- Email: [email protected]
Corruption and the effects of economic freedom
European Journal of Political Economy 29 (2013) 54–72
Contents lists available at SciVerse ScienceDirect
European Journal of Political Economy journal homepage: www.elsevier.com/locate/ejpe
Corruption and the effects of economic freedom L. Pieroni a,⁎, G. d'Agostino b a b
University of Perugia, Via Pascoli, 20‐06123 Perugia, Italy University of Roma Tre, via Silvio D'Amico 77, 00145 Roma, Italy
a r t i c l e
i n f o
Article history: Received 6 September 2011 Received in revised form 24 August 2012 Accepted 26 August 2012 Available online 31 August 2012 JEL classification: H11 H50 K20 Keywords: Corruption Economic freedom Multilevel models
a b s t r a c t The prediction that economic freedom is beneficial in reducing corruption has not been found to be universally robust in empirical studies. The present work reviews this relationship by using firms' data in a cross-country survey and argues that approaches using aggregated macro data have not been able to explain it appropriately. We model cross-country variations of the microfounded economic freedom–corruption relationship using multilevel models. Additionally, we analyse this relationship by disentangling the determinants for several components of economic freedom because not all areas affect corruption equally. The results show that the extent of the macro-effects on the measures of (micro)economic freedom for corruption, identified by the degree of economic development of a country, can explain why a lack of competition policies and government regulations may yield more corruption. Estimations for Africa and transition economy subsamples confirm our conjectures. © 2012 Elsevier B.V. All rights reserved.
1. Introduction Much attention has recently been devoted to testing the relationship between economic freedom and corruption, with the expectation that economic freedom leads to reduced levels of corruption. 1 While empirical studies do tend to support this expectation (see Chafuen and Guzmàn, 2000; Paldam, 2002b; Graeff and Mehlkop, 2003; Shen and Williamson, 2005; Carden and Verdon, 2010), there has been some inconsistency in the results. Certainly, economic freedom is found to reduce corruption when it is regressed on the entire distribution of corruption subdivided by its components, and when the relationship is tested for subsamples of countries. However, Billger and Goel (2009) found that, among the most corrupt nations, greater economic freedom does not reduce corruption and may even exacerbate it, indicating that nations respond differently to levels of economic freedom, depending on their level or stage of development conditions. 2 In addition, Graeff and Mehlkop (2003), while arguing against the use of an aggregate indicator of economic freedom to evaluate their effects, find support for the counter-intuitive influence of the size of government on corruption, as a specific component of economic freedom. The literature has also invoked an institutional dimension to explain some unexpected results. Lambsdorff (2007) argued that government intervention may not be sufficient if it aims only to avoid market failures, and that “good” government regulations may be important in reducing corruption. Thus, the effect of aspects of economic freedom on corruption may depend on how efficient government intervention is in countries which are more highly “regulated” or “have high levels of economic freedom”. ⁎ Corresponding author at: University of Perugia, Department of Political Science, Italy. Tel.: +39 0755855280; fax: +39 0755855299. E-mail addresses: [email protected] (L. Pieroni), [email protected] (G. d'Agostino). 1 Berggren (2003) defines economic freedom as “a composite that attempts to characterise the degree to which an economy is a market economy, that is, the degree to which it entails the possibility to enter into voluntary contracts within the framework of a stable and predictable rule of law that upholds contracts and protects private property, with a limited degree of interventionism in the form of government ownership, regulations and taxes”. 2 Lambsdorff (2007) provides a review of studies in which the relationship between economic freedom and corruption is insignificant or in which the effects of economic freedom increase corruption. 0176-2680/$ – see front matter © 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.ejpoleco.2012.08.002
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
55
This paper, in contrast with most of the existing research focusing on cross-country studies, reviews the relationship between economic freedom and corruption using firm level data, following the approach originally developed by Milgrom and Roberts (2008). It also considers macro-effects, such as the level of economic development and the quality of institutions (which are known to provide a fertile breeding ground for corruption), by means of a multilevel modelling framework. This mitigates the mismatch between macro- and micro-evidence, with the aim of keeping micro-relationships robust and identifying the socio-economic and institutional factors which are responsible for cross-country differences. It thus contributes to empirical work investigating whether more freedom lowers corruption (Chafuen and Guzmàn, 2000; Paldam, 2002a,b; Apergis et al., 2011), although it follows from Graeff and Mehlkop (2003) in considering how different aspects of economic freedom affect corruptive practices. This paper is also influenced by the extensive literature which considers how corruption can undermine the strength of public institutions and hamper economic growth and development (and vice versa), including Shleifer and Vishny (1993), Mauro (1995), Bardhan (1997), Treisman (2000) and Méon and Sekkat (2005), the more limited empirical analyses which use micro-data to investigate corruption and its determinants (Swamy et al., 2001; Svensson, 2005; Mocan, 2008) and the work of Hodgson (2006) and Hodgson and Jiang (2007), who consider the role of institutions in affecting corruption. We begin our work by documenting basic facts regarding the key variables of our empirical model and discuss the conceptual framework to explore why some dimensions of economic freedom are likely to affect corruption at firm level. We then provide information concerning data extracted from the World Business Environment Survey (WBES) and how we derive variables for empirical analysis. Section 4 presents multilevel models, a more realistic framework in which to test the hypotheses of our work. Estimating the micro-economic links in a multilevel model allows us to assume that the unexplained, insignificant or mitigated relationships of corruption with the components of economic freedom may be explained by differences in country characteristics. Thus, unlike models, which include conditioning variables to explain why the effect of economic freedom on corruption is not linear, we specify the broadest class of multilevel models and propose a selective strategy in which nested and non-nested multilevel models are tested. The rest of the paper estimates the parameters of these models. Although the results in the full sample are sufficiently in line with standard economic predictions, attention to firms' heterogeneous responses and cross-country effects is crucial in explaining empirical variations of outcomes. The results highlight the fact that government regulations in African countries and transition economies may reduce corruption although, as suspected, the large variability across countries is identified not only by differences in economic prosperity but also by quality institutions. 2. Economic freedom and links with corruption When we aim at measuring corruption in empirical studies, we find a number of possible indicators. The highest profile measure is provided by Transparency International, which uses survey methods to produce the corruption perception index (CPI). Although such perception-based indices may not provide robust estimates of corruption within countries, they do remain informative when used for cross-country comparisons across countries. 3 The economic freedom index published by the Economic Freedom of the World (EFW) is designed to measure the quality of a country's institutions and policies (Gwartney and Lawson, 2000, 2003) and to affect the effectiveness of individual incentives, productive effort, and the effectiveness of resource allocation (North and Thomas, 1973; de Haan and Sturm, 2003). The key components of the index are ‘objective’ measures, such as government consumption as a share of total consumption, and transfers and subsidies as a share of GDP, which are collected from external sources (International Monetary Fund, World Bank and World Economic Forum), and ‘subjective’ measures based on surveys, expert panels and case studies. 4 The scores for each country are then given for five areas of economic freedom: i) size of government; ii) legal structure and security of property rights; iii) access to finances and sound money; iv) freedom to trade internationally, and v) regulation of credit, labour, and business. The mean scores give the aggregate EFW index, which ranges from 1 to 10, where 10 is the maximum degree of economic freedom. The EFW index shows that economic freedom has improved over time, as the average value has increased from 5.8 in 1990 to 6.6 in 2000, and then rises in the first decade of the 21st century. For some idea of the probable relation with corruption, Fig. 1 shows the EFW index and the CPI corruption index (measured inversely in terms of good governance) across world countries for the year 2000, and illustrates the clear positive association between the indices, implying that higher economic freedom is associated with lower corruption. 5 While cross-country regressions of economic freedom and corruption give results in line with the prevailing view, that corruption is reduced by an increase in economic freedom, this may hide heterogeneity at sectoral level, particularly in the degree and nature of government intervention. Graeff and Mehlkop (2003) looking at the components of the economic freedom index find a more complex picture, with ambiguous correlations among the components of economic freedom and corruption. This suggests that the weight of the components of the aggregate index influences the effect on corruption, implying that the results are sample-sensitive. In addition, Lambsdorff (2007) argues that corruption may also arise in sectors with greater economic freedom, as not all aspects of economic freedom deter corruption. Some government regulations may have the effect of increasing
3
For a discussion on this topic see, for example, Gorodnichenko and Sabirianova (2007). The number of variables used to construct this index is not constant over time. For example, 21 components of economic freedom were considered in 2005, and 23 in 2008. 5 For an application of the association between economic freedom and growth, see de Jong and Bogmans (2011). 4
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
4
6
R2=0.58
2
Corruption index
8
10
56
4
5
6
7
8
9
Economic Freedom index Fig. 1. Cross-country relationship between economic freedom and (less) corruption.
the transaction costs of corruption and in such cases, policy-makers are unresponsive to the demand for regulation, and “free” competition and lack of government regulations would lead to a failure of anti-corruption policy. When combined with weak legal apparatus and enforcement mechanisms, as is often the case in less developed countries or transition economies, corruption may become rife, as the spontaneous mechanisms of economic freedom are conditioned by local rules which impose private bribes, frequently under the guise of taxation. 2.1. Conceptual framework We now discuss in detail which dimensions of economic freedom are likely to affect corruption at firm level. We do not include the size of government, a dimension which naturally affects corruption at aggregate level, but do include market competition, expected positively to affect the probability of corruption of a firm. Here, we list the five dimensions of economic freedom assumed to be related to corrupt activities. 2.1.1. Dimension I: market competition This paper focuses on the micro-dimensions which refer to market competition. Non-competitive markets may exhibit rent-seeking behaviour, bureaucrats acting independently of market forces, and constraints on the flow of information, all limiting competition and encouraging corruption beyond firms' control. While modelling the interaction between corrupt government officials and industrial firms, Emerson (2006) shows corruption to be antithetical to competition, and the competition–corruption link may easily suffer from reverse causality or sometimes even have a positive effect (Lambsdorff, 2007). The latter author considers that the prospect of money from corruptive practices may motivate private firms to pay bribes, thus creating endogeneity in the relationship; if quality-based rather than price-based competition prevails in the market, firms may be driven by myopic behaviour in delivering substandard quality products, meaning that corruption hampers competition and, in general, international trade. 6 The literature also uses proxies of (non)-market dominance, such as the effectiveness of non-competitive practices and the degree of information available on laws and regulations, as subjective measures of obstacles to the growth of firms' business. But further endogeneity problems may arise (Lambsdorff, 2007). 2.1.2. Dimension II: government regulation of private entrepreneurial activity Corruption is assumed to be influenced by aspects of government regulation, although it would be wrong to perceive corruption as the consequence of excessive regulation, or to assume that complete laissez-faire will always be the answer (Bliss and Di Tella, 1997). The view that government regulation hampers productive effort, encourages rent-seeking and increases the discretionary power of a few public officials still reigns throughout government, institutions and the literature (e.g., Graeff and Mehlkop, 2003; Paldam, 2002a). However, when government regulation is weak or almost absent, an increase in market rules is still recognised as crucial in developing a solid productive sector (Hodgson, 2003). Developing countries are often characterised by weak law enforcement, a large informal sector, underdeveloped capital markets, and informal credit and insurance networks. 7 This means that results from developed countries should not be directly extended to those from developing countries without serious reflection on their differences, both the trade-off between the benefits and disadvantages of state intervention in dealing with market failures (Acemoglu and Verdier, 2000) and the different context regarding the negative externalities of corruption.
6
For a discussion, see Compton et al. (2011). For example, informal arrangements, such as family networks of credit and insurance, have been found to influence the consequences of interventions to a great extent, limiting the onset of corruption. 7
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
57
Second, and even more importantly, empirical research has tended to focus on the effect of the overall size of the government budget relative to GDP, often neglecting the interactions of the various components of government regulation. 8 It also suggests that particular types of government intervention may lead to corruption. In an environment in which heterogeneous goods and services may be supplied in free markets at high transaction costs, which may be reduced by government regulations, government quantity and price interventions may later lead to serious problems of corruption. Indeed, Svensson (2003) finds that firms typically have to pay bribes when dealing with public officials whose actions directly affect firms' business, particularly their investments. 2.1.3. Dimension III: ability of the financial system to support private firms It has been postulated that the ability of a formal financial system to provide financing to the private sector reduces the effects of corruption. This relationship may be mitigated by the different quality of financial institutions across countries and, in the extreme case, the quality of institutions may reflect the level of corruption. Brunetti et al. (1997) found that the ability of the financial system reduced corruption, by ranking levels of corruption from a worldwide survey and observing that the second most significant impediment to doing business was lack of financing. However, even when the financial system provides adequate finance for the private sector, inefficiencies in business investments can lead to higher intermediation costs, which are associated with corruption payments (Ahlin and Pang, 2008). 2.1.4. Dimension IV: property rights and the protection of contracts The ability of the legal system to protect property rights is widely believed to reduce corruption, whereas the failure of the legal system to enforce contracts undermines the free market and reduces incentives for agents to participate in productive activities (Acemoglu and Verdier, 1998), leading to increases in corruption. 2.1.5. Dimension V: regulation of exports Government regulations on exports and international trade are usually assumed to increase the level of corruption and, there is a fair amount of empirical evidence to back this up. There are three main explanations. One is the relation between the rents of firms in a non-competitive market and import licensing Krueger (1974). When the number of licences is fixed and firms are encouraged to compete, a rational firm may shift productive plans to rent intensive activities and may then turn to bribery. Another explanation is that trade barriers may favour inefficient local firms rather than more efficient foreign ones and this may lead to corruptive practices Faber and Gerritse (2012). Thirdly, import barriers may create an artificial scarcity of specific commodities, channelling some of the non-competitive higher prices towards corrupt bureaucrats, a practice which, in developing countries or transition economies, may lead to underground economies. However, the empirical evidence of a negative sign in the relation between the extension of international trade and corruption is questionable (Treisman, 2000; Torrez, 2002). In general, it is important to recognise that the micro-economic relationship between aspects of economic freedom and corruption may be affected by factors that operate at more macro- or even country level. For example, economic development is argued to be negatively affected by corruption and vice versa (Mauro, 1995), whereas the nature and quality of institutions within a country can have a direct effect on development and corruption (Acemoglu et al. 2001). Thus, an increase in the quality of institutions would reduce the propensity to corruption. The aim of our empirical framework is to include these macro-level effects in investigating the relationship between economic freedom and corruption. 3. Data and variables In this paper, the data are mainly taken from the WBES cross-sectional survey of industrial and service enterprises ‘Voices of the Firms 2000’, conducted in mid-1999 by the World Bank and other agencies. This WBES survey, the most comprehensive source of micro-data for analysing both corruption and economic freedom by firms' perception responses, covers 67 countries with, on average, more than 100 firms per country. Appendix 1 lists the countries by region and the number of firms interviewed. The survey provides detailed information on taxation, government regulation, the financial sector, and perceptions of corruption in business (based on the opinions of firms' managers). Responses to questions on corruption include not only direct experience, but also perceptions of the behaviour of neighbouring firms in the same environment. Firms' responses to questions concerning corruption were grouped and values were assigned to them. Thus, in response to questions such as “How usual is it for firms to have to pay some kind of irregular extra payment in order to obtain services from the public administration?”, 6 indicates “always” and 1 “never”. Lower corruption was then defined as values from 1 to 3 inclusive, and higher corruption from 4 to 6. Gorodnichenko and Sabirianova (2007) suggested that this binary categorisation may lead to a better distinction between higher and lower degrees of corruption. It may also be explained by looking at the empirical distribution of firms' responses to the survey question. As Fig. 2 shows, the responses were not evenly distributed, but asymmetrically moved distribution towards higher values, meaning that intermediate values make it difficult to distinguish between higher or lower degrees of corruption, and thus makes the relationship with economic freedom more difficult to interpret from the policy-makers' point of view. 8 The test that size of government is positively correlated with level of corruption is weak: Elliot (1997) and Adsera (2003) obtained contrasting findings, and Graeff and Mehlkop (2003) and Billger and Goel (2009) reported ambiguous results, which could only partially interpreted by observing the relation in a subsample of countries, or conditioning the distribution of corruption variables across countries, respectively.
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
1.5 1 0
.5
Density
2
2.5
58
1
2
3
4
5
6
Corruption index Fig. 2. Corruption index by response category. Note: Aggregate score responses of category of corruptive practices from 1 to 6: 1 (never), 2 (seldom), 3 (sometimes), 4 (frequently), 5 (usually) and 6 (always).
With regard to economic freedom, the absence of market dominance was measured by considering the number of enterprises and the existence of anti-competitive practices. For the former, a scale from 1 to 3 was used, 3 implying a low degree of market dominance; anti-competitive practices were graded from 1 to 4, 1 indicating none. Following Ades and Di Tella (1999), an indicator of the degree of market competition was constructed, based on firms' ability to obtain information on laws and regulations, ranging from 6 if the firm respondent disagreed that information was easy to obtain, and 1 otherwise. To compare the areas of Gwartney and Lawson (2002) with our micro-dimensions of economic freedom, Dimensions II and III above are equivalent to Gwartney and Lawson's (2002) fifth area (regulation of credit, labour, and business). In Dimension II, respondents answer the question “How often does the government intervene … in decisions by your firm?”, and indicators of government regulation in private investment, employment sales and prices were used to measure the magnitude of this intervention. For all four indices, firms were graded on a scale from 1 to 6, where 6 represents government intervention in private activities. Dimension III includes indicators of financial constraint and obstacles. For constraints, a binary variable measuring the lack of financing in the private sector is used, whereas the nature of financial obstacles is ranked from 1 to 4, where 4 represents major obstacles. In line with the legal structure and security of property rights dimension in Gwartney and Lawson (2002), an index was constructed which measures the ability of the legal system to protect “property rights and contracts”, where the maximum 3 is a firm's perception that it is completely able and 1 that it is unable. Lastly, Dimension V uses an indicator for export regulation in line with the Freedom to Exchange with Foreigners, although a functional classification based on government intervention might also include this in Dimension II. Table 1 presents the complete description of the corruption index and the micro-covariates of economic freedom. Figs. 3 and 4 show the bivariate relationships between the components of economic freedom and corruption at country level, after aggregating firms' responses. Aggregate-level data were created to allow comparisons with the usual empirical analyses in the literature. 9 Unlike the scale of the Transparency International index, corruption increases along the y-axis, whereas economic freedom decreases right of the x-axis. In many of the panels, a clear-cut picture of the statistical significance between economic freedom indices and corruption cannot be defined. For non-market dominance, information constraints (Fig. 3, panels a and c), components of government regulation (Fig. 3, panels d–e; Fig. 4, panels a–b) and property rights and contract protection (Fig. 4, panels d), the patterns do not support the relationship suggested by standard theory, i.e., that more freedom reduces corruption. The graphs also clearly highlight the existence of clusters of countries, suggesting differences in the effects of the components of economic freedom on corruption across groups of countries. A multilevel modelling framework was adopted to explain these clusters assuming that, on average, firms' responses differ across countries. Without losing generality, this framework may be extended to the remaining economic freedom components, i.e., the absence of anti-competitive behaviour, more freedom in the financial system and export regulation, although in aggregate they consistently show that more freedom reduces corruption (Fig. 3, panel b; Fig. 4, panels c and e). This means that the possibility that unexplained variability at micro-level may result in different or non-significant responses in the relationship at macro-level. Note that, in line with the above classification, the impact on corruption of property rights and the protection of contracts at micro-level is modelled. 10 9 Our aim was to give some descriptive insights of the relation to be tested. This is statistically equivalent to aggregating all individual level variables to group level and carrying out ordinary least squares, e.g., regression over group means. Criticism of this procedure is discussed in Kreft and de Leeuw (1998). 10 One criticism in modelling the quality of institutions on corruption at micro-level is based on the low variability of firms' preferences and expectations. Mocan (2008) developed a model which assumes that an increase in the quality of the institutions of a country, which would increase the probability of apprehension, would in turn reduce the propensity to ask for bribes.
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
59
Table 1 Data and descriptions of variables. Variables Dependent variable Corruption index
Description
Range
“It is common for firms in my line of business to have to pay some irregular additional payments to get things done.”
