ON THE DETERMINANTS OF FISCAL DECENTRALIZATION: EVIDENCE FROM THE EU

We empirically analyze the determinants of the fiscal decentralization in the European Union. Our approach consists on a quantile regression for the period 2005-2017. The results show the differences in the impact of explanatory variables on the fiscal decentralization by quantiles. Specifically, while GDP per capita or corruption are not significant in a linear modelling, both are relevant in the quantile approach with different effects along the distribution of fiscal decentralization. And other variables as population, density or inequality do not have neither the same impact among quantiles, denoting the limitations of using linear approaches for complex issues as fiscal decentralization.


Introduction
The allocation of expenditure and revenue among the different levels of government, central and subcentral -local and regional or state in some countries-, is a major issue in Public Economics and, specifically, in Fiscal Federalism. The classical public functions of efficiency -allocation-, redistribution -equity-and stabilization (Musgrave, 1959) are permanently under the debate of the extent of fiscal decentralization, with large differences across nations, arranged as federal or unitary countries. That extent of fiscal decentralization depends on a wide array of factors, namely socioeconomic, institutional and, obviously, historical, in each territory. It should be also noted that the interest of focusing on fiscal decentralization processes is even greater in recent decades, with the need of undertaking fiscal adjustments to control the public debt, especially after the Great Recession, the integration processes as the eurozone, or the secession movements in countries as United Kingdom or Spain, among others.
This paper contributes to the fiscal decentralization literature with the first study with quantile regression, to deal with potential nonlinearities and primarily to explore different impact of determining factors along the distribution of fiscal decentralization. In addition, we analyze the European case, not studied before with this aim † , especially interesting because it is an integrated economic area, although Member States enjoy a large degree of autonomy within their territories. Concretetly, we consider 26 European countries and for the period 2005-2017.
The rest of the paper is organized as follows. Section 1 reviews the literature on determinants of fiscal decentralization. Section 2 describes the empirical strategy, based on quantile regression. Section 3 presents data, including a sigma convergence analysis of the fiscal decentralization measures, and main results. Finally, we offer the main conclusions and potential extensions of the work. The annex includes data for the initial and final years of the period (2005 and 2017).

Theoretical background and literature review
The theoretical background is based on seminal works on Fiscal Federalism and the role of levels of government, as Tiebout (1956) Oates (1972Oates ( , 2005 and Zodrow and Mieszkowski (1986). Theoretically, subcentral levels of government should play a relevant role in the allocative function, basically in terms of local public goods § with the Decentralization Theorem (Oates, 1972) as main reference, while their function in the redistribution of income and stabilization of the economy would be less pronounced. In this manner, the optimal degree of fiscal decentralization has attracted theoretical research since those seminal works, as Janeba and Wilson (2011) or more recently Aslim and Neyapti (2017) and Bellofatto and Besfamille (2018).
And specifically, some papers addressed theoretically the determinants of fiscal decentralization, as Panizza (1999) or Arzaghi and Henderson (2005). Concretely, Panizza (1999), in his theoretical model, concluded that the level of fiscal centralization was inversely correlated with country size, income per capita, tastes differentiation and level of democracy. Subsequently, Arzaghi and Henderson (2005) offered a theoretical model based on the proposal by Panizza (1999) and also the literature on secession, where fiscal decentralization in a country was higher with higher income, larger population, larger in spatial terms, greater degree of local democratic culture and population concentrated in the hinterland.
As stated above, the determinants of fiscal decentralization have been analyzed in relatively few empirical papers to date. Table no. 1 summarizes the papers on this topic. It should be noted that the empirical evidence, short, is some contradictory about the significance of some variables and their impact (positive or negative), reaching mixed results depending the model, determining factors, countries and period considered. We will comment on these results in Section 3 with the results of the study. Hence, our empirical evidence for the European case will help to better understand the determining factors of the distribution of expenditure and revenue among government levels. Positive relation with level of economic development (GDP per capita and urbanization) and population Panizza (1999Panizza ( ) 55 countries 1975Panizza ( , 1980Panizza ( , 1985 Negative effect of country size, income per capita, ethnic fractionalization and level of democracy Cerniglia (2003) OECD countries Positive relation of area, population, degree of urbanization and income per capita ‡ Brueckner (2004) carried out a numerical simulation to explore the fiscal decentralization in terms of Tiebout versus tax competition, taking into account the contrary postulations of both approaches. He concluded that under certain conditions, namely the curvature of the production function and the dispersion of preferences are high, the decentralization is desirable. § See Besley and Coate (2003) for a political economy approach to the trade-off between centralized and decentralized provision of local public goods.

