palma ratio will contribute to existing research by solely focusing on the top-and-bottom income inequality, in contrast to those using Gini. Lastly, the measurement adds an intriguing element since it reflects the current discussions on increasing inequality between the top and the bottom deciles (Piketty, 2014). The robustness of the Palma has been empirically confirmed (Cobham et al., 2015) and the explanatory power of the Palma is statistically significantly equal to that of the Gini (Cobham and Sumner, 2013).
2.3. THE INFORMAL SECTOR
The concept of informality can be hard to grasp from a statistical and econometric standpoint, due to the diffuseness inherent to the concept. How can one officially measure what by definition is informal?
Although the concept can be difficult to measure, the definition is somewhat easier7. Drawing from Portes and Castells (1991), the informal sector is defined as a part of a country’s economy that is “unregulated by the institutions of society, in a legal and social environment in which similar activities are regulated” (Portes and Castells, 1991: p. 12).
The underlying explanation to the phenomenon is somewhat unclear. As suggested by Loayza (1996), the emergence of the informal sector stems from weak governmental governance. It occurs if an economy is excessively regulated and taxed, and the government apparatus is not sufficiently competent to enforce the system. This generates a lack of public incentive from actors in the economy to comply with the system, thus exiting the formal economy.
The lack of governmental governance does not fully in itself explain the appearance of an unofficial economy. As suggested by de Soto (1989) and backed by Loayza (1996), the existence of an informal economy
7 (for example De Soto, 1989; Portes and Castells, 1991; Loayza, 1996; Dell’Anno, 2003)
essentially comes down to the cost for the actors in the economy. Firstly, it is dependent on the cost of accessing and remaining in the formal sector. A heavily regulated economy will have costs associated to it, both in terms of time and assets. Strenuous financial pressure on the actors in the formal sector makes the choice of staying formal a rational standpoint regarding the trade-off between costs and benefits. Secondly, costs do not appear to be limited to the formal sector. As identified by de Soto (1989), there is an unproportionately high unofficial financial pressure on actors in the informal economy. This causes actors to scale down business, leading to inefficiency.
The general cost associated with the informal sector makes it appear as if informality in an economy is inherent only to the bottom groups of the income distribution. However, Portes and Castells (1991) claim that informality is present at all hierarchical stages of the economy. Although multiple socio-economic working groups do fall under the broad general definition, the presence of the lowest income deciles is over- represented in the informal economy (Dobson and Ramlogan-Dobson, 2012b).
For the thesis, an econometric approach has been opted for when defining informality in a state’s economy.
Using one of the three definitions from the Instituto de Pesquisa Econômica Aplicada (IPEA) (eng. Institute of Applied Economic Research), the degree of informality is defined as the ratio of unprotected workers relative to the working population. The choice of a labor approach rather than the conventional monetary approach is based on the ambition to have a microeconomic perspective, and increases the understandability of the results on an economic level closer to the individual.
2.4. LITERATURE REVIEW
The literature on corruption and its effects on inequality is inconclusive, although most research is unanimous in that corruption deteriorates economic efficiency8. However, some researchers argue that corruption is solely corrosive while others have reached different conclusions when also considering other determining factors such as initial political and economic functionality.
The literature has discussed how corruption increases inequality in both direct and indirect ways. Beginning with the direct benefits, they are primarily captured by rich, well-connected, and well-informed individuals.
In a corrupt society, initial wealth also enables bribing policymakers for personal gain. Therefore, the rich are relatively better off in a corrupt society. In contrast, the unstable and uncertain rules of corruption add a risk
8 (for example Mauro, 1995; Gupta et al., 2002; Foellmi and Oechslin, 2007)
premium to any investment decision taken by poor and less-informed individuals, decreasing their relative return to investment (Gupta et al., 2002).
The effect of corruption on inequality also works through several indirect channels. Firstly, studies have shown that corruption decreases growth, mainly through dropped aggregate investment caused by capital flight, capital consumption (Gyimah-Brempong and Munoz de Camacho, 2006), and lowered investment rates (Mauro, 1995). The effect is further intensified through increased transaction costs due to uncertainty (Shleifer and Vishny, 1993) as well as the misallocation of talent and entrepreneurship (Svensson, 2005). As changes in growth have an inherently greater impact on the poor relative to the rich, reduced growth leads to increased inequality (Gyimah-Brempong and Munoz de Camacho, 2006).