Discrete variable from 1 to 6: 1 (never), 2 (seldom), 3 (sometimes), 4 (frequently), 5 (mostly), and 6 (always).
Components of economic freedom Non-market dominance “Regarding your firm's major product line, how many competitors do you face in your market?” Anti-competitive “Please judge on a four-point scale how behaviour problematic the following factors are for the operation and growth of your business.” Information constraints “In general, information on the laws and regulations affecting my firm is easy to obtain.” Private investment regulation
Employment regulation
Sales regulation
Prices regulation
Financial constraints
Financial obstacles
Property rights and contract protection Export regulation
“How often does the government intervene in the following types of decisions by your firm?” “How often does the government intervene in the following types of decisions by your firm?” “How often does the government intervene in the following types of decisions by your firm?” “How often does the government intervene in the following types of decisions by your firm?” “What is the ability of the financial system to provide financing to the private sector?” “Please judge on a four-point scale how problematic the following factors are for the operation and growth of your business.” “Please, judge the ability of the legal system to protect property rights and contracts” “How often does the government intervene in the following types of decisions by your firm?”
Firm's covariates Firm size
“Number of full-time employees”
Firm sector
“Principal sector of business”
Legal organisation of firm
“What is the legal organisation of this company?”
Macro covariates GDP INV GOV GINI CIVIL
Gross domestic product per capita Private investments as share of GDP Government spending as share of GDP Gini coefficient of distribution of income Civil liberty index
Discrete variable from 1 to 3: 1 (no competitors), 2 (3 or fewer), and 3 (more than 3). Discrete variable from 1 to 4: 1 (no obstacle), 2 (minor obstacle), 3 (moderate obstacle), and 4 (major obstacle). Discrete variable from 1 to 6: 1 (fully agree), 2 (agree in most cases) 3 (tend to agree), 4 (tend to disagree), 5 (disagree in most cases), and 6 (totally disagree). Discrete variable from 1 to 6: 1 (never), 2 (seldom), 3 (sometimes), 4 (frequently), 5 (usually), and 6 (always). Discrete variable from 1 to 6: 1 (never), 2 (seldom), 3 (sometimes), 4 (frequently), 5 (usually), and 6 (always). Discrete variable from 1 to 6: 1 (never), 2 (seldom), 3 (sometimes), 4 (frequently), 5 (usually), and 6 (always). Discrete variable from 1 to 6: 1 (never), 2 (seldom), 3 (sometimes), 4 (frequently), 5 (usually), and 6 (always). Binary variable: 0 (financing), and 1 (no financing). Discrete variable from 1 to 4: 1 (no obstacle), 2 (minor obstacle) 3 (moderate obstacle), and 4 (major obstacle). Discrete variable from 1 to 3: 1 (totally disagree), 2 (tend to agree), and 3 (fully agree). Discrete variable from 1 to 3: 1 (never), 2 (seldom), 3 (sometimes), 4 (frequently), 5 (usually), and 6 (always).
Discrete variable from 1 to 3: 1 (small, 50 or fewer employees), 2 (medium, 51–500 employees), and 3 (large, 501 or more employees) Discrete variable from 1 to 5: 1 (manufacturing), 2 (services), 3 (other), 4 (agriculture), and 5 (construction) Discrete variable from 1 to 5: 1 (single proprietorship), 2 (partnership), 3 (cooperative), 4 (corporation, privately held), and 5 (corporation listed on stock exchange)
Ordered variable from 1 (no civil liberties) to 6 (full civil liberties)
Dimension
Source WBES2000
Dimension I
WBES2000
Dimension I
WBES2000
Dimension I
WBES2000
Dimension II
WBES2000
Dimension II
WBES2000
Dimension II
WBES2000
Dimension II
WBES2000
Dimension III
WBES2000
Dimension III
WBES2000
Dimension IV
WBES2000
Dimension V
WBES2000
WBES2000
WBES2000
WBES2000
WDI WDI WDI WDI WAO
Notes: Labels of sources are: “WBES2000”, World Business Environment Survey, World Bank, 2000; “WDI”, World Development Indicators; and “WAO” World Audit Organisation.
60
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
(b) Anti-competitive behaviour (Dimension I) 3 2.5 2 1.5
Corruption index
2.5 2
R2=0.63
1
1.5 1
Corruption index
3
(a) Non-market dominance (Dimension I)
1
1.5
2
2.5
3
1.5
2
Non-market dominance
3
3.5
(d) Private investment regulation (Dimension II)
2.5 2
Corruption index
2
1
1.5 1
1.5
2.5
3
3
(c) Information constraints (Dimension I)
Corruption index
2.5
Anti-competitive behaviours
1
2
3
1
4
1.5
Information constraints
2
2.5
3
Private investment regulation
2.5 2 1.5 1
Corruption index
3
(e) Employment regulation (Dimension II)
1
1.5
2
2.5
3
Employment regulation Fig. 3. Components of economic freedom and corruption, WBES sample. Notes: Figures built by aggregating individual data of economic freedom components and corruption at country level.
The variables involved in the multilevel model are, first, control variables, to account for heterogeneity in firm behaviour, allowing for the possibility that the extent of a firm's self-reported corruption effects may depend on its size, legal organisation, and the sector in which it operates (Beck et al., 2002). Firm size is the number of full-time employees and classifies a firm into three categories, large (3), if there are more than 500 full-time employees; medium (2) if there are between 51 and 500 employees; and small (1) when the number of employees is under 50. The main sectors of business are subdivided into five traditional categories: manufacturing, services, agriculture, construction, and other, from 1 to 5, respectively; the legal organisation of the firm is also coded from 1 to 5: single proprietorship (1), partnership, cooperative, privately held corporation, or listed on the stock exchange (5). These variables are also described in Table 1. The existence of clusters suggested that macro-characteristics may influence the economic freedom/corruption link. Assuming that differences in economic conditions and institutional characteristics (in Hodgson's sense) can explain differences in levels of corruption, in line with Acemoglu et al. (2001) and La Porta et al. (2008), macro-economic indicators, per capita GDP, log (GDP) 11 private investment as a share of GDP (INV) and the Gini coefficient of distribution of income (GINI) were all considered. As usual, 11
Capital letters distinguish macro-covariates.
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
2.5 2
Corruption index
1
1.5
2.5 2 1.5 1
Corruption index
3
(b) Prices regulation (Dimension II)
3
(a) Sales regulation (Dimension II)
61
1
1.5
2
2.5
3
1
1.5
Sales regulation
2
2.5
3
Prices regulation
(d) Property rights and contract protection (c) Financial constraints (Dimension III)
3 2.5 2
Corruption index
1
1.5
2.5 2 1.5
R2=0.49
1
Corruption index
3
(Dimension IV)
1.5
2
2.5
3
1.5
3.5
Financial Constraints
2
2.5
3
Property rights and contract protection
2.5 2 1.5
2 R =0.19
1
Corruption index
3
(e) Exportregulation (DimensionV)
2
2.5
3
3.5
4
Export regulation Fig. 4. Economic freedom components and corruption, WBES sample. Notes: Figures built by aggregating individual data of components of economic freedom and corruption variables at country level. Financial constraints are not reported, as they are defined by a binary variable.
0 in the Gini index represents perfect equality, and 100 perfect inequality. The share of government spending in GDP (GOV) is also considered, as La Porta et al. (1999) found that the size of the state and its composition are key variables in explaining differences in good governance across countries. In addition, a civil liberties index (CIVIL) was included as a proxy of the level of democracy of a country (Bliss and Di Tella, 1997) according to the World Audit Organisation rankings, and was based on a checklist of political rights from 1 to 6,where 1 is taken as not free. A strong democratic regime has been found to enforce the reliability of public action, decrease firms' market power and reduce illegal profits, whereas increases in democracy have been found to decrease corruption (Emerson, 2006; Herzfeld and Weiss, 2003; Treisman, 2000) and, in general, economic growth (Bardhan, 1997; Dreher and Herzfeld, 2005). This implies that, as democracy increases, corruption is weakened, irrespective of its level in a country. The significant influence of democracy on the components of economic freedom is also well-known, e.g., the ability of the legal system to protect property rights and prevent corruption (Lundstrom, 2005). However, estimates should not ignore the existence of endogeneity issues, as lower levels of corruption at country level may be responsible for a reduction in the strength of property rights. 12 12
See Acemoglu and Verdier (1998) for a discussion.
62
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
4. Econometric specification This section describes the multilevel probit model, which is an extension of basic linear and non-linear estimation techniques for applications in which clusters of data result from non-independent observations. Neglecting clustering in the data can lead to bias estimates and invalid inferences. If yij is a binary response for the corruption propensity within firm i for country j, and xij is a set of explanatory variables at firm level, a latent continuous variable, yij ∗, is assumed to underlie yij. Although binary response variable yij is observed directly, this is not the case for yij ∗. We know that yij = 1 if yij ∗ > 0, and yij = 0 if yij ∗ ≤ 0. Thus, we can write the multilevel model for yij ∗ as: yij ¼ β0j þ β1 xij þ
H X βh xi0 þ eij
where eij xij ∼Nð0; πÞ
ð1Þ
h¼1
where β0j is the country-specific intercept and β1 is the regression coefficient of each economic freedom component. As we are interested in assessing separately the influences on corruption of various types of economic freedom, Eq. (1) is considered as a companion matrix which includes economic freedom indicators xij in the diagonal, and is otherwise zero. In addition Eq. (1) includes other firms' covariates, xi0, evaluated by parameters βh; eij is the first-level residual term. Under the hypothesis of a random-effect model, we can write β0j as: 1 ð2Þ β0j ¼ γ00 þ γ01 w0j þ u0j where u0j xij ∼Nð0; ψÞ where γ00 is the intercept and γ01 are the coefficients of the vector of the observed second-stage macro-covariates which identify 1 . u0j is the random effect of two-level model and is related to country-specific intercept β0j. The second-level variations w0j 1 does not avoid the possibility that the assumption that part of the variability can be identified by between-country covariates w0j unobserved variability of country effects may generate dependence between firms' economic freedom components xij. As a consequence, we turn to the specification of an extended model containing parameters associated with issues of endogeneity and implement nested restriction tests. normal distribution. It is Conditional on random effect u0j, a probit model is specified by assuming that eij has standard commonly assumed that cluster j is independent, that the covariance between various firms, Cov eij ei′ j ¼ 0, and that two-level error terms are not correlated, Cov(u0j,eij) = 0, so that we can write the reduced form of the model as 13: 1
yij ¼ γ 00 þ γ 01 w0j þ β1 xij þ
H X βh xi0 þ u0j þ eij :
ð3Þ
h¼1
Assuming that u0j is normally distributed, the strategy for estimating the model parameters is to integrate unobserved random effect u0j: 1 1 f yj xj ; wj ¼ ∫f yj xj ; wj ; u0j g u0j du0j ; ð4Þ where g(.) represents the normal density function. 14 As a result, the unconditional estimate does not lead to a closed expression. Maximum likelihood estimation must resort to approximation procedures, such as numerical integration. RabeHesketh et al. (2002) proposed an algorithm using the posterior mean and variance of random effects, which were calculated by building on the work of Naylor and Smith (1982). 15 If the assumed distribution is normal, the numerical quadrature approach yields a deviance ^ , where lnf y˜ ϑ is the (ϒ) which can easily be used for likelihood ratio tests. This statistic is given as ϒ ¼ 2 ln f y˜ϑ − ln f yϑ ^ log-likelihood of the saturated model and lnf y ϑ is the log-likelihood of the model of interest. Some nested specifications can be obtained by imposing parameter restrictions requiring a simple likelihood ratio test on parameter restrictions, as in: LR ¼ 2 ln f full yij jθ≠0 − ln f restr yij jθ ¼ 0
ð5Þ
which has an approximate χ 2 distribution with a number of degrees of freedom equal to the imposed restrictions on the parameters. As anticipated, the multilevel specification in Eq. (3) may imply that the unexplained variability among different countries' corruption propensity includes the effect of omitted macro-variables related to economic and political institutions which, in turn, may be correlated with xij. 13 The assumption that unobserved country characteristics are not correlated with unobserved firm characteristics excludes non-frequent cases in which firms are very successful in operating in corrupt countries. 14 For the sake of simplicity, we include micro covariates of the first level in intercept parameter γ00. 15 Although marginal quasi-likelihood (MQL) and penalised quasi-likelihood (PQL) are largely used in the statistical literature, they are found to generate downwardly biassed estimates (Hedeker, 2001).
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
63
Fig. 5. Strategy for testing nested and non-nested models.
While the standard approach to handling endogeneity is to estimate parameters by instrumental variables, in the present context, in which regressions are carried out on single economic freedom components, a consistent solution is to estimate the within-unit effects of xij, which can be done by controlling for cluster mean covariates of interest. 16 Following Snijders and Berkhof (2004), the correlation between covariates and random effects may expressed as a regression: 2
w0j ¼ α 00 þ α 01 x⋅j þ ε0j
ð6Þ
where x⋅j is the cluster mean of xij. By inserting Eq. (6) into Eq. (3), the random intercept model now also depends on x⋅j , while the reduced form of the model is given as:
1
yij ¼ γ 00 þ γ 01 w0j þ α 01 x⋅j þ β1 xij þ
H X βh xi0 þ u0j þ eij
ð7Þ
h¼1 ∗ ∗ = γ00 + α00 and u0j = u0j + ε0j. Note that the inclusion of the mean of the covariate of interest can account for reverse where γ00 causality between economic freedom indicators and corruption (see Section 2), since unobservable heterogeneity is mitigated by controlling for within-firm heterogeneity. Excluding from the analysis cluster mean x⋅j , when α01 ≠ 0, yields a biassed estimator of β1. Conversely, when α01 = 0, we obtain the exogenous model of Eq. (3) as a restricted specification. In many cases, however, it is of great interest to model dependence more extensively. Since endogeneity is often the result of one or more unobserved variables, which influence both explanatory and dependent variables in the equation, restricting the model in Eq. (7) by imposing vector γ01 = 0 allows us to identify the importance of the conditional macro-covariates in estimated relationships. We obtain an endogenous random intercept model, which only includes micro-covariates as controls, minimising the effects of unobservable country effects. Formally:
yij ¼ γ 00 þ α 01 x⋅j þ β1 xij þ
H X βh xi0 þ u0j þ eij :
ð8Þ
h¼1
Fig. 5 shows the nested relationships among models and the relative restrictions on the likelihood function of Eq. (7). A sequential strategy of the model selection process can reasonably be implemented by accounting for dependence between micro-covariates and random effects. A double route for testing nested models arises with respect to study aims, because we do not determine a priori whether macro-covariates or unobserved confounding, which includes the cluster of mean covariates as a control, can significantly produce endogeneity. We therefore partition the vector of the variables which should control for dependence across 1 2 ;w0j ]. By assuming that the restrictions of the vector of parameters θ1 = θ|γ01 = 0 and θ2 = θ|α01 = 0 are not rejected levels, W0j = [w0j separately, before passing to the next step for testing the restricted models against the benchmark random intercept model (a basic random intercept model without other micro-covariates except those of interest), we must decide whether a best model exists. By defining the conditional function density for restricted models, f(y|x, θ1) and g(y|x, theta2), conventional and adjusted (Vuong, 1989) h i Þ LR tests were used for these non-nested specifications (step 2, Fig. 5). The null hypothesis of model equivalence, H0: E log gf ððyjx;θ yjx;θ Þ ¼ 0, h i h i Þ f ðyjx;θ Þ was tested against the competing model, H1 : E log gf ððyjx;θ yjx;θ Þ > 0 or H1 : E log g ðyjx;θ Þ b0. If H0 is rejected in the first hypothesis, we prefer 1
2
1
1
2
2
f(.) to g(.), and vice versa if the result is in line with the second hypothesis. The best model is then tested against the basic “random H βhxi0 = 0, if a restricted model with γ01 = 0 or α01 = 0 is found. intercept” model by adding ∑ h=1 5. Results We test the models presented in Section 3 and shown in Fig. 5 to choose the most appropriate specification. Parsimonious models are obtained by testing restrictions on the likelihood function, and may be interpreted as nested cases of endogenous multilevel probit models (7). The conventional and adjusted likelihood ratio formulation is shown in Table 2. 16
The instrumental variables method was developed for multilevel analysis by Kim and Frees (2007).
64
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
Table 2 Random intercept models: specification tests, full sample. Non-market AntiPrivate Employment Sales Prices Financial Financial Property Export dominance and competitive investment regulation regulation regulation constraints obstacles rights and regulation information behaviour regulation contract constraints protection Extended endogenous model (Eq. (7)) vs. restricted endogenous model (Eq. (8)) Extended endogenous model (Eq. (7)) vs. exogenous model (Eq. (3)) Restricted endogenous model (Eq. (8)) vs. basic model Exogenous model (Eq. (3)) vs. basic model Exogenous model (Eq. (3)) vs. restricted endogenous model (Eq. (8))
LR test
1421.671 (0.00)
586.19 (0.00)
811.83 (0.00)
821.0637 (0.00)
809.92 (0.00)
813.35 (0.00)
904.48 (0.00)
932.81 (0.00)
836.65 (0.00)
896.28 (0.00)
LR test
2.58 (022)
2.08 (021)
1.33 (024)
2.21 (0.13)
1.62 (0.20)
2.14 (0.14)
0.04 (0.83)
1.75 (0.18)
2.70 (0.10)
2.20 (0.13)
LR test
406.41 (0.00)
2560.29 (0.00)
3326.89 (0.00)
3495.53 (0.00)
3467.49 (0.00)
3514.36 (0.00)
3603.52 (0.00)
2880.63 (0.00)
3753.51 (0.00)
3295.41 (0.00)
LR test
1809.97 (0.00)
2914.48 (0.00)
4137.39 (0.00)
4314.38 (0.00)
4275.79 (0.00)
4325.57 (0.00)
4507.96 (0.00)
3715.54 (0.00)
4683.63 (0.00)
4189.50 (0.00)
24.677 (0.00)
17.67 (0.00)
18.00 (0.00)
15.47 (4.20)
11.47 (0.00)
8.47 (0.00)
8.05 (0.00)
−3.74 (0.00)
10.17 (0.00)
Vuong's 8.432 test (0.00)
Note: p-Values in brackets. Vuong's (1989) test for non-nested models is carried out under standard normal distribution. Extended endogenous model (Eq. (7)) includes endogeneity, macro- and firm covariates. Restricted endogenous model (Eq. (8)) includes endogeneity and firm covariates. Exogenous model (Eq. (3)) includes macro- and firm covariates.
