Spatial spillover effects of fiscal decentralization on governance and public administration quality

ABSTRACT This study examines how fiscal decentralization affects the quality of governance and public administration. Using a balanced panel dataset of 63 provinces in Vietnam over the 2011–18 period and employing the spatial Durbin model, we find that governance and public administration quality develop gradually, leading to spatial spillovers across provinces. Importantly, revenue decentralization exerts a negative spatial effect on participation, transparency, accountability and control of corruption, but exhibits a positive spatial effect on public administrative procedures across regions. By contrast, fiscal transfers improve transparency, accountability and control of corruption. That is, fiscal decentralization drives local governments in ‘a race to the bottom’ when they address budget shortfall pressures.


INTRODUCTION
Over recent decades, numerous developing countries pursuing poverty-reduction strategies have recognized fiscal decentralization 1 as a vital policy instrument for improving governance (Altunbaş & Thornton, 2011;Sepulveda & Martinez-Vazquez, 2011). By contrast, a different stream of literature argues that fiscal decentralization could give rise to poor governance. Fiscal decentralization requires central governments to devolve fiscal powers to local authorities. When local authorities encounter problems of revenue shortfalls, fiscal transfers become the primary source of local revenue. As a result, local authorities may deliberately worsen the quality of governance (Rodden, 2004) to compete for a high tax base as 'a race to the bottom' and to attract increased outside investment by offering less transparency and accountability (van der Kamp et al., 2017). However, little attention has been paid to spatial interdependence associated with fiscal decentralization and public governance.
This study sheds light on this issue by investigating the spatial spillover effects of fiscal decentralization on governance and public administration quality in the context of a specific emerging economy, namely Vietnam. In recent decades, Vietnam has tried to improve various aspects of public governance, such as increasing transparency and accountability, reducing corruption, simplifying administrative procedures, and improving public service delivery (CECODES et al., 2019;Thang et al., 2015). Previous empirical studies show that the central government has moved towards increased fiscal decentralization to local authorities (Thanh & Canh, 2020). However, the quality of local governance remains weak, primarily owing to serious corruption which impedes local policy outcomes (Maruichi & Abe, 2019;Nguyen et al., 2018). Moreover, the Vietnamese central government has implemented a high expenditure decentralization model along with highly concentrated power (Rab et al., 2015), but it decides on all tax bases and rates; thus, local authorities lack own-revenue resources to finance local expenditure. In such a case, a relevant question is whether fiscal decentralization improves governance quality across Vietnamese provinces. However, undeniable difficulties exist in estimating causal effects in public goods provision owing to the spatial aspect of local public goods (Koide, 1988). In a recent study, Thanh and Canh (2019) examined the spillover effects of fiscal decentralization in Vietnam. However, the authors' research focuses on the impact of fiscal decentralization on government size and spending composition. They found that fiscal decentralization in Vietnam is in line with the Leviathan hypothesis, in which the decentralization of the fiscal system can reduce government spending and government size. Therefore, the study fails to account for the impact of fiscal decentralization on the effectiveness of local government. To this end, the present study fills the gap by examining the effects of fiscal decentralization on the quality of governance and public administration. Specifically, it investigates the spatial spillover effects of fiscal decentralization on the quality of governance and public administration in Vietnamese provinces. The question involves whether fiscal decentralization influences local governments to relax public governance and administration quality to increase tax competition, resulting in a 'race to the bottom' across provinces. To attain this objective, fiscal decentralization is proxied by two indicators, namely, revenue decentralization and fiscal transfer. Governance and public administration quality are measured in six dimensions: citizen participation; transparency; accountability; control of corruption; public administrative procedures; and public service delivery. We apply spatial econometric models to investigate the spatial spillover effects of fiscal decentralization on the quality of governance and public administration. According to LeSage and Dominguez (2012), spatial econometric models may be suitable for empirical studies related to public choice and public goods provision because the spillover of local public goods in a region affects the performance of neighbouring regions. We apply maximum likelihood estimators (MLEs) to the spatial Durbin model (SDM) with two matrices: (1) an inverse distance-based weighting matrix; and (2) a rowstandardized contiguity-based weighting matrix. To obtain an unbiased estimator, we conduct robust variance estimation using local fiscal decentralization clusters.
The remainder of the article is structured as follows. Section 2 provides a brief overview of fiscal decentralization and governance in Vietnam. Section 3 reviews the literature on fiscal decentralization and governance, and section 4 presents the research model, methodology and datasets. Empirical findings are discussed in section 5. Finally, conclusions and policy implications are presented in section 6.