Secondly, corruption is known to affect government spending through raised operating costs and weakened tax administration. The effects on the tax system tend to disproportionately favor the rich, thus creating a regressive effective tax system. Further, the social expenditure might be skewed as corruption can misallocate government resources from the poor to the rich. This can manifest itself in e.g. reduced funds for poverty reduction, healthcare, and education, as well as socially inefficient investments (Andres and Ramlogan- Dobson, 2011; Gupta et al., 2002).
Lastly, corruption creates imperfect credit markets, which in turn increase inequalities in the distribution of assets. The interconnection lies in that wealth operates as collateral under imperfect credit. Corruption increases the costs of borrowing, which leads to lowered aggregate demand, hence lower capital costs for the remaining debtors. This favors the already wealthy while those depending on external capital for financing are hit relatively worse. The effect is strengthened if the crowding-out effect skews competition and efficiency in the product market. Moreover, higher concentration of asset ownership, in turn, increases income inequality and creates lobbying opportunities for the rich (Foellmi and Oechslin, 2007; Gupta et al., 2002).
A second approach to corruption was mentioned in 1964 by Nathaniel Leff. He stated that corruption per se says nothing about social and economical development. Rather, he argued that corruption only indicates the existence of an extra-legal institution where groups or individuals can gain influence. Therefore, it is the interests and actions of the original policy makers and those getting political access in the corrupt system that determine the effect of corruption. In underdeveloped countries the formal political system is seldom greater at promoting development than the effective – corrupt – political system. In these cases, corruption instead
operates as a safeguard against bad economic policies. By increasing investment and competition when bureaucracy is inefficient and formal institutions are unstable, corruption can work as social collateral. This point of view has become known as the “greasing-the-wheels-hypothesis”, which states that corruption in a second-best world is actually beneficial for society as a whole. Empirical evidence supporting the hypothesis has since then been found. In a study on 69 developing and developed countries, Meón and Weill (2010) showed how corruption is less destructive to efficiency where governance is low. The study also found indications that corruption could be positively related to efficiency in countries where institutions are very ineffective.
In 2000, Li et al. estimated an inverted U-shaped effect by using an indirect corruption index and inequality expressed as Gini. They found that countries with high and low corruption levels experienced low inequality, while countries with mid-level corruption were more unequal. They also found that the effect of corruption was bigger in Latin America relative to the rest of the world, a finding later supported by Gyimah-Brempong and Munoz de Camacho (2006). Low levels of corruption had a greater impact on inequality in Latin America relative to other continents, and the marginal effect decreased faster as corruption increased (Li et al., 2000).
Chong and Calderón (2000) found a quadratic relationship between income distribution and institutional quality, a variable partially built on corruption in governments. In rich countries the two were negatively linked, but in poor countries the relationship was inverse. In other words, increased institutional quality was shown to increase inequality in poor countries. They suggested the existence of large informal sectors in less- developed countries as a possible explanation of the results. Institutional reforms entail high initial transaction costs, which will hit the informal workers hardest due to their need to learn new mechanisms of survival.
Building on this research, Dobson and Ramlogan-Dobson (2010) studied the corruption-inequality relationship in countries with widespread informal sectors. They chose Latin American countries partly since these countries had started working actively against corruption, but also since the informal sector produced around 25-30% of aggregate output. The study estimated that there was a trade-off between corruption and inequality by using an index of corruption and Gini. The study further proposed that institutional reforms might lead to more capital intense government operations and thereby raised unemployment. In 2011,
Andres and Ramlogan-Dobson published a very similar, but more ample study, using the same data9. In a further study by Dobson and Ramlogan-Dobson (2012a), the size of the informal sector, in terms of percent of official GDP, was included. They estimated that the marginal effect of corruption on inequality was a negative linear function of informality, decreasing the effect as the informal sector grows. The positive effect became statistically insignificant when informality exceed approximately 20%, corresponding to roughly 80% of the sample.
The same year, Dobson and Ramlogan-Dobson (2012b) published a continued elaboration on their previous research and found that the effect of corruption on inequality was insignificant for all developing countries in their sample. They argued that in countries with weak institutions, corruption is normality. This stems from two main things. Firstly, the absence of immorality linked to corruption, and secondly that corruption, in reference to Leff (1964), sometimes is the only way for formal businesses to remain efficient. Further, the study argued that anti-corruption policies imposes increased operating costs for individuals and firms associated with the informal sector, through new procedures, improved tax collection, and the erosion of established corrupt networks. Therefore, the informal sector shrinks. As it often supplies the poorest of society with an income, jobs and incomes are lost amongst the poor. Decreased corruption hence reduces the wealth of the poor relative to the rich in countries with large informal sector, and inequality increases (Dobson and Ramlogan-Dobson, 2012b).