In the first row, model (7) is tested against model (8), in which macro-covariate restrictions are imposed (i.e. γ01 = 0). The results of the LR test for the ten econometric specifications corresponding to each economic freedom variable clearly reject the hypothesis test, and indicate that these variables are important in identifying corruption differences across countries. Instead, as shown in the second row, endogeneity restriction α01 = 0 is never rejected, because the empirical LR test is always lower than the critical value at the usual percentile. To complete the analysis, we test specification (3) against the basic random effect model, in which firms' covariates and country-identifying variables are restricted to zero (βh = 0 and γ01 = 0). The LR test rejects the restricted basic model, confirming that Eq. (3) is the best model to rationalise the data. Voung's (1989) test completes the selection strategy by evaluating the exogenous random effect model (3) against the model with endogeneity (8) restricted for macro-covariates. Vuong's test statistic leads to the rejection of model equivalence for each specification, selecting model (3). We also evaluate restrictions with respect to the benchmark specification; the test provides further support to the model chosen (Table 2, row five). Maximum-likelihood estimates are listed in Table 3. Note that the use of a multilevel probit instead of a probit specification ensures that misleading significance effects due to violation of the assumption of independent errors with constant variance can be avoided. This effect was confirmed in our regression results, in which the multilevel regressions display lower levels of significance when compared with the restricted probit regression. 17 To support this result, we report the intra-class correlation (ρ) for each estimated specification. On average, for each equation, about 30% of the total variability of the link between the components of corruption and economic freedom is attributable to cross-country differences. To emphasise the interpretative role of multilevel models, we identify this source of variability in each relationship by macro-indicators and although there are considerable differences in the sizes of the macro-covariate coefficients, both their signs and significance are generally consistent with expectations. Accordingly, the degree of economic prosperity (GDP) can explain gains in the efficiency of the economic and social system and control of corruption (Mauro, 1995). In contrast with the results of Smelser (1971), our estimates suggest that more income inequality would reduce corruption. While this result may be thought counter-intuitive, there are several possible explanations. First, there is a composition effect of income inequalities on corruption, as pairwise correlations show that African and transition countries have a negative sign, which instead is positive for Latin American countries. Second, our perception index is closer bureaucratic corruption than the Transparency International index, which mainly accounts for political corruption. Inequalities may favour lobbying politicians rather than bureaucrats, reducing the importance of bureaucratic corruption 18 (see, for example, Harstad and Svensson, 2011; Campos and Giovannoni, 2007). In addition, the importance of the civil liberty index (CIVIL) in reducing corruption is in line with theoretical expectations, whereas an oversized state (GOV) affects the efficiency of expenditure, leading to increasing corruption. As with country characteristics, we find very significant coefficients for control covariates at firm level, except for the legal organisation of firms. The probability of corruptive practices in larger firms is consistently found to be 12–13% lower, irrespective
17
The results of estimates obtained from probit models are available from the authors. In order to support this hypothesis, we take corruption data for our sample of countries from Transparency International and calculate correlations with the Gini index. We find a positive relation for both all and subsample countries. The estimated correlations are available upon request from the authors. 18
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
65
Table 3 Corruption and components of economic freedom, full sample. Non-market AntiPrivate Employment dominance competitive investment regulation and behaviour regulation information Non-market dominance Information constraints
Sales regulation
Prices Financial Financial regulation constraints obstacles
0.012 (0.044) 0.075*** (0.018)
Anti-competitive behaviour Private investment regulation Employment regulation
0.148*** (0.026) 0.084** (0.038) 0.131** (0.038)
Sales regulation
0.081** (0.039)
Prices regulation
0.122** (0.037) −0.018 (0.063)
Financial constraints
−0.113** (0.024)
Financial obstacles
−0.218** (0.035)
Property rights and contract protection Export regulation Firm size Firm sector Legal organisation of firm GDP GINI INV GOV CIVIL Constant ρbenchmark ρid _macro N
Property Export rights and regulation contract constraints protection
−0.113*** (0.035) −0.046* (0.027) −0.013 (0.016) −0.270*** (0.000) −0.024*** (0.007) −0.050*** (0.015) 0.175*** (0.042) −0.097* (0.052) 1.084* (0.654) 0.297** 0.124** 3331
−0.101** (0.040) −0.111*** (0.037) −0.004 (0.018) −0.219*** (0.000) −0.021** (0.009) −0.040** (0.018) 0.105** (0.050) −0.177** (0.070) 1.447* (0.784) 0.244** 0.130** 2568
−0.144** (0.037) −0.056** (0.028) −0.017 (0.016) −0.217*** (0.000) −0.020** (0.009) −0.037** (0.018) 0.173** (0.044) −0.123** (0.058) 0.928 (0.764) 0.286** 0.093** 2959
−0.140** (0.036) −0.057** (0.027) −0.020 (0.016) −0.220*** (0.000) −0.019** (0.009) −0.035** (0.017) 0.172** (0.043) −0.119** (0.057) 0.753 (0.753) 0.289** 0.090** 2989
−0.135** (0.037) −0.053* (0.028) −0.017 (0.016) −0.213*** (0.000) −0.020** (0.009) −0.036** (0.018) 0.174** (0.044) −0.124** (0.058) 0.901 (0.769) 0.288** 0.093** 2962
−0.139** (0.036) −0.054** (0.028) −0.016 (0.016) −0.266*** (0.000) −0.018** (0.009) −0.035** (0.017) 0.170** (0.043) −0.121** (0.057) 0.735 (0.753) 0.282** 0.091** 2974
−0.139** (0.036) −0.051* (0.027) −0.010 (0.017) −0.268*** (0.000) −0.024** (0.007) −0.050** (0.015) 0.182** (0.043) −0.095* (0.052) 1.363** (0.641) 0.328** 0.088** 3132
−0.113** (0.036) −0.045 (0.027) −0.014 (0.016) −0.265*** (0.000) −0.025** (0.008) −0.050** (0.016) 0.174** (0.045) −0.086 (0.055) 1.691** (0.671) 0.315** 0.098** 3128
−0.123** (0.034) −0.048* (0.026) −0.015 (0.016) −0.269*** (0.000) −0.025** (0.007) −0.048** (0.015) 0.165** (0.042) −0.098* (0.051) 1.952** (0.633) 0.311** 0.087** 3410
0.082** (0.025) −0.125** (0.035) −0.058** (0.027) −0.016 (0.016) −0.280*** (0.000) −0.023** (0.008) −0.055** (0.015) 0.184** (0.044) −0.092* (0.054) 1.210* (0.667) 0.317** 0.098** 3192
Notes: Dependent variable is binary index of corruption. Standard errors in brackets; asterisks indicate significant p-value levels (*pb 0.1, **p b 0.05, *** p b 0.01). Two measures of intraclass correlations correspond to benchmark model without firm variables and macro-covariates (ρbenchmark) and to model with restrictions in macro-covariates (ρid _macro).
of which aspect of economic freedom is used. In addition, the industrial sector has a higher propensity (on average 5%) to be subject to corruption, perhaps reflecting more market independence of services and the smaller size of businesses. Government regulations on investment, employment, sales and prices increase corruption (Table 3, columns 3–6), as inefficiencies caused by an over-regulated economic system can distort private productive activity and influence corruptive behaviour. Because of the critical role played by the labour market in the process of economic development, understanding whether labour market regulations actually help or hinder corruption stands out as an important task (Caballero and Hammour, 2000; Foster et al., 2002; Bartelsman et al., 2004). Looking at individual components of government regulations gave very similar results across countries, and it is noticeable that the protection of employment had the highest propensity (0.131) to affect corruption. Adding government employment regulations can weigh down the processes of agreement and facilitate corruptive practices, while mediating among firms, workers and unions. Similarly, the other government regulation indicators were found to affect the reduction of corruption significantly, in line with Svensson (2003). Columns 7 and 8 of Table 3 show the estimated influences of the financial market and illustrate compositional effects for countries with different degrees of financial development. Constraints in private and public financing projects increase the probability of corruption, whereas – rather surprisingly – the efficiency of the financial system does not seem to affect it. Trade regulation, as expected, increase corruption, as found by Ades and Di Tella (1997, 1999), Sung and Chu (2003) and Gerring and Thacker (2005).
66
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
Table 4 Random intercept models: specification tests, subsample of African countries.
Extended endogenous model (Eq. (7)) vs. restricted endogenous model (Eq. (8)) Extended endogenous model (Eq. (7)) vs. exogenous model (Eq. (3)) Restricted endogenous model (Eq. (8)) vs. basic model Exogenous model (Eq. (3)) vs. basic model Exogenous model (Eq. (3)) vs. restricted endogenous model (Eq. (8))
Non-market Private Employment Sales Prices Financial Financial dominance and investment regulation regulation regulation constraint obstacles information regulation constraints
Property Export right and regulation contract protection
LR test
509.25 (0.00)
520.87 (0.00)
517.44 (0.00)
516.18 (0.00)
522.39 (0.00)
527.15 (0.00)
472.14 (0.00)
515.48 (0.00)
499.57 (0.00)
LR test
8.22 (0.01)
0.37 (0.54)
6.58 (0.01)
1.23 (0.26)
0.42 (0.51)
2.80 (0.09)
3.21 (0.07)
6.53 (0.01)
1.61 (0.20)
LR test
152.67 (0.00)
355.28 (0.00)
365.61 (0.00)
349.49 (0.00)
347.02 (0.00)
342.25 (0.00)
351.47 (0.00)
395.47 (0.00)
336.40 (0.00)
LR test
636.33 (0.00)
875.78 (0.00)
876.47 (0.00)
864.45 (0.00)
868.99 (0.00)
866.60 (0.00)
820.41 (0.00)
904.43 (0.00)
834.36 (0.00)
Vuong's test 16.81 (0.00)
6.61 (0.00)
94.35 (0.00)
75.23 (0.00)
33.69 (0.00)
40.38 (0.00)
23.37 (0.00)
12.93 (0.00)
27.73 (0.00)
Notes: p-Values in brackets. Vuong's (1989) test for non-nested models is carried out under standard normal distribution. Extended endogenous model (Eq. (7)) includes endogeneity, macro- and firm covariates. Restricted endogenous model (Eq. (8)) includes endogeneity and firm covariates. Exogenous model (Eq. (3)) includes macro- and firm covariates.
So far, the largest and most significant effect on corruption turns out to be exerted by the legal system protecting property rights and contracts (− 0.218). Limiting the possibility of confiscating private property or repudiating contracts may produce positive externalities and determine general improvements in the quality of institutions. However, this result closely depends on the difficulty of enforcing complex private contracts and on the potential advantages of a parallel developed framework for organising private transactions. There is no evidence that the absence of market dominance leads to fewer corruptive practices (column 1). Conversely, the coefficient on the information constraints on laws and regulations is positive and significant at the 1% level. This indicates that firms in countries where information is effective in checking non-competitive practices have lower rents and lower corruption. This result is also consistent with the significant coefficient of anti-competitive behaviour, although the lack of this variable for Africa excluded it from comparative estimates. Table 5 Random intercept models: specification tests, subsample of transition economies. Non-market AntiPrivate Employment Sales Prices Financial Financial Property Export dominance and competitive investment regulation regulation regulation constraint obstacles rights and regulation information behaviour regulation contract constraints protection Extended endogenous model (Eq. (7)) vs. restricted endogenous model (Eq. (8)) Extended endogenous model (Eq. (7)) vs. exogenous model (Eq. (3)) Restricted endogenous model (Eq. (8)) vs. basic model Exogenous model (Eq. (3)) vs. basic model Exogenous model (Eq. (3)) vs. restricted endogenous model (Eq. (8))
LR test
255.16 (0.00)
233.79 (0.00)
230.42 (0.00)
239.79 (0.00)
244.58 (0.00)
247.68 (0.00)
257.20 (0.00)
241.48 (0.00)
255.47 (0.00)
235.25 (0.00)
LR test
0.20 (0.90)
0.17 (0.67)
0.09 (0.76)
0.15 (0.69)
3.58 (0.05)
3.00 (0.08)
0.97 (0.32)
6.20 (0.01)
0.36 (0.54)
1.87 (0.17)
LR test
277.95 (0.00)
260.58 (0.00)
294.99 (0.00)
298.15 (0.00)
285.59 (0.00)
295.65 (0.00)
294.71 (0.00)
251.08 (0.00)
290.79 (0.00)
271.98 (0.00)
LR test
530.74 (0.00)
492.65 (0.00)
525.32 (0.00)
537.78 (0.00)
526.60 (0.00)
540.33 (0.00)
550.95 (0.00)
486.35 (0.00)
545.90 (0.00)
505.36 (0.00)
Vuong's 63.68 test (0.00)
52.66 (0.00)
84.77 (0.00)
85.76 (0.00)
7642 (0.00)
103.51 (0.00)
24.88 (0.00)
18.72 (0.00)
−39.25 (0.00)
31.44 (0.00)
Notes: p-Values in brackets. Vuong's (1989) test for non-nested models is carried out under standard normal distribution. Extended endogenous model (Eq. (7)) includes endogeneity, macro- and firm covariates. Restricted endogenous model (Eq. (8)) includes endogeneity and firm covariates. Exogenous model (Eq. (3)) includes macro- and firm covariates.
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
67
We tested the robustness of our results by estimating two subsamples of countries for Africa and the transition economies in Eastern Europe, and show that aggregation within the five economic freedom dimensions produces consistent estimates of individual relationships. These are useful tests as the Transparency International index for the years around 2000 shows that the African region gives about 3.35/10, whereas the world mean is about 4.00/10, and the transition countries are interesting because of their move to pass to freer market economies and democratic regimes during the 1990s. With the exception of Estonia, Transparency International rates most former communist countries as being highly corrupt and, although liberalisation policies have been extremely important during the last decade, political crises and economic recessions have inhibited the establishment of a market system with clear and general rules affecting economic legality (Graeff and Mehlkop, 2003). One specific cause of this
Table 6 Corruption and economic freedom components, subsample of African countries. Non-market dominance and information constraints Non-market dominance Mean of competitors Information constraints Mean of information constraints
Private investment regulation
Employment regulation
Sales regulation
Prices regulation
Financial constraint
Financial obstacles
−0.289** (0.104) −0.150 (0.097) 2.489** (0.969)
Employment regulation Mean employment regulation
−0.288** (0.134)
Sales regulation
−0.198* (0.105)
Prices regulation
−0.426** (0.162)
Financial constraints
−0.133** (0.059)
Financial obstacles
−0.217** (0.074) −2.065** (0.814)
Property rights and contract protection Mean of property rights and contract protection Export regulation
Firm sector Legal organisation of firm GDP GINI INV GOV CIVIL Constant ρbenchmark ρid _macro N
Export regulation
0.144 (0.107) 4.685 (2.467) 0.115** (0.041) 0.101** (0.625)
Private investment regulation
Firm size
Property rights and contract protection
−0.078 (0.078) 0.055 (0.046) −0.097* (0.042) 0.102 (0.207) −0.026** (0.024) −0.103* (0.045) 0.126** (0.439) 0.020 (0.083) −10.825 (8.110) 0.200** 0.000** 571
−0.160** (0.075) 0.013 (0.044) −0.088** (0.040) 0.926 (0.460) −0.020 (0.014) −0.057** (0.023) 0.368** (0.111) 0.043 (0.063) 0.258 (1.465) 0.237** 0.000** 590
−0.137* (0.074) 0.027 (0.044) −0.095** (0.040) 0.178 (0.201) −0.028** (0.014) −0.056** (0.023) 0.429** (0.116) 0.042 (0.063) −7.148** (3.119) 0.133** 0.000** 595
−0.115 (0.075) 0.023 (0.044) −0.071* (0.040) 0.144 (0.200) −0.021 (0.014) −0.044* (0.023) 0.387** (0.112) 0.025 (0.064) −0.211 (1.434) 0.226** 0.000** 589
−0.133* (0.074) 0.027 (0.044) −0.074* (0.040) 0.118 (0.204) −0.018 (0.013) −0.044* (0.023) 0.391** (0.112) 0.042 (0.064) −0.656 (1.410) 0.223** 0.000** 593
−0.105 (0.076) 0.040 (0.045) −0.044 (0.042) 0.166 (0.215) −0.028** (0.014) −0.045* (0.024) 0.321** (0.113) −0.004 (0.064) 0.133 (1.448) 0.210** 0.000** 578
−0.104 (0.079) 0.037 (0.046) −0.091** (0.043) 0.071 (0.197) −0.025* (0.014) −0.046* (0.025) 0.393** (0.119) 0.041 (0.069) −0.553 (1.476) 0.234** 0.000** 545
−0.136* (0.074) 0.039 (0.044) −0.097** (0.040) −0.017 (0.200) −0.048** (0.016) 0.034 (0.038) −0.161 (0.207) 0.087 (0.067) 8.018** (3.295) 0.035** 0.000** 612
0.248** (0.059) −0.091 (0.076) 0.039 (0.045) −0.076* (0.041) −0.280*** (0.079) −0.014 (0.014) −0.050** (0.024) 0.352** (0.113) 0.014 (0.066) −1.422 (1.451) 0.225** 0.000** 582
Notes: Dependent variable is binary index of corruption. Standard errors in brackets; asterisks indicate significant p-value levels (*pb 0.1, **p b 0.05, *** p b 0.01). Two measures of intraclass correlations correspond to benchmark model, without firm variables and macro-covariates (ρbenchmark) and to model with restrictions in macro-covariates (ρid _macro).
68
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
Table 7 Corruption and economic freedom components, subsample of transition countries. Non-market AntiPrivate Employment Sales Prices Financial Financial dominance and competitive investment regulation regulation regulation constraint obstacles information behaviour regulation constraints Non-market dominance Information constraints
0.136** (0.063) 0.065*** (0.022)
Anti-competitive behaviour Private investment regulation Employment regulation
0.151*** (0.032) −0.194** (0.058) −0.223** (0.068) −0.223** (0.060) 1.161* (0.614)
Sales regulation Mean of sales regulation
−0.123** (0.053)
Prices regulation
−0.021 (0.075)
Financial constraints
−0.161** (0.031) −0.717** (0.288)
Financial obstacles Mean of financial efficiency Property rights and contract protection Export regulation Firm size Legal organisation of firm GDP GINI INV GOV CIVIL Constant ρbenchmark ρid _macro N
Property Export rights and regulation contract protection
−0.215** (0.051)
−0.085 (0.059) −0.065*** (0.024) −0.116 (0.094) −0.016* (0.009) 0.009 (0.017) 0.140*** (0.023) −0.094* (0.049) −0.589 (0.612) 0.098** 0.00** 1704
−0.118* (0.060) −0.058** (0.025) 0.152 (0.201) −0.017* (0.009) 0.016 (0.018) 0.144*** (0.024) −0.089* (0.052) −0.502 (0.617) 0.098** 0.000** 1569
−0.145** (0.064) −0.077** (0.026) 0.926 (0.460) −0.022** (0.010) 0.006 (0.018) 0.137** (0.024) −0.114** (0.054) 0.981 (0.645) 0.100** 0.000** 1408
−0.148** (0.061) −0.066** (0.025) 0.178 (0.201) −0.019** (0.009) 0.007 (0.017) 0.139** (0.024) −0.100* (0.053) 0.864 (0.618) 0.109** 0.000** 1517
−0.163** (0.063) −0.062** (0.026) 0.144 (0.200) −0.037** (0.015) −0.008 (0.021) 0.149** (0.024) −0.216** (0.085) −1.072 (1.004) 0.107** 0.000** 1522
−0.149** (0.061) −0.068** (0.025) 0.118 (0.204) −0.018* (0.009) 0.013 (0.017) 0.154** (0.024) −0.111** (0.052) 0.345 (0.592) 0.106** 0.000** 1539
−0.121** (0.061) −0.075** (0.025) 0.166 (0.215) −0.016* (0.009) 0.009 (0.017) 0.148** (0.023) −0.103** (0.050) 0.059 (0.607) 0.111** 0.000** 1670
−0.134** (0.064) −0.057** (0.027) 0.071 (0.197) 0.005 (0.013) 0.033* (0.020) 0.193** (0.029) −0.049 (0.059) 0.727 (0.647) 0.103** 0.000** 1432
−0.107* (0.059) −0.062** (0.024) −0.017 (0.200) −0.012 (0.009) 0.019 (0.017) 0.152** (0.023) −0.067 (0.050) −0.121 (0.582) 0.107** 0.000** 1710
0.127** (0.036) −0.118* (0.064) −0.066** (0.026) −0.280*** (0.079) −0.016 (0.010) 0.016 (0.018) 0.142** (0.024) −0.141** (0.053) −0.178 (0.642) 0.107** 0.000** 1429
Notes: Dependent variable is binary index of corruption. Standard errors in brackets; asterisks indicate significant p-value levels (*p b 0.1, **p b 0.05, *** p b 0.01). Two measures of intraclass correlations correspond to benchmark model, without firm variables and macro-covariates (ρbenchmark) and to model with restrictions in macro-covariates (ρid _macro).
failure has been linked with reforms in public sector activities, i.e. privatisation, which did not achieve even one of the objectives of reducing corruption. 19 We repeated the selection strategy for the subsamples shown in Fig. 5. Complete results are shown in Tables 4 and 5. We found support for an exogenous multilevel specification for almost all the corruption equations of these subsamples, except in the cases discussed below, in which the endogeneity of economic freedom indicators linked with corruption was substantial. As discussed in Section 4, to assess this possibility empirically, we included the mean of firms' economic freedom components, which estimates the extended endogeneity random intercept model (Eq. (7)). Tables 6 and 7 show maximum-likelihood estimates. Surprisingly, we find that significant second-level residual variance exceeded 20% in many equations for Africa, showing – as expected – that country variability plays an important role in explaining corruption. This is also the case for transition countries, although the significant variation of the intercept of country level is on 19 The privatisation process in the former USSR, characterised by the sale of state assets, was marked by an increase in corruption, because many traditionalist ideas were upheld and not included in the criteria of the decision-making process of rationality and effectiveness. This offered the opportunity to accumulate illegal fortunes (Sachs, 2005).
