Spatial interdependence and spillovers of fiscal grants in Benin: Static and dynamic diffusions

This paper investigates the spatial diffusion of an intergovernmental grant in Benin. Using static and dynamic spatial models, we estimate the spillover effects of the Fonds d’Appui au Développement des Communes (FADeC) on per capita local government expenditure in the 77 municipalities from 2008 to 2015. Neighborliness – a measure of interdependence – is captured through geographic and distance-based spatial weighting schemes. In addition, we constructed a measurement of ethnic afﬁnity as an alternative spatial weighting scheme to test for the existence of an ethno-spatial interdependence in local public ﬁnance in Benin. The empirical results suggest that a statistically signiﬁcant share of the total effects of the FADeC stems from indirect elasticities or the diffusion process of grants received by neighboring juris-dictions, regardless of how we measure neighborliness. The results also conﬁrm the existence of a robust ethno-spatial interdependence and complementarity in local government expenditure in Benin. The spil-lovers across ethnic neighbors are estimated to be 13.9% of the total effects in the short-run and 15.5% in the long-run. Put differently, the effects of the FADeC in a given municipality are inﬂuenced by the transfers received by its ethnic (and linguistic) neighbors. The ﬁndings point to the appeal of inter-governmental transfers for the decentralized ﬁnancing of public services, especially in low-income countries where local bureaucratic capacity in raising own-source revenues might be limited. Supporting local governments with well-structured grants can not only be a channel to foster local public provision but also contribute to pushing geographic or ethnic neighboring localities to increase their own spending and generate positive spillovers that are crucial for wholistic regional development. (cid:1) 2022 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses


Introduction
Decentralization and devolution of the public sector have been at the center stage of national policy agendas in many African countries in the past decades. 1Despite this trend, sub-national governments continue to rely on top-down fiscal transfers to finance local public goods and services.According to the World Observatory on Subnational Finance and Investment, fiscal grants and subsidies represent, on average, 61% of sub-national governments' revenues in Sub-Saharan Africa, while taxes stand at only 20.9% (OECD & UCLG, 2019).The scarcity of fiscal resources in low-income economies and the steady relevance of fiscal grants in the financing of public services motivate continuing empirical inquiries into the design and underlying mechanisms of such financing schemes and their effects on socio-economic and political outcomes.Moreover, studying the spillovers of fiscal transfers is of utmost importance for understanding and reaping the benefits of positive externalities across jurisdictions.
This paper investigates a fiscal grant's static and dynamic spatial diffusions in Benin, a low-income country in West Africa.More specifically, we estimate the total, direct, and spillover effects of the Fonds d'Appui au Développement des Communes (hereafter FADeC) on per capita local government expenditure in the 77 municipalities from 2008 to 2015.The FADeC constitutes the largest pool of fiscal transfers in Benin to date.It accounted for 50% of total local government revenue in 2016.The pool includes a recurrent grant disbursed to finance operational expenditure of local governments (such as salaries and intermediate consumption) and an investment grant which finances public goods and services within respective local jurisdictions.This paper focuses on the latter.Since 2011, the FADeC finances approximately 80% of local spending on public goods and services, including roads, educational services, and other responsibilities assigned to local authorities.In 2015, the FADeC investment grant alone represented 39% of the total local government revenue, a ratio that outweighs that of tax revenue (24%) in that same year.Nevertheless, there remains a lack of (or no) empirical research on its contribution to local socio-economic dynamics and outcomes.
The contributions of the paper are two-fold.First, using spatial static and dynamic diffusion models, we provide prime evidence of the total, direct, and spillover effects of the FADeC grant on per capita local government expenditure over the period of 2008 to 2015.
To date, the literature on the incidence of fiscal transfers is limited in its consideration of the spatial spillovers of such financing mechanisms.By highlighting the indirect effects of the fiscal grant, this paper shows that a significant share of the effects can be attributed to indirect elasticities (i.e., transfers to neighboring jurisdictions), thus channeled through spatial interactions among municipalities in Benin.
Second, we introduce a measure of neighborliness based on the ethnic affinity of the municipalities, thereby testing the relevance of an ethno-spatial interdependence in local fiscal policies.According to Alesina et al. (2003)'s research, Benin is the 18th most ethnically fragmented country in the world.The existing literature also suggests that ethnic structure drives political affiliation and preferences for public goods in Benin (Wantchekon & Vermeersch, 2011;Wantchekon, 2003;Battle & Seely, 2010).We therefore postulate that municipalities that share the same ethnic structure or the same dominant ethnic group might have close or similar preferences in demands for public goods and services, which could, in turn, foster local government spending strategies.If we assume that individuals from a given ethnic group living in different municipalities carry semi-homogeneous preferences, then local authorities might well mimic the policies of their 'ethnic' neighbors, and the theoretical reasoning put forward by the spatial dependence literature would be valid in this context.Therefore, we consider an ethnic-based spatial weighting scheme (augmented by a linguistic-spatial weight) to test whether ethnic affinity is reflected in the interdependence in local fiscal policies and whether the fiscal transfers spillover to ethnic (or linguistic) neighbors.To the best of our knowledge, this paper is the first to consider ethnic affinity in the study of spatial interdependence in local public finance in Benin. 2he empirical estimates are derived using static and dynamic Spatial Durbin models (Lee & Yu, 2010a;Debarsy et al., 2012;Elhorst, 2012), in which we control for the heterogeneity of municipalities, the spatial interdependence in per capita local government spending, and local economic overflows.With the dynamic diffusion model, we control for temporal persistence in per capita local spending and account for global common factors.We adopt standard econometric test procedures, such as the CD-test by Pesaran (2015a) and the a-test of Bailey, Kapetanios and Pesaran (2016), to test for weak against strong cross-sectional dependence or for strong cross-sectional dependence in models with and without the common factors.These procedures, performed with insights from the most recent literature on spatial panel models with common factors (see for e.g.Ciccarelli & Elhorst, 2018;Elhorst, 2021), confirm the reliability of our estimates.
The results point to a positive spatial diffusion of the FADeC grant and its multiplier effects on per capita local public spending.In other terms, a large share of the total effects results not from direct elasticities but from the diffusion process of FADeC transfers received by neighboring jurisdictions, regardless of how we measure neighborliness.After accounting for the time persistence in per capita local government expenditure in the dynamic models, it is estimated that 9-15% of the total effects of the grant in the short-run and 10-17% in the long-run are due to such spillovers.The findings also confirm the existence of an ethno-spatial interdependence in local public finance.The ethno-spatial spillovers are estimated to be 13.9% in the short-run and 15.5% in the long-run.Strictly speaking, the effects of the fiscal grant in a given municipality are influenced by the transfers received by its ethnic neighbors (and its linguistic neighbors: those with dialects of similar roots).Overall, the paper shows that the FADeC carries a strong multiplier effect.From a policy-standpoint, such evidence points to the appeal of intergovernmental transfers as a means of financing local public services in low-income countries where local bureaucratic capacity in raising own-source revenues might be limited.Supporting disadvantaged local governments with wellstructured and targeted grants can not only be a channel to foster local public provision but can also contribute to pushing geographic or ethnic neighboring localities to further up their own spending.The resulting simultaneous spending can facilitate a wholistic regional development while reducing regional inequality.
In what follows, Section 2 reviews the existing literature and establishes the theoretical basis of the paper.Section 3 describes Benin's institutional context and the local public finance structure.Section 4 details the empirical framework, including the data sources, the estimation approaches, and a description of the spatial weighting schemes.Section 5 reports and discusses the empirical results, while concluding remarks are captured in Section 6.