SUMMARY OF FISCAL DECENTRALIZATION IN VIETNAM
Since the enactment of the 2002 State Budget Law, the central government has pursued a model of high expenditure decentralization in terms of highly concentrated authority. According to the State Budget Law, local governments are assigned a variety of revenue sources, such as 100% retained revenues, shared revenues and local borrowing. At the same time, the central government addresses the local fiscal imbalance with fiscal transfer mechanisms, including fiscal balancing and targeted transfers. Revenues retained 100% by local governments comprise taxes and fees related to local lands as well as local users' fees and charges. Revenues shared between central government and local governments consist of valueadded tax, corporate income tax, personal income tax and excise tax on domestic goods. In reality, only a few provinces have the ability to share revenues with central government (see Table A2 in the Appendix in the supplemental data online). The rate of revenue-sharing transferred to the centre has increased over time, whereas public expenditure has been devoted increasingly to local governments. As a result, although the provinces' share of total state revenue increased from 25% to 35% during the 2000-15 period, their share of total state spending increased from 45% to 54% over the same period. This situation increases the dependence of local governments on fiscal transfers from the centre, owing to the much lower local revenue compared with local spending. This condition is true for approximately 50 provinces that receive fiscal transfers from the centre (Anh, 2016). In certain respects, such as expenditure responsibilities and economic development, local governments need to boost their revenues by attracting additional investors from outside their regions to increase tax competition, and, consequently, their revenues (van der Kamp et al., 2017). However, in raising mobile tax bases, each province considers only economic activities within its jurisdiction, leading to a lack of coordination between provinces. As a result, the administrative boundary of each province becomes its economic border. Vietnamese provinces have ignored economic policies and institutional arrangements to take advantage of economies of scale to solve external problems (Anh, 2016).

LITERATURE REVIEW
Over the past two decades, much attention in the literature has been paid to the spillover effect of local public goods on fiscal decentralization (Bloch & Zenginobuz, 2007;De Siano & D'Uva, 2017;Ferraresi et al., 2018;Que et al., 2018a). De Siano and D'Uva (2017) argue that local governments engage in local public delivery by considering the behaviour of neighbouring governments. This practice means that decisions on local public delivery are determined not only by traditional factors such as tax revenue, fiscal transfers, and the demographic, socio-economic and political characteristics of jurisdictions but also on the fiscal policies of neighbouring governments. In other words, to maximize their welfare, local governments may choose fiscal policies that influence the welfare of neighbouring jurisdictions (Ferraresi et al., 2018). Theoretically, several models have been developed in the specialized literature to explain spatial interactions between jurisdictions, including the political yardstick competition, tax competition, spillover effects of local public goods, Spatial spillover effects of fiscal decentralization on governance and public administration quality 479 political trends and knowledge diffusion models (Ferraresi et al., 2018).
In the political yardstick competition model, voters are hypothesized to lack information on expenditure and tax policies within their jurisdiction, thereby comparing them with those of neighbouring jurisdictions and voting for the penalty of incumbent elective officials if decisions on local expenditure and tax policies are not in line with those of neighbouring jurisdictions. The spatial interaction effects of political yardstick competition are found in several empirical studies that are significant (Benz, 2012;Boarnet & Glazer, 2002;Elhorst & Freret, 2009). In the tax competition model, tax bases are mobile and thus may increase spatial interactions in tax competition behaviour across jurisdictions. Local governments may pursue tax competition strategies based on differences between their own and their neighbours' tax rates (Bayindir-Upmann, 1998;Ndiaye, 2018;Shon, 2018;Xiao & Wu, 2020;Bucovetsky & Smart, 2006). Regarding the spillover effects of public goods, public spending in a region may generate positive or negative spillover beyond its borders. Therefore, local governments may underspend on local public goods/services that create positive spillover across jurisdictions but overspend on local public goods/services that produce negative externalities in neighbouring jurisdictions. As a result, decisions on public expenditure in a jurisdiction are likely to depend on the level of public expenditure of neighbouring jurisdictions (Baicker, 2005;Caldeira et al., 2015;De Siano & D'Uva, 2017;Thanh & Canh, 2019). With regard to political interactions, a ruling incumbent politician is assumed to consistently consider his/her political party ideology. Therefore, he/she can choose tax and expenditure policies in his/her jurisdiction by considering those of his/her neighbours that are affiliated with his/her party. As a result, decisions on fiscal policy in a jurisdiction depend on the political trends of an incumbent affiliated party across jurisdictions (Altunbaş & Thornton, 2011;Bosch & Solé-Ollé, 2007;Caldeira et al., 2015;Elhorst & Freret, 2009;Faguet, 2014;Santolini, 2008). Similarly, another side of fiscal interaction may stem from knowledge diffusion and learning. In this model, the local government of a jurisdiction is assumed to be well-informed on the costs and benefits of tax rates and public expenditure of neighbouring governments. By learning and imitating, a local government may modify policy formulations to create optimal tax rates and efficient public expenditure allocation in its jurisdiction (Kim, 2013;Mitchell et al., 2017).
Interestingly, from the empirical perspective, debate on the impact of fiscal decentralization on government quality is ongoing. Intergovernmental fiscal competition with movable tax bases could restrain government monopoly on taxation and help align local public expenditure with people's preferences, thereby reducing public resource misallocation and improving policy outcomes (Weingast, 2009). According to Rodden (2004), under fiscal decentralization, expenditure responsibilities are delegated to local governments, but central government maintains rigid control over tax bases and local revenue sources, which means that local governments have limited authority in terms of expenditure. Moreover, Garman et al. (2001) stated that when political power is centralized in central government, fiscal transfers may be enhanced, and the autonomy of local spending is restricted. Consequently, governance likely degrades when local governments lack adequate resources to satisfy the assignment of fiscal responsibilities. Thus, local authorities must indulge the policy priorities of central government at the expense of local interests to gain increased fiscal transfers. Garman et al. documented that local governance reforms in Latin American countries fail because central governments tightly control national revenue sources and determine how funds are allocated, and thus have wide discretion in providing fiscal transfers to local governments. Van der Kamp et al. (2017) argue that although China's decentralization promotes inter-jurisdictional competition, it has potential drawbacks in slowing the progress of centrally mandated governance reforms; specifically, environmental transparency policies. Revenue pressures can lead to a race to the bottom, thereby defying governance reforms in China.