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
69
Table 8 Estimates of aggregate dimensions of economic freedom on corruption. Dimensions of economic freedom
Market competition constraints Government regulation Financial system ability
Multilevel model
Probit model
Full sample
African countries
Transition countries
Full sample
African countries
Transition countries
0.214*** (0.44) 0.174*** (0.049) −0.153*** (0.050)
0.258*** (0.052) −0.382*** (0.091) −0.284*** (0.058)
0.243*** (0.074) −0.534*** (0.170) −0.333*** (0.108)
0.216*** (0.042) 0.093*** (0.048) −0.167*** (0.043)
0.293*** (0.501) −0.390*** (0.084) −0.275*** (0.056)
0.218*** (0.057) −0.375*** (0.120) −0.192** (0.088)
Notes: Multilevel model adopts specification described in Eq. (3); probit model includes only country dummy and firm control variables. Asterisks indicate significant p-value levels (*p b 0.1, **p b 0.05, *** p b 0.01).
average slightly more than 10%. Covariates at country level have the expected significant effects in most cases, with some exceptions. Neither GDP for African countries nor INV for transition economies were found to exert any particular influence on corruption, while large differences in the size of government GOV appeared to explain most of the African variability. Its positive effect strengthens the hypothesis that inefficiencies generated by large expenditures, such as the misuse of recurring government budgets, feed corruption. In addition, we anticipate that including the country mean of employment regulation and property rights protection for Africa, and sales regulation and financial constraints for transition economies may cause them to lose their significance. The estimated relationships for these subsamples reveal changes in the components of regulation, suggesting that these interventions are more likely to hinder corruption. Not surprisingly, with respect to the non-standard literature, the subsamples reveal the positive and significant influence of anti-competitive behaviour on corruption, in line with the results of Gupta et al. (2001), in which increases in competition among local firms, which share similar norms and rules, encourages them to pay bribes to enhance profitability. Attractive as this result sounds, we emphasise that the nature of this correlation for African countries, although not excessively significant, may capture the simultaneity bias of corruptive practices. The negative sign on property rights strengthens the perception that institutional rules help to sustain economic activities and reduce corruption in African countries. As expected, the control for endogeneity produces insignificant estimated coefficients for GOV and CIVIL. This implies that not only properly functioning property rights rules but also sustained political interventions are necessary to reduce the problems of corruption and defend the structure of institutional rules, because such rules do not avoid large social costs. However, in the regressions, which account for endogeneity, some micro-relationships show the opposite sign in the estimated coefficient to that obtained using aggregate means. This is evident for government regulation on employment for Africa, as well as for sales in transition economies. Although the costs of corruption for Africa are difficult to reduce by micro-regulation of employment (as shown by the insignificant value of the coefficient), minimisation of corruption should depend on the effectiveness of institutional labour designs, irrespective of the development level of a country. In contrast, transition economies may internalise the advantages of sales regulation in terms of saving on corruption costs by applying policies which incentivise efficient sales regulation. As indicated by the strategy tests, constraints on financial system in transition countries are estimated with the additional country mean regressors. The significance of this parameter, together with that at firm level, reinforces the idea that better financial systems are beneficial for combating corruption and enhancing economic growth. Although this result is generally accepted, when we analyse corruption outcomes for transition economies, they turn out to be more complex, due to interactions with investments in the private sector. It is well-known that, soon after periods of reform, the lack of institutional rules for financial systems in the former communist economies led to increasing investments and simultaneously corruption. The significance of the parameter of private investments (INV) may also explain the existence of detrimental financial system effects on corruption, at least in the short term. A potential source of concern in this literature is the over-dispersion effect of specific economic freedom indicators in explaining corruption. The results may indeed reflect the influence of sub-categories of economic freedom, which specifically affect some aspects of corruption but not others. Here, we present the multilevel and probit estimates of economic freedom indices collected by the weighted mean of estimated components within dimensions. Thus, we aggregate the response categories in Dimensions I, II and III, and further select the best model to rationalise the data in the multilevel regression. Leaving out Dimensions IV and V, which have a single modality, the estimated coefficients of the first three dimensions in Table 8 are consistent – as weighted means – with those reported for disaggregate indicators within each dimension, irrespective of sample and statistical method used. The result of this sensitivity analysis proves the correctness of the association with the economic freedom areas and, in our work, the goodness of our previous conclusions.
6. Concluding remarks Standard economic theory predicts that government intervention transfers resources away from the private sector and provides opportunities for corruption. If economic freedom increases sufficiently, then the level of corruption tends to fall, and continues to fall as the quality of institutions gradually improves. This mechanism has received support from conventional
70
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
cross-country estimates, although (Graeff and Mehlkop, 2003) document that the response of corruption to the components of economic freedom appears to be heterogeneous and sometimes contradictory. In addition, the recent corruption literature has also emphasised the importance of modelling micro-level relationships as determinants of corruption (Mocan, 2008). The purpose of this paper was to complement previous approaches by estimating economic freedom and corruption relationships based on multilevel models. To this end, a vast sample of firm-level data from developing and developed countries was used to estimate corruption/economic freedom in these relationships, modelling unobserved cross-country variability. The results show that cross-country differences account on average for one-third of the economic freedom/corruption link and identify the reasons for these differences in some traditional macro-economic and institutional indicators. By accounting for clusters of countries with common features, we explain why the effects of indicators of market competition may change when estimates are carried out on subsamples which only include developing or transition economies. This suggests that market competition may be bad for corruption when institutions are weak, as they often are in developing countries although, in general, the aggregate index confirms that competition reduces corruption. In addition, we can correctly identify when government regulatory interventions encourage and when they discourage corruption. Although fighting corruption is one of the main objectives of modern society, this paper does suggest that a lack of government regulations may actually yield more corruption in less developed countries, whereas standard recipes for greater freedom can be applied in developed countries. Another novel contribution of this paper is its emphasis on the important role played by the political economy of countries and institutions on different aspects of economic freedom, which complements the effects of restricting market competition and government regulation on corruption. These results are consistent with theories of corruption which argue that complementarities exist between economic development and the quality of institutions and, in particular, with the nature of the African and transition economy samples in which, counter-intuitively, high levels of deregulation lead to more corruption. In the context of less developed economies, in economic and institutional terms, the costs of corruption resulting from greater regulation (through bureaucratic powers) seem to be less important than those related to the transition costs for firms which try to escape by paying bribes and thus cause even more corruption. Our model also links financial systems and property rights, which are considered positive mechanisms in combating corruption. The results suggest that these components of economic freedom reduce corruption and they do show high and significant variability across countries. The inclusion of identifying country effects makes the relationship more robust and confirms that improved civil liberties and macro-economic indicators lead to increased efficiency in fighting corruption. The implications for policy-makers in less developed countries and transition economies seem to be clear, although powerful interests, distant from theoretical discussions concerning the trade-off between government intervention, have persuaded several governments to take no action at all, making them almost unresponsive to the need to regulate some sectors. Policies should be introduced to implement complementary strategies to reduce corruption and the costs of economic growth, using suitable government interventions within appropriate sectors. Acknowledgements We would like to thank Paul J. Dunne, for helpful comments. Moreover, we are grateful to the anonymous referees for numerous helpful suggestions. Appendix 1. List of countries by macro regions Country Albania Argentina Bangladesh Belize Bolivia Botswana Brazil Bulgaria Cameroon Canada Chile China Colombia Costa Rica Cote d'Ivoire Croatia Czech Republic Dominican Republic Ecuador Egypt El Salvador Estonia
Africa 0 0 0 0 0 101 0 0 57 0 0 0 0 0 97 0 0 0 0 0 0 0
MENA-region 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 102 0 0
Transition economies 163 0 0 0 0 0 0 125 0 0 0 0 0 0 0 127 137 0 0 0 0 132
East Asia 0 0 0 0 0 0 0 0 0 0 0 101 0 0 0 0 0 0 0 0 0 0
South Asia 0 0 50 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
Latin America 0 100 0 50 100 0 201 0 0 0 100 0 101 100 0 0 0 111 100 0 104 0
OECD 0 0 0 0 0 0 0 0 0 101 0 0 0 0 0 0 0 0 0 0 0 0
Total 163 100 50 50 100 101 201 125 57 101 100 101 101 100 97 127 137 111 100 102 104 132
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
71
Appendix 1 (continued) Country
Africa
MENA-region
Transition economies
East Asia
South Asia
Latin America
OECD
Total
France Germany Ghana Guatemala Haiti Honduras Hungary India Indonesia Italy Kenya Lithuania Madagascar Malawi Malaysia Mexico Namibia Nicaragua Nigeria Pakistan Panama Peru Philippines Poland Portugal Romania Russian Federation Senegal Singapore Slovak Republic Slovenia South Africa Spain Tanzania Thailand Trinidad and Tobago Tunisia Turkey Uganda Ukraine United Kingdom United States Venezuela Zambia Zimbabwe Total
0 0 119 0 0 0 0 0 0 0 113 0 116 55 0 0 95 0 93 0 0 0 0 0 0 0 0 124 0 0 0 121 0 83 0 0 0 0 137 0 0 0 0 84 129 1524
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 52 0 0 0 0 0 0 0 0 154
0 0 0 0 0 0 129 0 0 0 0 112 0 0 0 0 0 0 0 0 0 0 0 225 0 125 525 0 0 129 125 0 0 0 0 0 0 150 0 225 0 0 0 0 0 2429
0 0 0 0 0 0 0 0 100 0 0 0 0 0 100 0 0 0 0 0 0 0 100 0 0 0 0 0 100 0 0 0 0 0 422 0 0 0 0 0 0 0 0 0 0 923
0 0 0 0 0 0 0 210 0 0 0 0 0 0 0 0 0 0 0 103 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 363
0 0 0 106 103 100 0 0 0 0 0 0 0 0 0 100 0 100 0 0 100 108 0 0 0 0 0 0 0 0 0 0 0 0 0 101 0 0 0 0 0 0 100 0 0 1985
100 100 0 0 0 0 0 0 0 100 0 0 0 0 0 0 0 0 0 0 0 0 0 0 100 0 0 0 0 0 0 0 104 0 0 0 0 0 0 0 102 100 0 0 0 807
100 100 119 106 103 100 129 210 100 100 113 112 116 55 100 100 95 100 93 103 100 108 100 225 100 125 525 124 100 129 125 121 104 83 422 101 52 150 137 225 102 100 100 84 129 8185
References Acemoglu, D., Johnson, S., Robinson, A.J., 2001. The Colonial Origins of Comparative Development: An Empirical Investigation. American Economic Review 91, 1369–1401. Acemoglu, D., Verdier, T., 1998. Property rights, corruption and the allocation of talent: a general equilibrium approach. The Economic Journal 108, 1381–1403. Acemoglu, D., Verdier, T., 2000. The choice between market failures and corruption. American Economic Review 90, 194–211. Ades, A., Di Tella, R., 1997. The new economics of corruption: a survey and some new results. Political Studies 45, 496–515. Ades, A., Di Tella, R., 1999. Rents, competition, and corruption. American Economic Review 89, 982–993. Adsera, A., 2003. Are you being served? Political accountability and quality of government. Journal of Law, Economics and Organization 19, 445–490. Ahlin, C., Pang, J., 2008. Are financial development and corruption control substitutes in promoting growth? Journal of Development Economics 86, 414–433. Apergis, N., Dincer, O., Payne, J., 2011. Live free or bribe: on the causal dynamics between economic freedom and corruption in U.S. states. European Journal of Political Economy 28, 215–226. Bardhan, P., 1997. Corruption and development: a review of issues. Journal of Economic Literature 35, 1320–1346. Bartelsman, E.J., Haltiwanger, J., Scarpetta, S., 2004. Microeconomic evidence of creative destruction in industrial and developing countries. Tinbergen Institute Discussion Papers 04-114/3. Tinbergen Institute. Beck, T., Demirguc-Kunt, A., Maksimovic, V., 2002. Financial and legal constraints to firm growth — does size matter? Policy Research Working Paper Series 2784. The World Bank. Billger, S.M., Goel, R.K., 2009. Do existing corruption levels matter in controlling corruption? Cross-country quantile regression estimates. Journal of Development Economics 90, 299–305. Bliss, C., Di Tella, R., 1997. Does competition kill corruption? Journal of Political Economy 105, 1001–1023. Brunetti, A., Kisunko, G., Weder, B., 1997. Institutional obstacles to doing business: region-by-region results from a worldwide survey of the private sector. Policy Research Working Paper Series 1759. The World Bank.
72
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
Caballero, R.J., Hammour, M.L., 2000. Creative destruction and development: institutions, crises, and restructuring. NBER Working Papers, 7849. National Bureau of Economic Research. Campos, N., Giovannoni, F., 2007. Lobbying, corruption and political influence. Public Choice 131, 1–21. Carden, A., Verdon, L.L., 2010. When is corruption a substitute for economic freedom? Law and Development Review 3, 40–63. Chafuen, A.A., Guzmàn, E., 2000. Economic freedom and corruption. Index of Economic Freedom. The Heritage Foundation, Washington, pp. 51–63. Compton, R.A., Giedeman, D.C., Hoover, G.A., 2011. Panel evidence on economic freedom and growth in the United States. European Journal of Political Economy 27, 423–435. de Haan, J., Sturm, J.E., 2003. Does more democracy lead to greater economic freedom? New evidence for developing countries. European Journal of Political Economy 19, 547–563. de Jong, E., Bogmans, C., 2011. Does corruption discourage international trade? European Journal of Political Economy 27, 385–398. Dreher, A., Herzfeld, T., 2005. The economic costs of corruption: a survey and new evidence. Public Economics 0506001, EconWPA. Elliot, K.A., 1997. Corruption as an International Policy Problem: Overview and Recommendations. Institute for International Economics, Washington DC, pp. 175–223. Emerson, P.M., 2006. Corruption, competition and democracy. Journal of Development Economics 81, 193–212. Faber, G., Gerritse, M., 2012. Foreign determinants of local institutions: Spatial dependence and openness. European Journal of Political Economy 28, 54–63. Foster, L., Haltiwanger, J., Krizan, C.J., 2002. The link between aggregate and micro productivity growth: evidence from retail trade. NBER Working Papers, 9120. National Bureau of Economic Research. Gerring, J., Thacker, S.C., 2005. Do neoliberal policies deter political corruption? International Organization 59, 233–254. Gorodnichenko, Y., Sabirianova, P.K., 2007. Public sector pay and corruption: measuring bribery from micro data. Journal of Public Economics 91, 963–991. Graeff, P., Mehlkop, G., 2003. The impact of economic freedom on corruption: different patterns for rich and poor countries. European Journal of Political Economy 19, 605–620. Gupta, S., de Mello, L., Sharan, R., 2001. Corruption and military spending. European Journal of Political Economy 17, 749–777. Gwartney, J., Lawson, R., 2000. Economic Freedom of the World 2000, Annual Report. The Fraser Institute, Vancouver. Gwartney, J., Lawson, R., 2002. Economic Freedom of the World 2002, Annual Report. The Fraser Institute, Vancouver. Gwartney, J., Lawson, R., 2003. The concept and measurement of economic freedom. European Journal of Political Economy 19, 405–430. Harstad, B., Svensson, J., 2011. Bribes, lobbying and development. American Political Science Review 105, 46–63. Hedeker, D., 2001. Multinomial models for ordinal and categorical variables. Handbook of Multilevel Analysis. Springer. Herzfeld, T., Weiss, C., 2003. Corruption and legal (in)effectiveness: an empirical investigation. European Journal of Political Economy 19, 621–632. Hodgson, G.M., 2003. The hidden persuaders: institutions and individuals in economic theory. Cambridge Journal of Economics 27, 159–175. Hodgson, G.M., 2006. Institutions, recessions and recovery in the transitional economies. Journal of Economic Issues 90, 1043–1061. Hodgson, G., Jiang, S., 2007. The economics of corruption and the corruption of economics: an institutionalist perspective. Journal of Economic Issues 91, 1043–1061. Kim, J., Frees, E., 2007. Multilevel modeling with correlated effects. Psychometrika 72, 505–533. Kreft, I., de Leeuw, J., 1998. Introducing Multilevel Modeling. Sage, Thousand Oaks, CA. Krueger, A.O., 1974. The political economy of the rent-seeking society. American Economic Review 64, 291–303. La Porta, R., Lopez-de Silanes, F., Shleifer, A., Vishny, R., 1999. The quality of government. Journal of Law, Economics and Organization 15, 222–279. La Porta, R., Lopez-de-Silanes, F., Shleifer, A., 2008. The Economic Consequences of Legal Origins. Journal of Economic Literature 46, 285–332. Lambsdorff, J.G., 2007. The Institutional Economics of Corruption and Reform: Theory, Evidence and Policy. Cambridge University Press, Cambridge. Lundstrom, S., 2005. The effect of democracy on different categories of economic freedom. European Journal of Political Economy 21, 967–980. Mauro, P., 1995. Corruption and growth. Quarterly Journal of Economics 110, 681–712. Méon, P.G., Sekkat, K., 2005. Does corruption grease or sand the wheels of growth? Public Choice 122, 69–97. Milgrom, P., Roberts, J., 2008. Motivation: contracts, information and incentives. Economics, Organization and Management. Prentice Hall. Mocan, N., 2008. What determines corruption? International evidence from microdata. Economic Inquiry 46, 493–510. Naylor, J.C., Smith, F.M.A., 1982. Applications of a method for the efficient computation of posterior distributions. Applied Statistics 31, 214–225. North, D.C., Thomas, R.P., 1973. The Rise of the Western World: A New Economic History. Cambridge University Press, Cambridge. Cambridge. Paldam, M., 2002a. The cross-country pattern of corruption: economics, culture and the seesaw dynamics. European Journal of Political Economy 18, 215–240. Paldam, M., 2002b. The big pattern of corruption, economics, culture and the seesaw dynamics. Economics Working Papers 1999–11. School of Economics and Management, University of Aarhus. RabeHesketh, S., Skrondal, A., Pickles, A., 2002. Reliable estimation of generalized linear mixed models using adaptive quadrature. The Stata Journal 2, 1–21. Sachs, J., 2005. The End of Poverty: How We Can Make It Happen in Our Lifetime. Penguin Books. Shen, C., Williamson, J.B., 2005. Corruption, democracy, economic freedom, and state strength: a cross-national analysis. International Journal of Comparative Sociology 46, 327–345. Shleifer, A., Vishny, R.W., 1993. Corruption. Quarterly Journal of Economics 108, 599–617. Smelser, N.J., 1971. Stability, Instability and the Analysis of Political Corruption. Little Brown, Boston, pp. 7–29. Snijders, T.A.B., Berkhof, J., 2004. Diagnostic checks for multilevel models. Handbook of Quantitative Multilevel Analysis. Sage, Thousand Oaks, CA. Sung, H., Chu, D., 2003. Does participation in the global economy reduce political corruption? An empirical inquiry. International Journal of Comparative Criminology 3, 94–118. Svensson, J., 2003. Who Must Pay Bribes And How Much? Evidence From A Cross Section Of Firms. The Quarterly Journal of Economics 118, 207–230. Svensson, J., 2005. Eight questions about corruption. Journal of Economic Perspectives 19, 19–42. Swamy, A., Knack, S., Lee, Y., Azfar, O., 2001. Gender and corruption. Center for Development Economics, 158. Department of Economics, Williams College. Torrez, J., 2002. The effect of openness on corruption. Journal of International Trade & Economic Development 11, 387–403. Treisman, D., 2000. The causes of corruption: a cross-national study. Journal of Public Economics 76, 399–457. Vuong, Q.H., 1989. Likelihood ratio tests for model selection and non-nested hypotheses. Econometrica 57, 307–333.