Background literature
To highlight the spatial diffusion and multiplier effects of fiscal grants in Benin, this paper intersects two strands of literature.The first relates to the political economy of intergovernmental transfers, their underlying rationales, and the existing evidence on the effects of such transfers on socio-economic and policy outcomes.The second is centered around the spatial interdependence in fiscal policies.We discuss both strands and highlight their synthesis, which lays the ground for assessing the extent to which the effects of the FADeC grant on per capita local government spending are influenced by the transfers to neighboring municipalities.

Political economy of intergovernmental grants
In most countries, the devolution of taxing powers and of public spending to lower-tier governments occurs at a different pace, resulting in a vertical fiscal imbalance in local government budgets.Without external financial support, lower-tier authorities tend to be incapacitated in fulfilling their mandates, especially on expenditure in poverty-sensitive areas such as primary education or housing.Through fiscal grants and subsidies, central governments ensure that sub-national authorities meet their spending requirements (Wallis, 1996;Holcombe & Zardkoohi, 1981;Smart, 2007).Besides correcting the vertical fiscal imbalance, such transfers are designed and executed with other underlying rationales, such as lessening regional disparities, economic stabilization, and fostering positive externalities.
Revenue mobilization in a jurisdiction relies on its socioeconomic characteristics and resource endowments.Disparities in revenues, rooted in available resources, are not only bound to create efficiency and equity concerns in public provision but can also be conflict-ridden (Boadway & Flatters, 1982;Shankar & Shah, 2003).Fiscal grants are thus often conceived with an equalization mechanism to correct regional disparities.This mechanism ensures that minimum standards are met in all jurisdictions and in critical public policy areas.Transfers are also designed to internalize externalities (see for e.g.Dahlby, 1996).Neighboring jurisdictions benefit from services for which they do not bear any cost.The imperfect mismatch between the spatial dimension of the locality that provides a good and the areas that benefit from it is likely to give rise to free incentives, causing the provision of such a good to fall below the Pareto-efficient level.
Fiscal transfers, especially matching grants, often aim to incentivize local spending in areas that yield positive externalities (Pigouvian subsidies). 3When the economy is flopping, intergovernmental transfers can also act as economic stabilizers (Rodden & Wibbels, 2010); they can simulate local activities which might positively affect overall economic performance.In many ways, grant allocation constitutes an easier alternative to changes in tax rates, which often involve political negotiations in a period of economic stress.If inter-regional migration does not lead to equilibrium and an increase in national welfare, grants might also be a preventive mechanism for controlling the outward migration of the regional labor force.Beyond the economic rationales, it has often been argued that political influence and affiliation are the most concrete underlying rationales for intergovernmental transfers (see Boex & Martinez-Vazquez, 2004 for further discussion).According to Grossman (1994), party affiliation and the size of the political majority in the parliament are strong determinants of an increase in fiscal transfers.Biswas and Marjit (2005) and Rao and Singh (2005) also show that political compromise and lobbying efforts further impact the distribution of intergovernmental transfers in India.It is thus implied that subnational governments with political powers can expect to receive a larger share of intergovernmental transfers than their counterparts with lesser political influence.
On the ramifications of intergovernmental grants, academics have generally studied two types of effects: an income effect, which implies that the grant increases local resources, and a price or substitution effect, which infers that the grant reduces the marginal cost of an additional unit of the funded good while resources increase (see Shah, 2007Shah, , 1994;;Inman, 2008).Pioneer research on the effects of fiscal grants has predominantly focused on the income side by looking at the spending side of local governments' budgets.They paved the way for the ''flypaper effect" literature through which authors have demonstrated that a dollar increase in a general-purpose grant increases spending by more than what a dollar would increase in residents' income (Hines & Thaler, 1995;Bailey & Connolly, 1998;Inman, 2008).Since the late 1990s, a wave of empirical studies has focused on the substitution effect by highlighting the disincentivizing effects of grants.Many have portrayed fiscal transfers as windfall resources that distort local fiscal capacity, exacerbate moral hazard and engender rent-seeking behaviors (Baretti et al., 2002;Büttner & Wildasin, 2006;Mogues & Benin, 2012).However, these findings are inconclusive as alternative results point to a crowd-in-effect of fiscal transfers on subnational revenue generation (Dahlberg et al., 2008;Caldeira, Emilie & Rota-Graziosi, Gregoire, 2014).According to Smart (2007Smart ( , 1998)), different intergovernmental grants yield different effects on local revenue and expenditure, and the ultimate impact may depend on other institutional parameters and contexts.Still, while the ramifications of intergovernmental grants have been extensively studied, a review of the empirical literature suggests that it has often ignored the spatial interdependence structure among local jurisdictions and under-investigated the multiplier effects of such transfers, which are channeled through such interdependence.Spatial interactions and interdependence among jurisdictions -be it geographical or otherwise -imply that the effects of fiscal grants on socio-economic or political outcomes not only result from direct elasticities but may also be affected by the diffusion process of the same transfers to neighboring jurisdictions.

Spatial interdependence in local fiscal policies
In recent decades, researchers in economics and political sciences have investigated the spatial interaction and interdependence of territorial units.In this line of work, it is postulated or demonstrated that local governments' spending or tax policy decisions are driven not only by their own determinants -such as income, socio-demographic or political characteristics -but also by the decisions and characteristics of neighbors.
The spatial interdependence in fiscal policies is explained through several hypotheses.For instance, Besley and Case (1995) propose the yardstick competition theory, according to which voters benchmark the performance of their region against their neighbors.As a result, elected governments will attempt to stay in power by mimicking the policies of such neighbors.A second hypothesis is grounded on the theory of proximity and contagion and the externalities of public provision.First, goods and services financed by one jurisdiction may be enjoyed by residents in neighboring jurisdictions, resulting in free riding in competition to reduce welfare benefits whose costs are not borne by neighboring authorities (Saavedra, 2000).Second, an economic shock or investment in a specific jurisdiction is not just felt by its neighbors but also alters their policies due to local spillovers, mobility, and migration.For instance, as the labor force commutes, a publicly financed manufacturing plant may create a demand for public investment in housing in neighboring jurisdictions that contribute to the labor force.As a result, towns strategically decide their level of expenditure based on their expectations of neighboring policies and plans.The political economy literature has also suggested that strategic interactions in taxes and expenditure among local governments can be explained by political interactions, namely the political ideology of neighboring jurisdictions or the political affiliation of local authorities (Foucault et al., 2008;Solé Ollé, 2003).
Decades of empirical evidence concerning different country contexts have corroborated the theoretical perspectives on spatial interdependence in fiscal policies.4Among others, Case et al. (1993) found that states' per capita expenditure is positively and significantly impacted by their neighbors' spending patterns.Similar outcomes are reported in Baicker (2005).Foucault et al. (2008), using panel data on French municipalities covering 1983-2002, suggest that spending interactions exist between municipalities that have the same political affiliation.To date, however, empirical evidence related to low-income countries is limited -a gap that can be partly attributed to the lack of coherent data on local governments' finance and limited decision-making at the subnational level, especially on 3 According to Shah (2007), specific-purpose open-ended matching grants are the most appropriate for increasing spending on assisted function; however, if governments are results-oriented, then specific-purpose non-matching transfers are preferable to other types of transfers as they respect local autonomy while providing incentives to improve service delivery.tax matters. 5Studies on spatial interactions have thus so far focused on the spending side (see for e.g.Caldeira et al., 2015;Arif et al., 2019), where local governments have greater discretion in decisions.