3. EMPIRICAL FRAMEWORK 3.1. EMPIRICAL SPECIFICATION
Estimations are conducted with a fixed effects ordinary least squares (OLS) regression, using panel data for every other year between 2006-2014 (T = 5) and 19 of 27 states in Brazil (n = 19)10. The data is balanced due to missing values consistently being estimated as the previous t’s value11. OLS estimates coefficients by minimizing the sum of the squared prediction mistakes, and renders it possible to estimate the effect of Xk
on Y while keeping other regressors constant (Stock and Watson, 2011). To control for bias caused by omitted constant differences across states or national changes over time, entity and time fixed effects are
9 Dobson and Ramlogan-Dobson (2010) used four-year panel data from 1984-2003 for 19 Latin American countries, provided by ICRG, while Andres and Ramlogan-Dobson (2011) had the same data from 1982-2002.
10 A sensitivity analysis on the number of states used is described in section 5. Discussion.
11 This is applicable to the Palma ratio, Informality, Urbanization, Schooling, GDP/capita, and GDP as observations for 2010 were missing for all IPEA and IBGE based sources. It is also applicable to 2010’s corruption values for Alagoas, Amazonas, Espirito Santo, Maranhão, Mato Grosso du Sul, Paraiba, Rio Grande du Norte, and Sergipe.
used. Standard errors are clustered by states and heteroskedasticity is allowed for. The following multiple regression model is used:
𝑃
!"= 𝛽
!𝐶
!"+ 𝛽
!𝐼
!"+ 𝛽
!𝐶
!"× 𝐼
!"+ 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠
!"+ 𝛾
!+ 𝛿
!+ 𝑢
!"(1)
In equation (1), the dependent variable 𝑃!" is the Palma ratio in state i, year t. 𝐶!" represents corruption, i.e.
the share of the population in state i who during the last year, t, has been asked to pay a bribe. A higher value indicates more corruption. 𝐼!" measures the degree of informality in state i and year t. The marginal effect of corruption on income inequality is then:
!!!"
!!!"
= 𝛽
!+ 𝛽
!× 𝐼
!"(2)
If the effect of corruption is conditional on the informal sector, as suggested in the research question, 𝛽! will be significantly different from zero. If there is an effect of corruption on income inequality, given the size of the informal sector, the total of 𝛽!+ 𝛽! × 𝐼!" will be significantly different from zero.
3.2. DATA
12The data for the Palma ratio is provided by the government agency Instituto Brasileiro de Geografia e Estatistíca (IBGE) (eng. Brazilian Institute of Geography and Statistics) and contains all individuals 10 years or older working in the formal sector13. Income inequality is, as previously mentioned, more rigid than consumption inequality (Haughton and Khandker, 2009), but considerably more volatile than endowment inequality.
The Corruption variable measures the share of state i’s population that has paid or been asked to pay a bribe to one or several institutions during the last 12 months14. The data is taken from five data sets created bi- annually between 2006-2014 by the Latin American Public Opinion Project (LAPOP). By using time fixed effects possible changes between the rounds of surveys are controlled for. By looking at experienced rather than perceived corruption it is assumed that respondents give more objective answers, which leads to less errors-in-variable bias. Furthermore, since the measurement contains information specifically about the last year, it is possible to see trends over time as well as allow for some time lag in the effect of corruption on
12 Variable definitions and sources can be found in Appendix 2.
13 Working in the formal sector refers to receiving income from all sources with an income statement, excluding retirees, household workers, and relatives to household workers.
14 Included institutions are the police, a public official, a judge, at the city hall, at work, at a public hospital or health center, or an educational institution.
income inequality15. For a more elaborated description of how the variable is created and modified, see Appendix 1.
The control variables are chosen a priori with several things in mind. Variables acting as main channels through which corruption leads to inequality are not controlled for since it might skew the estimates to keep them constant. Examples are GDP growth, social spending, domestic credit to the private sector, and the distribution of land16.
Although informality is considered one of the main channels, the choice of allowing the effect of corruption on inequality to interact with this variable instead of any other is based on the its academic relevance in relation to Latin America17. The Informality variable describes the number of people employed in the informal sector in relation to the number of people employed in either the informal or the formal sector, in state i, in year t.
