Advertisement

The Annals of Regional Science

, Volume 62, Issue 2, pp 247–264 | Cite as

The technical efficiency of local economies in Mexico: a failure of decentralized public spending

  • Alejandro U. Becerra-Ornelas
  • Hector M. NuñezEmail author
Original Paper

Abstract

This work analyses the effect of the institutional design of public spending on technical efficiency. The model controls for technical efficiency using institutional variables for earmarked and autonomous revenues and assesses them using two stochastic frontier models. The main findings show that expenditures by Mexican municipalities, regardless of type, reduce the technical efficiency of local production. Results support the Brennan–Buchanan collusion hypothesis that decentralization generates an increase in government spending, but it does not translate to higher social welfare.

JEL Classification

R1 R5 H7 C2 

1 Highlights

  • Public expenditure by municipalities reduces the technical efficiency of local production

  • Earmarked transfers have a bigger negative effect on technical efficiency than non-earmarked ones

  • Brennan–Buchanan collusion hypothesis holds true for Mexican municipalities

2 Introduction

Municipalities in Mexico are the closest level of government to the local economies within the federal system. The country has a centralized federal system, which distributes part of the tax collection to the subnational governments with non-earmarked and earmarked transfers while municipalities provide local public services and collect property taxes within their jurisdictions. Hence, it is quite important to understand how autonomous local public expenditure favors local industry development. Economically, this relationship could be analyzed by studying their technical efficiency, which allows measurement of the output gap given a certain amount of inputs (Farrel 1957). The government has direct and indirect influence over the exogenous factors that determine the economic efficiency, but public expenditure by itself does not influence the aggregate production (Adkins et al. 2002). For example, government can improve efficiency through public investments such as roads, well-designed institutions, and a legal framework that guarantees and facilitates business. There are several studies that analyze the municipal institutional design for Mexico and its coordination mechanisms with the State and Federal levels of government (García del Castillo 1995; Cabrero and Carrera 2000; Cabrero 2001), but none of them examine their relationship with the aggregate production within their jurisdictions.

Previous studies that have analyzed the fiscal system focus on the transparency of their processes (Cejudo and Gerhard 2010), on the effects of its design to achieve their objectives (Arellanes Ramirez 2011) or on poverty alleviation (Hernández Trillo 2016), but not on how efficient is their relationship with aggregate production. Most studies have focused in the institutional quality of the current efforts to decentralize the federal system, but few of them in its outputs. The importance of knowing this relationship lies in the fact that a design that incentivizes efficiency in local economies would generate positive spillovers, such as an improvement in wages. In other words, the institutional design of municipal finances could be a main determinant to boost regional development and deal with structural problems such as poverty.

To measure the effect of the institutional design in the local economies, this work controls the technical efficiency with two institutional variables of revenues following the theoretical statements of Shadbegian (1999) and assesses them using stochastic frontier models that are commonly used in recent literature. By measuring such efficiency, this work seeks to answer two questions simultaneously. The first is, what is the effect of municipal fiscal design of municipalities on the technical efficiency of aggregate production? Connected to this question, the second is, which are the determinants of technical efficiency and specifically, what effect does the municipal revenue structure have in this efficiency?

This work analyses a sample of 328 municipalities that belong to all metropolitan areas in the country. We have chosen these municipalities because they have a more developed administrative infrastructure. Consequently, they have more proficiency to take over new responsibilities granted by the federal government or to deliver tagged resources of the federal government with specific policy destinations.

This paper has six sections in total. Following this introduction, the next section makes a literature review and presents the contributions of this work. The third section describes the technical efficiency models. The fourth explains the dataset and the empirical strategy. Next, the fifth presents the results and the last section offers the conclusions and policy discussions.

3 Technical efficiency and the decentralization debate in Mexico

The Mexican state is a federal system composed of three levels of government: federation, states and municipalities. In practice, decision-making has been centralized in the federal government (Hernández 2008). Decentralization was introduced in the political agenda in the early eighties and municipalities were recognized as a government level in the late nineties.1 The debate about power distribution among government levels is still open and is part of the political agenda, so a short explanation of the legislation is useful here.

In 1980, the government created an allocation in the federal budget for states and municipalities called Ramo 28 (Peña Ahumada 2011), which aimed the simplification and coordination of the tax system. Under it, the collection of broad-base taxes was centralized in the federal government and then distributed in a compensatory process to the states following an established criterion by law. However, as subnational governments are sovereigns, they are not accountable to the federation for the use of these resources.

This reform was followed by two more to decentralize the education system in 1992 and the health system in 1996 (Peña Ahumada 2011). However, additional resources were not available to finance the new decentralized powers. To address this problem, the government allocated a new portion of earmarked transfers in 1997 called Ramo 33, which is composed by eight funds designed to provide financial support to the local governments to take over their new responsibilities (Guizar Jiménez 2004). However, as these new transfers are earmarked, municipalities must follow established criteria by the federation. In sum, new powers were decentralized, but not the budget control to exercise them.

In parallel with Ramo 33, in 1983 the government started a decentralization process through a series of constitutional reforms. The first reform gave municipalities the command of a set of local public services and the power to collect the property tax. Hence, revenues coming from either property tax collection or non-earmarked transfers could be freely spent by municipalities. Later, in 1999, there was another constitutional reform which gave municipalities the level of government instead of just local managers (Faya Viesca 2004). Even though it appeared to be a small reform, this change generated unprecedented policy implications for local governments.