RESEARCH MODEL
By applying the numerical simulation method, Elhorst (2010) demonstrated that the spatial Durbin model (SDM) developed by LeSage and Pace (2009) can generate unbiased estimations compared with traditional spatial models such as the spatial autoregressive (SAR) model, spatial autocorrelation model, spatial error model (SEM) and other spatial lag models. The objective of this study is to investigate the spillover effects of fiscal decentralization on governance and public administration quality in Vietnamese provinces. In the spirit of LeSage and Pace (2009), this study applies the SDM as follows: where i is the province (1, . . . , N ); t is time (1, . . . , T ); m i is provincial fixed effects; 1 it represents residuals N (0, d 2 ); Y it is governance and public administration quality; and X is a set of independent variables. W is the spatial weight matrix, with i, j [ [1, N ] and i = j. Parameters r, b and g refer to regression coefficient vectors.
Equation (1) involves a spatially lagged dependent variable (Wy) and spatially lagged independent variables (WX ). The presence of the matrix W reflects the strength of spatial interactions between jurisdiction I (the row of the matrix) and jurisdiction j (the column of the matrix). According to LeSage and Pace (2011), spatial effects depend on the exogenous chosen matrix; hence, in this study, we use two types of matrix for the robustness checks: . Row-standardized contiguity-based weighting matrix (W bin i,j ): this matrix is an adjacent binary matrix. If provinces i and j are bordering provinces, then they are . Inverse-distance-based weighting matrix (W dis i,j ): this matrix is calculated based on the distance of provincial administrative centres (i.e., provincial capital). The inverse-distance weighting matrix is identified as follows:  (1) can be rearranged as follows: where R refers to the intercept and errors terms. The matrix of partial derivatives for equation (3) is specified as: Following Golgher and Voss (2016), we present a simple example with three neighbouring regions, and use the row-normalized contiguity weight matrix to derive the matrix of partial derivatives as follows: The DE are measured by the average sum of the diagonal terms: The IE are calculated by the average sum of the offdiagonal terms: The TE are given as follows: The DE are determined in the diagonal terms of the matrix resulting from a change in the independent variable of a jurisdiction. The IE are found in the off-diagonal terms, indicating that a change in the dependent variable of a jurisdiction is affected by a change in the independent variable of several other jurisdictions. The results of the Spatial spillover effects of fiscal decentralization on governance and public administration quality 481 DE and IE are determined by parameters r, b k and g k . Notably, in the SDM model, the IE can be divided into two parts; that is, local and global effects. Local effects affect spatial interactions between close neighbours relating to W bin i,j , whereas global effects affect spatial interactions extending to an entire territory relating to W dis i,j . The relationship between fiscal decentralization and local governance may be driven by reverse causality (Altunbaş & Thornton, 2011). For example, corrupt local officials may require central government to increase fiscal powers in a local government, which would strengthen their rent-seeking ability. Moreover, a potential endogeneity issue exists in the SDM model (Gibbons & Overman, 2012). We follow previous methods (e.g., Lyytikäinen, 2012;Ferraresi et al., 2018) to deal with potential endogeneity. First, based on Belotti et al. (2017), we employ MLEs for spatial panel data models implemented in Stata. To obtain unbiased MLE, we conduct robust variance estimation using local fiscal decentralization clusters (Guermat & Harris, 2002). Although the generalized method of moments (GMM) estimators are likely considered as a substitute, the GMM estimators added by the spatially lagged independent variables as a precondition will not allow us to estimate the spatial spillovers (Del Bo & Florio, 2012;Que et al., 2018a). Similarly, several studies have applied MLE to estimate the SDM model to obtain robust results (Del Bo & Florio, 2012;Que et al., 2018aQue et al., , 2018bYu et al., 2013).
According to Elhorst (2010), equation (1) can be estimated by fixed effects (FE) or random effects (RE) for spatial panel data. While a dummy variable is applied for each spatial unit in the FE model, m i is assumed to be identically distributed with N (0, d 2 ui ) and independent of 1 it in the RE model. Furthermore, the spatial data are arranged first by time (T) and then spatial units (N). In this way, when the observations of Y it and X it variables are sacked, we can obtain an (NT,1) matrix for Y it and (NT,K) matrix for X it (Elhorst, 2010). The Hausman test is applied to select FE or RE. Interestingly, the common problem of spatial panel data models is that the Hausman test often does not meet its asymptotic assumptions, especially in the case of small samples. MLE can overcome this issue since it directly accounts for the coefficients of FE and RE (Belotti et al., 2017).