Moreover, highly corrupt countries tend to have lower urbanization as corruption has been observed to thrive in a rural economy, rather than in an urban (Li et al., 2000). Urbanization is also shown to be a determinant of inequality (Kuznets, 1955; Kanbur and Zhuang, 2013) and is therefore included to avoid omitted variable bias. Further, we control for GDP per capita, mean years of schooling, agricultural output in relation to GDP, and trade openness, which are all standard variables in the literature on the relation between corruption and inequality18. To ensure quality data, the data for all control variables is retrieved from Brazilian governmental institutions such as IBGE, IPEA, and Ministério do Desenvolvimento, Indústria e Comércio Exterior (MDICE) (eng. Ministry of Development, Industry, and Foreign Trade).
Table 1 shows descriptive statistics for all used variables and GDP in millions of Brazilian real (BRL), as several variables are built on GDP. The range of values for Corruption is quite large, spanning from 0% to 42%. The data is modified by only removing extreme values in the upper range. This is done because it was perceived that the exclusion of the zero-observations could increase the risk for bias more than the inclusion,
15 Due to the relatively large volatility of income inequality, the effect of corruption is assumed to be rather instant.
Hence, the choice of not time lagging corruption further is motivated by the assumption that the inherent time lag in the corruption variable is sufficient.
16 (Shleifer and Vishny; 1993, Mauro; 1995, Gupta et al., 2002; Gyimah-Brempong and Munoz de Camacho, 2006;
Svensson, 2005; Foellmi and Oechslin, 2007; Andres and Ramlogan-Dobson, 2011)
17 See Literature review in section 2.4.
18 (for example Gupta et al., 2002; Ferreira de Mendonça and Martins Esteves, 2014; Dobson and Ramlogan-Dobson, 2010; 2012a; 2012b; Andres and Ramlogan-Dobson, 2011)
contrary to the opinion regarding the high extreme values. In fact, the range of values is large for all variables, which is reasonable in such a vast and diverse country as Brazil (CIA, 2016). A correlation matrix for all variables used can be found in Appendix 3.
Table 2 provides descriptive statistics for selected variables for the 19 federal states used in the estimates.
Statistics refers to state averages between 2006-2014.
Table 2. Descriptive statistics of states - means
State Population
(1000’s)
Palma Palma
(Std. Dev)
Corruption Informality
1 Alagoas 3 220.6 3.76 0.530 6.83 61.44
2 Amazonas 3 589 2.78 0.073 15.86 55.75
3 Bahia 14 700 3.56 0.112 13.14 67.16
4 Ceará 8 540.8 3.54 0.264 17.08 67.49
5 Distrito Federal 2 599.8 5.48 0.360 19.33 40.72
6 Espírito Santo 3 686.2 3.08 0.120 3.44 50.45
7 Goiás 6 129.4 2.94 0.166 19.71 53.09
8 Maranhão 6 577 4.14 0.508 10.85 75.83
9 Mato Grosso do Sul 2 479.4 3.34 0.160 12.17 50.47
10 Minas Gerais 20 100 3.04 0.117 20.01 48.70
11 Paraná 10 700 3.02 0.191 23.90 44.02
12 Paraiba 3 810.6 3.98 0.302 7.06 69.67
13 Pará 7 608 2.88 0.080 24.05 67.27
14 Rio Grande do Norte 3 254.2 3.60 0.192 14.90 62.08
15 Rio Grande do Sul 11 000 2.98 0.120 14.16 47.01
16 Rio de Janeiro 16 000 3.40 0.110 19.99 44.82
17 Santa Catarina 6 338.4 2.56 0.133 12.62 39.10
18 Sergipe 2 113 3.44 0.163 6.20 63.37
19 São Paulo 42 400 3.02 0.128 15.20 37.44
Table 1. Descriptive statistics
Obs Mean Std. Dev. Min Max Median
Palma 95 3.40 0.80 2.20 6.30 3.20
Corruption 95 14.55 9.71 0.00 42.11 13.33
Informality 95 55.05 11.73 33.85 81.66 53.80
GDP/capita (K. BRL) 95 17.17 11.39 4.52 62.52 13.66
Urbanization 95 81.40 9.38 58.90 97.30 82.50
Schooling 95 6.94 1.19 4.68 10.08 6.87
Trade openness 95 27.77 17.57 2.17 71.15 27.55
Agricultural output 95 11.27 10.21 0.40 46.97 9.09
GDP (M. BRL) 95 176 684 266 529.7 15 124.27 1 408 904 90 131.72
Lastly, the finished dataset was studied in order to fulfill the assumptions for fixed effects regression on panel data (Stock and Watson, 2011). Therefore, the regression was plotted to ensure that no outliers or influential data affected the estimates. Further, to test for normality of the residual, a Kernel density test was conducted on the regression’s residual with good results; this motivates the choice of fixed effects over e.g.
bootstrapping.
