Fiscal Decentralization and the Efficiency of Public Service Delivery
Author: Moussé Sow and Mr. Ivohasina F Razafimahefa

Contributor Notes

Authors’ E-Mail Addresses: msow@imf.org and irazafimahefa@imf.org
Publication Date:
17 Mar 2015
eISBN:
9781484351116
Language:
English
Keywords:
WP; public expenditure; fiscal decentralization; significance level; decentralization process; infant mortality rate; government fractionalization; decentralization of expenditure; Real GDP pc; FD instrumentation test
Download PDF (4 MB)

This paper explores the impact of fiscal decentralization on the efficiency of public service delivery. It uses a stochastic frontier method to estimate time-varying efficiency coefficients and analyzes the impact of fiscal decentralization on those efficiency coefficients. The findings indicate that fiscal decentralization can improve the efficiency of public service delivery but only under specific conditions. First, the decentralization process requires adequate political and institutional environments. Second, a sufficient degree of expenditure decentralization seems necessary to obtain favorable outcomes. Third, decentralization of expenditure needs to be accompanied by sufficient decentralization of revenue. Absent those conditions, fiscal decentralization can worsen the efficiency of public service delivery.

Abstract

This paper explores the impact of fiscal decentralization on the efficiency of public service delivery. It uses a stochastic frontier method to estimate time-varying efficiency coefficients and analyzes the impact of fiscal decentralization on those efficiency coefficients. The findings indicate that fiscal decentralization can improve the efficiency of public service delivery but only under specific conditions. First, the decentralization process requires adequate political and institutional environments. Second, a sufficient degree of expenditure decentralization seems necessary to obtain favorable outcomes. Third, decentralization of expenditure needs to be accompanied by sufficient decentralization of revenue. Absent those conditions, fiscal decentralization can worsen the efficiency of public service delivery.

I. Introduction

This paper analyzes the impacts of fiscal decentralization on the efficiency of public service delivery. It contributes to existing studies by focusing explicitly on the efficiency of public service delivery instead of the policy outcome. The policy outcome can be improved by augmenting policy inputs (for instance, spending allocation); in contrast, efficiency is measured as the difference in policy outcomes—across countries and over time—under a similar set of policy inputs. This paper also covers a large sample of countries, including developed, emerging, and developing economies.2 Last, it uses recent empirical techniques to reach the findings and ascertain their robustness.

The paper’s findings suggest that fiscal decentralization can serve as a policy tool to improve performance, but only under specific conditions. Our findings focus on the efficiency of spending on education and health and indicate that an adequate institutional environment is needed for decentralization to improve public service delivery. Such conditions include effective autonomy of local governments, strong accountability at various levels of institutions, good governance, and strong capacity at the local level. Moreover, a sufficient degree of expenditure decentralization seems necessary to obtain a positive outcome. And finally, decentralization of expenditure needs to be accompanied by sufficient decentralization of revenue to obtain favorable outcomes. Absent those conditions, fiscal decentralization can worsen the efficiency of public service delivery. The paper is structured as follows. Section II reviews the existing literature and summarizes the merits and risks of fiscal decentralization. Section III presents the empirical analysis. Section IV concludes with the main policy recommendations.

II. Literature Review and Theoretical Background

Fiscal decentralization can improve the efficiency of public service delivery through preference matching and allocative efficiency. Local governments possess better access to local preferences and, consequently, have an informational advantage over the central government in deciding which provision of goods and services would best satisfy citizens’ needs (Hayek, 1945; Tiebout, 1956; Musgrave, 1969). When provided by the jurisdiction that has the control over the minimum geographic area, costs and benefits of public services are fully internalized, which is expected to improve allocative efficiency (Oates, 1972).

Fiscal decentralization can also ameliorate efficiencies by fostering stronger accountability. Geographical closeness of public institutions to the local population (final beneficiaries) fosters accountability and can improve public service outcomes, particularly in social sectors such as education and health (Ahmad, Brosio, and Tanzi, 2008; Cantarero and Pacual Sanchez, 2006). Local accountability is expected to put pressure on local authorities to continuously search for ways to produce and deliver better public service under limited resources, leading to “productive effeciency.” Accountability can foster larger spending in public investment and in growth-enhancing sectors, such as education and health (Keen and Marchand, 1997; Arze del Granado and others, 2005; Bénassy-Quéré and others, 2007; Kappeler and Valila, 2008; Fredriksen, 2013). Local accountability can be strengthened through a direct election of local authorities by the local population.