1995-2006
Negative effect of density, and nonsignificant impact of GDP per capita and openness Letelier-Saavedra and Saez-Lozano Geographical fragmentation and area are significantly and positively related to fiscal decentralization

Empirical strategy
We employ a quantile regression approach to capture different patterns along the distribution of fiscal decentralization. Contrary to linear regression, which summarizes the average relationship between the regressors and the dependent variable, this semiparametric approximation, proposed by Koenker and Basset (1978) and revised in Buchinsky (1998), Koenker and Hallock (2001), Koenker (2017) and Waldmann (2018), minimizes the deviations in absolute value with asymmetric weighting, instead of minimizing the squares of the errors as in Ordinary Least Squares (OLS).
In this way, in the quantile regression approach, with the 0.05, 0.25, 0.50, 0.75 and 0.95 quantiles considered, the estimated marginal effects from the estimates of β would indicate how the 5, 25, 50, 75 and 95 per cent conditional quantile would be affected at all x values.
In methodological terms, the quantile regression estimator can be more efficient than OLS if errors deviate from normality and, in addition, the quantile estimators are less sensitive to outliers. Besides, quantile regression provides a richer characterization of the data and is invariant to monotonic transformations.
Regarding the fiscal decentralization measures ** , we consider both the expenditure and revenue sides (FD Expenditure and FD Revenue), taking the percentage of non-central levels of government expenditure or revenue over the total government as indicators, the most employed measures in the literature of fiscal decentralization. The data have been obtained from Eurostat † † .
As explanatory variables, and following the literature on this topic, we consider the level of development through the GDP per capita, population, density, inequality, corruption and a dummy variable for federal versus unitary countries. The model also includes year dummies.
 GDP per capita: the literature does not predict a clear relationship with fiscal decentralization, although mostly positive: while Bahl and Nath (1986), Letelier (2005), Treisman (2006), Martinez-Vazquez and Timofeev (2009) and Bodman and Hodge (2010) found a positive relation between economic development and fiscal decentralization, Oates (1972) and Panizza (1999) concluded a negative relation, while Wu and Wang (2013) did not reach a significant impact. Source: Eurostat.
 Population: although there are reasons for both a positive as a negative effect, we expect a positive relationship, according to Litvack and Oates (1971); as population grows, the rising costs of congestion at the local level of government will tend to increase the noncentral government's expenditures relative to the central government's ones. Data are in millions of inhabitants. Source: Eurostat.
 Density: again, we find arguments in favor of both positive as negative impact on fiscal decentralization. Our expectation, following Letelier (2005), is a negative relationship. On one hand, a lower density will lower public spending, reducing the government's marginal benefit of centralization; on the other hand, as the median voter's marginal utility is decreasing in government expenditures, a lower government budget involves a higher marginal utility of public goods against private consumption. And the assumption is a more significant negative effect of more centralization on the median voter's demand for spending, raising the marginal cost of centralization. Source: Eurostat.
 Inequality ‡ ‡ : we include the Gini coefficient, reflecting the income inequality, but not in wealth. The expected sign is negative, in line with Sepulveda and Martínez-Vázquez (2011), who distinguished direct and indirect effects of fiscal decentralization on income inequality, derived from changes in the implementation of public policies or in the behavior of relevant economic agents, and those observed after the decentralization process has interacted with the socioeconomic framework, respectively. They found empirically that ** For a discussion about fiscal decentralization measures, see Stegarescu (2005), Martínez-Vázquez and Timofeev (2009), Dziobek et al. (2011) or more recently Martínez-Vázquez et al. (2017). † † Government revenue, expenditure and main aggregates (gov_10a_main), available at Eurostat (2020). ‡ ‡ In a related interesting study, Sacchi and Salotti (2014) investigated the effects of fiscal decentralization on income inequality for 23 OECD countries in 1971-2000, concluding that a higher degree of tax decentralization is associated with higher household income inequality within a country. And Kyriacou et al. (2017), for a sample of 23 OECD countries for the period 1984-2005, found that fiscal decentralization, together with measures to improve the quality of government, was an effective strategy for reducing regional inequalities. But they stated that these results have been obtained for rich and advanced economies, and it would be useful to analyze the case of less developed countries.
this impact is negative if the general government represents a significant share of the economy, above 20 per cent GDP. Source: Eurostat.
 Corruption: we consider the Corruption Perceptions Index by Transparency International, an indicator ranging from 0 (highly corrupt) to 100 (very clean). We would expect a negative relationship with fiscal decentralization, according to the hypothesis of Arikan (2004) and Lessmann and Markwardt (2010). Arikan (2004) developed a theoretical model where, as the number of competing jurisdictions rises, the level of corrupt earnings, or tax revenue appropriated by bureaucrats, falls; and found some evidence of this negative relationship in a cross-section data set of 40 countries. In Lessmann and Markwardt (2010), they analyzed the impact of fiscal decentralization on corruption taking into account the degree of freedom of the press, and they did not identify a robust impact of decentralization on corruption, but negative in countries with an eff ective monitoring through a free and independent press.  Federal: this dummy variable assigns value 1 to the three federal countries, Belgium, Austria, Germany, and also to Spain, with three levels of government -central, regional and local-despite it is not formally or politically a federal state, and one of the most decentralized country in the world, even including a fourth level of government represented by the provinces, with limited competences. Of course, the expected sign is positive

