Political Budget Cycles in Early Versus Regular Elections: The Case of Serbia

Abstract

The paper analyses Political Budget Cycles in the context of a young post-communist democracy, Serbia. The authors deploy well-established methodological (time series) approaches to examine the general government budget balance (fiscal deficit) in conjunction with elections. The findings suggest that there is clear evidence of higher fiscal deficit prior to elections—however, this is the case only for regular (scheduled) elections and not so for snap (early called) elections. The paper contributes to the PBC literature by revealing different incumbent behaviour in regular versus early elections, thus highlighting the importance of distinguishing between these types of elections in the domain of PBC research.

Introduction

Before elections, an incumbent government may engage in expansionary fiscal policies, to improve the economic situation in order to maximize electoral outcome, giving rise to the so called Political Business Cycles. Since the publication of the seminal paper of Nordhaus (1975), there has been a growing body of empirical research on this topic.

Traditional Political Business Cycle theory characterizes politicians as opportunistic. The driving force incentivizing both the incumbent’s behaviour and their economic policies is to remain in power. Voters are assumed to be both myopic and naive and as such prone to vote for incumbents when times are good prior to the election. Hibbs (1977) provided an alternative view incorporating partisan preferences over policy outcomes, arguing that left-wing parties tend to prefer low unemployment at the expense of higher inflation, while right-wing parties have the opposite preferences. While a partisan approach can fit well to (many) western countries or industrial democracies, there is a common understanding that the opportunistic model best describes (some) developing countries, where elections are often considered referenda on specific rulers and recent economic conditions (Block 2002).

The incumbent government may use monetary as well as fiscal policies/instruments to stimulate the economy in conjunction with elections. In the case of monetary instruments, the incumbent can persuade the central bank to engage in expansionary monetary policies prior to elections, known also as political monetary cycle (PMC). While the occurrence of PMC requires that the incumbent has certain levels of control over the central bank (which is not always the case), the alternative economic policy instrument which is under direct control/domain of the incumbent government is fiscal policy. The government can engage in expansionary fiscal policies, which can take the form of lower taxation and/increased spending prior to elections, which eventually results in increased debt—this phenomenon is known as Political Budget Cycle (PBC) (Lami and Imami 2019).

PBCs are more pronounced in emerging economies or new democracies. Brender and Drazen (2005), analysing 68 low- and high-income democracies, demonstrate that new democracies are more vulnerable to PBCs, resulting in higher public spending and larger deficits before elections. Klomp and de Haan (2013a, b), based on panel data analysis, confirm the existence of PBCs being reflected in fiscal balances (as well as public sending) both in young and old democracies. They find far stronger political cycles in the latter category.

Lack of (or limited) voter experience can explain why voters in new democracies do not behave rationally or punish such incumbent behaviour (Praščević 2020; Hanusch and Keefer 2014; Rose 2006; Shi and Svensson 2006; Brender and Drazen 2005). On the other hand, this is more likely to be the case in developed economies where voters have experience and understanding of the costs of PBC. Thus, in this context, fiscal manipulation in advanced democracies may be punished rather than rewarded (Brender and Drazen 2005). Differences in a PBC pattern or its magnitude between developing and developed countries may be due to differences in institutional environments, which are related to corruption, rent-seeking activities, and access to free media (Shi and Svensson 2006).

Alt and Lassen (2006a, b) highlight the importance of transparency—the higher the degree of fiscal transparency the lower are public debt and deficits during electoral periods. In this context, the media play a crucial role in providing trustworthy and accurate information to citizens (Hong 2016). New democracies are often characterized by an underdeveloped media sector which is under the strong influence of the government and characterized by non-transparent media ownership and money flows associated with high levels of corruption. These are the characteristics of the media market in Serbia (see Kisić 2015).

PBC patterns may be influenced by the timing of elections, e.g. whether elections are called earlier than it has been scheduled (snap/early elections) or on time (scheduled regular elections). In some countries (e.g. the USA), election timing is imposed constitutionally. In many countries early elections are possible and often common. If election timing is not set for a fixed date by the constitution, the incumbent can call early elections for political or economic reasons (Lächler 1982). This implies that the incumbent, instead of engaging in expansionary economic (e.g. fiscal) policies, may use a positive (external) sector supply shock (e.g. high growth, low unemployment, and inflation) as an opportunity to call an early election (Ito and Park 1988), whereas when early elections are called for other reasons (e.g. loss of confidence or a coalition breaks down), the incumbent may not have sufficient time to plan and implement expansionary polices—vice versa, if elections take place as planned (the case of regular elections), the incumbent government has more (sufficient) time to plan and execute expansionary policies with the aim that they will yield the expected “positive effect” in the economy prior to elections¹. Thus, it is very important to distinguish between regular (scheduled) and early (snap) elections when analysing PBCs (Imami et al. 2020).