4. RESULTS
“No doubt, [the] analysis will be misunderstood. So be it.”
– Nathaniel H. Leff (1964: p. 1)
The estimates of the OLS regressions are presented in Table 3. Model 1 shows a regression of Palma on corruption with only entity fixed effects, while Model 2 has both entity and time fixed effects. Model 3 shows the effect of corruption on Palma including control variables, but without taking the informal sector or the interaction between the informal sector and corruption into account. These are instead considered in Model 4, which is estimated with equation (1). In terms of the adjusted coefficient of determination, this first rises notably between Model 1 and 2. It then increases further when adding the informality variable and the informality-corruption interaction term. As for statistical significance, only estimates in Model 4 seems to be significant, further indicating the importance of the degree of informality.
The estimated coefficients of Urbanization, Trade openness, and Agricultural output are to be interpreted as the average marginal effect on the Palma ratio given a one-percentage point increase for each respective variable, holding all other constant. The coefficients of GDP/capita and Schooling tell the aforementioned effect expressed in thousands of BRL and a one-year change respectively. Lastly, the coefficients of Corruption and Informality are stated as the effect given a one-percentage point increase, but cannot solely be interpreted as partial derivatives due to the interaction term.
The estimated coefficient of the Interaction term 𝛽! is -0.00085. Since the estimate is significantly different from zero we provide evidence supporting that the effect of corruption on inequality is conditional on the degree of informality. Since the estimate is negative, the marginal effect of corruption on inequality will decrease as the degree of informality increases.
Table 3. Regression analysis
Palma Model 1 Model 2 Model 3 Model 4
Corruption 0.00242
(0.00641)
-0.00418 (0.00515)
-0.00343 (0.00513)
0.03859*
(0.01393)
GDP/capita 0.02143
(0.02819)
0.01799 (0.02080)
Urbanization 0.01244
(0.02838)
0.02795 (0.01707)
Schooling -0.46391
(0.47368)
-0.37093 (0.43972)
Trade openness 0.00550
(0.01214)
0.00412 (0.00884)
Agricultural output 0.01905
(0.01837)
0.01949 (0.01175)
Informality 0.15413***
(0.03713)
Interaction term 0.00085*
(0.00031)
Constant 3.36165****
(0.09333)
3.9050***
(0.14319)
7.50575*
(3.14415)
-5.55695 (3.57166)
State Fixed Effects Yes Yes Yes Yes
Year Fixed Effects No Yes Yes Yes
n 19 19 19 19
R-sqr 0.002 0.489 0.529 0.710
adj. R-sqr -0.009 0.460 0.473 0.668
2-tailed significance level, *p<0.05, **p<0.01, ***p<0.001. Regressions are estimated using OLS. Robust standard errors clustered by state in parentheses.
Using Table 3 and equation (2), the marginal effect of corruption on income inequality is expressed in equation (3). 𝐼!" is stated in a scale from 0-100.
!!!"
!!!"
= 𝛽
!+ 𝛽
!× 𝐼
!"= 0.03859 − 0.00085 × 𝐼
!"(3)
It is however, not possible to see whether the effect of corruption on inequality is statistically significant based on Table 3 alone. The only information provided above is that the marginal effect is statistically significant when informality is non-existent. Therefore the estimated marginal effect for all theoretically possible degrees of informality, as well as their 95% confidence interval, is calculated. The latter is defined as:
𝑏
!+ 𝑏
!× 𝐼 ± 1,96 𝑣𝑎𝑟 𝑏
!+ 𝑣𝑎𝑟 𝑏
!× 𝐼
!+ 2𝑐𝑜𝑣 𝑏
!𝑏
!× 𝐼
!,!(4)𝑏! and 𝑏! are point estimates of 𝛽! and 𝛽!. 𝑣𝑎𝑟 𝑏! , 𝑣𝑎𝑟 𝑏! and 𝑐𝑜𝑣 𝑏!𝑏! are estimated variances and covariance of corruption and informality. The outcome is visualized in Figure 2.
Figure 2. The marginal effect of corruption on the Palma ratio for all theoretically possible degrees of informality.