As a new level of government, municipalities have the power to release regulations and to take over the command as a principal of local bureaucracies. Likewise, with this reform, municipalities are able to design and implement regional development programs, grant construction licenses, among other faculties to promote productivity, employment and entrepreneurship. With these faculties, local governments are now an agent of their citizens. Therefore, at first, they should be incentivized to improve their tax collection system so they could be able to respond to their population demands.

Within the broad public finance literature, few works have aimed to study the relationship between local government finances and the technical efficiency, and to the best of our knowledge there are no studies that focus on that relationship for the Mexican fiscal structure. Our study aims to involve public spending in a direct form and follows the work of Shadbegian (1999) to construct hypotheses about the effects of the fiscal structure in technical efficiency. Shadbegian (1999) presents his hypothesis from the spending point of view and tests the principles of Traditional Fiscal Federalism and Public Choice in the case of the USA. From the Fiscal Federalism perspective, the author makes use of the work of Wallis and Oates (1988). Following the decentralization theorem of Oates (1972), Wallis establishes that decentralization increases local government sizes because they have to attend several heterogeneous demands. Whereas from the Public Choice perspective, the author makes use of the collusion hypothesis of Brennan and Buchanan (1980), which argues that all levels of government would behave as a cartel in order to maximize their revenues, without improving public services or, in this case, the technical efficiency. Furthermore, in their comparative statistics approach, Neyapti and Bulut-Cevik (2014) argue that earmarked transfers and increments in the tax rate have a negative relationship with fiscal efficiency or overall welfare, formally demonstrating the failure of the decentralization theorem.

In this way, this work aims to test both hypotheses for the Mexican case considering the relationship between revenues and spending versus technical efficiency. Thus, the resources in which local governments have more power and the last decision (non-earmarked transfers and local tax collection) are considered as a proxy of Wallis–Oates hypothesis. While earmarked transfers constrain discretion with resources and are classified as a proxy by the Brennan and Buchanan hypothesis.

Inefficiency is measured using the output gap between the actual and potential output, given an amount of inputs (Farrel 1957). The potential production of a firm or, in this case, a municipality, could be shaped as a frontier (Álvarez Pinilla et al. 2003). There are two main ways to measure the production frontier: stochastic frontier models or nonparametric methods. In this paper, we will use the first one due to its flexibility to include explanatory variables. For the case of local government finances, previous works have constrained their scope by not including heterogeneity controls for the inefficiency term. Hence, the main contribution of this paper is to analyze the effect of local government finances on the technical efficiency by including local institutional variables (among others) on the inefficiency side. Additionally, a second contribution is to control for heteroscedasticity in the idiosyncratic error, which is usually ignored in this type of studies.

For the Mexican case, there are some sectorial and regional works of stochastic frontiers, but none of them is directly related to public expenditure (e.g., Bannister and Stolp 1995; Becerril-Torres et al. 2010; Braun and Cullmann 2011; Aguilar Gutiérrez 2011; Chávez and López Ornelas 2013) and few of them explore the determinants of the inefficiency. Hence, this research presents a third contribution to the empirical literature about the case of Mexico. That is to control for several variables that have an impact on technical efficiency, such as: public capital, labor capital, infrastructure and geographic controls, but specifically municipal expenditure variables.

4 Models for measuring technical efficiency

As mentioned previously, technical efficiency is measured with the output gap, given an amount of inputs. In their seminal work Aigner et al. (1977), proposed a model that divided the error term into two parts to model the stochastic component as the technical inefficiency. Pitt and Lee (1981) presented an extension for modeling panel data. However, their specification did not allow time variation for the stochastic term. To address this deficiency, Battese and Coelli (1995) proposed a model that overcomes the problem by allowing exogenous variables to explain the inefficiency term. Nevertheless, their model did not allow to control for heteroscedasticity in the error term, which could result in biased estimators as it overestimates the intercept and underestimates coefficients slope (Caudill and Ford 1993).

Reifschneider and Stevenson (1991) were the first in attending the heteroscedasticity problem by allowing variances in the mean of the error in one side. They assumed the inefficiency error as nonnegative and truncated to zero. Caudill et al. (1995) extended this model assuming multiplicative heteroscedasticity and incorporating it directly to the variance of the stochastic error (from now on, this model would be referred as CFG). Considering the previous, the model is shown in Eqs. (1) and (2).
$$ y_{it} = x_{it} \beta + v_{it} - u_{it} $$
(1)
$$ \sigma_{{u_{it} }}^{2} = \exp \left( {z_{it} \gamma } \right) $$
(2)
where yit represents the production for periods t = 1, 2, …, T for the observations \( i = 1,2, \ldots ,N \cdot x_{it} \) are the explanatory variables in the frontier. uit is the inefficiency term that follows a half-normal distribution, \( N^{ + } \left( {0, \sigma_{{u_{it} }}^{2} } \right) \) and zit are the control variables in the stochastic term. Finally, vit is the idiosyncratic error that follows a normal distribution \( N\left( {0, \sigma_{v}^{2} } \right) \).
Lastly, Hadri (1999) uses the CFG model for the inefficiency term uit and expands the specification to allow to control for heteroscedasticity in the idiosyncratic error vit, where the variance, \( \sigma_{{v_{it} }}^{2} \), follows an exponential distribution and is explained for a set of exogenous variables, hit, as in the inefficiency error, as shown in Eq. (3).
$$ v_{it } \sim N\left( {0, \sigma_{{v_{it} }}^{2} } \right),\quad \sigma_{{v_{it} }}^{2} = \exp \left( {h_{it} \theta } \right) $$
(3)
For this research, we will consider CFG (Eqs. 1 and 2) and Hadri (Eqs. 1, 2 and 3) models. Further research should be aimed to implement most recent stochastic frontier specifications such as Greene (2005) to add fixed effects to separate the variant inefficiency in each period from the unobserved heterogeneity that remains unchanged throughout time. Despite the flexibility of the later model, a dataset with at least ten-time periods is necessary for estimation due to the large number of parameters (Belotti and Ilardi 2012).