Furthermore, fiscal decentralization can improve efficiency through the “voting with one’s feet” hypothesis. Decentralization gives voters more electoral control over the authorities (Seabright, 1996; Persson and Tabellini, 2000; Hindriks and Lockwood, 2005). It encourages competition across local governments to improve public services; voters can use the performance of neighboring governments to make inferences about the competence or benevolence of their own local politicians (Bordignon and others, 2004). Fiscal decentralization may lead to a decrease in lobbying by interest groups, distorting policy choices and increasing waste of public funds.

However, fiscal decentralization can worsen public service delivery if scale economy is important. Devolution of public service delivery to a small-scale local government can decrease efficiency and increase costs if economies of scale are important in the process of production and provision of some specific public goods. For instance, shifting the production and provision of public services to a municipality with a small size of government officials (producers and providers) and a small population (beneficiaries) can reduce efficiency.

Fiscal decentralization can also obstruct the redistribution role of the central government. To guarantee a minimum level of public service and basic needs (or standard of living) for the entire population (regardless of their geographical location), the central government often carries out equalization transfers, which would be disrupted in cases of insufficient leverage on resources (Ter-Minassian, 1997). When a large share of revenue and expenditure is shifted to local governments, the central government does not possess sufficient resources to ensure a minimum equity across the entire territory.

Fiscal decentralization can also hinder public service delivery if accountability is loose. If accountability is not broadly anchored in a local democratic process, but instead is based on rent-seeking political behavior, local governments would be tempted to allocate higher decentralized expenditure to non-productive expenditure items (such as wages and goods and services instead of capital expenditure). This can hinder efficiency, economic growth, and overall macroeconomic performance (Davoodi and Zou, 1998; Woller and Phillips, 1998; Zhang and Zou, 1998; Rodriguez-Pose and others, 2009; Gonzalez Alegre, 2010; Grisorio and Prota, 2011).

III. Empirical Analysis

A. Methodology

This paper investigates the efficiency, rather than just the outcome, of public service delivery in health and education. Policy outcome is the directly measurable impact of public service delivery; outcome indicators can include infant mortality rate and school enrollment rate. Policy outcomes can be improved by augmenting policy inputs, such as expenditure allocation for health and education. However, the efficiency analysis focuses on the improvement in outcome while keeping inputs unchanged.3 This approach allows analyzing the impact of policies other than inputs in improving the provision of public goods and services; such policies can include fiscal decentralization.

The methodology is based on a two-step approach, estimating efficiency coefficients and analyzing the impact of fiscal decentralization on the latter. In a first step, the efficiency of public service delivery is estimated using stochastic frontier techniques. These techniques provide time-varying coefficients that measure the distance of the public services in a specific country at a specific year to the best public services provided using similar inputs in the sample of countries considered in this analysis. In a second step, this paper estimates the effects of fiscal decentralization on the estimated efficiencies. Instrumental variable methods are used to obtain bias-corrected coefficients. These methods address concerns about endogeneity associated with the decentralization process; they can also tackle reverse causality that could plague the estimated parameters.

In a first step, efficiency coefficients are estimated from stochastic frontier techniques. Methodologies on efficiency estimates can be grouped in two main approaches: (i) a parametric approach (Battese and Coelli, 1988; Jayasuriya and Wodon, 2003; Grigoli and Kapsoli, 2013) and (ii) a non-parametric approach (Gupta and Verhoeven, 2001; Herrera and Pang, 2005; Gupta and others, 2007). This paper uses the parametric approach-based stochastic frontier analysis (SFA). The SFA allows estimating models with multiple inputs, as opposed to non-parametric models that do not take into account the effect of exogenous factors on the outcome variable because of the restriction on the number of variables. As the outcome variables in this paper, that is, infant mortality and enrollment ratio, are plausibly affected by structural factors other than public expenditure, such as socioeconomic characteristics of the country, a multivariable model is better suited for the analysis. Moreover, the SFA allows estimating country-specific and time-varying coefficients.

The SFA techniques assume that no economic agent (i.e., country) can exceed the ideal “frontier.” The frontier refers to the optimum output—infant mortality rate or enrollment rate—produced with limited inputs, such as public expenditure. The deviation of the output in a specific country at a specific time from this frontier represents the individual measure of efficiency of that country. Efficient governments are those operating at, or very close to, the frontier as they try to reduce the infant mortality rate or improve the enrollment rate, given a limited amount of public expenditure.

