ANALYZING POLITICAL AND SYSTEMIC DETERMINANTS OF FINANCIAL RISK IN LOCAL GOVERNMENTS

Abstract Studies have shown that political variables can infl uence the volume of government debt and have recommended investigating the joint effects of diverse factors on the risk of local government default. Considering the relation between economic management and political constraints, this paper examines the joint infl uence of political and systemic factors on the risk of loan default by Spanish local governments. To do so, we analyze 148 city councils for the period 2006-2011, using a logit model with panel data and an artifi cial neural network. The empirical results indicate that the fi nancial risk of local governments is affected both by political factors specifi c to each case and, simultaneously, by systemic variables for the country. Specifi cally, political variables such as the mayor not having economics-related university studies, the under-representation of female councilors in the municipal corporation, municipal government by a party with a progressive ideology, and ideological alignment between the municipal and the regional government are all associated with greater fi nancial risk. Moreover, rising national unemployment, an increased sovereign risk premium, the impact of the electoral cycle, and that of declining economic growth are all factors that may increase the risk of default. The fi ndings presented are of great potential interest for governments, managers, national and international fi scal authorities, fi nancial regulators, and citizens at large, because an understanding of the signifi cance of these variables can help authorities make appropriate decisions to prevent and/or overcome problems related to municipal insolvency.


Introduction
 e economic recession that began in 2008 led to high levels of bank debt and budget defi cits in the public sector, reducing solvency and restricting access to the credit market, as well as jeopardizing the sustainability of public services.  e debt crisis produced great concern about credit risks, among policymakers, fi nancial regulators, and fi scal authorities. Researchers concluded that it was necessary to study the causes of high levels of local government (LG) default in order to design and implement corrective and preventive policies and thus put government fi nances on a sound footing and in a position to meet debt and defi cit targets ( Beetsma and Vermeylen, 2007).
 e debt crisis has been particularly worrying in countries such as Italy, Ireland, Portugal, Greece, and Spain. International organizations have concurred with academic studies that public debt in these countries is of such a magnitude that major problems of repayment may arise and that there are worrying diff erences between public revenue and expenditure (European Commission, 2012; Aldasoro and Seiferling, 2014; Balaguer-Coll, Prior and Tortosa-Ausina, 2016; Navarro-Galera et al., 2015).
In this context, taking into account that fi nancial institutions are the main creditors of LGs, researchers have considered it especially interesting to study the risk of loan default (Arbatli and Escolano, 2015; Balaguer-Coll, Prior and Tortosa-Ausina, 2016; Elgin and Uras, 2013; Guillamón, Bastida and Benito, 2011). Various studies, including Balaguer-Coll, Prior and Tortosa-Ausina (2016), Elgin and Uras (2013) and Geys and Revelli (2011), have analyzed the infl uence of variables such as absolute majority government and political fragmentation on the volume of bank debt and on sustainability. However, these papers examined only individual factors pertaining to each LG and did not address the causes of default risk.
Despite the valuable conclusions presented in the above papers, many organizations in the fi eld of public fi nances argue that a comprehensive analysis of government credit risk should include, besides political factors (which are specifi c to each entity and may be dependent on election outcomes), systemic ones, such as the macroeconomic cycle, fi scal policy, and the electoral cycle, in view of the vulnerability of individual government entities to macroeconomic changes and the volatility of non-controllable variables ( In view of the above considerations, the aim of this paper is to advance our understanding of the factors that infl uence LG credit risk. Assuming an interconnection between political decisions and economic management, we study the joint eff ects of political and systemic variables on the probability of bank loan default, following the Basel II rules (BCBS, 2006). To do so, we analyze 148 large Spanish LGs during the period 2006-2011, using a logit model with panel data and an artifi cial neural net-work. From the results obtained, we identify political and systemic factors underlying the risk of LG insolvency.  e conclusions drawn provide useful new knowledge for policymakers, managers, fi nancial analysts, regulators, national and international tax authorities, voters, users of public services, citizens at large, and other stakeholders.