Synthesizing the two strands
The alignment of these two strands of literature serves two purposes.First, while a positive effect of the FADeC grant on local government expenditure is expected (in view of the existing literature on the conceptual impact of fiscal transfers) 6 , it remains very much under-researched, albeit it being the most important financing source of local services in Benin in the past years (see section 3.2).Second, under the assumption of strategic interactions and spatial interdependence in local government spending in Benin, it is worth investigating the spatial diffusion of the grant and testing its multiplier effects -i.e., whether the incidence of the grant can be attributed to spillovers from neighboring jurisdictions.As argued above, the literature on the effects of grants has often left aside the interdependence component to focus on direct elasticities.In pure empirical terms, a failure to account for such interdependence results in an omitted variable bias which could mislead conclusions on the real impact of such financing mechanisms.Therefore, aligning these two strands sets the framework within which we estimate the direct and indirect effects of the FADeC grant on per capita local government spending in the 77 municipalities of Benin.This paper also moves beyond a geography-based diffusion to analyze the relevance of an ethno-spatial interdependence in local government spending and the diffusion process of the grant through the ethnic affinity of the municipalities.While it is empirically challenging to consider socio-economic measures in spatial weighting techniques (see Getis, 2009;Fiaschi & Fiaschi, 2015;Bouayad-Agha et al., 2018), ethnicity and ethnic structure are shown to be critical drivers of public demands and policy preferences in Benin (see for e.g.Wantchekon, 2003;Wantchekon & Vermeersch, 2011;Battle & Seely, 2010).Therefore, we explore the existence of ethno-spatial interactions and assess whether such ethnic interdependence contribute to influencing local government fiscal policies within and across jurisdictions.

Institutional context
Benin is a French-speaking West-African country bordered by Nigeria on the East, Togo on the West, and Burkina Faso and Niger on the North (see Fig. A.1).The country remains one of the least developed economies with a per capita GDP of only USD (PPP) 3,505.4 and ranked 158 out of 189 in the Human Development Index in 2020 (UNDP, 2020;World Bank Group, 2020).Benin is an ethnically fragmented country with more than 42 tribes.Data from the recent national census point to an average of six different ethnic representations within each municipality. 7The fractionalization indicator of Alesina et al. (2003) also ranks Benin as the 18th most ethnically fragmented country in the world.The 77 municipalities vary significantly in socioeconomic and demographic profiles.In 2015, the population density ranged from 9.63 persons per km 2 in the municipality of Kari-Mama to 12,505.5 persons per km 2 in the economic capital of Cotonou.While the headcount poverty ratio was 64% in the municipality of Copargo in 2015, it only stood at only 22% in Savalou (INSAE, 2015).

The legal and institutional framework of decentralization in Benin
Decentralization has been one of Benin's major economic reforms of the last decades.The reform has brought several institutional changes to the political arena and territorial management since the first municipal elections of 2003.The municipalities are the unique and fully decentralized level in Benin.They were born out of the prefectures that existed before 2003 and are regrouped into twelve regions (départements) and subdivided into districts (arrondissements) and villages (MDGLAAT, 2010b;Thomas, 2013). 8The municipalities are governed by a council, elected every five years, whereas representatives of the central government oversee the regional council.
The Constitution of 1990 establishes, in its articles 150-153, the basic legal framework for decentralized governance.It confers the rights to sub-national entities to be autonomous.However, the functioning of local governments was held back throughout the 1990s.Several decrees were nevertheless enacted and came in corroboration of the constitutional provisions.For instance, in 1993, the government developed the first strategies for local development (see details in Thomas, 2013).In 1999, the parliament ratified five legal texts, establishing the 77 municipalities and the 12 regional governments (MDGLAAT, 2010b).The series of legal texts also outline the competencies to be transferred from central to local governments.Law N o 98-007, among others, sketches the legal directives on the financial management of the municipalities; Law N o 98-006 guides the electoral process, and Law N o 97-029 elaborates on the administrative governance of the municipal councils.
The constitutional provisions, legal texts on decentralization, and other strategies for local development became effective in 2003 with the first municipal elections and the establishment of the first elected municipal councils.Central authorities thus began the process of transferring technical and fiscal resources to local authorities.Several policy formulations on territorial governance and decentralization followed, including the policy document on territorial management (DEPONAT) adopted in July 2006. 9The DEPONAT introduced a series of norms and tools for municipal planning and incorporated the national vision for decentralization over fifteen years (Thomas, 2013).It also established some rules regulating intergovernmental relations, notably on transferring public financial resources from central to local governments through the state-territory program for development.Following the DEPONAT, the country adopted, in 2008, the first national decentralization and de-concentration policy document, also known as the PONADEC, which further defines the attributions of different government tiers and details the programs and policies to be implemented to foster good governance at the local level. 10

Local public finance and the FADeC grant
In the aftermath of the first municipal elections in 2003, central planning was predominant as local governments' own-source revenues were relatively low.Important municipal projects were financed through grant-contracts established between the central and local governments, which took into consideration each municipality's socio-economic and demographic characteristics (Thomas, 2013;MDGLAAT, 2010a).In 2007, following the adoption of the DEPONAT, the government introduced the Fonds d'Appui au Développement des Communes (FADeC), a budgetary mechanism that gathers all donations and financial resources to be transferred 6 See Smart (2007) for an overview on the conceptual impact of intergovernmental grants.
8 MDGLAAT stands for Ministère de la Décenralisation, de la Gouvernance Locale, de l'Administration et de l'Aménagement du Territoire.The FADeC is the largest pool of fiscal transfers in Benin and the most significant financing source of local public goods and services since its first operationalization in 2008.The fiscal transfers from the FADeC are divided into a general-purpose grant, which funds recurrent expenditures of local governments, such as salaries and intermediate consumption, and an investment grant disbursed to finance local public goods and services in respective municipalities.
The FADeC investment grant comprises an earmarked grant and a formula-based non-matching grant.The earmarked investment grant targets specific public policy areas such as health and education and is channeled through respective ministries and state institutions.However, audit reports by the National Commission on Local Finance (CONAFIL) indicate that the earmarked grant represents a tiny share (less than 5%) of the total FADeC investment transfers (CONAFIL, 2017).In contrast to the matching fund, the formula-based grant is at the discretion of the local governments, provided that the funds finance local public goods and services.The allocation formula for the non-matching grant includes socio-economic indicators: area, population size, and poverty level, and second, the administrative and financial performance of the municipalities in preceding fiscal years.These criteria are described in the Manuel de procédures du Fonds d'Appui au Développement des Communes (FADeC) (MDGLAAT, 2015(MDGLAAT, , 2008)).