The first-step model is specified as follows:

Yit=α+γPEit−1+∑k=1KφkZk,it−1+εit(1)
{εit=ωit±ηitηit=g(t)□ηi and g(t)=exp[−λ(t−Ti)](2)

The dependent variable Yit in equation (1) represents public expenditure outcomes on health and education, namely the infant mortality rate and the secondary school enrollment rate, with subscripts i and t denoting respectively country and time dimensions. The interest variable PEit−1 corresponds to public expenditure on health and education as a percent of GDP. A set of control variables Zk, it are added and are likely to influence the infant mortality rate or the enrollment rate. The error term εit in equation (1) has two components as shown in equation (2); ωit represents an idiosyncratic disturbance, capturing measurement error or any other classical noise, and the remaining part ηit is a one-sided disturbance capturing the country-specific and time-varying efficiencies of public expenditure.4 Equations (1) and (2) allow obtaining the country-specific and time-varying efficiencies of public expenditure, following the formula provided by Battese and Coelli (1988) and Jondrow and others (1988).

The second step consists of measuring the extent to which fiscal decentralization affects the estimated efficiencies. The impact of fiscal decentralization is analyzed through a direct channel, a non-linear relationship, and interactions with political and institutional variables. The baseline model is the following:

η^it=α+δfdit−1+φGDPit−1+ψit(3)

The dependent variable η^it is the country-specific and time-varying efficiencies estimated from equations (1) and (2), α is a common constant term, and fdit−1 measures fiscal decentralization. To explore non-linearities in the relationship between fiscal decentralization and public expenditure efficiency, a quadratic specification is added—i.e., squared fiscal decentralization (fdit−1)2—as shown in equation (4). Non-linearities, if any, are detected by computing the derivatives:

η^it=α+δ1fdit−1+δ2(fdit−1)2+φGDPit−1+ψit(4)

Furthermore, the impact of the political and institutional environment on the relationship between decentralization and the efficiency of public service delivery is investigated. Political and institutional variables are introduced additively (Iit-1) but also in interaction with fiscal decentralization (fdit-1 × Iit-1), as shown in equation (5).

η^it=α+δfdit−1+τ(fdit−1×Iit−1)+ρIit−1+φGDPit−1+ψit(5)

Parameter ρ corresponds to the direct effect of political and institutional variables on efficiency. Parameters δ and τ correspond respectively to the effect of fiscal decentralization on the efficiency and the influence of the political and institutional environments on the causal link between fiscal decentralization and public service efficiency. ψit in equations (3)–(5) is a composite error term, taking into account country-specific characteristics.

Fiscal decentralization is measured as the share of subnational fiscal variables over general government fiscal variables.5 The main estimates in this paper are based on the expenditure side of fiscal decentralization, using the share of subnational expenditure to general government expenditure.6 The main focus is on expenditure as it is directly linked to health and education outcomes and efficiency (as opposed to revenue). However, to ensure a comprehensive study, the paper also analyzes the impacts of revenue decentralization on the efficiency of public service delivery, using the share of local government revenue to general government revenue.7 The political and institutional variables focus on the level of corruption, the degree of autonomy of the regions, the strength of the democracy, and the constitutional regime (presidential or parliamentary). Control variables in the stochastic frontier analysis comprise the real GDP per capita as a measure of the level of development, the density and population size, and the average years of primary and secondary schooling. All these variables are considered to influence the infant mortality rate and the secondary school enrollment rate.8 It would be insightful to use the share of subnational expenditure on health and education to general government expenditure in each of the two sectors; however, such data are not available for many of the countries in the sample. Furthermore, the efficiency is influenced by factors beyond expenditure, and an analysis using aggregate expenditure ratio allows clearer comparison with the analysis using aggregate revenue ratio.