Contents lists available at SciVerse ScienceDirect
European Journal of Political Economy journal homepage: www.elsevier.com/locate/ejpe
Corruption and the effects of economic freedom L. Pieroni a,⁎, G. d'Agostino b a b
University of Perugia, Via Pascoli, 20‐06123 Perugia, Italy University of Roma Tre, via Silvio D'Amico 77, 00145 Roma, Italy
a r t i c l e
i n f o
Article history: Received 6 September 2011 Received in revised form 24 August 2012 Accepted 26 August 2012 Available online 31 August 2012 JEL classification: H11 H50 K20 Keywords: Corruption Economic freedom Multilevel models
a b s t r a c t The prediction that economic freedom is beneficial in reducing corruption has not been found to be universally robust in empirical studies. The present work reviews this relationship by using firms' data in a cross-country survey and argues that approaches using aggregated macro data have not been able to explain it appropriately. We model cross-country variations of the microfounded economic freedom–corruption relationship using multilevel models. Additionally, we analyse this relationship by disentangling the determinants for several components of economic freedom because not all areas affect corruption equally. The results show that the extent of the macro-effects on the measures of (micro)economic freedom for corruption, identified by the degree of economic development of a country, can explain why a lack of competition policies and government regulations may yield more corruption. Estimations for Africa and transition economy subsamples confirm our conjectures. © 2012 Elsevier B.V. All rights reserved.
1. Introduction Much attention has recently been devoted to testing the relationship between economic freedom and corruption, with the expectation that economic freedom leads to reduced levels of corruption. 1 While empirical studies do tend to support this expectation (see Chafuen and Guzmàn, 2000; Paldam, 2002b; Graeff and Mehlkop, 2003; Shen and Williamson, 2005; Carden and Verdon, 2010), there has been some inconsistency in the results. Certainly, economic freedom is found to reduce corruption when it is regressed on the entire distribution of corruption subdivided by its components, and when the relationship is tested for subsamples of countries. However, Billger and Goel (2009) found that, among the most corrupt nations, greater economic freedom does not reduce corruption and may even exacerbate it, indicating that nations respond differently to levels of economic freedom, depending on their level or stage of development conditions. 2 In addition, Graeff and Mehlkop (2003), while arguing against the use of an aggregate indicator of economic freedom to evaluate their effects, find support for the counter-intuitive influence of the size of government on corruption, as a specific component of economic freedom. The literature has also invoked an institutional dimension to explain some unexpected results. Lambsdorff (2007) argued that government intervention may not be sufficient if it aims only to avoid market failures, and that “good” government regulations may be important in reducing corruption. Thus, the effect of aspects of economic freedom on corruption may depend on how efficient government intervention is in countries which are more highly “regulated” or “have high levels of economic freedom”. ⁎ Corresponding author at: University of Perugia, Department of Political Science, Italy. Tel.: +39 0755855280; fax: +39 0755855299. E-mail addresses: [email protected] (L. Pieroni), [email protected] (G. d'Agostino). 1 Berggren (2003) defines economic freedom as “a composite that attempts to characterise the degree to which an economy is a market economy, that is, the degree to which it entails the possibility to enter into voluntary contracts within the framework of a stable and predictable rule of law that upholds contracts and protects private property, with a limited degree of interventionism in the form of government ownership, regulations and taxes”. 2 Lambsdorff (2007) provides a review of studies in which the relationship between economic freedom and corruption is insignificant or in which the effects of economic freedom increase corruption. 0176-2680/$ – see front matter © 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.ejpoleco.2012.08.002
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
55
This paper, in contrast with most of the existing research focusing on cross-country studies, reviews the relationship between economic freedom and corruption using firm level data, following the approach originally developed by Milgrom and Roberts (2008). It also considers macro-effects, such as the level of economic development and the quality of institutions (which are known to provide a fertile breeding ground for corruption), by means of a multilevel modelling framework. This mitigates the mismatch between macro- and micro-evidence, with the aim of keeping micro-relationships robust and identifying the socio-economic and institutional factors which are responsible for cross-country differences. It thus contributes to empirical work investigating whether more freedom lowers corruption (Chafuen and Guzmàn, 2000; Paldam, 2002a,b; Apergis et al., 2011), although it follows from Graeff and Mehlkop (2003) in considering how different aspects of economic freedom affect corruptive practices. This paper is also influenced by the extensive literature which considers how corruption can undermine the strength of public institutions and hamper economic growth and development (and vice versa), including Shleifer and Vishny (1993), Mauro (1995), Bardhan (1997), Treisman (2000) and Méon and Sekkat (2005), the more limited empirical analyses which use micro-data to investigate corruption and its determinants (Swamy et al., 2001; Svensson, 2005; Mocan, 2008) and the work of Hodgson (2006) and Hodgson and Jiang (2007), who consider the role of institutions in affecting corruption. We begin our work by documenting basic facts regarding the key variables of our empirical model and discuss the conceptual framework to explore why some dimensions of economic freedom are likely to affect corruption at firm level. We then provide information concerning data extracted from the World Business Environment Survey (WBES) and how we derive variables for empirical analysis. Section 4 presents multilevel models, a more realistic framework in which to test the hypotheses of our work. Estimating the micro-economic links in a multilevel model allows us to assume that the unexplained, insignificant or mitigated relationships of corruption with the components of economic freedom may be explained by differences in country characteristics. Thus, unlike models, which include conditioning variables to explain why the effect of economic freedom on corruption is not linear, we specify the broadest class of multilevel models and propose a selective strategy in which nested and non-nested multilevel models are tested. The rest of the paper estimates the parameters of these models. Although the results in the full sample are sufficiently in line with standard economic predictions, attention to firms' heterogeneous responses and cross-country effects is crucial in explaining empirical variations of outcomes. The results highlight the fact that government regulations in African countries and transition economies may reduce corruption although, as suspected, the large variability across countries is identified not only by differences in economic prosperity but also by quality institutions. 2. Economic freedom and links with corruption When we aim at measuring corruption in empirical studies, we find a number of possible indicators. The highest profile measure is provided by Transparency International, which uses survey methods to produce the corruption perception index (CPI). Although such perception-based indices may not provide robust estimates of corruption within countries, they do remain informative when used for cross-country comparisons across countries. 3 The economic freedom index published by the Economic Freedom of the World (EFW) is designed to measure the quality of a country's institutions and policies (Gwartney and Lawson, 2000, 2003) and to affect the effectiveness of individual incentives, productive effort, and the effectiveness of resource allocation (North and Thomas, 1973; de Haan and Sturm, 2003). The key components of the index are ‘objective’ measures, such as government consumption as a share of total consumption, and transfers and subsidies as a share of GDP, which are collected from external sources (International Monetary Fund, World Bank and World Economic Forum), and ‘subjective’ measures based on surveys, expert panels and case studies. 4 The scores for each country are then given for five areas of economic freedom: i) size of government; ii) legal structure and security of property rights; iii) access to finances and sound money; iv) freedom to trade internationally, and v) regulation of credit, labour, and business. The mean scores give the aggregate EFW index, which ranges from 1 to 10, where 10 is the maximum degree of economic freedom. The EFW index shows that economic freedom has improved over time, as the average value has increased from 5.8 in 1990 to 6.6 in 2000, and then rises in the first decade of the 21st century. For some idea of the probable relation with corruption, Fig. 1 shows the EFW index and the CPI corruption index (measured inversely in terms of good governance) across world countries for the year 2000, and illustrates the clear positive association between the indices, implying that higher economic freedom is associated with lower corruption. 5 While cross-country regressions of economic freedom and corruption give results in line with the prevailing view, that corruption is reduced by an increase in economic freedom, this may hide heterogeneity at sectoral level, particularly in the degree and nature of government intervention. Graeff and Mehlkop (2003) looking at the components of the economic freedom index find a more complex picture, with ambiguous correlations among the components of economic freedom and corruption. This suggests that the weight of the components of the aggregate index influences the effect on corruption, implying that the results are sample-sensitive. In addition, Lambsdorff (2007) argues that corruption may also arise in sectors with greater economic freedom, as not all aspects of economic freedom deter corruption. Some government regulations may have the effect of increasing
3
For a discussion on this topic see, for example, Gorodnichenko and Sabirianova (2007). The number of variables used to construct this index is not constant over time. For example, 21 components of economic freedom were considered in 2005, and 23 in 2008. 5 For an application of the association between economic freedom and growth, see de Jong and Bogmans (2011). 4
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
4
6
R2=0.58
2
Corruption index
8
10
56
4
5
6
7
8
9
Economic Freedom index Fig. 1. Cross-country relationship between economic freedom and (less) corruption.
the transaction costs of corruption and in such cases, policy-makers are unresponsive to the demand for regulation, and “free” competition and lack of government regulations would lead to a failure of anti-corruption policy. When combined with weak legal apparatus and enforcement mechanisms, as is often the case in less developed countries or transition economies, corruption may become rife, as the spontaneous mechanisms of economic freedom are conditioned by local rules which impose private bribes, frequently under the guise of taxation. 2.1. Conceptual framework We now discuss in detail which dimensions of economic freedom are likely to affect corruption at firm level. We do not include the size of government, a dimension which naturally affects corruption at aggregate level, but do include market competition, expected positively to affect the probability of corruption of a firm. Here, we list the five dimensions of economic freedom assumed to be related to corrupt activities. 2.1.1. Dimension I: market competition This paper focuses on the micro-dimensions which refer to market competition. Non-competitive markets may exhibit rent-seeking behaviour, bureaucrats acting independently of market forces, and constraints on the flow of information, all limiting competition and encouraging corruption beyond firms' control. While modelling the interaction between corrupt government officials and industrial firms, Emerson (2006) shows corruption to be antithetical to competition, and the competition–corruption link may easily suffer from reverse causality or sometimes even have a positive effect (Lambsdorff, 2007). The latter author considers that the prospect of money from corruptive practices may motivate private firms to pay bribes, thus creating endogeneity in the relationship; if quality-based rather than price-based competition prevails in the market, firms may be driven by myopic behaviour in delivering substandard quality products, meaning that corruption hampers competition and, in general, international trade. 6 The literature also uses proxies of (non)-market dominance, such as the effectiveness of non-competitive practices and the degree of information available on laws and regulations, as subjective measures of obstacles to the growth of firms' business. But further endogeneity problems may arise (Lambsdorff, 2007). 2.1.2. Dimension II: government regulation of private entrepreneurial activity Corruption is assumed to be influenced by aspects of government regulation, although it would be wrong to perceive corruption as the consequence of excessive regulation, or to assume that complete laissez-faire will always be the answer (Bliss and Di Tella, 1997). The view that government regulation hampers productive effort, encourages rent-seeking and increases the discretionary power of a few public officials still reigns throughout government, institutions and the literature (e.g., Graeff and Mehlkop, 2003; Paldam, 2002a). However, when government regulation is weak or almost absent, an increase in market rules is still recognised as crucial in developing a solid productive sector (Hodgson, 2003). Developing countries are often characterised by weak law enforcement, a large informal sector, underdeveloped capital markets, and informal credit and insurance networks. 7 This means that results from developed countries should not be directly extended to those from developing countries without serious reflection on their differences, both the trade-off between the benefits and disadvantages of state intervention in dealing with market failures (Acemoglu and Verdier, 2000) and the different context regarding the negative externalities of corruption.
6
For a discussion, see Compton et al. (2011). For example, informal arrangements, such as family networks of credit and insurance, have been found to influence the consequences of interventions to a great extent, limiting the onset of corruption. 7
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
57
Second, and even more importantly, empirical research has tended to focus on the effect of the overall size of the government budget relative to GDP, often neglecting the interactions of the various components of government regulation. 8 It also suggests that particular types of government intervention may lead to corruption. In an environment in which heterogeneous goods and services may be supplied in free markets at high transaction costs, which may be reduced by government regulations, government quantity and price interventions may later lead to serious problems of corruption. Indeed, Svensson (2003) finds that firms typically have to pay bribes when dealing with public officials whose actions directly affect firms' business, particularly their investments. 2.1.3. Dimension III: ability of the financial system to support private firms It has been postulated that the ability of a formal financial system to provide financing to the private sector reduces the effects of corruption. This relationship may be mitigated by the different quality of financial institutions across countries and, in the extreme case, the quality of institutions may reflect the level of corruption. Brunetti et al. (1997) found that the ability of the financial system reduced corruption, by ranking levels of corruption from a worldwide survey and observing that the second most significant impediment to doing business was lack of financing. However, even when the financial system provides adequate finance for the private sector, inefficiencies in business investments can lead to higher intermediation costs, which are associated with corruption payments (Ahlin and Pang, 2008). 2.1.4. Dimension IV: property rights and the protection of contracts The ability of the legal system to protect property rights is widely believed to reduce corruption, whereas the failure of the legal system to enforce contracts undermines the free market and reduces incentives for agents to participate in productive activities (Acemoglu and Verdier, 1998), leading to increases in corruption. 2.1.5. Dimension V: regulation of exports Government regulations on exports and international trade are usually assumed to increase the level of corruption and, there is a fair amount of empirical evidence to back this up. There are three main explanations. One is the relation between the rents of firms in a non-competitive market and import licensing Krueger (1974). When the number of licences is fixed and firms are encouraged to compete, a rational firm may shift productive plans to rent intensive activities and may then turn to bribery. Another explanation is that trade barriers may favour inefficient local firms rather than more efficient foreign ones and this may lead to corruptive practices Faber and Gerritse (2012). Thirdly, import barriers may create an artificial scarcity of specific commodities, channelling some of the non-competitive higher prices towards corrupt bureaucrats, a practice which, in developing countries or transition economies, may lead to underground economies. However, the empirical evidence of a negative sign in the relation between the extension of international trade and corruption is questionable (Treisman, 2000; Torrez, 2002). In general, it is important to recognise that the micro-economic relationship between aspects of economic freedom and corruption may be affected by factors that operate at more macro- or even country level. For example, economic development is argued to be negatively affected by corruption and vice versa (Mauro, 1995), whereas the nature and quality of institutions within a country can have a direct effect on development and corruption (Acemoglu et al. 2001). Thus, an increase in the quality of institutions would reduce the propensity to corruption. The aim of our empirical framework is to include these macro-level effects in investigating the relationship between economic freedom and corruption. 3. Data and variables In this paper, the data are mainly taken from the WBES cross-sectional survey of industrial and service enterprises ‘Voices of the Firms 2000’, conducted in mid-1999 by the World Bank and other agencies. This WBES survey, the most comprehensive source of micro-data for analysing both corruption and economic freedom by firms' perception responses, covers 67 countries with, on average, more than 100 firms per country. Appendix 1 lists the countries by region and the number of firms interviewed. The survey provides detailed information on taxation, government regulation, the financial sector, and perceptions of corruption in business (based on the opinions of firms' managers). Responses to questions on corruption include not only direct experience, but also perceptions of the behaviour of neighbouring firms in the same environment. Firms' responses to questions concerning corruption were grouped and values were assigned to them. Thus, in response to questions such as “How usual is it for firms to have to pay some kind of irregular extra payment in order to obtain services from the public administration?”, 6 indicates “always” and 1 “never”. Lower corruption was then defined as values from 1 to 3 inclusive, and higher corruption from 4 to 6. Gorodnichenko and Sabirianova (2007) suggested that this binary categorisation may lead to a better distinction between higher and lower degrees of corruption. It may also be explained by looking at the empirical distribution of firms' responses to the survey question. As Fig. 2 shows, the responses were not evenly distributed, but asymmetrically moved distribution towards higher values, meaning that intermediate values make it difficult to distinguish between higher or lower degrees of corruption, and thus makes the relationship with economic freedom more difficult to interpret from the policy-makers' point of view. 8 The test that size of government is positively correlated with level of corruption is weak: Elliot (1997) and Adsera (2003) obtained contrasting findings, and Graeff and Mehlkop (2003) and Billger and Goel (2009) reported ambiguous results, which could only partially interpreted by observing the relation in a subsample of countries, or conditioning the distribution of corruption variables across countries, respectively.
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
1.5 1 0
.5
Density
2
2.5
58
1
2
3
4
5
6
Corruption index Fig. 2. Corruption index by response category. Note: Aggregate score responses of category of corruptive practices from 1 to 6: 1 (never), 2 (seldom), 3 (sometimes), 4 (frequently), 5 (usually) and 6 (always).
With regard to economic freedom, the absence of market dominance was measured by considering the number of enterprises and the existence of anti-competitive practices. For the former, a scale from 1 to 3 was used, 3 implying a low degree of market dominance; anti-competitive practices were graded from 1 to 4, 1 indicating none. Following Ades and Di Tella (1999), an indicator of the degree of market competition was constructed, based on firms' ability to obtain information on laws and regulations, ranging from 6 if the firm respondent disagreed that information was easy to obtain, and 1 otherwise. To compare the areas of Gwartney and Lawson (2002) with our micro-dimensions of economic freedom, Dimensions II and III above are equivalent to Gwartney and Lawson's (2002) fifth area (regulation of credit, labour, and business). In Dimension II, respondents answer the question “How often does the government intervene … in decisions by your firm?”, and indicators of government regulation in private investment, employment sales and prices were used to measure the magnitude of this intervention. For all four indices, firms were graded on a scale from 1 to 6, where 6 represents government intervention in private activities. Dimension III includes indicators of financial constraint and obstacles. For constraints, a binary variable measuring the lack of financing in the private sector is used, whereas the nature of financial obstacles is ranked from 1 to 4, where 4 represents major obstacles. In line with the legal structure and security of property rights dimension in Gwartney and Lawson (2002), an index was constructed which measures the ability of the legal system to protect “property rights and contracts”, where the maximum 3 is a firm's perception that it is completely able and 1 that it is unable. Lastly, Dimension V uses an indicator for export regulation in line with the Freedom to Exchange with Foreigners, although a functional classification based on government intervention might also include this in Dimension II. Table 1 presents the complete description of the corruption index and the micro-covariates of economic freedom. Figs. 3 and 4 show the bivariate relationships between the components of economic freedom and corruption at country level, after aggregating firms' responses. Aggregate-level data were created to allow comparisons with the usual empirical analyses in the literature. 9 Unlike the scale of the Transparency International index, corruption increases along the y-axis, whereas economic freedom decreases right of the x-axis. In many of the panels, a clear-cut picture of the statistical significance between economic freedom indices and corruption cannot be defined. For non-market dominance, information constraints (Fig. 3, panels a and c), components of government regulation (Fig. 3, panels d–e; Fig. 4, panels a–b) and property rights and contract protection (Fig. 4, panels d), the patterns do not support the relationship suggested by standard theory, i.e., that more freedom reduces corruption. The graphs also clearly highlight the existence of clusters of countries, suggesting differences in the effects of the components of economic freedom on corruption across groups of countries. A multilevel modelling framework was adopted to explain these clusters assuming that, on average, firms' responses differ across countries. Without losing generality, this framework may be extended to the remaining economic freedom components, i.e., the absence of anti-competitive behaviour, more freedom in the financial system and export regulation, although in aggregate they consistently show that more freedom reduces corruption (Fig. 3, panel b; Fig. 4, panels c and e). This means that the possibility that unexplained variability at micro-level may result in different or non-significant responses in the relationship at macro-level. Note that, in line with the above classification, the impact on corruption of property rights and the protection of contracts at micro-level is modelled. 10 9 Our aim was to give some descriptive insights of the relation to be tested. This is statistically equivalent to aggregating all individual level variables to group level and carrying out ordinary least squares, e.g., regression over group means. Criticism of this procedure is discussed in Kreft and de Leeuw (1998). 10 One criticism in modelling the quality of institutions on corruption at micro-level is based on the low variability of firms' preferences and expectations. Mocan (2008) developed a model which assumes that an increase in the quality of the institutions of a country, which would increase the probability of apprehension, would in turn reduce the propensity to ask for bribes.
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
59
Table 1 Data and descriptions of variables. Variables Dependent variable Corruption index
Description
Range
“It is common for firms in my line of business to have to pay some irregular additional payments to get things done.”
Discrete variable from 1 to 6: 1 (never), 2 (seldom), 3 (sometimes), 4 (frequently), 5 (mostly), and 6 (always).