5 Data and the empirical model for fiscal structure and technical efficiency

We use a balanced panel dataset2 composed of 328 Mexican municipalities and four years of observations of the 59 metropolitan areas defined by the National Population Council (CONAPO) 2010 classification.3 The four-time periods correspond to the economic censuses taken by the National Statistics and Geography Institute (INEGI) for 1993, 1998, 2003 and 2008. The advantage of this span is that we have information for two periods before and two after the reform that granted autonomy to municipalities in 1999. As described in Sect. 2, before that Constitutional reform, municipalities were only administrative entities with some operational faculties. In contrast, after the reform, municipalities became a real level of government with decision-making power over their jurisdictions. Therefore, this analysis is aiming to assess the effect of these legal modifications in these two independent periods of the Mexican federal system.

Metropolitan areas are defined as the set of two or more municipalities with 50,000 or more inhabitants and with high level of socioeconomic concentration. They are composed of either central municipalities, municipalities defined by statistical and geographic criteria, or exterior municipalities defined following urban and political planning criteria. Municipalities in metropolitan areas were chosen as they are the local governments with more administrative infrastructure and that have more proficiency to take over new responsibilities granted by the federal government or to implement tagged resources of the federal government with policy destinations. All the metropolitan areas account for about 57% of the total population of the country (CONAPO 2010) and contribute 85% of domestic total value added (INEGI 2008). Descriptive analysis in the following section strengths the importance of choosing municipalities of metropolitan areas. Despite being just 13% of the total number of municipalities in 2009, they received 25% of total transfers (INEGI 2015) by then. Hence, this analysis has an impact on the majority of the Mexican population and its economy.

Regarding the variables of the model, firstly, we use as proxy of production (y), the total value added from the four principal economic sectors: manufactures, mining, trade and services in 2008 prices. Labor (l) represents the sum of total number of workers in the same four economic sectors. Similarly, capital (k) is the total fixed assets value in 2008 prices. All the information was gathered from the Economic Censuses of INEGI. Table 1 presents the summary statistics of these and the rest of variables. Value-added measurements give a mean of 573 million Mexican pesos (MXN). There are municipalities with zero value, but others reached MXN 26,000 million. Capital and labor display means of MXN 737 million and 23,719 workers and maximums value of MXN 26,600 million and 614,547 workers, respectively.
Table 1

Summary statistics

Variable

Mean

SD

Min

Max

VA (1000 MXN) (y)a

573,000

1,810,000

0

26,000,000

Number of workers (l)b

23,719

61,024

0

614,547

Total fixed assets (1000 MXN) (k)c

737,000

1,800,000

0

26,600,000

Average education (years) (educ)

7.83

1.51

3.10

13.50

Specialization index (specialization)

0.31

0.20

0.00

0.92

Autonomous

0.62

0.26

0.00

0.99

Earmarked

0.18

0.20

0.00

0.83

Distance to the border (km) (distance)

977.83

346.27

1.00

2315.64

Infrastructure (HDI) (infra)

0.85

0.05

0.59

0.93

Population density (density)

966.27

1855.90

2.10

19,357.83

Infrastructure km of roads + railroads (rr_infra)

11,747

5194

1860

22,900

aSeven municipalities have negative value-added values in different periods for two main reasons. The first one establishes that they are auxiliary jurisdictions in other words, that are places where firms have their warehouses, which represents costs, but not revenues. The second one suggests that important investments were implemented during those periods. In order to be specified as natural logarithms, their values were equaled to zero as shown in the summary statistics table

bSome municipalities registered zero workers as their most important economic sectors are not part of the four principal ones (manufactures, mining, trade and services) that are included in this estimation and their other workers are working in other municipalities

cThe municipalities with negative value added also registered the same problem with fixed assets. Natural logarithms were specified as well

Following Aschauer (1989), our model includes controls for variables for public and human capital, labor specialization and population density. Infrastructure (infra) is used as a proxy of public capital and it is constructed as the Human Development Index (HDI) published by INEGI (it includes electricity, piped water and drained).4 This variable does not include highways, airports and ports, as this kind of hard infrastructure is not within the municipal constitutional powers. This variable reports a standard deviation of 0.05 points, where the most disadvantaged municipality has an index of 0.592, whereas the best one has an index of 0.927 as shown in Table 1. In this sense, as an additional control, we include another infrastructure variable (rr_infra) that is composed of the addition of roads and railroads available in each state. The mean of this variable is 11,747 km and has an important standard deviation of 5194 km.