 e impact of political and systemic variables on credit risk under Basel II rules
We study the causes of credit risk by considering the probability of default (PD) as a fi nancial indicator. In line with previous research in this fi eld, our analysis takes into account the defi nition of default, or breach of bank payment commitments, established by the Basel II Banking Supervision Commi ee (Castrén, Dées and Zaher, 2010;Bluhm and Overbeck, 2003;Gordy, 2003), according to which a higher probability of default is associated with a greater expected loss, a greater need for capital and, therefore, a higher risk-adjusted rate of interest.
Considering the diff erent default scenarios considered under Basel II, our study incorporates a dependent variable addressing these possibilities through an ability-to-pay process (APP), which measures LGs' capacity to meet their credit liabilities (Bluhm and Overbeck, 2003).  is APP depends on the quality of LG assets and fi nancial resources and is a latent random variable that is not directly observable, but which can be estimated using a nonlinear discrete-choice approach, namely the logit panel data model, which is an appropriate means of considering the factors that contribute to the likelihood of debt default (Bonfi m, 2009; Jacobson, Lindé and Roszbach, 2013). In addition, to study the phenomenon of government insolvency, we construct an artifi cial neural network (ANN), in the knowledge that previous researchers have used ANNs as a complement and an advance on parametric techniques. In most cases, this approach enhances the analytical process.
In this respect, the standard known as Basel II (BCBS, 2006) is a highly signifi cant advance in the international fi nancial system as this model helps ensure the soundness and stability of credit institutions, making it possible to assess the fi nancial risks associated with the institutions, such as governments, to which these entities make loans. In particular, the Basel II model seeks to ensure that banks implement new tools to strengthen the capital requirements arising from their credit risk operations, focusing both on the market and on operational aspects. Accordingly, the model proposes tools with which fi nancial institutions can estimate the default risk on the loans made to their customers.
According to Gordy (2003), a local government (LGi) is in default if its ability to pay at any given time APP it is below a certain level of credit liability (c it ). Under this approach, default by LG i in the period t is a random dichotomous variable Y it such that: where the probability of LG i default at time t is equal to Following the methods used in previous studies of credit risk in the business sector (Castro, 2013;Mileris, 2012), we analyze two types of credit risk factor, as variables expected to infl uence the probability of default: political factors and systemic factors. Political factors impacting on credit risk (Z it ) are specifi c to each local government. In contrast, systemic credit risk factors (X t ) include aspects of the macroeconomic cycle, fi scal policy, and the electoral cycle, with respect to the country.
In this paper, the following initial premises are assumed: (1) local governments form a homogeneous segment within the public sector; (2) the systemic factors X t that infl uence credit risk aff ect all local governments at time t (t = 1, … , T); (3) the idiosyncratic political factors Z it (i = 1, … , Nt, t = 1, … , T) that infl uence credit risk individually aff ect each LG i ; (4) the idiosyncratic LG factors (individual or LG-specifi c factors) are not entirely independent of systemic eff ects, an aspect that is particularly signifi cant in economic recessions (Bonfi m, 2009).  us, the APP it variable of the i-th LG at time t is a function of the political and systemic variables, as in expression (3): where β j and δ k are the parameter vectors estimated by a linear panel data model and u it is the random perturbation.  en, following Rösch (2003) and Bonfi m (2009) a borrower L g is considered to be unpaid if its APP it falls below the level c it (credit obligations). Although the variable APP it is latent and not directly observable, the explanatory variables X t and Z it and the systemic and political factors, together with the independent binary variable Y it , the indicator of default, are directly observable, from the sample data. In our study, credit risk was determined via the estimation of PD, using a logistic regression model and an ANN.  is procedure was adopted for several reasons. First, discrete-choice models are held to be appropriate when the research aim is to analyze the determinants of the probability of an individual economic agent (Jacobson, Lindé and Roszbach, 2013). Second, the two models we use meet all the statistical requirements specifi ed in the Basel II rules (BCBS, 2006) and in Bank of Spain Circular 3/2008 for calculating PD.  ird, the European Commission (2015), Aldasoro and Seiferling (2014), the World Bank Group (2015) and the US Department of the Treasury (2013) have all recognized the need to study the joint eff ect of idiosyncratic factors (or individual ones, for a single entity) and systemic factors (such as the macroeconomic cycle, fi scal policy, and the electoral cycle) in the measurement of government credit risk.

