«No. 2008/11 ISSN 1478-9396 CORRUPTION, PRIVATISATION AND THE DISTRIBUTION OF INCOME IN LATIN AMERICA Antonio RODRIGUEZ ANDRES and Carlyn ...»
No. 2008/11 ISSN 1478-9396
CORRUPTION, PRIVATISATION AND THE DISTRIBUTION OF
INCOME IN LATIN AMERICA
Antonio RODRIGUEZ ANDRES and
DISCUSSION PAPERS IN ECONOMICS
The economic research undertaken at Nottingham Trent University covers various fields of economics. But, a large part of it was grouped into two categories, Applied Economics and Policy and Political Economy.
This paper is part of the new series, Discussion Papers in Economics.
Earlier papers in all series can be found at:
http://www.ntu.ac.uk/research/school_research/nbs/overview/working_papers/index.html Enquiries concerning this or any of our other Discussion Papers should be addressed
to the Editor:
Dr Juan Carlos Cuestas Division of Economics Nottingham Trent University Burton Street Nottingham, NG1 4BU
Key words: corruption, Latin America, income inequality, instrumental variables, panel data, privatisation.
JEL classifications: O15, O54 Corruption, privatisation and the distribution of income in Latin America
1. Introduction Surveys of public opinion in Latin America highlight corruption and inequality as major problems facing the region, along with unemployment and crime (Lagos, 2003).
Though corruption is perceived to be a problem throughout the region, the International Country Risk Guide (ICRG) 1 reports that during the period 1980-2000, several countries, including Bolivia, Ecuador, Guatemala and El Salvador, showed evidence of declining corruption. In 2001, 90 per cent of the population considered the distribution of income in the region to be unfair or very unfair (see Lopez and Perry, 2008).
The origins of corruption and income inequality in Latin America go back to the early post colonial period and the development of key institutions (Engerman and Sokoloff, 2002;
Acemoglu et al, 2002). At this time, a privileged few controlled the profitable activities and to protect their interests, institutions were structured in such a way that most of the population were denied access to land, education and political power. The pattern of non-representative and exclusionary institutions survived the move to independence across the region as the Creole elite gained control of key institutions and shaped them to their advantage. This elite group was able to wield significant influence on the formation and implementation of government policies. For example, the failure to expand public education helped to protect the vested interests of the elite group. 2 This neglect continued into the 20th century with education being of low quality 3 and patterns of social exclusion and discrimination persisting (Lopez and Perry, 2008).
The opening up of the international economy exacerbated rather than reduced income disparities because the gains accrued to landholders (the elite). These gains were exaggerated by the fact that Latin America is rich in natural resources, the abundant productive factor in the region. Natural resources (rather than labour 4 ) were more intensively used in the production of exportable goods. Consequently, returns to land grew relative to those of labour. Since the majority of the population were excluded from owning property, the income distribution problem worsened as the wealth of landowners increased. The natural outcome was that inequality increased over the early period of globalisation (Williamson, 1999) 5.
While the above sheds light on the roots of inequality in Latin America, it also illustrates the close link between corrupt practices, institutions and inequality. It seems reasonable to conclude that if there had been less preferential treatment towards the few in the early colonial period, the outcome with respect to inequality may well have been different.
The discussion also highlights the fact that corruption is entrenched in the political and economic operations of the region.
Bourguignon and Morrisson (2002) suggest that the distribution of income did not change from the time of independence to the mid 20th century, while Morley (2000) argues that since World War 2 the situation has worsened. Londono and Szekeley (2000) argue that inequality levels in the 1990s were similar to those in the 1930s. De Ferranti et al (2004) note that, as in the 19th century, authoritarianism, may be the primary reason for the persistence of inequality in the 20th century. Although democratisation has taken place, the process is unconsolidated and the authors conclude that correcting institutional failures along with direct polices are essential to reduce inequality. Perry et al (2006) confirm the findings of De Ferranti et al and after examining the evidence conclude that Latin America entered the 20th century with high levels of inequality which persisted for the rest of the century. This conclusion is highlighted in a study of Argentina; Calvo et al (2002) indicate that inequality levels changed little during the 20th century.
According to economic theory, corruption is expected to worsen income inequality (Mauro, 1997; Jain, 2001; Gupta et al, 2002). Corruption, in the form of tax evasions and exemptions, reduces tax revenues and funds for social programmes, including education and health. Furthermore, since the beneficiaries of tax evasion and exemptions are more likely to be the relatively wealthy, the tax burden falls almost exclusively on the poor, making the effective tax system regressive. The impact on social programmes can be more direct as funds may be siphoned out of poverty alleviation programmes in order to extend benefits to relatively wealthy population groups. Even when social programmes are not reduced, corruption may change the composition of social spending in a manner that benefits the rich at the expense of the poor; for example, expenditure on tertiary rather primary education. In a corrupt system, the allocation of public procurement contracts may lead to inferior public infrastructure, which also has implications for inequality and welfare. In sum, corruption in a government allows for polices which favour the higher income groups and hence promotes greater inequality.