Dependent variable
The Vietnam provincial governance and public administration quality variable (Y ) is measured by the public administration performance index (PAPI; PAPI). PAPI covers information on citizen judgements of provincial performance in terms of governance and public administration quality based on six dimensions: (1) citizen participation at local levels (PART); (2) transparency in policymaking (TRAN); (3) vertical accountability (ACCOU); (4) control of corruption (CORRUP); (5) public administrative procedures (PUBAD); and (6) public service delivery (PUBSER). Each PAPI dimension has several subdimensions compiled from several estimated components from the collected data (see section A3 in the Appendix A in the supplemental data online). This study uses the six PAPI dimensions successively to measure the spillover effects of fiscal decentralization on governance and public administration quality, in which PART, TRAN, ACCOU and CORRUP can be grouped into governance quality, and PUBAD and PUBSER can be grouped into public administration quality.

Independent variables
We use two independent variables; that is, provincial government decentralized revenue per capita (FDREV) and fiscal transfers per capita (TRANSFER), to estimate equation (1). These measures capture the dynamics of local governments in raising local revenue to finance public expenditure. The efficiency of provincial expenditure is assumed to significantly contribute to local governance quality (De Mello & Barenstein, 2001). According to the 2002 and 2015 Vietnam State Budget Law, provincial government revenues include 100% retained and shared revenues. The revenue-sharing rate is fixed to a three-or five-year period or the stability period. Considering the incentives given by central government, local governments may increase own-source and shared revenues to overcome budgetary constraint and finance the needs of their expenditure during the stability period. Thus, decentralized revenue indicates potential local differences and local fiscal capacity in raising local revenue to finance local expenditure (Rodden, 2002).
Fiscal transfers are important in addressing horizontal and vertical fiscal imbalances across provinces, generating the equal distribution of fiscal resources between provinces. However, such fiscal transfers could generate a common pool problem, leading to poor public service provisions and governance (Garman et al., 2001). Therefore, this variable is included to test how fiscal transfers affect governance and public administration quality in Vietnamese provinces.

Control variables
Under fiscal decentralization, local governance can be influenced by other factors, such as demographic, socioeconomic, and political characteristics (Altunbaş & Thornton, 2011). Given that Vietnamese politics is unitary, we include only a set of control variables that characterizes the demographic and socio-economic situation of Vietnamese provinces. Specifically, we include provincial population (POP), which may control public service needs and thus affect local governance and public administration quality. Regarding the relationship between fiscal decentralization and tax base competition, we include private sector investment (PRIDEV), which indicates private investment flows between neighbours to capture tax base competition under revenue pressure. Thus, changes in private investment flows may influence local governance and public administration quality (van der Kamp et al., 2017).

482
Su Dinh Thanh et al.
PRIDEV is calculated by the share of new private investment flows to the gross provincial product (GPP). Similarly, we include provincial leadership arrangements (DYNAM), indicating provincial proactivity in solving problems in the local private sector (McCulloch et al., 2013). DYNAM is measured by the proactive and creative provincial leadership score index, which is an index built in the Provincial Competitiveness Index ranging from zero to one hundred percentage. 2 For local economic structure change, we use a provincial industrial production variable (ECOSTRU), which is measured by the share of provincial industrial production in the GPP. This variable is expected to improve the quality of governance and public administration (Cao & Zheng, 2016).

Research data
Our empirical study uses a balanced panel dataset of 63 Vietnamese provinces during the 2011-18 period. 3 Variables such as FDREV, TRANSFER, POP, PRIDEV and ECOSTRU are from the annual provincial Statistical Yearbook. 4 DYNAM is from the provincial competitiveness index (PCI), which is the Vietnam Chamber of Commerce and Industry (VCCI). The unweighted scores of the PAPI and its six dimensions are collected from PAPI reports during the 2011-18 period. 5 The logarithm is obtained for all the variables except for DYNAM to reduce heteroscedasticity. Data on the weighting matrices (W bin i,j and W dis i,j ) is collected from DIVA-GIS (diva.gis.org/data), which is transformed with the spmat command in Stata.
This study selects the 2011-18 period because the PAPI was piloted in 2008 and extended to 30 provinces in 2009. From 2011 onwards, a total of 63 Vietnamese provinces have been added to the PAPI ranking list. Table 1 reports the definition, descriptive statistics and sources of the variables in equation (1). Figure 1 presents a map of the FDREV during the 2011-18 period. It illustrates the maps of the six dimensions of the PAPI during the same period. Surprisingly, we find that provinces with the highest FDREV do not match those with the highest scores in the six PAPI dimensions.