Endogeneity and causality concerns are addressed through lag and instrument techniques that motivate the introduction of additional variables. An initial attempt at reducing any bias consists of introducing all explanatory variables, including fiscal decentralization, with a one-period lag. Furthermore, two-stage least squares techniques are applied for the fiscal decentralization variable, using three instrumental variables. First, the population size is considered a significant variable affecting the decentralization process because larger countries generally tend to be more decentralized despite some counter examples (Dziobek et al., 2011; Jiménez-Rubio, 2011; Escolano et al., 2012). The rationale is that in countries with large populations, it is more difficult for central authorities to have sufficient information to target citizens’ needs, which leads to decentralization. Second, the existence of natural resources can act as an obstacle to decentralization, because of possible rent-seeking behaviors of fiscal authorities that benefit directly from the resource windfalls. Under such circumstances, embarking on a fiscal decentralization process would imply a subsequent private loss for incumbent authorities. On the other hand, residents of resource-rich regions can claim larger shares of resources through accelerated decentralization. Moreover, natural resources might be seen as a blessing, triggering the decentralization process because windfalls may constitute an additional source of revenue to share with the subnational governments. Third, government fractionalization and fractionalization in the legislative system can affect the decentralization process. Fractionalization is measured as the probability that two deputies randomly picked either from the government or the legislature will be from different parties. Higher fractionalization may either act against the decentralization process, owing to political motives, or accelerate decentralization. The expected signs of these two last instrumental variables on the decentralization process cannot be determined a priori.

B. Data

The sample covers an unbalanced panel of 64 countries, including advanced, emerging, and developing economies, during 1990–2012. Data are taken from various sources, including the IMF’s Governments Financial Statistics, the World Bank’s World Development Indicators, Eurostat, and OCED databases, among others. Annexes I and II present the full sample, variable definitions, and sources.

Fiscal decentralization is larger in advanced economies than in emerging economies and developing countries, but it has accelerated in the latter two groups in recent decades. Table 1 and Figure 1 provide descriptive statistics of the main variables used in this analysis. On average, about 30 percent of public expenditure is implemented by subnational governments. This share is about 40 percent for advanced economies compared to about 25 percent for emerging economies and developing countries. On the revenue side, the share of subnational governments is about 27 percent; 37 percent in advanced economies; and 23 percent in emerging economies and developing countries. The legislative system appears to be much more fractionalized than the government. The probability that two deputies come from two different parties is 65 percent, whereas it is only 29 percent for members of governments. A higher corruption index indicates a more corrupt system; corruption seems more pervasive in emerging economies and developing countries. The political system index is a binary variable, taking a value of one for parliamentary regimes and zero for presidential regimes; advanced economies appear more parliamentary based than emerging economies and developing economies. A higher democracy score indicates a higher degree of democracy. The “autonomy” indicator is a dummy variable taking the value of one when constitutionally autonomous regions exist in the country.

Figure 1.
Figure 1.

Share of Subnational Government Expenditure/Revenue

(Percent of general government expenditure/revenue)

Citation: IMF Working Papers 2015, 059; 10.5089/9781484351116.001.A001

Source: Authors’ calculations.
Table 1.

Descriptive Statistics

article image
Source: Authors’ calculations.

C. Efficiency Estimates

The average efficiency of the countries in the sample is at about 85 percent of the production frontier. The predicted efficiencies from the stochastic frontier analysis are about 82.2 percent on average for health and 87.8 percent for education (Table 2). An efficiency score of x percent implies that the country delivers x percent of the possible objective (reducing infant mortality rate or increasing school enrollment rate) as compared to a fully efficient country using similar input values (such as public expenditure). The benchmark efficiency estimates—columns (1) and (4) in Table 2—are based on the approach proposed by Battese and Coelli (1988). To check the robustness of the findings, two other methodologies are applied. Efficiency estimates based on Jondrow and others (1982) are presented in columns (2) and (4); and the estimates that take into account heterogeneity and heteroskedasticity are shown in columns (3) and (6). The estimates from those various approaches are highly correlated.

Table 2.

Stochastic Frontier Estimates of Public Service Efficiency

article image
Note: Columns 1 and 4 use the Battese and Coelli (1988) method to estimate the efficiency score, while columns 2 and 5 draw upon the alternative Jondrow et al. (1982) methodology. We allow for heterogeneity and heteroscedasticity while estiamting the efficiency scores in columns 3 and 6.Source: Authors’ calculations.