Corruption is shown to have a statistically significant positive effect on formal inequality if the informal sector employs less than 37.97% of the total workforce. This means that increased (decreased) corruption leads to increased (decreased) income inequality. Under these circumstances, corruption is shown to disproportionately favor the top 10 relative the bottom 40 percent of the income distribution. However, this only corresponds to 5.26% of the sample19. If the degree of informality exceeds 55.34% corruption instead has a negative effect on inequality, with the relationship being inversed. This corresponds to 46.32% of the observations, and under these societal conditions corruption favors the poor relative the rich in the formal sector. In states with a degree of informality between 37.97% and 55.34% no statistically significant effect is estimated.
If looking at the point estimates, only 22.11% of the total observations have a degree of informality below the threshold of 42.54% informality, where the marginal effect of corruption on inequality turns from positive to negative. 26.32% of the states have a lower mean over the whole time period. On average, 55.05% of all employees work in the informal sector. In this case, a one-percentage point increase in corruption corresponds to a 0.0082 unit decrease in the Palma ratio. In a state experiencing a one standard deviation
19 The sample distribution of Informality is shown in Appendix 4.
Marginal effect of corruption on the Palma ratio
Informality
higher informal sector (66.77%) the effect is -0.0182, or approximately 2%20 of the total Palma standard deviation. In a state with an informal sector one standard deviation lower than the mean (43.32%), the effect is once again positive on 0.0018. However, not enough evidence is found to reject that this effect is significantly different from zero. Furthermore, this only represents 0.2%21 of the Palma standard deviation.
The economic significance of the estimated marginal effect of corruption on inequality is hence minimal for a considerable part of our sample, if looking at a one-unit change. In reality, most Latin American countries have actively fought corruption since the turn of the century (Shepherd, 1998). Therefore, it is reasonable to instead consider a ten-unit decrease in corruption. If done in a state with 66.77% informality, the average increase in the Palma ratio is 0.182, similar to most state standard deviations expressed in Table 2.
What is of further economic significance is the nature of the estimates. Our results support the findings of Dobson and Ramlogan-Dobson (2010; 2012a; 2012b) and Andres and Ramlogan-Dobson (2011) regarding the interconnections between corruption, inequality, and informality. As in line with previous literature we find statistical evidence that the marginal effect of corruption on inequality is a linear function of the degree of informality. Moreover, the majority of our observations have informal sectors large enough for the effect of corruption on inequality to become negative. If applied to the actual situation in Brazil, as well as other parts of the world with large informal sectors, our findings are very much in line with Leff’s (1964) theories about corruption as economic lubricant in an uncertain, inefficient economical and political environment.
In a second-best world, corruption appears to be more beneficial for the poor than for the rich.
5. DISCUSSION
One concern for the validity of the estimates is sample selection bias caused by the usage of 19 Brazilian states instead of all 27, done due to lack of data on corruption22. This bias occurs when data is missing based on the values of the regressand or the residual, and it is therefore of essence that the exclusion of 8 states does not exclude Palma values in a systematic way. To study potential sample selection bias, the average Palma ratio across the original and modified sample23 was plotted for each year. The trend lines shown in Figure 3
20 0.0182/0.802 = 0.02269
21 0.0018/0.802 = 0.00224
22 Values for Corruption are missing for 2006 in Mato Grosso and Pernambuco, 2006 and 2008 in Acre and Rondónia, 2006, 2008 and 2010 in Amapá and Roraima, and all years for Piauí and Tocantins. These states were therefore dropped.
23 All Palma values for 2010 are the values of 2008, due to missing data for all states in 2010. Hence, both the data set including 27 states and 19 states are modified in this way.
indicate that the trend is not heavily affected by removing states from the data. The trend line of the modified sample (solid line) is somewhat more elastic than the one of the original sample (dotted line).
However, their similarity motivates the assumption that the sample is free from sample selection bias caused by data modification.
Figure 3. Average Palma ratio for the original and modified samples over time. The dotted trend line represents the non- modified dataset and the solid line illustrates the trend after data modification.
Robustness tests are conducted in four ways. First, as an alternative to the Palma ratio, Gini is used as a dependent variable in Model 5 (Appendix 5). The estimates for the variables of interest are statistically significant, but the actual values are hard to compare to the ones in Model 4 since they describe the effect on different scales. The general effect is identical to when using Palma. When the degree of informality is sufficiently small corruption increases income inequality, but the marginal effect is diminishing as the informal sector grows. At a degree of informality of more than 48.18%, the marginal effect turns negative, which is very similar to the corresponding level when using the Palma ratio.