Education (educ), defined as the average years of education of the employed population, was used as a proxy of human capital (INEGI 2009). Summary statistics table shows that the minimum average value of education is 3.1 years, which is equal to incomplete elementary school. While the maximum average value is 12.1, which is equal to high school level. Labor specialization (specialization) and population density (density) complete the set of control variables. On one side, the specialization index is defined as the ratio between number of workers in the manufacturing sector over the total number of workers. On the other side, we included a variable that divides the population over the area of each jurisdiction in order to control for the municipality size. The mean density is 1856 inhabitants per square kilometer.

Finally, our main variables of interest, i.e., fiscal structure, were constructed following the power over the resource criteria. Non-earmarked transfers and municipalities’ own taxes are combined together so they could be spent in an autonomous way and are associated with Wallis–Oates hypothesis. On the contrary of earmarked transfers, which are closer to the Brennan and Buchanan perspective. Both variables are defined as in Eqs. (4) and (5), respectively.
$$ {\text{autonomous}} = \frac{{{\text{own}}\,{\text{taxes}} + {\text{non - earmarked transfers}}}}{{{\text{total}}\,{\text{revenues}}}} $$
(4)
$$ {\text{earmarked}} = \frac{{{\text{earmarked}}\,{\text{transfers}} }}{{{\text{total}}\,{\text{revenues}}}} $$
(5)
Additionally, two institutional variables that consider an origin criterion were included. As mentioned above, revenues can come from both the municipal tax collection and the federation, so this criterion divides the autonomous variable into two different variables. The first one divides own tax revenues over total revenues (r_taxes). The second one divides non-earmarked transfers over total revenues (r_nonear). These are incorporated together in some specifications in substitution of the autonomous variable. These institutional variables were constructed with annual data from the State and Municipal System of Data Bases (SIMBAD) from INEGI (2015). Figure 1 presents the evolution over time of earmarked and autonomous variables for the aggregated municipalities included in this study. It is possible to identify a significant reduction in municipal own revenues beginning in 1997 after Ramo 33 reform.
Fig. 1

Municipal revenues

Some endogeneity issues can arise between the institutional variables and local production. However, homogeneity among selected local governments minimizes reverse causality suspects because earmarked transfers were created once the economic features of the metropolitan areas were already established. Table 2 shows further that there is no strong linear correlation between the institutional variables and the other controls, including the main targets of the conditional transfer, namely, education and infrastructure. Similarly, population density and total number of workers are weakly correlated with the non-earmarked transfers. Additionally, residuals were tested for spatial autocorrelation using global Moran’s I test. Since the null hypothesis was rejected,5 a spatial lag of the dependent variable (spatial_lag) is added to the model as a covariate, aiming to address the potential spatial autocorrelation bias. The spatial lag was generated using an inverse distance weighted matrix.
Table 2

Correlations

 

y

l

k

edu

infra

density

spec

auton

earm

rr_infra

y

1

         

l

0.91

1

        

k

0.78

0.72

1

       

educ

0.33

0.33

0.39

1

      

infra

0.20

0.20

0.26

0.75

1

     

density

0.25

0.29

0.27

0.34

0.26

1

    

spec

0.10

0.04

0.05

− 0.07

0.00

0.07

1

   

auton

− 0.14

− 0.14

− 0.14

− 0.35

− 0.28

− 0.08

0.05

1

  

earm

0.02

0.03

0.03

0.23

0.16

0.00

− 0.07

− 0.78

1

 

rr_infra

− 0.03

− 0.06

− 0.02

0.02

− 0.15

0.04

− 0.12

− 0.21

0.20

1

Finally, we consider eight variables in the error terms. Six of them are the same as in the frontier: specialization, density, r_taxes, r_nonear, autonomous and earmarked. Additionally, we include two other explanatory variables. The first one is a trend (trend) that is a categorical variable that controls for technical change. The second one is a geographic variable (distance), which is defined as the number of kilometers between the center of each municipality and the nearest border point to the USA. This variable follows the new economic geography principles about the importance of the proximity among the units under analysis (Krugman and Lizas-Elizondo 1996). For its specification, we used the application Traza tu Ruta of the Ministry of Communications and Transportation (SCT 2015). We use a Cobb–Douglas production function (Cobb and Douglas 1928) for the empirical specification, resulting in the stochastic frontier model in Eq. (6).
$$ \begin{aligned} \ln \,y_{it} & = \beta_{0} + \gamma_{t} t_{t} + \theta_{i} r_{i} + \beta_{1} \ln k_{it} + \beta_{2} \ln \left( {edu_{it} *l_{it} } \right) + \beta_{4} \ln \,idh_{{services_{it} }} \\ & \quad + \beta_{5} density_{it} + \beta_{6} specialization_{it} + \beta_{7} \ln rr_{{infra_{it} }} \\ & \quad + \beta_{9} \ln \,spatial_{{lag_{it} }} + \beta_{10} r_{{taxes_{it} }} + \beta_{11} r_{{nonear_{it} }} + \beta_{12} earmarked_{it} + v_{it} \left( \cdot \right) - u_{it} \left( \cdot \right) \\ \end{aligned} $$
(6)
where i refers to municipalities and t for time. The variable y refers to production, k capital, l labor, xjit is a set of control variables, all mentioned above, while tt and ri are time and regional fixed effects,6 respectively. Finally, uit(·) is the term for measuring inefficiency and vit(·) is the idiosyncratic error, both error terms are referred as the stochastic part of the model as in CFG and Hadri models (Eqs. 2 and 3, respectively) and include the variables described previously.