In our paper, we empirically investigate PBCs in Serbia, a post-communist economy, characterized by a limited democratic history and voter experience, a weak institutional framework, and a high level of corruption, as explained in the following section. This paper adds value to the literature from several perspectives. First, this is the first paper to target PBC phenomena in Serbia. Until now, Serbia was rather part of a broader analysis of transition countries (see Pavlović and Bešić 2019) but previous studies do not provide insight on the occurrence of PBC. The case of Serbia is of interest also because it is one of the last countries embracing transition from a planned economy and dictatorship to market economy and democracy. Second, the analysis puts an emphasis on distinguishing between early/snap elections and regular elections in conjunction with PBC, which is relevant and of interest for other countries too. Even in the context of a relaxed/weak institutional environment, which characterizes Serbia (see more details later in the paper), the incumbent faces limitations on the timely execution of fiscal expansionary policies. It is even more important, since the changes in the institutional framework for conducting monetary policy considerably reduce the possibility of monetary policy misuse, also in transition economies (Praščević 2020). The findings suggest that the elections-related deterioration of the fiscal balance is clearly present in Serbia—this phenomenon is driven by regular (scheduled) elections, while there is no significant increase in the deficit prior to early (snap) elections. Our findings are in line with previous research dealing with PBC in context of weak institutions and now democracies (see, for example, Hallerberg et al. 2002; Brender and Drazen 2005; Block 2002). Moreover, the use of recent monthly data provides more accurate insight into the PBC dynamics.

The structure of the paper is as follows. The next section describes the country context by presenting an extensive analysis, which is important to understand the political, institutional, and economic context, and also provides useful information per se, given the scarcity of such insights about Serbia in the academic literature. The third section describes the methodology, the fourth section provides the findings while in the last section concluding remarks are presented.

Country Context

In the early 1990s, the Yugoslav state entered political and economic chaos ending up in the bloody decay of the Yugoslav socialist state, where the Federal Republic of Yugoslavia (FRY), comprising Serbia and Montenegro, became a successor. While all the other former communist countries were confronted with steep economic downturn at the beginning of transition, in the case of FRY it was even more pronounced: it was facing economic and political isolation, destructive economic policies, and a hostile political and regulatory environment. This all culminated in the Kosovo war and subsequent NATO bombing of FRY in 1999. When the communist block started to collapse in 1989, FRY was in relatively favourable starting conditions, including substantial experience with market-oriented reforms, major economic (and political) decentralization, greater openness to the West as well as better integration in world economy (Uvalić 2007), although still belonging (arguably) to more rigorous communist regimes (Havrylyshyn 2006). Nevertheless, Serbia entered transformation towards a market economy and democracy as a latecomer and under highly unfavourable conditions at the beginning of the third millennium. All the fundamental policy inputs for achieving desirable economic and social goals: market transformation, state of democracy, and media freedom, were at an extremely low level (Havrylyshyn 2006). Vojislav Koštunica won the presidential election against Slobodan Milošević on 24 September 2000, while the democratic opposition took 2/3 of the seats in the parliamentary elections on 23 December 2000. At that time, Serbia was one of the poorest European countries with GDP at half of the level in 1989. Moreover, within all former Yugoslav republics, only Serbia and Bosnia and Herzegovina have lower level of GDP when compared to the time before the Berlin wall fall down (Jović 2022).

One may argue that Serbia as a latecomer had a chance for more efficient transition, avoiding the mistakes made by other transition countries and learning from their experience. This did not happen. The countervailing forces were stronger: path dependency, adverse initial conditions, and broad and deeply rooted institutional failure. The informal institutional structure, without the intellectual and cultural roots in classical liberalism “based on the principles of self-interest, self-determination, self-responsibility and free market competition” (Pejovich 2003, 350), made an additional long-lasting burden in the successful transformation towards a market economy and supporting political institutions.

While the transition during the 1990s may be characterized as a blocked transformation, retaining a lot of former communist characteristics and only a facade of democratic institutions (Trifunović 2015), the 2000s marked real transition towards a market economy. However, while the initial conditions as well as political and economic heritage were seen as a disadvantage, the institutional setting formed during transition was equally disappointing. State capture, weak rule of law, uneven distribution of political and economic power created strong incentives towards rent-seeking and partisan behaviour, leading to high economic uncertainty and political instability (Resimić 2022).

Two rather different periods of institutional development may be identified since 2000. The first period is between 2000 and 2012. During this time Serbia was confronted with several tectonic political shifts: the assassination of the Prime Minister and leading figure of the democratic movement, Zoran Đinđić (2003); reform of the joint state with Montenegro, which was an introduction to its dissolution (2006); the non-transparent and accelerated process of adoption of a new constitution (2006); and unilateral proclamation of Kosovo independence (2008) (Kovačević 2019). Internal economic conditions were characterized by high unemployment and slow pace of reforms and were additionally worsened by the Global financial crisis in 2008–2009 causing a recession in 2009, making the government-debt-to-GDP ratio unsustainable (Andrić et al. 2016).

In this period, progress was made in setting up and upgrading the institutional structure (Vladisavljević 2019) as reflected in the rising values of democracy indices such as Freedom House or Bertelsmann Foundation, including the most important formal rules of a market economy—credible and stable property rights, freedom of contract, an independent judiciary, and constitutional protection (Pejovich 2003). However, most of this progress remained to a large extent non-functional and the discretionary power of bureaucrats were not reduced. The end result was a captured state, flourishing corruption and an institutional structure that amplified it. This development demonstrates an important feature to be observed in many failed transition countries: de jure institutional building was not followed by de facto institutional development.