The municipal administrative and financial performance of the local governments is assessed by the National Commission on Local Finance (CONAFIL), whereas socio-economic indicators are provided by the National Institute of Statistics (INSAE).
The share of the FADeC investment grant in total local public revenues has significantly increased over the years.Since 2012, it alone had surpassed all other revenue sources of local governments and stood at an average of 39% in 2015 (see Fig. 1).While in the early years of the decentralization reform, local public goods and services were primarily financed through residual funds from operating revenues, the FADeC investment grant has substituted all other financing sources in recent years.Fig. 2 suggests that the grant accounts for more than 70% of funding allocated to public investment spending and even reached 91% in 2013.The prominence of the FADeC in a low-income country like Benin, where fiscal resources are scarce, calls for empirical inquiries into its design and effects on local socio-economic development and political outcomes.

Data sources and description
This paper benefits from a comprehensive panel dataset on local public finance and characteristics of the 77 municipalities in Benin covering the period of 2003 to 2015.The empirical estimations with the FADeC grant cover the period of 2008 to 2015 (i.e., from its institution in 2008).The local public finance data are provided by the National Commission on Local Finances (CONAFIL) and encompass information on (i) local government revenue streams (taxes, fees and charges, grants, and donations) as well as revenues per budgetary function; (ii) expenditure, including items such as wages and salaries, and spending in schools, and investment spending in infrastructure works, sanitary networks, roads and equipment, and other local public services; (iii) and municipal savings and debt repayment.To the local public finance data, we adjoin geo-coded data on climatic and socio-economic characteristics of respective municipalities.This latter is compiled and made available by AidData using countries' official geographical and administrative boundaries of territorial units (Tierney et al.,   2011; Goodman et al., 2019). 11To limit the issues of endogeneity and bi-directional causality, the covariates of the models are restricted to primarily geographical and climatic conditions.These include the mean annual temperature, the mean annual precipitation, and the annual value of the normalized difference in vegetation which is created using vegetation data from NASA's long-term data records.These variables allow us to control for features that may explain variations in municipal government policy decisions over time.The description of these variables and their summary statistics are provided in Table A.1 and Table A.2, respectively.Fig. 3 provides an overview of the spatial distribution of per capita local government expenditure in 2008 (the year when the FADeC was implemented) and 2014.It suggests that local spending per inhabitant has increased significantly across all municipalities.Per capita spending is the highest in the economic capital of Cotonou, both in 2008 and 2014.In 2014, municipal expenditure was also very high in Tanguieta, Toucountouna, and Save.These were close to the level of per capita spending in major cities in Southern Benin.Fig. 4 highlights that the amount of the FADeC grant has increased across municipalities, although there are some variations in the quantile distribution when comparing 2008 to 2014. 12The literature on the conceptual impact of fiscal grants points to a potential positive effect of the grant on municipal spending (Shah, 2007;Gamkhar & Shah, 2007;Smart, 2007).The FADeC falls under the taxonomy of block grants that provide broad support while local authorities retain discretion.This grant type generally stands in a grey area between general-purpose and specific-purpose non-matching transfer.As argued by Smart (2007), they are often preferable to other types of transfers as they respect local autonomy while providing incentives to improve service delivery.While a positive effect is expected, the potential magnitude of such an increase and the likely spillover of the grant both in the short and long terms remain unaddressed in the context of Benin.

Modeling strategy: from static to dynamic panel estimations
We begin with a spatial lag panel model in which local government per capita expenditure in municipality i is driven by a set of observed characteristics and the level of per capita expenditure in neighboring jurisdictions.Second, we extend it into an unconstrained Spatial Durbin Model by introducing the spatial lags of some covariates, namely population density, vegetation index, and climatic characteristics (precipitation and temperature).The local-spatially variables allow us to control for the effects of geographical overflow related to inter-municipal migration (Kelejian & Robinson, 1997) and local economic overflows, especially those related to agricultural production.According to the Food and Agriculture Organization (FAO), 80% of the Beninese population earns a living in agriculture. 13Mainly, cotton is an essential contributor to the country's exports and represents approximately 35% of total export revenues.Located in a tropical region, Benin is also often subject to floods and other climate-related stress.It is thus expected that changes in climate conditions in adjacent areas would spillover locally by, for instance, shifting the labor force, influencing the availability of work or public capital in neighboring municipalities, and impacting public provisions, agricultural production, and other economic activities. 1411 Data from the AidData platform have been widely used in recent publications in economic and political science (see for e.g.Blair & Roessler, 2021;Eichenauer et al., 2021), especially with respect to low-income and middle income countries for which data access remains a challenge. 12See Figure A.2 in the Appendix.Given that 95% of the total amount is based on objective criteria (MDGLAAT, 2008(MDGLAAT, , 2015)), the changes in certain municipalities' quantile position suggest that those criteria are not static but change over time. 13See https://www.fao.org/agriculture/ippm/projects/benin/en/. 14It is also noted that most empirical studies with spatial panel also include localspatially lagged independent variables.See for instance Elhorst (2014a), Debarsy et al. (2012) and Ciccarelli and Elhorst (2018) for a more recent application.Following LeSage and Pace (2009), the unconstrained Spatial Durbin Model (SDM) can be written as in Equation ( 1), where i ¼ 1; Á Á Á ; N refers to the municipalities and i ¼ 1; Á Á Á ; T the time period.l i denotes the municipal fixed effects which capture time-invariant confounders such as the communes' background conditions, size, initial endowments, administrative status, and geographical locations; and t t the time fixed effects which implicitly account for year-based shocks or events that affect all communes in a given year (e.g., such as economic downturn).The fixed effects parameters (and common factors in dynamic estimations, see section below) account for heterogeneity across the municipalities and reduce spatial error dependence which arises through spatial auto-correlation of omitted variables.
The parameter q indicates the spatial interdependence of per capita local government spending across jurisdictions.If positive and statistically significant, it suggests mimicking patterns in spending policies.In contrast, if it is negative and significant, it points to diverging patterns in local municipal spending.w ij refers to the spatial weighting matrix of dimensionN Â N. The per capita local government expenditure of municipality i is also a function of x it , a 1 Â k vector of exogenous explanatory variables, and z jt a 1 Â k 1 subset of x it , which are spatially lagged as we control for local spillovers k 1 < k ð Þ .b and h are the vector of coefficient parameters to be estimated.e it is the error term that is assumed to be independent and identically distributed e it $ N 0; r 2 e it .
To simply, equation ( 1) can be written in a stacked form where Y t is the N Â 1 vector of the dependent variable at time t; X t denotes the N Â k matrix of exogenous explanatory variables at time t, Z t is the N Â k 1 matrix of a subset of X t and which contains the variables that are spatially lagged; W N is the N Â N nonstochastic weighting matrix; l is the N Â 1 vector of municipal fixed effects and e t the error term.