Sample selection
 is empirical study focuses on large LGs in Spain, in the view that international organizations (European Commission, 2012; Aldasoro and Seiferling, 2014) and previous research (Balaguer-Coll, Prior and Tortosa-Ausina, 2016; Navarro-Galera et al., 2015; Guillamón, Bastida and Benito, 2011) have all concluded that bank debt in local and regional governments in Spain is too high (in fact, it is among the highest in the Eurozone).
Following Guillamón, Bastida and Benito (2011) and the criteria of the Local Government Modernization Act no. 57/2003, we selected 148 municipalities, all with over 50,000 inhabitants or, otherwise, provincial capitals, taking data for the period 2006-2011, i.e., from three years before the economic crisis until three years a er its appearance.
 is sample is appropriate for our research aim for the following reasons: (a) the introduction to the Local Government Rationalization and Sustainability Act no. 27

The dependent variable
In this study, the dependent variable is the Y it of the LGs in the sample, assessed according to the four fi nancial indicators that determine when APP i is lower than the credit liability, according to the defi nition of default in the Basel II rules and in accordance with Spanish legislation on local government (Royal Legislative Decree 2/2004, of 5 March, approving the consolidated text of the Local Finance Regulatory Act). Hence, following the criteria used in previous studies of local government fi nance (Moody's, 2013, 2008; Navarro-Galera et al., 2015), we assume that APP i is lower than the credit liability and, therefore, that a local government is at risk of default (or at risk of defaulting on loan repayment obligations) when it meets at least one of the situations identifi ed by the four fi nancial indicators stipulated in Table 1, in line with the recommendations of the BCBS (2006). Table 2 shows the political and systemic variables analyzed jointly in this study, together with their defi nitions and the expected sign of the relationship between the explanatory variables and PD. As discussed below, these variables were selected taking into account previous research on LG debt and ratings (Gaillard, 2009)     Ministry of Finance and Public Administration +

Political context
Ideological alignment between the city council and the regional government LGs. In this area, neither Massolo (1991) nor Guillamón, Bastida and Benito (2011) found any evidence that the mayor's gender might infl uence the evolution of the volume of LG debt. In the present study, following Navarro-Galera et al.
(2017) and Benito, Bastida and Muñoz (2010), we include three possible explanatory variables with respect to the mayor's profi le: gender, university background and academic profi le (studies related to economics). Based on previous research fi ndings, we expected a negative relationship between these variables and PD, i.e. when the mayor is male, has no university degree or has a degree in a subject unrelated to economics, each of these circumstances will increase PD.
Regarding the LG profi le, we selected fi ve variables for analysis, in line with previous research fi ndings. In this respect, Lago and Lago (2009) and Ashworth, Geys and Heyndels (2005) all concluded that political competition (defi ned as the absence of an absolute majority in the municipal corporation) is associated with greater fi scal pressure, higher levels of defi cit and debt, and fewer available resources.  ese fi ndings suggest that political competition may have a harmful impact on PD. Taking into account that political competition could increase the volume of debt, we expect to fi nd a positive sign in the corresponding estimator; in other words, governments with no absolute majority are more likely to experience loan default.
Geys and Revelli (2011) and Roubini and Sachs (1989) concluded that greater political fragmentation can worsen the budget defi cit and, therefore, increase municipal debt. Consequently, we examine the specifi c infl uence of this variable on PD and expect to obtain a positive sign in the estimator, showing that the greater the fragmentation, the greater the PD.
Various studies (Cabaleiro-Casal and Buch-Gomez, 2015; Guillamón, Bastida and Benito, 2011; Lago and Lago, 2009; Ryan, Pini and Brown, 2005; Ashworth, Geys and Heyndels, 2005) have concluded that greater strength in the governing political party is associated with higher levels of debt and defi cit. However, Galli and Padovano (2002) found that a lower degree of political strength leads to greater indebtedness and, therefore, to greater diffi culty in meeting loan maturities. To resolve this question, we use the Herfi ndahl index to examine whether the political strength of the governing party aff ects PD; in this respect, either a positive or a negative sign may be expected in the estimator.
Another interesting political factor is that of gender. Piotrowski and Van Ryzin (2007) and Jennings (1983) concluded that male councilors tend to present a more active political commitment than their female counterparts. In another study, Geys and Revelli (2011) observed that the presence of a higher proportion of women in the governing party is associated with higher municipal tax revenues, which suggests that the fi scal pressure and the defi cit would be lower in these cases. Although the la er studies did not analyze the infl uence of the percentage of female councilors on PD, their fi ndings suggest that this variable may be of interest, and therefore we included it for analysis, expecting a negative sign in the estimator, in the sense that the greater presence of women in the governing party would be associated with a lower PD.
According to some studies, the political ideology of the governing party (conservative versus progressive) can infl uence fi nancial decision making (Cabaleiro-Casal and Buch-Gomez, 2015; Bastida, Benito and Guillamón, 2009). In this respect, Balaguer-Coll, Prior and Tortosa-Ausina (2016) and Bel and Miralles (2010) concluded that le -wing parties are more likely than conservative ones to adopt expansive spending policies, thus increasing levels of debt and leading to greater diffi culties in meeting payment obligations.  erefore, for this variable we expect to fi nd a positive sign in the estimator.
Finally, Solé-Ollé and Sorribas-Navarro (2008) studied variables related to the political context and reported fi nding a positive relationship between the ideological alignment of local and regional governments and the volume of transfers received.  is circumstance could reduce the need for debt and, consequently, the risk of insolvency. However, Bastida, Benito and Guillamón (2009) recorded a negative infl uence between this ideological alignment on spending and fi scal revenues. In the present study we analyzed two variables, corresponding to the ideological alignment of the local government with the regional government and with the central government, expecting to fi nd that, in each case, this would increase PD.