Secondly, Table 2 shows how the population size varies significantly over the states. São Paulo, being the biggest state, has 42 million inhabitants, whereas the least populous state, Sergipe, has 2 million24. To test whether the population differences affect the estimates in Model 4, equation (1) is run excluding the four
24 Referring to the average population over 2006-2014.
most populous states in the data set25. The estimated coefficients for Corruption, Informality, and the Interaction term are still statistically significant (Model 6, Appendix 5). Their values are moreover nearly identical to the ones in Model 4, further indicating the robustness of the main empirical specification.
Thirdly, the path of effect is tested. This is to assure that it is not the informal sector given a certain level of corruption that is estimated, but rather the effect of corruption on inequality given a certain degree of informality. Therefore the third sensitivity check is estimated with state levels of informality averaged over time (Model 7, Appendix 5). By doing this, Informality will be dropped due to zero within variation, and it is possible to compare how the estimates differ between Model 4 and 7. Due to the loss of information in the data, the standard errors are increased in Model 7 and the coefficients of interest lose their significance26. The point estimates are however strikingly similar to the ones in Model 4. This provides satisfactory support that it is the effect of corruption on inequality, for a given level of informality, that is being estimated by the main regression.
Lastly, to check for functional form misspecification, a squared corruption polynomial is added in Model 8 (Appendix 5) in reference to Li et al.’s (2000) inverted U-shaped relationship between corruption and inequality. The estimates for Corruption, Informality, and the Interaction term are still statistically significant and do show the same general pattern as Model 4, even though the estimates and standard errors are marginally bigger in absolute terms. The polynomial is not statistically significant, which means no evidence of a non-linear effect based on the value of the corruption term is found. This, in combination with the aforementioned sensitivity checks, provides strong indications of that the coefficients estimated in Model 4 are robust.
Throughout the literature on the effect of corruption on inequality, multiple instrumental variables have been used to check for simultaneous causality. None of them have been considered strong enough to use in this thesis. To exemplify, democracy has been suggested as one potential instrument by multiple sources27. However, as argued by Gradstein et al. (2001), it is flawed as equality in political power is linked to equality in income. Andres (2011) discusses the usage of latitude of the country and the mortality rate of colonial settlers as instruments. Even though they could potentially be applicable to Brazil, they are not viable since they are best suited for pure cross-sectional data. Mauro (1995) and Ramlogan-Dobson (2012a; 2012b) used
25 Bahia, Minas Gerais, São Paulo, and Rio de Janeiro.
26 Corruption p-value=0.301, Interaction term p-value=0.293.
27 (Gupta et al., 2002; Ramlogan-Dobson, 2012a; 2012b)
ethnic fractionalization as an instrument. However, this instrument is deemed as more appropriate in studies on growth rather than this thesis (Andres, 2011).
Given the absence of a check for simultaneous causality, the largest risk of biased estimates stems from possible endogeneity. Therefore, cautiousness is encouraged when interpreting the results presented above as an absolute causal effect of corruption on inequality, given the size of the informal sector. Despite this, the authors still consider the results to be of academic as well as policy relevance since they contribute to the understanding of corruption’s effect in different societal environments.
6. CONCLUSION
So, is corruption inherently bad? This thesis’ results cannot give a clear answer to that question – however, important findings have been made. Given the data set containing 19 Brazilian federal states between every other year between 2006 and 2014, an effect of corruption on income inequality has been found. In line with the research question, this effect is conditional on the size of the informal sector and it alters character depending on the given degree of informality in a state. If the informal sector employs less than 38% of the total workforce corruption increases inequality, while the effect is reversed if the degree of informality exceeds 55%.
Despite the lack of data on income distribution in the unofficial economy, it is reasonable to believe that inequality, poverty, and corruption all permeate society. Consequently, conclusions drawn from the formal sector can to some extent be extrapolated to the informal sector. Therefore, in general terms, it is conceivable that the evidence presented here indicates that corruption in both the formal and informal sector does have an effect on the distribution in society as a whole.
These findings are important as they are in line with previous research, even though different measurements have been applied. As the thesis has had its base in a microeconomic perspective, the key indicators corruption, income inequality, and degree of informality have all been expressed in microeconomic terms.
When comparing to previous literature, which mainly utilizes a macro perspective, this thesis reaches similar conclusions. Together with the gained insight of the relationship between corruption and income inequality in Brazil, the microeconomic aspect of the results is the main academic contribution of the thesis.