ð1:1Þ
We use the maximum likelihood function approach developed by Lee and Yu (2010b) to estimate the model.The interpretation of the coefficients is not straightforward due to the spatial dependence, as pointed out in related publications (see for e.g.LeSage, 2008;LeSage & Pace, 2013).Any change in one explanatory factor for one observation can affect the dependent variable for all other observations.Hence, the parameters must be interpreted in terms of direct, indirect, and total effects, as described in Elhorst (2014a) and LeSage and Pace (2021).
Moving beyond the static analysis, we extend Eq (1.1) and estimate a Spatial Dynamic Durbin Panel Model.First, we introduce a time-lagged variable for per capita local government expenditure -Y tÀ1 with its corresponding coefficient s as we argue that local governments' budgets are time-dependent, and there is persistence in per capita local government expenditure (Lee & Yu, 2010a;Elhorst, 2012;Bouayad-Agha et al., 2018;Debarsy et al., 2012).It is noted that spatial dynamic panel estimations have become prominent in recent years, notably in urban economics and regional sciences literature.Lee and Yu (2010b) and Elhorst (2014b) provide an extensive summary of the literature and insights into the interpretation of coefficient estimates, which is not straightforward as they do not represent the marginal effects of the covariates.
Besides including the time-recursive parameter, we introduce a common factor to equation (1.1) in lieu of the time fixed-effects parameter (see for e.g.Ciccarelli & Elhorst, 2018;Elhorst, 2021). 15The common factor is intended to control for strong cross-sectional dependence.Although the FADeC allocation formula primarily takes into account socio-demographic measures such as poverty and population and the fiscal performance of the municipalities --as described in the guidelines and audit reports (MDGLAAT, 2008(MDGLAAT, , 2015;;CONAFIL, 2017), the grant amount transferred to each municipality relies on the availability of financial resources at the national level.We suggest a model that allows one period to take effect and reflects the level of information asymmetries across tiers of governments.The common factor is defined as the weighted population average of per capita expenditure across the 77 communes in the sample at time t À 1 as: . Using common factors in panel estimations (non-spatial static and dynamic models) is attributed to Pesaran (2006Pesaran ( , 2015b;;Pesaran et al., 2013).The author argues that these averages may be treated as exogenous under the assumption that the added value of each unit to the crosssectional average is null as the number of units (N) goes to infinity.The extension of common factors to the spatial econometrics literature was pioneered by Bailey, Holly and Pesaran (2016), and subsequently adopted in various empirical applications (Ciccarelli & Elhorst, 2018;Halleck Vega & Elhorst, 2016;Ertur & Musolesi, 2017;Elhorst et al., 2020).
In this paper, we follow Elhorst (2021)'s suggested strategy in filtering out the effectiveness of added common factors.As suggested by the author, we recur to the cross-sectional dependence test (CD) developed by Pesaran (2015a) to test whether the models are free of any additional cross-sectional dependence.The null hypothesis of the CD test is that the residuals are only weakly cross-sectionally dependent (as opposed to strongly crosssectionally dependent), also better known as the local spatial dependence.As Ciccarelli and Elhorst (2018) argued, one key advantage of the CD-test is that it is not based on any specific spatial weight matrices.Following this latter publication, we also recur to the exponent a-test of Bailey, Holly and Pesaran (2016), which allows testing for weak against strong cross-sectional dependence in estimations without common factors, and for weak cross-sectional dependence after the inclusion of the common factors.This a-component indicates the degree of cross-sectional dependence, i.e., the rate at which the average pair-wise correlation coefficient measured over all N units varies with N when this latter goes to infinity.If a ¼ 1, it indicates strong spatial and the need to account for common factors.Any value of a between 0 and 1 points to moderate cross-sectional dependence with a ¼ 3 4 considered as a turning point.If a falls within the range of 3 4 ; 1 Â Á , the average correlation coefficient tends to go to zero very slowly and points to the presence of common factors.If within the range of 0; 1 2 Â Á -which we aim for -the average correlation coefficient tends to go to zero very fast, pointing to local spatial dependence comparable to spatial arrangements in a binary contiguity matrix. 16Lastly, we assess the stationarity of the dependent variable in estimations with the common factor in two ways.First, we use the stability test based on the restriction that the sum of the temporal and spatial lags is lower than 1 to test whether the per capita local government expenditure is stationary over time and space based on the common factor included in the model.The p-value from testing this relationship s þ q À 1 < 0 is reported in the result tables. 15The results tables provide test results that corroborate the selection of this approach. 16See details on the a test in Bailey, Holly and Pesaran (2016), notably in the appendix.

Spatial weighting schemes and neighborliness
Defining neighborliness is the first step in testing spatial interactions and policy interdependences across jurisdictions.Examples of spatial weighting schemes include geographical contiguity (Case et al., 1993;Baicker, 2005), demographic characteristics such as the percentage of the black population (Case et al., 1993), interstate mobility (Figlio et al., 1999), or trade flows between countries (Aten, 1996).Unfortunately, the literature on the most appropriate spatial weighting scheme is nonunanimous.Many critics have suggested that the results are sensitive to the choice of matrices (Bell & Bockstael, 2000;Getis, 2009;Harris et al., 2011).Notwithstanding, research by LeSage and Pace (2009) suggests that the estimated effects are robust to the chosen weights, even more so than the coefficients, which are, in many cases, not directly interpretable.As the spatial econometrics literature has yet to indicate the ideal approach to choosing the most appropriate weighting matrix, our empirical strategy combines five different measures of neighborliness or interconnectedness based on geography or the ethno-cultural patterns of the communes. 17

Geography-based neighborliness
The first three spatial weighting schemes are based on geographical contiguity and the Euclidean distance between the municipalities.First, we consider the most straightforward and prominent spatial weighting scheme based on geographical contiguity, also known as queen contiguity.Two jurisdictions are contiguous if they share a common border such that w ij ¼ 1 or w ij ¼ 0 if otherwise.Second, we consider a weighting scheme based on the Euclidean distance between two municipalities: the Delaunay triangulation (also known as Thiessen polygons, proximity polygons, or Voronoi diagrams).This geometric method connects points into triangles such that the number of triangles is maximized.The Delaunay triangulation is often regarded as the sphere of influence of one particular unit (Bellefon et al., 2018) and has previously been used in spatial economic research (see for e.g.Anselin & Le Gallo, 2006).Third, we select the five kernel nearest neighbors of each municipality.While the neighboring list is not always symmetrical, the nearest neighbors' schemes provide additional robustness to our analysis.