Data and results
We study the European Union, concretely 26 countries due to unavailability of data for Croatia and Romania, for the period 2005-2017. The main statistics are reported in Table  no. 2. We observe a great gap in fiscal decentralization, both in expenditure as in revenue, ranging from near zero in Malta to around 72 per cent in Germany. Other few decentralized countries are Ireland and United Kingdom, with indicators under 10 per cent, and among the highly decentralized nations we must also mention France, Finland and Spain, with measures above 50 per cent § § . It should be noted that these three last countries are not federal in political terms, but their degrees of fiscal decentralization are larger than federal countries as Belgium or Austria.  Blanco et al. (2020) analyze the convergence of fiscal decentralization in the European Union. In addition, Finzgar and Oplotnik (2013) revise the fiscal decentralization systems in the EU. Amfiteatru Economic With the aim at deepening the evolution of fiscal decentralization measures in the period, we carry out a sigma convergence analysis. This approach consists on the measurement of the dispersion of the variable through the coefficient of variation, evidencing a sigma convergence process if that dispersion diminishes over time, and sigma divergence if the dispersion increases. We plot this sigma convergence measure along with the average fiscal decentralization to better understand the evolution in our sample (Figure no. 1).
With regard to expenditure, we can observe a growing trend in the (unweighted) average after 2010 and, especially, after 2013. And, however, in the dispersion, we differentiate two main stages: a general growing trend until 2013, followed by a decline to end the period above the initial level of dispersion. Hence, the expenditure does not exhibit a sigma convergence process in the EU.
In relation to the revenue, the path of the average is, to some extent, contrary to the case of expenditure: a decreasing trend since 2009, broken in 2015, ending the period practically at the same level of 2005. Meanwhile, the behaviour of the dispersion is similar to the observed for the expenditure, and we can not conclude a sigma convergence neither in the revenue.