In addition to contributing to the research on the relationship between corruption and income inequality, this thesis has contributed to affirming the robustness of the Palma. Although this is not the main objective of the thesis, the results and robustness tests indicate that the Palma ratio might be an intuitively agreeable and statistically viable measurement of income inequality. It is however suggested that this robustness is tested further, and that the Palma should be extensively compared to empirically conventional measurements.
Furthermore, it is suggested to further investigate the importance of the informal sector in the relationship between corruption and income inequality. The results indicate that the informal sector does play a major part in determining the characteristic of the effect. The underlying cause of this is however not yet statistically, or to some extent empirically, determined. To achieve truly policy relevant studies, the intuitive understanding of the underlying structures is essential. However, by increasing the knowledge of the uniqueness of the Latin American continent, a piece of the puzzle that explains the relationship between corruption and inequality has been identified.
In conclusion, the thesis has shown that, given the size of the informal sector, corruption does have an effect on income inequality. The question still remains whether or not this supports the notion that corruption is inherently bad. The results indicate that Leff (1964) might not have been that far off from reality and that corruption might have a beneficial effect on society. However, as this only applies to a second-best world, it cannot be concluded whether or not corruption is inherently bad in praxis. What can be concluded, on the other hand, is that the enigma that is corruption will continue to intrigue for years to come.
7. REFERENCES
Andres, A.R. & Ramlogan-Dobson, C., 2011. Is Corruption Really Bad for Inequality? Evidence from Latin America.
Journal of Development Studies, 47(7), pp.959–976.
Atkinson, A.B., 1970. On the Measurement of Inequality. Journal of Economic Theory, 2, pp.244–263.
Castells, M. & Portes, A., 1991. World Underneath: The Origins, Dynamics, and Effects of the Informal Economy. In:
Portes, A., Castells, M. & Benton, L., ed. 1991. The Informal Economy - Studies in Advanced and Less Developed Countries, 2nd ed., London: The Johns Hopkins University Press. Pp. 11-41
Central Intelligence Agency (CIA), 2016. Country Profile: Brazil. CIA World Factbook. Available at:
https://www.cia.gov/library/publications/the-world-factbook/geos/print_br.html [Accessed April 4, 2016].
Chong, A. & Calderón, C.A., 2000. Institutional quality and income distribution. Economic Development and Cultural Change, 48(4), pp. 761–786.
Cobham, A., Schlogl, L. & Sumner, A., 2015. Inequality and the Tails: The Palma Proposition and Ratio Revisited. DESA Working Paper no.143. New York.
Cobham, A. & Sumner, A., 2013. Is It All About the Tails? The Palma Measure of Income Inequality, Center for Global Development Working Paper no. 343. Washington DC.
De Soto, H., 1989. The other path: The economic answer to terrorism, Basic Books: New York
Dell’Anno, R., 2003. A structural equation approach, University of Aarhus Working Paper no. 2003-07. Aarhus.
Dobson, S. & Ramlogan-Dobson, C., 2010. Is there a trade-off between income inequality and corruption? Evidence from Latin America. Economics Letters, 107(2), pp.102–104.
Dobson, S. & Ramlogan-Dobson, C., 2012a. Inequality, corruption and the informal sector. Economics Letters, 115(1), pp.104–107.
Dobson, S. & Ramlogan-Dobson, C., 2012b. Why is Corruption Less Harmful to Income Inequality in Latin America?
World Development, 40(8), pp.1534–1545.
European Union (EU), 2016. Facts and Figures. Living in the EU. Available at: http://europa.eu/about-eu/facts- figures/living/index_en.html [Accessed April 22, 2016].
Ferreira de Mendonça, H. & Martins Martins Esteves, D., 2014. Income inequality in Brazil: What has changed in recent years? Cepal Review, (112), pp.107–123.
Foellmi, R. & Oechslin, M., 2007. Who gains from non-collusive corruption? Journal of Development Economics, 82(1), pp.95–119.
Gradstein, M., Milanovic, B. and Ying, Y., 2001. Democracy and income inequality: an empirical analysis. World Bank Policy Research Working Paper no. 2561. New York.
Gupta, S., Davoodi, H.R. & Alonso-Terme, R., 2002. Does Corruption Affect Income Inequality and Poverty?
Economics of Governance, 3(1), pp.23–45.
Gyimah-Brempong, K. & Munoz de Camacho, S., 2006. Corruption, growth, and income distribution: Are there regional differences? Economics of Governance, 7(3), pp.245–269.