6 Results

Two different specifications were estimated for the stochastic frontier models. The first one follows the origin criterion for the institutional variables, including both r_taxes and r_nonear as shown in Eq. (6). The second specification follows the spending independence criteria, so it uses the autonomous variable instead as specified in Eq. (4). Results of the frontier estimates are presented in Table 3. There is not a significant difference between CFG and Hadri models estimation results in the frontier.7 Likewise, the specification of institutional variables seems to not affect coefficients sign and magnitude, therefore we will describe them without emphasizing a model or specification.
Table 3

Stochastic frontier estimations for the natural logarithm of production (Frontier)

Model

CFG-1

Hadri-2

CFG-3

Hadri-4

Variables

Equation (6)

Equation (6) modified with (4)

Central

− 0.04

(0.09)

− 0.04

(0.09)

− 0.01

(0.09)

0.00

(0.08)

North

− 0.24*

(0.13)

− 0.26**

(0.13)

− 0.19

(0.13)

− 0.23*

(0.12)

South

− 0.19

(0.13)

− 0.22*

(0.13)

− 0.16

(0.13)

− 0.20*

(0.12)

1998

− 0.62***

(0.07)

− 0.62***

(0.08)

− 0.65***

(0.08)

− 0.60***

(0.08)

2003

− 0.13

(0.11)

− 0.12

(0.11)

− 0.16

(0.11)

− 0.03

(0.10)

2008

− 0.32**

(0.14)

− 0.27**

(0.14)

− 0.32***

(0.14)

− 0.07

(0.12)

k

0.09***

(0.01)

0.09***

(0.01)

0.09***

(0.01)

0.09***

(0.01)

l*educ

1.17***

(0.01)

1.16***

(0.01)

1.16***

(0.01)

1.15***

(0.01)

Infra

− 1.08

(0.68)

− 1.14*

(0.68)

− 1.20*

(0.71)

− 1.05

(0.70)

Density

− 0.00*

(0.00)

− 0.00***

(0.00)

− 0.00

(0.00)

− 0.00

(0.00)

Specialization

0.73***

(0.15)

0.48**

(0.20)

0.89***

(0.16)

− 0.49***

(0.17)

RR_infra

0.08*

(0.04)

0.08**

(0.04)

0.07*

(0.04)

0.08**

(0.04)

Spatial_lag

− 0.00

(0.04)

− 0.00

(0.04)

− 0.00

(0.04)

− 0.03

(0.04)

r_taxes

3.12***

(0.48)

3.05***

(0.48)

r_nonear

2.43***

(0.23)

2.41***

(0.26)

Autonomous

2.22***

(0.24)

1.47***

(0.21)

Earmarked

2.04***

(0.32)

2.01***

(0.33)

1.78***

(0.32)

1.20***

(0.27)

Constant

2.71**

(1.23)

2.98***

(1.23)

3.45***

(1.24)

4.53***

(1.19)

Observations

1312

1312

1312

1312

Log-likelihood

− 1813.76

− 1796.39

− 1830.19

− 1803.06

H0: constant returns to scale

   

Reject: β1 + β2 > 1

Models 1 and 2 include the institutional variables following the origin criteria. Meanwhile, the models 3 and 4 include the aggregate specifications of centralized and decentralized revenues

*p < 0.10; **p < 0.05; ***p < 0.01. Standard error in parenthesis

The border region is taken as the reference one in the regional fixed effects and as expected, all other regions present negative coefficients. Mexican border region has both a strong entrepreneurial culture and an advantageous proximity to the USA. Monterrey City and its eleven metropolitan municipalities are a notorious example of this situation. As the third most populated and the second largest metropolitan area in the country, it has multinational firms such as Cemex and FEMSA group, it has also a strong manufacturing sector and includes a municipality that reports the highest HDI in the country. The central region, where Mexico City is located, is the one with the closest levels to the reference one. Yet only the northern region shows a statistically significant coefficient at the 10% level. In the same way, the year 1993 is the reference point for the time fixed effects and all other time periods display a negative coefficient, being year 2003 the smallest one. This could be explained due to the Tequila crisis in 1995 affecting the 1998 period, and the economic recession of 2008–2009 due to the Financial crisis impacting the 2008 observation.

Coefficients for capital and labor are significant at a 1% level in all specifications and report positive expected signs. Likewise, null hypothesis of constant returns to scale is rejected. Municipal infrastructure reports a negative value, but it is weakly statistically significant at the 10% level. While this sign might appear to be counterintuitive, the results are supported by Duran-Fernández and Santos (2014). The authors explain that spillovers generated by domestic infrastructure have already been internalized by the industry. In contrast, State and Federal infrastructure (roads and railroads) present a positive and statistically significant effect in all models. Labor specialization also is statistically significant and presents the positive expected sign in most of the models. The population density coefficient is almost zero, but nonetheless statistically significant. Likewise, spatial lag variable reports a small negative effect but is not statistically significant across all specifications.

