I Escola de Verão em Economia do Desenvolvimento – FEA/USP
Transcrição
I Escola de Verão em Economia do Desenvolvimento – FEA/USP
I Escola de Verão em Economia do Desenvolvimento FEA/USP – 06 a 10 de Fevereiro de 2012 Minicurso: Eficiência do Gasto Público e Desenvolvimento Econômico Docente: Fabiana Rocha Este arquivo inclui os seguintes itens bibliográficos: Adam, A., Delis, M.. Kammas, P. 2008. Fiscal Decentralization and Public Sector Efficiency: Evidence from OECD Countries, CESIFO Working Paper No. 2364. Afonso, A., Schuknecht, L., Tanzi, V. 2006. Public sector efficiency: an international comparison, Public Choice 123, 321-347. Angelopoulos, K., Philippopoulos,A., Tsionas, E. 2008. Does public sector efficiency matter? Revisiting the relation between fiscal size and economic growth, Public Choice 132, 245-278. Souza, I., Nishijima, M., Rocha, F. 2010. Eficiência do setor hospitalar nos municípios paulistas, Economia Aplicada, 14(1). Fiscal Decentralization and Public Sector Efficiency: Evidence from OECD Countries ANTONIS ADAM MANTHOS D. DELIS PANTELIS KAMMAS CESIFO WORKING PAPER NO. 2364 CATEGORY 1: PUBLIC FINANCE AUGUST 2008 An electronic version of the paper may be downloaded • from the SSRN website: www.SSRN.com • from the RePEc website: www.RePEc.org • from the CESifo website: www.CESifo-group.org/wp T T CESifo Working Paper No. 2364 Fiscal Decentralization and Public Sector Efficiency: Evidence from OECD Countries Abstract This paper attempts to identify the effect of fiscal decentralization on public sector efficiency (PSE). We employ data envelopment analysis on a panel of 21 OECD countries over the period 1970-2000 to construct two alternative PSE indicators that reflect the governmental goals of economic performance and stability. In turn, using a novel technique that merges the methodologies of Simar and Wilson (2007) and Khan and Lewbel (2007), we regress the PSE scores obtained on an extensive set of alternative fiscal decentralization measures. Backed by strong empirical results, obtained from a number of different specifications, we contend that PSE is increasing with fiscal decentralization. JEL Code: C14, C24, H11, H50. Keywords: public sector efficiency, fiscal decentralization, semi-parametric models. Antonis Adam Department of Economics University of Cyprus P.O. Box 20537 1678 Nicosia Cyprus [email protected] Manthos D. Delis Athens University of Economics and Business 76 Patission Street 10434, Athens Greece [email protected] Pantelis Kammas Department of Economics University of Cyprus P.O. Box 20537 1678 Nicosia Cyprus [email protected] 23, February 2008 We are indebted to Jon Fiva and to Efthymios Tsionas for valuable suggestions and to Jakob de Haan, Richard Jong-A-Pin, Jochen Mierau, Dan Stegarescu and Jose Tavares for generously giving us access to their data. We also thank Konstantinos Angelopoulos, Vangelis Dioikitopoulos, George Economides, Thomas Moutos, Apostolis Philippopoulos and Nikos Tsakiris. Any remaining errors are ours. 1. Introduction It has long been recognized that governments differ significantly in the efficiency of delivering public services (see e.g. Tanzi and Schuknecht, 1998; Afonso et al., 2005). Some are extremely wasteful and ineffective in performing basic activities, whereas others achieve their objectives in a systematic and comprehensive way. The strive to increase government/ public sector efficiency (PSE hereafter) has spawned a vigorous theoretical literature on channels that may affect it, with a quite prominent one being the design of fiscal relation across the levels of government. A strand of the ongoing debate argues that fiscal decentralization is positively associated with government efficiency and attributes this effect either to increased electoral control – that comes as a result of increased decentralization (see e.g. Seabright, 1996) – or to yardstick competition among local governments (see e.g. Besley and Case, 1995; Besley and Smart, 2007).1 In contrast, other scholars note that local politicians and bureaucrats are likely to face increased pressure from local interest groups (see e.g. Bardhan and Mookherjee, 2000) and argue that fiscal decentralization, under these or similar state of affairs, undermines government efficiency (Prud’homme, 1995).2 In the recent years, there is a small, albeit growing, body of empirical work that aims at identifying the effect of fiscal decentralization on the quality of government (see e.g. Fisman and Gatti, 2002a; Enikolopov and Zhuravskaya, 2007). In most of these studies, the dependent variable is some internationally comparable outcome of government policy – usually captured by socioeconomic indices like infant mortality, the literacy ratio, immunization of population etc. – and the key explanatory variable is fiscal decentralization, measured as the ratio of sub-national government expenditures (resp. tax revenues) to total public spending (resp. tax revenues).3 Yet, the theoretical hypotheses postulated above are 1 Another branch of the literature argues that fiscal decentralization restricts the governments’ Leviathan behavior and the consequent overspending by the politicians, through inter-jurisdictional fiscal competition (see e.g. Brennan and Buchanan, 1980; Edwards and Keen, 1996). 2 This argument goes back to Alexander Hamilton, John Jay and James Madison who argued that the lower the level of government, the greater is the extent of vulnerability to vested interest and the less protected the minorities and poor tend to be [The Federalist Papers, 1787]. 3 Fisman and Gatti (2002a) and Mello and Barenstein (2001) find that increased decentralization (measured as the budgetary share of subnational governments) is associated with lower levels of corruption. In a similar vein, Fisman and Gatti (2002b) and Henderson and Kuncoro (2004) using sub-national data for the US and Indonesia, respectively, show that decentralization of public expenditure is effective in reducing corruption only if it is accompanied by increased power to raise revenue (i.e. increased tax autonomy). Robalino et al. (2001) and Khalegian (2003) in cross-country studies, also find support that fiscal decentralization is associated with lower infant mortality rates and immunization rates (taken as measures of the quality of governance). Finally, Enikolopov and Zhuravskaya (2007) examine the effect of decentralization on a set of four indicators of governance quality (namely the three indicators used in studies reviewed above plus the illiteracy ratio) and conclude that the effects of fiscal decentralization are beneficial only in countries that are also characterized by a high degree of political centralization. 2 not comprehensively addressed simply by employing socioeconomic indicators as measures of “good governance”. This is because these measures do not encompass the size of government spending and thus fail to reflect the level of efficiency in delivering government services. In the words of Barankay and Lockwood (2007) “[…] these regressions do not estimate government “production functions” because they do not control for the inputs to the output that is the dependent variable. […] In the absence of controls for these inputs, these regressions cannot tell us much about the efficiency of government as any observed correlation between decentralization and government output can be due to omitted variable bias.” In an effort to construct a plausible connection between theory and identification, the purpose of this paper is to generate an empirical model that analyzes the relationship between fiscal decentralization and PSE. Therefore, we opt for direct measures of PSE, derived nonparametrically at a first stage of analysis. In particular, we use data envelopment analysis (DEA) on a panel of 21 OECD countries that covers the period 1970-2000 to construct two alternative PSE indicators that reflect the goals of economic performance and stability. By doing this, we implicitly assume that these indicators are derived from an underlying government production relationship, where public spending serves as the input in the production of public services.4 In the subsequent stage of analysis, we regress the PSE scores obtained on a set of alternative fiscal decentralization measures following a technique that merges the methodologies of Simar and Wilson (2007) and Khan and Lewbel (2007).5 Given this methodological novelty, the main contribution of our study is that our dependent variable allows for differences in the size of government spending and, therefore, does not give an unfair credit to wasteful governments, even when the latter achieve better outcomes. Our main finding is that government efficiency increases with the degree of fiscal decentralization. This result appears to be robust to a number of different specifications and fiscal decentralization measures. More precisely, we employ alternatively fiscal decentralization measures as developed by Stegarescu (2005b), the measures reported in the IMF’s Government Finance Statistics (2002), and measures of fiscal autonomy (reflecting vertical fiscal imbalance and taxation autonomy) and we show that the positive relationship between fiscal decentralization and PSE survives in all different specifications. 4 This is as in Tanzi and Schuknecht (1998) and Afonso et al. (2005). We have resorted to this technique mainly because the second-stage analysis may not be robustly carried out with conventional econometric methods. For details see Section 4. 5 3 The structure of the rest of the paper emerges along the following lines. Section 2 presents the theoretical considerations. In Section 3 we describe the data used in our empirical analysis, as well as the DEA technique employed to obtain the government efficiency estimates. In Section 4 we illustrate the econometric methodology used to regress the government efficiency estimates on fiscal decentralization measures. In Section 5 we present and discuss the empirical results and, finally, Section 6 concludes. 2. Theoretical considerations The theoretical literature on fiscal federalism identifies two benchmark channels through which fiscal decentralization is expected to affect positively the efficiency of governments. These are (i) increased electoral control and (ii) yardstick competition among local governments that comes as a result of decentralization.6 On the other hand, it has been also pointed out that fiscal decentralization may be negatively associated with government efficiency. In the presence of economies of scale (see e.g. Stein, 1997) or differences in the quality of human capital between national and sub-national bureaucracies (Prud’homme, 1995), decentralization may lead to higher costs and thus increased inefficiency in the delivery of public services. In the present section, we briefly review these mechanisms and we set out the main testable hypotheses of our paper. According to the electoral control mechanism, decentralization reduces the incentives for officials to divert rents and increases the probability of “bad” incumbents to be voted out of office, therefore affecting the overall efficiency of the government positively (Hindriks and Lockwood, 2005). More precisely, Seabright (1996) shows that rent-seeking politicians, when contesting in decentralized elections, face incentives to please the voters in each (local) constituency, whereas in national elections they should please the voters only in a majority of localities to get re-elected. Similar results are obtained by Persson and Tabellini (2000), Hindriks and Lockwood (2005) and Myerson (2006). The second path through which fiscal decentralization can alter the incentives and the selection effects of elections is via yardstick competition. According to this theory (see e.g. Shleifer, 1985; Salmon, 1987; Besley and Case, 1995), citizens have an advantage in evaluating the performance of their policy makers when they are able to compare the policy 6 Barankay and Lockwood (2007) suggest an additional mechanism through which fiscal decentralization may lead to increased efficiency, namely the decrease in lobbying by interest groups. However, since the theoretical literature (see e.g. Bardhan and Mookherjee, 2000; Bordignon et al., 2003; Redoano, 2003) appears to be rather inconclusive on this issue (mainly because under certain conditions there may be more lobbying with decentralization), we prefer not to include this mechanism in the ones we refer to as benchmark. 4 choices of their own political representatives with the corresponding choices of neighbor regions’ policy makers.7 Therefore, fiscal decentralization may raise PSE, since it provides citizens the chance to compare public services and taxes across jurisdictions and helps them to judge whether their government wastes resources through low human capital capacity or rent-seeking (Besley and Smart, 2007). However, fiscal decentralization may also exert a negative impact on the efficiency of government. This impact may be attributed to a number of potential advantages of the provision of public goods by central governments. First, in the presence of economies of scale, decentralization may lead to higher costs (see e.g. Stein, 1997). Second, national government bureaucracies are more likely to offer talented people better careers and possibilities of promotion, which may in turn attract higher quality individuals (Prud’homme, 1995). Finally, other scholars underline the potential danger that local politicians and bureaucrats are likely to face increased pressure from local interest groups (see e.g. Bardhan and Mookherjee, 2000; Prud’homme, 1995). In view of these contradictory theoretical underpinnings, we provide below an empirical framework to analyze the relationship between fiscal decentralization and PSE. 3. The data 3.1. Public sector efficiency estimates using DEA The measurement of PSE and the resulting comparison of individual countries in terms of the efficient functioning of their public sectors, present a number of difficulties related to the scarcity of publicly available data and the complicated problems that may emerge in the estimation procedure. In the present study, we opt for a direct estimation of productive PSE using Data Envelopment Analysis (DEA).8 DEA is a non-parametric programming technique that provides a linear piecewise frontier, by enveloping the observed data points, and yields a convex production possibilities set.9 As such, it does not require the explicit specification of a functional form of the 7 The theory of yardstick competition is also studied by Bordignon et al. (2004), Belleflamme and Hindriks (2003), Besley and Smart (2007) and Bodenstein and Ursprung (2001). 8 Only recently a number of studies cultivated an effort towards the computation of PSE indicators. Concerning OECD economies, Afonso et al. (2005) employed a nonparametric method to estimate relative efficiency scores for several parts of the public sector during the 1980s and the 1990s, while Afonso and St. Aubyn (2005) focused on the efficiency of government spending on education and health. Using similar techniques, Gupta and Verhoeven (2001), Sijpe and Rayp (2007) and Afonso et al. (2006) focused on developing countries. Finally, Balaguer-Coll et al. (2007) considered using DEA to analyze the efficiency of local governments in Spain. 9 For an excellent account on DEA, see Coelli et al. (2005). 5 underlying production relationship. To introduce some notation, let us assume that for N observations there exist M inputs in the production of public goods, yielding S outputs. Hence, each observation x n = ( x1n , x2n ,..., xmn ) ∈ R+M to n uses a produce a nonnegative nonnegative vector vector of of inputs denoted outputs, denoted y n = ( y1n , y2n ,..., ySn ) ∈ R+S . Production technology, F = {( y, x) : x can produce y} , describes the set of feasible input-output vectors, and the input sets of production technology, L( y ) = {x : ( y, x) ∈ F } describe the sets of input vectors that are feasible for each output vector (Kumbhakar and Lovell, 2000). To measure productive efficiency we use the following input-oriented DEA model,10 where the inputs are minimized and the outputs are held at their current levels: θ * = min θ , s.t. n ∑λ x j =1 ji ≤ θ xi 0 i = 1, 2,..., m; j rj ≥ yr 0 j =1 j n ∑λ y j =1 n ∑λ j =1 λj ≥ 0 r = 1, 2,..., s; (1) j = 1, 2,..., n; where public sector 0 represents one of the N public sectors under evaluation, and xi0 and yr0 are the ith input and rth output for public sector 0, respectively. If θ* = 1, then the current input levels cannot be proportionally reduced, indicating that public sector 0 is on the frontier. Otherwise, if θ* < 1, then public sector 0 is inefficient and θ* represents its inputoriented efficiency score. Finally, λ is the activity vector denoting the intensity levels at which the N observations are conducted. Note that this approach, through the convexity constraint Σλ = 1 (which accounts for variable returns to scale) forms a convex hull of intersecting planes, since the frontier production plane is defined by combining some actual production planes. 10 DEA may be computed either as input or output oriented. Input-oriented DEA shows by how much input quantities can be reduced without varying the output quantities produced. Output-oriented DEA assesses by how much output quantities can be proportionally increased without changing the input quantities used. The two measures provide the same results under constant returns to scale but give slightly different values under variable returns to scale. Nevertheless, both output and input oriented models will identify the same set of efficient/inefficient public sectors (see Coelli et al., 2005). Also, a constant returns to scale assumption is only appropriate when all public sectors are operating in an optimal scale (imperfections, asymmetries, etc. are not present), and therefore we opt for a variable returns to scale specification. 6 Obviously, to measure PSE some “performance measures” are required that could be interpreted as outputs of total public spending (which naturally serves as the input of the production of public services) and should reflect the objectives (or alternatively the tasks) of government. Following the rationale of the relevant literature (see e.g. Afonso et al., 2005; Angelopoulos and Philippopoulos, 2005), we utilize two well-established performance indicators that reflect benchmark tasks of government. These are: (i) the economic performance indicator (EcPerf) and (ii) the economic stability indicator (EcStab).11 The EcPerf measure assumes that the government output is composed by the unemployment rate, GDP per capita and the annual GDP growth rate. More precisely, lower scores in the unemployment rate and higher scores in GDP per capita and GDP growth reflect better economic performance. Data for the unemployment rate are taken from OECD’s Economic Outlook (2005), whereas data for GDP per capita and GDP growth are obtained from World Bank’s Development Indicators (2004). On the other hand, the EcStab indicator consists of the standard deviation of the GDP growth rate, which is interpreted as a measure of economic fluctuations, and of the inflation rate. In this case, lower scores in both measures reflect higher economic stability. Data for standard deviation of the GDP growth rate are taken from OECD Economic Outlook: Annual and Quarterly Data (2007)12 and data for the inflation rate are obtained from World Bank’s Development Indicators (2004). Finally, data for total public spending (the input) are also obtained from the World Bank’s Development Indicators (2004). Our dataset consists of 21 OECD countries and spans the period 1970-2000.13 Space constraints prevent reporting the yearly results of estimation of Program 1 and, therefore, we present 10-year averages for each country on the Figures in Appendix B.14 The first set of four graphs presents the relative efficiency scores when the government target is economic performance and the second set the equivalent when the government target is economic stability. Scores where data for any of the input or output of the production process are missing are not derived for the particular country. Sensitivity analyses performed on 11 As Tanzi and Schuknecht (2000, pp. 75) state: “It is difficult or even impossible to consider all the social and economic objectives (and thus all the socioeconomic indicators) that the governments might want to influence with this spending. By necessity, the analysis will include fewer indicators than might have been desirable to include”. 12 Annual data on standard deviation of the growth rate is obtained by utilizing the quarterly data available in the Economic Outlook. 13 The countries in our sample are: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Japan, Luxembourg, Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, UK and USA. 14 The full set of Stage 1 results is available upon request. 7 Program 1 showed that efficient public sectors remained efficient to any simultaneous data changes in the respective inputs (for a detailed discussion of the sensitivity analysis on DEA estimates see Zhu, 2003). When the government goal is economic performance, the PSE frontier is mainly shaped by Japan (if data is available) and Switzerland, with USA, Luxembourg, Canada and, most importantly, Ireland gaining significant ground towards the end of the period. Besides these countries, and although the ranking of public sectors appears to change through time, we note that Australia and Norway are characterized by relatively efficient public sectors, whereas Belgium, France, Greece, Italy and Ireland (in the beginning of the sample period) are the poor performers. The results are strikingly similar when the government goal is assumed to be economic stability. The frontier is shaped by exactly the same countries and the patterns of change remain unaltered, with Ireland, Canada and USA substantially improving their PSE scores by the end of the sample period. A noticeable development is that most countries tend to achieve higher PSE scores towards the end of the period examined, even if they are among the relatively poor performers. Overall, these results seem to be reasonable approximations of prior academic belief and are aligned with findings of previous research (see e.g. Afonso et al., 2005). 3.2. Fiscal decentralization measures The best approximation of the degree of fiscal decentralization has been an issue of considerable disagreement among empirical studies. Usually, it is proxied by the budgetary share of sub-national units as recorded by the IMF’s Government Financial Statistics (GFS).15 However, this widely employed measure bears major shortcomings, since it fails to integrate vital aspects of intergovernmental relations. Most importantly, it fails to capture the real degree of sub-national governments’ autonomy that is to reflect the extent to which decisions regarding revenues and expenditures are truly assigned to lower levels of government (see Ebel and Yilmaz, 2003; Stegarescu, 2005b; Barankay and Lockwood, 2007). In particular, it has been pointed out that the GFS measure tends to overestimate the share of government expenditure and tax revenues that is controlled by sub-national governments and that it does so in a way that varies widely across countries (Ebel and Yilmaz, 2003). For example, consider the extreme case of a country where all taxes are set nationally, but where the revenues are shared with local governments via a fixed formula. 15 Previous empirical studies based on the GFS measure of fiscal decentralization are Jin and Zou (2002), Davoodi and Zou (1998), Fisman and Gatti (2002a) and Enikopolov and Zhuravskaya (2007). 8 The share of tax revenues going to sub-national governments is measured in the GFS statistics as sub-national revenue, even though local governments have no control over the tax rate and the tax base. Similar problems arise on the expenditure side from policies that are controlled by central government, but implemented by sub-national governments (Stegarescu, 2005b; Barankay and Lockwood, 2007).16 In view of these difficulties, Stegarescu (2005b) developed new measures of fiscal decentralization and sub-national tax autonomy, based on the detailed data provided by OECD (1999). The advantage of the OECD (1999) survey is that it classifies in a very analytical way the taxes of sub-national governments according to the degree of decisionmaking autonomy.17 More precisely, it separates taxes that are set by sub-central governments (i.e. sub-central governments determine the tax rate and the corresponding tax base) from those that are determined by the central government at a national level and in turn shared with sub-national units. To this end, Stegarescu’s measures of fiscal decentralization reflect the “real” tax-raising autonomy of sub-national units, since they count as local tax revenues only those strictly determined by sub-national governments.18 In this study, we employ both the decentralization measures developed by Stegarescu (2005b) and the decentralization indicators of the public finance statistics reported in the GFS database. More precisely, we employ (i) the tax revenue decentralization indicator, (TaxRevDec) and (ii) the revenue decentralization indicator as constructed by Stegarescu (RevDec);19 (iii) the GFS expenditure decentralization measure (DecGFS1) and (iv) the GFS revenue decentralization measure (DecGFS2).20 Finally, following the paradigm of Jin and Zou (2002), we also use measures of revenue and expenditure autonomy of the local governments. These measures are: (i) TaxAut, defined as sub-central government own tax revenue as a share of sub-central government total tax revenue (obtained from Stegarescu, 2005b); (ii) TaxAutGFS, defined as sub-national tax revenue as a share of sub-national revenue and grants (taken from GFS, 2002); (iii) VertImb1, defined as transfers from other levels of government as a share of sub-national 16 Stegarescu (2005b) finds that the GFS measure of tax revenue decentralization overestimates the extent of fiscal decentralization. This is particularly the case for Austria (28.4% versus 3.5%), Belgium (44.4% versus 24.6%) and Germany (49.4% versus 7.3).The percentages refer to data from 1999 and 2000. 17 See Appendix A for details. 18 To our knowledge, empirical application of these measures is limited to Fiva (2006), Stegarescu (2005a) and Lessmann (2006). 19 The RevDec indicator represents the vertical structure of all the sources of public revenue. Thus, compared with the TaxRevDec index, it additionally accounts for the structure of non-tax revenue (e.g. user charges or operational surplus of public enterprises). 20 See Appendix A for details and definitions and Appendix C for correlations between these measures. 9 expenditures; and (iv) VertImb2, defined as transfers from other levels of government as a share of sub-national revenues and grants.21 The latter two indices (both taken from the GFS database) are measures of vertical fiscal imbalance and are expected to affect PSE negatively. This is because voters view intergovernmental grants and “own resources” through different lenses and they are more likely to sanction overspending by politicians when local governments are purely financed by intergovernmental grants (see e.g. Oates 1979, 1991).22 In contrast, TaxAut and TaxAutGFS are expected to affect PSE positively, as higher values for both variables indicate higher degree of tax autonomy. 3.3. Other controls To ensure correct econometric identification, we use a series of additional controls. First, we employ a standard demographic variable, namely the dependency ratio of the population (i.e. the share of population aged below 16 and above 65 to total population), denoted as depend. Depend is expected to exert a negative impact on PSE. This is because a higher proportion of economically dependent population generates fiscal needs for programs targeted towards the dependant group. Note that these programs are mainly transfers that do not directly affect (what is assumed here to be) government output, while at the same time they increase total government spending.23 Data for depend are obtained from the WDI (2004). In order to control for the overall level of productivity in the economy, we employ a total factor productivity growth index (denoted as TFP), which is estimated as the residual of the regression of the growth rate of per capita capital on the growth rate of per capita output, for each country in our sample (see Solow, 1957; Barro and Sala-i-Martin, 2004). Countries that present higher productivity growth are expected to be characterized by higher productivity in their public sectors as well. Note, that the causality between TFP and PSE may be reverse (i.e. higher PSE may lead to increased overall productivity). Therefore, in the empirical specification, we assume that these variables are endogenous. 21 See Appendix A for details and definitions and Appendix C for correlations between these measures. As it is usually suggested, intergovernmental grants create the picture that local public spending is funded by non-residents. This is because local voters within a central legislature receive benefits from grants without internalizing their full cost (Weingast el al 1988, Rattsø 2000). 23 This effect holds only for the measures of government output we employ here. Certainly, as transfer programs are expected to create additional effects in the economy (e.g. changes in inequality) they also affect other government activities and, therefore, may represent forms of government output. Had we measured the efficiency of government in providing such outputs, the relationship between Depend and government efficiency may have been different from the one suggested in the text. 22 10 Moreover, we employ two measures of the constraints that the government faces, (i) a simple openness indicator (i.e. total trade over GDP) corrected for country size24 (Open) and (ii) an index of government regulation (EconFreedom) as measured by Gwartney and Lawson (2004), in which higher values reflect less interventionist governments. For three main reasons we expect both variables to be positively associated with PSE. First, more trade and domestic restrictions create rents and therefore higher waste through rent seeking activities (Krueger, 1974; Gatti, 1999). Second, lower international openness and greater government intervention imply lower product market competition within the country, which is also associated with increased government waste (Ades and Di Tella, 1999). Finally, within an international setting, the domestic government must be more efficient if it seeks to attract foreign investors (Wei, 2000). To measure the propensity of the state to employ redistributive policies, we use a measure of population heterogeneity. According to Alesina et al. (2003), La Porta et al. (1999) and Alesina and La Ferrara (2005), countries with high ethno-linguistic fractionalization are expected to exhibit inferior government performance. Once again many reasons have been put forward to justify this relationship. First, high ethnic fractionalization results into pressures for redistribution between groups (Easterly and Levine, 1997). Moreover, it may lead to high demand for publicly provided private goods, especially those that can be targeted towards specific groups (Alesina et. al., 2003). It is also possible that a relationship between fractionalization and corruption is formed. Finally, in more extreme circumstances, increased ethnic fractionalization may lead to ethnic hatred and ultimately to violent civil wars that disrupt the workings of government (see Fearon, 2003). Following Easterly and Levine (1997), we control for ethno-linguistic fractionalization using a Herfindahl index (named Ethnolig), which is calculated on the basis of the share of each 24 There are several reasons why the uncorrected for size measure of openness does not correctly reflect the constraints that the economy faces from the international environment. The first is a simple statistical bias: when a large and a small country trade with each other, the volume of trade is the same for both, but trade shares as a portion of GDP differ. Second, in the presence of increasing returns to scale in production, the market size affects the overall level of productivity. This argument goes back to Adam Smith who argued that the size of the market imposes a constraint on the division of labor. Therefore, small countries that are relatively closed to international trade must experience a lower level of productivity. Ades and Glaeser (1999), Wacziarg (2001) and Alesina et al. (2000) provide empirical evidence consistent with these ideas: large countries experience smaller dynamic gains from trade. Finally, according to Frankel and Romer (1999, p.382), “smaller countries may engage in more trade with other countries simply because they engage in less within-country trade”. For further details on the relationship between openness and country size see also Alesina and Spolaore (2003, chapter 6) and Alesina and Wacziarg (1998, pp.306-307). To correct for this bias we run a regression with total trade over GDP as the dependent variable and the share of country i's GDP to the average GDP of our sample (at the same time period) as the independent variable. Then, we use the residuals from this regression as an indicator for openness (see Bretschger and Hettich, 2002). 