Systemic variables
Regarding macroeconomic variables related to the political cycle and to monetary and fi scal policy, Van Der Burgt (2009) argued that, under the Basel II regulatory framework, PD should be calculated as a long-term average of default rates, in the view that this time scale corresponds to that of the economic cycle.  erefore, and following Balaguer-Coll, Prior and Tortona-Ausina (2016), we employed a dichotomous variable (the economic cycle) that defi nes the periods of expansion or recession of macroeconomic indicators.  e la er authors suggested that this variable could infl uence the volume of debt and be related to LG insolvency; as such, this variable was included for analysis as a possible determinant of PD, and its sign was expected to be positive (thus, in the years of economic expansion, PD would be lower). In addition, Arbatli and Escolano (2015), Navarro-Galera et al. (2015) and Balaguer-Coll, Prior and Tortona-Ausina (2016) have suggested that a higher level of economic activity and a higher rate of national unemployment may be associated with increased LG debt.  erefore, our empirical study included the variables rate of growth of the national economy (GDP) and rate of unemployment, expecting a negative sign in the fi rst case and a positive one in the second.
 e electoral cycle has also been identifi ed as one that may infl uence LG spending. According to Blais and Nadeau (1992), LGs spend more in the years preceding electoral processes and less in post-electoral years, which could aff ect levels of government defi cit and debt. If this were so, the electoral cycle could aff ect PD, and we included this variable for analysis, expecting a positive sign for its estimator. Finally, statements by agencies such as the IMF (Aldasoro and Seiferling, 2014) and the European Commission (2015), as well as previous research fi ndings (Mackey, 2014) suggest that the national risk premium should be included for analysis, and that a positive sign should be expected, i.e. a higher risk premium is associated with a greater likelihood of loan default.

Statistical models
 e empirical results required to calculate local government PD were obtained using a parametric technique, that of logistic regression with panel data, together with a non-parametric one with a particular ANN, the multilayer perceptron (MLP).
 e main purpose of our logistic regression model is to accurately predict the outcome category for individual cases, using the most parsimonious model. To achieve this goal, we designed a model that included all the variables expected to be useful for predicting the dependent variable.  e variables can be introduced into the model by stepwise regression, following the order specifi ed in previous studies, and testing the fi t of the model a er the inclusion of each coeffi cient.
 e logit panel data parametric technique is used to analyze the correlation between unobserved factors. According to Train (2003), this approach eliminates the bias derived from the existence of unobservable and time-invariant heterogeneity, making it very appropriate for the characteristics of our sample. Since the dependent variable is binary, and starting from equations (2) and (3) Next, a particular artifi cial neural network, the MLP, was designed with a sigmoid activation function, calculated with the logistic activation function .  is is also used in the hidden layer of the MLP, taking arguments of real value and transforming them into the range (0,1).  e output layer contains the target (dependent) variables. Accordingly, the output of the neural network, from a vector of inputs (x 1 ,…,x p ), is: Finally, we designed two statistical models with the following characteristics: (a) in Model 1, the independent variables are exclusively the political variables (pro-fi le of the mayor, profi le of the government and political context); (b) in Model 2, the independent variables are the same political variables plus the systemic ones (macroeconomic variables, electoral cycle, and fi scal policy).