Serbia, at that time, was one of the few countries in the world in which the seats in Parliament belonged to a party and not to elected politicians (Novaković 2010). The supremacy of the party over institutions was achieved. Party domination was additionally fuelled by a high level of state centralization. The government directly controlled 17 public companies, some of them being the largest companies in Serbia, and many agencies, foundations, and services. Overall, it is estimated that there were about 40,000 positions in Serbia to be appointed by party officials of the ruling parties at all levels (Pešić 2007).

Alongside this centralized state comes an authoritarian and oligarchic organization of the participants in the political market. To accomplish partisan goals—whether it was employment of party people, friends, and relatives in the public sector, or “exchange services” with tycoon businesses, party leaders usurped state institutions, and functions by politicizing them and preventing their impartial and legal action. The basic democratic institutions have also been shaken, and the power of the parliament has weakened most in relation to the power of the executive. In consequence, the political system became overburdened because it performed functions that do not belong to it, which undermines democratization, political, and economic competition (Pešić 2007).

The combined factors of political instability, financial crisis, slow process of European integration, and overall failure of a new political elite after 2000 to establish democratic and independent institutions (Bieber 2018), and the slow pace of economic reforms, where the private sector share was only 60% of GDP in 2010 compared with 40% in 2000 (Uvalić 2013), strongly influenced the change in government in 2012, where the second period of institutional transformation started. A younger generation of Milošević-era politicians gained power and found unrestricted access to public funds in a weak institutional environment (Pavlović 2020). An old elite from the 1990s came back with a changed political and economic narrative. Almost from the beginning of this period, Serbia started moving from moderate party pluralism towards a one-party system. Political and economic pressures on media started to rise, and the independence of institutions, pressures on judiciary, and the fight against corruption were crumbled further (Lončar 2017). Control over the media goes hand-in-hand with economic incentives especially where the role of public resources comes to light: whether through direct financing (for example, regional TV Novi Pazar) or tax write-offs (for example, TV Pink) (Pavlović 2020).

Everything is subordinated to the interests of the incumbents. For example, ruling parties were abusing the governance of the public company “Pošta Srbije”, which refused to post the list of opposition parties 10 days before the elections in 2016. The “catch all” strategy of the incumbent party (Spasojević 2017) spurred vast misuse of resources because of the many diverse interests to be satisfied. The unresolved Kosovo question, migration crisis, the almost frozen process of European integration, and finally the COVID-19 pandemic additionally diverted attention from economic issues.

With an uneven distribution of power, along with institutional weaknesses came an excessive concentration of power and “presidentialization” of politics, although the electoral and political system provides incentives for a more balanced power sharing between different political actors (Spasojević 2021). The newly formed political landscape is described as “competitive authoritarianism” (Castaldo 2020), a hybrid system where de jure democratic procedures and de facto authoritarian rule merge with an ultimate aim of conquering and consolidating power. The abuse of economic policy and public resources to promote the ruling parties, especially during election campaigns, has intensified. Ruling parties have become mechanisms for clientelist networks, substituting or replacing formal institutions with negative impact on long-term democracy development and state capacity. The political party (organization) in power is not only the instrument of power but is also a central force in maintaining and protecting power (Günay and Dzihic 2016). The result has been the restructuring and rebuilding of the political system to serve the incumbent party interests, similar to tendencies in Poland, Hungary, or other neighbouring Balkan countries (Keil 2018). Dismantling the democratic institutions was followed by a weakening of the legitimacy of the electoral process. Participation of citizens in the electoral process is in constant decline: it was relatively high in 2008—reaching over 60%, and it dropped to 48.93% in 2020 (Ilić 2021).

While the privatization process is almost finished, an important channel for misuse of public resources after 2012 continued through secret contracts where the most important information on rights and obligations between the state and private investors are hidden. Secret contracts are against constitutional provisions (Article 51), according to which public investment contracts are not allowed to be hidden or blocked (Pavlović 2020).

The weak institutional development and impeding political environment were followed by poor economic performance. Since 2000, Serbia has had four recessions: 2009, 2012, 2014, and 2020. All four downturns were initiated by exogenous shocks—global financial crisis, European crisis of sovereign debt, and COVID-19 pandemic, but further worsened by internal economic and institutional drawbacks. That has resulted in increased need for external funding as well as higher public debt.

Through the whole period after 2000, Serbia was never an over-indebted country. At the beginning of transition there were generous write-offs of a large part of liabilities by the London and Paris Club (Đukić and Nikolić 2012), but the global financial crisis in 2008 caused significant deterioration of the indebtedness position. The foreign debt of Serbia during this period reached about 80 per cent of GDP. Also, according to the ratio of external debt to the value of exports of goods and services Serbia was a highly indebted country until 2011; however, after this period, due to the strong inflow of foreign direct investments and an increase in exports, it switched to being a medium indebted country (Božić Miljković 2019). In the years immediately following the global financial crisis, the level of indebtedness became a lesser problem compared to its growth dynamics. Between 2009 and 2012, the absolute value of debt doubled (Šojić 2013).