11 separate ethno-linguistic group over total population (data are obtained from La Porta et al., 1999). The final set of controls we employ includes variables that refer to the structure of the elected government. Therefore, we use the variable NSM, taken from Mierau et al. (2007), which reports the number of ministers that directly use part of the government budget (i.e. the total number of ministers excluding the minister of finance). Since we expect that these ministers care about the size of the budget they control,25 the relationship between NSM and PSE should be negative.26 Finally, the variable coalition, taken from Tavares (2004), is a dummy variable taking the value 1 if a coalition cabinet that includes ministers from two or more parties is in power. As the number of parties involved in the government increases, the accountability of each of the parties usually diminishes, thus providing fewer incentives for efficiency. In addition, coalition governments are typically associated with a shorter life span (see Schofield, 1993; Müller and Strøm, 2000) and therefore are less concerned with superior performance. 4. Empirical methodology Based on the theoretical considerations of the previous section, we estimate the following empirical model: pit = α 0 + β k zit + uit (2) where pit are government efficiency estimates derived from Program (1) and zit are the set of explanatory variables described above. Unfortunately, it turns out that estimation of Eq. (2) is not a trivial econometric issue. In particular, when non-parametrically derived measures (like the DEA efficiency scores) are regressed against a number of determinants, conventional censored regressions (such as Tobit regressions) yield biased results.27 Only very recently Simar and Wilson (2007) proposed a robust procedure to overcome the associated difficulties. Specifically, they offer an algorithm, comprised of subsequent steps, that begins with a truncated regression and ends with the estimation of confidence intervals. Still, as discussed above, the total factor productivity variable may be endogenous in the PSE measures. To account for this 25 Ministers care about the size of the budget they receive for many reasons, which may include participation in rent-seeking activities, increase in the size of the bureau they control (Niskanen, 1973) and the ability to make income transfers as a means for controlling a larger political clientele. 26 This effect is consistent with the idea that there may exist diseconomies of scale in the administration of government (see Stein, 1997). 27 This is mainly due (but not limited) to the fact that DEA efficiency estimates are serially correlated (for a proof and further details, see Simar and Wilson, 2007). 12 endogeneity we follow the methodology put forth by Khan and Lewbel (2007), who for the first time suggested a truncated regression model with endogenous regressors.28 To this end, we merge the algorithm suggested by Simar and Wilson (2007) with the two-stage least squares truncated regression model put forth by Khan and Lewbel (2007). We consider all observations as cross sections and therefore we drop subscript t in Eq. (2). Consequently, the following procedure may be used to provide inference on the determinants of PSE: ) 1. Obtain maximum likelihood estimates αˆ k of α k and σ u of σ u in the endogenous truncated regression of pˆ i on its k determinants (zi) in Eq. (2), where pˆ i ≤1. The instrument used is the one period lag of the endogenous variables (i.e. lagged one period TFP).29 2. Loop over the next three steps L=2000 times to obtain a set of bootstrap L estimates Βi = (αˆ * , σˆ u* )b b =1 For each i=1,…,m, draw ui from the N (0, σˆ u2 ) distribution with left-truncation at (1 − zi aˆ ) . For details on how to draw from a left-truncated normal distribution see the Appendix of Simar and Wilson (2007). Again for each i=1,…,m, compute pi* = ziαˆ + ui . Use the maximum likelihood method to estimate the endogenous truncated regression of pi* on zi , yielding estimates µ µ* ,νν* . 3. Use the bootstrap values in B and the original estimates α , σ u to construct estimated confidence intervals for each element of α and for σ u . This is done by using the jth element of each bootstrap value α̂ * to find values µπ* ,ν π* such that Pr −ν π ≤ (αˆ *j − αˆ j ) ≤ µπ* ≈ 1 − π , for some small conventional value of π , π = 0.05 in the present analysis. The approximation improves as L → ∞ . Substituting µπ* ,ν π* for µπ ,ν π in Pr −ν π ≤ (αˆ j − α j ) ≤ µπ = 1 − π leads to an estimated confidence interval (αˆ j + µπ* , αˆ j +ν π* ) . 28 Their simulation results show that their new estimator performs well, while they specifically state that their method is applicable in general contexts involving two-stage analyses with a nonparametric first stage, such as ours. 29 The results of the paper remain intact if more lags as instrumental variables. 13 5. Results In this section we discuss the results obtained by estimating Eq. (2), using the data described in Section 3 and the empirical methodology presented in the previous section. The baseline results are presented in Tables 1 and 2, while the extensive sensitivity analyses performed are presented in Tables 3 and 4.30 5.1. Basic results Table 1 reports the results of the regressions of PSE on alternative measures of fiscal decentralization. In column 1, the government efficiency estimates are regressed on TaxRevDec, as well as on our set of control variables (i.e. Coalition, NSM, EconFreedom, Depend, TFP, Open, Ethnolig). In all estimated equations, we include regional dummies (see Appendix A) and a time trend. Evidently, the coefficient on TaxRevDec bears a positive sign and is significant at the 1% level, suggesting a strong positive link between fiscal decentralization and PSE. This result is aligned with the propositions of the theoretical debate discussed in Section 2. Focusing on the rest of the explanatory variables, we observe that coefficients on both Coalition and NSM present negative signs and are significant at conventional levels, indicating that coalition governments and large cabinets exert a negative impact on PSE. In contrast, EconFreedom is positive and highly significant, whereas the coefficient on Open is positive and marginally significant. These results may be explained by the beneficial effects of internal and external market constraints on the function of governments. Finally, Depend and Ethnolig appear to be insignificant determinants of PSE. In column 2, we re-estimate the model by using Revdec, instead of TaxRevDec, as a proxy for fiscal decentralization. As we have already pointed out, the Revdec indicator is a more general measure of fiscal decentralization, since it accounts for the vertical structure of non-tax revenues (such as user charges and operational surpluses of public enterprises), which is not encompassed in TaxRevDec.31 Markedly, the main result of this second specification remains unaffected. The coefficient on Revdec is positive and highly significant, thus confirming the positive effect of fiscal decentralization on government efficiency. As regards the rest of the explanatory variables, we observe that only slight changes in the results emerge (compared to those presented in column 1). Specifically, the coefficients on Coalition and TFP appear to lose their statistical significance, whereas 30 Note that in all estimated equations presented in the first two tables, we include regional dummies (see Appendix A), which are not reported to save space. The full set of results is available upon request. 31 For details on this see Appendix A. 14 Ethnolig enters with a positive and significant coefficient. It should be noted that this result is not in line with our theoretical priors, since it implies that the relationship between ethnolinguistic fractionalization and PSE is positive. In column 3 we employ as a proxy for fiscal decentralization the GFS expenditure decentralization measure (DecGFS1), which is also found to be positively and significantly related to government efficiency. As a final test, in column 4 we use DecGFS2, reaching similar conclusions. In these last two specifications, the behavior of the control variables is much similar to that observed in column 2. Table 2 presents the results of the regressions of PSE on the two alternative measures of tax autonomy and the two alternative measures of vertical fiscal imbalance. In column 1, the DEA government efficiency estimates are regressed on the Stegarescu (2005b) tax autonomy measure (denoted as TaxAut), as well as on the rest of the explanatory variables. The results suggest that TaxAut bears a positive sign and appears to be highly significant, which is consistent with our theoretical priors outlined in Section 2. Indeed, increased tax autonomy (or alternatively decreased dependency of local governments on intergovernmental transfers) explains, at least in OECD economies, higher levels of PSE. As suggested above, this relationship is probably associated with the adverse effect of tax autonomy due to the “common pool” problem. Turning to the rest of the explanatory variables, we observe that our results are similar to those presented in Table 1. More precisely, the coefficients on Coalition and NSM are negative and significant, whereas the coefficients on EconFreedom, TFP and Open are positive and significant. Finally, Depend and Ethnolig appear to be insignificant at conventional levels of statistical significance. In column 2 we employ the GFS tax autonomy measure (denoted as TaxAutGFS) instead of the TaxAut and we re-estimate Eq. (2). As expected, TaxAutGFS enters with a positive sign and is significant at the 1% level, validating the positive relationship between tax autonomy and PSE. Concerning the rest of the variables, our results remain practically unaffected, with the exception of Coalition (which loses its significance) and Ethnolig (which presents a positive and significant effect on PSE). In column 3, we employ a vertical fiscal imbalance measure (denoted as VertImb1) in order to capture the fiscal dependency of local governments on intergovernmental transfers. We observe that the coefficient on VertImb1 bears a negative sign and is significant at the 1% level. This result is in accordance with our previous findings on the relationship between tax autonomy and PSE, as well as the dominant view of the theoretical literature (see e.g. Oates 1979, 1991). A similar result is reached by using an alternative fiscal dependency measure, 15 (i.e. VertImb2 in column 4). In both estimations presented in columns 3 and 4, the behavior of the control variables is similar to that observed in column 2. 5.2. Sensitivity analysis In this section we inquire into the robustness of our results. First, we examine the sensitivity of our estimates with respect to individual outliers, as well as with respect to regional characteristics. Next, we re-estimate our model using five year averages, in order to ensure that our results are not driven by the noise generated by annual data. Finally, we reestimate our benchmark model using an alternative measure of government efficiency, namely the economic stability (EcStab) indicator, as defined above.32 Seeing that our sample consists of 21 OECD countries, which are quite heterogeneous in many aspects, we first examine the sensitivity of our estimates to individual outliers or to regional characteristics. To account for the first issue, we re-estimate our benchmark model, this time excluding all observations with an error term in the upper or lower 5th percentile (i.e. we drop 10% of our sample). The results, presented in column 1 and 2 of Table 3, indicate that the conclusions presented in Sections 5.1 are firm as regards the influence of individual outliers. The second issue that relates to the potential effect of regional characteristics has been (partially) addressed by including the three dummies Scandinavian, Anglo-Saxon, Mediterranean in the empirical models. Here we further examine whether the results change when we exclude each of these groups in turn. The results, presented in columns 5 to 10, suggest that the effects of Taxrevdec and TaxAut on PSE remain positive and significant at conventional levels. Another potential drawback of the analysis of the previous section is the annual nature of the dataset. This was the preferred choice with the aim of increasing the size of our sample, which may however come at the expense of also increasing the noise in the data.33 This would imply that observed changes in PSE may be due to random factors (such as the business cycle), which are not necessarily related to changes in the explanatory variables. For this reason, we re-examine our two propositions using simple five-year averages of our data. The results, reported in column 3 and 4 of Table 3, suggest that even though for most of the control variables the statistical and economic significance drops, the variables of main 32 For expositional brevity, in Table 3 we present the results from using only Taxrevdec as the measure of decentralization. We have verified, however, that our results carry through to the rest of the measures used in Table2 1 and 2. 33 Rodden (2003) underlines the importance of panel studies in examining fiscal decentralization and argues that cross-national studies fail to capture important aspects of this issue, since they do not account for the fact that the process of decentralization unfolds overtime. 16 interest remain practically unaffected. We attribute the increase in the standard errors to the fact that our sample is now confined to about 100 observations and it is well-known that maximum likelihood estimators usually produce a bias in small samples, with this bias diminishing as the sample increases. Nonetheless, as the coefficients on TaxRevDec and TaxAut remain significant, we are confident that the positive relationship in hand is robust to the larger time span of our observations. An important sensitivity analysis involves estimation of Eq. (2) using a different dependent variable (namely EcStab) that looks into economic stability as the ultimate goal of governments.34 Once again, we use a number of alternative specifications that capture the relationship between fiscal decentralization and PSE and between fiscal dependency and PSE. The results, reported in Table 4, suggest no discrepancy from previous findings: Increased fiscal decentralization and tax autonomy exert a positive impact on PSE, with the results in some cases being enhanced compared to their counterparts of Tables 1 and 2.35 Concerning the rest of the explanatory variables our results also remained analogous to our previous findings. 6. Conclusions In this paper we specified an empirical framework to investigate the effect of fiscal decentralization on public sector efficiency. With this aim we (i) directly measured PSE using DEA, thereby specifying an underlying production process of public goods; and (ii) examined the impact of the variables of interest on PSE via the amalgamation of two prominent semi-parametric techniques. Therefore, we proceeded in two stages. The first involved estimation of PSE, in terms of assuming governments to aim for either economic performance or stability, while the second entailed regressing the PSE scores derived in stage 1 on a number of well-established indicators for fiscal decentralization. The analysis was carried out on a panel that included 21 OECD economies over the period 1970-2000. Backed by strong empirical results, obtained from a number of different specifications and sensitivity analyses, we contend that public sector efficiency is increasing with fiscal decentralization. This relationship calls for a deeper understanding of the inter- and intra-country mechanisms that shape it; however, before moving on to another issue, we have better bring this entry to a close. 34 Note that until now we assumed that governments aim at improved economic performance. The vertical fiscal imbalance indicators enter the estimated equations (columns 5 and 6) with negative and significant coefficients, much like in Table 2. 35 17 Table 1 Public sector efficiency and fiscal decentralization PSE(EcPerf) Taxrevdec (1) (2) (3) 0.005*** (5.99) Revdec 0.004*** (4.19) DecGFS1 0.002** (2.35) DecGFS2 Coalition NSM EconFreedom Depend TFP Open Ethnolig Obs Wald Sigma (4) -0.068** (-2.46) -0.008*** (-2.94) 0.104*** (4.90) 0.227 (0.94) 1.802*** (2.98) 0.003** (1.97) -0.166 (-1.34) 495 216.72 0.237 -0.015 (-0.55) -0.020*** (-8.32) 0.146*** (8.01) 0.311 (1.34) 0.769 (1.57) -0.001 (-0.75) 0.373*** (2.99) 398 363.87 0.193 -0.042* (-1.70) -0.018*** (-7.84) 0.135*** (7.28) 0.191 (0.86) 0.447 (0.02) 0.001 (0.78) 0.561*** (5.23) 468 308.51 0.205 0.003*** (2.69) -0.038 (-1.53) -0.019*** (-7.90) 0.138*** (7.38) 0.239 (1.06) 0.446 (0.02) 0.001 (0.33) 0.511*** (4.60) 468 316.53 0.204 Note: **,*** denote statistical significance at 5% and 1% level of statistical significance respectively. Country dummies are included in all estimated equations 18 Table 2 Public sector efficiency and fiscal dependency of local governments PSE(EcPerf) TaxAut (1) (2) (3) 0.001** (2.62) TaxAutGFS 0.002*** (3.49) VertImb1 -0.002*** (-2.77) VertImb2 Coalition NSM EconFreedom Depend TFP Open Ethnolig Obs Wald Sigma (4) -0.088*** (-3.55) -0.013*** (-5.34) 0.119*** (5.91) -0.267 (-1.17) 1.441*** (2.99) 0.006*** (3.43) 0.188* (1.75) 581 214.60 0.245 -0.039 (-1.52) -0.018*** (-7.67) 0.119*** (6.05) -0.045 (-0.21) 1.000** (2.16) 0.001* (1.86) 0.361*** (3.16) 473 268.04 0.214 -0.011 (-0.45) -0.019*** (-8.16) 0.128*** (7.10) 0.147 (0.71) 1.286*** (2.88) 0.000 (-0.15) 0.610*** (5.56) 469 331.05 0.203 -0.003*** (-3.46) -0.012 (-0.45) -0.017*** (-6.86) 0.115*** (5.96) 0.078 (0.36) 1.231*** (2.69) 0.001** (2.41) 0.465*** (4.08) 469 289.10 0.210 Note: *, **, *** denote statistical significance at 10%, 5% and 1% level of statistical significance respectively. Country dummies are included in all estimated equations 19 Table 3 Sensitivity analysis I: Accounting for outliers, short-run dynamics and regional effects PSE(EcPerf) Taxrevdec (1) 0.005*** (5.95) TaxAut Coalition NSM EconFreedom Depend TFP Open Ethnolig dscand das dmed Obs Wald Sigma (2) -0.069** (-2.51) -0.008*** (-2.82) 0.118*** (5.01) 0.322 (1.21) 1.722*** (2.71) 0.003* (1.90) -0.141 (-1.22) 0.055* (1.83) 0.046 (1.12) 0.171*** (3.29) 446 233.72 0.214 (3) (4) 0.041*** (2.68) 0.001*** (2.77) -0.075*** (-3.10) -0.011*** (-4.40) 0.116*** (5.63) -0.130 (-1.22) 1.426** (2.39) 0.005** (2.48) 0.139 (1.40) 0.101* (1.80) 0.048 (1.31) 0.164*** (3.02) 523 213.75 0.234 -0.125** (-2.17) -0.011* (-1.81) 0.071** (2.21) 0.176 (0.43) 3.900* (1.79) 0.007** (2.24) 0.329* (1.67) 0.103* (1.93) 0.141* (1.82) 0.254*** (3.92) 100 82.92 0.192 0.001** (2.34) -0.107* (-1.82) -0.007 (-1.28) 0.053* (1.82) 0.398 (0.96) 6.199*** (3.39) 0.006 (1.58) 0.667*** (3.38) 0.046 (0.80) 0.121* (1.85) 0.186*** (2.69) 117 82.18 0.198 (5) (6) (7) 0.008*** (7.52) 0.005*** (5.39) 0.003*** (3.22) -0.029 (-0.76) -0.004 (-1.26) 0.136*** (5.32) 0.652** (2.26) 3.134*** (3.93) 0.002 (0.95) -0.400*** (-3.01) 0.003 (0.07) 0.250*** (4.77) 356 242.67 0.238 -0.083*** (-2.79) -0.009*** (-3.18) 0.111*** (4.96) 0.099 (0.39) 1.969*** (3.03) 0.003* (1.74) -0.092 (-0.66) 0.069** (2.04) 0.030 (0.73) 450 220.15 0.240 -0.116*** (-3.45) -0.015*** (-4.27) 0.116*** (4.10) 1.132*** (3.07) 0.732 (1.02) 0.018*** (3.98) 0.166 (1.05) 0.100** (2.59) 0.159*** (3.19) 352 168.97 0.237 (8) (9) (10) 0.001** (2.43) -0.073** (-2.27) -0.012*** (-4.75) 0.144*** (6.03) -0.270 (-1.05) 1.840*** (3.21) 0.005*** (3.16) 0.158 (1.41) 0.001*** (3.56) -0.111*** (-3.74) -0.013*** (-5.24) 0.129*** (5.97) -0.428* (-1.69) 1.391** (2.64) 0.004*** (2.76) 0.266** (2.23) 0.116*** (3.62) -0.007 (-0.20) 0.001** (2.17) -0.088*** (-2.85) -0.017*** (-6.12) 0.136*** (5.36) 0.606* (1.88) 1.050** (1.98) 0.019*** (4.88) 0.267* (1.82) 0.127*** (3.59) -0.002 (-0.05) 0.229*** (5.09) 440 223.73 485 219.52 0.259*** (5.66) 436 194.93 Note: *, **, *** denote statistical significance at 10%, 5% and 1% level of statistical significance respectively. Country dummies are included in all estimated equations 20 Table 4 Sensitivity analysis II: Using economic stability as a measure of government performance PSE(EcSstab) Taxrevdec Revdec DecGFS1 DecGFS2 VertImb1 VertImb2 TaxAut (1) (2) (3) (4) (5) (6) (7) 0.007*** (7.26) 0.007*** (6.80) 0.004*** (3.23) 0.002** (2.09) -0.002** (-2.18) -0002** (-2.59) 0.003*** (8.04) TaxAutGFS Coalition NSM EconFreedom Depend TFP Open Ethnolig Obs Wald Sigma (8) -0.008 0.029 -0.016 -0.015 -0.006 -0.004 -0.044* (-0.27) (0.93) (-0.56) (-0.52) (-0.19) (-0.13) (-1.68) 0.004 0.001 -0.001 -0.001 -0.001 -0.001 0.001 (1.49) (0.18) (-0.37) (-0.46) (-0.46) (-0.40) (0.24) 0.147*** 0.155*** 0.153*** 0.149*** 0.136*** 0.137*** 0.102*** (6.13) (7.13) (6.99) (6.73) (6.26) (6.11) (4.75) 0.803*** 0.728** 0.548** 0.457* 0.296 0.216 -0.099 (2.97) (2.64) (2.10) (1.72) (1.19) (0.84) (-0.41) 1.532** 1.247** 1.233** 1.206** 1.261** 1.265** 1.567*** (2.25) (2.13) (2.35) (2.28) (2.35) (2.29) (3.06) 0.002 0.000 0.004* 0.004* 0.004** 0.005** 0.007*** (1.10) (0.00) (1.90) (1.71) (2.13) (2.39) (3.73) 0.031 0.444*** 0.883*** 0.871*** 1.022*** 0.931*** 0.403*** (0.22) (2.99) (7.00) (6.60) (7.73) (6.91) (3.53) 495 398 468 468 469 469 581 245.16 287.78 259.97 254.68 258.75 215.93 281.69 0.264 0.233 0.243 0.244 0.244 0.259 0.003*** (2.87) -0.012 (-0.40) -0.001 (-0.31) 0.144*** (6.44) 0.212 (0.83) 1.135** (2.06) 0.005** (2.41) 0.803*** (6.03) 473 212.18 Note: *, **, *** denote statistical significance at 10%, 5% and 1% level of statistical significance respectively. Country dummies are included in all estimated equations 21 Appendix A : Data Sources and Descriptive Statistics Description Obs. Mean Std. min max Sources Dev. DEA efficiency scores when the output is Economic Performance DEA efficiency scores when the output is Economic Stability Sub-Central Government own tax revenue as a share of General Government total tax revenue. Sub-Central Government own tax and non-tax revenue as a share of General Government total tax revenue Sub-national Expenditures as a share of total expenditures Sub-national Revenues as a share of total revenues Sub-Central Government own tax revenue as a share of Sub-Central Government total tax revenue Sub-national tax revenues as a share of sub-national revenues and grants Transfers from other levels of government as a share of subnational expenditures. Transfers from other levels of government as a share of subnational revenues and grants Dummy variable taking the value 1 if a coalition cabinet is in power Number of Spending Ministers Index of Economic Regulation 630 0.58 0.29 0.28 1 Own calculations based on Afonso et al. (2005). 630 0.61 0.32 0.27 1 Own calculations based on Afonso et al. (2005). 522 22.40 17.09 0.27 61.50 Stegarescu (2005b) 403 25.23 15.89 4.13 64.69 Stegarescu (2005b) 481 31.70 13.76 1.45 59.18 481 23.20 14.21 1.61 54.60 Government Financial Statistics. IMF (2002) Government Financial Statistics. IMF (2002) 623 77.27 33.91 2.83 100 Stegarescu (2005b) 486 40.59 17.09 2.15 108.73 Government Financial Statistics. IMF (2002) 483 40.04 16.61 8.39 82.00 Government Financial Statistics. IMF (2002) 483 41.42 17.70 8.45 86.41 Government Financial Statistics. IMF (2002) 609 0.55 0.49 0.00 1.00 Tavares (2004) 633 15.30 4.79 5.00 33.00 Mierau et al. (2007) 651 5.89 0.81 4.30 8.30 Population 16- and 65+ as a share of total population 651 0.52 0.05 0.44 0.74 TFP Total Factor Productivity 630 0 0.02 -0.20 0.10 Open Residuals from regression of Size on (Exports+Imports)/GDP 630 0 10.77 -43.35 78.83 Gwartney and Lawson (2006) World Bank Development Indicators(WBDI) (2004) Own Calculations as described in Section 3.3. Data from Penn World Tables (2006) Own Calculations as described in Section 3.3. Data from WBDI (2004) Index of ethno-linguistic fractionalization 651 0.13 0.11 0.003 0.376 PSE (EcPerf) PSE (EcStab) Taxrevdec Revdec DecGFS1 DecGFS2 TaxAut TaxAutGFS VertImb1 VertImb2 Coalition NSM EconFreedom Depend Ethnolig La Porta et al. (1999) 22 Appendix B : Graphs of PSE(EcPerf) and PSE(EcStab) DEA efficiency scores for EcPerf (10-years averages, period 1970-1979) 1.20 1.00 0.80 0.60 0.40 SWI UK USA SWI UK USA SWI UK USA SWI UK USA SWE SWE SWE SWE SPA SPA SPA SPA POR POR POR POR NLD NOR NOR LUX JPN ITA IRL GRE FRA GER FIN DNK CAN BEL AUT 0.00 AUS 0.20 DEA efficiency scores for EcPerf (10-years averages, period 1980-1989) 1.20 1.00 0.80 0.60 0.40 LUX NLD JPN ITA IRL GRE FRA GER FIN DNK CAN BEL AUT 0.00 AUS 0.20 DEA efficiency scores for EcPerf (10-years averages, period 1990-1999) 1.20 1.00 0.80 0.60 0.40 0.20 NLD NOR LUX JPN ITA IRL GRE GER FRA FIN DNK CAN BEL AUT AUS 0.00 DEA efficiency scores for EcPerf, year 2000 1.20 1.00 0.80 0.60 0.40 NLD NOR LUX JPN ITA IRL GRE GER FRA FIN DNK CAN BEL AUT 0.00 AUS 0.20 Note: Estimates of PSE(EcPerf) for Gernany prior to 1991 and for Japan after 1994 are not available due to the unavailability of the data for unemployment and public spending respectively. 23 DEA efficiency scores for EcStab (10-years averages, period 1970-1979) 1.20 1.00 0.80 0.60 0.40 0.20 UK USA SWE SWE SWI SPA POR SPA POR NOR NLD NOR LUX JPN ITA IRL GRE GER FRA FIN DNK CAN BEL AUT AUS 0.00 DEA ef f iciency scores f or EcStab (10-y ears av erages, period 1980-1989) 1.20 1.00 0.80 0.60 0.40 0.20 UK SPA SWE SWI UK USA SPA SWE SWI UK USA USA SWI POR POR NLD LUX JPN ITA IRL GRE GER FRA FIN DNK CAN BEL AUT AUS 0.00 DEA efficiency scores for EcStab (10-years averages, period 1990-1999) 1.20 1.00 0.80 0.60 0.40 0.20 NLD NOR LUX JPN ITA IRL GRE GER FRA FIN DNK CAN BEL AUT AUS 0.00 DEA efficiency scores for EcStab, year 2000 1.20 1.00 0.80 0.60 0.40 0.20 NLD NOR LUX JPN ITA IRL GRE GER FRA FIN DNK CAN BEL AUT AUS 0.00 Note: Estimates of PSE(EcStab) for Austria, for Germany prior to 1991 and for Japan after 1994 are not available due to the unavailability of the data for standard deviation of growth, inflation and standard deviation of growth and public spending respectively. 24 Appendix C : Correlation Matrix DecGFS1 DecGFS2 Taxrevdec Revdec VetrImb VertImb2 TaxAut TaxAut GFS Coalition NSM Depend Econ Freedom TFP Open DecGFS1 1 DecGFS2 0.93 1 Taxrevdec 0.71 0.78 1 Revdec 0.81 0.87 0.97 1 VetrImb -0.45 -0.70 -0.52 -0.57 1 VertImb2 -0.47 -0.73 -0.53 -0.59 0.97 1 TaxAut 0.19 0.033 0.45 0.35 0.29 0.29 1 TaxAutGFS 0.44 0.66 0.57 0.57 -0.92 -0.86 -0.21 1 Coalition -0.02 -0.06 -0.13 -0.08 0.11 0.13 -0.13 -0.07 1 NSM -0.14 -0.12 -0.08 -0.16 -0.06 0.01 0.11 0.17 -0.25 1 Depend -0.30 -0.35 -0.22 -0.27 0.22 0.27 0.19 -0.27 -0.23 0.09 1 Econ Freedom 0.32 0.26 0.30 0.32 0.09 0.04 0.42 -0.09 -0.21 0.07 -0.07 1 TFP -0.02 -0.02 -0.05 -0.04 0.07 0.06 -0.05 -0.05 -0.03 -0.02 -0.03 0.04 1 Open 0.24 0.31 0.20 0.27 -0.18 -0.21 0.045 0.12 -0.16 -0.20 -0.01 0.29 0.02 1 Ethnolig 0.10 0.20 0.45 0.41 -0.09 -0.07 0.22 0.14 -0.11 -0.04 -0.20 0.19 -0.06 0.063 Ethnolig 1 25 References Ades, A., di Tella, R., (1999). Rents, competition and corruption. American Economic Review 89, 982-993. Ades, A., Glaeser, E.L., (1999). Evidence on growth, increasing returns and the extent of the market. Quarterly Journal of Economics 114, 1025-1046. Alesina, A., Devleeschauwer, A., Easterly, W., Kurlat, S., Wacziarg, R., (2003). Fractionalization. Journal of Economic Growth 8, 155-194. Alesina, A., La Ferrara, E., (2005). Preferences for redistribution in the land of opportunities. Journal of Public Economics 89, 897-931. Alesina, A., Spolaore, E., (2003). The Size of Nations. Cambridge: MIT Press. Alesina, A., Spolaore, E., Wacziarg, R., (2000). Economic integration and political disintegration. American Economic Review 90, 1276-1296. Alesina, A., Wacziarg, R., (1998). Openness, country size and government. Journal of Public Economics 69, 305-321. Afonso, A., St. Aubyn, M., (2005). Non-parametric approaches to education and health: Expenditure efficiency in OECD countries. Journal of Applied Economics 8, 227-246. Afonso, A., Schuknecht, L., Tanzi, V., (2005). Public sector efficiency: An international comparison. Public Choice 123, 321-347. Afonso, A., Schuknecht, L., Tanzi, V., (2006). Public sector efficiency: Evidence for new EU member states and emerging markets. ECB Working Paper no. 581. Angelopoulos, K., Philippopoulos, A., (2005). The role of government in anti-social redistributive activities. CESifo Working Paper no. 1427. Angelopoulos, K., Philippopoulos, A., Tsionas, E., (2007). Does public sector efficiency matter? Revisiting the relation between fiscal size and economic growth in a world sample. Mimeo. Balaguer-Coll, M.T., Prior, D., Tortosa-Ausina, E., (2007). On the determinants of local government performance: A two-stage nonparametric approach. European Economic Review 51, 425-451. Barankay, I., Lockwood, B., (2007). Decentralization and the productive efficiency of government: Evidence from Swiss cantons. Journal of Public Economics 91, 11971218. Bardhan, P., Mookerjee, D., (2000). Capture Governance at Local and National Levels. American Economic Review 90, 135-139. Barro, R., Sala-i-Martin, X., (2004). Economic Growth. Second edition. Cambridge: MIT Press. 26 Belleflamme, P., Hindriks, J., (2005). Yardstick Competition and Political Agency Problems, Social Choice and Welfare 24, 155-169. Besley, T., Case, A., (1995). Incumbent Behavior: Vote-Seeking, Tax-Setting, and Yardsick Competition. American Economic Review 85, 25-45. Besley, T., Smart, M., (2007). Fiscal Restraints and Voter Welfare. Journal of Public Economics 91, 755-773. Bodenstein, M., Ursprung, H., (2001). Political yardstick competition, economic integration, and constitutional choice in a federation. CESifo Working Paper no. 501. Bordignon, C., Colombo, L., Galmarini, U., (2003). Fiscal Federalism and Endogenous Lobbies' Formation, CESifo Working Paper no. 1017. Brennan, G., Buchanan, J., (1980). The power to tax: analytical foundations of a fiscal constitution. New York: Cambridge University Press. Bretschger, L., Hettich, F., (2002). Globalization, capital mobility and tax competition: Theory and evidence for OECD countries. European Journal of Political Economy 18, 695-716. Coelli, T., Rao, D.S.P., O'Donnell, C.C., Battese, G.E., (2005). An Introduction to Efficiency and Productivity Analysis. New York: Springer. Davoodi, H., Zou, H., (1998). Fiscal decentralization and economic growth: A crosscountry study. Journal of Urban Economics 43, 244–257. Easterly, W., Levine, R., (1997). Africa’s growth tragedy: Policies and ethnic divisions. Quarterly Journal of Economics 111, 1203-1250. Ebel, R. D., Yilmaz, S., (2003). On the measurement and impact of fiscal decentralization, in J. Martinez-Vazquez and J. Alm (eds), Public Finance in Developing and Transitional Countries: Essays in Honor of Richard Bird. Cheltenham: Elgar. Edwards, J., Keen, M., (1996).Tax competition and Leviathan. European Economic Review 40, 113-134. Enikolopov, R., Zhuravskaya, E., (2007). Decentralization and political institutions. Journal of Public Economics 91, 2261-2290. Fearon, J., (2003). Ethnic and cultural diversity by country. Journal of Economic Growth 8, 195-222. Frankel, J.A., Romer, D., (1999). Does trade cause growth? American Economic Review 89, 379-399. Fisman, R., Gatti, R., (2002a). Decentralization and corruption: Evidence across countries. Journal of Public Economics 83, 325–346. 27 Fisman, R., Gatti, R., (2002b). Decentralization and corruption: Evidence from US transfer programs. Public Choice 113, 25-35. Fiva, J., (2006). New evidence on the effects of fiscal decentralization on the size and the composition of government spending. FinanzArchiv 62, 250-280. Gatti, R., (1999). Explaining corruption: Are open countries less corrupt? The World Bank, mimeo. Gupta, S., Verhoeven, M., (2001). The efficiency of government expenditure: Experiences from Africa. Journal of Policy Modelling 23, 433-467. Gwartney, J., Lawson, R., (2004). Economic Freedom of the World: 2004 Annual Report. Vancouver: The Fraser Institute. Hamilton, A., Jay, J., Madison, J., [1787] (1982). The Federalist Papers. New York: Bantam Classic Books. Henderson, J.V, Kuncoro, A., (2004). Corruption in Indonesia. NBER Working Paper, no. 10674. Herrera, S., Pang, G., (2005). Efficiency of public spending in developing countries: An efficiency frontier approach. World Bank Policy Research Working Papers, no. 3645. Hindriks, J., Lockwood, B., (2005). Decentralization and electoral accountability: Incentives, separation, and voter welfare. Warwick Working Papers in Economics, no. 717. IMF, (2002). Government Finance Statistics, CD-ROM. Washington D.C., USA. Jin, J., Zou, H., (2002). How does fiscal decentralization affect aggregate, national and subnational government size? Journal of Urban Economics 52, 270-293. Khalegian, P., (2003). Decentralization and public services: The case of immunization. World Bank Policy Research Working Paper, no. 2989. Khan, S., Lewbel, A., (2007). Weighted and two stage least squares estimation of semiparametric truncated regression models. Econometric Theory 23, 309-347. Krueger, A., (1974). The political economy of the rent-seeking society. American Economic Review 64, 291-303. Kumbhakar, S.C., Lovell, C.A.K., (2000). Stochastic Frontier Analysis. Cambridge: Cambridge University Press. 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. Lessmann, C., (2006). Fiscal decentralization and regional disparity: A panel data approach for OECD countries. Ifo Working Paper No.25. 28 Mello, L., Barenstein, M., (2001). Fiscal decentralization and governance: A crosscountry approach. IMF Working Paper, no. 01/71. Mierau, J., Jong-A-Pin, R., de Haan, J., (2007). Do political variables affect fiscal policy adjustment decisions? New empirical evidence. Public Choice 133, 297-320. Müller, W., Strøm, K., (2000). Coalition Governments in Western Europe. Oxford: Oxford University Press. Myerson, R., (2006). Federalism and incentives for success of democracy. Quarterly Journal of Political Science 1, 3-23. Niskanen, W. A. (1973). Bureaucracy: Servant or master? London, UK: The Institute of Economic Affairs. Oates, W., (1979). Lump-Sum Intergovernmental Grants Have Price Effects. In Fiscal Federalism and Grants-in-Aid, ed. P. Mieszkowski and W. Oakland. Washington, D.C.: The Urban Institute. Oates, W., (1991). On the Nature and Measurement of Fiscal Illusion: A Survey. In Studies in Fiscal Federalism, ed.Wallace Oates. Brookfield: Edward Elgar. OECD (1999). Taxing Powers of State and Local Government. OECD Tax Policy Studies no. 1, Paris. OECD (2002). Fiscal Design Surveys across Levels of Government. OECD Tax Policy Studies no. 7, Paris. OECD (2005). Economic Outlook. OECD (2007). Economic Outlook: Annual and Quarterly Data no.81. Persson, T., Tabellini, G., (2000). Political Economics: Explaining Economic Policy. Cambridge, MA: MIT Press. Prud’homme, R., (1995). On the dangers of decentralization. World Bank Research Observer 10, 201-220. Rattsø, J., (2000). Spending growth with vertical fiscal imbalance: Decentralized government spending in Norway: 1880-1990. Typescript. Norwegian University of Science and Technology. Redoano, M., (2003). Does centralization affect the number and the size of lobbies? Warwick Economic Research Paper, no. 674. Robalino, D.A., Picazo, O.F., Voetberg, V.A., (2001). Does fiscal decentralization improve health outcomes? Evidence from a cross-country analysis. World Bank Policy Research Working Paper, no. 2565. Rodden, J., (2002). The dilemma of fiscal federalism: Grants and fiscal performance around the world. American Journal of Political Science 46, 670-687. 