Components of economic freedom Non-market dominance “Regarding your firm's major product line, how many competitors do you face in your market?” Anti-competitive “Please judge on a four-point scale how behaviour problematic the following factors are for the operation and growth of your business.” Information constraints “In general, information on the laws and regulations affecting my firm is easy to obtain.” Private investment regulation
Employment regulation
Sales regulation
Prices regulation
Financial constraints
Financial obstacles
Property rights and contract protection Export regulation
“How often does the government intervene in the following types of decisions by your firm?” “How often does the government intervene in the following types of decisions by your firm?” “How often does the government intervene in the following types of decisions by your firm?” “How often does the government intervene in the following types of decisions by your firm?” “What is the ability of the financial system to provide financing to the private sector?” “Please judge on a four-point scale how problematic the following factors are for the operation and growth of your business.” “Please, judge the ability of the legal system to protect property rights and contracts” “How often does the government intervene in the following types of decisions by your firm?”
Firm's covariates Firm size
“Number of full-time employees”
Firm sector
“Principal sector of business”
Legal organisation of firm
“What is the legal organisation of this company?”
Macro covariates GDP INV GOV GINI CIVIL
Gross domestic product per capita Private investments as share of GDP Government spending as share of GDP Gini coefficient of distribution of income Civil liberty index
Discrete variable from 1 to 3: 1 (no competitors), 2 (3 or fewer), and 3 (more than 3). Discrete variable from 1 to 4: 1 (no obstacle), 2 (minor obstacle), 3 (moderate obstacle), and 4 (major obstacle). Discrete variable from 1 to 6: 1 (fully agree), 2 (agree in most cases) 3 (tend to agree), 4 (tend to disagree), 5 (disagree in most cases), and 6 (totally disagree). Discrete variable from 1 to 6: 1 (never), 2 (seldom), 3 (sometimes), 4 (frequently), 5 (usually), and 6 (always). Discrete variable from 1 to 6: 1 (never), 2 (seldom), 3 (sometimes), 4 (frequently), 5 (usually), and 6 (always). Discrete variable from 1 to 6: 1 (never), 2 (seldom), 3 (sometimes), 4 (frequently), 5 (usually), and 6 (always). Discrete variable from 1 to 6: 1 (never), 2 (seldom), 3 (sometimes), 4 (frequently), 5 (usually), and 6 (always). Binary variable: 0 (financing), and 1 (no financing). Discrete variable from 1 to 4: 1 (no obstacle), 2 (minor obstacle) 3 (moderate obstacle), and 4 (major obstacle). Discrete variable from 1 to 3: 1 (totally disagree), 2 (tend to agree), and 3 (fully agree). Discrete variable from 1 to 3: 1 (never), 2 (seldom), 3 (sometimes), 4 (frequently), 5 (usually), and 6 (always).
Discrete variable from 1 to 3: 1 (small, 50 or fewer employees), 2 (medium, 51–500 employees), and 3 (large, 501 or more employees) Discrete variable from 1 to 5: 1 (manufacturing), 2 (services), 3 (other), 4 (agriculture), and 5 (construction) Discrete variable from 1 to 5: 1 (single proprietorship), 2 (partnership), 3 (cooperative), 4 (corporation, privately held), and 5 (corporation listed on stock exchange)
Ordered variable from 1 (no civil liberties) to 6 (full civil liberties)
Dimension
Source WBES2000
Dimension I
WBES2000
Dimension I
WBES2000
Dimension I
WBES2000
Dimension II
WBES2000
Dimension II
WBES2000
Dimension II
WBES2000
Dimension II
WBES2000
Dimension III
WBES2000
Dimension III
WBES2000
Dimension IV
WBES2000
Dimension V
WBES2000
WBES2000
WBES2000
WBES2000
WDI WDI WDI WDI WAO
Notes: Labels of sources are: “WBES2000”, World Business Environment Survey, World Bank, 2000; “WDI”, World Development Indicators; and “WAO” World Audit Organisation.
60
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
(b) Anti-competitive behaviour (Dimension I) 3 2.5 2 1.5
Corruption index
2.5 2
R2=0.63
1
1.5 1
Corruption index
3
(a) Non-market dominance (Dimension I)
1
1.5
2
2.5
3
1.5
2
Non-market dominance
3
3.5
(d) Private investment regulation (Dimension II)
2.5 2
Corruption index
2
1
1.5 1
1.5
2.5
3
3
(c) Information constraints (Dimension I)
Corruption index
2.5
Anti-competitive behaviours
1
2
3
1
4
1.5
Information constraints
2
2.5
3
Private investment regulation
2.5 2 1.5 1
Corruption index
3
(e) Employment regulation (Dimension II)
1
1.5
2
2.5
3
Employment regulation Fig. 3. Components of economic freedom and corruption, WBES sample. Notes: Figures built by aggregating individual data of economic freedom components and corruption at country level.
The variables involved in the multilevel model are, first, control variables, to account for heterogeneity in firm behaviour, allowing for the possibility that the extent of a firm's self-reported corruption effects may depend on its size, legal organisation, and the sector in which it operates (Beck et al., 2002). Firm size is the number of full-time employees and classifies a firm into three categories, large (3), if there are more than 500 full-time employees; medium (2) if there are between 51 and 500 employees; and small (1) when the number of employees is under 50. The main sectors of business are subdivided into five traditional categories: manufacturing, services, agriculture, construction, and other, from 1 to 5, respectively; the legal organisation of the firm is also coded from 1 to 5: single proprietorship (1), partnership, cooperative, privately held corporation, or listed on the stock exchange (5). These variables are also described in Table 1. The existence of clusters suggested that macro-characteristics may influence the economic freedom/corruption link. Assuming that differences in economic conditions and institutional characteristics (in Hodgson's sense) can explain differences in levels of corruption, in line with Acemoglu et al. (2001) and La Porta et al. (2008), macro-economic indicators, per capita GDP, log (GDP) 11 private investment as a share of GDP (INV) and the Gini coefficient of distribution of income (GINI) were all considered. As usual, 11
Capital letters distinguish macro-covariates.
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
2.5 2
Corruption index
1
1.5
2.5 2 1.5 1
Corruption index
3
(b) Prices regulation (Dimension II)
3
(a) Sales regulation (Dimension II)
61
1
1.5
2
2.5
3
1
1.5
Sales regulation
2
2.5
3
Prices regulation
(d) Property rights and contract protection (c) Financial constraints (Dimension III)
3 2.5 2
Corruption index
1
1.5
2.5 2 1.5
R2=0.49
1
Corruption index
3
(Dimension IV)
1.5
2
2.5
3
1.5
3.5
Financial Constraints
2
2.5
3
Property rights and contract protection
2.5 2 1.5
2 R =0.19
1
Corruption index
3
(e) Exportregulation (DimensionV)
2
2.5
3
3.5
4
Export regulation Fig. 4. Economic freedom components and corruption, WBES sample. Notes: Figures built by aggregating individual data of components of economic freedom and corruption variables at country level. Financial constraints are not reported, as they are defined by a binary variable.
0 in the Gini index represents perfect equality, and 100 perfect inequality. The share of government spending in GDP (GOV) is also considered, as La Porta et al. (1999) found that the size of the state and its composition are key variables in explaining differences in good governance across countries. In addition, a civil liberties index (CIVIL) was included as a proxy of the level of democracy of a country (Bliss and Di Tella, 1997) according to the World Audit Organisation rankings, and was based on a checklist of political rights from 1 to 6,where 1 is taken as not free. A strong democratic regime has been found to enforce the reliability of public action, decrease firms' market power and reduce illegal profits, whereas increases in democracy have been found to decrease corruption (Emerson, 2006; Herzfeld and Weiss, 2003; Treisman, 2000) and, in general, economic growth (Bardhan, 1997; Dreher and Herzfeld, 2005). This implies that, as democracy increases, corruption is weakened, irrespective of its level in a country. The significant influence of democracy on the components of economic freedom is also well-known, e.g., the ability of the legal system to protect property rights and prevent corruption (Lundstrom, 2005). However, estimates should not ignore the existence of endogeneity issues, as lower levels of corruption at country level may be responsible for a reduction in the strength of property rights. 12 12
See Acemoglu and Verdier (1998) for a discussion.
62
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
4. Econometric specification This section describes the multilevel probit model, which is an extension of basic linear and non-linear estimation techniques for applications in which clusters of data result from non-independent observations. Neglecting clustering in the data can lead to bias estimates and invalid inferences. If yij is a binary response for the corruption propensity within firm i for country j, and xij is a set of explanatory variables at firm level, a latent continuous variable, yij ∗, is assumed to underlie yij. Although binary response variable yij is observed directly, this is not the case for yij ∗. We know that yij = 1 if yij ∗ > 0, and yij = 0 if yij ∗ ≤ 0. Thus, we can write the multilevel model for yij ∗ as: yij ¼ β0j þ β1 xij þ
H X βh xi0 þ eij
where eij xij ∼Nð0; πÞ
ð1Þ
h¼1
where β0j is the country-specific intercept and β1 is the regression coefficient of each economic freedom component. As we are interested in assessing separately the influences on corruption of various types of economic freedom, Eq. (1) is considered as a companion matrix which includes economic freedom indicators xij in the diagonal, and is otherwise zero. In addition Eq. (1) includes other firms' covariates, xi0, evaluated by parameters βh; eij is the first-level residual term. Under the hypothesis of a random-effect model, we can write β0j as: 1 ð2Þ β0j ¼ γ00 þ γ01 w0j þ u0j where u0j xij ∼Nð0; ψÞ where γ00 is the intercept and γ01 are the coefficients of the vector of the observed second-stage macro-covariates which identify 1 . u0j is the random effect of two-level model and is related to country-specific intercept β0j. The second-level variations w0j 1 does not avoid the possibility that the assumption that part of the variability can be identified by between-country covariates w0j unobserved variability of country effects may generate dependence between firms' economic freedom components xij. As a consequence, we turn to the specification of an extended model containing parameters associated with issues of endogeneity and implement nested restriction tests. normal distribution. It is Conditional on random effect u0j, a probit model is specified by assuming that eij has standard commonly assumed that cluster j is independent, that the covariance between various firms, Cov eij ei′ j ¼ 0, and that two-level error terms are not correlated, Cov(u0j,eij) = 0, so that we can write the reduced form of the model as 13: 1
yij ¼ γ 00 þ γ 01 w0j þ β1 xij þ
H X βh xi0 þ u0j þ eij :
ð3Þ
h¼1
Assuming that u0j is normally distributed, the strategy for estimating the model parameters is to integrate unobserved random effect u0j: 1 1 f yj xj ; wj ¼ ∫f yj xj ; wj ; u0j g u0j du0j ; ð4Þ where g(.) represents the normal density function. 14 As a result, the unconditional estimate does not lead to a closed expression. Maximum likelihood estimation must resort to approximation procedures, such as numerical integration. RabeHesketh et al. (2002) proposed an algorithm using the posterior mean and variance of random effects, which were calculated by building on the work of Naylor and Smith (1982). 15 If the assumed distribution is normal, the numerical quadrature approach yields a deviance ^ , where lnf y˜ ϑ is the (ϒ) which can easily be used for likelihood ratio tests. This statistic is given as ϒ ¼ 2 ln f y˜ϑ − ln f yϑ ^ log-likelihood of the saturated model and lnf y ϑ is the log-likelihood of the model of interest. Some nested specifications can be obtained by imposing parameter restrictions requiring a simple likelihood ratio test on parameter restrictions, as in: LR ¼ 2 ln f full yij jθ≠0 − ln f restr yij jθ ¼ 0
ð5Þ
which has an approximate χ 2 distribution with a number of degrees of freedom equal to the imposed restrictions on the parameters. As anticipated, the multilevel specification in Eq. (3) may imply that the unexplained variability among different countries' corruption propensity includes the effect of omitted macro-variables related to economic and political institutions which, in turn, may be correlated with xij. 13 The assumption that unobserved country characteristics are not correlated with unobserved firm characteristics excludes non-frequent cases in which firms are very successful in operating in corrupt countries. 14 For the sake of simplicity, we include micro covariates of the first level in intercept parameter γ00. 15 Although marginal quasi-likelihood (MQL) and penalised quasi-likelihood (PQL) are largely used in the statistical literature, they are found to generate downwardly biassed estimates (Hedeker, 2001).
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
63
Fig. 5. Strategy for testing nested and non-nested models.
While the standard approach to handling endogeneity is to estimate parameters by instrumental variables, in the present context, in which regressions are carried out on single economic freedom components, a consistent solution is to estimate the within-unit effects of xij, which can be done by controlling for cluster mean covariates of interest. 16 Following Snijders and Berkhof (2004), the correlation between covariates and random effects may expressed as a regression: 2
w0j ¼ α 00 þ α 01 x⋅j þ ε0j
ð6Þ
where x⋅j is the cluster mean of xij. By inserting Eq. (6) into Eq. (3), the random intercept model now also depends on x⋅j , while the reduced form of the model is given as:
1
yij ¼ γ 00 þ γ 01 w0j þ α 01 x⋅j þ β1 xij þ
H X βh xi0 þ u0j þ eij
ð7Þ
h¼1 ∗ ∗ = γ00 + α00 and u0j = u0j + ε0j. Note that the inclusion of the mean of the covariate of interest can account for reverse where γ00 causality between economic freedom indicators and corruption (see Section 2), since unobservable heterogeneity is mitigated by controlling for within-firm heterogeneity. Excluding from the analysis cluster mean x⋅j , when α01 ≠ 0, yields a biassed estimator of β1. Conversely, when α01 = 0, we obtain the exogenous model of Eq. (3) as a restricted specification. In many cases, however, it is of great interest to model dependence more extensively. Since endogeneity is often the result of one or more unobserved variables, which influence both explanatory and dependent variables in the equation, restricting the model in Eq. (7) by imposing vector γ01 = 0 allows us to identify the importance of the conditional macro-covariates in estimated relationships. We obtain an endogenous random intercept model, which only includes micro-covariates as controls, minimising the effects of unobservable country effects. Formally:
yij ¼ γ 00 þ α 01 x⋅j þ β1 xij þ
H X βh xi0 þ u0j þ eij :
ð8Þ
h¼1
Fig. 5 shows the nested relationships among models and the relative restrictions on the likelihood function of Eq. (7). A sequential strategy of the model selection process can reasonably be implemented by accounting for dependence between micro-covariates and random effects. A double route for testing nested models arises with respect to study aims, because we do not determine a priori whether macro-covariates or unobserved confounding, which includes the cluster of mean covariates as a control, can significantly produce endogeneity. We therefore partition the vector of the variables which should control for dependence across 1 2 ;w0j ]. By assuming that the restrictions of the vector of parameters θ1 = θ|γ01 = 0 and θ2 = θ|α01 = 0 are not rejected levels, W0j = [w0j separately, before passing to the next step for testing the restricted models against the benchmark random intercept model (a basic random intercept model without other micro-covariates except those of interest), we must decide whether a best model exists. By defining the conditional function density for restricted models, f(y|x, θ1) and g(y|x, theta2), conventional and adjusted (Vuong, 1989) h i Þ LR tests were used for these non-nested specifications (step 2, Fig. 5). The null hypothesis of model equivalence, H0: E log gf ððyjx;θ yjx;θ Þ ¼ 0, h i h i Þ f ðyjx;θ Þ was tested against the competing model, H1 : E log gf ððyjx;θ yjx;θ Þ > 0 or H1 : E log g ðyjx;θ Þ b0. If H0 is rejected in the first hypothesis, we prefer 1
2
1
1
2
2
f(.) to g(.), and vice versa if the result is in line with the second hypothesis. The best model is then tested against the basic “random H βhxi0 = 0, if a restricted model with γ01 = 0 or α01 = 0 is found. intercept” model by adding ∑ h=1 5. Results We test the models presented in Section 3 and shown in Fig. 5 to choose the most appropriate specification. Parsimonious models are obtained by testing restrictions on the likelihood function, and may be interpreted as nested cases of endogenous multilevel probit models (7). The conventional and adjusted likelihood ratio formulation is shown in Table 2. 16
The instrumental variables method was developed for multilevel analysis by Kim and Frees (2007).
64
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
Table 2 Random intercept models: specification tests, full sample. Non-market AntiPrivate Employment Sales Prices Financial Financial Property Export dominance and competitive investment regulation regulation regulation constraints obstacles rights and regulation information behaviour regulation contract constraints protection Extended endogenous model (Eq. (7)) vs. restricted endogenous model (Eq. (8)) Extended endogenous model (Eq. (7)) vs. exogenous model (Eq. (3)) Restricted endogenous model (Eq. (8)) vs. basic model Exogenous model (Eq. (3)) vs. basic model Exogenous model (Eq. (3)) vs. restricted endogenous model (Eq. (8))
LR test
1421.671 (0.00)
586.19 (0.00)
811.83 (0.00)
821.0637 (0.00)
809.92 (0.00)
813.35 (0.00)
904.48 (0.00)
932.81 (0.00)
836.65 (0.00)
896.28 (0.00)
LR test
2.58 (022)
2.08 (021)
1.33 (024)
2.21 (0.13)
1.62 (0.20)
2.14 (0.14)
0.04 (0.83)
1.75 (0.18)
2.70 (0.10)
2.20 (0.13)
LR test
406.41 (0.00)
2560.29 (0.00)
3326.89 (0.00)
3495.53 (0.00)
3467.49 (0.00)
3514.36 (0.00)
3603.52 (0.00)
2880.63 (0.00)
3753.51 (0.00)
3295.41 (0.00)
LR test
1809.97 (0.00)
2914.48 (0.00)
4137.39 (0.00)
4314.38 (0.00)
4275.79 (0.00)
4325.57 (0.00)
4507.96 (0.00)
3715.54 (0.00)
4683.63 (0.00)
4189.50 (0.00)
24.677 (0.00)
17.67 (0.00)
18.00 (0.00)
15.47 (4.20)
11.47 (0.00)
8.47 (0.00)
8.05 (0.00)
−3.74 (0.00)
10.17 (0.00)
Vuong's 8.432 test (0.00)
Note: p-Values in brackets. Vuong's (1989) test for non-nested models is carried out under standard normal distribution. Extended endogenous model (Eq. (7)) includes endogeneity, macro- and firm covariates. Restricted endogenous model (Eq. (8)) includes endogeneity and firm covariates. Exogenous model (Eq. (3)) includes macro- and firm covariates.