Weak institutions not only enabled rent-seeking activities, but they also attracted political agents who were prone to using public resources for private gains, so dismissing pluralistic institutions and violating democratic procedures (Pavlović 2020)². Weak institutions favour powerful incumbents (Havrylyshyn 2006), which in turn creates a highly constraining vicious circle damaging long-term economic performance. With political agents, where weak vertical and even more squeezed horizontal accountability exists, the use of public resources for partisan purposes is a standard tool in maximizing and consolidating political power. The State became a “money wasting machine” (Pavlović 2016) hampering long-term productivity and economic growth.

In this institutional and political context, the incumbent is expected to have strong incentives and relatively high degrees of freedom to engage in PBC. Higher public spending prior to elections may provide tangible benefits to voters (e.g. through increased employment), while higher spending means also an opportunity to incentivize or fuel clientele (such as businesses attached to the media, that may benefit somehow from public spending) to increase their support for the incumbent.

Data and Method

We statistically test the hypothesis that the general government budget balance (fiscal balance) deteriorates significantly before general (parliamentary) elections in Serbia and such deterioration is attributed to regular (scheduled) elections only—not to early (snap) elections. Fiscal balance is defined as general government budget revenues minus general government budget spending. If revenues are smaller (greater) than the respective spending, then the budget is running a fiscal deficit (surplus). When observed in annual terms, Serbia’s budget was in deficit for 14 out of 17 fiscal years and in surplus for the other three. Observations on a monthly basis shows the government ran a deficit in about 69 per cent of the observed months (143 out of 206) and a surplus for about 31 per cent. To test our hypothesis, we employ monthly time series data on the overall budget balance obtained from the government fiscal statistics.³ Monthly data, in addition to providing more robust statistical results, due to a higher number of observations (compared to annual data), most importantly allows for the inclusion of any intra-annual election effects. Empirical analysis based on annual data has been a serious drawback of many empirical studies analysing several aspects of PBCs (see Streb et al. 2012; Akhmedov and Zhuravskaya 2004). On the other hand, one of the potential problems associated with monthly time series (or, generally, with any intra-annual frequency data) is the possible existence of seasonality patterns, which if not addressed could distort the results. We address this potential drawback, as explained below.

The available time series of the fiscal balance includes 206 observations, from January 2005 to February 2022. The data are denominated in billions of Serbian Dinar (RSD) and deflated by the Consumer Price Index (CPI). In our empirical analysis strategy, we distinguish between early (snap) elections and regular (scheduled) elections which as a rule should take place every four years.⁴ Six parliamentary (general) elections were held during this period—out of which three were regular elections and three early elections—whose expected effect on fiscal balance is statistically captured by several dummy variables, constructed as explained below. Parliamentary elections were held on 21 January 2007 (regular)⁵; 11 May 2008 (early); 6 May 2012 (regular); 16 March 2014 (early); 24 April 2016 (early); and 21 June 2020 (regular)⁶.

We test the hypothesis of this paper by utilizing Intervention Analysis as the main econometric tool, which is based on the Box and Tiao (1975) methodology. This econometric approach has been applied in several similar works on PBCs or other fields with the same statistical inquiry objective of analysing the impact of “a known event” on a social or a natural time process.⁷ There are not many appropriate controlling variables available at a monthly frequency for this analysis. Therefore, another main reason we employ Intervention Analysis as our primary statistical framework is its advantage of enabling reliable econometric modelling even in the absence of such explanatory variables, as the time process could be modelled by its own autoregressive and moving average components (ARMA). However, as explained below, we conduct robustness checks for our findings by replicating the analysis using linear regressions, and by estimating models with the data collapsed to quarterly frequency so we can utilize additional and more appropriate control variables available at quarterly intervals.

Basically, the test in the Intervention Analysis proceeds by modelling the variable of interest (i.e. the fiscal balance) by an appropriate autoregressive moving average model (ARMA) and an intervention term. The intervention term models the time distance to each election day and captures any potential effect of elections on the variable of interest. The intervention term that models “the event”—the approaching elections in this case—could be considered as an explanatory variable capturing the dynamics of the dependent variable in addition to its “natural” pattern, which is modelled by an ARMA(p,q) specification (where p refers to the order—number of lags—of the autoregressive component, and q to the order of the moving average component). Intervention terms employed in this analysis consist of several dummy variables modelling different periods before and after elections. We call these variables “Electoral dummies” (EDs). Therefore, if the estimated parameter of a particular ED variable was to both prove statistically significant and have the anticipated sign, that would be considered as empirical evidence in support of the hypothesis of this study.