29 Rodden, J., (2003). Revisiting Leviathan: Fiscal federalism and the growth of government. International Organization 57, 695-729. Salmon, P., (1987). Decentralization as an incentive scheme. Oxford Review of Economic Policy 3, 24–43. Schofield, N., (1993). Political competition and multiparty coalition governments. European Journal of Political Research 23, 1-33. Seabright, P., (1996). Accountability and decentralization in government: An incomplete contracts model. European Economic Review 40, 61-91. Shleifer, A., (1985). A theory of yardstick competition. Rand Journal of Economics 16, 319-327. Sijpe, N., Rayp, G., (2007). Measuring and explaining government efficiency in developing countries. Journal of Development Studies 43, 360-381. Simar, L., Wilson, P., (2007). Estimation and inference in two-stage, semi-parametric models of production processes. Journal of Econometrics 136, 31–64. Solow, R., (1957). Technical change and the aggregate production function. Review of Economics and Statistics 39, 312-320. Stegarescu, D. (2005a). Costs, Preferences, and Institutions: An Empirical Analysis of the Determinants of Government Decentralization, ZEW Discussion Paper No. 05-39, Mannheim Stegarescu, D. (2005b). Public Sector decentralization: Measurement concepts and recent international trends. Fiscal Studies 26, 301-333. Stein, E., (1997). Fiscal decentralization and government size in Latin America. In: Fukasaku, K., Hausmann, R., (Eds.), Democracy, Decentralization and Deficits in Latin America. IDB-OECD. Tanzi, V., Schuknecht, L., (1998). Can small governments secure economic and social well-being? In: Grubel, H. (Ed.), How to Spend the Fiscal Dividend: What is the Optimal Size of Government? Vancouver: Fraser Institute. Tanzi, V., Schuknecht, L., (2000). Public Spending in the 20th Century: A Global Perspective. Cambridge: Cambridge University Press. Tavares, J., (2004). Does right or left matter? Cabinets, credibility and fiscal adjustments. Journal of Public Economics 88, 2447-2468. Treisman, D., (2000). The causes of corruption: A cross-national study. Journal of Public Economics 76, 399-457. Treisman, D., (2002). Decentralization and the quality of government. Mimeo, University of California. 30 Tulkens, H., (1993). On FDH efficiency analysis: Some methodological issues and applications to retail banking, courts, and urban transit. Journal of Productivity Analysis 4, 183-210. Wacziarg, R., (2001). Measuring the dynamic gains from trade. World Bank Economic Review 15, 393-429. Wei, S.J, (2000). Local corruption and global capital flows. Brookings Papers on Economic Activity 2, 303-354. Weingast, B., Kenneth, S., Christopher, J., (1988). The political economy of benefits and costs: A neoclassical approach to distributive politics. Journal of Political Economy 89, 642-664. World Bank, (2004). World Bank Development Indicators, CD-ROM. Washington D.C., USA. Zhu, J., (2003). Quantitative Models for Performance Evaluation and Benchmarking. New York: Springer. 31 CESifo Working Paper Series for full list see www.cesifo-group.org/wp (address: Poschingerstr. 5, 81679 Munich, Germany, [email protected]) ___________________________________________________________________________ T T 2301 Harald Badinger and Peter Egger, GM Estimation of Higher Order Spatial Autoregressive Processes in Panel Data Error Component Models, May 2008 2302 Jan K. Brueckner, Slot-Based Approaches to Airport Congestion Management, May 2008 2303 Sören Blomquist, Vidar Christiansen and Luca Micheletto, Public Provision of Private Goods and Nondistortionary Marginal Tax Rates, May 2008 2304 Dan Anderberg and Alessandro Balestrino, The Political Economy of Post-Compulsory Education Policy with Endogenous Credit Constraints, May 2008 2305 Tomer Blumkin, Yoram Margalioth and Efraim Sadka, The Role of Stigma in the Design of Welfare Programs, May 2008 2306 Vesa Kanniainen and Paolo M. Panteghini, Tax Neutrality: Illusion or Reality? The Case of Entrepreneurship, May 2008 2307 Thomas Dohmen, Armin Falk, David Huffman and Uwe Sunde, The Intergenerational Transmission of Risk and Trust Attitudes, May 2008 2308 Guglielmo Maria Caporale and Mario Cerrato, Using Chebyshev Polynomials to Approximate Partial Differential Equations, May 2008 2309 Peter Egger and Doina Maria Radulescu, Labour Taxation and Foreign Direct Investment, May 2008 2310 Laurent Linnemer, Dissipative Advertising Signals Quality even without Repeat Purchases, May 2008 2311 Jordi Jofre-Monseny and Albert Solé-Ollé, Which Communities should be afraid of Mobility? The Effects of Agglomeration Economies on the Sensitivity of Firm Location to Local Taxes, May 2008 2312 Andreas Haufler and Ferdinand Mittermaier, Unionisation Triggers Tax Incentives to Attract Foreign Direct Investment, May 2008 2313 Ronel Elul and Piero Gottardi, Bankruptcy: Is it enough to Forgive or must we also Forget?, May 2008 2314 Andreas Irmen and Johanna Kuehnel, Productive Government Expenditure and Economic Growth, May 2008 2315 Beate Henschel, Carsten Pohl and Marcel Thum, Demographic Change and Regional Labour Markets: The Case of Eastern Germany, May 2008 2316 Gabriel Felbermayr, Wido Geis and Wilhelm Kohler, Restrictive Immigration Policy in Germany: Pains and Gains Foregone?, May 2008 2317 Michael Hofmann, Gerhard Kempkes and Helmut Seitz, Demographic Change and Public Sector Budgets in a Federal System, May 2008 2318 Paul De Grauwe, Macroeconomic Modeling when Agents are Imperfectly Informed, June 2008 2319 Johann K. Brunner and Susanne Pech, Optimum Taxation of Inheritances, June 2008 2320 Thomas Eichner and Marco Runkel, Corporate Income Taxation of Multinationals in a General Equilibrium Model, June 2008 2321 Rainald Borck and Matthias Wrede, Subsidies for Intracity and Intercity Commuting, June 2008 2322 Patricia Apps and Ray Rees, Testing the Pareto Efficiency of Household Resource Allocations, June 2008 2323 Amihai Glazer, Vesa Kanniainen and Panu Poutvaara, Firms’ Ethics, Consumer Boycotts, and Signalling, June 2008 2324 Claudia M. Buch, Jörg Döpke and Kerstin Stahn, Great Moderation at the Firm Level? Unconditional vs. Conditional Output Volatility, June 2008 2325 Helmuth Cremer, Philippe De Donder, Dario Maldonado and Pierre Pestieau, Forced Saving, Redistribution and Nonlinear Social Security Schemes, June 2008 2326 M. Hashem Pesaran and Paolo Zaffaroni, Optimal Asset Allocation with Factor Models for Large Portfolios, June 2008 2327 Harald Badinger and Peter Egger, Horizontal versus Vertical Interdependence in Multinational Activity, June 2008 2328 Jan K. Brueckner and Harris Selod, A Theory of Urban Squatting and Land-Tenure Formalization in Developing Countries, June 2008 2329 Paolo M. Panteghini, Corporate Debt, Hybrid Securities and the Effective Tax Rate, June 2008 2330 Guglielmo Maria Caporale, Juncal Cuñado and Luis A. Gil-Alana, Modelling Long-Run Trends and Cycles in Financial Time Series Data, June 2008 2331 Avi Ben-Bassat and Momi Dahan, Social Identity and Voter Turnout, June 2008 2332 Martin R. West and Ludger Wößmann, “Every Catholic Child in a Catholic School”: Historical Resistance to State Schooling, Contemporary Private Competition, and Student Achievement across Countries, June 2008 2333 Erkki Koskela and Panu Poutvaara, Outsourcing and Labor Taxation in Dual Labor Markets, June 2008 2334 Philippe Choné and Laurent Linnemer, Optimal Litigation Strategies with Signaling and Screening, June 2008 2335 Albert Solé-Ollé and Pilar Sorribas-Navarro, Does Partisan Alignment Affect the Electoral Reward of Intergovernmental Transfers?, June 2008 2336 Antonio Cabrales and Piero Gottardi, Markets for Information: Of Inefficient Firewalls and Efficient Monopolies, June 2008 2337 Sumon Majumdar and Sharun W. Mukand, The Leader as Catalyst – on Leadership and the Mechanics of Institutional Change, June 2008 2338 Ulrich Hange, Tax Competition, Elastic Labor Supply, and Growth, June 2008 2339 Guy Laroque and Bernard Salanié, Does Fertility Respond to Financial Incentives?, June 2008 2340 Adriano Paggiaro, Enrico Rettore and Ugo Trivellato, The Effect of Extending the Duration of Eligibility in an Italian Labour Market Programme for Dismissed Workers, June 2008 2341 Helmut Seitz, Minimum Standards, Fixed Costs and Taxing Autonomy of Subnational Governments, June 2008 2342 Robert S. Chirinko, Leo de Haan and Elmer Sterken, Asset Price Shocks, Real Expenditures, and Financial Structure: A Multi-Country Analysis, July 2008 2343 Wolfgang Leininger, Evolutionarily Stable Preferences in Contests, July 2008 2344 Hartmut Egger and Udo Kreickemeier, Fairness, Trade, and Inequality, July 2008 2345 Ngo Van Long and Bodhisattva Sengupta, Yardstick Competition, Corruption, and Electoral Incentives, July 2008 2346 Florian Baumann, Employment Protection: The Case of Limited Enforceability, July 2008 2347 Alessandro Balestrino, Cinzia Ciardi and Claudio Mammini, On the Causes and Consequences of Divorce, July 2008 2348 Dirk Schindler and Benjamin Weigert, Insuring Educational Risk: Opportunities versus Income, July 2008 2349 Lammertjan Dam and Ben J. Heijdra, The Environmental and Macroeconomic Effects of Socially Responsible Investment, July 2008 2350 Avner Greif, Contract Enforcement and Institutions among the Maghribi Traders: Refuting Edwards and Ogilvie, July 2008 2351 Helmuth Cremer, Philippe De Donder, Dario Maldonado and Pierre Pestieau, Habit Formation and Labor Supply, July 2008 2352 Francesco Menoncin and Paolo M. Panteghini, The Johansson-Samuelson Theorem in General Equilibrium: A Rebuttal, July 2008 2353 Michael Kaganovich and Itzhak Zilcha, Alternative Social Security Systems and Growth, July 2008 2354 Keith Blackburn, Kyriakos C. Neanidis and M. Emranul Haque, Corruption, Seigniorage and Growth: Theory and Evidence, July 2008 2355 Edward Castronova, A Test of the Law of Demand in a Virtual World: Exploring the Petri Dish Approach to Social Science, July 2008 2356 Harald Badinger and Peter Egger, GM Estimation of Higher-Order Spatial Autoregressive Processes in Cross-Section Models with Heteroskedastic Disturbances, July 2008 2357 Wolfgang Buchholz and Jan Schumacher, Discounting the Long-Distant Future: A Simple Explanation for the Weitzman-Gollier-Puzzle, July 2008 2358 Luca Anderlini, Leonardo Felli and Alessandro Riboni, Statute Law or Case Law?, July 2008 2359 Guglielmo Maria Caporale, Davide Ciferri and Alessandro Girardi, Are the Baltic Countries Ready to Adopt the Euro? A Generalised Purchasing Power Parity Approach, July 2008 2360 Erkki Koskela and Ronnie Schöb, Outsourcing of Unionized Firms and the Impacts of Labour Market Policy Reforms, July 2008 2361 Francisco Alvarez-Cuadrado and Ngo Van Long, A Permanent Income Version of the Relative Income Hypothesis, July 2008 2362 Gabrielle Demange, Robert Fenge and Silke Uebelmesser, Financing Higher Education and Labor Mobility, July 2008 2363 Alessandra Casarico and Alessandro Sommacal, Labor Income Taxation, Human Capital and Growth: The Role of Child Care, August 2008 2364 Antonis Adam, Manthos D. Delis and Pantelis Kammas, Fiscal Decentralization and Public Sector Efficiency: Evidence from OECD Countries, August 2008 EUROPEAN CENTRAL BANK WO R K I N G PA P E R S E R I E S WORKING PAPER NO. 242 PUBLIC SECTOR EFFICIENCY: AN INTERNATIONAL COMPARISON BY ANTÓNIO AFONSO, LUDGER SCHUKNECHT AND VITO TANZI July 2003 EUROPEAN CENTRAL BANK WO R K I N G PA P E R S E R I E S WORKING PAPER NO. 242 PUBLIC SECTOR EFFICIENCY: AN INTERNATIONAL COMPARISON1 BY ANTÓNIO AFONSO2, LUDGER SCHUKNECHT3 AND VITO TANZI4 July 2003 1 We are grateful to Carlos Barros, Juergen von Hagen, José Marin, Pierre Pestieau, Philipp Rother, Miguel St. Aubyn, Rolf Strauch, an anonymous referee, and participants at the ZEI Workshop, University of Bonn, at the 2003 European Public Choice Society conference in Aarhus, at the 2003 French Economics Association conference in Lille, for helpful comments and Gerhard Schwab for valuable research assistance. Any remaining errors are the responsibility of the authors.The opinions expressed herein are those of the author(s) and do not necessarily represent those of the European Central Bank. This paper can be downloaded without charge from http:// www.ecb. int or from the Social Science Research Network electronic library at: http://ssrn.com/abstract_id=434002. 2 European Central Bank, Kaiserstrasse 29, D-60311 Frankfurt am Main, Germany, email: [email protected] ISEG/UTL - Technical University of Lisbon, R. Miguel Lúpi 20, 1249-078 Lisbon, Portugal, email: [email protected] 3 European Central Bank, Kaiserstrasse 29, D-60311 Frankfurt am Main, Germany, email: [email protected] 4 Sottosegretario, Ministero del tesero,Via XX Settembre 97, 00187 Roma, Italy © European Central Bank, 2003 Address Postal address Telephone Internet Fax Telex Kaiserstrasse 29 D-60311 Frankfurt am Main Germany Postfach 16 03 19 D-60066 Frankfurt am Main Germany +49 69 1344 0 http://www.ecb.int +49 69 1344 6000 411 144 ecb d All rights reserved by the author/s. Reproduction for educational and non-commercial purposes is permitted provided that the source is acknowledged. The views expressed in this paper do not necessarily reflect those of the European Central Bank. ISSN 1561-0810 (print) ISSN 1725-2806 (online) Contents Abstract 4 Non-technical summary 5 1. Introduction 7 2. Public sector performance indicators 8 3. Public sector expenditure efficiency analysis 14 4. Measuring input and output efficiency via an FDH analysis 4.1 The FDH analysis 4.2 FDH-based expenditure efficiency analysis 18 18 20 5. Conclusion 23 Appendices 25 References 28 European Central Bank working paper series 35 ECB Working Paper No 242 July 2003 3 Abstract We compute public sector performance (PSP) and efficiency (PSE) indicators, comprising a composite and seven sub-indicators, for 23 industrialised countries. The first four sub-indicators are “opportunity” indicators that take into account administrative, education and health outcomes and the quality of public infrastructure and that support the rule of law and a level playing-field in a market economy. Three other indicators reflect the standard “Musgravian” tasks for government: allocation, distribution, and stabilisation. The input and output efficiency of public sectors across countries is then measured via a non-parametric production frontier technique. Keywords: Government expenditure, Efficiency, Free Disposable Hull, Production possibility frontier. JEL Classification Numbers: C14, H50. 4 ECB Working Paper No 242 July 2003 Non-technical summary In this paper we study the performance and the efficiency of the public sectors of 23 industrialised OECD countries. We compute public sector performance (PSP) and efficiency indicators (PSE) for the government as whole and for its core functions. When deriving performance indicators we distinguish the role of government in providing “opportunities” and a level playing field in the market process and the traditional “Musgravian” tasks of government. “Opportunity” indicators look at administrative, education, health, and public infrastructure outcomes. “Musgravian” indicators assess governments’ performance in allocation, distribution, and stabilisation. A number of socio-economic indicators serve as proxies for performance. In assessing the efficiency of public sectors, we look at total public spending and a number of spending categories as proxies for resource use. These are set in relation to performance indicators as they can be seen as reflecting the opportunity costs of public sector activities. The ratio of performance indicators and public spending yields indicators of efficiency for each country. Finally, we use a non-parametric framework to compute a so-called production possibility frontier, and calculate input efficiency and output efficiency scores in order to rank the sample countries in terms of public spending efficiency. We find that the difference in overall performance is moderate across the sample countries. Countries with “small” public sectors on average report the highest scores for overall performance, and especially for administrative and economic performance. Countries with large public sectors show more equal income distribution. Some countries managed to deliver a significant relative improvement in ECB Working Paper No 242 July 2003 5 public sector performance over the last decade (notably, Greece, Portugal, Spain and Ireland). Regarding public sector efficiency, countries with small public sectors report significantly higher indicators than countries with medium-sized or big public sectors. Overall efficiency is highest in Japan, Luxembourg, Australia, the United States and Switzerland. The results of the FDH analysis suggest that “average inefficiency” is about 20%. However, all the results have to be seen as indicative and need to be interpreted with great care. Besides the occasional difficulty of data comparability, it is also not easy to accurately identify the effects of public sector spending on outcomes and separate the impact of spending from other influences. For instance, it is difficult to assess to what extent does higher life expectancy reflect public intervention rather than other factors such as climate, dietary habits, etc. Robustness analysis that emulates the effect of different preferences as to the role of government by giving different weights to sub-indicators suggested that the overall results are not sensitive to moderate changes in the weights of sub-indicators. Finally, the discussion focuses on the overall indicators, while the comparison of the different sub-indicators across countries may provide further and more specific insights and lessons. 6 ECB Working Paper No 242 July 2003 1. Introduction The debate on the role of the state has shifted in recent years towards empirical assessments of the efficiency and usefulness of public sector activities. A growing academic literature has been investigating the stabilisation, allocation and distribution effects of public expenditure. It has also been assessing the role of rules and institutions, and the scope for privatising public sector activities (see e.g., Mueller (1997), Persson and Tabellini (2001), Shleifer and Vishny (1998), Strauch and Von Hagen (2000), Tanzi and Schuknecht (1997, 2000), Rodrik (2000), Gwartney et al. (2002)). Most studies conclude that public spending could be much smaller and more efficient than today. However, for this to happen, governments should adopt better institutions and should transfer many non-core activities to the private sector. The measurement of public sector performance (defined as the outcome of public sector activities) and efficiency (defined as the outcome relative to the resources employed), however, is still very limited. The objective of this paper is to provide a proxy for measuring public sector performance and efficiency. To do this we will put together a number of performance indicators in the government’s core functions. These include the summary functions defined by Musgrave (allocation, distribution, stabilisation) and a number of specific indicators that promote equality of opportunity in the market place. Economic philosophers from Adam Smith to Hayek and Buchanan have stressed the importance of rules of law in promoting “good” government and the “wealth of nations”. Naturally they assume that the rules are “good” rules. We will set these indicators in relation to the costs of achieving them. We will, hence, derive simple performance and efficiency indicators for 1990 and 2000 for the public sectors of 23 industrialised OECD countries. The performance index is then also used in a Free Disposable Hull (FDH) analysis, a rarely used non-parametric production frontier technique to estimate the extent of slack in government expenditures. Note, however, that it is not only public expenditure but also tax and regulatory policies that affect the efficiency of the public sector. While expenditure is also a relatively good proxy of the tax burden, we ignore the composition of tax revenue and ECB Working Paper No 242 July 2003 7 other characteristics of tax systems.5 Public spending may be closely related to regulation because large civil services, that often accompany large public spending, are likely to generate much regulation and vice versa.6 The paper is organised as follows. In Sections two and three we discuss and compute the public sector performance (PSP) and efficiency (PSE) indicators. Section four extends the efficiency analysis with the help of an FDH analysis and section five provides conclusions. 2. Public sector performance indicators The study looks at 23 OECD countries for which we compiled data on various public expenditure categories and socio-economic variables, reflecting the effects/outputs/outcomes of government policies.7 Assume that public sector performance (PSP) depends on the values of certain economic and social indicators (I). If there are i countries and j areas of government performance which together determine overall performance in country i, PSPi, we can then write n PSPi = ∑ PSPij , (1) j =1 with PSPij = f ( I k ) . Therefore, an improvement in public sector performance depends on an improvement in the values of the relevant socioeconomic indicators: 5 For exemple, tax collection may impose significant welfare and compliance costs on taxpayers. However, Brennan (2000) and Tanzi (1998) have argued that regulations and tax expenditures can also become a substitute of public spending, and thereby be negatively correlated with the size of the public sector as measured by the level of public spending. 7 One should be aware of the distinction between output and outcome. The number of hospital days per 1000 people is an output but full recovery from illness or life expectancy is an outcome. Even though we try to approximate outcomes rather than output (e.g. red tape, life expectance) the distinction is not always possible and we use both terms in an interchangeable way. 6 8 ECB Working Paper No 242 July 2003 n ∆PSPij = ∑ i =k ∂f ∆I k . ∂I k (2) Reasonably, the greater the positive effect of public expenditure on any of the selected sub-indicators, the greater will be the envisaged improvement in the PSP indicator. Accordingly, the changes that might occur in the economic and social indicators may be seen as changes in public sector performance. As a first step, we define 7 sub-indicators of public performance. The first four look at administrative, education, health, and public infrastructure outcomes. A good public administration, with a well-functioning judiciary and a healthy and well-educated population, could be considered a prerequisite for a level playing field with wellfunctioning markets and secure property rights, where the rule of law applies, and opportunities are plenty and in principle accessible to all. High-quality public infrastructure is conducive to attaining the same objectives. These indicators, thereby, try to reflect the quality of the interaction between fiscal policies and the market process and the influence on individual opportunities this has. They could be called “process” or “opportunity” indicators. We adopt the latter terminology in the following. The three other sub-indicators reflect the “Musgravian” tasks for government. These try to measure the outcomes of the interaction with and reactions to the market process by government. Income distribution is measured by the first of these indicators. An economic stability indicator illustrates the achievement of the stabilisation objective. The third indicator tries to assess allocative efficiency by economic performance. The conceptual separation is of course somewhat artificial, as for example health and education indicators could also be seen as indicators of allocative efficiency. Finally all sub-indicators are put together in a public sector performance indicator. ECB Working Paper No 242 July 2003 9 Figure 1. Total public sector performance (PSP) indicator Opportunity indicators Standard “Musgravian” indicators Corruption Distribution Income share of 40% poorest households Stability of GDP growth (coeff. of variation) Red tape Administrative Stability Quality of judiciary Inflation (10 years average) Shadow economy GDP per capita (PPP) Secondary school enrolment Economic performance GDP growth (10 years average) Education Education achievement Unemployment (10 years average) Infant mortality Health Life expectancy Quality communication & transport infrast. Public Infrastruc -ture Total public sector performance Before showing the result it is worthwhile illustrating how we derive these performance indicators. Figure 1 shows the socio-economic indices on which government has a significant if not exclusive influence and which, therefore, reflect as close as possible the outcomes of public policies (Annex Tables A and B provide primary data). In as much as possible we provide data for 1990 and 2000 (or the nearest available year), and in some instances, 10-year averages. This is because we are not so much interested in annual fluctuations but in structural changes in public sector performance. Many indices reflect “stocks” which change only very slowly 10 ECB Working Paper No 242 July 2003 over time so that observations every 10 years suffice to reflect such structural changes. A case in point is for example per capita GDP and secondary school enrolment. Other indices, such as inflation or GDP growth, vary strongly and a 10year average seems the best way to capture long-term trends and structural changes.8 Figure 1 also displays the composition of PSP indicators. As to the “opportunity indicators”, administrative performance of government is measured as a composite of the following indices: corruption, red tape, quality of the judiciary, and the size of the shadow economy. The education indicator contains secondary school enrolment and the OECD educational attainment indicators in order to measure both the quantity and quality of education. The health performance indicator contains infant mortality and life expectancy. The public infrastructure indicator contains a measure of the communication and transport infrastructure quality. All these indicators change slowly so that observations every 10 years provide a good impression of changes over time except in the case of public infrastructure where period averages have been used. As to the standard “Musgravian” general indicators, income distribution is proxied by the income share of the poorest 40 per cent of the households. Economic stability is measured by the stability of output growth (coefficient of variation) and average inflation (10-year average). Economic performance comprises per-capita GDP (PPP), GDP growth (10-year average), and unemployment (10-year average). The total PSP indicator combines the seven sub-indicators. Note that some indices also capture the effect of regulation rather than expenditure policies and some indices are only partly the result of government policies (for example, private provision and financing of health and education play an important role in some countries). We compile the performance indicators from the various indices giving equal weight to each of them. For example, red tape, efficiency of the judiciary, corruption and size of the shadow economy each contribute 25 per cent to the administrative performance indicator. This of course introduces a strong assumption. For those indicators where higher numbers are less favourable (e.g., infant mortality, inflation), 8 There are few instances where actual and trend growth deviate by 0.4/0.5% for the 10-year averages. However, when using trend rather than actual growth in the calculation of indices, results change very little even for the economic performance indicator. ECB Working Paper No 242 July 2003 11 we use the inverse of the original values. In order to facilitate the compilation, we normalised the values and set the average for all indices equal to 1. The values for each country are then recalculated relative to the average. Table 1 presents the results for the constructed PSP indicators for the year 2000. Table 1. Public sector performance (PSP) indicators (2000) Opportunity indicators Standard “Musgravian” Total public indicators sector Adminis- Education Health Infra- Distribu- Stability Economic performance (equal Country tration structure tion perform. weights 1/) Australia 1.17 1.02 0.94 1.00 0.87 1.31 1.00 1.04 Austria 1.21 1.00 0.98 1.10 1.22 1.28 1.01 1.12 Belgium 0.73 1.00 0.94 0.91 1.17 1.10 0.83 0.95 Canada 1.11 1.05 0.95 1.16 0.92 1.00 0.92 1.02 Denmark 1.16 1.00 1.03 1.03 1.19 1.10 0.91 1.06 Finland 1.26 1.07 1.04 1.18 0.75 0.73 1.01 France 0.72 1.03 1.03 1.01 0.90 1.12 0.70 0.93 Germany 1.02 0.98 1.01 1.01 0.98 0.91 0.81 0.96 Greece 0.60 0.94 0.93 0.81 0.97 0.55 0.69 0.78 Iceland 1.02 0.98 1.25 0.59 1.29 1.03 Ireland 1.06 0.94 0.88 1.00 0.89 1.22 1.40 1.05 Italy 0.52 0.96 0.93 0.84 1.10 0.76 0.69 0.83 Japan 0.87 1.09 1.12 1.09 1.20 1.40 1.18 1.14 Luxembourg 1.05 0.81 0.95 1.22 2.04 1.21 Netherlands 1.16 1.04 0.97 1.09 1.00 1.42 1.06 1.11 New Zealand 1.18 1.03 0.89 0.62 0.99 0.84 0.93 Norway 0.97 1.04 1.09 0.94 1.17 1.45 1.26 1.13 Portugal 0.54 0.94 0.90 0.75 0.92 0.64 0.92 0.80 Spain 0.77 1.00 1.10 0.86 1.02 0.82 0.67 0.89 Sweden 1.16 1.07 1.19 1.10 1.17 0.69 0.91 1.04 Switzerland 1.32 0.97 1.14 1.23 0.95 0.79 1.09 1.07 United Kingdom 1.00 1.05 0.91 0.99 0.79 0.78 0.84 0.91 United States 1.15 1.00 0.82 1.08 0.76 1.14 1.20 1.02 Average 0.99 1.00 1.00 1.00 1.00 1.00 1.00 1.00 Small govs 2/ 1.11 1.01 0.98 1.08 0.94 1.17 1.17 1.07 Medium govs 0.93 0.98 1.00 0.93 0.92 0.89 1.03 0.97 Big govs 0.99 1.02 1.01 1.01 1.12 1.03 0.85 1.01 EU 15 3/ 0.88 1.00 0.99 0.98 0.98 0.93 0.80 0.94 Euro area 3/ 0.84 0.99 1.00 0.97 1.00 0.96 0.78 0.93 1/ Each sub-indicator contributes 1/7 to total indicator. 2/ Small governments: public spending <40% of GDP in 2000. Big governments: public spending >50% of GDP in 2000. Medium governments: 40%< public spending <50% of GDP in 2000. 3/ Weighted averages according to the share of each country GDP in the relevant group. Indicators suggest notable but not extremely large differences in public sector performance across countries (with a few exceptions). Countries with the highest values for sub-indicators include Switzerland (administration and infrastructure), Japan (education), Iceland (health), Austria (distribution), Norway (economic stability) and Luxembourg (economic performance). Countries such as Luxembourg, 12 ECB Working Paper No 242 July 2003 Japan, Norway, Austria, and the Netherlands report high total PSP indicators. The latter is true both for a PSP indicator with equal weights for the sub-indicators and for different weighting, suggesting that the findings are relatively robust to moderate changes in weighting.9 Looking at country groups, small governments (industrialised countries with public spending below 40 % of GDP in 2000) on balance report better economic performance than big governments (public spending above 50 % of GDP) or medium sized governments (spending between 40 and 50 percent of GDP). Big governments feature more even income distribution whereas small governments perform better especially in the administrative, stability and economic performance domains. These results are consistent with those found in Tanzi and Schuknecht (2000). When comparing the main economic “players” of today, it is noteworthy that the US and particularly Japan report above-average performance in most sub-indices and for the total PSP measure. By contrast, the EU (weighted average) performs below average. Taking advantage of the data set available, we performed a comparison between the PSP for 2000 and for 1990, in order to assess how public sector performance has changed over time10 and the results are presented in Figure 2. 9 For example, giving alternative weights to the sub-indicators does not change much the results in most cases. In the Appendix (Table A1) we present alternative weighting schemes. Rank correlations for PSP indicators with the tested changes in weights are in the [0.95 0.99] range. This weigthing of the variables is quite straightforward and economically intuitive (even though it is still somewhat ad hoc). It avoids the problem of lack of economic justification of a more complex statistical approach such as principal component analysis that might come to mind in this context. 