Moran's I test
To determine whether spatial connections exist between governance and public administration quality, we conduct Moran's I test for spatial autocorrelations between the six PAPI dimensions (i.e., PART, TRAN, ACCOU, COR-RUP, PUBAD and PUBSER), which are calculated based on the global Moran's I for W dis i,j with different clusters (e.g., 160, 230, 320 km and the entire panel sample). Table 2 provides the spatial autocorrelation results of the six PAPI dimensions during the 2011-18 period. The observations show that spatial connections exist between the six PAPI dimensions, which is favourable to the use of spatial econometric models.
Additionally, we perform an exploratory spatial data analysis (ESDA) to detect patterns in spatial data. First, we diagnose the spatial pattern of the PAPI for Vietnam in the period 2012-18. Figure A1 in the Appendix in the supplemental data online shows that provinces with high PAPI scores are mostly located in the Red River Delta region and the Central Coast region. Second, based on two matrices (W bin i,j and W dis i,j ), we test the marginal spillover effects of PAPI that are caused by changing provincial revenue decentralization. Because the number of Vietnamese provinces is large, we only estimate the marginal spillover effect of PAPI for two provinces of interest: Hanoi and Ho Chi Minh City. As Figures A2 and A3 in the Appendix online, the marginal spillover effects of PAPI are estimated for W bin i,j and W dis i,j when there is a change in revenue decentralization in Ho Chi Minh City (Hanoi). For W bin i,j and W dis i,j , we find that a change in revenue decentralization in Ho Chi Minh City (Hanoi) will change PAPI in Ho Chi Minh City (Hanoi) and its neighbours. Interestingly, we find that the speed of the spillover effects seems to be the same from W bin i,j and W dis i,j . The spatial spillovers from two locations diminish corresponding to the distance from bordering provinces. This finding suggests the significance of the spatial structure with geographic proximity to spillovers in PAPI in Vietnam.
First, we test equation (1) using the FE model. The result of Pesaran's test shows the presence of spatial correlations (see section A4 in the Appendix in the supplemental data online). By testing the order of the SAR model, the SEM and SDM, we obtain an applicable spatial model based on the Akaike information criterion (see A5 in the Appendix online). If the results support the SAR model, then we will use SDM to further evaluate the spatial spillovers. Table 3 presents the SDM results estimated by the two matrices (W dis i,j and W bin i,j ) with local revenue decentralization clusters. Columns (1)-(6) are for W dis i,j , and columns (7)-(12) are for W bin i,j . Testing hypotheses H 1 0 := 0 and H 2 0 := −rb is necessary to identify whether or not a generalized SDM is proper. The results of the Wald test reject the H 1 0 and H 2 0 hypotheses for the two matrices at the significant 1% and 5% levels, indicating that SDM is the most appropriate. Second, the Hausman test is used to decide between the FE or RE models.
The spatial spillover effects are analysed by the spatial lag of the dependent variable r (rho) and the spatial lags of WX g. The spatial autocorrelation coefficient r is significant in all cases. Furthermore, the magnitude of coefficient r for W dis i,j (see columns 1-6) is larger than that for W bin i,j (see columns 7-12), indicating that all interactions between Vietnamese provinces result from the reforms on governance and public administration of the entire nation. Moreover, Vietnamese provinces are characterized by positive spatial interactions between PART, TRAN, ACCOU, CORRUP and PUBAD, with the exception of PUBSER, which is negative. This finding suggests that the enhancement of governance and public administration quality among neighbours promotes the improvement of a province's governance and public administration quality. The negative spatial interaction of PUBSER suggests that public Spatial spillover effects of fiscal decentralization on governance and public administration quality 483 services provided by neighbours can substitute a province's public service delivery.
Regarding the spillover effects of WX , the results in Table 3 provide an idea of interactions between provinces. Thus, Table 4 reports the signs and magnitudes of the DE, IE and TE to display the spatial spillover effects of fiscal decentralization on governance and public administration quality.

Direct and indirect effects
The theoretical framework proposes the spillover effects of fiscal decentralization. However, whether such spillover effects occur must be further assessed by IEs. Table 4 presents the estimation results of the DEs and IEs in the SDM for W dis i,j and W bin i,j for the entire sample.