We define four ED variables for different time intervals preceding elections (all elections) and four others for symmetrical time intervals after elections. Likewise, we also define ED variables separately for each category of elections part of our hypothesis, namely for regular and early elections. Each set of EDs is formally defined as follows:

• (i) EDs for “all elections” (regular and early elections altogether)

  $$ED_{ \pm j,t} = \left\{ {\begin{array}{*{20}l} {1:} \hfill & \begin{gathered} for\;all\;months\;up\;to\;and\;including\;the \pm jth\;month \hfill \\ before\;( - j)\;or\;after\;( + j)\;elections \hfill \\ \end{gathered} \hfill \\ {0:} \hfill & {otherwise} \hfill \\ \end{array} } \right.,\quad j \in [3;6;9;12]$$

• (ii) EDs for “regular elections” only

  $$ED\_regular_{{\left( { \pm j,t} \right)}} = \left\{ {\begin{array}{*{20}l} {1:} \hfill & \begin{gathered} for\;all\;months\;up\;to\;and\;including\;the\;jth\;month \hfill \\ before\;( - j)\;or\;after\;( + j)\;``regular\;elections^{\prime\prime} \hfill \\ \end{gathered} \hfill \\ {0:} \hfill & {otherwise} \hfill \\ \end{array} } \right.,\quad j \in [3;6;9;12]$$

• (iii) EDs for “early elections” only

  $$ED\_early_{{\left( { \pm j,t} \right)}} = \left\{ {\begin{array}{*{20}l} {1:} \hfill & \begin{gathered} for\;all\;months\;up\;to\;and\;including\;the\;jth\;month \hfill \\ before\;( - j)\;or\;after\;( + j)\;``early\;elections^{\prime\prime} \hfill \\ \end{gathered} \hfill \\ {0:} \hfill & {otherwise} \hfill \\ \end{array} } \right.,\quad j \in [3;6;9;12]$$

The methodology allows also for augmentation of the statistical model with other explanatory variables, which, referring to economic theory or common sense, could be considered relevant to explain any degree of variation in the dependent variable. These augmented models are known as ARMAX(p,q,m), where X denotes the presence of (m) other explanatory variables. We employ this type of augmented model as the main statistical setting of our analysis. The additional explanatory variables we include are the industrial production index (IPI) in volume terms; retail trade index (RTI) in volume terms; the number of unemployed persons (Un_Per); RSD/USD nominal exchange rate (NER); and a dummy variable to control for the COVID-19 pandemic shock, which takes the value “1” from February 2020 to March 2021 and value “0” otherwise (Covid_dum).⁸ Based on theoretical and intuitive reasoning, the explanatory variables are included either with a time lag of one period (in the case of monthly data) or as time contemporary variables (when quarterly collapsed data were employed). Descriptions of variables used in analysis are presented in Table 1, while detailed description with additional clarifications regarding syntax and transformations employed in estimations for monthly and quarterly data are to be found in Table 3of “Appendix”).⁹

Table 1 List of variables and definitions

In the absence of monthly time series data on real output growth (gross domestic production—GDP), which would be the most appropriate variable to control for real economic activity that might affect fiscal balance, the IPI and RTI in volume terms are reasonable proxy variables.¹⁰ The number of unemployed persons is available at monthly frequency, and we employ this variable to control for possible fiscal balance variation due to labour market dynamics (e.g. to control for potential influences on the fiscal deficit through certain budget items such as unemployment state assistance), whereas nominal exchange rate (RSD/USD) controls for any potential variation due to dynamics in currency exchange markets which also, theoretically and intuitively, might affect fiscal balance.

In the Box–Jenkins methodology of ARMA modelling (Box and Jenkins 1976), one key prerequisite is the stationarity and non-presence of seasonality of the time process being modelled (i.e. the dependent variable), as well as all explanatory variables in the model, if any. Initially, we deflated the original time series of monthly fiscal balance with the Consumer Price Index (CPI) to remove inflation effects and then test the deflated series for any presence of seasonality.¹¹ The series contains strong patterns of seasonality based on all seasonality tests employed (i.e. F-test; nonparametric Kruskal–Wallis test¹²; moving seasonality test; and combined test).

The same is the case for the time series of our explanatory variables. Therefore, first we seasonally adjusted all of the original series.¹³ Then, we tested again for the stationarity of each seasonally adjusted time series, utilizing several unit root tests. The seasonally adjusted series of the dependent variable (i.e. fiscal balance) results in a stationary time process according to all of the statistical tests employed (i.e. augmented Dickey–Fuller test; Phillips–Perron test; and Kwiatkowski–Phillips–Schmidt–Shin test).¹⁴ Conversely, all seasonally adjusted series of explanatory variables were non-stationary processes according to all tests. Therefore, in order to obtain a stationary series, we further transformed explanatory variables into their respective first lag differences of the natural logarithms, which are approximately the monthly growth rates of the original series.¹⁵

The right-hand graph in Fig. 1 presents the time series of the seasonally adjusted monthly fiscal balance in constant prices, whereas the left-hand graph shows the time series only adjusted for prices but not seasonally adjusted, both measured in RSD billions. The seasonal patterns are also visible from the left-hand graph of Fig. 1. The “regular election” dates are depicted by the solid grey vertical lines, and the “early elections” dates are depicted by the dashed lines. Already, from an eyeballing of the right-hand graph in Fig. 1, it is possible to discern deteriorating (decreasing) patterns during certain time periods anticipating certain regular elections and a pickup afterwards.