10 One should bear in mind that data are not fully comparable. E.g., some data are not available for some countries. For example the OECD PISA report on education achievement only covers 2000. ECB Working Paper No 242 July 2003 13 Figure 2. Public sector performance: 1990 and 2000 1.50 Improvement (+) 1.25 Luxembourg Ireland Japan Switzerland 1.00 2000 Spain US 0.75 Portugal Greece Euro area Italy UK 0.50 0.25 0.25 Worsening (-) 0.50 0.75 1.00 1.25 1.50 1990 One can easily see that while some countries managed to deliver a relative improvement in public sector performance (all the countries located above and to the left of the diagonal line), some other countries showed a decrease in public sector performance (countries below and to right of the diagonal). Examples of the first group of countries are Greece, Portugal, Spain and Ireland. However, only Ireland succeeded in placing itself above the average of the 23 OECD country sample. Some countries experienced reductions in public sector performance. Especially Japan and Switzerland saw their performance fall in 2000 compared to 1990. This is also true for the EU and the euro area as a whole. However, one should be aware that progress in public sector performance made by the different countries over time is measured relative to other countries and not relative to its own past performance. 3. Public sector expenditure efficiency analysis Public expenditure, expressed as a share of GDP, can be assumed to reflect the opportunity costs of achieving the public sector performance estimated in the previous 14 ECB Working Paper No 242 July 2003 section.11 In addition to total public spending we looked at average spending on goods and services, transfers, functional spending on education and health, and public investment. Data for 1990 and 2000 for these categories across countries are reported in Annex Table C. Public expenditures differ considerably across countries. Average total spending in the 1990s ranged from around 35 percent of GDP in the US to 64 percent of GDP in Sweden. The difference is mainly due to more or less extensive welfare programs. Public spending on health and education and on goods and services differs much less strongly across countries. Based on the framework of equations (1) and (2), we now compute indicators of Public Sector Efficiency (PSE). We weigh performance (as measured by the PSP indicators) by the amount of relevant public expenditure, PEX, that is used to achieve a given performance level. The overall PSE indicator for any country i, is given by: PSEi = PSPi , PEX i (3) with n PSP PSPi ij =∑ . PEX i j =1 PEX ij (4) Positive but declining marginal productivity of public expenditure would imply: ∂PSEij ∂PEX ij > 0, ∂ 2 PSEij ∂PEX ij2 < 0. (5) In order to compute efficiency indicators, public spending was normalised across countries, with the average taking the value of one for each of the six categories specified above. We focus on average expenditure over the 1990s, as we would assume a lagged effect from spending on performance. For example, public spending 11 Proceeds from the sale of UMTS mobile telephone licences have been excluded from total expenditure since they were recorded as a temporary decline in expenditure. ECB Working Paper No 242 July 2003 15 on education (at least) over the previous decade, is assumed to affect educational achievement in the late 2000. Before putting public sector performance and expenditure together it is worth stressing that not all expenditure categories are equally suitable indices for measuring the efficiency with which a certain performance is achieved. Goods and services spending are a rather crude approximation for what is needed to achieve administrative efficiency. Health and education spending seem better measures of the public sector inputs in these domains.12 Similarly, transfers (social payments only) are probably suitable approximations for government spending to promote income equality, and public investment is likely to be closely connected with infrastructure quality.13 Total spending may be a useful proxy for government stabilisation efforts because automatic stabilisers are larger in countries with “big governments” (Van den Noord (2000), Bouthevillain et al (2001)). Total spending is generally financed by distortive taxation. It can, hence, be used as a proxy for the efficiency (or inefficiency) of the state in affecting economic performance. Before turning to Table 2, which reports the ratio of performance and expenditure indices as so-called Public Sector Efficiency (PSE) indicators it is worthwhile stressing a few caveats. Public spending across countries is not always fully comparable even though much progress has been achieved in this regard. For example, some countries’ transfer payments are taxed, thereby overstating public spending compared to countries where such benefits are not taxed. Nevertheless, it is not possible to systematically assess and correct such problems. Moreover, comparing expenditure ratios across countries implicitly assumes that production costs for public services are proportionate to GDP per capita. While this approximation is likely to be quite good for labour intensive services (such as education or administrative efficiency) it is likely to be less so for infrastructure quality. In the absence of cross12 Notice however, that it is not easy to accurately identify the effects of public sector spending on outcomes and separate the impact of spending from other influences. For instance, it is difficult to assess to what extent does higher life expectancy reflect public intervention rather than other factors such as climate, dietary habits, etc. The same argument could be made regarding infant mortality. On that line of reasoning, adverse geographical conditions may also impair on the quality and cost of a country communications infrastrucutre. 13 Income distribution and stabilisation is also affected by the progressivity of the tax system, but this effect is very difficult to assess due to the lack of comparable and detailed enough data. 16 ECB Working Paper No 242 July 2003 country data of different public service sector costs, this is nevertheless the best possible approximation. Table 2. Public sector efficiency (PSE) indicators (2000) 1/ Opportunity indicators Standard “Musgravian” Total public indicators sector Adminis- Education Health Infra- Distribu- Stability Economic efficiency (equal Country tration structure tion perform. weights 2/) Australia 1.21 1.06 1.05 1.05 1.80 1.59 1.22 1.28 Austria 1.22 0.93 1.07 0.98 0.93 1.17 0.92 1.03 Belgium 0.64 0.96 0.85 1.11 0.71 0.87 0.65 0.83 Canada 1.00 0.84 0.86 1.27 1.39 1.01 0.93 1.04 Denmark 0.86 0.74 0.76 1.62 1.05 0.89 0.74 0.95 Finland 1.22 1.07 1.03 1.19 0.79 0.77 1.01 France 0.61 0.99 0.90 1.00 0.64 1.01 0.63 0.83 Germany 1.01 1.09 0.93 1.27 0.85 0.88 0.78 0.97 Greece 0.79 2.25 1.05 0.87 1.04 0.61 0.78 1.06 Iceland 1.06 1.12 0.65 1.42 0.85 Ireland 1.10 0.90 0.88 0.96 0.90 1.20 1.38 1.05 Italy 0.54 1.11 0.93 0.75 0.95 0.68 0.62 0.80 Japan 1.25 1.12 1.34 0.68 1.60 1.99 1.68 1.38 Luxembourg 1.10 0.88 0.98 1.19 1.99 1.23 Netherlands 0.90 0.85 0.95 1.52 0.56 1.15 0.85 0.97 New Zealand 1.20 1.02 0.85 0.00 0.68 0.97 0.82 0.93 Norway 0.95 0.86 0.96 0.88 1.32 1.40 1.22 1.09 Portugal 0.74 1.31 1.46 0.66 1.28 0.73 1.05 1.03 Spain 0.97 1.49 1.33 0.81 1.12 0.95 0.78 1.06 Sweden 0.81 0.75 0.83 1.19 0.94 0.51 0.68 0.82 Switzerland 1.86 1.01 1.21 1.07 1.68 1.05 1.45 1.33 United Kingdom 0.94 1.10 1.01 1.68 0.98 0.84 0.91 1.06 United States 1.30 0.92 1.05 1.40 1.15 1.46 1.55 1.26 Average 1.01 1.06 1.01 1.09 1.08 1.03 1.04 1.04 Small govs 3/ 1.34 1.00 1.11 1.03 1.43 1.46 1.45 1.26 Medium govs 0.98 1.19 1.05 1.06 1.08 0.92 1.07 1.03 Big govs 0.85 0.93 0.92 1.17 0.87 0.88 0.73 0.90 EU 15 4/ 0.84 1.09 0.97 1.18 0.87 0.88 0.77 0.94 Euro area 4/ 0.82 1.11 0.97 1.06 0.84 0.90 0.74 0.92 1/ These indicators are the expenditure weighted “counterparts” of the indicators of Table 1. 2/ Each sub-indicator contributes 1/7 to total indicator. 3/ Small governments: public spending <40% of GDP in 2000. Big governments: public spending >50% of GDP in 2000. Medium governments: 40%< public spending <50% of GDP in 2000. 4/ Weighted averages according to the share of each country GDP in the relevant group. We find significant differences in public sector efficiency across countries. Japan, Switzerland, Australia, the United States and Luxembourg show the best values for overall efficiency. Looking at country groups, “small” governments post the highest ECB Working Paper No 242 July 2003 17 efficiency amongst industrialised countries. Differences are considerable as “small” governments on average post a 40 percent higher scores than “big” governments.14 In summary, we find that differences in efficiency are much more pronounced than in performance across countries, with “small” governments clearly outranking the others. This illustrates that the size of government may be too large in many industrialised countries, with declining marginal products being rather prevalent. But given the non-extreme differences in performance as outlined above, the incidence of “negative” marginal products of public spending may be more limited. 4. Measuring input and output efficiency via an FDH analysis 4.1 The FDH analysis In a final step, we use the information from previous sections to measure the “wastefulness” of public spending across countries, i.e. the input and output efficiency of expenditure. To this end, we apply a so-called FDH analysis, which is a nonparametric technique that was first proposed by Deprins, Simar, and Tulkens (1984).15 In the FDH framework it is possible to rank the efficiency of producers by comparing each individual performance with a production possibility frontier. Along this production possibility frontier one can observe the highest possible level of output/outcome for a given level of input. Conversely, it is possible to determine the lowest level of input necessary to attain a given level of output/outcome. This allows identifying inefficient producers both in terms of input efficiency and in terms of output/outcome efficiency. A few other studies that apply FDH analysis to assess public spending efficiency include Vanden Eeckhaut, Tulkens and Jamar (1993) who studied the efficiency of public spending in Belgian municipalities, and Fakin and Crombrugghe (1997) who assessed the efficiency of government expenditures as regards some specific public 14 The PSE indicators are also quite robust to different weightings as can be seen in the Appendix (Table A2). 15 For an overview of the FDH analysis see for instance Tulkens (1993). Another non-parametric approach that might be used to assess public expenditure efficiency would be Data Envelopment Analysis (DEA). This technique, developed by Charnes, Cooper and Rhodes (1978), implies a convex production frontier, an hypothesis which is not required in the FDH approach. For an overview of non-parametric approaches see for instance Simar and Wilson (2003). 18 ECB Working Paper No 242 July 2003 services in OECD and Central Europe countries. Gupta and Verhoeven (2001) use FDH analysis to measure the efficiency of government expenditure on education and health in a set of countries in Africa. Clements (2002) assessed the efficiency of education spending in the European Union. St. Aubyn (2002) reports results of FDH analysis applied to education and health spending in OECD countries. The FDH methodology can be well illustrated graphically (Figure 3). Figure 3. Production possibility frontier Assume four countries, A, B, C and D that use a certain amount of public expenditures, measured on the horizontal axis in monetary units. The countries are then assumed to achieve a certain level of public spending performance, measured on the vertical axis. The efficiency of the four countries is obviously different. For instance, country B uses more input than country A [X(B)>X(A)], but produces less output [Y(B)<Y(A)]. Therefore country B is relatively inefficient in comparison with country A. On the other hand, country A is efficient in relation to country B, and it is placed on the production possibility frontier. This means there are no other countries besides country A that deliver the same level of output with a lower level of input. Similarly, ECB Working Paper No 242 July 2003 19 countries C and D are efficient and are also on the production possibility frontier. No other country is inefficient compared to them.16 This framework allows the calculation of the production possibility frontier, and input efficiency and output efficiency scores in order to rank the several countries in terms of public spending efficiency. These efficiency scores will be set between 0 and 1, and all the countries placed on the production possibility frontier will be assigned the maximum score of 1. Note that this approach is likely to underestimate inefficiencies, as the countries on the production possibility frontier are efficient by definition (even though they too may have scope for savings). The input efficiency score of a given country indicates how much less input this country could use to achieve the same level of output. Additionally, the output efficiency score of a given country would tell how much more output the country should be able to produce with the same amount of resources that it is currently using. 17 4.2. FDH-based expenditure efficiency analysis We now conduct an FDH efficiency analysis of public expenditure to our sample of 23 OECD countries. Public spending as a percentage of GDP in 2000 measures the input and as output we use the public sector performance indicator already determined in section 2. The production possibility frontier for our set of countries is presented in Figure 4.18 One can see that the most efficient countries, positioned on the production possibility frontier, are the US, Japan, and Luxembourg. Australia, Ireland and 16 Gupta and Verhoeven (2001) would call countries such as C and D “independently efficient”, and country A “not independently efficient.” 17 Figure 3 illustrates that country B’s input efficiency score is given by X(A)/X(B), which is 0.5, smaller than one, since B is the interior of the production possibility frontier. This implies that the excess use of inputs by inefficient country B is 50 per cent of the necessary inputs to achieve the same level of performance of country A. Country B’s output efficiency score is Y(B)/Y(A). In this case, the loss of output of country B relative to the most efficient country turns out to be also 50 per cent (since for country B one can calculate Y(B)/Y(A)=5/10=0.5). The production possibility frontier for the example in Figure 3 is as 0, X < 100 10, 100 ≤ X < X (C ) follows: Y = f ( X ) = . ≤ < ( ), ( ) ( ) Y C X C X X D Y ( D), X ≥ X ( D) 18 One must be aware of the scaling when interpreting the chart. A doubling in PSP is not necessarily a doubling of welfare or utility. 20 ECB Working Paper No 242 July 2003 Switzerland come very close to the frontier while the other countries are further removed and therefore less “efficient”. Figure 4. Production possibility frontier, 23 OECD countries, 2000 Total public sector performance (PSP index) 1.50 Luxembourg 1.25 Austria Japan Sweden Switzerland US 1.00 Australia Germany UK Italy 0.75 Greece Production possibility frontier 0.50 25 30 35 40 45 50 55 60 Total public expenditures/GDP (%) The figure shows that the EU countries are mostly well inside the production possibility frontier. They mostly report a much higher ratio of public expenditure-toGDP than the US, but nevertheless often report lower public sector performance indicators. The results both for input efficiency and output efficiency are presented in Table 3, where we report the respective efficiency scores along with each country’s ranking. ECB Working Paper No 242 July 2003 21 Table 3. Efficiency scores: public expenditures as a % of GDP in 2000 and Public Sector Performance indicator (see Table 1) Country Australia Austria Belgium Canada Denmark Finland France Germany Greece Iceland Ireland Italy Japan Luxembourg Netherlands New Zealand Norway Portugal Spain Sweden Switzerland United Kingdom United States Average EU15 average Non-EU15 average Small governments 1/ Medium governments 1/ Big governments 1/ EU 15 2/ Euro area 2/ Input efficiency Score Rank 0.99 4 0.67 17 0.66 19 0.75 12 0.62 21 0.61 22 0.64 20 0.72 16 0.73 14 0.87 7 0.96 5 0.66 18 1.00 1 1.00 1 0.72 15 0.83 9 0.73 13 0.79 11 0.80 10 0.57 23 0.95 6 0.84 8 1.00 1 0.79 0.73 0.89 0.98 0.81 0.65 0.72 0.70 Output efficiency Score Rank 0.92 7 0.92 8 0.79 18 0.84 13 0.87 11 0.83 14 0.77 20 0.79 17 0.65 23 0.90 10 0.93 6 0.68 22 1.00 1 1.00 1 0.91 9 0.81 15 0.93 5 0.70 21 0.78 19 0.86 12 0.94 4 0.80 16 1.00 1 0.85 0.82 0.92 0.96 0.82 0.83 0.78 0.78 The values in bold signal the countries located on the production possibility frontier. 1/ See notes of Tables 1 and 2. 2/ Weighted averages according to the share of each country GDP in the relevant group. The Table shows that input efficiency scores start at 0.57 and output efficiency scores at 0.65. The average input efficiency of the 15 EU countries is 0.73 meaning that they should be able to attain the same level output using only 73 per cent of the inputs they are currently using (or about 35% of GDP rather than close to 50%). The output efficiency score implies that with given public expenditures, public sector performance is 82 percent (or 18 percent less) of what it could be if the EU was on the production possibility frontier (and more if the countries on the production possibility frontier also have scope for expenditure savings). By contrast, the non-EU OECD 22 ECB Working Paper No 242 July 2003 countries report more public expenditure efficiency. An average input efficiency score of 0.89 implies only roughly 11 percent “waste”. It is also now possible to focus on some specific interesting cases, such as Sweden. It reports a PSP indicator of 1.04, above the average of the country sample. High public spending pushes down the PSE indicator to a value of only 0.82, well below the average. The input efficiency score of 0.57 suggests that little more than half the current spending would be sufficient to achieve the same public sector performance. The situation is similar in some of the other countries with “big governments”, namely France, Germany and Italy where public expenditures account for around 50 per cent of GDP. Indeed, with the exception of Luxembourg, all two other countries located on or near the production possibility frontier belong to the group of “small government” countries, with a public expenditures-to-GDP ratio below the 40 per cent threshold. 5. Conclusion We developed indicators of public sector performance for 23 industrialised countries. For that purpose we used a number of socio-economic indicators as proxies for performance, and total spending and a number of spending categories as proxies for resource use. We find moderate differences in the public sector performance (PSP) indicators across industrialised countries. Unsurprisingly, countries with small public sectors report the “best” economic performance while countries with large public sectors show more equal income distribution. When weighing performance by the resources used to achieve it, i.e. public expenditure, there are important differences across countries in the resulting public sector efficiency (PSE) indicators. Countries with small public sectors report significantly higher PSE indicators than countries with medium-sized or big public sectors. All these findings suggest diminishing marginal products of higher public spending. The results that we get from the production-frontier-related FDH analysis, which uses the PSP indicators, are also in line with the aforementioned conclusions. Small ECB Working Paper No 242 July 2003 23 governments tend to show better results. Spending in big governments could be, on average, about 35 per cent lower to attain the same public sector performance. The calculations also point out that EU 15 countries show relatively low public sector efficiency when compared with the US and also the average of the other OECD countries in the sample. EU 15 countries are using 27 per cent more public spending than the “most efficient” countries with similar PSP indicators. Spending for the average of the other OECD countries is “only” 11 percent higher than necessary. However, all the results have to be seen as indicative and need to be interpreted with great care for the reasons outlined above. In our interpretation, we mainly focussed on the overall PSP and PSE indicators to which we also applied the FDH analysis. This is appropriate to gain an overall impression. The comparison of the different opportunity and standard “Musgravian” sub-indicators across countries and the detailed assessment of differences may provide further and more specific insights and lessons. Finally, it seems important to bear in mind that by using a non-parametric approach, and in spite of FDH being an established and valid methodology, differences across countries are not statistically assessed, which can be considered as a limitation of such methodology. Additionally, scale economies may also play a role in public sector policies being able to deliver better outcomes. 24 ECB Working Paper No 242 July 2003 Appendix In order to assess the sensitivity of the results for public sector performance and efficiency, we used alternative weighting schemes. We computed PSP and PSE indicators that can give more weight to, inter alia, opportunity, equality, stability and economic performance sub-indicators. One could argue that these indicators emulate people with different intensities of preferences. The results, presented in Table A1 and in Table A2, confirm that the conclusions presented in the main text are generally not changed. Rank correlations with the tested changes in weights are in the [0.95 0.99] range for PSP indicators and in the [0.96 0.99] range for PSE indicators. ECB Working Paper No 242 July 2003 25 Table A1 – Total public sector performance (PSP), 2000, different weights Country Australia Austria Belgium Canada Denmark Finland France Germany Greece Iceland Ireland Italy Japan Luxembourg Netherlands New Zealand Norway Portugal Spain Sweden Switzerland United Kingdom United States Average Small govs Medium govs Big govs EU 15 * Euro area * Baseline 1) 1.04 1.12 0.95 1.02 1.06 1.01 0.93 0.96 0.78 1.03 1.05 0.83 1.14 1.21 1.11 0.93 1.13 0.80 0.89 1.04 1.07 0.91 1.02 1.00 1.07 0.97 1.01 0.94 0.93 Weighting of sub-indicators with emphasis on: Opportunity 2) Equality 3) Stability 4) Economic performance 5) 1.04 1.01 1.10 1.03 1.11 1.14 1.15 1.09 0.94 1.00 0.99 0.93 1.03 1.00 1.01 1.00 1.06 1.09 1.07 1.03 1.05 1.04 0.96 0.95 0.93 0.92 0.97 0.88 0.97 0.96 0.95 0.92 0.79 0.82 0.73 0.76 1.04 1.03 0.95 1.07 1.04 1.02 1.09 1.13 0.83 0.89 0.81 0.80 1.12 1.15 1.20 1.15 1.17 1.21 1.22 1.35 1.10 1.08 1.18 1.09 0.96 0.86 0.94 0.91 1.11 1.14 1.20 1.16 0.80 0.83 0.76 0.83 0.90 0.92 0.87 0.84 1.06 1.07 0.96 1.01 1.09 1.04 1.01 1.07 0.93 0.88 0.88 0.89 1.02 0.96 1.05 1.06 1.00 1.00 1.00 1.00 1.06 1.04 1.09 1.09 0.97 0.97 0.95 0.97 1.01 1.03 1.01 0.97 0.94 0.95 0.93 0.91 0.94 0.95 0.94 0.90 1) Equal weights assigned to each sub-indicator (1/7), as in Table 1. 2) 2/3 assigned to opportunity indicators and 1/3 to "Musgravian indicators". This means 1/6 assigned to each of the 4 opportunity indicators and 1/9 to each of the 3 "Musgravian indicators". 3) 1/3 assigned to the distribution indicator and 2/3 to the other indicators. This means that each of the other 6 indicators will have a weight of 1/9. 4) 1/3 assigned to the stability indicator and 2/3 to the other indicators. This means that each of the other 6 indicators will have a weight of 1/9. 5) 1/3 assigned to the economic performance indicator and 2/3 to the other indicators. This means that each of the other 6 indicators will have a weight of 1/9. * Weighted averages according to the share of each country GDP in the relevant group. 26 ECB Working Paper No 242 July 2003 Table A2 – Total public sector efficiency (PSE), 2000, different weights Country Australia Austria Belgium Canada Denmark Finland France Germany Greece Iceland Ireland Italy Japan Luxembourg Netherlands New Zealand Norway Portugal Spain Sweden Switzerland United Kingdom United States Average Small govs Medium govs Big govs EU 15 * Euro area * Baseline 1) 1.28 1.03 0.83 1.04 0.95 1.01 0.83 0.97 1.06 0.85 1.05 0.80 1.38 1.23 0.97 0.93 1.09 1.03 1.06 0.82 1.33 1.06 1.26 1.04 1.26 1.03 0.90 0.94 0.92 Weighting of sub-indicators with emphasis on: Opportunity 2) Equality 3) Stability 4) Economic performance 5) 1.24 1.40 1.35 1.27 1.04 1.01 1.06 1.01 0.84 0.80 0.84 0.79 1.03 1.12 1.04 1.02 0.96 0.97 0.94 0.90 1.04 1.05 0.97 0.96 0.84 0.79 0.87 0.78 0.99 0.94 0.95 0.93 1.10 1.05 0.96 0.99 0.83 0.85 0.82 0.95 1.03 1.01 1.08 1.12 0.80 0.83 0.77 0.76 1.32 1.43 1.52 1.45 1.19 1.23 1.22 1.35 0.99 0.88 1.01 0.94 0.96 0.88 0.93 0.91 1.05 1.14 1.16 1.12 1.04 1.09 0.97 1.04 1.08 1.08 1.04 1.00 0.84 0.84 0.75 0.79 1.32 1.41 1.27 1.36 1.09 1.05 1.01 1.03 1.24 1.24 1.31 1.33 1.04 1.05 1.04 1.03 1.23 1.30 1.30 1.30 1.04 1.04 1.01 1.03 0.92 0.90 0.90 0.87 0.96 0.93 0.93 0.90 0.93 0.90 0.91 0.88 1) Equal weights assigned to each sub-indicator (1/7), as in Table 2. 2) 2/3 assigned to opportunity indicators and 1/3 to "Musgravian indicators". This means 1/6 assigned to each of the 4 opportunity indicators and 1/9 to each of the 3 "Musgravian indicators". 3) 1/3 assigned to the distribution indicator and 2/3 to the other indicators. This means that each of the other 6 indicators will have a weight of 1/9. 4) 1/3 assigned to the stability indicator and 2/3 to the other indicators. This means that each of the other 6 indicators will have a weight of 1/9. 5) 1/3 assigned to the economic performance indicator and 2/3 to the other indicators. This means that each of the other 6 indicators will have a weight of 1/9. * Weighted averages according to the share of each country GDP in the relevant group. ECB Working Paper No 242 July 2003 27 References Bouthevillain, C.; Cour-Thimann, P.; Van Den Dool, G; Hernandez de Cos, P.; Langenus, G.; Mohr, M.; Momigliano, S. and Tujula, M. (2001). “Cyclically Adjusted Budget Balances: An Alternative Approach,” ECB Working Paper 77. Brennan, G. (2000). “The Political Economy of Regulation,” Australian National University, mimeo. Charnes, A.; Cooper, W. and Rhodes, E. (1978). “Measuring the efficiency of decision making units,” European Journal of Operational Research, 2 (6), 429–444. Clements, B. (2002). “ How Efficient is Education Spending in Europe?” European Review of Economics and Finance, 1 (1), 3-26. Deprins, D., Simar, L., and Tulkens, H. (1984). “Measuring labor-efficiency in post offices,” in: Marchand, M.; Pestieau, P. and Tulkens, H. (Eds.), The performance of public enterprises: concepts and measurement. Amsterdam: North-Holland. Fakin, B. and de Crombrugghe, A. (1997). “Fiscal adjustment in transition economies: social transfers and the efficiency of public spending: a comparison with OECD countries,” Policy Research Working Paper 1803). Washington, DC: World Bank. Gwartney, J.; Lawson, R.; Park, W.; Wagh, S.; Edwards, C. and de Rugy, V. (2002). Economic Freedom of the World: 2002 Annual Report. Vancouver, the Fraser Institute. Gupta, S. and Verhoeven, M. (2001). “The efficiency of government expenditure Experiences from Africa,” Journal of Policy Modelling 23, 433– 467. Mueller, D. (1997). (ed.) Perspectives on Public Choice: A Handbook, Cambridge: Cambridge University Press. Persson, T. and Tabellini, G. (2001). “Political Institutions and Policy Outcomes: What are the Stylized Facts?” Mimeo. Rodrik, D. (2000). “Institutions for High-Quality Growth: What they are and how to Acquire them,” NBER Working Paper 7540. Schneider, F. (2002). “The Size and Development of the Shadow Economies and Shadow Labor Force of 22 Transition and 21 OECD Countries: What do We Really Know?” mimeo, prepared for the Round Table Conference: “On the Informal Economy,” Sofia, Bulgaria, April 2002. Shleifer, A. and Vishny, R. (1998). The Grabbing Hand: Government Pathologies and their Cures, Cambridge: Harvard University Press. Simar, L. and Wilson, P. (2003). Efficiency analysis: the statistical approach, lecture notes, January. 28 ECB Working Paper No 242 July 2003 St. Aubyn, M. (2002). “Evaluating efficiency in the Portuguese health and education sectors,” mimeo presented at the conference on Desenvolvimento Económico Português no Espaço Europeu: Determinantes e Políticas organised by Bank of Portugal, 24-25 May 2002. Strauch, R. and Hagen, J. (2000). Institutions, Politics and Fiscal Policy, Boston: Kluwer Academic Publishers. Tanzi, V. (1998). “Government Role and the Efficiency of Policy Instruments,” in Sorenson. P. (ed.) Public Finance in a Changing World, Macmillan Press, 51-79. Tanzi, V. and Schuknecht, L. (1997). “Reconsidering the Fiscal Role of Government: The International Perspective,” American Economic Review, 87 (2), 164-168. Tanzi. V. and Schuknecht, L. (2000). Public Spending in the 20th Century: A Global Perspective, Cambridge: Cambridge University Press. Tulkens, H. (1993). “On FDH analysis: some methodological issues and applications to retail banking, courts and urban transit,” Journal of Productivity Analysis 4, 183– 210. Vanden Eeckhaut, P., Tulkens, H., and Jamar, M.-A. (1993). “Cost-efficiency in Belgian municipalities,” in Fried, H.; Lovell, C. and Schmidt, S. (eds.), The Measurement of Productive Efficiency: Techniques and Applications. New York: Oxford Univ. Press. Van den Noord, P. (2000). “The Size and Role of Automatic Fiscal Stabilisers in the 1990s and Beyond,” OECD Working Paper 230. Data references International Institute for Management Competitiveness Yearbook 2001. Development (2001). The World OECD (2001a). Economic Outlook Database. OECD (2001b). Education at a Glance 2001. OECD (2001c). Knowledge and Skills for Life – First Results from Pisa 2000, Paris. OECD (2001d). Social Expenditure Database. OECD (2001e). Main Economic Indicators Database. World Bank, several years. World Development Report. World Bank (2001). World Development Indicators 2001. World Economic Forum (1990). The World Competitiveness Report 1990. ECB Working Paper No 242 July 2003 29 30 ECB Working Paper No 242 July 2003 Red tape 1/ Quality of judiciary 1/ Shadow Economy 2/ School enrolment 3/ Education achievement 1990 78.8 77.9 78.0 79.0 75.9 77.3 78.5 77.0 77.9 79.2 76.1 78.3 80.6 76.9 77.7 77.4 78.5 75.4 77.9 79.3 79.6 77.2 76.9 76.2 2000 78.9 78.2 78.2 78.9 76.4 77.5 78.9 77.4 77.9 79.5 76.3 78.7 80.7 77.0 77.9 78.2 78.6 75.6 78.2 79.6 79.7 77.3 77.1 78.1 Infant mortality Life expectancy 1990 2001 1990 2001 1990 2001 1989/90 1999/2000 1990 1998 1995 2000 1990 2000 Australia 6.57 8.21 4.13 4.94 7.84 8.51 10.1 14.3 78.6 88.9 519 530 8.0 5.3 Austria 5.24 6.92 4.86 4.12 7.33 9.04 6.9 9.8 91.1 88.2 514 7.8 4.8 Belgium 5.52 5.22 3.76 2.78 6.18 5.70 19.3 22.2 87.7 88.0 550 508 7.9 5.3 Canada 7.50 7.78 4.59 4.63 8.44 8.49 12.8 16.0 88.7 93.7 521 532 6.8 5.2 Denmark 9.16 9.03 4.74 5.04 8.42 8.59 10.8 18.0 86.8 89.5 497 7.5 4.3 Finland 7.79 9.53 5.46 6.38 8.42 8.70 13.4 18.1 93.0 94.8 540 5.6 4.2 France 6.03 4.22 4.09 1.76 6.20 5.85 9.0 15.2 85.8 94.2 507 7.3 4.4 Germany 7.58 6.91 4.81 3.87 8.17 8.23 11.8 16.0 87.8 487 7.0 4.5 Greece 2.82 3.00 1.90 2.34 5.00 6.15 22.6 28.7 82.7 86.4 460 9.7 5.4 Iceland 9.03 6.28 8.28 85.4 506 5.9 3.1 Ireland 6.98 5.54 5.30 5.64 8.09 7.57 11.0 15.9 79.9 77.0 514 8.2 5.9 Italy 2.56 3.53 2.76 1.97 3.16 3.56 22.8 27.1 88.3 491 473 8.2 5.3 Japan 5.46 4.27 5.32 2.62 7.76 6.25 8.8 11.2 96.8 98.6 581 543 4.6 3.8 Luxembourg 5.52 7.37 3.76 4.11 6.18 7.47 67.6 436 7.3 5.0 Netherlands 8.13 7.97 5.42 4.69 8.13 8.28 11.9 13.1 83.6 92.6 529 7.1 4.9 New Zealand 8.43 8.76 6.27 4.34 7.89 8.26 9.2 12.8 85.0 90.3 501 531 8.3 5.9 Norway 7.35 8.07 4.00 3.03 8.20 8.30 14.8 19.1 87.7 96.4 501 6.9 3.9 Portugal 4.51 3.89 3.32 2.22 8.03 2.70 15.9 22.7 87.6 456 10.9 5.5 Spain 3.78 5.57 3.18 3.97 2.89 4.43 16.1 22.7 91.6 487 7.6 3.9 Sweden 7.63 8.61 4.63 5.58 7.06 8.52 15.8 19.2 85.3 99.5 513 6.0 3.4 Switzerland 7.89 7.16 6.11 5.36 8.70 8.02 6.7 8.6 79.8 83.1 506 6.8 3.7 United Kingdom 8.00 6.83 5.97 3.14 7.51 7.40 9.6 12.7 79.1 93.7 498 528 7.9 5.6 United States 6.53 6.55 5.31 3.73 7.61 7.07 6.7 8.7 85.8 90.2 492 499 9.4 7.1 Average 6.4 6.7 4.5 4.0 7.1 7.2 12.7 16.7 85.7 89.3 520.2 518.2 7.5 4.8 1/ Scale 1-10. 2/ In percentage of GDP. 3/ Ratio of the number of children of official school age enrolled in school, to the population of the corresponding official school age. Corruption 1/ Annex Table A - Opportunity indicators Annex – Data and sources 3.3 3.4 2.8 2.2 2.6 3.3 3.7 3.0 3.3 3.0 3.0 2.5 3.3 2.8 2.3 3.5 3.2 2.0 2.3 3.3 3.6 2.9 3.8 3.0 3.0 3.0 2.4 2.8 3.4 2.6 Public Communic. and transports quality 1980-89 1990-95 3.5 3.0 3.3 2.8 2.7 3.1 3.5 3.1 3.1 ECB Working Paper No 242 July 2003 31 Coefficient of Average inflation Per capita income 2/ Average economic variation of growth growth 1980s 3/ 1990s 3/ 1980s 1990s 1980s 1990s 1990 2000 1980s 1990s Australia 15.50 17.90 1.2 2.5 8.4 2.5 15530 25420 3.1 3.6 Austria 25.20 1.6 2.3 3.8 2.4 15710 24690 2.4 2.4 Belgium 21.60 24.10 1.3 1.6 4.9 2.1 15530 24910 2.0 2.2 Canada 18.95 1.1 1.3 6.5 2.2 17400 27320 2.9 2.9 Denmark 17.40 24.50 0.8 1.5 6.9 2.1 15820 27070 1.6 2.3 Finland 18.40 24.20 2.1 0.5 7.2 2.2 15220 23200 3.1 2.1 France 18.60 2.2 1.4 7.4 1.9 15970 21980 2.5 1.9 Germany 20.10 1.2 1.3 2.9 2.6 17010 23630 2.2 1.9 Greece 19.90 0.3 1.3 19.5 11.1 8680 15250 0.7 2.3 Iceland 0.9 0.9 39.2 4.3 16210 27070 2.8 2.7 Ireland 18.30 1.5 2.1 9.3 2.3 10940 26610 3.6 7.3 Italy 18.80 22.70 2.1 1.4 11.2 4.2 15180 22890 2.3 1.6 Japan 21.90 24.80 3.1 1.0 2.5 1.2 16950 24920 4.1 1.5 Luxembourg 1.4 1.9 4.8 2.2 22320 43110 5.0 5.4 Netherlands 20.65 1.2 2.8 2.9 2.4 15390 26310 2.3 2.9 New Zealand 15.90 12.70 1.0 1.2 11.9 2.1 12360 18740 1.9 2.8 Norway 19.00 24.00 1.2 2.9 8.3 2.4 16220 30730 2.4 3.6 Portugal 18.90 1.1 1.3 17.6 6.0 9120 16590 3.3 2.8 Spain 21.05 1.5 1.6 10.2 4.2 11320 18230 2.9 2.7 Sweden 21.20 24.10 1.7 1.0 8.0 3.5 16320 22940 2.2 2.3 Switzerland 16.90 19.60 1.2 0.7 3.3 2.3 19670 28360 2.1 0.9 United Kingdom 16.35 1.4 1.3 7.4 3.7 14860 23290 2.7 2.3 United States 15.70 15.70 1.4 2.2 5.6 3.0 21340 35030 3.2 3.2 Average 18.4 20.6 1.4 1.6 9.1 3.2 15438 25143 2.7 2.8 1/ Share of 40% poorest.2/ GDP at current market prices per head of population (in 1000 PPS). 2/ GDP at market prices per head of population (in 1000PPS). 