Fiscal decentralization from the perspectives of revenue decentralization
The W dis i,j estimation results show that the spatial spillover effect of FDREV is significant in three of the six dimensions of governance and public administration quality. Specifically, the negative DE of FDREV on ACCOU is significant, implying that revenue decentralization reduces vertical accountability within a province. The negative IE of FDREV on ACCOU is likewise significant, indicating that an increase in revenue decentralization among neighbours will generate a reduction in a province's vertical accountability. The coefficient of the spillover effect of FDREV on ACCOU (0.125) is larger than that of the DE (0.044) (  all neighbours. Similarly, the observations show that the positive DE of FDREV on CORRUP is significant, indicating that revenue decentralization increases the control of corruption within a province. The negative IE of FDREV on CORRUP is significant, implying that an increase in revenue decentralization among neighbours will lead to a reduction in a province's control of corruption. As a result, the TE of FDREV on CORRUP is negative and significant, suggesting that revenue decentralization could reduce control of corruption among Vietnamese provinces (Table 4, column 4). Interestingly, the magnitude of coefficient of the TE of FDREV on ACCOU (−0.169) is larger than that on COR-RUP (−0.070), thus indicating that revenue decentralization could significantly reduce accountability, while it might minimize corruption among Vietnamese provinces. Moreover, the IE of FDREV on PUBSER is positive and significant, suggesting that an increase in revenue decentralization among neighbours will generate an increase in a province's public goods delivery. This finding implies that when a local government enhances its financial ability, it can spend more on public provision, thereby definitely increasing spillover public service/goods demand. The W bin i,j estimation results indicate that the spatial spillover effect of FDREV is significant in five of the six dimensions of governance and public administration quality. Specifically, the negative spatial spillover effects of FDREV on PART, TRAN and ACCOU are significant, suggesting that an increase in revenue decentralization will generate a reduction in a province's citizen participation, vertical transparency, and accountability in local policymaking, respectively. In addition, the positive spatial spillover effect of FDREV on PUBAD and PUBSER is significant, indicating that an increase in revenue decentralization will generate an increase in a province's public administrative procedures and public service delivery, respectively. In general, our empirical findings provide evidence that the spillover impact of revenue decentralization can reduce citizen participation, transparency, accountability and control of corruption, but increase public administration procedures, thus leading to a race to the bottom.

Fiscal decentralization from the perspective of fiscal transfers
The W dis i,j estimation results show that the spatial spillover effect of TRANSFER is significant in three of the six dimensions of governance and public administration quality (Table 4). The positive spatial spillover effect of TRANSFER on TRAN is significant, suggesting that an increase in fiscal transfers among neighbours will generate an increase in a province's transparency in local policymaking. However, the TE of TRANSFER on TRAN is insignificant (Table 4, column 2). Fiscal transfers thus fail to contribute significantly to improving local transparency in Vietnamese provinces. Surprisingly, the coefficient of the spatial spillover effect of TRANSFER on PART is 0.144 and significantly negative (Table 4, column 1), suggesting that an increase in fiscal transfers among neighbours can lead to a reduction in a province's citizen participation at the local level. Thus, the coefficient of the TE of TRANSFER on PART is −0.177 and significant, indicating that fiscal transfers strongly reduce local participation among Vietnamese provinces. Furthermore, we find that the positive IE of TRANSFER on PUBSER is significant (Table 4, column 6), suggesting that an increase in fiscal transfers among neighbours will generate an increase in a province's public goods delivery. This finding implies that fiscal transfers increase the provision of local public services/goods.

Control variables
Control variables are crucial in revealing the determinants of governance and public administration. First, our empirical findings on the W dis i,j results show that the TE of POP on PART, CORRUP and PUBSER is significantly positive. In the W bin i,j results, the TE of POP on ACCOU is positive and significant. This suggests that population growth would increase the demand for public goods, leading to a complementary relationship between public governance and public goods provision. Second, in the W dis i,j results, the TE of PRIDEV on PART and public goods provision is positive and significant, indicating that private sector development could improve citizen participation and increase public goods provision. This change may come about because, after more than three decades of socio-economic reforms, the private sector plays an essential role in the Vietnamese economy ( Van Thang & Freeman, 2009). As a result, the private sector can improve governance and public administration quality (Han & Baumgarte, 2000). However, the TE of PRIDEV on TRAN and ACCOU is significantly negative, but that on PUBAD is significantly positive. For the W bin i,j results, we further find that the TE of PRIDEV on CORRUP is negative and significant. Indeed, given that own-resource revenue is important to the local budget, and to overcome local budgetary constraints, local governments must indulge the interests of local firms, ignoring governance and public administration reforms and generating ineffective decentralization policies (van der Kamp et al., 2017). As a result, we find that the TE of DYNAM on PART, TRAN and CORRUP is positive, suggesting that local governments with increased proactivity could boost citizen participation and transparency in local policymaking, leading to enhanced control of corruption. However, local government proactivity reduces accountability in local policymaking and increases public administrative procedures (for the W bin i,j results), demonstrating that local government proactivity generates decentralization policies that mismatch governance and public administration reforms. Third, the TE of ECOSTRU on CORRUP is positive and significant, implying that economic structure change has an important contribution in the control of corruption. Similarly, economic structure change can lead to a race to the top in promoting citizen participation, transparency, and public goods delivery across Vietnamese provinces. However, the TE of industrial sector growth on PUBAD Note: ***, **, *Significant levels at 1%, 5% and 10%, respectively. Standard errors are adjusted for clusters in FDREV (robust in brackets).
Spatial spillover effects of fiscal decentralization on governance and public administration quality 489 is significantly positive, suggesting that provinces with a high level of growth in the industrial sector fail to reduce public administrative procedures.
6.6. Testing for spatial independence Based on the estimation results in Table 4, we find that the significance, magnitude and sign of the spatial spillover effects of fiscal decentralization on governance and public administration quality in the W bin i,j estimation are stronger than those in the W dis i,j estimation. This finding implies that the spillover effects of fiscal decentralization are mainly limited to the county level, owing to the huge distances between provinces.
To further confirm the above results, we test for spatial independence using different local clusters based on distance. Table 5 reports the results of the DEs and IEs in the SDM for W dis i,j truncated by two different clusters based on distance; that is, (1) 160 km (linkages from two to 19 bordering provinces); and (2) 320 km (linkages from seven to 27 bordering provinces). Our empirical results show that the spatial autocorrelation coefficients r for the two clusters have signs and are statistically significant, which is consistent with those in Table 3 for the W dis i,j estimation of the entire sample. However, the magnitude of the spatial autocorrelation coefficients r gradually diminishes as the distance between provinces varies from 160 to 320 km, suggesting that spatial interactions between neighbouring provinces are stronger than those between provinces that are located far apart. The magnitude of the significant spillover coefficients for the cluster with a 320 km distance is larger than that for the cluster with a 160 km distance. This result may be indirect evidence of fiscal decentralization competition between neighbouring provinces. Moreover, the IE is found to be larger than the DE.