D. Direct Channel and Non-Linear Relationship

Through a direct channel, expenditure decentralization seems to improve the efficiency of public service delivery in advanced economies but has a negative impact in emerging economies and developing countries. Estimating equation (3), the first step of the two-stage least squares points to the appropriateness of the instrument variables. The latter are significantly correlated with the endogenous regressor in almost all cases (the associated p-values are < 0.05). Besides, using the Kleibergen-Paap’s p values, the null hypothesis that “the equations are underidentified” can be rejected at the 5 percent level. The results of the second step are presented in Table 3. Pooling the advanced economies, emerging markets, and developing economies, it appears that fiscal decentralization has no significant effect on the efficiency of public expenditure (columns 1 and 6). Considering that the various countries exhibit dissimilar levels of decentralization (as shown in the previous section), the sample is divided in two groups: (i) advanced economies, and (ii) emerging markets and developing economies.9 For advanced economies, fiscal decentralization shows positive impacts on the efficiency of public expenditure on health (column 2). To quantify this effect, one could say that a 5 percent increase in fiscal decentralization would lead to 2.9 percentage points of efficiency gains in public service delivery. The coefficient is statistically insignificant for education (column 7). In contrast, for emerging markets and developing economies, the impacts are negative (columns 3 and 8). These positive and negative effects of decentralization, respectively for the first and second group of countries, are robust to the inclusion of time dummies, albeit with a slight reduction in the magnitude of the parameters (columns 4,5,9, and 10). This seems to confirm that the results are not driven by common shocks hitting all countries at the same time, nor by a time-trend evolution of the efficiency scores.

Table 3.

Fiscal Decentralization and Public Expenditure Efficiency

article image
Note: (*), (**) and (***) denote statistical significance level of 10%, 5% and 1% percent respectively. Robust t-statistics are shown in parentheses. Fisher statistic presents a test of join significance of estimated coefficients. Hansen OID and Kleibergen-Paap (KP) test respectively the over-identification restriction and the hypothesis that equations are underidentified. FD instrumentation test, with a lower p-value indicates that endogenous regressors (fiscal decentralization) are significantly correlated with the instrumental variables proposed (political and government fractionalization, and natural resource.Source: Authors’ calculations.

A non-linearity analysis seems to indicate that a sufficient degree of expenditure decentralization is required to bring about positive impacts. The non-linearity is investigated through equation (4), and the results are presented in Table 4. For the entire sample, the fiscal decentralization variable and its squared term affect significantly the efficiency of public services (columns 1 and 4). Interestingly, the coefficient of the former is negative whereas that of the latter is positive. This seems to suggest that the relationship between fiscal decentralization and the efficiency of public service delivery is not linear, but U-shaped. A low level of fiscal decentralization seems to be harmful; it needs to exceed about 35.7 percent for health and 35.4 percent for education to bring about improvements in the efficiency of public services.10 At least, about one third of public expenditure would need to be shifted to the local authorities to obtain positive outcomes from fiscal decentralization. This non-linear relationship might imply the importance of the scale economy in the production and delivery of public services. As many public services require substantial initial fixed costs, if the scale of public services shifted to the local level is too small, the local authorities might have to reduce the provision of services to reduce the variable costs to cover the large initial fixed costs. Note, however, that the sufficient level of fiscal decentralization likely differs across countries, depending on country-specific considerations.

Table 4.

Fiscal Decentralization and Public Expenditure Efficiency (Non-linearity)

article image
Note: (*), (**) and (***) denote statistical significance level of 10%, 5% and 1% percent respectively. Robust T-statistics are shown in parentheses.Source: Authors’ calculations.

The U-shaped relationship is confirmed when the sample observations are split below and above the indicative threshold. For health, when the fiscal decentralization ratio is below the estimated indicative threshold of 35.7 percent, a 1 percent increase in fiscal decentralization ratio reduces the efficiency by about 0.8 percentage point (column 2 of Table 4). In contrast, when the decentralization ratio reaches or exceeds the indicative threshold, decentralization improves the efficiency of public service delivery. A 1 percent increase in the decentralization ratio increases the efficiency by 0.2 percentage point (column 3 of Table 4). For education, the coefficients of the fiscal decentralization are not statistically significant when the sample observations are divided.

The findings on the U-shape relationship are supported by the dissimilar impacts of fiscal decentralization in advanced economies and in emerging markets and developing countries. As shown in Table 3, fiscal decentralization positively affects the efficiency of public services in advanced economies and negatively affects efficiency in emerging markets and developing countries. Interestingly, the level of expenditure decentralization is on average about 40 percent in advanced economies, which is above the mentioned indictive threshold of about 35 percent. In contrast, the average level of expenditure decentralization is only about 25 percent in emerging markets and developing countries, far below the indicative threshold of 35 percent.