3/ Or nearest available year. Precise year varies and depends on data availability. Income distribution 1/ Annex Table B – Standard “Musgravian” indicators 1980s 7.5 3.3 9.5 9.4 7.1 4.9 9.0 6.8 6.6 0.8 14.2 8.4 2.5 1.4 8.0 4.3 2.8 7.7 17.5 2.5 0.7 9.6 7.3 6.6 1990s 8.9 5.2 8.7 9.5 7.4 11.9 11.2 7.7 9.5 3.3 12.0 10.7 3.0 2.5 5.8 7.9 4.8 5.6 19.6 6.2 3.4 7.9 5.8 7.8 Average Unemployment Annex Table C – Expenditures categories (% of GDP) Total expenditure Goods and Education 1/ services 1980s 1990s 1980s 1990s 1980s 1990s Australia 37.4 36.7 19.1 18.6 5.1 5.1 Austria 49.7 53.8 19.4 19.9 5.6 5.6 Belgium 57.9 52.5 22.6 21.2 5.5 4.6 Canada 45.1 45.9 21.7 21.2 6.6 6.7 Denmark 56.3 58.3 26.6 25.9 7.1 7.8 Finland 43.4 56.3 20.3 23.0 5.2 7.1 France 50.3 53.6 23.0 23.6 5.5 5.8 Germany 47.1 48.2 19.8 19.5 4.7 4.7 Greece 40.5 47.3 15.0 14.7 2.2 2.7 Iceland 41.2 41.7 18.8 22.0 4.6 5.5 Ireland 46.1 37.7 18.9 16.0 5.5 5.1 Italy 50.6 52.2 18.9 18.8 4.5 4.4 Japan 31.9 36.2 13.7 15.0 5.1 3.6 Luxembourg 46.6 44.0 18.8 17.7 4.8 3.5 Netherlands 56.3 50.1 25.5 23.5 6.4 5.1 New Zealand 46.4 41.7 19.2 18.5 5.3 6.9 Norway 46.8 49.3 20.1 21.5 6.4 7.7 Portugal 39.5 43.7 14.5 18.9 3.8 5.2 Spain 39.0 43.4 15.6 17.9 3.5 4.5 Sweden 60.8 63.5 28.0 27.8 7.4 7.6 Switzerland 34.1 38.2 13.9 15.1 5.0 5.6 United Kingdom 42.3 40.9 20.9 19.5 5.0 5.2 United States 35.3 34.5 17.4 15.4 5.7 5.1 Average 45.4 46.5 19.6 19.8 5.2 5.4 Health 1980s 5.0 5.1 6.1 6.2 7.5 5.6 6.4 6.1 4.9 5.6 5.6 4.7 5.4 5.7 5.8 6.3 3.4 4.6 8.0 5.3 5.0 4.4 5.6 1990s 5.6 5.8 6.6 6.7 6.9 6.1 7.3 7.7 4.7 6.8 5.2 5.9 5.3 5.7 6.3 6.1 6.8 4.7 5.5 7.1 7.0 5.7 6.0 6.2 Social transfers 1980s 7.2 19.6 24.6 9.8 16.9 14.7 21.0 17.0 13.8 7.0 14.6 17.3 11.2 20.5 26.7 13.4 13.1 10.7 13.6 18.5 8.4 12.0 9.9 14.8 1990s 8.6 19.6 19.3 12.0 19.2 20.8 20.0 18.4 15.4 7.5 11.8 17.9 10.0 15.4 18.7 13.6 15.3 12.7 14.1 20.4 11.2 13.7 11.3 15.1 Public investment 1980s 1990s 3.0 2.5 3.6 2.6 2.6 1.6 2.9 2.5 2.0 1.8 3.7 3.0 3.2 3.2 2.5 2.3 3.0 3.4 4.3 4.1 3.3 2.5 3.5 2.5 5.1 5.7 4.7 4.5 2.3 2.6 2.1 2.1 3.4 3.3 3.6 3.9 3.4 3.6 2.9 2.8 3.7 3.1 1.9 1.6 2.5 2.6 3.2 3.0 1/ All general government, averages for the period. 32 ECB Working Paper No 242 July 2003 Annex Table D – Variables and series Variable Corruption Sources, notes Series World Economic Forum: The World Values divided by 10 for better comparison. Competitiveness Report 1990, item "10.22 Corruption (for 1990) World Economic Forum, The World Competitiveness Yearbook 2001, item 2.3.16 Bribing and corruption (for 2001). Red tape World Economic Forum: The World Values divided by 10 for better comparison. Competitiveness Report 1990, item "6.21 Regulatory environment (for 1990) World Economic Forum, The World Competitiveness Yearbook 2001, "Bureaucracy" (for 2001). Efficient judiciary World Economic Forum: The World Values divided by 10 for better comparison. Competitiveness Report 1990, item "10.04 Confidence in administration o justice" (for 1990) World Economic Forum, The World Competitiveness Yearbook 2001, "Justice" (for 2001). Size shadow economy Schneider (2002) Currency demand approach, (in % of official GDP), reciprocal value (1/x). Secondary school enrolment based on WDI 2001 Secondary school enrolment Education achievement OECD, Education at a glance, 2001 Mathematical achievement, grade eight (page 309). PISA report, 2000 Simple average of reading, mathematics and science scores. Infant mortality WDI 2001 Mortality rate, infant (per 1,000 live births), reciprocal value (1/x). Life expectancy WDI 2001 Life expectancy at birth, total (years). Communications and transport quality Center for Institutional Reform and the Informal Sector (IRIS) based on reports from Business Environmental Risk Intelligence (BERI). Income distribution Worldbank: World Development Report 1995, 2000/2001 Poorest 40 % (when two surveys within the time range of 86-98 were available the average was calculated). 2000 Annual Report (for 1990), 2002 Annual Report (for 2000). Coefficient of European Commission, Ameco variation of growth Based on GDP at constant market prices (1.1.0.0.ovgd), reciprocal value (1/x). Standard deviation of inflation OECD, Main Economic Indicators Based on "CPI, all items" (CPALTT01.IXOB), reciprocal value (1/x). Per capita income European Commission, Ameco Ameco, GDP at current market prices per head of population (in 1000 PPS) (1.0.212.0.hvgdp). Average economic growth European Commission, Ameco Based on GDP at constant market prices (1.1.0.0.ovgd). ECB Working Paper No 242 July 2003 33 Unemployment OECD, Economic Outlook Unemployment rate (UNR), reciprocal value (1/x). Total public expenditure European Commission, Ameco Total expenditure; general government (UUTG/UUTGF). Goods and services European Commission, Ameco Final consumption expenditure of general government at current prices (UCTG). Public education Based on WDI 2001 Public spending on education, total (% of GNI, UNESCO). Public health OECD, Social Expenditure database Public expenditure on health (item 11) (for 1980 - 1999). Transfers and subsidies European Commission, Ameco Social transfers other than in kind (UYTGH/UYTGHF) Public investment European Commission, Ameco Gross fixed capital formation at current prices; general government (UIGG). 34 ECB Working Paper No 242 July 2003 (XURSHDQ&HQWUDO%DQNZRUNLQJSDSHUVHULHV )RUDFRPSOHWHOLVWRI:RUNLQJ3DSHUVSXEOLVKHGE\WKH(&%SOHDVHYLVLWWKH(&%·VZHEVLWH KWWSZZZHFELQW ´$JJUHJDWHORDQVWRWKHHXURDUHDSULYDWHVHFWRUµE\$&DO]D00DQULTXHDQG-6RXVD -DQXDU\ ´0\RSLFORVVDYHUVLRQGLVDSSRLQWPHQWDYHUVLRQDQGWKHHTXLW\SUHPLXPSX]]OHµE\ ')LHOGLQJDQG/6WUDFFD-DQXDU\ ´$V\PPHWULFG\QDPLFVLQWKHFRUUHODWLRQVRIJOREDOHTXLW\DQGERQGUHWXUQVµE\ /&DSSLHOOR5)(QJOHDQG.6KHSSDUG-DQXDU\ ´5HDOH[FKDQJHUDWHLQDQLQWHUWHPSRUDOQFRXQWU\PRGHOZLWKLQFRPSOHWHPDUNHWVµE\ %0HUFHUHDX-DQXDU\ ´(PSLULFDOHVWLPDWHVRIUHDFWLRQIXQFWLRQVIRUWKHHXURDUHDµE\'*HUGHVPHLHUDQG %5RIILD-DQXDU\ ´$FRPSUHKHQVLYHPRGHORQWKHHXURRYHUQLJKWUDWHµE\)5:UW]-DQXDU\ ´'RGHPRJUDSKLFFKDQJHVDIIHFWULVNSUHPLXPV"(YLGHQFHIURPLQWHUQDWLRQDOGDWDµE\ $$QJDQG$0DGGDORQL-DQXDU\ ´$IUDPHZRUNIRUFROODWHUDOULVNFRQWUROGHWHUPLQDWLRQµE\'&RVVLQ=+XDQJ '$XQRQ1HULQDQG)*RQ]iOH]-DQXDU\ ´$QWLFLSDWHG5DPVH\UHIRUPVDQGWKHXQLIRUPWD[DWLRQSULQFLSOHWKHUROHRILQWHUQDWLRQDO ILQDQFLDOPDUNHWVµE\66FKPLWW*URKpDQG08ULEH-DQXDU\ ´6HOIFRQWURODQGVDYLQJVµE\30LFKHODQG-39LGDO-DQXDU\ ´0RGHOOLQJWKHLPSOLHGSUREDELOLW\RIVWRFNPDUNHWPRYHPHQWVµE\(*ODW]HUDQG 06FKHLFKHU-DQXDU\ ´$JJUHJDWLRQDQGHXURDUHD3KLOOLSVFXUYHVµE\6)DELDQLDQG-0RUJDQ)HEUXDU\ ´2QWKHVHOHFWLRQRIIRUHFDVWLQJPRGHOVµE\$,QRXHDQG/.LOLDQ)HEUXDU\ ´%XGJHWLQVWLWXWLRQVDQGILVFDOSHUIRUPDQFHLQ&HQWUDODQG(DVWHUQ(XURSHDQFRXQWULHVµE\ +*OHLFK)HEUXDU\ ´7KHDGPLVVLRQRIDFFHVVLRQFRXQWULHVWRDQHQODUJHGPRQHWDU\XQLRQDWHQWDWLYH DVVHVVPHQWµE\0&D·=RU]LDQG5$'H6DQWLV)HEUXDU\ ´7KHUROHRISURGXFWPDUNHWUHJXODWLRQVLQWKHSURFHVVRIVWUXFWXUDOFKDQJHµE\-0HVVLQD 0DUFK ECB Working Paper No 242 July 2003 35 ´7KH]HURLQWHUHVWUDWHERXQGDQGWKHUROHRIWKHH[FKDQJHUDWHIRUPRQHWDU\SROLF\LQ -DSDQµE\*&RHQHQDQG9:LHODQG0DUFK ´([WUDHXURDUHDPDQXIDFWXULQJLPSRUWSULFHVDQGH[FKDQJHUDWHSDVVWKURXJKµE\ %$QGHUWRQ0DUFK ´7KHDOORFDWLRQRIFRPSHWHQFLHVLQDQLQWHUQDWLRQDOXQLRQDSRVLWLYHDQDO\VLVµE\05XWD $SULO ´(VWLPDWLQJULVNSUHPLDLQPRQH\PDUNHWUDWHVµE\$'XUUp6(YMHQDQG53LOHJDDUG $SULO ´,QIODWLRQG\QDPLFVDQGVXEMHFWLYHH[SHFWDWLRQVLQWKH8QLWHG6WDWHVµE\.$GDPDQG 03DGXOD$SULO ´2SWLPDOPRQHWDU\SROLF\ZLWKLPSHUIHFWFRPPRQNQRZOHGJHµE\.$GDP$SULO ´7KHULVHRIWKH\HQYLVjYLVWKH´V\QWKHWLFµHXURLVLWVXSSRUWHGE\HFRQRPLF IXQGDPHQWDOV"µE\&2VEDW55IIHUDQG%6FKQDW]$SULO ´3URGXFWLYLW\DQGWKH´V\QWKHWLFµHXURGROODUH[FKDQJHUDWHµE\&2VEDW)9LMVHODDUDQG %6FKQDW]$SULO ´7KHFHQWUDOEDQNHUDVDULVNPDQDJHUTXDQWLI\LQJDQGIRUHFDVWLQJLQIODWLRQULVNVµE\ /.LOLDQDQG60DQJDQHOOL$SULO ´0RQHWDU\SROLF\LQDORZSDVVWKURXJKHQYLURQPHQWµE\70RQDFHOOL$SULO ´0RQHWDU\SROLF\VKRFNV²DQRQIXQGDPHQWDOORRNDWWKHGDWDµE\0.ODHIILQJ0D\ ´+RZGRHVWKH(&%WDUJHWLQIODWLRQ"µE\36XULFR0D\ ´7KHHXURDUHDILQDQFLDOV\VWHPVWUXFWXUHLQWHJUDWLRQDQGSROLF\LQLWLDWLYHVµE\ 3+DUWPDQQ$0DGGDORQLDQG60DQJDQHOOL0D\ ´3ULFHVWDELOLW\DQGPRQHWDU\SROLF\HIIHFWLYHQHVVZKHQQRPLQDOLQWHUHVWUDWHVDUHERXQGHG DW]HURµE\*&RHQHQ$2USKDQLGHVDQG9:LHODQG0D\ ´'HVFULELQJWKH)HG·VFRQGXFWZLWK7D\ORUUXOHVLVLQWHUHVWUDWHVPRRWKLQJLPSRUWDQW"µE\ (&DVWHOQXRYR0D\ ´7KHQDWXUDOUHDOUDWHRILQWHUHVWLQWKHHXURDUHDµE\1*LDPPDULROLDQG19DOOD 0D\ ´8QHPSOR\PHQWK\VWHUHVLVDQGWUDQVLWLRQµE\0/HyQ/HGHVPDDQG30F$GDP 0D\ ´9RODWLOLW\RILQWHUHVWUDWHVLQWKHHXURDUHDHYLGHQFHIURPKLJKIUHTXHQF\GDWDµE\ 1&DVVRODDQG&0RUDQD-XQH 36 ECB Working Paper No 242 July 2003 ´6ZLVVPRQHWDU\WDUJHWLQJWKHUROHRILQWHUQDOSROLF\DQDO\VLVµE\*5LFK-XQH ´*URZWKH[SHFWDWLRQVFDSLWDOIORZVDQGLQWHUQDWLRQDOULVNVKDULQJµE\2&DVWUpQ00LOOHU DQG56WLHJHUW-XQH ´7KHLPSDFWRIPRQHWDU\XQLRQRQWUDGHSULFHVµE\5$QGHUWRQ5(%DOGZLQDQG '7DJOLRQL-XQH ´7HPSRUDU\VKRFNVDQGXQDYRLGDEOHWUDQVLWLRQVWRDKLJKXQHPSOR\PHQWUHJLPHµE\ :-'HQKDDQ-XQH ´0RQHWDU\SROLF\WUDQVPLVVLRQLQWKHHXURDUHDDQ\FKDQJHVDIWHU(08"µE\,$QJHORQLDQG 0(KUPDQQ-XO\ 0DLQWDLQLQJSULFHVWDELOLW\XQGHUIUHHIORDWLQJDIHDUOHVVZD\RXWRIWKHFRUQHU"µE\ &'HWNHQDQG9*DVSDU-XO\ ´3XEOLFVHFWRUHIILFLHQF\DQLQWHUQDWLRQDOFRPSDULVRQµE\$$IRQVR/6FKXNQHFKWDQG 97DQ]L-XO\ ECB Working Paper No 242 July 2003 37 Does public sector efficiency matter? Revisiting the relation between fiscal size and economic growth in a world sample Konstantinos Angelopoulos University of Glasgow Apostolis Philippopoulos Athens University of Economics & Business, University of Glasgow, and CESifo Efthymios Tsionas Athens University of Economics & Business January 2008 Abstract: This paper revisits the relationship between fiscal size and economic growth. Our work differs from the empirical growth literature because this relationship depends explicitly on the efficiency of the public sector. We use a sample of 64 countries, both developed and developing, in four 5-year time-periods over 1980-2000. Building on the work of Afonso, Schuknecht and Tanzi (2005), we construct a measure of public sector efficiency in each country and each time-period by calculating an output-to-input ratio. In addition, we get an estimate of technical efficiency of public spending for 52 countries for the time-period 1995-2000 by employing a stochastic frontier analysis. Using these two measures, we find evidence of a non-monotonic relation between fiscal size and economic growth that depends critically on the size-efficiency mix. Keywords: Fiscal policy, government efficiency, growth. JEL classification: H1, E6, 04. Corresponding author: Apostolis Philippopoulos, Department of Economics, Athens University of Economics & Business, 76 Patission Street, Athens 10434, Greece. Tel. +30-210-8203357. Fax: +30-210-8203301. Email: [email protected] Acknowledgements: We thank S. Kalyvitis, P. Kammas, M. Katsimi, J. Malley, M. Ntelis, H. Park, E. Tzavalis and seminar participants at the University of Stirling and the 2007 Scottish Economics Society Conference for discussions and comments. Any errors are ours. The first coauthor is grateful to the "Foundation Propondis" for their support. 1. Introduction The relationship between government size and economic growth is not expected to be monotonic. On one hand, governments provide public goods and services and correct market failures. On the other hand, policy intervention generates its own distortions, as it requires taxes and distorts incentives. There is thus a tradeoff depending on the size-efficiency mix of the public sector. By efficiency, we mean the ability of the government to transform its revenues into public goods and services that benefit the economy and promote growth. After a critically large size, or a critically low efficiency, the costs of a larger public sector outweigh the benefits.1 This paper revisits the relation between fiscal size and economic growth. Our work differs from the empirical growth literature because this relation depends explicitly on the efficiency of the public sector. We use a sample of 64 countries, both developed and developing, in four 5-year periods over 1980-2000. To obtain a measure of government efficiency, we follow the methodology of Afonso, Schuknecht and Tanzi (2005) for the OECD and construct measures of public sector efficiency (PSE). This index measures the efficiency of public sector in reaching a range of objectives of government intervention. It is basically the ratio of performance indicators (output) to a measure of public expenditure related to those indicators (input), based on the assumption that the input is used to achieve that output. We construct such indexes of public sector efficiency for four policy areas: administration, stabilization, infrastructure and education. In addition to this measure, focusing on 52 countries for the sub-period 1995-2000 during which more data are available, we also obtain an estimate of the so-called technical efficiency (TE) of the public sector by applying a stochastic production frontier analysis (see e.g. Kumbhakar and Lovell, 2000, and Greene, 2005). The ranking of countries according to the TE measure does not differ substantially from that implied by the PSE measure. We then incorporate these two measures (PSE or TE) into a simple econometric model in which the size-growth relationship is non-monotonic depending on the size-efficiency mix. This novel feature is included into an otherwise standard growth regression (see e.g. Barro and Sala-iMartin, 2004, chapter 12). Our main finding is that, when the fiscal size is measured by the government consumption share in GDP, the size-efficiency mix is significant in explaining the size-growth relationship. The latter is indeed non-monotonic as discussed above. This result holds for both efficiency measures 1 A simple and popular conceptual framework is provided by Barro’s (1990) model, where there is a trade-off between growth-promoting public goods and the distorting taxes required to finance them. When the government size and its associated tax burden are high (resp. small) relative to the productivity of public sector, a larger size is bad (resp. good) for growth. See also Hillman (2003) and Mueller (2003) on the market failures vs policy distortions trade-off. 1 constructed and is robust to a number of changes in the econometric specification, as well as to dividing the world sample into two sub-samples consisting of “high-income” and “developing” countries. Among other things, the model provides an endogenously determined efficiency threshold below (resp. above) which the size-growth relationship is negative (resp. positive). In general, this relationship is found to be negative in most countries and time periods. When we use, for instance, the PSE as a measure of efficiency in our world sample for all four 5-year periods, our estimates imply that only in 34 out of 159 observations (different countries in different periods) the size-growth relationship is positive.2 Our results imply that what really matters to growth is not the government size per se, but the size-efficiency mix. They can also help to explain why the evidence on the growth effects of the overall fiscal size has so far been mixed (see e.g. Levine and Renelt, 1992, Tanzi and Zee, 1997, Gemmel and Kneller, 2001, and Mueller, 2003, chapter 22). Essentially, our results suggest that it is difficult to obtain a “robust” effect of the overall fiscal size on economic growth when important elements that shape the size-growth relationship (in our case, the efficiency of the public sector) are omitted from the analysis.3 In sum, as Levine and Renelt (1992, p. 951) point out, “using simple expenditure data without accounting for government efficiency may yield inaccurate measures of the actual delivery of public services”. The rest of the paper is as follows. Section 2 develops measures of government efficiency. Section 3 studies the growth effects of the size-efficiency mix. Conclusions are in Section 4. 2. Measures of government efficiency In this section, we present two measures of government efficiency. 2.1 Public sector efficiency Following Afonso et al. (2005, 2006), we construct sub-indices of relative Public Sector Efficiency (PSE) in certain policy areas in each country and each time period, and then take the average of these sub-indices to obtain an index of aggregate government efficiency in each country and each time period. 2 Regarding the causal effect of fiscal size on economic growth, a concern has been the potential endogeneity of fiscal size. The literature so far has not provided a “credible” identification of fiscal size in growth regressions (see e.g. Agell et al., 2006). Although the aim of our paper is not to resolve the causality issue, we also provide some evidence that it can be easier to find a credible identification of the size-efficiency mix, rather than of size alone, in growth regressions. 3 An additional potential explanation that has received a lot of empirical support is that the overall size of government cannot capture the different implications of different government activities. As has been shown (see e.g. Devarajan et al., 1996, Kneller et al., 1999, and Angelopoulos et al., 2007), the growth effects of the different components of government expenditure, as well as of the various types of tax instruments, are not the same. See also Angelopoulos and Philippopoulos (2007) for a single country, time-series study that also supports the result that both the composition and efficiency of the government matter. 2 Afonso et al. have constructed PSEs for seven policy areas for OECD countries over the eighties and nineties. Here, we focus on four policy areas (education, administration, infrastructure and stabilization) for 64 countries, both industrialized and developing, and four 5-year time-periods, over 1980-2000 (obviously, due to data availability, there is a trade-off between the number of countries and the number of policy areas).4 We keep only those observations for which indexes of government efficiency in all four areas are available. Since the methodology is in Afonso et al. (2005, 2006), here we only discuss the basic insight and point out where we differ. The basic insight of this methodology is to compare the performance of government in certain areas of economic activity (where these areas are influenced directly by government intervention) to the associated expenditure that the government allocates to achieve this particular performance. Thus, to construct a PSE index, we need a measure of Public Sector Performance (PSP) and a measure of the associated Public Sector Expenditure (PEX) for each country in each policy area and each time-period. Then, the PSE will be the ratio of PSP to PEX. More details about the construction of PSP and PSE indexes in each policy area are in our Appendix. To make these PSP and PEX measures (expressed in different units of measurement) comparable across countries, we follow Afonso et al. by expressing each country’s PSP and PEX relative to the average PSP and PEX of all countries in each period, and this is done for all periods and indexes. In other words, each country’s PSP and PEX are expressed as percentages of the respective average (normalized to be 1), and in turn the PSE is obtained as the ratio of these relative PSP and PEX.5 Therefore, the resulting PSE is an index that measures the efficiency of a country’s government relative to governments in other countries in each period in a particular policy area. The larger the value, the more efficient the country’s government is. This is the notion of relative efficiency in Afonso et al. Table A.1 in the Appendix reports the relative PSPs, and the resulting PSEs, in the four policy areas for the countries and the time-periods that data are available. The order of countries is alphabetical. The second-from-the-end column in Table A.1 reports the (relative) aggregate efficiency of a country’s government obtained as the average of the four (relative) sub-indices. As expected, high-income OECD countries get on average better scores, although the public sectors in economies like Korea, Thailand or Malaysia appear to be particularly efficient. The most efficient governments during 1995-2000 are those of Korea (2.221), Canada (2.039), the USA (1.938) and 4 Greene (2005) has measured the efficiency of public spending in developing countries focusing on the areas of health and education. Afonso et al. (2006) have also constructed measures of public sector efficiency for a group of 24 uppermiddle income countries for the late nineties. 5 Since the averages of PSP and PEX are both normalized to be 1, the resulting PSE has an average around 1 (specifically, the PSEs in education and stabilization have an average of about 1.1, whereas the PSEs in infrastructure and administration have an average of about 1.25). 3 Switzerland (1.813) that are twice as efficient as the average countries, e.g. United Kingdom or France. At the bottom end, Namibia (0.483), Nicaragua (0.447) and Yemen (0.35) score about half of the average score.6 Of course, we have to be cautious with these estimates. For instance, in rich countries, like Finland or Sweden, the cost of resources used for providing public education or capital is higher than in say Uruguay or Lebanon, and this may result in an overestimation of relative efficiency in the latter group of countries. In addition, government performance in a certain policy area may be overestimated when private resources are used to complement government policy; this is especially the case of education in many countries (e.g. Greece). In sum, the main advantage of the above output-to-input approach is its simplicity and logical coherence, which allow a meaningful comparison across countries. Its main weakness is that several assumptions have to be made to calculate such a composite index (for a critical assessment of different methodologies and measures of public sector efficiency, see e.g. Afonso et al., 2005 and 2006, as well as the special issue of European Economy, no. 3, 2004, on “Public finances in EMU 2004”). 2.2 A stochastic production frontier methodology As an alternative approach to measuring government efficiency, we estimate a stochastic production frontier for the public sector and then obtain an estimate of the so-called Technical Efficiency ( TE ) of this sector. For a review of this methodology, see Kumbhakar and Lovell (2000). Our stochastic frontier model is of the form: ln y i = β 0 + β 1 ln x i + v i − u i (1) where y i is a measure of public sector output in country i , xi is a measure of public sector input, ui is the nonnegative technical inefficiency component of the error term, and v i is the noise component assumed to be distributed normally and independently of u i . Both error components are assumed to be independent of the regressors. 6 Two countries score suspiciously high in this Table. Paraguay, which seems to be the most efficient country in the world, and Argentina, which seems to be the second most efficient country in the last time-period. Regarding Paraguay, this result is driven by a very high score in the variable Electric Power Transmission and Distribution Losses (see the Appendix), which results in a very high PSP in infrastructure. This score may reflect measurement errors or unusual circumstances, so we drop Paraguay from our regressions in the next section. Regarding Argentina, the high efficiency score for 1995-2000 is probably due to the extended stabilization program implemented by the country in this period. We also choose not to include Argentina in our analysis in the next section. We report, however, that including these two countries does not have a significant effect on the econometric results presented later. 4 After estimating equation (1) by maximum likelihood, a measure of technical efficiency for each country i ( T E i ) is defined as: T E i = E [ ex p { − u i } / ε i ] (2) where ε i = vi − ui (see Kumbhakar and Lovell, 2000, chapter 3, for details). This efficiency score is bounded between zero and one. To apply the above, we need to measure public sector outputs and inputs ( y i and xi , respectively). We use the average of the PSP indices as a measure of y i . As a measure of xi , we use Total Government Expenditure (as a share of GDP) which is available from the World Development Indicators. We estimate (1)-(2) under the assumption that u i is characterized by a nonnegative half-normal distribution (we have also examined the case where u i is assumed to follow a truncated normal distribution but, since this gives very similar results, we discuss only the nonnegative half-normal case). Results for each country’s technical government efficiency ( TE i ) during the 1995-2000 sub-period (where we again look at a 5-year period average, as we did with the PSE measure above) are reported in Table A.2.7 The ranking results look sensible again. In this cross-section world sample during 1995-2000, Switzerland’s government scores the best being followed by Sweden and Finland. Again, as probably expected, governments in OECD countries are more efficient than those in developing countries, although public sectors in fast-growing economies like Thailand, Malaysia, Cyprus and especially Korea get high scores. Algeria, Nicaragua and Yemen have now the least efficient governments. Therefore, the ranking of countries using the PSE measure does not differ substantially from that using the TE measure (recall that this refers to the 1995-2000 period during which both measures are available) with the correlation coefficient being 0.75. In this sample, an LR test of the null that σ u2 = 0 gives a value of 5.64, which rejects the null (the respective p-value of the test is 0.009).8 This implies that government technical efficiency differs significantly across countries during 1995-2000. We report that we have also estimated government TE during the three time-periods before 1995 (i.e. the three 5-year periods between 1980 and 1995). However, there are significantly less data available for these earlier years 7 To examine whether the TE i estimates in Table A.2 are not biased due to heteroskedasticity in either v i or u i (see Kumbhakar and Lovell, 2000), we have tested whether the variance functions of vi or u i depend (linearly) on govexp. Since this is rejected, we can have some faith in the homoskedasticity assumption. The limiting distribution of the LR test statistic is a mixture of a chi-square with zero degrees of freedom, i.e. a point mass at zero, and a chi-square with 1 degree of freedom (see e.g. Kumbhakar and Lovell, 2000). The p-value of the test reported here takes this into account. 8 5 (especially in the eighties when the sample size drops to around 25-30, i.e. it mainly consists of the OECD countries). Not surprisingly, we have not been able to reject the null σ u2 = 0 for any of these early periods. Hence, concerning the TE measure, we concentrate on the 1995-2000 period. In sum, the TE measure has obvious advantages but, on the other hand, it depends on the assumptions made about the error term. The assumption that government expenditure is uncorrelated with the error term may be strong when governments respond to negative shocks by increasing their expenditures. In any case, the TE measure of government efficiency provides a useful alternative measure also used below to check the importance of the size-efficiency mix. 3. The size-efficiency nexus matters to growth This section tests whether there is a non-monotonic relationship between government size and economic growth with this relationship driven by the size-efficiency mix. 3.1 Econometric model We use the above constructed measures of government efficiency (PSE or TE) in a growth regression of the following form (see Dutt and Mitra, 2002, for a similar specification in a trade policy context): growthit = α 0 + α 1 sizeit + α 2 sizeit * eff it + X it β + ε it (3) where growthit is the growth rate of country i at time t , sizeit is a measure of government size, eff it is a measure of government efficiency (PSE or TE) and X it includes control variables usually included in growth regressions (see below). The partial derivative with respect to sizeit is simply: ∂growthit = α 1 + α 2 eff it ∂sizeit (4) where we expect α 2 to be positive in the sense that the more efficient the public sector, the larger the positive effect of government on growth. We also expect α 1 to be negative to catch the adverse effects of government size on growth. As long as the estimated coefficients α 1 and α 2 in (3) are statistically significant and have the right signs, so that the size-efficiency nexus matters to growth, the above specification can also 6 give an estimate of a (common to all countries) critical level of efficiency, eff * , where eff * ≡ −(α 1 / α 2 ) > 0 makes the partial in (4) equal to zero. When an individual country’s efficiency, eff it , is higher (resp. lower) than eff * , the positive (resp. negative) effects dominate and the country is placed on the positively (resp. negatively) sloped part of the size-growth curve; this, of course, requires eff * to lie within the range of values of eff it in the data. Note that (3)-(4) imply that the growth effects of fiscal size can differ among countries and time-periods.9 3.2 Data and variables used in the regressions For the eff variable, we use the two measures of government efficiency (TE and PSE) constructed in section 2 above. The rest of the variables are as in most of the literature. We work with 5-year period averages as we did with our eff measures (5-year periods are also used in the growth literature, especially the literature on the growth effects of fiscal policy, see e.g. Folster and Henrekson, 2001, and Kneller et al, 1999). The main datasets used are the Penn World Tables (PWT) version 6.1 (see Heston et al., 2002) and the World Development Indicators (WDI) developed by the World Bank. Our dependent variable, the growth rate of per capita GDP, is from the PWT. In particular, the PWT dataset provides us with the real GDP per capita in constant prices, which is then used to obtain the five-year average of annual growth rates (denoted as growth in our regressions). The PWT also provides us with consumption of the general government as a share of GDP in constant prices, which is averaged over 5-year periods to give a variable denoted as govshare in our regressions. This will be our primary measure of government size.10 An alternative measure of government size, which is also used below, is total expenditures of the central government as a share of GDP (denoted as govexp in our regressions and obtained from WDI). This variable includes transfers and interest payments on public debt, in addition to government consumption (note that to avoid double counting, we do not include government investment in our govexp measure, as government investment is included in the investment share in GDP used as a separate regressor (see below). We have also examined a specification like growthit = α 0 + α 11 size it + α 12 size it 2 + X it β + ε it , which gives a partial as a function of size, so that an “optimal” size can be calculated given the estimated coefficients irrespectively of efficiency. We report that estimation of this equation does not give meaningful results (coefficients are not significant and in some regressions they have wrong signs). 10 This is the general government consumption component of GDP. It does not include public investment, interest payments, subsidies and other transfers. Public investment is included in PWT in the variable “investment share in GDP” (see below). Note however that a large part of government spending on goods and services, included in govshare, has investment features (e.g. salaries of teachers, professors and doctors and spending on police or the judiciary system). The variable govshare is closer to what Tanzi and Schuknecht (2000) refer to as a measure of “real government expenditure”. 9 7 Concerning the above two measures of fiscal size, an advantage of govshare over govexp is that it refers to the general government and can thus capture better the full trust of fiscal size on economic growth; moreover, it is PPP adjusted and therefore more suitable for international comparisons. The advantage of govexp, on the other hand, is that it allows us to examine whether including more types of government expenditure (at the disadvantage of using data at the central level only) gives different results regarding the effect of fiscal size on growth. Ideally, we would like to have a measure of general government spending for all types of government expenditure, but unfortunately, such a measure does not, as far as we know, exist for all the countries and time periods in our world sample. Finally, the fiscal size of government can be also measured by tax revenue or the budget balance, both as shares of GDP (see e.g. Tanzi and Zee, 1997, and Persson and Tabellini, 2003); see below in subsection 3.4 for details.11 In our choice of the control variables included in X in equation (3) above, we will follow most of the literature (see e.g. Barro and Sala-i-Martin, 2004, chapter 12, and the review papers mentioned above). Thus, we use the logarithm of the initial level of GDP per capita (denoted as lgdp), obtained from PWT, to control for convergence effects; the initial (or the value closest to the beginning of the period) secondary school enrolment rate (denoted as enrol), obtained from WDI, to proxy for human capital;12 the investment share of GDP (denoted as investment), obtained from PWT and averaged over the 5-year period; the logarithm of the fertility ratio (denoted as fertility), obtained from WDI; a measure of openness (denoted as openness), obtained from PWT and defined as the sum of exports and imports over GDP.13 Finally, we include in our regressions time dummies, as well as regional dummies for countries in Sub-Saharan Africa, East Asia, Latin America and the economies in transition. 3.3 Basic results Results using the PSE measure of efficiency for the sample of 64 countries over 1980-2000 are presented in Table 1. We report standard errors obtained under the assumption of spherical errors and standard errors that are robust to arbitrary heteroskedasticity and arbitrary intra-country serial 11 The tax revenue-to-GDP ratio is generally not preferred to fiscal spending measures, mainly because of tax evasion problems (see e.g. Tanzi and Zee, 1997). The same can be said about the budget-to-GDP ratio since it includes tax revenue. 