The empirical literature on corruption and income inequality finds that higher levels of corruption increase income inequality. In a few studies a number of Latin American countries have been included as part of a larger sample of both developing and developed countries (e.g. Li et al, 2000; Gupta et al, 2002; Gyimah-Brempong and Muñoz de Camacho, 2006).
However, no study has yet examined inequality and corruption across Latin America. The region has seen financial crises, periods of positive and negative growth, huge external borrowing, closed market policies and pro-market reforms, yet high inequality persists and our understanding of income inequality remains limited. In this study, we present new evidence on income inequality in Latin America, focusing in particular on the relationship between inequality and corruption. In contrast to other work and a priori expectation, we find that lower levels of corruption are associated with higher levels of inequality. This surprising finding is explained by the privatisation process in the region (see Section 3).
The structure of the paper is as follows. In Section 2 the model specification and data are described. The empirical results are presented and discussed in Section 3. Section 4 reports some robustness tests and Section 5 concludes.
2. Model Specification and Data Econometric estimation is conducted using four-year panel data over the period 1981for 19 Latin American countries, 6 with each observation of the dependent variable being the relevant four year average value. Panel data provides more degrees of freedom than crosssection and time series data. Furthermore, panel data analysis controls for omitted variable bias, thus improving the accuracy of parameter estimates. This approach also has the advantage of capturing possible idiosyncratic differences in income inequality by means of the time invariant individual effects. A priori, a fixed effects model is preferred to a random effects model since we expect the explanatory variables to be correlated with the unobserved individual effects. All the countries of the region for which data is available are included in the study. There are some missing observations in the data so the panel is unbalanced.
The empirical specification is similar to that in previous empirical research (see Li et
al, 1998; Barro, 2000; Lundberg and Squire, 2003):
where I is a measure of income inequality for country i at time t. Xit is a vector of explanatory variables which vary across time and countries. It includes a corruption variable (corupt) among other explanatory variables. The parameter Ai contains a constant and individualspecific variables that are invariant over time (for example, geographical factors), and εit is the classical error term.
The dependent variable is a standard measure of income inequality, the Gini coefficient. The data on inequality is drawn from the United Nations World Income Inequality Database (WIID) (UNU-WIDER, 2005). 7 We use the new quality label provided in Version 2a of the WIID, which combines and improves the quality ratings in Deininger and Squire (1996) with older versions of the WIID. Data classified as the lowest quality is excluded.
Furthermore, only data which covers the entire population is employed. Gini coefficients are based on income rather than on consumption because of data limitations. For each country, we have formed the longest possible series of observations.
The measure of corruption adopted is the International Country Risk Guide (ICRG) corruption index which is collected and published annually by Political Risk Services (PRS).
This measure focuses on corruption in government and has been used in the development economics literature (e.g. Fisman and Gatti, 2002). The corruption variable is intended to capture the likelihood that high level government officials will demand special payments, and the extent to which illegal payments are expected throughout lower levels of government (Knack and Keefer, 1995). Compared to the Corruption Perception Index (CPI), this measure has the advantage of having the broadest coverage for Latin American countries for the study period and it is appropriate here because we are interested in examining the role of corruption in government. The ICRG measure takes values from zero (most corrupt) to six (least corrupt), so a priori, a rise in the corruption index (less corruption) is expected to lead to a fall in the Gini coefficient (a negative sign on the variable corupt). The privatisation variable (priv) is taken from Lora (2001) and is defined as cumulative privatisation as a percentage of GDP.
The natural logarithm of real output per capita (lgdp) and real output per capita squared (lgdp2) are included to test the classical Kuznets hypothesis (Kuznets, 1955; Lewis, 1954). According to this hypothesis, inequality rises with income at low levels but falls once income reaches a critical level. In line with other studies (e.g. Bourguignon and Morrison, 1998; Li et al, 1998; De Janvry and Sadoulet, 2000; Morley, 2000; Reuveny and Li, 2003;
Breen and García-Peñalosa, 2005), the model also includes the following variables: primary (primary) and secondary (secondary) gross school enrolment rates, the share of agriculture in total output (aggdp), the ratio of broad money to output (m2gdp) and a variable to represent the distribution of land resources (land). Both land and education represent investment in assets (physical and human) and should contribute to lowering inequality. Because of its labour intensive nature, an expansion of the agriculture sector is expected to increase employment levels and contribute to reducing inequality. Finally, m2gdp is included as an indicator of financial development. Greater financial development is expected to lower inequality by alleviating credit constraints and by making investment opportunities more available to low income households. Data for all these variables is taken from the Penn World Table, Version 6.1 (Heston, Summers, and Aten, 2002), World Bank’s World Development Indicators (2003) and Frankema (2005).
An important potential issue in estimating equation (1) is the endogeneity of the control variables. Incorporating time invariant fixed effects into the model addresses this issue to some extent, but the inclusion of time varying factors means omitted variable bias is still a potential problem. Furthermore, if there is correlation between at least one explanatory variable and the error term, OLS estimates will suffer from simultaneity bias.8 In order to deal with both potential problems, an instrumental variable (IV) methodology is adopted. It should be noted, however, that because of data limitations we only instrument for the corruption variable.