In the first row, model (7) is tested against model (8), in which macro-covariate restrictions are imposed (i.e. γ01 = 0). The results of the LR test for the ten econometric specifications corresponding to each economic freedom variable clearly reject the hypothesis test, and indicate that these variables are important in identifying corruption differences across countries. Instead, as shown in the second row, endogeneity restriction α01 = 0 is never rejected, because the empirical LR test is always lower than the critical value at the usual percentile. To complete the analysis, we test specification (3) against the basic random effect model, in which firms' covariates and country-identifying variables are restricted to zero (βh = 0 and γ01 = 0). The LR test rejects the restricted basic model, confirming that Eq. (3) is the best model to rationalise the data. Voung's (1989) test completes the selection strategy by evaluating the exogenous random effect model (3) against the model with endogeneity (8) restricted for macro-covariates. Vuong's test statistic leads to the rejection of model equivalence for each specification, selecting model (3). We also evaluate restrictions with respect to the benchmark specification; the test provides further support to the model chosen (Table 2, row five). Maximum-likelihood estimates are listed in Table 3. Note that the use of a multilevel probit instead of a probit specification ensures that misleading significance effects due to violation of the assumption of independent errors with constant variance can be avoided. This effect was confirmed in our regression results, in which the multilevel regressions display lower levels of significance when compared with the restricted probit regression. 17 To support this result, we report the intra-class correlation (ρ) for each estimated specification. On average, for each equation, about 30% of the total variability of the link between the components of corruption and economic freedom is attributable to cross-country differences. To emphasise the interpretative role of multilevel models, we identify this source of variability in each relationship by macro-indicators and although there are considerable differences in the sizes of the macro-covariate coefficients, both their signs and significance are generally consistent with expectations. Accordingly, the degree of economic prosperity (GDP) can explain gains in the efficiency of the economic and social system and control of corruption (Mauro, 1995). In contrast with the results of Smelser (1971), our estimates suggest that more income inequality would reduce corruption. While this result may be thought counter-intuitive, there are several possible explanations. First, there is a composition effect of income inequalities on corruption, as pairwise correlations show that African and transition countries have a negative sign, which instead is positive for Latin American countries. Second, our perception index is closer bureaucratic corruption than the Transparency International index, which mainly accounts for political corruption. Inequalities may favour lobbying politicians rather than bureaucrats, reducing the importance of bureaucratic corruption 18 (see, for example, Harstad and Svensson, 2011; Campos and Giovannoni, 2007). In addition, the importance of the civil liberty index (CIVIL) in reducing corruption is in line with theoretical expectations, whereas an oversized state (GOV) affects the efficiency of expenditure, leading to increasing corruption. As with country characteristics, we find very significant coefficients for control covariates at firm level, except for the legal organisation of firms. The probability of corruptive practices in larger firms is consistently found to be 12–13% lower, irrespective
17
The results of estimates obtained from probit models are available from the authors. In order to support this hypothesis, we take corruption data for our sample of countries from Transparency International and calculate correlations with the Gini index. We find a positive relation for both all and subsample countries. The estimated correlations are available upon request from the authors. 18
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
65
Table 3 Corruption and components of economic freedom, full sample. Non-market AntiPrivate Employment dominance competitive investment regulation and behaviour regulation information Non-market dominance Information constraints
Sales regulation
Prices Financial Financial regulation constraints obstacles
0.012 (0.044) 0.075*** (0.018)
Anti-competitive behaviour Private investment regulation Employment regulation
0.148*** (0.026) 0.084** (0.038) 0.131** (0.038)
Sales regulation
0.081** (0.039)
Prices regulation
0.122** (0.037) −0.018 (0.063)
Financial constraints
−0.113** (0.024)
Financial obstacles
−0.218** (0.035)
Property rights and contract protection Export regulation Firm size Firm sector Legal organisation of firm GDP GINI INV GOV CIVIL Constant ρbenchmark ρid _macro N
Property Export rights and regulation contract constraints protection
−0.113*** (0.035) −0.046* (0.027) −0.013 (0.016) −0.270*** (0.000) −0.024*** (0.007) −0.050*** (0.015) 0.175*** (0.042) −0.097* (0.052) 1.084* (0.654) 0.297** 0.124** 3331
−0.101** (0.040) −0.111*** (0.037) −0.004 (0.018) −0.219*** (0.000) −0.021** (0.009) −0.040** (0.018) 0.105** (0.050) −0.177** (0.070) 1.447* (0.784) 0.244** 0.130** 2568
−0.144** (0.037) −0.056** (0.028) −0.017 (0.016) −0.217*** (0.000) −0.020** (0.009) −0.037** (0.018) 0.173** (0.044) −0.123** (0.058) 0.928 (0.764) 0.286** 0.093** 2959
−0.140** (0.036) −0.057** (0.027) −0.020 (0.016) −0.220*** (0.000) −0.019** (0.009) −0.035** (0.017) 0.172** (0.043) −0.119** (0.057) 0.753 (0.753) 0.289** 0.090** 2989
−0.135** (0.037) −0.053* (0.028) −0.017 (0.016) −0.213*** (0.000) −0.020** (0.009) −0.036** (0.018) 0.174** (0.044) −0.124** (0.058) 0.901 (0.769) 0.288** 0.093** 2962
−0.139** (0.036) −0.054** (0.028) −0.016 (0.016) −0.266*** (0.000) −0.018** (0.009) −0.035** (0.017) 0.170** (0.043) −0.121** (0.057) 0.735 (0.753) 0.282** 0.091** 2974
−0.139** (0.036) −0.051* (0.027) −0.010 (0.017) −0.268*** (0.000) −0.024** (0.007) −0.050** (0.015) 0.182** (0.043) −0.095* (0.052) 1.363** (0.641) 0.328** 0.088** 3132
−0.113** (0.036) −0.045 (0.027) −0.014 (0.016) −0.265*** (0.000) −0.025** (0.008) −0.050** (0.016) 0.174** (0.045) −0.086 (0.055) 1.691** (0.671) 0.315** 0.098** 3128
−0.123** (0.034) −0.048* (0.026) −0.015 (0.016) −0.269*** (0.000) −0.025** (0.007) −0.048** (0.015) 0.165** (0.042) −0.098* (0.051) 1.952** (0.633) 0.311** 0.087** 3410
0.082** (0.025) −0.125** (0.035) −0.058** (0.027) −0.016 (0.016) −0.280*** (0.000) −0.023** (0.008) −0.055** (0.015) 0.184** (0.044) −0.092* (0.054) 1.210* (0.667) 0.317** 0.098** 3192
Notes: Dependent variable is binary index of corruption. Standard errors in brackets; asterisks indicate significant p-value levels (*pb 0.1, **p b 0.05, *** p b 0.01). Two measures of intraclass correlations correspond to benchmark model without firm variables and macro-covariates (ρbenchmark) and to model with restrictions in macro-covariates (ρid _macro).
of which aspect of economic freedom is used. In addition, the industrial sector has a higher propensity (on average 5%) to be subject to corruption, perhaps reflecting more market independence of services and the smaller size of businesses. Government regulations on investment, employment, sales and prices increase corruption (Table 3, columns 3–6), as inefficiencies caused by an over-regulated economic system can distort private productive activity and influence corruptive behaviour. Because of the critical role played by the labour market in the process of economic development, understanding whether labour market regulations actually help or hinder corruption stands out as an important task (Caballero and Hammour, 2000; Foster et al., 2002; Bartelsman et al., 2004). Looking at individual components of government regulations gave very similar results across countries, and it is noticeable that the protection of employment had the highest propensity (0.131) to affect corruption. Adding government employment regulations can weigh down the processes of agreement and facilitate corruptive practices, while mediating among firms, workers and unions. Similarly, the other government regulation indicators were found to affect the reduction of corruption significantly, in line with Svensson (2003). Columns 7 and 8 of Table 3 show the estimated influences of the financial market and illustrate compositional effects for countries with different degrees of financial development. Constraints in private and public financing projects increase the probability of corruption, whereas – rather surprisingly – the efficiency of the financial system does not seem to affect it. Trade regulation, as expected, increase corruption, as found by Ades and Di Tella (1997, 1999), Sung and Chu (2003) and Gerring and Thacker (2005).
66
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
Table 4 Random intercept models: specification tests, subsample of African countries.
Extended endogenous model (Eq. (7)) vs. restricted endogenous model (Eq. (8)) Extended endogenous model (Eq. (7)) vs. exogenous model (Eq. (3)) Restricted endogenous model (Eq. (8)) vs. basic model Exogenous model (Eq. (3)) vs. basic model Exogenous model (Eq. (3)) vs. restricted endogenous model (Eq. (8))
Non-market Private Employment Sales Prices Financial Financial dominance and investment regulation regulation regulation constraint obstacles information regulation constraints
Property Export right and regulation contract protection
LR test
509.25 (0.00)
520.87 (0.00)
517.44 (0.00)
516.18 (0.00)
522.39 (0.00)
527.15 (0.00)
472.14 (0.00)
515.48 (0.00)
499.57 (0.00)
LR test
8.22 (0.01)
0.37 (0.54)
6.58 (0.01)
1.23 (0.26)
0.42 (0.51)
2.80 (0.09)
3.21 (0.07)
6.53 (0.01)
1.61 (0.20)
LR test
152.67 (0.00)
355.28 (0.00)
365.61 (0.00)
349.49 (0.00)
347.02 (0.00)
342.25 (0.00)
351.47 (0.00)
395.47 (0.00)
336.40 (0.00)
LR test
636.33 (0.00)
875.78 (0.00)
876.47 (0.00)
864.45 (0.00)
868.99 (0.00)
866.60 (0.00)
820.41 (0.00)
904.43 (0.00)
834.36 (0.00)
Vuong's test 16.81 (0.00)
6.61 (0.00)
94.35 (0.00)
75.23 (0.00)
33.69 (0.00)
40.38 (0.00)
23.37 (0.00)
12.93 (0.00)
27.73 (0.00)
Notes: p-Values in brackets. Vuong's (1989) test for non-nested models is carried out under standard normal distribution. Extended endogenous model (Eq. (7)) includes endogeneity, macro- and firm covariates. Restricted endogenous model (Eq. (8)) includes endogeneity and firm covariates. Exogenous model (Eq. (3)) includes macro- and firm covariates.
So far, the largest and most significant effect on corruption turns out to be exerted by the legal system protecting property rights and contracts (− 0.218). Limiting the possibility of confiscating private property or repudiating contracts may produce positive externalities and determine general improvements in the quality of institutions. However, this result closely depends on the difficulty of enforcing complex private contracts and on the potential advantages of a parallel developed framework for organising private transactions. There is no evidence that the absence of market dominance leads to fewer corruptive practices (column 1). Conversely, the coefficient on the information constraints on laws and regulations is positive and significant at the 1% level. This indicates that firms in countries where information is effective in checking non-competitive practices have lower rents and lower corruption. This result is also consistent with the significant coefficient of anti-competitive behaviour, although the lack of this variable for Africa excluded it from comparative estimates. Table 5 Random intercept models: specification tests, subsample of transition economies. Non-market AntiPrivate Employment Sales Prices Financial Financial Property Export dominance and competitive investment regulation regulation regulation constraint obstacles rights and regulation information behaviour regulation contract constraints protection Extended endogenous model (Eq. (7)) vs. restricted endogenous model (Eq. (8)) Extended endogenous model (Eq. (7)) vs. exogenous model (Eq. (3)) Restricted endogenous model (Eq. (8)) vs. basic model Exogenous model (Eq. (3)) vs. basic model Exogenous model (Eq. (3)) vs. restricted endogenous model (Eq. (8))
LR test
255.16 (0.00)
233.79 (0.00)
230.42 (0.00)
239.79 (0.00)
244.58 (0.00)
247.68 (0.00)
257.20 (0.00)
241.48 (0.00)
255.47 (0.00)
235.25 (0.00)
LR test
0.20 (0.90)
0.17 (0.67)
0.09 (0.76)
0.15 (0.69)
3.58 (0.05)
3.00 (0.08)
0.97 (0.32)
6.20 (0.01)
0.36 (0.54)
1.87 (0.17)
LR test
277.95 (0.00)
260.58 (0.00)
294.99 (0.00)
298.15 (0.00)
285.59 (0.00)
295.65 (0.00)
294.71 (0.00)
251.08 (0.00)
290.79 (0.00)
271.98 (0.00)
LR test
530.74 (0.00)
492.65 (0.00)
525.32 (0.00)
537.78 (0.00)
526.60 (0.00)
540.33 (0.00)
550.95 (0.00)
486.35 (0.00)
545.90 (0.00)
505.36 (0.00)
Vuong's 63.68 test (0.00)
52.66 (0.00)
84.77 (0.00)
85.76 (0.00)
7642 (0.00)
103.51 (0.00)
24.88 (0.00)
18.72 (0.00)
−39.25 (0.00)
31.44 (0.00)
Notes: p-Values in brackets. Vuong's (1989) test for non-nested models is carried out under standard normal distribution. Extended endogenous model (Eq. (7)) includes endogeneity, macro- and firm covariates. Restricted endogenous model (Eq. (8)) includes endogeneity and firm covariates. Exogenous model (Eq. (3)) includes macro- and firm covariates.
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
67
We tested the robustness of our results by estimating two subsamples of countries for Africa and the transition economies in Eastern Europe, and show that aggregation within the five economic freedom dimensions produces consistent estimates of individual relationships. These are useful tests as the Transparency International index for the years around 2000 shows that the African region gives about 3.35/10, whereas the world mean is about 4.00/10, and the transition countries are interesting because of their move to pass to freer market economies and democratic regimes during the 1990s. With the exception of Estonia, Transparency International rates most former communist countries as being highly corrupt and, although liberalisation policies have been extremely important during the last decade, political crises and economic recessions have inhibited the establishment of a market system with clear and general rules affecting economic legality (Graeff and Mehlkop, 2003). One specific cause of this
Table 6 Corruption and economic freedom components, subsample of African countries. Non-market dominance and information constraints Non-market dominance Mean of competitors Information constraints Mean of information constraints
Private investment regulation
Employment regulation
Sales regulation
Prices regulation
Financial constraint
Financial obstacles
−0.289** (0.104) −0.150 (0.097) 2.489** (0.969)
Employment regulation Mean employment regulation
−0.288** (0.134)
Sales regulation
−0.198* (0.105)
Prices regulation
−0.426** (0.162)
Financial constraints
−0.133** (0.059)
Financial obstacles
−0.217** (0.074) −2.065** (0.814)
Property rights and contract protection Mean of property rights and contract protection Export regulation
Firm sector Legal organisation of firm GDP GINI INV GOV CIVIL Constant ρbenchmark ρid _macro N
Export regulation
0.144 (0.107) 4.685 (2.467) 0.115** (0.041) 0.101** (0.625)
Private investment regulation
Firm size
Property rights and contract protection
−0.078 (0.078) 0.055 (0.046) −0.097* (0.042) 0.102 (0.207) −0.026** (0.024) −0.103* (0.045) 0.126** (0.439) 0.020 (0.083) −10.825 (8.110) 0.200** 0.000** 571
−0.160** (0.075) 0.013 (0.044) −0.088** (0.040) 0.926 (0.460) −0.020 (0.014) −0.057** (0.023) 0.368** (0.111) 0.043 (0.063) 0.258 (1.465) 0.237** 0.000** 590
−0.137* (0.074) 0.027 (0.044) −0.095** (0.040) 0.178 (0.201) −0.028** (0.014) −0.056** (0.023) 0.429** (0.116) 0.042 (0.063) −7.148** (3.119) 0.133** 0.000** 595
−0.115 (0.075) 0.023 (0.044) −0.071* (0.040) 0.144 (0.200) −0.021 (0.014) −0.044* (0.023) 0.387** (0.112) 0.025 (0.064) −0.211 (1.434) 0.226** 0.000** 589
−0.133* (0.074) 0.027 (0.044) −0.074* (0.040) 0.118 (0.204) −0.018 (0.013) −0.044* (0.023) 0.391** (0.112) 0.042 (0.064) −0.656 (1.410) 0.223** 0.000** 593
−0.105 (0.076) 0.040 (0.045) −0.044 (0.042) 0.166 (0.215) −0.028** (0.014) −0.045* (0.024) 0.321** (0.113) −0.004 (0.064) 0.133 (1.448) 0.210** 0.000** 578
−0.104 (0.079) 0.037 (0.046) −0.091** (0.043) 0.071 (0.197) −0.025* (0.014) −0.046* (0.025) 0.393** (0.119) 0.041 (0.069) −0.553 (1.476) 0.234** 0.000** 545
−0.136* (0.074) 0.039 (0.044) −0.097** (0.040) −0.017 (0.200) −0.048** (0.016) 0.034 (0.038) −0.161 (0.207) 0.087 (0.067) 8.018** (3.295) 0.035** 0.000** 612
0.248** (0.059) −0.091 (0.076) 0.039 (0.045) −0.076* (0.041) −0.280*** (0.079) −0.014 (0.014) −0.050** (0.024) 0.352** (0.113) 0.014 (0.066) −1.422 (1.451) 0.225** 0.000** 582
Notes: Dependent variable is binary index of corruption. Standard errors in brackets; asterisks indicate significant p-value levels (*pb 0.1, **p b 0.05, *** p b 0.01). Two measures of intraclass correlations correspond to benchmark model, without firm variables and macro-covariates (ρbenchmark) and to model with restrictions in macro-covariates (ρid _macro).
68
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
Table 7 Corruption and economic freedom components, subsample of transition countries. Non-market AntiPrivate Employment Sales Prices Financial Financial dominance and competitive investment regulation regulation regulation constraint obstacles information behaviour regulation constraints Non-market dominance Information constraints
0.136** (0.063) 0.065*** (0.022)
Anti-competitive behaviour Private investment regulation Employment regulation
0.151*** (0.032) −0.194** (0.058) −0.223** (0.068) −0.223** (0.060) 1.161* (0.614)
Sales regulation Mean of sales regulation
−0.123** (0.053)
Prices regulation
−0.021 (0.075)
Financial constraints
−0.161** (0.031) −0.717** (0.288)
Financial obstacles Mean of financial efficiency Property rights and contract protection Export regulation Firm size Legal organisation of firm GDP GINI INV GOV CIVIL Constant ρbenchmark ρid _macro N
Property Export rights and regulation contract protection
−0.215** (0.051)
−0.085 (0.059) −0.065*** (0.024) −0.116 (0.094) −0.016* (0.009) 0.009 (0.017) 0.140*** (0.023) −0.094* (0.049) −0.589 (0.612) 0.098** 0.00** 1704
−0.118* (0.060) −0.058** (0.025) 0.152 (0.201) −0.017* (0.009) 0.016 (0.018) 0.144*** (0.024) −0.089* (0.052) −0.502 (0.617) 0.098** 0.000** 1569
−0.145** (0.064) −0.077** (0.026) 0.926 (0.460) −0.022** (0.010) 0.006 (0.018) 0.137** (0.024) −0.114** (0.054) 0.981 (0.645) 0.100** 0.000** 1408
−0.148** (0.061) −0.066** (0.025) 0.178 (0.201) −0.019** (0.009) 0.007 (0.017) 0.139** (0.024) −0.100* (0.053) 0.864 (0.618) 0.109** 0.000** 1517
−0.163** (0.063) −0.062** (0.026) 0.144 (0.200) −0.037** (0.015) −0.008 (0.021) 0.149** (0.024) −0.216** (0.085) −1.072 (1.004) 0.107** 0.000** 1522
−0.149** (0.061) −0.068** (0.025) 0.118 (0.204) −0.018* (0.009) 0.013 (0.017) 0.154** (0.024) −0.111** (0.052) 0.345 (0.592) 0.106** 0.000** 1539
−0.121** (0.061) −0.075** (0.025) 0.166 (0.215) −0.016* (0.009) 0.009 (0.017) 0.148** (0.023) −0.103** (0.050) 0.059 (0.607) 0.111** 0.000** 1670
−0.134** (0.064) −0.057** (0.027) 0.071 (0.197) 0.005 (0.013) 0.033* (0.020) 0.193** (0.029) −0.049 (0.059) 0.727 (0.647) 0.103** 0.000** 1432
−0.107* (0.059) −0.062** (0.024) −0.017 (0.200) −0.012 (0.009) 0.019 (0.017) 0.152** (0.023) −0.067 (0.050) −0.121 (0.582) 0.107** 0.000** 1710
0.127** (0.036) −0.118* (0.064) −0.066** (0.026) −0.280*** (0.079) −0.016 (0.010) 0.016 (0.018) 0.142** (0.024) −0.141** (0.053) −0.178 (0.642) 0.107** 0.000** 1429
Notes: Dependent variable is binary index of corruption. Standard errors in brackets; asterisks indicate significant p-value levels (*p b 0.1, **p b 0.05, *** p b 0.01). Two measures of intraclass correlations correspond to benchmark model, without firm variables and macro-covariates (ρbenchmark) and to model with restrictions in macro-covariates (ρid _macro).
failure has been linked with reforms in public sector activities, i.e. privatisation, which did not achieve even one of the objectives of reducing corruption. 19 We repeated the selection strategy for the subsamples shown in Fig. 5. Complete results are shown in Tables 4 and 5. We found support for an exogenous multilevel specification for almost all the corruption equations of these subsamples, except in the cases discussed below, in which the endogeneity of economic freedom indicators linked with corruption was substantial. As discussed in Section 4, to assess this possibility empirically, we included the mean of firms' economic freedom components, which estimates the extended endogeneity random intercept model (Eq. (7)). Tables 6 and 7 show maximum-likelihood estimates. Surprisingly, we find that significant second-level residual variance exceeded 20% in many equations for Africa, showing – as expected – that country variability plays an important role in explaining corruption. This is also the case for transition countries, although the significant variation of the intercept of country level is on 19 The privatisation process in the former USSR, characterised by the sale of state assets, was marked by an increase in corruption, because many traditionalist ideas were upheld and not included in the criteria of the decision-making process of rationality and effectiveness. This offered the opportunity to accumulate illegal fortunes (Sachs, 2005).