12 A better proxy for human capital could be a measure of the average years of schooling (see e.g. Barro and Sala-iMartin, 2004). However, such measures are not available for all the countries in our sample and we do not want to restrict our sample for any other reasons than the requirements for the efficiency measure. Hence, we use the enrol variable, also used by Levine and Renelt (1992). 13 We have also used the average annual growth rate of the labour force, obtained from the WDI, in the growth regressions, but it is always insignificant. 8 correlation (see e.g. Wooldridge, 2002). The first three columns report estimates when using govshare as a measure of fiscal size and the last three when using govexp.14 Table 1 around here In column 1 of Table 1, we start with a standard growth regression: the coefficient of govshare is significantly negative. In column 2, we add the PSE measure of government efficiency, which is positive but marginally significant, while the coefficient of govshare remains significantly negative. To examine whether it is government efficiency that shapes the size-growth relationship, we move to column 3, which presents results for our key equation (3) above.15 Both estimates of govshare and govshare*eff are significant with the expected sign (negative and positive respectively), indicating a heterogeneous across countries size-growth relationship depending on government efficiency. Actually, the estimates imply a threshold of eff * = 1.358 , which means that only in 34 out of 159 observations (different countries in different time periods), the size-growth relationship is positive. The estimated coefficients α 1 and α 2 also allow us to calculate the growth effect of fiscal size in each country and each time period, as implied by equation (4). Results are reported in the last column of Table A.1. As can be seen, the estimated effect differs substantially across countries. There is a small group of countries where public sectors are efficient meaning a positive growth effect from fiscal size. This group includes Canada, Japan, Korea and Switzerland in all time periods we have data for; and Australia, Finland and the USA in most time periods (here we report those countries with more than one observation/time period; see Table A.1 for all countries). However, for most countries and time periods, this effect is negative. Therefore, the general picture that emerges is that fiscal sizes have grown too much - relative to public sector efficiency - in the last decades. This finding is similar to the arguments made in e.g. Gwartney et al. (1998) and Tanzi and Schuknecht (2000) although these papers do not take account of efficiency explicitly. Regarding the control variables that enter significantly, lgdp is negative, implying (conditional) convergence, while investment and openness are positive. The effect of fertility is negative (this is as in Barro and Sala-i-Martin, 2004, chapter 12) but not robustly significant. The effect of enrolment is positive but not significant. Regarding the regional dummies, those for the economies in transition are significantly negative, while those for Latin American countries are 14 We do not include a dummy for each country (and thus we do not estimate fixed effects regressions) as this would result in losing all cross-country variation. This is important because the measure of efficiency developed here is a relative one across countries. It would make little sense to use this variation to explain differences within countries only. 9 negative but not significant when we use robust standard errors. An interesting result is the negative dummy for East Asian countries, as this variable usually has a positive effect in similar regressions (see e.g. Barro and Sala-i-Martin, 2004, chapter 12). However, East Asian countries, in general, are ranked highly in our efficiency measures (see Table A.1), so that a large part of the positive regional effect has been already controlled for by our fiscal measure. The results are less clear when we use the other widely used measure of fiscal size, govexp (see the last three columns in Table 1). The coefficient of govexp is negative but not robustly so (see column 5 that includes pse). More importantly, in column 6, there is no significant evidence of a non-linear relationship like the one found in column 3; namely, the coefficient of govexp*pse is not significant (although it has the right sign). Recall that the key difference between govshare and govexp is that the latter includes redistributive transfers and interest payments on public debt. Both items (i.e. transfers and interest payments) do not involve a direct use of real resources by the state sector (recall the economy’s resource constraint). We thus do not find it surprising that govexp does not give as clear results as govshare. In a sense, these new results indicate that both the sizeefficiency mix and the composition of government expenditure matter to growth.16 3.4 Robustness of basic results We now examine the robustness of the basic results above by extending the empirical specification in two dimensions. First, we test whether our results - regarding the importance of the sizeefficiency mix on growth - are sensitive to the financing assumption of government spending (see e.g. Miller and Russek, 1997, and Kneller et al., 1999). Given that we do not have detailed tax and spending data for all the countries and time periods in our sample, we use a general form of government budget that equates aggregate spending to tax revenue and deficit (see e.g. Miller and Russek, 1997). In principle, in the absence of Ricardian equivalence, the effect of spending on growth can be different depending on whether higher spending is financed by more tax revenues or by a larger budget deficit (higher debt). If, for instance, we include a measure of taxation, together with spending, in a growth regression, we would expect the effect of the tax measure to be negative 15 We do not include eff together with size * eff in the same regression, as they are highly correlated and as a result both eff and size * eff become insignificant. In this specification, the growth effect of government efficiency takes place only via government size, assuming that efficiency is independent of size. 16 We have also used another potential measure of the extent of government involvement in the economy, the so-called Economic Freedom index as developed by the Fraser Institute (see e.g. Gwartney et al., 2006). The Economic Freedom (EF) index is a rather general measure of government involvement than includes the size of government; the degree of regulation of credit, labor and business by the government; the legal structure; the security of property rights; the freedom to trade; etc. We report that, when we use the EF index as a measure of fiscal size in our regressions for the world sample (i.e. instead of govshare and govexp), then (a) it has a negative growth effect (see also De Haan et al., 2006) although this effect is not always significant (b) the estimated α 2 is not significant in equations (3)-(4) above. We believe this is not surprising given that this index contains more variables than the size of the government, while equations like (3) test whether the growth effect of size depends on the size-efficiency mix. Besides, the EF index may be correlated with government efficiency. 10 capturing the adverse implications of a larger fiscal size, whereas the effect of the spending measure to be positive capturing the positive effects of e.g. more public good provision. It is therefore interesting to see whether our results are robust to the inclusion of a finance instrument (obviously, because of multi-collinearity problems, we cannot include both tax revenues and public deficits in the regressions). For our sample, we obtain data for tax revenues, as a share of GDP, from the WDI database (we denote the respective measure, which is again expressed in 5-year period averages, as tax). We then rerun the basic regressions of Table 1 by including tax as an additional explanatory variable. Results for the main variables are shown in Table 2 (since the estimates for the control variables are not generally affected, we do not include them in Table 2 to save on space - these results are available upon request). As can be seen, the results of Table 1 remain essentially unchanged when we include tax, which, itself, is not significant. We report that these results again do not change if we use deficits instead of taxes.17 Tables 2 and 3 around here Second, we also test whether the inclusion of lagged growth rates changes our results. Although our basic specification (see Table 1) is common in the empirical growth-policy literature working with 5-year averages (see e.g. Kneller et al., 1999, and Folster and Henrekson, 2001), dynamic effects from past growth may persist even after five years. Therefore, we now examine whether the size-efficiency mix retains its significance in explaining economic growth, even after controlling for lagged growth rates (see also Miller and Russek, 1997). Results obtained from including the lagged-once growth rate (denoted as grolag in our regressions) as an explanatory variable in the regressions of Table 1 are reported in Table 3 (again, we present results for the main variables only to save on space). Note that the sample size drops from 159 to 98 observations (there are now 46 instead of 62 countries). The lagged-once growth rate is generally significant, but the results for the main variables of interest are not qualitatively affected. Actually, in column (3), where we present our key results by using govshare, grolag is not found to be significant. 17 Notice, when we compare Tables 1 and 2, that the inclusion of tax does not alter the negative effects of govshare and govexp in columns (1) and (4) respectively. Thus, the effect of government size itself, as measured by govshare or govexp, remains negative even if we add a measure of the tax burden, tax. This is probably because tax revenues, as an ex post measure, is not an ideal proxy for the distortions imposed by the tax system; higher tax revenue may e.g. reflect less tax evasion and better institutions (see Tanzi and Zee, 1997 and Angelopoulos et al., 2007, for discussion and references). Thus, the basic size-efficiency specification in Table 1 appears to be good enough to capture the trade-offs in fiscal policy at least in our sample. In other words, to the extent that we allow the effect of the fiscal size to depend on the size-efficiency mix, we view our basic specification as an alternative to including both spending (see positive effects) and taxation (see negative effects) to capture the trade-off in fiscal policy. 11 Finally, in Table 4, we present results for the main variables by including both grolag and tax in our regressions. As can be seen, the previous results and analysis remain robust to this specification as well. Table 4 around here 3.5 High-income and developing countries So far - although we allowed for the effect of fiscal size to differ across countries depending on the efficiency of the public sector in each country - we have studied rich and developing countries jointly in a single sample. We now divide countries into two subgroups to study whether the sizeefficiency mix matters differently in high income and developing countries (where we classify countries as high income following the classification in the WDI dataset). For each group, we first calculate the measure of public sector efficiency (PSE) separately, repeating the steps described in sub-section 2.1 above (since the efficiency measure is re-constructed for more homogeneous groups of countries, this can provide an additional robustness test). Tables 5 and 6 around here Using these new PSE measures, Tables 5 and 6 rerun the basic regressions of Table 1 for high income and developing countries respectively (again, we present results for the main variables only to save on space). As can be seen, the results remain practically unchanged for the subgroup of high-income countries in Table 5. For the subgroup of developing countries in Table 6, the main story, regarding the importance of the size-efficiency mix, is again supported when we use govshare as a measure of fiscal size (see column (3) in Table 6), which is as in the world sample above. It is interesting to note that, for developing countries, public expenditure is not significantly related to economic growth in the first two columns, but significance is restored in column (3) that explicitly allows for the size-efficiency mix. All this suggests that in both subgroups, our story - that the sizeefficiency mix matters - is confirmed by the data. We finally report that these results are robust to the inclusion of tax as an explanatory variable (see subsection 3.4 above). On the other hand, including grolag reduces the sample size in both subgroups too much to give any reliable results. 3.6 Can the size-efficiency mix help with endogeneity? When looking for a causal effect from fiscal policy in a growth regression, a usual concern is that there might be a reverse causality when e.g. governments respond to negative shocks by increasing 12 their expenditure (see e.g. Tanzi and Zee, 1997, and Agell et al., 2006). Although this problem is to some extent mitigated here since we work with 5-year averages, such reverse causality cannot be excluded. In addition, our fiscal size variables, and especially the measure of government efficiency, may be correlated with the error term due to omitted variables or measurement error. The natural approach to dealing with such an endogeneity is to use instruments for the endogenous variables in IV methods. A fundamental concern with IV regression methods, however, is whether the instruments are valid and relevant. As far as we know, the relevant literature has not yet provided a credible identification of fiscal policy so that the instruments used are both exogenous and strongly correlated with the endogenous variables (see e.g. Agell et al., 2006). We now investigate whether accounting for the size-efficiency mix can help in this direction. We will build upon the basic specification of subsection 3.3. We need instruments for size and size*eff in 2SLS regressions. As such instruments, we use variables usually considered as potential determinants of fiscal policy (see e.g. Person and Tabellini, 2003, chapter 3). In particular, we use the age dependency ratio (agedep) and two measures of country size (population and surface, denoted respectively as pop and surface). All these three variables are obtained from WDI and, except for surface, are averaged over the 5-year periods. In Table 2, we present results for the core variables when we re-estimate the basic regressions of Table 1 by using these instruments in 2SLS methods (the results for the control variables do not change significantly, so we do not present them to save on space). Table 7 around here We start again with the govshare variable. When we do not account for efficiency (column 1 in Table 7), the Sargan over-identifying restrictions test rejects the null that the instruments are uncorrelated with the error term. However, when efficiency is included as an endogenous variable, either on its own (column 2) or multiplicatively with govshare (column 3), the null clearly cannot be rejected (the p-value is very low in both cases). Therefore, in this sample, the instruments affect growth only indirectly through the size-efficiency mix. Note also that the Anderson (1984) canonical correlations, and the Cragg and Donald (1993) tests of whether the equation is underidentified, reject the null thus lending some support to the relevance of the instruments.18 More importantly, the first-stage F-statistic is very high for the govshare*eff variable, which indicates that the instruments are strongly correlated with this variable. Although the first-stage F-statistic for govshare is not as high, it is clear that the diagnostics favor the key regression in column 3 that controls for the size-efficiency mix. In this regression (in column 3), the critical eff * = 1.238 18 These tests have been implemented using the routines written by Baum et al. (2006). 13 implies that in 46 countries/periods there is a positive effect on growth from govshare. The fact that the critical efficiency level is lower in the 2SLS regressions indicates that the estimate of fiscal size is biased downwards when endogeneity is not accounted for, so that the “true” effect of fiscal size may in fact be less negative (or more positive) than implied in Table 1 for many countries. As in Table 1 above, the results are not so promising when we use the govexp variable as a measure of government size. Although the Sargan test does not reject the validity of the instruments, the Anderson (1984) canonical correlations and the Cragg and Donald (1993) tests cannot reject the null that the equation in column 6 of Table 7 is under-identified. Therefore, although further research is clearly required concerning the issue of causality in the fiscal policy-growth relation in cross-country growth regressions, our results suggest that taking account of the size-efficiency mix can help in identifying the growth effects of fiscal policy. 3.7 An alternative measure of government efficiency To further examine the robustness of our results, we also use the TE measure of efficiency instead of PSE. Again, we will build upon the basic specification of subsection 3.3. As explained in section 2, we have been able to obtain the TE measure for the 1995-2000 period only. In Table 8, we present results focusing on this period. Actually, in this table, we report results for both the PSE and TE indices of government efficiency, and both the govshare and govexp measures of fiscal size. This has the additional advantage of checking whether there has been a structural break in the size-efficiency-growth relationship of equation (3). The regressions in Table 8 are the same as those in Table 1, except that now we do not include time dummies. Table 8 around here We start again with govshare (columns 1-3). The average effect of govshare is negative (column 1), while the size efficiency mix (when we use the PSE measure for efficiency) is important (column 2). Thus, the non-monotonic relationship holds for both the whole period and the 1995-2000 sub-period. The critical level of efficiency is now eff * = 1.216 , which implies that for 24 out of 51 countries in this period the size-growth relationship is positive. Note also that the regression with the size-efficiency mix is much better that the regression without it, as can be seen by both the increase in R 2 and the fact that the coefficients of lgdp, openness and East Asia become significant. Regarding lgdp, in particular, this implies that the size-efficiency mix is an important long-run determinant of economic growth that has to be conditioned upon so that convergence can be captured in the data (see e.g. Barro and Sala-i-Martin, 2004, chapter 12, for conditional convergence). 14 Then, we estimate equation (3) for the 1995-2000 sample by using TE as the efficiency measure. Results are in column (3). The coefficients are again significant with the right signs. The critical efficiency is now eff * = 0.889 , which implies that only in 8 out of 51 countries in this period the size-growth relationship is positive (see the last column in Table A.2 for the estimated growth effect in each country in this case). These are Finland, Korea, Sweden and Switzerland, as well as (but only marginally) Canada, Germany, Iceland and Uruguay. Note, however, that the regression with the PSE measure in the size-efficiency mix explains about 10% more of the variation in the growth rate than the regression with the TE measure. In columns 4-6 of Table 8, we repeat the same regressions by using govexp as a measure of government size. As before, govexp is negative and significant, while the size*eff variables have a positive sign but are not significant. As we did in Table 7, we have also run 2SLS regressions for the equations in Table 8 by using the same set of instruments for the size-efficiency mix. The estimated coefficients are again supportive of the importance of the size-efficiency nexus, at least for the govshare measure, but the first stage regression diagnostics reveal that the instruments are not strongly correlated with the endogenous variables. Since the small sample size does not help us to draw any safe conclusions, we find the results of Table 7 to be more reliable. In any case, as discussed above, the identification of fiscal policy remains a challenge in this literature. Finally, we report that with the TE measure of efficiency, we cannot divide countries into rich and developing, as we did in subsection 3.5 (the sub-samples are now too small). Concerning the addition of tax in the regressions (as we did in subsection 3.4 above), we report that once more the main results are not affected. Therefore, the main result from this subsection is that the relationship between the sizeefficiency mix and economic growth is robust to the time period and the measure of government efficiency used. 4. Concluding remarks We revisited the relationship between fiscal size and economic growth and provided evidence that this relationship depends on the size-efficiency mix of the public sector. The policy implication is that what matters to growth is not the size per se, but the size-efficiency mix. Of course, improving the efficiency of the public sector is not an easy task. It requires, among other things, the reallocation of government resources, as well as the effective and efficient use of those resources towards identified and transparent strategic priorities. The measurement of government efficiency is still an open issue. The measures developed here, although plausible, cannot be treated as definitive. Future research may provide alternative 15 measures to test the robustness of our results. Further research is also needed to investigate the causal effects of fiscal policy on growth in cross-country regressions. We nevertheless believe that we have contributed to these important policy issues. 16 TABLE 1: Growth regressions using PSE: 62 countries, 1980-2000 Dep. Variable: growth rate govshare (1) (2) (3) (4) (5) (6) -0.054 [0.022]** (0.026)** - - - - - - govexp - - -0.106 [0.031]** (0.037)** 0.078 [0.030]** (0.039)** - - govshare*pse -0.052 [0.023]** (0.027)* - govexp*pse - - - -0.049 [0.023]** (0.025)* - -0.037 [0.028] (0.028) - Pse - - - Lgdp -2.108 [0.503]** (0.720)** 0.119 [0.039]** (0.045)** 0.015 [0.014] (0.020) -1.677 [0.813]** (1.012) 0.011 [0.004]** (0.005)** -0.227 [0.996] (0.706) -0.838 [0.720] (0.916) -0.873 [0.523]* (0.751) -3.601 [0.742]** (0.974)** 19.327 [4.878]** (6.248)** 0.919 [0.477]* (0.567) -2.392 [0.519]** (0.671)** 0.107 [0.039]** (0.047)** 0.016 [0.014] (0.019) -1.522 [0.810]* (1.003) 0.013 [0.004]** (0.005)** -0.204 [0.987] (0.690) -1.794 [0.869]** (1.101) -0.994 [0.522]* (0.739) -3.478 [0.738]** (1.011)** 20.898 [4.902]** (5.908)** -2.325 [0.501]** (0.636)** 0.118 [0.038]** (0.047)** 0.019 [0.014] (0.017) -1.275 [0.814] (0.985) 0.012 [0.004]** (0.005)** -0.051 [0.980] (0.683) -1.629 [0.773]** (0.883)* -0.956 [0.515]* (0.678) -3.329 [0.736]** (0.984)** 20.269 [6.609]** (5.685)** -1.736 [0.479]** (0.686)** 0.109 [0.039]** (0.049)** 0.025 [0.015] (0.021) -2.136 [0.810]** (1.041)** 0.011 [0.004]** (0.005)** -0.372 [0.992] (0.708) -1.156 [0.784] (0.944) -0.801 [0.519] (0.737) -3.280 [(0.756]** (1.191)** 16.342 [4.592]** (5.898)** 0.471 [0.576] (0.596) -1.879 [0.576]** (0.664)** 0.106 [0.040]** (0.049)** 0.024 [0.015] (0.021) -2.017 [0.824]** (1.064)* 0.011 [0.004]** (0.005)** -0.414 [0.995] (0.720) -1.449 [0.863] (1.085) -0.797 [0.520] (0.738) -3.289 [(0.757]** (1.170)** 16.826 [4.635]** (5.763)** -0.053 [0.025]** (0.028)* 0.009 [0.028] (0.027) - -1.804 [0.521]** (0.670)** 0.108 [0.040]** (0.050)** 0.025 [0.015] (0.021) -2.078 [0.830]** (1.067)* 0.011 [0.004]** (0.005)** -0.365 [0.995] (0.704) -1.207 [0.801] (0.978) -0.729 [0.522] (0.729) -3.240 [(0.768]** (1.211)** 16.801 [4.807]** (5.840)** 0.378 0.394 0.405 0.374 0.377 0.375 investment enrolment fertility openness Sub-Saharan Africa East Asia Latin America Transition Economies constant R2 Notes: 1. The estimation method is Least Squares. The sample consists of 62 countries, in 5-year periods over 1980-2000. There is a total of 159 observations. All regressions include time dummies. 2. Standard errors obtained under the assumption of spherical errors are shown in brackets below the estimated coefficients. Standard errors that are robust to arbitrary heteroskedasticity and arbitrary intra-country serial correlation are shown in parentheses. 3. An asterisk denotes significance at the 10% level and two asterisks at the 5% level. 17 TABLE 2: Growth regressions using PSE (controlling for tax): 62 countries, 1980-2000 Dep. Variable: (1) (2) (3) (4) (5) (6) growth rate -0.057 -0.056 -0.103 govshare govshare*pse [0.023]** (0.026)* - [0.023]** (0.026)** - govexp - - [0.032]** (0.041)** 0.072 [0.035]** (0.048) - govexp*pse - - pse - tax -0.039 [0.026] (0.030) 0.748 [0.594] (0.699) -0.015 [0.032] (0.038) 0.388 0.395 R2 - - - - -0.067 [0.037]* (0.038)* - -0.060 [0.037] (0.037) - - - -0.011 [0.029] (0.036) 0.024 [0.040] (0.050) 0.656 [0.609] (0.693) 0.039 [0.042] (0.055) -0.078 [0.042]* (0.049) 0.016 [0.029] (0.032) - 0.405 0.376 0.381 0.032 [0.042] (0.056) 0.377 Notes: 1. The estimation method is Least Squares. The sample consists of 62 countries, in 5-year periods over 1980-2000. There is a total of 159 observations. All regressions include time dummies, regional dummies and the control variables of the regressions in Table 1. 2. Standard errors obtained under the assumption of spherical errors are shown in brackets below the estimated coefficients. Standard errors that are robust to arbitrary heteroskedasticity and arbitrary intra-country serial correlation are shown in parentheses. 3. An asterisk denotes significance at the 10% level and two asterisks at the 5% level. TABLE 3: Growth regressions using PSE (controlling for grolag): 46 countries, 1985-2000 (1) (2) (3) (4) (5) (6) Dep. Variable: growth rate -0.071 -0.082 -0.165 govshare govshare*pse [0.033]** (0.038)* - [0.033]** (0.036)** - govexp - - [0.053]** (0.049)** 0.103 [0.0474]** (0.057) - govexp*pse - - pse - grolag 0.198 [0.104]* (0.139) 1.120 [0.679] (0.674) 0.182 [0.103]* (0.135) 0.458 0.475 R2 - - - - -0.054 [0.031]* (0.024)** - -0.047 [0.036] (0.031) - - - 0.131 [0.106] (0.124) 0.225 [0.104]** (0.132)* 0.282 [0.796] (0.837) 0.223 [0.105]** (0.131)* -0.055 [0.035] (0.030)* 0.001 [0.034] (0.039) - 0.225 [0.105]** (0.137) 0.487 0.448 0.448 0.448 Notes: 1. The estimation method is Least Squares. The sample consists of 46 countries, in 5-year periods over 1985-2000. There is a total of 98 observations. All regressions time dummies, regional dummies and the control variables of the regressions in Table 1. 2. Standard errors obtained under the assumption of spherical errors are shown in brackets below the estimated coefficients. Standard errors that are robust to arbitrary heteroskedasticity and arbitrary intra-country serial correlation are shown in parentheses. 3. An asterisk denotes significance at the 10% level and two asterisks at the 5% level. 18 TABLE 4: Growth regressions using PSE (controlling for tax and grolag): 46 countries, 19852000 (1) (2) (3) (4) (5) (6) Dep. Variable: growth rate -0.074 -0.079 -0.190 govshare govshare*pse [0.034]** (0.036)** - [0.033]** (0.036)** - govexp - - [0.058]** (0.061)** 0.140 [0.057]** (0.077)* - govexp*pse - - pse - tax -0.011 [0.033] (0.030) 0.198 [0.104]* (0.139) 1.585 [0.868] (1.026) -0.035 [0.041] (0.046) 0.176 [0.104]* (0.138) 0.458 0.479 grolag R2 - - - - -0.151 [0.049]** (0.056)** - -0.150 [0.049]** (0.056)** - - - 0.043 [0.039] (0.043) 0.107 [0.108] (0.125) 0.127 [0.051]** (0.063)** 0.208 [0.101]** (0.128) 1.373 [0.847] (0.940) 0.168 [0.057]** (0.069)** 0.191 [0.101]* (0.129) -0.207 [0.063]** (0.072)** 0.054 [0.034] (0.049 - 0.168 [0.059]** (0.077)** 0.213 [0.101]** (0.135) 0.495 0.485 0.501 0.497 Notes: 1. The estimation method is Least Squares. The sample consists of 46 countries, in 5-year periods over 1985-2000. There is a total of 98 observations. All regressions include time dummies, regional dummies and the control variables of the regressions in Table 1. 2. Standard errors obtained under the assumption of spherical errors are shown in brackets below the estimated coefficients. Standard errors that are robust to arbitrary heteroskedasticity and arbitrary intra-country serial correlation are shown in parentheses. 3. An asterisk denotes significance at the 10% level and two asterisks at the 5% level. 19 TABLE 5: Growth regressions using PSE (high income countries): 26 countries, 1980-2000 Dep. Variable: (1) (2) (3) (4) (5) (6) growth rate -0.065 -0.057 -0.173 govshare govshare*pse [0.028]** (0.038)* - [0.027]** (0.032)* - govexp - - [0.042]** (0.050)** 0.124 [0.038]** (0.040)** - govexp*pse - - pse - 0.599 R2 - - - - -0.079 [0.016]** (0.020)** - -0.072 [0.022]** (0.024)** - 1.161 [0.358]** (0.345)** - - 0.211 [0.465] (0.314) -0.080 [0.016]** (0.020)** 0.004 [0.023] (0.019) - 0.649 0.649 0.673 0.674 0.674 Notes: 1. The estimation method is Least Squares. The sample consists of 26 countries, in 5-year periods over 1980-2000. There is a total of 85 observations. All regressions include time dummies and the control variables of the regressions in Table 1. 2. Standard errors obtained under the assumption of spherical errors are shown in brackets below the estimated coefficients. Standard errors that are robust to arbitrary heteroskedasticity and arbitrary intracountry serial correlation are shown in parentheses. 3. An asterisk denotes significance at the 10% level and two asterisks at the 5% level. TABLE 6: Growth regressions using PSE (developing countries): 36 countries, 1980-2000 Dep. Variable: (1) (2) (3) (4) (5) (6) growth rate -0.022 -0.010 -0.069 govshare govshare*pse [0.035] (0.037) - [0.036] (0.038) - govexp - - [0.041]* (0.049) 0.078 [0.037]** (0.044)* - govexp*pse - - pse - 0.191 R2 - - - - -0.026 [0.036] (0.040) - 0.012 [0.047] (0.045) - 1.127 [0.848] (1.131) - - 1.362 [1.072] (1.327) -0.056 [0.039] (0.046) 0.090 [0.050]* (0.057) - 0.213 0.243 0.192 0.212 0.238 Notes: 1. The estimation method is Least Squares. The sample consists of 36 countries, in 5-year periods over 1980-2000. There is a total of 74 observations. All regressions include time dummies and the control variables of the regressions in Table 1. 2. Standard errors obtained under the assumption of spherical errors are shown in brackets below the estimated coefficients. Standard errors that are robust to arbitrary heteroskedasticity and arbitrary intracountry serial correlation are shown in parentheses. 3. An asterisk denotes significance at the 10% level and two asterisks at the 5% level. 20 TABLE 7: Growth regressions using PSE: 2SLS for 62 countries, 1980-2000 Dep. variable: growth rate govshare (1) (2) (3) (4) (5) (6) -0.160 [0.082]* (0.096)* - - - - - - govexp - - -0.208 [0.071]** (0.081)** 0.168 [0.053]** (0.054)** - - govshare*pse -0.126 [0.067]* (0.088) - govexp*pse - - - -0.096 [0.050]* (0.059) - 0.028 [0.094] (0.123) - pse - 4.826 [1.902]** (1.775)** χ (21) = 0.110 - - χ (21) = 0.029 2 χ (2) = 9.894 4.880 [2.848] (2.718) χ (21) = 4.144 -0.225 [0.106]** (0.107)** 0.314 [0.198] (0.205) - χ (21) = 2.902 (0.739) (0.865) (0.007) χ (23) = 42.17 (0.041) (0.088) Sargan overidentification test Cragg-Donald Underidentification Anderson canonical correlations First-stage F (gov) First-stage F (pse) First-stage F (gov*pse) 2 χ (2) = 9.513 (0.008) χ (23) = 20.78 χ 2 ( 2) = 15.12 χ 2 ( 2) = 19.88 χ 2 ( 2) = 8.96 χ (22) = 5.56 (0.000) (0.000) (0.000) (0.000) (0.011) (0.061) χ (23) = 19.53 χ (22 ) = 14.44 χ (22) = 18.73 χ (23) = 37.40 χ (22) = 8.72 χ (22) = 5.469 (0.000) (0.000) (0.000) (0.000) (0.012) (0.064) F (3,143) = F (3,143) = F (3,143) = F (3,143) = F (3,143) = F (3,143) = 6.23 - 6.23 12.64 - 12.64 F (3,143) = 6.23 - F (3,143) = 12.64 - - 4.91 - F (3,143) = - 4.91 - F (3,143) = 11.95 3.59 Notes: Notes: 1. The estimation method is 2SLS. The sample consists of 62 countries, in 5-year periods over 19802000. There is a total of 159 observations. All regressions include time dummies, regional dummies and the control variables of the regressions in Table 1. 2. Standard errors obtained under the assumption of spherical errors are shown in brackets below the estimated coefficients. Standard errors that are robust to arbitrary heteroskedasticity and arbitrary intra-country serial correlation are shown in parentheses. 3. An asterisk denotes significance at the 10% level and two asterisks at the 5% level. 4. The instruments used are: agedep, pop, surface. 5. The Sargan test is a test of overidentifying restrictions. Under the null, the test statistic is distributed as chi-squared in the number of over-identifying restrictions (the p-value is reported in parenthesis). 6. The Anderson (1984) canonical correlation is a likelihood-ratio test of whether the equation is identified. The Cragg and Donald (1993) test statistic is also a chi-squared test of whether the equation is identified. Under the null of underidentification, the statistics are distributed as chi-squared with degrees of freedom=(L-K+1) where L=number of instruments (included + excluded) and K is the number of regressors (the p-values are reported in parentheses). 7. The 1st stage F-statistic tests the hypothesis that the coefficients on all the excluded instruments are zero in the 1st stage regression of the endogenous regressor on all instruments (the p-value is reported in parenthesis). 