Results of the stochastic section are presented in Table 4. As explained in Sect. 3, CFG 1995 and Hadri (1999) models allow the inclusion of explanatory variables in the inefficiency term uit(·), where a positive (negative) coefficient represents higher (lower) inefficiency. Hadri (1999) model also allows for explanatory variables in the idiosyncratic error vit(·). However, unlike the inefficiency one, covariates included in this term do not have an explanatory mean, but they do improve the efficiency of the estimation. Labor specialization has a negative and statistically significant coefficient; therefore, higher specialization of the municipality would imply an efficiency increase. On the contrary, population density coefficient is positive and significant. Hence, denser municipalities would be more inefficient. Although this could be counterintuitive because denser territories represent lower public services costs, this might happen as bigger municipalities have enough territory to afford big industries depending on the size of their jurisdictions.
Table 4

Stochastic frontier estimations for the natural logarithm of production (error term)

Model

CFG-1

Hadri-2

CFG-3

Hadri-4

Equation (6)

Equation (6) modified with (4)

Inefficiency—Eq. (3)

 Specialization

− 1.01**

(0.42)

− 2.08***

(0.65)

− 0.57

(0.37)

− 8.21***

(0.93)

 Density

0.00***

(0.00)

0.00***

(0.00)

0.00***

(0.00)

0.00***

(0.00)

 r_taxes

14.17***

(1.63)

15.58***

(1.51)

 r_nonear

9.57***

(1.01)

10.40***

(0.94)

 Autonomous

7.53***

(0.81)

6.81***

(0.89)

 Earmarked

9.91***

(1.10)

10.62***

(1.03)

7.82***

(0.92)

7.06***

(0.94)

 Distance

− 0.00***

(0.00)

− 0.00***

(0.00)

− 0.00***

(0.00)

− 0.00***

(0.00)

 Trend

− 0.00

(0.11)

0.04

(0.11)

− 0.16

(0.10)

0.22*

(0.11)

 Constant

− 6.98***

(0.93)

− 7.47***

(0.91)

− 4.67***

(0.72)

− 3.53***

(0.84)

Idiosyncratic error—Eq. (4)

 Specialization

 

0.97***

(0.35)

 

2.15***

(0.30)

 Density

 

− 0.00***

(0.00)

 

− 0.00***

(0.00)

 r_taxes

 

− 0.89

(1.46)

 

 r_nonear

 

− 0.03

(0.64)

 

 Autonomous

 

 

2.06***

(0.46)

 Earmarked

 

− 0.15

(0.73)

 

1.74***

(0.56)

 Distance

 

0.00

(0.00)

 

0.00***

(0.00)

 Trend

 

− 0.02

(0.09)

 

− 0.23***

(0.08)

 Constant

0.61***

(0.02)

− 1.18*

(0.63)

0.58***

(0.02)

− 2.68***

(0.49)

Models 1 and 2 include the institutional variables following the origin criteria. Meanwhile, the models 3 and 4 include the aggregate specifications of centralized and decentralized revenues

*p < 0.10; **p < 0.05; ***p < 0.01. Standard error in parenthesis

The time trend coefficient presents a negative sign, which means that technological changes diminish technical inefficiency. However, the coefficient is only statistically significant in one of the specifications. The distance to the US border coefficient is almost zero, but negative and statistically significant, which means that the closer to the US border, the lower the inefficiency. This result is consistent with the region fixed effects where being in the border region represents an advantage over the others, and it is also consistent with the Duran-Fernández and Santos (2014) hypothesis on the access to the US markets, as these estimations already control for spatial spillover effects.

Regarding our central variables, results in Table 4 show that institutional variables do not contribute positively to technical efficiency. No matter whether earmarked or autonomous revenues present positive signs, government spending increases inefficiency in the aggregated local production. These results are consistent also in the divided specification of the autonomous variable following the origin criteria. Municipalities with better tax collection are the ones with larger coefficients. Likewise, both in the frontier and in the inefficiency side, all the specifications are statistically significant at 1% of confidence. Although institutional variables present positive signs in the frontier, this could represent that they have a positive contribution to production, but not to efficiency. Figure 2 shows the results of the efficiency ranking for CFG model for the four-time periods. Maps show us that northern areas of the country have higher efficiency levels. Hence, regardless the origin and the level of autonomy, we find that local government spending diminishes technical efficiency, which confirms the collusion hypothesis of Brennan and Buchanan (1980) within Public Choice literature and contributes to an empirical validation of the more recent Neyapti and Bulut-Cevik (2014) formal model.
Fig. 2

Maps of the technical efficiency rankings from 1994 to 2009

7 Discussion and conclusions

Throughout their spending, local governments increase technical inefficiency regardless of whether the revenues are earmarked or autonomous. This conclusion supports the collusion hypothesis established by Brennan–Buchanan within the Public Choice field. The hypothesis establishes that decentralization generates an increment in government spending, but it is not translated into population welfare, as all levels of government are colluded and behave as a cartel. The main assumption of hypothesis collusion establishes that public servants are selfish individuals that seek to maximize their utilities, even contrary to the welfare of society. In a framework with weak checks and balances at the municipal level, it is not surprising that the inefficiency coefficient is even bigger for most independent municipalities.

This problem has already been identified by the literature (Hernández Trillo and Jarillo Rabling 2007; Timmons and Broid 2013; Cejudo and Gerhard 2010). From a horizontal perspective, weak internal checks allow selfish public servants to maximize their utility functions instead of citizens’ welfare. Meanwhile, from a vertical perspective, as transfers pass through state governments, discretionality and opacity bias the distribution of resources (Timmons and Broid 2013). However, earmarked transfers have a bigger negative effect on technical efficiency than non-earmarked ones. This confirms another already-identified problem of perverse incentives created by the institutional design (Arellanes Ramirez 2011).