Figure 1

Source: National Bank of Serbia–seasonal adjustment by the authors

Inflation adjusted monthly fiscal balance in RSD billion (left-hand panel); inflation and seasonally adjusted monthly fiscal balance in RSD billion (right-hand panel).

The formal representation of the intervention analysis in this study is:

$$y_{t} = a_{0} + \sum\limits_{i = 1}^{p} {a_{i} y_{t - i} } + \sum\limits_{i = 1}^{q} {\beta_{i} \varepsilon_{t - i} } + \omega_{ \pm j} ED_{ \pm j,t} + \sum\limits_{k = 1}^{m} {\sum\limits_{i = 0}^{n} {\phi_{k} x_{k,t - i} } ,\quad j \in [3;6;9;12]}$$

where y_t denotes the seasonally and price adjusted monthly fiscal balance measured in RSD billions and t indexes months; $a_{0}$ is the constant term; a_i and β_i are, respectively, the i autoregressive (AR) and moving average (MA) parameters of the p AR lags and q MA (ε) terms in the ARMAX(p,q,m) model, which model the “natural” dynamics of fiscal balance; ω_±j,t are the parameters that capture any opportunistic effects of approaching elections (i.e. “the event”) on the variable of interest, namely fiscal balance; and the parameters ϕ_k model the effect of x_k, where k is the number (m) of additional explanatory variables. The latter could be either contemporaneous variables (i = 0) or variables with a time lag (i =1,…, n). In this case, with monthly data, k = 5—i.e. IPI_{(t − 1)}; RTI_{(t − 1)}; Un_Per_{(t − 1)}; NER_{(t − 1)}; and Covid_dum_(t). Therefore, the parameters ω_±j,t measure the effects of the interventions (events) and are estimated along with the parameters of the ARMAX components. The estimation procedure provides estimates of ω_±j,t as well the corresponding confidence intervals. The probabilistic distribution of each estimator ω_±j,t is a t-distribution allowing for straightforward testing of our hypothesis.

We follow the Box–Jenkins methodology (Box and Jenkins 1976) to identify and estimate the most appropriate ARMAX(p,q,m) model for the time process of interest, namely the seasonally adjusted fiscal balance. The most appropriate ARMA(p,q) component of the ARMAX model tentatively found for the variable of interest was an ARMA(1,1) specification—i.e. one first lag auto regression term (AR1) and one first lag moving average term (MA1). We reached this econometric conclusion following the Box–Jenkins methodology, which consists of an iterative three-stage process of: (i) model identification; (ii) parameter estimation; and (iii) assessing the model’s diagnostics. Several conventional criteria and diagnostic tests were employed throughout this iterative procedure.¹⁶

Each pair of symmetrical pre- and post-elections dummy variables (EDs) as defined earlier were introduced one at a time in the “best” ARMA(1,1) model.¹⁷ Including also the monthly growth rates of IPI, RTI, Un_Per, NER (all four lagged by one period/month), and Covid_dum as additional controlling variables, all parameters of each final comprehensive ARMAX model were estimated simultaneously. If the respective ED estimates have the expected sign (in line with our hypothesis), then the statistical significance of the electoral dummy variables, tested through t tests, reveals whether there is indeed any supposed impact of the elections on the fiscal balance.

Robustness Checks

To check the robustness of the main estimated parameters of interest, firstly we run the whole analysis on the “second best” alternative competing model ARMA(1,0), as well as on specifications without any control variables but with ARMA components alone. We also run specifications including separately for pre-elections and post-elections EDs (i.e. in contrast to the simultaneous inclusion of symmetrical pairs of EDs before and after elections in the primary specification).

Secondly, we apply the intervention analysis in the framework of OLS linear regression modelling, employing the same transformed variables as in the ARMAX setting, given that the stationarity of time series (including non-presence of seasonality) is also a prerequisite for OLS regression.¹⁸ Appropriate dependent variable lags, as determined by standard statistical tests (i.e. the Durbin–Watson test, the Breusch–Godfrey LM test, etc.), are introduced as additional regressors to model the inherent autocorrelation in the fiscal balance. In all estimated regressions we utilize robust standard errors (i.e. the White S.E.) to address the potential presence of heteroscedasticity. The results and findings obtained from this approach are essentially the same as those obtained from ARMAX modelling.

Thirdly, we collapsed the monthly data to quarterly frequencies and carried out the analysis in both econometric settings, i.e. ARMAX and OLS linear regression. In this case, we substitute the control variables of IPI and RTI with quarterly GDP in constant prices, as a better variable to control for real economic activity.¹⁹ In order to ensure stationarity of the series, we transformed the original series of quarterly GDP in the same way as we did for the other explanatory variables already introduced in the monthly frequency modelling (as explained earlier). The following section presents the empirical results from all aforementioned primary and alternative specifications.

Empirical Results

The empirical analysis reveals clear evidence of election-related cycles in the fiscal balance of Serbia during the period January 2005–March 2022. When distinguishing between regular and early elections, we find that PBCs take place only in regular (scheduled) elections, while there is virtually no PBC whatsoever in snap (early called) elections.