21 TABLE 8: Growth regressions using PSE and TE: OLS for 51 countries, 1995-2000 Dep. variable: Growth rate govshare (1) (2) (3) (4) (5) (6) -0.225 [0.088]** (0.083)** - - - - - - govshare*te - -0.202 [0.055]** (0.055)** 0.166 [0.050]** (0.046)** - - govshare*pse -0.088 [0.049]* (0.048)* - - - - govexp - - 0.253 [0.137]* (0.130)* - govexp*pse - - - -0.138 [0.055]** (0.049)** - -0.187 [0.089]** (0.088)** - govexp*te - - - - -0.148 [0.047]** (0.056)** 0.047 [0.060] (0.045) - lgdp -1.143 [1.027] (1.344) 0.040 [0.076] (0.093) 0.015 [0.027] (0.034) 0.025 [1.679] (1.359) 0.012 [0.008] (0.010) -0.929 [1.803] (0.960) -1.903 [1.709] (1.814) -1.191 [0.980] (1.039) -3.617 [1.608]** (1.705)** 12.004 [9.003] (10.020) 0.291 -1.185 [0.943]* (1.094)* 0.040 [0.068] (0.089) 0.029 [0.025] (0.028) 0.502 [1.508] (1.260) 0.019 [0.008]** (0.009)** -0.535 [1.617] (1.000) -3.923 [1.645]** (1.893)** -1.470 [0.881] (0.903) -3.854 [1.440]** (1.590)** 16.093 [8.146]* (7.823) 0.446 -1.899 [1.079]* (1.275) 0.024 [0.074] (0.091) 0.014 [0.026] (0.033) -0.069 [1.632] (1.384) 0.012 [0.008] (0.011) -1.262 [1.761] (0.974) -2.774 [1.726] (1.934) -1.341 [0.956] (1.056) -3.738 [1.564]** (1.697)** 19.240 [9.588] (9.638) 0.347 -0.676 [0.908] (0.992) 0.035 [0.073] (0.095) 0.046 [0.030] (0.030) -1.388 [1.528] (1.105) 0.018 [0.009]** (0.010)* -0.348 [1.776] (0.877) -4.179 [2.020]** (1.768)** -1.316 [0.942] (0.958) -4.104 [1.551]** (2.063)** 8.906 [7.980] (7.423) 0.335 -0.940 [0.972] (1.078) 0.031 [0.074] (0.093) 0.039 [0.031] (0.032) -1.087 [1.582] (1.158) 0.017 [0.009]* (0.010)* -0.347 [1.785] (0.905) -3.895 [2.061]* (1.875)** -1.181 [0.962] (0.922) -3.871 [1.586]** (2.025)* 10.497 [8.257] (7.951)* 0.346 investment enrolment fertility openness Sub-Saharan Africa East Asia Latin America Transition Economies Constant R2 0.081 [0.113] (0.098) -0.994 [1.015] (1.091) 0.026 [0.075] (0.094) 0.043 [0.304] (0.031) -1.409 [1.538] (1.122) 0.018 [0.009]** (0.011)* -0.247 [1.793] (0.906) -4.122 [2.034]** (1.966)** -1.199 [0.976] (0.950) -3.768 [1.629]** (2.029)* 12.026 [8.986] (8.143)* 0.344 Notes: 1. The estimation method is OLS. The sample consists of 51 countries, 1985-2000. There are 51 observations. 2.-3. As in Table 1. 22 APPENDIX: construction of PSE We construct measures of public sector efficiency (PSE) for 64 countries, in four 5-year periods, over 1980-2000, as output-to-input ratios by working as in Afonso et al. (2005). Afonso et al. have focused on OECD countries, where the available data cover both government performance and the associated public expenditure. Although we have tried to follow Afonso et al. in the choice of policy areas and variables used, the construction of such a rich PSE for a broader group of countries runs into data limitations, especially when looking for decomposed public expenditure data. Thus, some deviations from the variables used by Afonso et al. are inevitable. Nevertheless, the variables used here are the same in spirit. In the policy area of education, the PSP can be measured by the variable Secondary School Enrollment, while the associated PEX is the average of the variable Public Spending in Education as a percentage of GDP (both variables are available from the World Development Indicators, WDI), where we use the end of period values (or the closest to the end available) of Secondary School Enrollment.19 The resulting PSE is then a measure of government efficiency in the policy area of education. In the policy area of administration, the PSP is measured by the end of period values of the variables Corruption in Government and Bureaucratic Quality (both obtained from the IRIS-3 dataset)20 with higher scores denoting better outcomes, while the PSE is obtained as in Afonso et al. (2005) by dividing this variable by the average public spending on goods and services (available from WDI). In the policy area of infrastructure, the PSP is measured by the average of Diesel Locomotives in Use as a percentage of total locomotives, and the average of the inverse of Electric Power Transmission and Distribution Losses (both variables are available from WDI). These measures have also been used by Tanzi and Davoodi (1998) as indicators of the quality of infrastructure (see also Angelopoulos and Philippopoulos, 2007). A problem here is that the relevant PEX for infrastructure quality, which has been used by Afonso et al. for the OECD countries, is not available for the larger group of countries we work with. We 19 Afonso et al. (2005) use the same PEX, but they also include a measure of the quality of education when they construct the PSP. 20 Afonso et al. (2005) have used very similar variables (measures of corruption, red tape, quality of judiciary and shadow economy). We prefer the IRIS-3 indexes because they are available for the counties and time periods we work with. 23 therefore choose to use Total Government Expenditure (as a percentage of GDP) for PEX (this is also available from WDI), again averaged over the 5-year period. Finally, in the policy area of stabilization, the PSP is measured by the average of the inverse of the variables Inflation Rate and Unemployment Rate (obtained from WDI), while the relevant PSE is calculated by dividing this PSP by Total Government Expenditure (as a percentage of GDP), averaged over the 5-year period. Afonso et al. also use total government spending as a measure of public sector expenditures that are associated with stabilization and economic performance indicators, such as inflation and unemployment. 24 Table A.1: Public Sector Efficiency (PSE) in 64 countries, 1980-2000 Country Period Algeria 19901995 19952000 19952000 19801985 19851990 19901995 19952000 19851990 19901995 19952000 19801985 19851990 19901995 19851990 19901995 19901995 19952000 19801985 19851990 19901995 19952000 19851990 19901995 19952000 19901995 19851990 19901995 Algeria Argentina Australia Australia Australia Australia Austria Austria Austria Belgium Belgium Belgium Bolivia Brazil Bulgaria Bulgaria Canada Canada Canada Canada Chile Chile Chile Colombia Costa Rica Costa Rica PSPAdmin 0.557 PSPEduc 0.793 PSPInfrast 0.797 PSPStabil 0.258 PSEAdmin 0.466 PSEEduc 0.648 PSEInfrast 0.749 PSEStabil 0.242 PSEaverage 0.526 Growth effect -0.065 0.586 0.803 0.452 0.233 0.522 0.73 0.459 0.237 0.487 -0.068 0.592 1.013 0.528 2.092 1.763 1.366 1.043 4.136 2.077 - 1.266 1.108 0.972 0.78 2.374 1.048 1.357 1.089 1.467 0.008 1.309 1.169 1.015 0.552 2.053 1.079 1.296 0.705 1.283 -0.006 1.224 1.226 0.997 1.316 1.754 1.122 1.243 1.64 1.44 0.006 1.306 1.222 1.26 1.149 1.954 1.218 1.603 1.462 1.559 0.016 1.309 1.354 1.214 1.409 1.257 1.074 0.981 1.139 1.113 -0.019 1.224 1.255 1.043 1.701 1.196 1.083 0.831 1.355 1.116 -0.019 1.257 1.212 1.247 1.709 1.197 1.042 0.956 1.31 1.127 -0.018 1.384 1.254 1.332 1.22 1.181 1.069 0.814 0.745 0.952 -0.032 1.285 1.303 1.279 1.108 1.243 1.12 0.791 0.686 0.96 -0.031 1.156 1.24 1.155 1.154 1.242 1.114 0.735 0.734 0.956 -0.031 0.359 0.437 0.691 0.23 0.404 0.935 1.584 0.527 0.863 -0.039 0.778 0.275 0.42 0.553 1.599 0.78 0.41 0.539 0.832 -0.041 0.78 1.056 0.724 0.24 0.565 0.934 0.499 0.166 0.541 -0.064 0.836 1.111 0.618 0.239 0.792 1.606 0.502 0.194 0.773 -0.046 1.384 1.247 1.107 0.728 2.715 0.921 1.559 1.026 1.555 0.015 1.429 1.318 1.105 0.686 2.712 0.924 1.432 0.889 1.489 0.010 1.336 1.284 1.037 1.379 2.457 0.873 1.197 1.592 1.53 0.013 1.428 1.288 1.187 1.266 3.654 1.13 1.631 1.741 2.039 0.053 0.715 0.811 0.638 0.374 0.936 1.08 0.783 0.459 0.815 -0.042 0.668 0.773 0.575 0.756 1.08 1.363 0.856 1.126 1.106 -0.020 0.952 0.966 0.962 0.748 1.5 1.411 1.377 1.071 1.34 -0.002 0.778 0.643 0.302 0.451 2.23 1.09 0.719 1.075 1.278 -0.006 0.956 0.531 0.917 0.524 0.757 0.568 1.145 0.655 0.781 -0.045 0.892 0.575 0.698 0.832 0.751 0.743 1.001 1.193 0.922 -0.034 25 Costa Rica Cyprus Cyprus Cyprus Cyprus Czech Rep. Czech Rep. Denmark Denmark Denmark Denmark Dominican Rep. Dominican Rep. Egypt Egypt El Salvador Finland Finland Finland France France France France Germany Germany Greece Greece Greece Greece 19952000 19801985 19851990 19901995 19952000 19901995 19952000 19801985 19851990 19901995 19952000 19901995 19952000 19901995 19952000 19952000 19851990 19901995 19952000 19801985 19851990 19901995 19952000 19901995 19952000 19801985 19851990 19901995 19952000 0.959 0.563 0.868 0.672 0.892 0.613 1.216 0.941 0.915 -0.035 0.805 1.116 1.248 1.251 0.565 1.481 1.326 1.328 1.175 -0.014 0.832 1.182 1.247 1.306 0.663 1.472 1.289 1.35 1.193 -0.013 1.113 1.204 1.139 1.845 0.897 1.438 1.084 1.757 1.294 -0.005 1.19 1.002 1.491 1.619 0.941 0.977 1.314 1.427 1.165 -0.015 0.89 1.228 0.995 1.244 1.043 1.148 0.827 1.033 1.013 -0.027 0.952 1.089 1.086 0.777 1.806 1.119 0.954 0.683 1.14 -0.017 1.384 1.179 0.952 0.705 1.579 0.829 0.758 0.562 0.932 -0.033 1.429 1.291 1.073 0.756 1.755 0.839 0.88 0.62 1.023 -0.026 1.336 1.236 1.093 1.791 1.58 0.77 0.837 1.371 1.139 -0.017 1.428 1.23 1.62 1.234 1.801 0.758 1.31 0.998 1.217 -0.011 0.668 0.405 0.239 0.29 1.302 1.169 0.528 0.639 0.909 -0.035 0.836 0.725 0.32 0.397 1.172 1.783 0.616 0.764 1.084 -0.021 0.757 0.919 0.8 0.521 0.61 1.011 0.723 0.471 0.704 -0.051 0.592 0.927 0.701 0.563 0.465 0.978 0.677 0.543 0.666 -0.054 0.647 0.515 0.648 0.694 0.539 1.089 1.287 1.378 1.073 -0.022 1.429 1.383 1.257 0.868 2.288 1.186 1.318 0.911 1.426 0.005 1.336 1.305 1.248 1.22 1.786 0.909 0.99 0.967 1.163 -0.015 1.312 1.304 1.773 1.598 1.911 0.905 1.485 1.338 1.41 0.004 1.384 1.156 1.163 0.667 1.092 1.067 0.91 0.521 0.898 -0.036 1.309 1.275 1.168 0.742 1.108 1.056 0.859 0.545 0.892 -0.036 1.112 1.331 1.113 1.477 0.953 1.117 0.782 1.038 0.972 -0.030 1.068 1.295 1.289 1.438 0.945 1.075 0.869 0.969 0.965 -0.031 1.336 1.255 1.592 1.33 1.356 1.272 1.588 1.326 1.386 0.002 1.306 1.207 1.547 1.467 1.234 1.272 1.445 1.371 1.331 -0.002 0.739 1.142 0.86 0.559 0.477 2.826 0.77 0.5 1.143 -0.017 0.98 1.229 0.875 0.41 0.555 2.431 0.64 0.3 0.981 -0.029 1.002 1.215 0.729 0.555 0.839 2.303 0.626 0.477 1.061 -0.023 1.074 1.188 1.172 0.577 1.058 1.971 1.144 0.563 1.184 -0.014 26 Hungary Hungary Hungary Hungary Iceland Iceland Iceland India Indonesia Indonesia Iran Ireland Ireland Ireland Ireland Israel Italy Jamaica Jamaica Jamaica Jamaica Japan Japan Japan Jordan Jordan Jordan Korea, Rep Korea, Rep 19801985 19851990 19901995 19952000 19851990 19901995 19952000 19952000 19851990 19901995 19901995 19801985 19851990 19901995 19952000 19952000 19952000 19801985 19851990 19901995 19952000 19801985 19851990 19901995 19851990 19901995 19952000 19801985 19851990 0.922 0.987 0.79 1.289 0.861 0.98 0.474 0.773 0.772 -0.046 1.049 1.111 0.777 0.342 0.964 0.93 0.438 0.193 0.631 -0.057 1.102 1.212 0.766 0.472 0.963 0.905 0.439 0.27 0.644 -0.056 1.19 1.166 0.72 0.456 1.545 1.224 0.498 0.315 0.895 -0.036 1.429 1.268 0.83 2.745 0.842 1.201 0.862 2.854 1.44 0.006 1.336 1.228 0.904 1.258 0.787 1.117 0.858 1.193 0.989 -0.029 1.428 1.174 1.228 1.638 0.881 1.003 1.238 1.652 1.193 -0.013 0.83 0.534 0.779 0.324 2.335 0.87 1.602 0.667 1.369 0.001 0.154 0.563 0.679 0.536 0.315 3.07 1.051 0.829 1.316 -0.003 0.668 0.597 0.721 0.703 1.487 2.196 1.3 1.268 1.563 0.016 0.89 1.003 0.621 0.293 0.744 1.038 0.898 0.424 0.776 -0.045 1.153 1.151 0.794 0.459 1.281 0.998 0.552 0.319 0.788 -0.045 1.191 1.188 0.934 0.632 1.429 0.986 0.656 0.443 0.878 -0.037 1.224 1.219 0.803 1.335 1.597 1.11 0.639 1.062 1.102 -0.020 1.183 1.058 1.013 1.083 1.878 1.079 0.905 0.968 1.207 -0.012 1.153 1.163 1.616 0.604 0.702 0.748 1.058 0.395 0.726 -0.049 1.068 1.214 1.089 0.771 1.269 1.272 0.738 0.522 0.95 -0.032 0.574 0.814 0.551 0.342 0.381 0.646 0.463 0.287 0.444 -0.071 0.594 0.945 0.408 0.24 0.282 0.851 0.334 0.196 0.416 -0.074 0.778 0.9 0.4 0.265 0.593 0.974 0.587 0.389 0.636 -0.056 0.83 1.085 0.819 0.309 0.451 0.905 0.721 0.272 0.587 -0.060 1.266 1.349 1.479 1.993 5.247 1.232 2.676 3.606 3.19 0.143 1.309 1.439 1.475 2.489 5.594 1.352 2.816 4.753 3.629 0.177 1.224 1.39 1.371 2.784 5.47 1.816 2.128 4.322 3.434 0.162 0.715 0.488 0.627 0.523 0.346 0.46 0.56 0.467 0.458 -0.070 0.89 0.584 0.85 0.859 0.455 0.368 0.807 0.816 0.611 -0.058 0.952 0.818 0.992 0.682 0.465 0.551 0.956 0.657 0.658 -0.055 0.687 1.191 1.193 0.811 0.89 1.54 2.266 1.539 1.559 0.016 0.711 1.275 1.263 1.152 1.131 1.542 2.542 2.319 1.883 0.041 27 Korea, Rep Korea, Rep Lebanon Luxembourg Luxembourg Luxembourg Luxembourg Malaysia Mexico Mexico Mexico Mexico Namibia Namibia Netherlands Netherlands Netherlands Netherlands New Zealand New Zealand New Zealand New Zealand Nicaragua Norway Norway Norway Norway Panama Panama 19901995 19952000 19952000 19801985 19851990 19901995 19952000 19952000 19801985 19851990 19901995 19952000 19901995 19952000 19801985 19851990 19901995 19952000 19801985 19851990 19901995 19952000 19952000 19801985 19851990 19901995 19952000 19801985 19851990 1.113 1.353 1.176 1.645 1.96 1.653 2.225 3.112 2.237 0.069 1.068 1.333 1.512 1.165 2.254 1.757 2.753 2.121 2.221 0.067 0.354 1.039 0.548 0.771 0.311 2.071 0.46 0.648 0.872 -0.038 1.372 0.936 1.136 1.704 1.418 0.8 0.867 1.3 1.096 -0.020 1.429 0.953 1.044 2.003 1.585 0.958 0.871 1.671 1.271 -0.007 1.336 0.944 0.418 2.34 1.426 1.396 0.331 1.85 1.251 -0.008 1.342 0.929 0.299 2.274 1.441 1.12 0.238 1.81 1.152 -0.016 0.952 1.277 0.933 1.559 1.012 1.357 1.378 2.301 1.512 0.012 0.624 0.649 0.755 0.165 1.003 0.741 1.165 0.254 0.791 -0.044 0.715 0.666 0.731 0.94 1.325 0.888 0.91 1.17 1.074 -0.022 0.668 0.722 0.725 1.077 1.319 0.811 1.467 2.18 1.444 0.007 0.592 0.771 0.726 0.964 1.51 0.764 1.455 1.931 1.415 0.004 1.113 0.504 1.128 0.403 0.417 0.269 0.991 0.354 0.508 -0.066 1.068 0.432 1.115 0.343 0.425 0.254 0.959 0.295 0.483 -0.068 1.384 1.254 1.501 1.114 1.65 0.907 0.912 0.677 1.037 -0.025 1.429 1.243 1.47 2.888 1.762 0.901 0.873 1.714 1.313 -0.004 1.336 1.279 1.291 1.528 1.744 1.115 0.799 0.945 1.15 -0.016 1.428 1.273 1.561 1.312 1.947 1.271 1.037 0.871 1.281 -0.006 1.384 1.185 0.86 0.744 1.235 1.165 0.7 0.605 0.926 -0.034 1.429 1.264 0.924 0.599 1.269 1.042 0.684 0.443 0.86 -0.039 1.336 1.256 0.875 1.583 0.893 0.885 0.704 1.273 0.939 -0.033 1.306 1.241 0.71 1.355 0.784 0.872 0.677 1.293 0.907 -0.035 0.721 0.449 0.315 0.355 0.588 0.626 0.271 0.305 0.447 -0.071 1.293 1.217 0.978 1.207 1.716 0.993 0.92 1.135 1.191 -0.013 1.312 1.304 1.017 1.06 1.763 0.916 0.853 0.889 1.105 -0.020 1.336 1.346 0.998 1.649 1.537 0.833 0.738 1.221 1.082 -0.022 1.306 1.325 1.129 1.488 1.698 0.849 0.949 1.251 1.187 -0.013 0.348 0.677 0.525 1.003 0.205 0.787 0.541 1.034 0.642 -0.056 0.359 0.754 0.363 3.949 0.209 0.712 0.419 4.555 1.474 0.009 28 Paraguay Paraguay Peru Peru Peru Philippines Philippines Philippines Poland Portugal Portugal Portugal Romania Romania South Africa South Africa South Africa Spain Sweden Sweden Sweden Switzerland Switzerland Switzerland Syria Thailand Trinidad & Tobago Trinidad & Tobago Tunisia 19851990 19901995 19801985 19901995 19952000 19801985 19851990 19901995 19901995 19851990 19901995 19952000 19901995 19952000 19851990 19901995 19952000 19952000 19851990 19901995 19952000 19801985 19901995 19952000 19801985 19952000 19901995 19952000 19851990 0.117 0.384 3.504 0.489 0.228 1.539 12.52 1.747 4.008 - 0.556 0.503 10.84 0.698 0.822 1.043 28.17 1.812 7.96 - 0.579 0.693 0.717 0.11 0.633 1.202 1.252 0.192 0.82 -0.042 0.557 0.749 0.356 0.356 0.97 1.137 0.603 0.603 0.828 -0.041 0.598 0.845 0.587 0.606 0.84 1.317 0.993 1.025 1.044 -0.025 0.192 0.701 1.417 0.604 0.259 1.952 3.714 1.583 1.877 0.040 0.359 0.842 0.449 0.485 0.467 1.768 0.946 1.021 1.05 -0.024 0.557 0.829 0.449 0.622 0.649 1.459 0.733 1.015 0.964 -0.031 1.058 1.191 0.731 0.271 0.93 1.127 0.547 0.203 0.702 -0.051 0.98 1.038 0.84 0.472 0.825 1.246 0.709 0.398 0.794 -0.044 0.946 1.095 0.849 0.926 0.577 1.058 0.643 0.701 0.744 -0.048 1.074 1.205 1.015 1.033 0.661 1.094 0.797 0.811 0.841 -0.040 0.669 1.029 0.736 0.419 0.58 1.495 0.663 0.377 0.779 -0.045 0.598 1.038 0.845 0.512 0.546 1.282 0.805 0.487 0.78 -0.045 1.309 0.751 1.302 0.234 0.805 0.626 1.361 0.244 0.759 -0.047 1.069 0.812 1.02 0.403 0.673 0.615 1.015 0.401 0.676 -0.053 1.074 0.768 1.17 0.352 1.381 0.605 1.186 0.357 0.882 -0.037 1.19 1.259 1.02 0.667 2.132 1.337 0.929 0.607 1.251 -0.008 1.429 1.267 1.105 1.299 2.461 0.824 0.863 1.015 1.291 -0.005 1.336 1.388 0.99 1.031 1.979 0.869 0.694 0.723 1.066 -0.023 1.428 1.368 1.234 2.248 2.129 0.877 0.894 1.628 1.382 0.002 1.384 1.119 1.239 4.532 2.379 1.128 2.089 7.642 3.31 0.152 1.336 1.178 1.099 1.95 1.672 1.009 1.346 2.389 1.604 0.019 1.306 1.142 1.41 2.735 1.575 1.006 1.59 3.083 1.813 0.035 0.461 0.72 0.767 0.8 0.224 0.642 0.577 0.602 0.511 -0.066 0.83 0.759 0.999 2.025 0.849 0.837 1.551 3.145 1.596 0.018 0.668 0.917 0.631 0.525 0.431 1.138 0.7 0.583 0.713 -0.050 0.714 0.996 0.984 0.557 0.488 1.51 1.089 0.616 0.926 -0.034 0.715 0.632 0.647 0.502 0.567 0.498 0.567 0.441 0.518 -0.066 29 Tunisia Turkey Turkey Turkey Turkey United Kingdom United Kingdom United Kingdom United Kingdom Uruguay USA USA USA USA Venezuela Venezuela Venezuela Venezuela Yemen 19952000 19801985 19851990 19901995 19952000 19801985 19851990 19901995 19952000 19952000 19801985 19851990 19901995 19952000 19801985 19851990 19901995 19952000 19952000 0.714 0.754 0.863 0.739 0.551 0.536 0.836 0.716 0.66 -0.055 0.692 0.51 0.728 0.258 0.798 1.076 1.169 0.413 0.864 -0.039 0.594 0.615 0.731 0.28 0.877 1.895 1.327 0.509 1.152 -0.016 0.846 0.723 0.728 0.419 0.772 1.201 1.058 0.609 0.91 -0.035 0.707 0.705 0.705 0.518 0.702 1.544 0.736 0.541 0.881 -0.037 1.384 1.132 0.963 0.673 1.074 1.06 0.782 0.547 0.866 -0.038 1.309 1.175 0.936 0.582 1.117 1.093 0.798 0.497 0.876 -0.038 1.224 1.29 0.894 0.991 0.995 1.17 0.677 0.751 0.898 -0.036 1.232 1.288 1.048 0.975 1.09 1.278 0.844 0.785 0.999 -0.028 0.83 0.901 0.472 0.382 0.89 1.662 0.484 0.392 0.857 0.001 1.266 1.291 1.044 0.851 1.778 1.001 1.494 1.218 1.373 0.006 1.309 1.275 1.161 0.878 1.821 1.093 1.612 1.219 1.436 0.014 1.224 1.268 0.962 1.287 1.955 1.154 1.307 1.748 1.541 0.045 1.183 1.24 1.288 1.282 2.637 1.218 1.954 1.944 1.938 -0.039 0.692 0.224 0.617 0.585 0.684 0.213 0.92 0.872 0.672 -0.054 0.715 0.276 0.462 0.283 0.996 0.265 0.72 0.44 0.606 -0.059 0.668 0.271 0.341 0.424 1.042 0.287 0.539 0.669 0.634 -0.057 0.714 0.307 0.603 0.305 1.534 0.302 0.978 0.494 0.827 -0.041 0.714 0.484 0.332 0.115 0.479 0.446 0.351 0.122 0.35 -0.079 Key: PSP: Public Sector Performance PSE: Public Sector Efficiency Admin: Administration Educ: Education Infrast: Infrastructure Stabil: Stabilization * See footnote 6 30 Table A.2: Technical Efficiency (TE) of public spending in 52 countries, 1995-2000 Growth effect -0.133 - Country Algeria Argentina (see fn 6) Australia Austria Bulgaria Canada Chile TE 0.363355 0.830471 0.875214 0.867984 0.466058 0.910333 0.672651 Costa Rica Cyprus Czech Republic Denmark Dominican Rep. Egypt El Salvador Finland France Germany Greece Hungary Iceland India Ireland Israel Italy Jamaica Jordan 0.56748 0.872052 0.653997 0.885186 0.453656 0.47918 0.500757 0.928959 0.802157 0.903573 0.686213 0.565051 0.906244 0.496827 0.727425 0.713245 0.657281 0.513118 0.589832 -0.081 -0.004 -0.060 -0.001 -0.110 -0.104 -0.098 0.010 -0.022 0.004 -0.051 -0.082 0.004 -0.099 -0.041 -0.045 -0.059 -0.095 -0.076 Korea, Rep Lebanon Luxembourg Malaysia Mexico Namibia 0.927815 0.454145 0.791004 0.866256 0.608441 0.496673 0.010 -0.110 -0.025 -0.006 -0.071 -0.099 Netherlands 0.866055 -0.006 -0.004 -0.005 -0.107 0.005 -0.055 31 Country New Zealand Nicaragua Norway Peru Portugal Romania South Africa TE 0.7823 0.312448 0.858392 0.509566 0.706914 0.513156 0.582557 Growth effect -0.027 -0.146 -0.008 -0.096 -0.046 -0.095 -0.078 Spain Sweden Switzerland Thailand Trinidad & Tobago Tunisia Turkey United Kingdom Uruguay USA Venezuela Yemen 0.697047 0.934942 0.965281 0.857647 0.573398 0.527729 0.461675 0.745181 0.451386 0.903279 0.372336 0.292314 -0.049 0.012 0.019 -0.008 -0.080 -0.091 -0.108 -0.036 0.004 -0.111 -0.131 -0.151 References Afonso A., L. Schuknecht and V. Tanzi (2005): Public sector efficiency: An international comparison, Public Choice, 123, 321-347. Afonso A., L. Schuknecht and V. Tanzi (2006): Public sector efficiency: Evidence for new EU member states and emerging markets, ECB Working Paper, no. 581. Agell J., H. Ohlsson and P. S. Thoursie (2006): Growth effects of government expenditure and taxation in rich countries: A comment, European Economic Review, 50, 211-218. Anderson T.W. (1984): Introduction to Multivariate Statistical Analysis, 2d edition, New York: John Wiley & Sons. Angelopoulos K., G. Economides and P. Kammas (2007): Tax-spending policies and economic growth: Theoretical predictions and evidence from the OECD, European Journal of Political Economy, 23, 885-902. Angelopoulos K. and A. Philippopoulos (2007): The growth effects of fiscal policy in Greece 19602000, Public Choice, 131, 157-175. Baum C. F., M. E. Schaffer and S. Stillman (2006): Stata module to Extended instrumental variables/2SLS, GMM and AC/HAC, LIML and k-class regression, http://ideas.repec.org/c/boc/bocode/s425401.html Barro R. (1990): Government spending in a simple model of economic growth, Journal of Political Economy, 98, S103-S125. Barro R. and X. Sala-i-Martin (2004): Economic Growth, Second edition, The MIT Press, Cambridge, Mass. Cragg J.G. and S.G. Donald (1993): Testing identfiability and specification in instrumental variables models, Econometric Theory, 9, 222-240. De Haan J., S. Lundstrom and J.-E. Sturm (2006): Market-oriented institutions and policies and economic growth: a critical survey, Journal of Economic Surveys, 20, 157-191. Devarajan S., Swaroop, V., Zoo H. (1996): The composition of public expenditure and economic growth, Journal of Monetary Economics, 37, 313-344. Dutt P. and D. Mitra (2002): Endogenous trade policy through majority voting: An empirical investigation, Journal of International Economics, 58, 107-133. Folster S. and M. Henrekson (2001): Growth effects of government expenditure and taxation in rich countries, European Economic Review, 45, 1501-20. Gemmel N. and R. Kneller (2001): The impact of fiscal policy on long-run growth, European Economy, 1, 98-129. Greene W. H. (2005): Efficiency of public spending in developing countries: A stochastic frontier approach, May 2005, mimeo. 32 Gwartney J., R. Holcombe and R. Lawson (1998): The scope of government and the wealth of nations, Cato Journal, 18, 163-190. Heston A., R. Summers and B. Aten (2002): Penn World Table Version 6.1, Center for International Comparisons at the University of Pennsylvania (CICUP). Hillman A. (2003): Public Finance and Public Policy: Responsibilities and Limitations of Government, Cambridge University Press, Cambridge. Kneller R., M. Bleaney and N. Gemmel (1999): Public policy and the government budget constraint, Journal of Public Economics, 74, 171-190. Kumbhakar S. C. and C. A. K. Lovell (2000): Stochastic Frontier Analysis, Cambridge University Press, Cambridge Levine R. and D. Renelt (1992): A sensitivity analysis of cross-country growth regressions, American Economic Review, 82, 942-63. Miller S. and F. Russek (1997): Fiscal structures and economic growth: international evidence, Economic Inquiry, XXXV, 603-613. Mueller D. (2003): Public Choice III, Cambridge University Press, Cambridge. Persson T. and G. Tabellini (2003): The Economic Effects of Constitutions. The MIT Press, Cambridge, Mass. Tanzi V. and H. R. Davoodi (1998): Corruption, public investment and growth, in The Welfare State, Public Investment and Growth, edited by H. Shibata and T. Ihori, Springer-Verlag, Tokyo. Tanzi V. and L. Schuknecht (2000): Public Spending in the 20th Century: A Global Perspective, Cambridge University Press, Cambridge. Tanzi V. and H. Zee (1997): Fiscal policy and long-run growth, IMF Staff Papers, 44, 179-209. Wooldridge J. W. (2002): Econometric Analysis of Cross Section and Panel Data, The MIT Press, Cambridge, Mass. 33 Economia Aplicada, v. 14, n. 1, 2010, pp. 51-66 EFICIÊNCIA DO SETOR HOSPITALAR NOS MUNICÍPIOS PAULISTAS Igor Viveiros Souza* Marislei Nishijima† Fabiana Rocha‡ Resumo O objetivo deste artigo é avaliar o grau de eficiência produtiva do setor público hospitalar em 366 municípios do estado de São Paulo entre os anos de 1998 e 2003. Para tanto é utilizado o método de fronteira estocástica de produção. O modelo estimado com a forma flexível de Fourier usa o complemento da taxa de mortalidade hospitalar como produto e o gasto público com profissionais e o número de leitos por município como insumos. Os resultados sugerem que os municípios mais eficientes são aqueles que contratam mais leitos de hospitais privados, que realizam o maior número de internações (efeito de economia de escala), que possuem menor população (efeito congestionamento) e que apresentam menor tempo médio de internação. Palavras-chave: hospitais, eficiência, fronteira estocástica, municípios. Abstract The purpose of this article is to assess the degree of productive efficiency of public sector hospitals in 366 municipalities of the state in São Paulo between the years 1998 and 2003. In2003. In order to do so we use the stochastic frontier of production approach. The model is estimated using the flexible form of Fourier and uses the complement of the rate of hospital mortality as output and public spending with professionals and the number of beds per municipality as inputs. The results suggest that the most efficient municipalities are those that hire more beds in private hospitals, the ones which perform the highest number of admissions (economies of scale effect), the ones with smaller population (congestion effect), and the ones which show lower average time of hospitalization. Keywords: hospitals, efficiency, stochastic frontier, municipalities. JEL classification: C23, H51, I10 * Fundação João Pinheiro † EACH/USP ‡ FEA/USP Recebido em 14 de dezembro de 2007 . Aceito em 8 de março de 2010. 52 1 Souza, Nishijima e Rocha Economia Aplicada, v.14, n.1 Introdução O setor de saúde brasileiro é um dos grandes demandantes de recursos públicos. Para se ter uma idéia o Governo despendeu uma média anual de 3,23% de seu PIB em saúde entre os anos de 1999 e 2006, conforme estatísticas do Banco Mundial. Em 2003, somente o estado de São Paulo teve um orçamento para o setor de saúde, considerando as três esferas de Governo, de aproximadamente 13,4 bilhões de reais, o que implica um gasto per capita em torno de R$ 346,211 . Apesar da Constituição de 1988 ter estabelecido como modelo a descentralização na provisão e no financiameanto das ações de saúde, por meio do Sistema Único de Saúde (SUS), visando maior eficiência no uso dos recursos, poucos trabalhos empíricos foram realizados para avaliar o desempenho dos municípios na gestão de tais recursos. Deste modo, este estudo busca contribuir para a redução desta lacuna através da investigação da eficiência produtiva dos gastos públicos no setor hospitalar dos municípios do estado de São Paulo. O período de análise vai de 1998 a 2003 e abrange somente os municípios de São Paulo por questão de disponibilidade e confiabilidade dos dados. São avaliados 366 municípios o que corresponde a 89% do total de municípios que possuem rede hospitalar pública ou privada contratada. Somente para esses tem-se informação disponível para todas as variáveis invetisgadas em 2003. Além disso, em 2003 a população desses 366 municípios representava 94% da população total do Estado. A literatura sobre fronteira estocástica aplicada especificamente a hospitais em geral trata de estimar fronteiras de eficiência de custos, em que são utilizados grandes volumes de informação sobre preços e quantidades de insumos e produtos, além de informações específicas sobre os pacientes para estimar relações microeconômicas. Bradford & Kleit (2001) utilizam dados de um hospital para comparar a eficiência produtiva de dois tipos de tratamentos alternativos em pacientes cardíacos. Smet (2007) compara a eficiência entre hospitais na Bélgica, usando uma função multi-produto para avaliar seus desempenhos na presença de uma demanda estocástica (dada a sua grande variabilidade). Bernet et al. (2008) verificam como o acesso ao financiamento de investimentos produtivos de hospitais (equipamentos e instalações) altera o grau de eficiência produtiva de uma amostra de hospitais. Finalmente, Brown III & Pagán (2006) avaliam como o sistema de saúde norte americano Managed Care impacta sobre o custo de uma amostra de hospitais em diferentes localidades. Considerando-se, contudo, a ausência de informações detalhadas sobre insumos utilizados pelos hospitais públicos brasileiros a alternativa adotada foi estimar modelos de fronteira estocástica de produção, seguindo a literatura de eficiência produtiva de gastos públicos baseada em dados agregados, usando informações hospitalares por município. A literatura sobre eficiência de gastos públicos é farta em estudos que comparam a eficiência relativa de diferentes países na provisão de saúde (Evans et al. (2000), Gupta & Verhoeven (2001),Jayasuriya & Wodon (2002), Greene (2003a), Afonso & St. Aubyn (2004), Herrera & Pang (n.d.), Herrera & Worldwide (2005)). Afonso et al. (2003), por sua vez, constrõem um indi1 Valores de 2003, informação retirada do DATASUS. Eficiência do setor hospitalar 53 cador de desempenho para o setor público como um todo para 23 países da OCDE. Esse indicador é composto por 7 sub-indicadores que captam a qualidade das funções administrativas, os resultados em educação, os resultados em saúde, a qualidade da infra-estrutura, o grau de desigualdade, a estabilidade econômica e o desempenho econômico. Especificamente para o Brasil, Sousa et al. (2005) que buscam calcular scores de eficiência sobre os serviços gerais dos municípios brasileiros e de Marinho (2001), que analisa a eficiência da prestação de serviços de saúde em municípios do Rio de Janeiro. A fim de estimar a eficiência dos hospitais será utilizado o método de Fronteira Estocástica (FE), mais especificamente o modelo implementado por Battese & Coelli (1995). Esse método permite a decomposição das variações do desempenho dos municípios na provisão de serviços hospitalares em relação à fronteira em variações na eficiência técnica e em choques puramente aleatórios. Além disso, permite a investigação empírica objetiva de prováveis variáveis explicativas do termo ineficiência. O uso desse método se constitui num diferencial analítico em relação aos trabalhos anteriormente listados que majoritariamente utilizam metodologias não paramétricas, quais sejam, o Free Disposable Hull (FDH) e a Data Envelopment Analysis (DEA).2 Além desta introdução, o artigo apresenta três seções. Na segunda seção, apresenta-se a base teórica e quantitativa sobre a qual este estudo repousa. Na terceira seção discute-se a base de dados e os principais resultados obtidos das estimativas. Por fim, na quarta seção resumem-se as principais conclusões. 2 Metodologia Para medir eficiência produtiva a teoria utiliza uma medida de distância entre o ponto de operação da unidade tomadora de decisão (neste caso, o município) e a fronteira tecnológica, medida que pode ser dividida em eficiência técnica e alocativa.3 O conceito de eficiência técnica, que remonta a Debreu (1951), foi consolidado por Farrel (1957), que, adicionalmente desenvolveu o conceito de eficiência alocativa. O conceito de eficiência técnica diz respeito estritamente às relações entre as quantidades produzidas de produto e as quantidades de fatores utilizados na produção. Quando a quantidade produzida por uma firma, dada uma combinação de fatores, fica aquém do máximo possível de ser atingido com aquela combinação tem-se a caracterização da ineficiência técnica. A eficiência alocativa - que se refere à escolha ótima da proporção de insumos dado o vetor de preços - não será mensurada neste estudo. De acordo com Farrel (1957), a ineficiência alocativa deve ser estudada quando se objetiva verificar se as firmas agem dentro de um arcabouço lucro-maximizador ou custo-minimizador. Como o objetivo aqui é apenas a obtenção de uma relação técnica da função de produção e, além disso, não se dispõe de informações 2 Gupta & Verhoeven (2001), por exemplo, usam uma abordagem FDH. Afonso & St. Aubyn (2004), Herrera & Pang (n.d.) e Herrera & Worldwide (2005) usam tanto DEA quanto FDH. Uma exceção do ponto de vista metodológico é Greene (2003b) que estima uma fronteira estocástica para avaliar a eficiência dos gastos em saúde usando os dados do World Health Organization (WHO). 3 A literatura que trata da eficiência do setor público seguiu de perto a literatura que trata da eficiência do setor privado e, assim, países, estados e municípios são considerados analiticamente como iguais às firmas. 54 Souza, Nishijima e Rocha Economia Aplicada, v.14, n.1 sobre preços, pois se trata de serviços prestados pelo setor público, estima-se apenas um modelo de fronteira de produção. 