This research has used a quantitative model to analyze technical efficiency. Nevertheless, in order to propose structural policy solutions, it is necessary to strengthen these conclusions with a qualitative methodology for a deeper knowledge of municipal features. Likewise, conclusions are constrained to the municipal role on technical efficiency, so the municipal performance in other responsibilities is not considered. However, following a Public Choice perspective, regardless of the level of government, bureaucracies, politicians and power groups, the result will be the same. Therefore, it is necessary to create an institutional design that minimizes opportunities for discretion.

To sum up, there are different theoretical perspectives on the studies of fiscal relations in a federal system and its effect on technical efficiency. The literature for the Mexican case has focused on institutional design of the transfers and different biases in their distribution but has neglected its role in economic development. Finally, our results show that current fiscal design and spending do not have a positive effect in local economic performance. This problem has a direct negative effect on population welfare as lower technical efficiency levels are related to lower levels of investment and wages. As a result, regional inequality is perpetuated among municipalities of both the most efficient and inefficient metropolitan areas.

Footnotes

  1. 1.

    For the sake of simplicity, municipalities, local governments, and jurisdictions are used flexibly.

  2. 2.

    The dataset analyzed during the current study is available from the corresponding author upon request.

  3. 3.

    Some municipalities did not exist in our base period 1993, so 13 new municipalities were collapsed with their origin jurisdictions. Mexico City delegations have another fiscal institutional design, so they were excluded of the analysis. Also, other municipalities were excluded due to the lack of information, all of them belong to the state of Oaxaca.

  4. 4.

    It follows a normal HDI specification, but changes weights with the public services indexes.

  5. 5.

    Statistic Moran I’s: P value 0.000 at alpha 0.05.

  6. 6.

    We split the country in four regions: border, north, center and south.

  7. 7.

    Results of the idiosyncratic error for Hadri model are different in comparison with the inefficiency error.

Notes

Acknowledgements

We would like to thank Antonio Álvarez Pinilla and Fausto Hernández Trillo for their valuable comments.

Compliance with ethical standards

Conflict of interest

Hector M. Nuñez declares that he has no conflict of interest. Alejandro U. Becerra-Ornelas declares that he has no conflict of interest.

Human and animal rights

This article does not contain any studies with human participants or animals performed by any of the authors.