Figure no. 1: Sigma convergence of fiscal decentralization
The main results are summarized in Table no. 3, concretely 3.a for expenditure and 3.b for revenue, being the results virtually identical in both cases. Hence, we will comment only on the expenditure case. The linear results, reported only for benchmarking, show non significant results for GDP per capita nor corruption, significant and positive for population and federal, and significant and negative for density and inequality. It should be noted that most of the signs are the expected ones. But the quantile results allows a richer analysis of the impacts along the distribution of fiscal decentralization.
Regarding GDP per capita, our results show a significant and negative effect for all quantiles except 0.25, and it is stronger in the extremes of the distribution, 0.05 and 0.95, namely the lowest and highest decentralized countries. This negative impact is in line with Panizza (1999), as stated in the review reported in previous section.
For the population, we conclude a significant and positive effect on all quantiles, result in line with Bahl and Nath (1986). In this case, the impact is weaker in the extremes, 0.05 and 0.95, that is, for lowest or highest-decentralized countries.
In the case of density, the effect is negative in all quantiles, and this impact is stronger as we move foreward the distribution of fiscal decentralization.
With regard to inequality, as expected, we find a negative relation with fiscal decentralization, but it is bigger in the first quantiles and finally non significant at 0.95. Hence, the negative impact of inequality is larger for low-decentralized countries, and this effect dissapears for high-decentralized cases.
With respect to corruption, the results reveal that the relationship is significant and negative at the first quantiles, low-decentralized countries, turning into significant and positive in the last ones, 0.75 and 0.95, namely high-decentralized cases.
Finally, the federal dummy presents a significant and positive coefficient in all quantiles, as expected.
The data and results achieved in this paper bring to light a huge variety of degrees of fiscal decentralization across European countries and their determining factors, with impacts varying along the distribution of fiscal decentralization, namely, the effects are not uniform for low, medium or high-decentralized countries. It should be also noted that, among the most decentralized European countries, we find federal and unitary cases, denoting also the relevance of historical issues to explain fiscal decentralization in Europe. We must also take into account that the integration *** and harmonization processes in the European Union do not include rules or recommendations about the fiscal decentralization within the Member States. In another perspective, it should be remarked that the proper European Union, as a whole, faces the centralization versus descentralization issue, in terms of fiscal policy, fiscal discipline and structural reforms (Wyplosz, 2015). *** In a related topic, Ermini and Santolini (2014) studied the effects of globalization on fiscal decentralization for OECD countries, concluding a positive impact of the overall index of globalization, concretely the KOF Globalisation Index, on both tax revenue and expenditure decentralization. Amfiteatru Economic Notes: ***, **, * denotes statistical significance at the 1 per cent, 5 per cent and 10 per cent levels, respectively. Standard errors in parentheses.

Conclusions
Fiscal decentralization, in the expenditure and revenue sides, have attracted researchers in the last decades to investigate issues as the relationship with economic growth, inequality, public efficiency or government size, but few papers have addressed the determinants of fiscal decentralization to date.
In this paper, we study the determining factors of fiscal decentralization in the European Union for the period 2005-2017, considering 26 Member States due to data unavailability for Croatia and Romania. Specifically, and in order to deal with potential nonlinearities and to enrich the analysis along the distribution of fiscal decentralization, we employ a quantile regression approach.
The results show a positive relationship between fiscal decentralization and population. In addition, the impact is negative for GDP per capita, density -with larger negative impact in high-decentralized countries relative to low-decentralized nations-and inequality -although the impact is non-significant in high-decentralization countries. Finally, the evidence for corruption is mixed, being negative for the low-decentralized countries and positive in high-decentralized cases. Hence, we observe how the effects vary at different parts of the fiscal decentralization distribution. It should be also noted that other factors, mainly historical issues, help to explain the variety in the degree of fiscal decentralization in the European Union Member States, taking into account that the integration and harmonization processes do not contain rules or recommendations about fiscal decentralization, prevailing the autonomy of the countries in this matter.
Regarding possible extensions of our study, we could extend the analysis for alternative measures of fiscal decentralization, beyond the two most commonly used in the literature followed in this work, as the Regional Authority Index -although this is available only for some years-, or with some disaggregation in those measures. In addition, it should be interesting to explore the effects of Great Recession on the degree of fiscal decentralization. And finally, to investigate the merging of this literature of fiscal decentralization determinants with club convergence analysis to explain the composition of clusters.