E. Political and Institutional Conditions

To support public expenditure efficiency, fiscal decentralization requires an adequate political and institutional environment. Table 5 presents the results of the estimation from model (5). It appears that the interactions of the decentralization and political and institutional variables are significantly associated with the efficiency of public service delivery. Corruption negatively affects the impacts of fiscal decentralization on the efficiency of public services. When taking into account the corruption variable, a 5 percent increase in the fiscal decentralization ratio is associated on average with a 2.5 percent decrease in the efficiency of public expenditure relative to the mean efficiency.11 This might be due to stronger power of interests groups at the local level. Local authorities may also have more discretion and fewer controls, giving room for leakage of public resources, as argued by Gauthier and Wane (2008).12 In contrast, the positive and statistically significant sign of the interaction between fiscal decentralization and the political system variables (FD × Parliamentary(t-1)) indicates that the combination of a parliamentary system and fiscal decentralization may boost public expenditure efficiency. Parliamentary regimes, as opposed to presidential regimes, have stronger institutional frameworks to limit the executive’s discretionary powers. Also, implementing decentralization in a more democratic environment can improve the efficiency of public service delivery. Furthermore, the existence of constitutionally autonomous regions also has positive and statistically significant impacts. Autonomous regions may be free of any vertical constraint that could come from the top level and influence the way public expenditure is implemented locally. The nonsignificance of real GDP per capita used as a control variable in most cases might be because the methodology already controlled for this variable in the first step, when estimating the efficiency.

Table 5.

Fiscal Decentralization and Political/Institutional Environments

article image
Note: (*), (**) and (***) denote statistical significance level of 10%, 5% and 1% percent respectively. Robust t-statistics in parentheses.Source: Authors’ calculations.

The role of the political and instutional environment is also confirmed when seperately analyzing advanced economies and emerging markets and developing countries. Table 6 displays the results of the estimations of equation (5) using two subsamples: (i) advanced economies and (ii) emerging markets and developing economies. First, looking at the coefficients of the fiscal decentralization variable itself, the results support the above findings that decentralization broadly improves the efficiency of public service delivery in advanced economies but worsens the efficiency in emerging markets and developing countries. Second, advanced economies and emerging economies and developing countries seem to broadly confirm that an adequate political and institutional environment improves the impact of fiscal decentralization on the efficiency of public service delivery. For both subgroups and for both health and education, corruption has negative impacts and the autonomy of regions has positive effects on the relationship between decentralization and public service efficiency. This is the expected result because weak governance at the local level might lead to misuse of decentralized resources and expenditure and worsen the efficiency of public service delivery. Sufficient autonomy of local authorities vis-à-vis the central government is needed to allow the preference matching and allocation efficiency to fully operate.

Table 6.

Fiscal Decentralization and Political/Institutional Environments (sub-groups)

article image
Note: (*), (**) and (***) denote statistical significance level of 10%, 5% and 1% percent respectively. Robust t-statistics in parentheses.Source: Authors’ calculations.

F. Robustness

A range of sensitivity analysis is performed to assess the robustness of the findings. Outliers are excluded from the baseline estimates. Then, the baseline model is reestimated using a dependent variable—efficiency of public service delivery—that is derived through alternative methodologies. Finally, the political and institutional variables are replaced with alternative indicators.

The results are robust to the exclusion of countries with extreme ratios of fiscal decentralization. The analysis is conducted using a narrowed sample. Countries totally or almost totally centralized, i.e., with decentralization ratios close to zero, are excluded. Also, countries that have extremely high degrees of decentralization, i.e., decentralization ratios exceeding 90 percent, are dropped. A comparison of the results displayed in Table 7 with those in Table 3 shows that the results are not driven by outliers. Regarding health, the impact of decentralization remains positive for advanced economies, and negative for emerging markets and developing economies, corroborating the baseline findings. The thrust of the results also remains unchanged for education despite a slight difference in the magnitude of the coefficients.

Table 7.

Fiscal Descentralization and Public Expenditure Efficiency: Excluding Outliers

article image
Note: (*), (**) and (***) denote statistical significance level of 10%, 5% and 1% percent respectively. Robust t-statistics in parentheses.Source: Authors’ calculations.

The findings are robust to alternative methodologies of efficiency estimates. Two methodologies are employed to compute alternative estimates of the efficiency of public service delivery: a variante of stochastic frontier analysis based on Jondrow and others (1982) and a methodology that takes into account the sample heterogeneity and heteroskedasticity. The results shown in Table 8 focus on the role of political and institutional variables, and confirm the findings from the baseline analysis.13 Under both alternative efficiency estimates, and for both health and education, corruption hinders—with high statistical significance—the impacts of fiscal decentralization on public service efficiency. The favorable role of parliamentary regimes and more democratic institutions in combination with fiscal decentralization is also confirmed, despite weak statistical significance in some cases. The positive impact of the autonomy of regions on the relationship between fiscal decentralization and efficiency of public service delivery is ascertained with high statistical significance in all cases (alternative efficiency estimates and health and education).