Ethnic neighborliness
The fourth spatial weighting scheme is based on the ethnic contiguity matrix of the municipalities and is constructed with data from the ethno-linguistic map of the African continent by Meur and Felix (2001).Two municipalities are ethnically contiguous if they carry the same predominant ethnic group.Fig. 5 displays the ethnic affinity of the municipalities.It must be noted that, in some cases, different (dominant) ethnic groups are from the same ethnic lineage (family) but have been disaggregated throughout history.As displayed in Fig. 5, the dominant ethnic groups in Benin also tend to be geographically close, although there are a few exceptions and a few communes without any ethnic neighbors.
Our consideration for ethnic neighborliness is based on existing research.First, ethnic fragmentation has been established as an important driver of decentralization and demands for public services in hinterlands (Easterly & Levine, 1997;Alesina et al., 1999;Panizza, 1999;Arzaghi & Henderson, 2005).Second, it is established that ethnicity and ethnic background shape public demands and policy preferences in Benin.For instance, Battle and Seely (2010) find that the concentration of a particular ethnic group in a region enhances the effect of ethnicity on political preferences.
Research findings from field experiments by Wantchekon (2003) and Wantchekon and Vermeersch (2011) also suggest that the demand for public goods is higher among voters from the same ethnic tribe as the dominant candidate.They argue that sharing an ethnic affiliation with political candidates may have rendered residents more receptive to public goods messages as they believe that their communities would be granted the opportunity to monitor the implementation of public services delivery and have their voices heard in national development planning.
Given the relevance of ethnicity and ethnic profiles, it is reasonable to postulate that municipalities that share the same ethnic structure or the same dominant ethnic group might have close or similar preferences in public goods and services, which could, in turn, foster local government spending strategies.If we assume that individuals from the same ethnic group living in different municipalities carry homogeneous preferences, mobility across municipalities that share the same dominant ethnic structure might be frequent and, to some extent, convenient.In such a case, local authorities might well mimic the policies of their 'ethnic' neighbors, and the theoretical reasoning put forward by the spatial interdependence literature would be valid in this context.Two municipalities are thus considered 'ethnically contiguous' if they host the same predominant ethnic group such that ew ij ¼ 1 or ew ij ¼ 0 if otherwise.In addition to the ethnic contiguity, we also explore the relevance of linguistic contiguity in robustness checks with data on the linguistic patterns of the communes.Two municipalities are considered linguistically contiguous if their local languages are from the same linguistic family, according to information from Meur and Felix (2001).

Spatial weight objects
Table 1 presents the summary statistics of the spatial weight objects.It shows a significant variation in the number of ties and observations with the most or the least connected objects.The Moran Statistics are the primer method for statistically testing spatial interdependence.Using the Kronecker product, the weighting matrices in the panel data structure are taken as I T bW N , a ðNT; NTÞ dimension matrix with N being the number of observations and T the time parameter.Using the spatial weights described in this sub-section, the Moran Statistics with 1000 permutations corroborate the existence of spatial interdependence in per capita local government expenditure in Benin.As noted in the last rows, such interdependence also exists when we measure neighborliness through the ethnic contiguity and the linguistic contiguity matrices of the communes.

Static spatial estimates
As described in subsection 4.1, we estimate our baseline static model as an unconstrained Spatial Durbin Model.The results are reported in Table 2.The parameter q is positive and statistically significant in all specifications, suggesting complementarity in local government spending regardless of the chosen geographical or Euclidean distance-based spatial weighting schemes.The coefficient estimates also point to significant local spillovers of population density and the vegetation index, thus corroborating our expectations on local demographic and economic overflows across municipalities.
While the magnitude varies with the spatial weighting scheme, the different columns point to a considerable increase in per capita local government expenditure.Every percent increase in the FADeC grant is associated with a 0.3% increase in per capita local government expenditure.However, more than 22% of that total increase is 17 Both the static and dynamic diffusion models are estimated using the package statistically due to spatial spillovers or the indirect effects of the grant received by neighboring municipalities.The magnitude of the spillover effect is estimated at 22.4% based on the binary geographical contiguity matrix (queen), 26.4% based on the Delaunay triangulation graph, and 29.7% based on the 5-kernel nearest neighbors.As discussed in subsection 4.2, it is expected that coefficient estimates are sensitive to spatial weighting matrices (see for e.g.Harris et al., 2011;Corrado & Fingleton, 2012;Bell & Bockstael, 2000).However, as pointed out by LeSage andPace (2009, 2014), coefficient estimates may differ due to differences in the scaling of spatial weighting schemes but the estimated effects and inferences are robust to the chosen weights if models are specified correctly.It is noted that the coefficient estimates and inferences are consistent despite differences in scale and construction approach of the spatial weighting schemes.
We perform two robustness checks on these baseline results.First, we re-estimated the models using only municipal fixed effects to test the sensitivity of the estimates to the time parame-ter, as we consider a cross-section average in the dynamic estimates.The results, reported in columns (1.1) to (3.1) of Table A.4, are consistent with previous findings.Second, we also test the robustness of the results by controlling for the nighttime light density as a proxy for local economic activity.As for population density, the nighttime light density is also spatially lagged to account for local economy spillovers.The use of nighttime light density as a proxy for local economic activity has been prominent in recent years, especially in contexts where data access is limited at the subnational level (Bruederle & Hodler, 2018;Mellander et al., 2015;Hodler & Raschky, 2014;Henderson et al., 2012).However, the DPMS-OLS nightlight density estimate is only available for 2008-2013, reducing our sample.Nonetheless, the estimates confirm the robustness of the results even when controlling further for economic activity at the municipality levels and local spatial spillovers of such activities.They point to the spatial dependence in local government expenditure ðqÞ, corroborate the positive and significant effect of the grant on per capita local government  expenditure, and confirm the multiplier effect of the grant as a significant share of that total effect is due to spatial spillovers across jurisdictions (see Table A.4).18

Dynamic spatial estimates
Moving beyond the static analysis, we estimate a dynamic panel model to account for the persistence in per capita local government expenditure through time.As described in section 4.1, we introduce a time-lagged version of the outcome variable (Lee & Yu, 2010a;Elhorst, 2012;Bouayad-Agha et al., 2018).Then, we test for cross-sectional dependence using the CD-test of Pesaran (2015a) and the exponent a-test of Bailey, Holly and Pesaran (2016), with insights from Elhorst (2021) and Ciccarelli and Elhorst (2018).Panel A and B of Table 3 report the statistical tests of estimations with and without the common factors for all specifications.In Panel B, the results of the CD-test on every specification without common factors are highly statistically significant, indicating that cross-sectional dependence needs to be accounted for in the dynamic models.The exponent a-test also surpasses the threshold of ¾ (i.e., a > 3 4 ) as stipulated in subsection 4.2, which indicates strong cross-sectional dependence.
When we introduce the common factor, namely the average local government per capita expenditure at time t À 1, and apply the CD-test to the models' residuals, we no longer find evidence favoring cross-sectional dependence.As we fail to reject the null hypothesis, the CD-test confirms that the residuals are only weakly cross-sectionally dependent, better known as the local spatial dependence, as opposed to strong cross-sectional dependence.Furthermore, the a-test points to values belonging to the range of 0; 1 2 Â Á , indicating that the average correlation coefficient tends to go to zero very fast, thus pointing to local spatial dependence comparable to spatial arrangements in a contiguity matrix.Furthermore, the standard errors of the exponent-a also drop significantly.With these two test results, we confidently confirm that both weak and strong cross-sectional dependence have been tackled.Regarding stationarity of the dependent variable, it is also shown that the sum of the spatial and temporal coefficients is smaller than 1, also confirmed by the p-value of the t-test, suggesting thereby that the hypothesis of stationarity is not rejected in favor of unit root (see for e.g.Ciccarelli & Elhorst, 2018).
Table 3 also reports the estimated short-term and long-term effects of the grant using the dynamic approach summed up in Equation (2).As to the static model, we recur to the existing literature for interpreting the direct, indirect, and total effects of the variable of interest.Table 3 shows that the FADeC grant positively affects per capita local government expenditure.Having controlled for the time-persistence in local spending, the results indicate that, for every percent increase in the grant amount, per capita local expenditure increases by 0.41% in the short-run and 0.46% in the long run.
The dynamic spatial estimates corroborate the multiplier effect of the grant, as illustrated by the statistical significance of the coefficient estimates for the indirect effects -both in the short and long run.In short-run, approximately 9% to 16% of the total effects are due to spillovers, whereas in the long-run, the ratio ranges between 10% and 17%, depending on the spatial weight object.The results are plausible; not only is there spatial interaction and complementarity in local public spending, but the grant also results in a multiplier effect and boosts local per capita expenditure further.The coefficient estimates for the common factor variable point to a diverging trend in per capita local government spending.In other terms, the higher the global average of per capita expenditure in the preceding fiscal year, the lesser the level of spending undertaken by local governments in any given fiscal year.