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
69
Table 8 Estimates of aggregate dimensions of economic freedom on corruption. Dimensions of economic freedom
Market competition constraints Government regulation Financial system ability
Multilevel model
Probit model
Full sample
African countries
Transition countries
Full sample
African countries
Transition countries
0.214*** (0.44) 0.174*** (0.049) −0.153*** (0.050)
0.258*** (0.052) −0.382*** (0.091) −0.284*** (0.058)
0.243*** (0.074) −0.534*** (0.170) −0.333*** (0.108)
0.216*** (0.042) 0.093*** (0.048) −0.167*** (0.043)
0.293*** (0.501) −0.390*** (0.084) −0.275*** (0.056)
0.218*** (0.057) −0.375*** (0.120) −0.192** (0.088)
Notes: Multilevel model adopts specification described in Eq. (3); probit model includes only country dummy and firm control variables. Asterisks indicate significant p-value levels (*p b 0.1, **p b 0.05, *** p b 0.01).
average slightly more than 10%. Covariates at country level have the expected significant effects in most cases, with some exceptions. Neither GDP for African countries nor INV for transition economies were found to exert any particular influence on corruption, while large differences in the size of government GOV appeared to explain most of the African variability. Its positive effect strengthens the hypothesis that inefficiencies generated by large expenditures, such as the misuse of recurring government budgets, feed corruption. In addition, we anticipate that including the country mean of employment regulation and property rights protection for Africa, and sales regulation and financial constraints for transition economies may cause them to lose their significance. The estimated relationships for these subsamples reveal changes in the components of regulation, suggesting that these interventions are more likely to hinder corruption. Not surprisingly, with respect to the non-standard literature, the subsamples reveal the positive and significant influence of anti-competitive behaviour on corruption, in line with the results of Gupta et al. (2001), in which increases in competition among local firms, which share similar norms and rules, encourages them to pay bribes to enhance profitability. Attractive as this result sounds, we emphasise that the nature of this correlation for African countries, although not excessively significant, may capture the simultaneity bias of corruptive practices. The negative sign on property rights strengthens the perception that institutional rules help to sustain economic activities and reduce corruption in African countries. As expected, the control for endogeneity produces insignificant estimated coefficients for GOV and CIVIL. This implies that not only properly functioning property rights rules but also sustained political interventions are necessary to reduce the problems of corruption and defend the structure of institutional rules, because such rules do not avoid large social costs. However, in the regressions, which account for endogeneity, some micro-relationships show the opposite sign in the estimated coefficient to that obtained using aggregate means. This is evident for government regulation on employment for Africa, as well as for sales in transition economies. Although the costs of corruption for Africa are difficult to reduce by micro-regulation of employment (as shown by the insignificant value of the coefficient), minimisation of corruption should depend on the effectiveness of institutional labour designs, irrespective of the development level of a country. In contrast, transition economies may internalise the advantages of sales regulation in terms of saving on corruption costs by applying policies which incentivise efficient sales regulation. As indicated by the strategy tests, constraints on financial system in transition countries are estimated with the additional country mean regressors. The significance of this parameter, together with that at firm level, reinforces the idea that better financial systems are beneficial for combating corruption and enhancing economic growth. Although this result is generally accepted, when we analyse corruption outcomes for transition economies, they turn out to be more complex, due to interactions with investments in the private sector. It is well-known that, soon after periods of reform, the lack of institutional rules for financial systems in the former communist economies led to increasing investments and simultaneously corruption. The significance of the parameter of private investments (INV) may also explain the existence of detrimental financial system effects on corruption, at least in the short term. A potential source of concern in this literature is the over-dispersion effect of specific economic freedom indicators in explaining corruption. The results may indeed reflect the influence of sub-categories of economic freedom, which specifically affect some aspects of corruption but not others. Here, we present the multilevel and probit estimates of economic freedom indices collected by the weighted mean of estimated components within dimensions. Thus, we aggregate the response categories in Dimensions I, II and III, and further select the best model to rationalise the data in the multilevel regression. Leaving out Dimensions IV and V, which have a single modality, the estimated coefficients of the first three dimensions in Table 8 are consistent – as weighted means – with those reported for disaggregate indicators within each dimension, irrespective of sample and statistical method used. The result of this sensitivity analysis proves the correctness of the association with the economic freedom areas and, in our work, the goodness of our previous conclusions.
6. Concluding remarks Standard economic theory predicts that government intervention transfers resources away from the private sector and provides opportunities for corruption. If economic freedom increases sufficiently, then the level of corruption tends to fall, and continues to fall as the quality of institutions gradually improves. This mechanism has received support from conventional
70
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
cross-country estimates, although (Graeff and Mehlkop, 2003) document that the response of corruption to the components of economic freedom appears to be heterogeneous and sometimes contradictory. In addition, the recent corruption literature has also emphasised the importance of modelling micro-level relationships as determinants of corruption (Mocan, 2008). The purpose of this paper was to complement previous approaches by estimating economic freedom and corruption relationships based on multilevel models. To this end, a vast sample of firm-level data from developing and developed countries was used to estimate corruption/economic freedom in these relationships, modelling unobserved cross-country variability. The results show that cross-country differences account on average for one-third of the economic freedom/corruption link and identify the reasons for these differences in some traditional macro-economic and institutional indicators. By accounting for clusters of countries with common features, we explain why the effects of indicators of market competition may change when estimates are carried out on subsamples which only include developing or transition economies. This suggests that market competition may be bad for corruption when institutions are weak, as they often are in developing countries although, in general, the aggregate index confirms that competition reduces corruption. In addition, we can correctly identify when government regulatory interventions encourage and when they discourage corruption. Although fighting corruption is one of the main objectives of modern society, this paper does suggest that a lack of government regulations may actually yield more corruption in less developed countries, whereas standard recipes for greater freedom can be applied in developed countries. Another novel contribution of this paper is its emphasis on the important role played by the political economy of countries and institutions on different aspects of economic freedom, which complements the effects of restricting market competition and government regulation on corruption. These results are consistent with theories of corruption which argue that complementarities exist between economic development and the quality of institutions and, in particular, with the nature of the African and transition economy samples in which, counter-intuitively, high levels of deregulation lead to more corruption. In the context of less developed economies, in economic and institutional terms, the costs of corruption resulting from greater regulation (through bureaucratic powers) seem to be less important than those related to the transition costs for firms which try to escape by paying bribes and thus cause even more corruption. Our model also links financial systems and property rights, which are considered positive mechanisms in combating corruption. The results suggest that these components of economic freedom reduce corruption and they do show high and significant variability across countries. The inclusion of identifying country effects makes the relationship more robust and confirms that improved civil liberties and macro-economic indicators lead to increased efficiency in fighting corruption. The implications for policy-makers in less developed countries and transition economies seem to be clear, although powerful interests, distant from theoretical discussions concerning the trade-off between government intervention, have persuaded several governments to take no action at all, making them almost unresponsive to the need to regulate some sectors. Policies should be introduced to implement complementary strategies to reduce corruption and the costs of economic growth, using suitable government interventions within appropriate sectors. Acknowledgements We would like to thank Paul J. Dunne, for helpful comments. Moreover, we are grateful to the anonymous referees for numerous helpful suggestions. Appendix 1. List of countries by macro regions Country Albania Argentina Bangladesh Belize Bolivia Botswana Brazil Bulgaria Cameroon Canada Chile China Colombia Costa Rica Cote d'Ivoire Croatia Czech Republic Dominican Republic Ecuador Egypt El Salvador Estonia
Africa 0 0 0 0 0 101 0 0 57 0 0 0 0 0 97 0 0 0 0 0 0 0
MENA-region 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 102 0 0
Transition economies 163 0 0 0 0 0 0 125 0 0 0 0 0 0 0 127 137 0 0 0 0 132
East Asia 0 0 0 0 0 0 0 0 0 0 0 101 0 0 0 0 0 0 0 0 0 0
South Asia 0 0 50 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
Latin America 0 100 0 50 100 0 201 0 0 0 100 0 101 100 0 0 0 111 100 0 104 0
OECD 0 0 0 0 0 0 0 0 0 101 0 0 0 0 0 0 0 0 0 0 0 0
Total 163 100 50 50 100 101 201 125 57 101 100 101 101 100 97 127 137 111 100 102 104 132
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
71
Appendix 1 (continued) Country
Africa
MENA-region
Transition economies
East Asia
South Asia
Latin America
OECD
Total
France Germany Ghana Guatemala Haiti Honduras Hungary India Indonesia Italy Kenya Lithuania Madagascar Malawi Malaysia Mexico Namibia Nicaragua Nigeria Pakistan Panama Peru Philippines Poland Portugal Romania Russian Federation Senegal Singapore Slovak Republic Slovenia South Africa Spain Tanzania Thailand Trinidad and Tobago Tunisia Turkey Uganda Ukraine United Kingdom United States Venezuela Zambia Zimbabwe Total
0 0 119 0 0 0 0 0 0 0 113 0 116 55 0 0 95 0 93 0 0 0 0 0 0 0 0 124 0 0 0 121 0 83 0 0 0 0 137 0 0 0 0 84 129 1524
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 52 0 0 0 0 0 0 0 0 154
0 0 0 0 0 0 129 0 0 0 0 112 0 0 0 0 0 0 0 0 0 0 0 225 0 125 525 0 0 129 125 0 0 0 0 0 0 150 0 225 0 0 0 0 0 2429
0 0 0 0 0 0 0 0 100 0 0 0 0 0 100 0 0 0 0 0 0 0 100 0 0 0 0 0 100 0 0 0 0 0 422 0 0 0 0 0 0 0 0 0 0 923
0 0 0 0 0 0 0 210 0 0 0 0 0 0 0 0 0 0 0 103 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 363
0 0 0 106 103 100 0 0 0 0 0 0 0 0 0 100 0 100 0 0 100 108 0 0 0 0 0 0 0 0 0 0 0 0 0 101 0 0 0 0 0 0 100 0 0 1985
100 100 0 0 0 0 0 0 0 100 0 0 0 0 0 0 0 0 0 0 0 0 0 0 100 0 0 0 0 0 0 0 104 0 0 0 0 0 0 0 102 100 0 0 0 807
100 100 119 106 103 100 129 210 100 100 113 112 116 55 100 100 95 100 93 103 100 108 100 225 100 125 525 124 100 129 125 121 104 83 422 101 52 150 137 225 102 100 100 84 129 8185
References Acemoglu, D., Johnson, S., Robinson, A.J., 2001. The Colonial Origins of Comparative Development: An Empirical Investigation. American Economic Review 91, 1369–1401. Acemoglu, D., Verdier, T., 1998. Property rights, corruption and the allocation of talent: a general equilibrium approach. The Economic Journal 108, 1381–1403. Acemoglu, D., Verdier, T., 2000. The choice between market failures and corruption. American Economic Review 90, 194–211. Ades, A., Di Tella, R., 1997. The new economics of corruption: a survey and some new results. Political Studies 45, 496–515. Ades, A., Di Tella, R., 1999. Rents, competition, and corruption. American Economic Review 89, 982–993. Adsera, A., 2003. Are you being served? Political accountability and quality of government. Journal of Law, Economics and Organization 19, 445–490. Ahlin, C., Pang, J., 2008. Are financial development and corruption control substitutes in promoting growth? Journal of Development Economics 86, 414–433. Apergis, N., Dincer, O., Payne, J., 2011. Live free or bribe: on the causal dynamics between economic freedom and corruption in U.S. states. European Journal of Political Economy 28, 215–226. Bardhan, P., 1997. Corruption and development: a review of issues. Journal of Economic Literature 35, 1320–1346. Bartelsman, E.J., Haltiwanger, J., Scarpetta, S., 2004. Microeconomic evidence of creative destruction in industrial and developing countries. Tinbergen Institute Discussion Papers 04-114/3. Tinbergen Institute. Beck, T., Demirguc-Kunt, A., Maksimovic, V., 2002. Financial and legal constraints to firm growth — does size matter? Policy Research Working Paper Series 2784. The World Bank. Billger, S.M., Goel, R.K., 2009. Do existing corruption levels matter in controlling corruption? Cross-country quantile regression estimates. Journal of Development Economics 90, 299–305. Bliss, C., Di Tella, R., 1997. Does competition kill corruption? Journal of Political Economy 105, 1001–1023. Brunetti, A., Kisunko, G., Weder, B., 1997. Institutional obstacles to doing business: region-by-region results from a worldwide survey of the private sector. Policy Research Working Paper Series 1759. The World Bank.
72
L. Pieroni, G. d'Agostino / European Journal of Political Economy 29 (2013) 54–72
Caballero, R.J., Hammour, M.L., 2000. Creative destruction and development: institutions, crises, and restructuring. NBER Working Papers, 7849. National Bureau of Economic Research. Campos, N., Giovannoni, F., 2007. Lobbying, corruption and political influence. Public Choice 131, 1–21. Carden, A., Verdon, L.L., 2010. When is corruption a substitute for economic freedom? Law and Development Review 3, 40–63. Chafuen, A.A., Guzmàn, E., 2000. Economic freedom and corruption. Index of Economic Freedom. The Heritage Foundation, Washington, pp. 51–63. Compton, R.A., Giedeman, D.C., Hoover, G.A., 2011. Panel evidence on economic freedom and growth in the United States. European Journal of Political Economy 27, 423–435. de Haan, J., Sturm, J.E., 2003. Does more democracy lead to greater economic freedom? New evidence for developing countries. European Journal of Political Economy 19, 547–563. de Jong, E., Bogmans, C., 2011. Does corruption discourage international trade? European Journal of Political Economy 27, 385–398. Dreher, A., Herzfeld, T., 2005. The economic costs of corruption: a survey and new evidence. Public Economics 0506001, EconWPA. Elliot, K.A., 1997. Corruption as an International Policy Problem: Overview and Recommendations. Institute for International Economics, Washington DC, pp. 175–223. Emerson, P.M., 2006. Corruption, competition and democracy. Journal of Development Economics 81, 193–212. Faber, G., Gerritse, M., 2012. Foreign determinants of local institutions: Spatial dependence and openness. European Journal of Political Economy 28, 54–63. Foster, L., Haltiwanger, J., Krizan, C.J., 2002. The link between aggregate and micro productivity growth: evidence from retail trade. NBER Working Papers, 9120. National Bureau of Economic Research. Gerring, J., Thacker, S.C., 2005. Do neoliberal policies deter political corruption? International Organization 59, 233–254. Gorodnichenko, Y., Sabirianova, P.K., 2007. Public sector pay and corruption: measuring bribery from micro data. Journal of Public Economics 91, 963–991. Graeff, P., Mehlkop, G., 2003. The impact of economic freedom on corruption: different patterns for rich and poor countries. European Journal of Political Economy 19, 605–620. Gupta, S., de Mello, L., Sharan, R., 2001. Corruption and military spending. European Journal of Political Economy 17, 749–777. Gwartney, J., Lawson, R., 2000. Economic Freedom of the World 2000, Annual Report. The Fraser Institute, Vancouver. Gwartney, J., Lawson, R., 2002. Economic Freedom of the World 2002, Annual Report. The Fraser Institute, Vancouver. Gwartney, J., Lawson, R., 2003. The concept and measurement of economic freedom. European Journal of Political Economy 19, 405–430. Harstad, B., Svensson, J., 2011. Bribes, lobbying and development. American Political Science Review 105, 46–63. Hedeker, D., 2001. Multinomial models for ordinal and categorical variables. Handbook of Multilevel Analysis. Springer. Herzfeld, T., Weiss, C., 2003. Corruption and legal (in)effectiveness: an empirical investigation. European Journal of Political Economy 19, 621–632. Hodgson, G.M., 2003. The hidden persuaders: institutions and individuals in economic theory. Cambridge Journal of Economics 27, 159–175. Hodgson, G.M., 2006. Institutions, recessions and recovery in the transitional economies. Journal of Economic Issues 90, 1043–1061. Hodgson, G., Jiang, S., 2007. The economics of corruption and the corruption of economics: an institutionalist perspective. Journal of Economic Issues 91, 1043–1061. Kim, J., Frees, E., 2007. Multilevel modeling with correlated effects. Psychometrika 72, 505–533. Kreft, I., de Leeuw, J., 1998. Introducing Multilevel Modeling. Sage, Thousand Oaks, CA. Krueger, A.O., 1974. The political economy of the rent-seeking society. American Economic Review 64, 291–303. La Porta, R., Lopez-de Silanes, F., Shleifer, A., Vishny, R., 1999. The quality of government. Journal of Law, Economics and Organization 15, 222–279. La Porta, R., Lopez-de-Silanes, F., Shleifer, A., 2008. The Economic Consequences of Legal Origins. Journal of Economic Literature 46, 285–332. Lambsdorff, J.G., 2007. The Institutional Economics of Corruption and Reform: Theory, Evidence and Policy. Cambridge University Press, Cambridge. Lundstrom, S., 2005. The effect of democracy on different categories of economic freedom. European Journal of Political Economy 21, 967–980. Mauro, P., 1995. Corruption and growth. Quarterly Journal of Economics 110, 681–712. Méon, P.G., Sekkat, K., 2005. Does corruption grease or sand the wheels of growth? Public Choice 122, 69–97. Milgrom, P., Roberts, J., 2008. Motivation: contracts, information and incentives. Economics, Organization and Management. Prentice Hall. Mocan, N., 2008. What determines corruption? International evidence from microdata. Economic Inquiry 46, 493–510. Naylor, J.C., Smith, F.M.A., 1982. Applications of a method for the efficient computation of posterior distributions. Applied Statistics 31, 214–225. North, D.C., Thomas, R.P., 1973. The Rise of the Western World: A New Economic History. Cambridge University Press, Cambridge. Cambridge. Paldam, M., 2002a. The cross-country pattern of corruption: economics, culture and the seesaw dynamics. European Journal of Political Economy 18, 215–240. Paldam, M., 2002b. The big pattern of corruption, economics, culture and the seesaw dynamics. Economics Working Papers 1999–11. School of Economics and Management, University of Aarhus. RabeHesketh, S., Skrondal, A., Pickles, A., 2002. Reliable estimation of generalized linear mixed models using adaptive quadrature. The Stata Journal 2, 1–21. Sachs, J., 2005. The End of Poverty: How We Can Make It Happen in Our Lifetime. Penguin Books. Shen, C., Williamson, J.B., 2005. Corruption, democracy, economic freedom, and state strength: a cross-national analysis. International Journal of Comparative Sociology 46, 327–345. Shleifer, A., Vishny, R.W., 1993. Corruption. Quarterly Journal of Economics 108, 599–617. Smelser, N.J., 1971. Stability, Instability and the Analysis of Political Corruption. Little Brown, Boston, pp. 7–29. Snijders, T.A.B., Berkhof, J., 2004. Diagnostic checks for multilevel models. Handbook of Quantitative Multilevel Analysis. Sage, Thousand Oaks, CA. Sung, H., Chu, D., 2003. Does participation in the global economy reduce political corruption? An empirical inquiry. International Journal of Comparative Criminology 3, 94–118. Svensson, J., 2003. Who Must Pay Bribes And How Much? Evidence From A Cross Section Of Firms. The Quarterly Journal of Economics 118, 207–230. Svensson, J., 2005. Eight questions about corruption. Journal of Economic Perspectives 19, 19–42. Swamy, A., Knack, S., Lee, Y., Azfar, O., 2001. Gender and corruption. Center for Development Economics, 158. Department of Economics, Williams College. Torrez, J., 2002. The effect of openness on corruption. Journal of International Trade & Economic Development 11, 387–403. Treisman, D., 2000. The causes of corruption: a cross-national study. Journal of Public Economics 76, 399–457. Vuong, Q.H., 1989. Likelihood ratio tests for model selection and non-nested hypotheses. Econometrica 57, 307–333.