Table 8.

Fiscal Decentralization and public expenditure efficiency: Alternative Efficiency Estimates

article image
Note: (*), (**) and (***) denote statistical significance level of 10%, 5% and 1% percent respectively. Robust t-statistics in parentheses.Source: Authors’ calculations.

The thrusts of the results remain unchanged under an approach that absorbs short-term fluctuations. Fiscal decentralization changes slowly over time and plausibly affects the efficiency of public services with time lags. Thus, it would be useful to check the robustness of the results using averages of the variables over a few-year period. Accordingly, all variables are averaged over a four-year period. In the efficiency of public service delivery and the fiscal decentralization variables, the latter is introduced with a one-period lag. The results, displayed in Table 9, support the baseline findings. Decentralization improves the efficiency of public expenditure in advanced economies (columns 2 and 8). The impact seems negative for emerging markets and developing countries, but it is not statistically significant. In terms of interactive variables, the negative impact of corruption is confirmed (columns 4 and 10). The favorable contribution of parliamentary regimes is also ascertained (columns 5 and 11). As for the autonomy of regions, the impact is positive but not statistically significant.

Table 9.

Fiscal Decentralization and Public Expenditure Efficiency: Absorbing Short-term Fluctuations

article image
Note: (*), (**) and (***) denote statistical significance level of 10%, 5% and 1% percent respectively. Robust t-statistics in parentheses.Source: Authors’ calculations.

Furthermore, the results are broadly robust to alternative political and institutional variables. The following alternative variables are employed: bureaucracy, political stability, and checks and balances.14 All those alternative variables lead to broadly similar inferences as under the baseline analysis; the signs of the coefficiencts are mostly as expected, although statistical significance is low in many cases (Table 10).

Table 10.

Fiscal Decentralization and Public Expenditure Efficiency: alternative political and institutional variables.

article image
Note: (*), (**) and (***) denote statistical significance level of 10%, 5% and 1% percent respectively. Robust t-statistics in parentheses.Source: Authors’ calculations.

G. Revenue Decentralization

Decentralization of revenue can contribute to public service efficiency. Revenue decentralization shows positive and statistically significant impacts on public service delivery for advanced economies and emerging economies and developing countries (Table 11). The findings are robust to alternative estimates of the efficiency variable, based on Jondrow (Table 12) or adjusting for heterogeneity (Table 12). The robustness is further ascertained by excluding outliers or by restricting the sample to only the countries that have revenue decentralization between zero and ninety percent (Table 13). For health and education, and for advanced economies and emerging economies and developing countries, revenue decentralization positively affects the efficiency of public service delivery. These findings might imply the need to accompany expenditure decentralization with sufficient revenue decentralization to ensure improvement of performance.15

Table 11.

Revenue Descentralization: Baseline and Country Specific Estimates

article image
Note: (*), (**) and (***) denote statistical significance level of 10%, 5% and 1% percent respectively. Robust t-statistics in parentheses.Source: Authors’ calculations.
Table 12.

Revenue Decentralization: Alternative Efficiency Estimates

article image
Note: (*), (**) and (***) denote statistical significance level of 10%, 5% and 1% percent respectively. Robust t-statistics in parentheses.Source: Authors’ calculations.
Table 13.

Revenue Decentralization: Excluding Outliers

article image
Note: (*), (**) and (***) denote statistical significance level of 10%, 5% and 1% percent respectively. Robust t-statistics in parentheses.Source: Authors’ calculations.

The importance of a favorable institutional environment is also confirmed by the analysis of revenue decentralization (Table 14). Corruption decreases the positive impact of revenue decentralization on the efficiency of public service delivery. Despite the negative influence of the regime variable, which accounts for the strength of the democracy, the overall effect of revenue decentralization remains positive. The checks and balances variable, which is incrementally coded with the existence of effective control over the executive and legislature in a presidential system, enhances the contribution of revenue decentralization.

Table 14.

Revenue Descentralization: Political/Institutional Interactions

article image
Note: (*), (**) and (***) denote statistical significance level of 10%, 5% and 1% percent respectively. Robust t-statistics in parentheses.Source: Authors’ calculations.