4 Para mensurar a eficiência técnica, a literatura nos apresenta técnicas paramétricas e não paramétricas. Dentre as técnicas paramétricas, a mais difundida, e com propriedades amplamente pesquisadas, é a fronteira estocástica de produção. Já dentre as técnicas não paramétricas, tem-se o Free Disposal Hull (FDH) e Data Envelopment Analisys (DEA). Neste estudo opta-se pelo uso de da fronteira estocástica, pois ela apresenta algumas propriedades que não estão disponíveis nos métodos não paramétricos de acordo com Coelli et all (2005). São elas: (1) a possibilidade de se realizar testes de hipóteses sobre os parâmetros das variáveis explicativas, (2) a possibilidade de se incluir variáveis de controle para explicar a ineficiência técnica em apenas um estágio e (3) permite a presença de ruídos aleatórios no ambiente em que a unidade tomadora de decisão opera5 . Embora as fronteiras estocásticas apresentem as vantagens descritas anteriormente, exigem a necessidade de imposição de uma forma funcional a priori e de hipóteses acerca da distribuição do termo de ineficiência6 . Estes são custos que a metodologia DEA não impõe uma vez que tal técnica simplesmente mapeia os pontos dados existentes e traça, sobre os mesmos, um envoltório convexo que contém todos os pontos observados. Dessa forma, a única restrição que o DEA impõe é a convexidade dos conjuntos de produção. Já o FDH, assim como o DEA, não apresenta as propriedades descritas para as fronteiras estocásticas. Adicionalmente, esta técnica descarta a hipótese de convexidade dos conjuntos de produção. As características positivas supracitadas das fronteiras estocásticas de produção fundamentam a escolha de tal metodologia. No entanto, na tentativa de atenuar os efeitos da imposição de uma forma funcional a priori que a escolha de tal método traz consigo, optou-se pelo uso da forma funcional Flexível de Fourrier7 . Não existe na literatura um consenso acerca de qual método é superior. Evidências apontam que ambos produzem, em muitos casos, resultados pouco robustos quando comparados entre métodos. Uma dessas evidências pode ser vista em Jacobs (2000). Em tal estudo, examinam-se as propriedades das fronteiras estocásticas e do DEA, em diversas especificações, para o cálculo de eficiência hospitalar no Reino Unido. Seus resultados apontam para uma robustez interna dos métodos, no tocante aos rankings de eficiência, mas não entre métodos. Conforme exposto, as fronteiras estocásticas de produção permitem a decomposição do resíduo em dois componentes: um relativo à ineficiência e outro relativo ao choque puramente aleatório, chamado de erro idiossincrático. Esse método foi introduzido na literatura econômica simultaneamente por Aigner et al. (1977) e Meeusen & Van de Broeck (1977) numa versão crosssection e sua especificação para dados em painel feita por Pitt & Lee (1981) é descrita em (1). 4 Apesar disso, a metodologia aqui aplicada permite afirmar que se uma unidade tomadora de decisão é tecnicamente ineficiente, então, ela não maximiza seu retorno (Kumbhakar e Lovell, 2000). 5 Métodos não paramétricos incorporam esses ruídos no valor da ineficiência técnica. 6 Maiores detalhes serão dados a seguir. 7 Propriedades da forma Flexível de Fourrier serão apresentadas mais adiante. Eficiência do setor hospitalar k ln yit = β0 + f (βk , xit ) + vit − uit , i = 1, . . . , I; k = 1, . . . , n; t = 1, . . . , T 55 (1) em que ln yit é o logaritmo da quantidade produzida pelo município i no período t; β0 é o intercepto da equação; f (βk , xik ) é a forma funcional adequada; βk é o vetor de coeficientes tecnológicos; k é o vetor de insumos utilizado na produção pelo município i no período t; xit k e uit e com distribuição vit é o choque aleatório não correlacionado com xit 2 N (0, σv ); uit é o termo de ineficiência não negativo do município i no período t também k . não correlacionado com xit Para a estimação dos coeficientes tecnológicos das fronteiras estocásticas de produção são empregados os estimadores de máxima verossimilhança8 , sendo necessário assumir hipóteses explícitas sobre a distribuição assimétrica do termo de ineficiência, uit . Aqui se utiliza o modelo proposto por Battese & Coelli (1995) que desenvolveram uma metodologia de estimação para painéis desbalanceados utilizando uma distribuição normal-truncada. A escolha desta distribuição resulta do fato de ser mais flexível do que outras formas funcionais mais simples como a semi-normal ou a exponencial.9 O modelo especificado por Battese & Coelli (1995) também permite a variação da ineficiência técnica ao longo do tempo, que pode ser modelada usandose características dos municípios que variam ao longo do tempo. A equação (2), em que o termo de ineficiência segue uma distribuição normal truncada uit ∼ N + (zit δ, σu2 ), mostra o termo zit δ que representa a média da ineficiência, sendo esta composta pelo vetor de variáveis específicas dos municípios zit e δo vetor de coeficientes associados a essas variáveis. Assim, o termo ineficiência é modelado como uma média condicional de um conjunto linear de covariadas pré-especificadas. O termo wit é uma variável correspondente à truncagem de uma normal com média zero e variância σu2 no ponto (− zit δ), ou seja, wit ≥ (− zit δ). uit = zit δ + wit (2) Conforme, já explicitado, escolheu-se a forma funcional Flexível de Fourier, que consiste numa aproximação global que inclui os termos padrões de uma trans-log mais os termos trigonométricos de Fourier, sua especificação é dada pela equação (3) abaixo.10 8 Sobre metodologia de estimação de fronteiras estocásticas ver Kumbhakar & Lovell (2000). 9 Para estimações com distribuição semi-normal ou exponencial ver Aigner, Lovell e Schmidt (1977) e para estimações com distribuições gama ver Greene (2000). 10 Foram feitos testes estatísticos para justificar a escolha da forma flexível de Fourier que serão apresentados e discutidos na seção 3 que trata dos resultados das estimativas. 56 Souza, Nishijima e Rocha ln yit =β0 + βt t + βtt t 2 Economia Aplicada, v.14, n.1 n X k βk ln xit + k=1 + + n X k βkt ln xit t+ k=1 n X n X k=1 q=k n XX j k βjk ln xit ln xit j≤k k=1 n X [φ k cos(hk ) + ωk sen(hk )] k=1 (3) [φ kq cos(hk + hq ) + ωkq sen(hk + hq )] + vit − uit , i = 1, . . . , I; k = 1, . . . , n; t = 1, . . . , T . Sendo assim é possível, a partir da estimação da equação acima, testar a adequação tanto de uma trans-log, quanto de uma Cobb-Douglas. De acordo com Berger & Mester (1997), esta forma funcional é uma aproximação global pelo fato dos termos cos(hk ), sen(hk ), cos(hk + hq ) e sen(hk + hq )serem mutuamente ortogonais no intervalo [0;2π], aproximando a função a ser estimada de seu verdadeiro caminho.11 Segue-se a sugestão dos autores para o corte de 10% de cada cauda do intervalo [0, 2π] evitando-se, desta maneira, problemas de aproximação nas fronteiras do intervalo. Como conseqüência, os termos hk e hq são calculados por (4). hk = 0, 2 × π − µ × a + µ × xk . (4) intervalo12 Sendo [a,b] o transformado em radianos e µ ≡ (0, 9 × 2π − 0, 1 × 2π)/(b − a). Portanto, para obter as estimativas dos parâmetros da equação acima por máxima verossimilhança, basta maximizar a função de verossimilhança deεit . Para tal, é preciso conhecer a distribuição de εit , sendo εit = vit − uit . Supondo vit e uit independentes e com suas respectivas distribuições conhecidas, a distribuição conjunta vit e uit é dada por f (vit , uit ) = f (uit ) × f (vit ) e desde que εit = vit − uit tem-se quef (vit , uit ) = f (uit + εit , uit ) = f (εit , uit ). A função de distribuição conjunta assume a expressão dada em (5). f (ε, u) = e − 21 ( (u−µ∗ )2 σ∗2 " + (ε−zδ)2 σv2 +σu2 ( ) 2πσu σv Φ (zδ/σu ) #) (5) onde: µ∗ = σv2 zδ − σu2 ε σv2 + σu2 σ2σ2 σ∗2 = 2u v 2 σv + σu e Φ(zδ/σu ) é a função de distribuição acumulada da normal padrão avaliada no ponto (z δ/σu ). Para obter a distribuição R ∞ do erro idiossincrático ε, integra-se a função (5) com respeito a u: f (ε) = 0 f (ε, u)du, para obter (6). 11 O que lhe confere a flexibilidade destacada anteriormente. 12 São respectivamente os valores mínimo e máximo de xk. Eficiência do setor hospitalar f (ε) = q e ( − 21 (ε+zδ)2 σv2 +σu2 ( ) ) 2π σv2 + σu2 [Φ (zδ/σu ) Φ (µ∗ /σ∗ )] −∞ ≤ ε ≤ ∞ , 57 (6) As estimativas dos parâmetros são obtidas a partir maximização do logaritmo da função de verossimilhança dada por (7). N L(β, δ, σu , σv ) = − N T i yit − xit β + zit δ 1 XX ln 2π + ln σs2 X Ti − 2 2 σs2 i=1 i=1 t=1 − Ti h N X X i=1 t=1 !2 i ln Φ(dit ) − ln Φ(dit∗ ) (7) onde: dit = zit δ √ σs2 γ µ∗it = (1 − γ)zit δ − γ(yit − xit β) γ≡ σu2 . σs2 dit∗ = µ∗ p it σs γ(1 − γ) σs2 ≡ σu2 + σv2 A parametrização (7) permite verificar a relevância do termo de ineficiência. Se o termo γ converge para 1 o termo de ineficiência predomina o erro idiossincrático e se γ converge para zero a ineficiência torna-se irrelevante para explicar o termo ε. No último caso o emprego da técnica de FE não traz ganhos em relação ao método de mínimos quadrados ordinários13 . Deste modo, caso não se rejeite a hipótese de ineficiência técnica, obtêm-se as estimativas destas a partir da distribuição condicional de uit em εit dada pela expressão (8). µ Φ σ∗ − σ∗ 2 ∗ µ E(e uit |εit ) = e (−µ∗ +σ∗ /2) (8) Φ σ∗ ∗ A partir da estimação da ineficiência é possível calcular os scores de eficiência que permitem ordenar os municípios de acordo com seu desempenho relativo. A preocupação, contudo, aqui é com os determinantes de ineficiência e não com a distância dos mesmos em relação à fronteira. 3 Dados e resultados econométricos A amostra é composta por 366 municípios paulistas em painel abrangendo os anos de 1998 a 2003, o que corresponde a 89% do total de municípios com rede hospitalar disponível. Além disso, a população desses 366 municípios corresponde a 94% dos residentes do Estado de São Paulo em 2003. 13 A aplicação de mínimos quadrados parte do pressuposto de que não existe correlação entre o termo residual e a matriz de variáveis independentes X. Se esta hipótese não puder ser sustentada, então torna-se necessário o uso de estimadores de efeito fixo ou mínimos quadrados de dois estágios. 58 Souza, Nishijima e Rocha Economia Aplicada, v.14, n.1 Tabela 1: Variáveis utilizadas Variáveis Obs. Média Desvio padrão Mín. Máx. Complemento mortalidade hospitalar Pessoal Leitos Tendência (T) População Internações Gestão PMDB PSDB PT PFL PTB Demais partidos Internações hospitais públicos Internações hospitais privados Internações hospitais universitários Internações cirurgia clínica Internações obstetrícia Internações clínica médica Internações pediatria Outras internações Gasto público Permanência 2135 2129 2135 2135 2135 2135 2135 2135 2135 2135 2135 2135 2135 2135 2135 2135 2135 2135 2135 2135 2135 2134 2129 97.2988 440191 274.87 3.46042 95696.8 6202.09 0.32758 0.16815 0.27681 0.06464 0.13068 0.10258 0.25714 0.21261 0.76802 0.01937 0.15145 0.19290 0.49633 0.12503 0.03429 268.59 6.75514 1.85975 2886995 1345.44 1.70676 558698 28076.5 0.44198 0.37409 0.44753 0.24594 0.33713 0.30348 0.43716 0.39081 0.39903 0.10904 0.11261 0.13439 0.20742 0.08572 0.11224 510.56 23.2544 76.92 23.57 6 1 3320 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 63.36 1.3 100 6, 77 × 107 28091.6 6 1, 07 × 107 561530 1 1 1 1 1 1 1 1 1 0.93467 0.61576 1 1 0.75521 1 20722.0 956.4 Fonte: DATASUS e IBGE O critério de seleção foi a existência de rede hospitalar no município ( SUS e particular contratado14 ) e a disponibilidade de informações. Supõe-se que pacientes de localidades onde não exista rede hospitalar se desloquem para aqueles municípios que possuem hospital, fenômeno este que também pode ocorrer devido às especialidades disponíveis em cada município. Para estimação da função de produção hospitalar dos municípios paulistas considerou-se como produto (ly) o logaritmo da variável complemento da taxa de mortalidade hospitalar por município (100 - taxa de mortalidade hospitalar)15 . Ainda que por questões de indisponibilidade de dados siga-se a literatura de eficiência de gastos públicos em saúde, esta utiliza como medida de produto a mortalidade infantil e a expectativa de vida. Como aqui a preocupação é com a eficiência dos hospitais de procura-se utilizar uma medida que capte a qualidade do atendimento hospitalar do SUS. Quanto menos pessoas morrem nos hospitais teoricamente melhor é a qualidade do atendimento hospitalar. Para mensurar as variáveis explicativas, trabalho (lw) e capital (lk), foram consideradas como proxies respectivamente, o valor dos serviços de profissionais empregados nas internações hospitalares e o número de leitos contratados pelo SUS em cada município. A variável de natureza monetária (lw) foi deflacionada pelo IPCA e está expressa em reais de 1998. Todas as informações sobre hospitais foram obtidas junto ao Datasus e as informações complementares foram obtidas junto ao IBGE e estão resumidas na Tabela 1. 14 O SUS aluga a capacidade produtiva do setor privado. 15 Usou-se o completo para tornar mais fácil a leitura das estimativas da função de produção. Eficiência do setor hospitalar 59 A Tabela 1 apresenta inicialmente as variáveis utilizadas na função de produção: taxa de sobrevivência, que corresponde ao complemento da variável taxa de mortalidade hospitalar; o valor de gastos com profissionais ligados diretamente às internações como proxy do insumo trabalho, o número de leitos (do SUS e contratados) como proxy do insumo capital e uma variável que descreve uma tendência no tempo (T). As demais variáveis são usadas como “controle”16 e , assim, aparecem como variáveis explicativas do termo ineficiência estimado. Foram usadas as seguintes variáveis explicativas por município: população; número de internações; gestão, variável com valores compreendidos entre 0 e 1 que descreve o percentual de internações realizadas sob gestão plena do município17 ; dummies de partidos que estavam à frente da prefeitura durante os anos de 1998 a 2003; variáveis que descrevem o percentual das internações que ocorreu em leitos públicos (do próprio SUS), privados contratados ou em hospitais universitários, portanto, variando entre 0 e 1; variáveis que descrevem os percentuais de internações realizadas em cirurgia clínica, obstetrícia, clínica médica, pediatria e demais tipos de internação18 ; o valor médio do gasto público por internação; e o tempo médio de permanência na internação. Considerando que as variáveis usadas como insumos não estão em valores per capita ou por quantidade de internações é preciso controlar pelo tamanho da população e pelo número de internações para tornar compatíveis os resultados. As dummies de partido e a variável de gestão municipal plena buscam captar efeitos da forma de gestão sobre a eficiência produtiva do setor, ou seja, verificar se a gestão descentralizada nas prefeituras é mais eficiente que a gestão mais central por parte dos Estados. As variáveis que descrevem os percentuais de internações em leitos públicos, privados e universitários buscam inferir se existem efeitos diferenciados entre a provisão de internações públicas diretamente pelo setor público e a produzida pelo setor privado. Já as variáveis de percentuais de tipos de internação buscam reduzir o grau de homogeneidade da informação sobre taxa de mortalidade hospitalar, uma vez que traz informações sobre o tipo de tratamento recebido na internação, controlando assim por diferenças possíveis de tratamentos envolvidos na internação, como por exemplo diferentes graus de complexidade. A variável valor médio da internação busca controlar gastos diferenciados sobre a taxa de sobrevivência hospitalar e a variável tempo médio de internação os tempos diferenciados dos tratamentos. Além disso, um termo de tendência foi incluído no vetor Z a fim de verificar como a média da eficiência técnica se comporta no tempo, isso é, se os municípios estão, em média, convergindo ou divergindo da fronteira. A Tabela 2 mostra as estimativas19 da fronteira estocástica para os hospi16 Colocamos entre aspas para lembrar que estas variáveis não entram diretamente na função de produção que em nosso caso é utilizada a forma flexível de Fourier. A especificação colocandoas como controle, entretanto, não alterou a direção (sinais dos coeficientes) dos valores das estimativas. 17 A alternativa compreende internações realizadas sob gestão estadual ou composição de gestão em que o município gere apenas a atenção básica em saúde. 18 Note que tanto estas variáveis como as anteriores equivalem a um conjunto de dummies centradas, ou seja, seus valores somam 1, mas não são compostas exclusivamente de zeros e uns. 19 As estimativas foram obtidas utilizando o pacote Stata. 60 Souza, Nishijima e Rocha Economia Aplicada, v.14, n.1 tais paulistas. O modelo (1) descreve o modelo inicialmente considerado para a análise. Os modelos de (2) a (5) atestam a robustez dos resultados obtidos, uma vez que consistem nas estimativas do modelo (1) com a retirada consecutiva dos “controles” da ineficiência. Na primeira parte da tabela estão os resultados das estimativas da função de produção usando a forma flexível de Fourier. Assim, além das variáveis de trabalho (Pessoal) e capital (Leitos) tem-se o produto do logaritmo do capital e do trabalho (Pessoal×leitos), o tempo ao quadrado (Tendência2 ), o quadrado do logaritmo do trabalho (Pessoal2 ) , o quadrado do logaritmo do capital (Leitos2 ), o tempo vezes o logaritmo do trabalho (Tempo×pessoal), o tempo vezes o logaritmo do capital (Tempo×leitos) que são os termos tradicionais de uma função trans-log. Os termos f_cos_w, f_cos_k, f_sen_w, f_sen_k, f_cos_2w, f_cos_2k, f_cos_wk, f_sen_2w, f_sen_2k e f_sen_wk são os termos de Fourier. Os resultados da Tabela 2 sugerem, como esperado, que capital e trabalho afetam positivamente a produção de sobrevivência hospitalar. Atestam também que o capital ao quadrado, que mede uma grande capacidade produtiva instalada, proporciona menor taxa de mortalidade hospitalar. As significâncias estatísticas dos vários termos de Fourier atestam a escolha adequada desta forma funcional. Na Tabela 3 são apresentados os coeficientes estimados das variáveis explicativas da ineficiência. Embora as estimativas da função de produção e da ineficiência tenham sido feitas conjuntamente, optou-se por apresentar os resultados separadamente para facilitar a análise. Inicialmente é importante observar que a hipótese de inexistência da ineficiência pode ser rejeitada. A estimativa gama é muito próxima da unidade, indicando a predominância do termo ineficiência20 sobre a variância do erro idiossincrático. Testes de adequação das formas funcionais também foram realizados, indicando que a escolha da forma funcional Flexível de Fourier é adequada (Tabela 4). Outro teste reportado nessa mesma tabela é o teste conjunto nos coeficientes que explicam a ineficiência. O objetivo de tal teste é verificar se a ineficiência é melhor descrita por uma constante do que condicionada a características individuais que poderiam explicá-las. Assim, as diferenças observadas entre firmas na ineficiência, seriam aleatórias. Esse teste também apresenta o resultado favorável do teste à modelagem da ineficiência. Assim sendo o conjunto de características descrito anteriormente parece ser eficaz na determinação do valor da ineficiência técnica. Os resultados das estimativas do termo ineficiência mostram que quanto maior a população do município maior é a ineficiência do setor hospitalar. Esse resultado pode refletir um efeito congestionamento. As indivisibilidades características da produção de certos serviços de saúde (por exemplo, serviços de alta complexidade) restringiriam a provisão desses serviços aos grandes centros urbanos, implicando um custo de congestionamento. Já com relação ao número de internações realizadas por um município quanto maior é este 20 A estatística LR não pode ser calculada para o teste conjunto de todos os coeficientes que compõem o termo de ineficiência porque não houve convergência dados os valores iniciais. Seus valores críticos estão em Koode e Palm (1986). Entretanto, estimando por MQO, o teste de Coelli comprova a ineficiência dos resíduos. Eficiência do setor hospitalar 61 Tabela 2: Estimativas da função de produção Pessoal Leitos Tendência Tendência2 Pessoal×leitos Pessoal2 Leitos2 Tempo×pessoal Tempo×Leitos F_cos_w f_sen_w f_cos_k f_sen_k f_cos_wk f_sen_wk f_cos_2w f_sen_2k Constante (1) (2) (3) (4) (5) 0.0135∗∗∗ (0.00453) 0.0189∗∗∗ (0.00453) 0.0147∗∗∗ (0.00452) 0.0187∗∗∗ (0.00452) 0.0156∗∗∗ (0.00453) 0.0179∗∗∗ (0.00453) 0.0170∗∗∗ (0.00448) 0.0178∗∗∗ (0.00448) 0.0196∗∗∗ (0.00437) 0.0188∗∗∗ (0.00437) 0.000234 (0.00157) 0.000409 0.00058 (0.00156) (0.00155) −0.000144 −0.000158∗ (0.00009) −0.00576∗∗∗ (0.00092) 9.83 × 10−5 (0.00033) 0.00517∗∗∗ (0.00074) −0.000148 (0.00009) −0.00578∗∗∗ (0.00093) (0.00009) −0.00572∗∗∗ (0.00093) 0.000274 0.000198 0.000149 (0.00033) 0.00506∗∗∗ (0.00075) 0.000139 (0.00021) (0.00033) 0.00504∗∗∗ (0.00075) 0.000179 (0.00021) (0.00033) 0.00503∗∗∗ (0.00074) 0.000158 (0.00021) (0.00027) −1.017∗∗∗ (0.29700) 1.088∗∗∗ (0.31000) 0.0273∗∗ (0.01300) −0.000513∗ (0.00027) −1.063∗∗∗ (0.29500) 1.136∗∗∗ (0.30800) 0.0297∗∗ (0.01280) −0.000487∗ (0.00027) −1.082∗∗∗ (0.29300) 1.156∗∗∗ (0.30600) 0.0308∗∗ (0.01270) −0.0322 −0.0335 −0.0342 −0.000391 (0.02130) −0.0522∗ (0.03020) −0.0784∗∗∗ (0.02580) −0.0699∗∗∗ (0.02460) 0.0668∗∗∗ (0.02480) 4.564∗∗∗ (0.02530) 0.000884 (0.00157) −0.000154∗ (0.00009) −0.00566∗∗∗ (0.00093) −0.00015 0.000758 (0.00157) (0.02110) −0.0533∗ (0.03000) −0.0816∗∗∗ (0.02550) −0.0727∗∗∗ (0.02440) 0.0702∗∗∗ (0.02460) 4.559∗∗∗ (0.02510) (0.02100) −0.0543∗ (0.02980) −0.0833∗∗∗ (0.02530) −0.0739∗∗∗ (0.02410) 0.0719∗∗∗ (0.02430) 4.557∗∗∗ (0.02500) 0.000134 (0.00020) −0.000480∗ (0.00026) −1.167∗∗∗ (0.29300) 1.249∗∗∗ (0.30600) 0.0341∗∗∗ (0.01260) −0.0379∗ (0.02100) −0.0594∗∗ (0.02960) −0.0879∗∗∗ (0.02520) −0.0796∗∗∗ (0.02410) 0.0778∗∗∗ (0.02410) 4.552∗∗∗ (0.02500) (0.00009) −0.00536∗∗∗ (0.00090) −0.000106 (0.00032) 0.00459∗∗∗ (0.00073) 0.000117 (0.00020) −0.000479∗ (0.00026) −1.278∗∗∗ (0.29300) 1.368∗∗∗ (0.30600) 0.0311∗∗ (0.01260) −0.0386∗ (0.02190) −0.0615∗∗ (0.03010) −0.0929∗∗∗ (0.02580) −0.0849∗∗∗ (0.02460) 0.0775∗∗∗ (0.02440) 4.547∗∗∗ (0.02520) Obs: ***, ** e * implicam significância estatística aos níves de 1%, 5% e 10%, respectivamente. número menor é a ineficiência hospitalar, sugerindo que municípios que possuem maior escala de produção são mais eficientes.21 Os coeficientes negativos das variáveis internações em leitos privados e internações em leitos universitários significam que municípios onde predominam hospitais com estas formas de gestão são mais eficientes (o sinal negativo implica que a variável tem um impacto de reduzir a ineficiência). Em relação à significância dos leitos em hospitais privados contratados pelo SUS, pode-se pensar que a provisão dos bens de saúde pelo setor privado tende a ser mais eficiente do que a oferta direta pelo setor público. Como discutido anteriormente, a justificativa para a inclusão de dummies de partido está no fato da saúde, atualmente, ser um serviço público descentralizado. Dessa forma, as prefeituras administram boa parte dos recursos repassados pelo SUS. Acredita-se que prefeituras administradas por um mesmo partido partilhem de práticas comuns de gestão resultantes, entre outras coisas, de programas partidários de Governo. Adicionalmente, para uma melhor inferência sobre gestão é necessário considerar se a gestão é municipal plena ou estadual. A insignificância estatística de todas estas variáveis para expli21 Para uma avaliação do papel do congestionamento e da presença de economias de escala na provisão de serviços públicos municipais de saúde e educação ver Mendes e Sousa (2006). O trabalho de Mendes e Sousa avalia essas questões olhando a função de demanda por serviços públicos e neste sentido é complementar ao presente estudo que enfatiza o lado da oferta. 62 Souza, Nishijima e Rocha Economia Aplicada, v.14, n.1 Tabela 3: Determinantes da ineficiência População Internações Tendência Internações hospitais púbicos Internações hospitais universitários PMDB PT PFL PTB Demais Gestão Internações obstetrícia Internações clínica médica Internações pediatria Outras internações Gastos públicos Permanência Constante ilgtgamma Cons lnsigma2 Cons σ2 γ σu2 σv2 Observações (1) (2) (3) (4) (5) 0.0960∗∗∗ (0.02090) −0.113∗∗∗ (0.02140) −0.0151∗∗ (0.00767) −0.181∗∗∗ (0.05230) −0.561∗ (0.32600) 0.0922∗∗∗ (0.02540) −0.103∗∗∗ (0.02500) 0.103∗∗∗ (0.02550) −0.123∗∗∗ (0.02530) 0.0840∗∗∗ (0.02060) −0.111∗∗∗ (0.02110) 0.101∗∗∗ (0.02200) −0.126∗∗∗ (0.02230) −0.0032 (0.00692) −0.00371 (0.00845) −0.00717 −0.00718 (0.07070) (0.07280) −0.223∗∗∗ (0.06710) − 686 (0.45700) 0.00389 −0.199∗∗∗ − 608 (0.38100) 0.0131 (0.04620) − 66 (0.03600) −0.0829∗ (0.04830) 0.0186 24 (0.03270) (0.03690) 0.000294 (0.03770) 0.00543 (0.03990) 0.0148 (0.03980) 23 −0.242∗∗∗ (0.46400) −0.0879 (0.05940) 0.0128 (0.04750) 0.0306 (0.04780) 0.0128 (0.03250) (0.03640) 0.0278 0.0253 (0.02660) − 177 (0.13200) − 164 (0.12700) −0.471∗∗∗ (0.16000) −0.670∗∗∗ (0.21700) − 678 0.0187 0.0374 − 124 (0.00711) (0.04340) (0.02870) (0.02560) (0.00767) (0.03190) (0.13700) − 134 (0.12300) −0.469∗∗∗ (0.17500) −0.0724 (0.14000) −0.00468 (0.04170) 0.120∗∗∗ (0.03630) − 384 − 324 (0.28100) 4.036∗∗∗ (0.25200) −5.556∗∗∗ (0.23200) (0.20700) 4.075∗∗∗ (0.33600) −5.522∗∗∗ (0.31600) 0.0038625 0.9826323 0.0037954 0.0000671 2129 0.003996 0.9832849 0.003929 0.0000668 2129 −0.547∗∗∗ −0.385∗∗ −0.516∗∗∗ 0.0051577 0.9870321 0.0050908 0.0000669 2129 0.0046625 0.9854348 0.0045946 0.0000679 2190 0.0042371 0.9848355 0.0041729 0.0000643 2190 (0.20100) 4.332∗∗∗ (0.26700) −5.267∗∗∗ (0.24800) (0.15800) 4.214∗∗∗ (0.27000) −5.368∗∗∗ (0.24900) (0.17100) 4.174∗∗∗ (0.24400) −5.464∗∗∗ (0.22100) Obs: ***, ** e * implicam significância estatística aos níves de 1%, 5% e 10%, respectivamente. car o termo de ineficiência sugere que os resultados eleitorais não alteram o padrão de eficiência produtiva do setor hospitalar para o período estudado. Além disso, para o caso particular da variável de gestão municipal plena, parece não existirem ganhos de eficiência decorrente da descentralização da provisão de bens de saúde. O termo de tendência nos coeficientes de ineficiência indica que os municípios ao longo do tempo vão melhorando sua eficiência, isto é, vão convergindo para a fronteira. As variáveis que descrevem os percentuais por tipos de internação - clínica cirúrgica, que é a categoria base, obstetrícia, clínica médica, pediatria e outros (cuidados crônicos prolongados, psiquiatria, tisiologia e reabilitação) indicam internações de pediatria e outros motivos de internação são estatisticamente significantes para redução da ineficiência em relação à clínica cirúrgica. Este resultado era esperado uma vez que a pediatria atende somente crianças en- Eficiência do setor hospitalar 63 Tabela 4: Testes para a forma funcional da função de produção e modelagem da ineficiência Testes para a Forma Funcional Estimativas por MQO Cobb× Translog Douglas Translog × Fourier MV Conclusão: LR p-valor LR p-valor 502,35 0,00 126,77 0,00 LR p-valor LR p-valor 188,17 0,00 51,44 0,00 teste est. p-valor 820.5996 0,000 Teste para a modelagem de eficiência* Rejeita Cobb-Douglas em favor da Translog Rejeita Translog em favor de Fourier Rejeita não modelagem de eficiência * Distribuição Qui-quadrado mista. Valores críticos em Koode e Palm (1986) quanto as demais modalidades atende a todas as idades, o que aumenta a probabilidade de morte durante o tratamento, pois esta aumenta com o avanço da idade. Pela composição da categoria outras internações também se explica o fato de estas terem menos impacto sobre a mortalidade hospitalar, resultando em maior eficiência para a produção. Por fim, a variável gasto médio por internação não se mostra estatisticamente significante para explicar o termo de ineficiência e a variável tempo médio de internação sugere que quanto maior o tempo de internação mais ineficiente é o município. Este último resultado, num primeiro momento poderia sugerir que uma redução no tempo de internação poderia ser ideal, entretanto, deve ser tomado com cautela, pois se deve ter em conta que um tempo maior de internação pode indicar um estoque de saúde muito baixo, de modo que o risco de morte durante a internação é alto. 4 Conclusões e considerações O objetivo deste trabalho é avaliar a eficiência técnica produtiva do setor hospitalar nos municípios do Estado de São Paulo utilizando como variável dependente a taxa de sobrevivência hospitalar (o complemento da taxa de mortalidade hospitalar) e os gastos com profissionais de saúde empregados nestas internações e o número de leitos disponíveis em cada município como variáveis explicativas/insumos. Os resultados obtidos, entretanto, devem ser analisados considerando que as informações hospitalares utilizadas são definidas por local de internação e não por moradia. Além disso, deve-se ter em conta o alto grau de agregação das informações e o fato de os modelos estimados não estabelecerem uma relação de causalidade entre as variáveis e a ineficiência. As estimativas indicam que os municípios com maior população são mais ineficientes, neste caso a população é a residente e as demais informações são de local de internação. Também indicam que são mais eficientes os municípios que: contratam um maior percentual de leitos de hospitais privados e de leitos universitários; possuem maior escala de atendimentos, ou seja, com 64 Souza, Nishijima e Rocha Economia Aplicada, v.14, n.1 maior número de internações por ano; atendem internação por pediatria e por uma categoria denominada outras internações (cuidados crônicos prolongados, psiquiatria, tisiologia e reabilitação); e que apresentam menor tempo médio de internação. Como o produto considerado é a taxa de sobrevivência hospitalar era esperado que internações por pediatria e outras internações se mostrassem mais eficientes, pois têm impactos diretos sobre a taxa de mortalidade. Sendo assim, estas variáveis podem ser vistas como controles para o grau de complexidade do tratamento oferecido na internação. Com relação à significância dos leitos privados e universitários pode-se pensar que a provisão pelo setor privado parece mais eficiente do que a oferta direta pelo setor público. A redução do tempo médio de internação, por sua vez, não deve ser vista como promotora da eficiência, pois pacientes que ficam internados por muito tempo devem na média apresentar capital-saúde baixo e alto risco de morte. O fato de o número de internações estar correlacionado com o termo ineficiência com sinal negativo sugere existirem ganhos de escala na produção de qualidade de atendimento hospitalar medida pela taxa de sobrevivência hospitalar. De fato, os municípios que se mostraram menos ineficientes possuem número de internações anuais muito baixos comparados com os demais. O fato de a população estar correlacionado com o termo de ineficiência com sinal positivo sugere existirem custos de congestionamento. Por último, variáveis de gestão (partidos políticos e percentual de gestão municipal plena) foram testadas e se mostraram não significantes. Essa ausência de correlação sugere que a qualidade dos produtos de internação hospitalar independe do ciclo político e que a gestão municipal não apresenta ganhos quando comparada com a estadual. Referências Bibliográficas Afonso, A., Schuknecht, L. & Tanzi, V. (2003), Public sector efficiency : an international comparison, Working Paper 242, European Central Bank. Afonso, A. & St. Aubyn, M. (2004), Non-parametric approaches to education and health expenditure efficiency in oecd countries, Economics Working Paper 1/2004/DE/CISEP/UECE, ISEG-UTL. Aigner, D., Lovell, C. A. K. & Schmidt, P. (1977), ‘Formulation and estimation of stochastic frontier production function models’, Journal of Econometrics 6, 21–37. Battese, G. & Coelli, T. (1995), ‘A model for technical inefficiency effects in a stochastic frontier production function for panel data’, Empirical Economics 20(2), 325–332. Berger, A. & Mester, L. (1997), ‘Inside the black box: What explains differences in the efficiencies of financial institutions’, Journal of Banking & Finance 21, 895–947. Bernet, P. M., Rosko, M. D. & Valdmanis, V. G. (2008), ‘Hospital efficiency and debt’, Journal of Health Care Finance 34(4), 66–88. Eficiência do setor hospitalar 65 Bradford, W. D. & Kleit, A. N. (2001), ‘Stochastic frontier estimation of cost models within the hospital’, The Review of Economics and Statistics 83(2), 302– 309. Brown III, H. S. & Pagán, J. A. (2006), ‘Managed care and the scale efficiency of us hospitals’, International Journal of Health Care Finance and Economics 6, 278–289. Debreu, G. (1951), ‘The coefficient of resource utilization’, Econometrica 19(3), 273–292. Evans, D., Tandon, A. & e J.A. Lauer, C. J. M. (2000), The comparative efficiency of national health systems in producing health : an analysis of 191 countries, GPE Discussion Paper Series 29, World Health Organization. Farrel, M. J. (1957), ‘The measurement of productive efficiency’, Journal of the Royal Statistical Society 120, 253–281. Series A, General. Greene, W. (2003a), Distinguishing between heterogeneity and inefficiency : stochastic frontier analysis of the world health organization´s panel data on national health care systems, NYU Working Paper EC-03-10, NYU. Greene, W. H. (2003b), Econometric Analysis, Prentice-hall. Gupta, S. & Verhoeven, M. (2001), ‘The efficiency of government expenditures experiences from África’, Journal of Policy Modeling 23, 433–467. Herrera, S. & Pang, G. (n.d.), Efficiency of public spending in developing countries : an efficiency frontier approach. Mimeo. Herrera, S. & Worldwide, W. W. (2005), Efficiency of public spending in developing countries : an efficiency frontier approach, Policy Research Working Paper 3645, The World Bank. Jacobs, R. (2000), Alternative methods to examine hospital efficiency: Data envelopment analysis and stochastic frontier analysis, Working papers 177chedp, Centre for Health Economics, University of York. Jayasuriya, R. & Wodon, Q. (2002), Measuring and explaining country efficiency in improving health and education indicators, Technical report, The World Bank. Kumbhakar, S. & Lovell, K. (2000), Sthocastic Frontier Analysis, Cambridge University Press. Marinho, A. (2001), Avaliação da eficiência técnica nos serviços de saúde dos municípios do estado do rio de janeiro, Texto para discussão 842, IPEA. Meeusen, W. & Van de Broeck, J. (1977), ‘Efficiency estimation from cobbdouglas production functions with composed error’, International Economic Review 18, 435–444. Pitt, M. & Lee, L. (1981), ‘The measurement and sources of technical inefficiency in indonesian weaving industry’, Journal of Development Economics 9, 43–64. 66 Souza, Nishijima e Rocha Economia Aplicada, v.14, n.1 Smet, M. (2007), ‘Measuring performance in the presence of stochastic demand for hospital services: an analysis of belgian general care hospitals’, Journal of Productivity Analysis 27, 13–29. Sousa, M. C. S., Cribari-Neto, F. & Stosic, B. D. (2005), ‘Explaining dea technical efficiency scores in an outlier corrected environment: The case of public services in brazilian municipalities’, Brazilian Review of Econometrics 25(2), 287–313.
Documentos relacionados
Fiscal Monitor - Jornal de Negócios
Natural resource-rich countries benefited from an exceptional commodity price boom during the 2000s, with metal and oil prices reaching historic highs. This provided a substantial boon to resource-...
Leia mais