The estimated parameters of most of the electoral dummy variables employed in the analyses strongly indicate that there is a statistically significant deterioration of the fiscal balance at various time intervals before elections, followed by normalizations or improvements thereafter. More interestingly, the election-related effect on fiscal balance is essentially driven only by regular elections, while there is no statistically significant deteriorating effect of the fiscal balance before snap elections, thus corroborating the hypothesis of this article. Fiscal balance cycles are obviously more pronounced during regular elections. The deterioration magnitude before these elections (i.e. the negative values of the respective estimated electoral dummy variables) are substantially higher than when all elections were considered together.

Furthermore, improvements of fiscal balance after elections (statistically significant at conventional levels) also appear mostly only in the estimated equations employing regular elections or all elections altogether and only a couple in the case of equations employing early elections. In contrast, in the case of early elections we find no statistically significant estimated ED coefficients for any of the time intervals before elections and only in two cases/EDs after elections in one of the alternative specifications.

These findings are robust to alternative econometric approaches and specifications, namely: (1) ARMAX modelling, including the alternative specifications within this modelling framework (i.e. specifications with the “second best” ARMA components, or without any controlling variables but ARMA components only, or with separate inclusion of pre- and post-elections EDs instead of pair inclusion of symmetrical pre- and post-elections EDs); (2) OLS linear regression modelling; and (3) specifications and estimations with quarterly collapsed data for the dependent variable (fiscal balance) and employing more adequate explanatory variables available at quarterly frequency (i.e. GDP).²⁰

Table 2 presents the econometric results for each set of elections separately: i.e. “all elections”; “regular elections”; and “early elections”. In each case, estimates are reported from each econometric approach (i.e. ARMAX and OLS linear regression modelling) and for each data frequency (i.e. monthly and quarterly). Table 2 is trimmed to present only the main variables of interest, i.e. the estimated parameters of the electoral dummy variables, while in “Appendix” (Tables 4, 5, 6, 7) we provide the complete econometric results for each estimated model.

Table 2 The impact of elections on fiscal balance

Most of the estimated parameters of EDs before “all elections”, estimated through both ARMAX and OLS modelling on monthly data, are significantly negative at either the five or one per cent level of significance. More specifically, prior to elections, when “all elections” are considered, we see a deterioration of the monthly fiscal balance ranging from RSD 2.3 billion in the twelve months before elections (ED_-12) estimated through OLS modelling to RSD 5.9 billion in the three months before elections (ED_-3) estimated through ARMAX, as shown, respectively, in the third and first columns of the “all elections” block in Table 2. Given that the overall sample mean of fiscal balance (monthly average of fiscal balance at constant prices) is RSD − 5.2 billion, these constitute substantial magnitudes of deterioration, from almost half of its long-term “natural” average to above a hundred per cent of this average.

Such deterioration in the fiscal balance is considerably larger when only “regular elections” are considered compared to the case of “all elections”, particularly for the most immediate time intervals before elections (i.e. 6 or 3 months before elections). As shown in the second block of Table 2, the deterioration in the monthly fiscal balance before regular elections ranges from RSD 1.9 billion in the twelve months before elections (ED_Regular_-12) estimated through OLS modelling to RSD 10.2 billion in the three months before those elections (ED_Regular_-3) estimated through ARMA modelling, statistically significant at conventional levels of significance. Interestingly, when only “regular elections” are considered, there seems to be a kind of intensifying monotonic trend of deterioration in fiscal balance as elections come closer (i.e. ED_regular_-12 > ED_regular_-9 > ED_regular_-6 > ED_regular_-3—noting that the inequality signs in this case mean that each succeeding ED is more negative than the preceding one).²¹ That is the case for both ARMAX and OLS regressions. Hence, the closer in time we are to regular elections the larger the deterioration of the fiscal balance. For both of the aforementioned categories of elections (i.e. “all elections” and “regular elections”), for all of the econometric settings employed (i.e. on monthly or quarterly data with ARMAX or OLS), the highest PBC effect results at the closest time-interval to elections, namely in the last three months or the last quarter before elections. In contrast, when only “early elections” are considered, none of the estimated parameters of the respective EDs before those elections result in statistically significant results at conventional levels, in any of the econometrical setting employed (see the “early elections” block in Table 2).

Consistently following these findings on what happens with fiscal balance before regular versus early elections, one could take a subtler view also on what happens after each of these elections’ categories. Indeed, even in the aftermath of elections, almost everything statistically significant regarding fiscal balance dynamics seems to happen only in regular elections and very little in the early ones. We find improvements of fiscal balance (i.e. EDs’ coefficients with a positive sign and statistically significant at conventional levels) three and six months after regular elections with OLS regression, respectively, with monthly and quarterly estimations, as well as three months after regular elections with monthly ARMAX estimations (see the respective EDs in the second block of Table 2). The magnitude of the fiscal balance improvement in these cases averages at around RSD 4 billion per month. We find only two post-elections EDs (estimated with the OLS quarterly setting) statistically significant for the defined time intervals after early elections (see the respective EDs in the third block of Table 2).