IV. Conclusions and Policy Implications

Fiscal decentralization can serve as a policy tool to improve the efficiency of public service delivery, but only under some conditions. Expenditure decentralization seems to have improved service delivery in advanced economies, but its impacts in emerging economies and developing countries seems rather mixed. The empirical findings in this paper indicate that expenditure decentralization needs to exceed an indicative threshold of about 35 percent to improve service delivery. However, revenue decentralization seems to have positive impacts accross all country groups. This seems to indicate the need to accompany the decentralization of responsibilities with sufficient decentralization of resources.

Findings under expenditure decentralization and under revenue decentralization point to the need for a favorable institutional and political environment. Effective autonomy of local governments is required to allow preference matching and the allocative efficiency hypothesis to operate. Strong accountability of local authorities vis-à-vis the local population is necessary to allow the productive efficiency hypthesis to operate. Corruption needs to be tackled to prevent misuse of public resources. And capacity needs to be strengthened at the local level. Absent those conditions, fiscal decentralization can worsen public service delivery.

An extension of this paper could include an analyis of an alternative indicator of policy outcome and an investigation of the impact of decentralization on other macroeconomic performance. Alternative outcome indicators, such as life expectancy at birth and adjusted primary education net enrollment rate, are presented in the paper to confirm the robustness of the results (Annex III). However, it would be insightful to conduct the analysis using life expectancy, school drop-out rates, or PISA scores as these variables might exhibit larger variance across countries and across time. Moreover, it would be important to analyze the impact of decentralization on key macroeconomic performance, such as fiscal outcome and GDP growth as improvements in public expenditure efficiency can be a channel through which decentralization ultimately influences those variables.

Annex I. Countries, Data Coverage, and Sources

article image

Annex II. Variables, Definitions and Data Sources

article image
Note: Expenditure and Revenue descentralization for European and OECD countries are taken respectively from Eurostat and OECD databases. For emerging economies and developing countries, data are from GFS and WEO.

Annex III. Alternative Policy Outcome Variables (Life expectancy at birth and adjusted primary education net enrollment rate)

Table A3.1.

Fiscal Decentralization and Public Expenditure Efficiency

article image
Note: (*), (**), and (***) denote statistical significance level of 10 percent, 5 percent and 1 percent respectively. Robust t-statistics inSource: Authors’ calculations.
Table A3.2.

Fiscal Decentralization and Political/Institutional Environment

article image
Note: (*), (**) and (***) denote statistical significance level of 10%, 5% and 1% percent respectively. Robust t-statistics in parentheses.Source: Authors’ calculations.

References

  • Ahmad, E., Brosio, G. and Tanzi, V. (2008) “Local Service Provision in Selected OECD Countries: Do Decentralized Operations Work Better?”, IMF Working Paper 08/67.

    • Search Google Scholar
    • Export Citation

  • Arze del Granado, F.A., Martinez-Vazquez, J. and McNab, R. (2005) “Fiscal Decentralization and The Functional Composition of Public Expenditures”, International Center for Public Policy Working Paper Series.

    • Search Google Scholar
    • Export Citation

  • Barankay, I. and Lockwood, B. (2007) “Decentralization and the Productive Efficiency of Government: Evidence from Swiss Cantons”, Journal of Public Economics, 91(5-6), pp. 1197–1218.

    • Search Google Scholar
    • Export Citation

  • Bardhan, P. and Mukherjee, D. (2002) “Decentralization of Governance and Development”, Journal of Economic Perspectives, pp. 185–205.

    • Search Google Scholar
    • Export Citation

  • Battese G.E. and Coelli, T.J. (1988) “Prediction of Firm-level Technical Efficiencies: With a Generalized Frontier Production Function and Panel Data”, Journal of Econometrics, 38, pp. 387–399.

    • Search Google Scholar
    • Export Citation

  • Battese, G. E. and Coelli, T. J. (1995) “A Model for Technical Inefficiency Effects in a Stochastic Frontier Production Function for Panel Data”, Empirical Economics, 20, pp. 325–332.

    • Search Google Scholar
    • Export Citation

  • Bénassy-Quéré A., Gobalraja N. and Trannoy A. (2007) “Tax and Public Input Competition”, Economic Policy, CEPR & CES & MSH, 22(4), pp. 385–430.

    • Search Google Scholar
    • Export Citation

  • Besley, T. and Smart, M. (2007) “Fiscal Restraints and Voter Welfare”, Journal of Public Economics, 91, pp. 755–773.

  • Bordignon, M., Cerniglia, F., and Revelli, F. (2004) “Yardstick competition in intergovernmental relationships: theory and empirical predictions”, Economics Letters, 83, pp. 325–333.