The relevance of ethnic affinity and contiguity
Table 4 reports our estimates testing the ethno-spatial interdependence in local public finance in Benin.As described in section 3, Benin is an ethnically fragmented nation, and previous research has established that ethnic affinity shapes political and public goods preferences in the country (Wantchekon, 2003;Battle & Seely, 2010;Wantchekon & Vermeersch, 2011).Therefore, we postulate that municipalities that share the dominant ethnic group would display similar patterns in public spending and that mobility across ethnically affiliated municipalities might be frequent and, to some extent, convenient.Hence, in their responses to public demands, local authorities might try to mimic the spending patterns of their 'ethnic' neighbors.
We derive the ethnic spatial weighting scheme from the ethnic contiguity matrix that indicates whether two municipalities are hosts to the same dominant ethnic group, with data from the ethno-linguistic map of the African continent (Meur & Felix, 2001).As there are some minor variations from ethnic to linguistic patterns, we also test for the existence of spatial-linguistic interactions among the municipalities in Benin using the same data source.In doing so, we assess whether the statistically significant multiplier effect of the grant holds with the spatial weight object measures linguistic neighborliness.
The results are reported in Table 4.In line with the previous tables, Columns (1) and ( 2) respectively report the coefficient estimates of the static and dynamic models.Consistent with previous findings, the magnitude of the indirect effects, or the ethno-spatial spillovers of the grant, is estimated to be 29.8% of the total effects in the static model -similar to the coefficient estimates using the 5-kernel nearest neighbors (as in column 3 of Table 2).Column (2) reports the coefficient estimates for the dynamic models with common factors.The magnitude of the ethno-spatial spillovers is estimated to be 13.9% in the short-run and 15.5% in the long-run.Similar to Table 3 on the dynamic spatial estimates, Panel B displays the outcomes of statistical tests for estimations without common factors.It is indicated that both strong and weak crosssectional dependence are to be addressed.Once we introduce the common factor, the results align with previous estimates.The CD-test of Pesaran (2015a) and the a-test of Bailey, Kapetanios and Pesaran (2016) confirm that weak and strong cross-sectional dependence have been properly addressed in the model.It is also noted that a falls within the range of 0; 1 2 Â Á and with smaller standard errors.Also, similar to previous reports, it is shown that the sum of the spatial and temporal coefficients is smaller than 1, suggesting thereby that the hypothesis of stationarity is not rejected in favor of unit root.
In Table A.3 in the appendix, we also report estimates using the linguistic contiguity of the municipalities and its resulting spatial weighting scheme to test the interactions among jurisdictions with similar linguistic roots and the spillover of the grant that might be therefrom derived.Similar to the ethno-spatial spillover, the coefficient estimates confirm the existence of spatial interactions among linguistic neighbors.Moreover, the results also corroborate that the grant indirectly and positively affects local spending in linguistic neighbors, with a spillover estimated at 11.5% and 12.9% of the total effects, respectively, in the short and long run.
In summary, the paper finds a positive effect of the FADeC grant on per capita local government spending of receiving jurisdictions.Most importantly, it evidences that a statistically significant share of that outcome does not result from direct elasticities but is embedded in spatial interactions and transfers to neighboring jurisdictions, regardless of whether neighborliness is measured in terms of geographical distance or ethno-linguistic affinity.

Concluding remarks
This paper investigates the total, direct, and spillover effects of a fiscal grant on per capita local government expenditure in Benin from 2008 to 2015.Local (municipal) governments in Benin are primarily financed through fiscal transfers, and the FADeC grant studied in this paper accounts for more than 70% of all financing of local public services since 2012 and 25% of all local government revenues since 2012.Our empirical strategy combines static and dynamic spatial diffusion models with different weighting schemes.First, we measure neighborliness through geographical contiguity (queen), the Euclidean distance between two municipalities, and the k-nearest neighbors with insights from the spatial economics literature.Second, we consider the ethnic contiguity of the municipalities as an alternative spatial weighting scheme.Drawing from previous research on the relevance of ethnic affinity in Benin, we argued that municipalities with the same ethnic structure or dominant ethnic group would have similar patterns in pub- lic demands.Thus local authorities, in their policy responses, might try to mimic the spending patterns of their 'ethnic' neighbors or strategically define their spending accordingly.Moreover, as ethnic affinity is often in line with linguistic affinity, we also test the relevance of a linguistic-spatial interdependence among the jurisdictions, and test whether the total effect of the grant to municipality i is influenced by the dynamics of that same grant received by its ethnic or linguistic neighbors.Using a Maximum-Likelihood approach, the static analysis confirms the existence of a spatial interdependence in per capita municipal expenditure and a total positive effect of the grant on local government spending per capita.After accounting for the time persistence in per capita local government spending in a dynamic diffusion model, the magnitude of the spillover effects of the grant is estimated between 9-15% in the short-run and 10-17% in the long-run, with variations linked to the spatial weight objects.The findings also confirm the existence of an ethno-spatial interdependence in local government spending.The ethno-spatial spillovers in the dynamic diffusion process are estimated to be 13.9% of the total effects in the short-run and 15.5% in the long-run.Our analysis also reveals that the grant statistically spillovers across linguistic neighbors.The linguistic-spatial spillover's magnitude is estimated at 11.5% in the short-run and 12.9% in the long-run.Our results prove to be consistent across all estimations and across all the different spatial weighting schemes that capture neighborliness.The findings of this paper align with the literature suggesting that local governments respond to the policy decisions of their neighbors.The literature also denotes that an economic shock or investment in a specific jurisdiction is not just felt by neighboring jurisdictions but also alters their policies.Our evidence shows that the FADeC carries a strong multiplier effect throughout the 77 municipalities.We show that the fiscal grant not only statistically increases per capita local government spending but a higher level of spending in a jurisdiction has excellent potential to encourage local authorities in neighboring towns to spend further.Such evidence points to the appealing nature of intergovernmental transfers as a mechanism for financing local public services, especially in least-developed economies.While supporting disadvantaged local governments with targeted grants helps them improve their development, it pushes neighboring local governments to further up their own development expenditure.The resulting simultaneous spending can facilitate a wholistic regional development while reducing regional inequality.Our findings thereby contribute to the broader debate on locally provided public goods by showing that, through spatial interactions, finances channeled to lower-tier governments could help to foster local development and create positive spillovers.