Analysis of the results
Our empirical results show that 486 cases (54.73 per cent) met the default condition, and that 402 (45.27 per cent) did not. Table 3 shows the estimated coeffi cients transformed into odds ratios or Exp (β) of the logistic regression with random eff ects, both for Model 1 (political variables) and for Model 2 (political and systemic variables). Note: (*) , (**) , (***) represent signifi cance at the 10 per cent, 5 per cent and 1 per cent levels.

Source: The authors
 e MLP results for the normalized importance of each variable (Figure 1) show that stronger weights are assigned to systematic variables and other variables found to be signifi cant when logistic regression by panel data is applied.  erefore, the variables included as signifi cant in Table 3 are those with the greatest explanatory capacity of local government PD, which demonstrates the robustness of the values and signs obtained.
Moreover, as shown in Table 5   e results obtained for Model 2, with each of the two techniques used ( Table  3), show that four variables are of a systemic nature and four are political, which confi rms the joint, balanced impact of both types of factors on PD. However, if we consider the number of variables of each type, the representativeness of the systemic variables (four out of fi ve) is greater than that of the idiosyncratic ones (four out of ten), which suggests that the impact of non-controllable factors is greater than that of controllable ones.
Our analysis by individual variables shows that the mayor's having a university degree in an economics-related subject is a signifi cant variable, with a sign equal to that expected; in other words, when a local government mayor has this educational background there is a lower probability of default.  is fi nding advances upon the conclusions of previous studies (Benito, Bastida and Muñoz, 2010; Guillamón, Bastida and Benito, 2011), which did not examine the relationship with PD, but only analyzed the volume of debt.
Second, we fi nd that the proportion of female councilors in the municipal corporation is inversely related to PD, which implies that increasing the number of women in the LG team could reduce the risk of default. Previous research has only reported evidence of the eff ect of this variable on tax revenues (Geys and Revelli, 2011), and therefore our results are novel, corroborating the specifi c infl uence of female representation in the municipal government on the risk of PD.
With respect to political ideology, we fi nd that LGs with a progressive ideology are associated with a greater PD, which extends the conclusions of previous research With respect to public policies, a comprehensive analysis of government solvency on the basis of default risk should include, as well as variables refl ecting the decisions of policymakers and citizens (idiosyncratic variables), systematic factors whose behavior is not subject to these circumstances, but which nevertheless should be understood and taken into account, in order to eff ectively manage credit risk. Our fi ndings suggest that policy makers need to be well informed about the detrimental eff ects on public ser-vices of an increased risk of municipal default, which may occur when policymakers do not have a background in economics, when the government has a progressive ideology and when female councilors are under-represented in the corporation. If politicians are sensitive to the possible negative consequences of these circumstances, a more prudent and viable approach may be adopted in their fi nancial planning.
With respect to the infl uence of the systemic variables considered, we identify four that may aff ect LG credit risk. As shown in Table 3, the fi rst systemic variable considered is that of the growth rate of the national economy (i.e., GDP). In this respect, a signifi cant, inverse association was observed; this was in line with our expectations and suggested that a fall in the GDP may increase LG credit risk.  is fi nding contrasts with previous studies (Arbatli and Escolano, 2015; Balaguer-Coll, Prior and Tortosa-Ausina, 2016), which reported the infl uence of GDP on the volume of government debt, but did not study its eff ect on PD. Another statistically signifi cant variable is the national unemployment rate.  is variable, too, was found to be inversely associated with PD, which is in accordance with previous research fi ndings (Arbatli and Escolano, 2015;Navarro-Galera et al., 2015).
 e third signifi cant systemic variable is the national risk premium, which presented a positive sign, as expected.  is result empirically corroborates the conclusions of Aldasoro and Seiferling (2014), the European Commission (2015) and previous research (Mackey, 2014), and suggests that an increase in the risk of sovereign debt is eventually refl ected in a greater PD in LGs, given the interrelations between local, regional, and central governments, and the current and capital transfers made. , we assume this factor is non-controllable. Our analysis also shows that the electoral cycle may infl uence the default risk of local governments (specifi cally that PD may increase during pre-election periods) as observed previously by Blais and Nadeau (1992).
 ese fi ndings for the systematic variables are relevant to the decisions of policy makers.  ese variables cannot be directly controlled by politicians and their significance leads us to consider that annual budgets should be prepared taking into account the foreseeable evolution of the macroeconomic variables, by means of multiannual planning techniques and scenario analyses. Trends in these variables should be taken as warning signs, with the potential to impact on access to bank credit and on the cost of borrowing.
Finally, these fi ndings may be useful for managers in other countries, too.  e signifi cance of the idiosyncratic variables considered (the ideology of the governing party, the proportion of female councilors and the educational background of the mayor) is in line with the conclusions of previous work on debt and fi nancial sustainability, carried out in countries such as Netherlands, Finland, Austria, Switzerland, Italy, Germany or Belgium (Jennings, 1983;Geys and Revelli, 2011). In addition, the explanatory variables (such as unemployment and the risk premium) were selected taking into account previous research in this fi eld in countries such as Norway, Italy or Canada (Arbatli and Escolano, 2015;Blais and Nadeau, 1992), bearing in mind that their eff ects tend to be generalized.