References

  1. Adkins LC, Moomaw RL, Savvides A (2002) Institutions, freedom, and technical efficiency. South Econ J 69:92–108CrossRefGoogle Scholar
  2. Aguilar Gutiérrez G (2011) Eficiencia industrial en las regiones de México. EconoQuantum 7(2):93–113CrossRefGoogle Scholar
  3. Aigner D, Lovell CAK, Schmidt P (1977) Formulation and estimation of stochastic frontier production function models. J Econom 6:21–37CrossRefGoogle Scholar
  4. Álvarez Pinilla A, Arias Sampedro C, Orea Sánchez L (2003) Introducción al Análisis Empírico de la Producción. Universidad de Oviedo and Universidad de León, EspañaGoogle Scholar
  5. Arellanes Ramirez A (2011) Una solución a la paradoja de las aportaciones federales para los municipios (Ramo 33 Municipal). CIDE: Tesis de para obtener el grado de Maestro en Administración y Políticas PúblicasGoogle Scholar
  6. Aschauer DA (1989) Does public capital crowd out private capital? J Monet Econ 24:171–188CrossRefGoogle Scholar
  7. Bannister GJ, Stolp C (1995) Regional concentration and efficiency in Mexican manufacturing. Eur J Oper Res 80:672–690CrossRefGoogle Scholar
  8. Battese G, Coelli T (1995) A model for technical inefficiency effects in a stochastic frontier production function for panel data. Empir Econ 20:325–332CrossRefGoogle Scholar
  9. Becerril-Torres O, Álvarez-Ayuso I, del Moral-Barrera L (2010) Eficiencia técnica de las entidades federativas de México. Economía, Sociedad y Territorio 10:485–511Google Scholar
  10. Belotti F, Ilardi G (2012) Consistent estimation of the “true” fixed-effects stochastic frontier model. CEIS Research Papers 231Google Scholar
  11. Braun FG, Cullmann A (2011) Regional differences of production and efficiency of Mexican manufacturing: an application of nested and stochastic frontier panel models. J Dev Areas 45:291–311CrossRefGoogle Scholar
  12. Brennan G, Buchanan JM (1980) The power to tax: analytical foundation of a fiscal constitution. Cambridge University Press, CambridgeGoogle Scholar
  13. Cabrero E (2001) La Experiencia Descentralizadora Reciente en México. Problemas y Dilemas. CIDE Working Papers División de Administración Pública 28Google Scholar
  14. Cabrero E, Carrera A (2000) Fiscal decentralisation and institutional constraints. Paradoxes of the Mexican Case. CIDE Working Papers División de Administración Pública 85Google Scholar
  15. Caudill SB, Ford JM (1993) Biases in frontier estimation due to heteroscedasticity. Econ Lett 41:17–20CrossRefGoogle Scholar
  16. Caudill SB, Ford JM, Gropper DM (1995) Frontier estimation and firm-specific inefficiency measures in the presence of heteroscedasticity. J Bus Econ Stat 13:105–111Google Scholar
  17. Cejudo G, Gerhard R (2010) La Rendición de Cuentas de Transferencias Intergubernamentales. El Caso del FAIS. UNAM: Instituto de Investigaciones JurídicasGoogle Scholar
  18. Chávez JC, López Ornelas LF (2013) Un enfoque no paramétrico para la descomposición de la productividad del trabajo en la industria manufacturera regional. Working Papers, Banco de México 2013-08Google Scholar
  19. Cobb C, Douglas P (1928) A theory of production. Am Econ Rev 18:139–172Google Scholar
  20. Duran-Fernández R, Santos G (2014) Road Infrastructure spillovers on the manufacturing sector in Mexico. Res Transp Econ 46:17–29CrossRefGoogle Scholar
  21. Farrel MJ (1957) The measurement of productive efficiency. J R Stat Soc Ser A 120:253–290CrossRefGoogle Scholar
  22. Faya Viesca J (2004) El Federalismo Mexicano. Régimen Constitucional del Sistema Federal. Porrúa, MéxicoGoogle Scholar
  23. García del Castillo R (1995) Análisis del municipio mexicano: diagnóstico y perspectivas. Partes I y II. CIDE Working Papers División de Administración Pública 32Google Scholar
  24. Greene W (2005) Reconsidering heterogeneity in panel data estimators of the stochastic frontier model. J Econom 126:269–303CrossRefGoogle Scholar
  25. Guizar Jiménez J (2004) Ramo 33: Fondo de Aportaciones Federales (Aspectos Normativos, Operativos y Financieros del Gasto Federalizado). Instituto para el Desarrollo Técnico de las Haciendas PúblicasGoogle Scholar
  26. Hadri K (1999) Estimation of a doubly heteroscedastic stochastic frontier cost function. J Bus Econ Stat 17:359–363Google Scholar
  27. Hernández R (2008) El centro dividido: la nueva autonomía de los gobernadores. El Colegio de MéxicoGoogle Scholar
  28. Hernández Trillo F (2016) Poverty alleviation in federal systems: the case of México. World Dev 87:204–214CrossRefGoogle Scholar
  29. Hernández Trillo F, Jarillo Rabling B (2007) Transferencias condicionadas federales en países en desarrollo: el caso del FISM en México. Estudios Económicos 22:143–184Google Scholar
  30. INEGI (2008) Censos Económicos 1994, 1999, 2004 y 2009 Instituto Nacional de Estadística y Geografía. http://www.inegi.org.mx/est/contenidos/Proyectos/ce/Default.aspx. Accessed 26 June 2017
  31. INEGI (2009) Censos de Población y Vivienda 1990, 1995, 2000, 2005, 2010 Instituto Nacional de Estadística y Geografía. http://www.inegi.org.mx/est/contenidos/proyectos/ccpv/default.aspx. Accessed 16 June 2017
  32. INEGI (2015) Sistema Estatal y Municipal de Bases de Datos. Instituto Nacional de Estadística y Geografía. http://sc.inegi.org.mx/cobdem/contenido.jsp?rf=false&solicitud. Accessed 1 May 2017
  33. Krugman P, Lizas-Elizondo R (1996) Trade policy and the Third World metropolis. J Dev Econ 49:137–150CrossRefGoogle Scholar
  34. Neyapti Bilin, Bulut-Cevik ZeynepBurcu (2014) Fiscal efficiency, redistribution and welfare. Econ Model 41:375–382CrossRefGoogle Scholar
  35. Oates WE (1972) Fiscal Federalism. Harcourt Brace Jovanovich, NYGoogle Scholar
  36. Peña Ahumada JA (2011) Manual de Transferencias Federales para Municipios. Secretaría de Gobernación, Instituto Nacional para el Federalismo y el Desarrollo MunicipalGoogle Scholar
  37. Pitt MM, Lee LF (1981) The measurement and sources of technical inefficiency in the Indonesian weaving industry. J Dev Econ 9:43–64CrossRefGoogle Scholar
  38. Reifschneider D, Stevenson R (1991) Systematic departures from the frontier: a framework for the analysis of firm inefficiency. Int Econ Rev 32:715–723CrossRefGoogle Scholar
  39. SCT (2015) Traza tu Ruta. Secretaría de Comunicaciones y Transportes. http://aplicaciones4.sct.gob.mx/sibuac_internet/ControllerUI?action=cmdEscogeRuta. Accessed 15 July 2017
  40. Shadbegian R (1999) Fiscal federalism, collusion, and government size: evidence from the states. Public Finance Rev 27:262–281CrossRefGoogle Scholar
  41. Timmons J, Broid D (2013) The political economy of municipal transfers: evidence from Mexico. Publius J Fed 43:551–579CrossRefGoogle Scholar
  42. Wallis JJ, Oates W (1988) Does economic sclerosis set in with age? An empirical study of the Olson hypothesis. Kyklos 41:397–417CrossRefGoogle Scholar

Copyright information

© Springer-Verlag GmbH Germany, part of Springer Nature 2019

Authors and Affiliations

  1. 1.Pardee RAND Graduate SchoolSanta MonicaUSA
  2. 2.Department of EconomicsCentro de Investigación y Docencia EconómicasAguascalientesMexico

Personalised recommendations