Fig. 2 .
Fig. 2. Financing Sources of Local Public Goods and Services between 2003 and 2015.Notes: This figure reports the financing sources of local public goods and services (budgetary allocation) between 2003 and 2015.Prior to 2008, local public provision was primarily financed through residual funds from operating revenues and loancontracts from the central government and, to a minor extent, donations from international organizations.Since the introduction of the FADeC in 2018, the investment grant has become the most significant financing source of local public goods and services.There are limited or no funding transfers from operating revenues, and most donations from international organizations are now channeled through the FADeC mechanism.Data Source: National Commission on Local Finance (CONAFIL).

Fig. 1 .
Fig. 1.Revenue Structure of Local Governments in Benin between 2008 and 2015.Notes: This figure conveys the revenue structure of the municipalities between 2008 and 2015.It compares the shares of tax and non-tax revenue (fees and charges), the FADeC grants, and other revenue in total local government revenues.Other revenues include donations from international organizations, shared revenues from mines and quarries, and the road tax.The figure shows that the share of the FADeC investment grant has increased significantly since 2012.Data Source: National Commission on Local Finance (CONAFIL).

Fig. 3 .
Fig. 3. Per Capita Local Government Expenditure in 2008 and 2014 (in CFA).Notes: This figure shows the spatial distribution of local government per capita expenditure in 2008 and 2014.It suggests that per capita expenditure has increased across the board in the 77 municipalities.Data Source: National Commission on Local Finance (CONAFIL).

Fig. 4 .
Fig. 4. FADeC Investment Grant in 2008 and 2014 (in million CFA).Notes: This figure shows the spatial distribution of the FADeC investment grant in million CFA francs in 2008 and 2014, respectively.It suggests that the amount transferred to the municipalities has increased across the board over the years.However, there are some minor variations in the quantile distribution, as suggested in Fig. A.2. Data Source: National Commission on Local Finance (CONAFIL).

Fig. 5 .
Fig. 5. Ethnic affinity of Beninese Municipalities.Notes: This figure shows the ethnic affinity of Beninese municipalities.Each color refers to the predominant ethnic group in each jurisdiction.The contiguity matrix that is therefrom derived is used to calculate the spatial weight object measuring ethnic neighborliness.

Fig. A1 .
Fig. A1.Geographical Location of Benin.Notes: This figure shows the geographical location of Benin in West Africa.Source: Nations Online Project.

Fig. A2 .
Fig. A2.Quantile Distribution of the FADeC Investment Grant in 2008 and 2014 (in million CFA).Notes: This figure shows the quantile distribution of the FADeC investment grant in million CFA francs in 2008 and 2014, respectively.It suggests that there are minor variations in the ranking of the recipients, suggesting thereby that some of the criteria in the grant allocation formula may have changed over time for some municipalities.Data Source: National Commission on Local Finance (CONAFIL).

Fig. A3 .
Fig. A3.Geographical and Euclidean distance-based neighborliness.Notes: This figure shows the connectedness of Beninese municipalities based on geographical and Euclidean distance-based neighborliness.Source: Authors with shapefiles from GADM.

Table 1
Summary of Spatial Weight Objects.This table summarizes the spatial weight objects.The Moran Statistics point to the existence of spatial interdependence in per capita local government expenditure, regardless of the weighting scheme.

Table 2
(LeSage & Pace, 2009)ce and Spillovers of Fiscal Grants in Benin Static Diffusion Models.This table presents the coefficient estimates from static spatial models assessing the total, direct, and spillover effects of the FADeC grant on per capita local government expenditure in Benin.Spatial interactions are captured through geographical spatial weight objects.The estimated effects of the covariates (temperature, vegetation, precipitation, population density) are estimated but not reported.The estimations follow the bias-corrected maximum likelihood approach byYu et al. (2008).Standard errors are computed using the Monte-Carlo approach with 1000 replications(LeSage & Pace, 2009).Standard errors are clustered at the municipal level.Significance level: * p < 0.10, ** p < 0.05, ***p < 0.01.

Table 3
(LeSage & Pace, 2009)ce and Spillovers of Fiscal Grants in Benin Dynamic Diffusion Models with Common Factors.This table presents the coefficient estimates from models assessing the total, direct, and spillover effects of the FADeC grant on per capita local government expenditure in Benin using dynamic spatial diffusion models.Spatial interactions are captured through geographical spatial weight objects.The estimated effects of the covariates (temperature, vegetation, precipitation, population density) are estimated but not reported.The estimations follow the bias-corrected maximum likelihood approach byYu et al. (2008).Standard errors are computed using the Monte-Carlo approach with 1000 replications(LeSage & Pace, 2009).

Table 4
(LeSage & Pace, 2009)ce and Spillovers of Fiscal grants in Benin Relevance of Ethnic Affinity (Spatial Weight: Ethnic Contiguity).This table presents the coefficient estimates from models assessing the total, direct, and spillover effects of the FADeC grant on per capita local government expenditure in Benin.Spatial interactions are captured through a measurement of ethnic neighborliness, derived from the ethnic contiguity matrix of the municipalities with data fromMeur and Felix (2001).The estimated effects of the covariates (temperature, vegetation, precipitation, population density) are estimated but not reported.The estimations in dynamic models follow the bias-corrected maximum likelihood approach byYu et al. (2008).The models are estimated using the Monte-Carlo approach with 1000 replications(LeSage & Pace, 2009).Standard errors are clustered at the municipal level.Significance level: * p < 0.10, ** p < 0.05, ***p < 0.01.
(LeSage & Pace, 2009)he summary statistics (in nominal and log values) of the variables used in the empirical estimations.Aside from the grant and per capita local government expenditure, all control variables are geographic or climate-related, thereby limiting issues of endogeneity and reverse0causality bias in the estimations.Spatial Interdependence and Spillovers of Fiscal grants in Benin Relevance of Linguistic Affinity (Spatial Weight: Linguistic Contiguity).This table presents the coefficient estimates from models assessing the total, direct, and spillover effects of the FADeC grant on per capita local government expenditure in Benin.Spatial interactions are captured through a measurement of linguistic neighborliness, derived from the linguistic family contiguity matrix of the municipalities with data fromMeur and Felix (2001).The estimated effects of the covariates (temperature, vegetation, precipitation, population density) are estimated but not reported.The estimations follow the bias-corrected maximum likelihood approach byYu et al. (2008).Standard errors are computed using the Monte-Carlo approach with 1000 replications(LeSage & Pace, 2009).Standard errors are clustered at the municipal level.Significance level: * p < 0.10, ** p < 0.05, ***p < 0.01.A4 Spatial Interdependence and Spillovers of Fiscal Grants in Benin Static Diffusion Models -Sensitivity Checks.This table presents the coefficient estimates from static spatial models assessing the total, direct, and spillover effects of the FADeC grant on per capita local government expenditure in Benin.Specifications from columns (1.1.)to(3.1)test the sensitivity of the results of Table2to the exclusion of time-fixed effects.Columns (2.2) to (3.2) include the nighttime light density for the years 2008-2013 (data availability) as an additional control for local economic activity.Spatial interactions are captured through geographical spatial weight objects.The estimated effects of the covariates (temperature, vegetation, precipitation, population density) are estimated but not reported.The estimations follow the bias-corrected maximum likelihood approach byYu et al. (2008).Standard errors are computed using the Monte-Carlo approach with 1000 replications(LeSage & Pace, 2009).Standard errors are clustered at the municipal level.Significance level: * p < 0.10, ** p < 0.05, ***p < 0.01.