Conclusions
 e present study highlights the interconnection between political decisions and economic-fi nancial management.  e empirical results obtained provide novel evidence of the infl uence of political and systemic variables on the risk of LG loan default.  e results show that some political variables (the mayor's lack of economics-related university studies, a low proportion of female councilors in the municipal corporation, the progressive ideology of the governing party and ideological alignment between the municipal and the regional government) may contribute to increasing the probability of loan default, thus heightening the fi nancial risk and problems of insolvency.
Our joint analysis of political and systemic variables shows that the la er are also relevant to the risk of default.  us, a rise in the national unemployment rate and in the sovereign risk premium, together with the impact of the electoral cycle and of declining economic growth, may all increase PD in local government.  ese fi ndings suggest that the infl uence of political factors, which in turn depend on the decisions taken by citizens in electoral processes, may be aff ected by the evolution of systemic variables, which do not depend on these decisions.
 is fi nding represents the empirical corroboration of statements by international organizations (European Commission, IMF and the World Bank) and of prior research in this fi eld, in that the insolvency problems facing governments arise from both controllable and non-controllable factors, such as systemic variables.
Our joint study of the infl uence of political and systemic variables on credit risk provides more complete and representative results than those derived from the individual analysis of political factors.  e incorporation of systemic variables into the analysis has produced a very signifi cant advance in our understanding of these questions.  us, the fi nancial risk behavior of LGs is subject to considerable uncertainty and the probability of default is aff ected by variables whose evolution is beyond the immediate control of policymakers, such as changes in the unemployment rate, in the risk premium and in the growth of the national economy.
In short, although systemic variables -which escape the control of citizens and of local governments -exert a signifi cant infl uence on default risk, the political decisions of voters, expressed in electoral processes, also impact on this risk, as do factors such as the political ideology of the governing party, the educational background of the mayor and the gender of municipal councilors.
Finally, our fi ndings represent an advance on those of previous research into the question of government debt. Although earlier studies have reported the impact of political ideology and of ideological alignment between local and regional government on the volume of LG debt, our fi ndings specifi cally identify the infl uence of these factors on the risk of debt default. On the other hand, we found no evidence of the infl uence on the risk of default by the mayor's gender or education background, by political fragmentation or by the governing party's ideological alignment with the central government.
 ese fi ndings can be very useful when policy makers must take concrete decisions, for the following reasons: (a) in the fi nancial planning of public policies, the eff ect of variables that are not controllable by the rulers must be recognized, measured and included in the budget; (b) to avoid or mitigate the risk of default, policy makers must be well informed of the detrimental eff ects on the viability of public services of factors such as the mayor's lack of university studies in an economics-related subject, governance by a party with a progressive ideology and the under-representation of female councilors; (c) the annual budget must be prepared taking into account the foreseeable evolution of macroeconomic variables, using multiannual planning techniques and scenario analyses; (d) the eff ect of political variables on the risk of default may be contrary to that of the systematic variables, and this in turn would impact on the fi nancial planning of public policies.