Robustness checks
To check robustness, we firstly ignore spatially weighted explanatory variables and use the SAR model to estimate the relationship between fiscal decentralization and governance and public administration quality for provinces with revenue-sharing arrangements with central government (see section A2 in the Appendix in the supplemental data online). Thus, we include a dummy variable (DUMMY), indicating provinces with revenue-sharing arrangements with central government (hereafter known as a revenue-sharing province). Given that local revenue is limited owing to the centre-provincial revenue-sharing arrangement, a revenue-sharing province must solve its budget problems by increasing tax base competition between its neighbours. Thus, it is hypothesized that to attract additional outside investors, a revenue-sharing province may choose to satisfy local firms' interests by relaxing governance (van der Kamp et al., 2017).
The estimation results of the SAR model are reported in Table 6. The observations show that, first, a revenuesharing province significantly increases public service provision within the province but fails to improve public administration procedures. Second, though the province significantly increases control of corruption within its borders, it fails to improve citizen participation, transparency in policymaking, and vertical accountability. Once again, these results imply that under revenue pressures, a revenue-sharing province must ignore the priorities of local governance reforms at the expense of local interests to obtain additional local tax bases. Third, the study added the Log of provincial GDP (in constant prices) as another control variable in the estimation to check for robustness. The results are reported in sections A6 and A7 in the Appendix in the supplemental data online. Income can be relevant in determining the level of provincial public governance, as individuals in rich provinces could have a high demand for good public services delivery. Table A6 in the Appendix online reports the estimation results of the SDM model similar to Table 3; and Table A7 online reports the direct, indirect and TE comparable with Table  4.
Fourth, the potential endogeneity in the SDM model is a serious issue. There are some potential solutions for endogeneity, such as using quasi-experimental factor as an instrumental variable (Gibbons & Overman, 2012). Lyytikäinen (2012) and Ferraresi et al. (2018) both use exogenous sources of variation in tax rates as instrumental variables. Specifically, Ferraresi et al. (2018) use exogenous variation in neighbours' expenditure induced by natural disaster; however, earthquakes are not common in Vietnam. We have followed the suggestion in Gibbons and Overman (2012) to consider other natural disasters such as flooding or storm. Once again, we do not find data availability for all provinces in Vietnam by year. Instead, there is some data on earthquakes, flooding or storms in some provinces or place by discontinuous years. On this basis, we have relied on the second approach in Ferraresi et al. (2018) using an internal instrument; namely, the change in average per capita expenditure of neighbours is instrumented by using its lag. For detail, a GMM estimator is applied. We did succeed in estimating our empirical models using two-step system GMM estimators. We report these results in Table A8 in the Appendix in the supplemental data online. Compared with the main results obtained in Table 3, the sign and significance of the spatial lag coefficients of the dependent variables are similar, except for the PUBSER equation (only when using the distance-weighting matrix). The coefficient is significant in the case of GMM estimators, while it is insignificant in the base specifications. For the spatial lag coefficients of FDREV, we confirm the robustness for the PART, ACCOU and PUBSERV equations. As for the spatial lags of TRANSFER, the robustness is confirmed in the CORRU and PUBSERV equations.