Therefore, based on these empirical results, one can take the view that, while in general there clearly exist PBC patterns in one of the main targeted parameters of fiscal policy in Serbia, namely general government overall budget balance, this incumbent behaviour takes place (statistically) only in regular (scheduled) elections and does not occur with snap (early called) elections.

Discussion of the Results and Conclusions

This is the first research work on PBCs in Serbia. As such it provides important insight into the less studied transition economy in the literature of the political economy of elections. The findings suggest the existence of PBCs in Serbia.

Since the system of checks and balances is weak in Serbia, incumbents could easily misuse their official state function and party engagement (OSCE 2020), thus creating the conditions for a boost in public spending for electoral purposes. There are several mechanisms through which higher public spending increases before elections. One important channel is the substantial increase in employment in the already large public sector in Serbia—providing jobs prior to elections is expected to increase electoral support (among the households of newly employed people). Serbia has had, after Belarus, the largest number of employees in the public sector—46% of all employees (Radio-0212014), and as such, public sector employment represents a heavy weight for the public budget. Although the government of Mirko Cvetković (2008–2012) was considered “a champion in party employment”, the other governments also followed the same behavioural pattern. It is interesting to note the case of RTB Bor (public company) where right before 2014 elections, 500 people were employed on one single day (corresponding to 10% increase in total employment in that company)—the company was managed by a politician affiliated to the ruling party (Pavlović 2016, 68). Another striking example is that of the City of Niš which before (regular) 2020 elections, the budgeted expansion of employment by 200 jobs increased the budget by about 170 millions RSD corresponding to 1.6% of the city budget (Stankov 2020). Another example is the cash transfer that the government gave to all adults corresponding to 70 billion of RSD, right before elections in June 2020 (Fiskalni Savet 2020).

The study findings are in line with the previous studies on PBCs such as the recent publication of Lami (2022) who analyses the case of Albania (another post-communist economy which is part of the Western Balkans like Serbia). However, this paper distinguishes between the type of elections by their timing, namely snap (early called) versus regular (scheduled) elections. The empirical findings suggest that while, in general, there clearly exist PBC patterns in one of the main targeted parameters of fiscal policy in Serbia, namely general government overall budget balance, this incumbent behaviour takes place (statistically) only in regular elections and does not occur in snap elections. Thus, the paper contributes to the political economy debate around the incumbent PBC related behaviour in early versus regular elections, by showing very different policy strategy followed by the incumbent in Serbia, thereby highlighting the importance of distinguishing between these types of elections when conducting research on PBCs.

Although our results undoubtedly imply that regular elections are exerting a strong effect on the fiscal balance (deficit), the capacity to carry out PBC is conditioned also by the overall economic situation and related circumstances. While during the elections in 2007 economic growth was deemed high (6.4%) (The World Bank 2022), in the case of the 2020 elections, the extraordinary powers that the government took over due to COVID-19 crisis provided an additional opportunity to increase spending (also in conjunction to elections) bypassing public scrutiny similar to other countries which experienced elections during COVID-19 (Imami et al 2022), whereas in the case of the 2012 elections, the economy was strongly hit by the Global Financial Crisis of 2009 and by the subsequent European sovereign debt crisis increasing Serbian public debt to unsustainable levels, thus limiting the space for fiscal manoeuvring prior to elections.

In order to reduce election driven debt/deficit trends, it is necessary to improve both the overall justice and institutional framework (e.g. courts, state audit, etc.) and the professionalization and independence of media, which is important to raise the awareness of voters. Indeed, an enhanced institutional framework is important not only to reduce election driven debt, but also to enable growth as previous research highlights that the development of an institutional framework has a significant positive impact on growth (Havrylyshyn and Van Rooden 2003; Havrylyskyn and Wolf 1999).

One of the limitations of this paper is the rather small number of elections covered by the analysis, conditioned by the focus on a single country. Nonetheless, the empirical findings are robust to alternative econometric settings and thus findings related to the distinction between the two types of election can be considered informative. Furthermore, the empirical research relies on monthly data on the fiscal balance. Most studies in this field of research rely on annual data which has been considered a serious drawback (Streb et al. 2012; Akhmedov and Zhuravskaya 2004). As such, the analysis is more solid.

In this paper, we do not explore the mechanisms behind increasing deficit prior to elections. That can be possibly caused by increased public expenditure (e.g. for infrastructure or increasing pensions, or increasing public sector employment as indicated above), or lower tax revenues (due to lower tax rates or lower tax collection performance), or both. For instance, Lami and Imami (2019) find that lower tax collection performance takes place before elections even in mature democracies and, however, that this is more common in “younger” democracies than in mature ones. Future research should consider these aspects.

Appendix

See Tables 3, 4, 5, 6, 7 and 8.

Table 3 Variable definitions

Table 4 Estimated equations with monthly data/ARMAX models

Table 5 Estimated equations with monthly data/OLS linear regression models

Table 6 Estimated equations with quarterly data/ARMAX models

Table 7 Estimated equations with quarterly data/OLS linear regression models

Table 8 Variable transformations + Unit-root tests (H0: unit root)