CONCLUSIONS
This study investigates the spatial spillover effects of fiscal decentralization on governance and public administration in Vietnamese provinces. By employing a panel dataset of 63 provinces during the 2011-18 period and applying SDM with a bias-corrected maximum likelihood method 490 Su Dinh Thanh et al. for two matrices (W dis i,j and W bin i,j ), we obtain the following key findings from the data.
First, we find that local effects on close neighbourhood clusters are important. The literature (e.g., Lewis, 2017) focuses on the DEs of local government spending on economic activities or public service delivery in one jurisdiction. Meanwhile, other studies (Chen, 2016) present evidence on labour mobility between regions during the process of fiscal decentralization, which implies interconnections among jurisdictions. Thus, the finding that the effects of local government spending should be concerned with the broad context of interdependence among neighbouring regions is significant to the literature. In such a finding, non-measurable factors, such as history, tradition and culture, play a key role in transmitting the spatial effects of fiscal decentralization on governance and public administration quality, which is in line with Que et al. (2018aQue et al. ( , 2018b. Therefore, based on the estimation results of W bin i,j , we discuss the main findings of this study. Second, a highly decentralized fiscal system can see an improvement in good governance by increasing citizen participation, transparency, and accountability, which can lead to the satisfactory control of corruption (Hankla, 2009;Huther & Shah, 1998). Recently, Védrine (2020) provided evidence on spatial interactions in fiscal decentralization in Europe, showing that a high degree of policy decentralization can induce spatial interactions. Surprisingly, we find that the spatial effect of revenue decentralization participation, transparency, and accountability is significantly negative, but the effect on the control of corruption is insignificant. These findings suggest that provinces with decentralized revenues are likely to reduce participation, transparency, and accountability, and ignore control of corruption, leading to a race to the bottom in improving local governance across Vietnamese provinces, or at least defiance against governance reforms. Previous studies (e.g., Lago et al., 2020) indicate that the extent and quality of institutions in local governments may create differences in political, economic and fiscal performance across countries. Our findings may be significant, as provinces that experience strong revenue pressures can choose to frustrate public governance reforms in the local budgeting process. This action can cause the spatial effect of revenue decentralization to increase public administrative procedures across Vietnamese provinces. Our findings do not support the hypothesis that decentralization with inter-jurisdiction competition would give rise to high participation, transparency, and accountability, and reduce corruption, as found in Arikan (2004) and Ivanyna and Shah (2011).
What can be attributed to this finding in the context of Vietnam? First, to ensure increased local revenue resources, local governments must raise local tax bases by accommodating local business concerns, further boosting complicity between local authorities and powerful business groups. As a result, local governments can pursue policies to protect the business sector from governance reforms. For this reason, according to the World Bank ranking in ease of doing business in recent years, Vietnam still lags behind other countries in the Asia region in terms of business environment and control of corruption (Maruichi & Abe, 2019). Second, tax competition is subject to economic constraints and induces dramatic fiscal congestion, because the money collected through local taxes cannot sufficiently finance the needs of public spending. For the fiscal congestion effect, local governments may resist implementing governance reforms, as such reforms may affect local businesses that play an important role in generating local revenue. Less transparency allows inter-jurisdiction to attract increased investments by offering businesses nonformal benefits, specifically, lax public governance (van der Kamp et al., 2017).
Third, the findings show that the spatial effect of revenue decentralization on public service delivery is positive in the provinces of Vietnam. This result implies that a complementary relationship may exist in the provision of public services between neighbouring provinces, which is in line with the findings of Ferraresi et al. (2018) and Thanh and Canh (2019).
Fourth, we find that provinces with additional fiscal transfers can generate increased public goods provision, which suggests that although fiscal transfers are important in solving vertical imbalances, they can increase public spending size, as proposed in the specialized literature (Duan & Zhan, 2011;Ferraresi et al., 2018). Surprisingly, we find that the spatial spillover effect of fiscal transfers on transparency, accountability and control of corruption is positive, but on public administration procedures is negative. These results suggest that fiscal transfers increase inter-jurisdiction competition among neighbours in improving governance and public administration quality. In other words, provinces with additional fiscal transfers significantly improve governance and public administration quality, leading to a race to the top. This finding is because provinces with increased fiscal transfers depend heavily on central government, and thus must obey the priorities of reform policies to obtain additional funds from central government.
Our findings imply that under fiscal decentralization, attempts to create good governance and public administration in terms of citizen participation, transparency, and accountability, and to increase control of corruption and reduce public administration procedures have tradeoffs, because local governments become involved in a race to the bottom owing to tax competition and lack of local revenue. Local governance and public administration quality should lay the groundwork for public sector reforms. Meanwhile, fiscal decentralization must respond appropriately to ensure that local governments have adequate local revenue resources to meet local expenditure demands (Huther & Shah, 1998). Moreover, local authorities should push inter-jurisdiction collaboration in fiscal management to advance good governance and foster regional sustainable development.
Lastly, it is worth noting that the issue of potential endogeneity from the estimates should be further examined in future studies. This study has dealt with this issue by using lags of neighbour expenditure as internal instruments in the GMM estimate. However, it would be more beneficial if future studies could consider exogenous variation in public spending through exogenous shocks such as natural disasters, when this data is available for all provinces in Vietnam.

DISCLOSURE STATEMENT
No potential conflict of interest was reported by the authors.