Electoral Cycles of Tax Performance in Advanced Democracies

Abstract

It is widely accepted that incumbents in democratic societies may use various economic policies to increase their chances of re-election. But incumbents in the most advanced democracies may be restrained in overtly manipulating economic policy instruments before elections, because more experienced voters could penalize them for such opportunistic behavior. Incumbents may embrace indirect and more ‘camouflaged’ means such as opportunistically relaxing the stance of tax revenue performance before elections, either by laxer collection efforts, additional tax exemptions or preferential treatments, or a combination of these. In this article, we present evidence of election-related cycles in the tax revenue performance of 25 Organization for Economic Co-operation and Development advanced democracies. We empirically analyze the collection effectiveness of Value Added Tax (VAT) around elections. The findings reveal significant deterioration of VAT revenue performance before elections.

1. Introduction

Sometimes it is called ‘election year economics’: the concept that long-run economic performance may be sacrificed to the immediate re-election concerns of incumbent governments. It is common knowledge, supported by research, that economic performance prior to elections can affect the likelihood of the incumbent re-election.¹ Thus, while economic factors may influence political outcomes, political factors may also influence economic outcomes; incumbent may use economic policy instruments to enhance the chances of re-election.

There is a long record of research on the use of fiscal and monetary instruments controlled by incumbents to ameliorate macroeconomic outcomes before elections. In the political economy literature this perspective is commonly referred to as the concept of the ‘political business cycle’. Governments may behave opportunistically prior to elections, engaging in expansionary economic policies to increase output and decrease unemployment in order to please voters. The political business cycles model of Nordhaus (1975) opened the way for many subsequent empirical and theoretical studies and remains a point of reference for this concept.

Over the years, the focus of the empirical research shifted from macroeconomic outcomes toward elections-related dynamics in economic policy instruments. This view is commonly summarized as ‘political budget cycles’ in the case of fiscal policy instruments, or ‘political monetary cycles’ in the case of monetary policy instruments. Over the past two decades, this field of political economy has been enriched by a significant amount of empirical research. There are many studies empirically showing electorally driven manipulation of the main fiscal policy instruments (i.e. public expenditures, tax revenues, or budget balance). However, the evidence consistent with such conventional wisdom is most convincingly obtained for developing countries with immature democracies.² In the case of developed countries with advanced democracies, the evidence on the hypothesized presence of political budget cycles is questionable. Recently, a widely accepted view is that incumbents in advanced democracies, more often than not, refrain from manipulation of economic policy instruments before elections. The thinking is that voters in advanced democracies, being both much more experienced and better informed, tend to punish rather than reward incumbents when they notice economic policy manipulation for political/electoral ends.

However, intuitively speaking, this does not necessarily mean that incumbents in advanced democracies become intrinsically more ‘benevolent’, nor that they hold back from all types of manipulative behavior within the arsenal of policy instruments they control. Instead, incumbents in advanced democracies may refrain from policy manipulation only for those policy instruments that are observable and within the understanding of the voters, in order to avoid any possible backfire. This leaves open the possibility that incumbents in advanced democracies might tend to channel their manipulative efforts toward other policy instruments that are not easily perceived by voters.

We hypothesize that the tax revenue performance could be one such area prone to incumbents’ electoral manipulations. Shifting toward a more relaxed attitude regarding tax effectiveness before elections, that is tax collection and control, tax exemptions or other preferential treatments, or a combination of these, could be both an effective way to create some additional economic stimulus and, at the same time, remain relatively beyond the scrutiny of the typical voter.

In this article, we bring evidence from a panel of 25 advanced democracies, members of the Organization for Economic Co-operation and Development (OECD), that the performance of value added tax (VAT) significantly deteriorates before elections. The findings could be viewed as an initial exploration in the need of further empirical research on the manipulation possibilities of economic policy instruments that may not be so easily observed by the typical voter, in addition to the classical ones that have been studied in the past. Moreover, this research is relevant for both more mature democracies as well as less mature ones.

In the next section, we present in more details the literature review and the hypothesis of the study. Section 3 explains the data and methods, while section four provides the findings as well as robustness scrutiny. The paper is wrapped up with concluding remarks in Section 5.

2. Background of the Hypothesis

There is a growing empirical research showing evidence of political business cycles or political budget cycles including both consolidated and young democracies. In the case of older democracies, there have been obtained mixed results. For example, Alesina et al. (1997) find only weak evidence of election-related cycles in the budget balance and no cycles in the main components of the government budget for 13 OECD economies from the 1960s to the 1990s. Keech and Pak (1989) show evidence of significant increase in veterans’ benefits before elections in the USA between 1961 and 1978, but not subsequently, and they do not find evidence of electoral cycles in total public spending. Persson and Tabellini (2002, 2003) find no political cycles neither in total expenditures, nor in transfers or the overall budget balance across a panel of 60 democracies during 1960–1998.³ By contrast, in pre-electoral periods in Portugal, the incumbent is found to increase total expenditures and change their composition, favoring items that are highly visible to the voters (Veiga and Veiga 2007). Meanwhile, Dahlberg and Mörk (2011) find that governments increase public employment in election years in Sweden and Finland—such an increase not only implies increased public expenditure but also improved public services for citizens.

In the case of young democracies, which by default tend to have weaker institutional framework and less relevant experience, political budget cycles are more likely to be observed, when compared to the consolidated democracies. Empirical research carried out by Brender and Drazen (2005) as well as Shi and Svensson (2006) confirm that young democracies are particularly vulnerable to political budget cycles. Brender and Drazen (2005) find evidence of political cycles in fiscal balances in a large panel of countries—they attribute this finding to the ‘new’ democracies (i.e. having democratic experience of up to 15 years) included in the sample. Once these are removed from the sample and only established democracies are considered, evidence of political budget cycles disappears. Interestingly, they find that total revenues in proportion to Gross Domestic Product (GDP) in a sub-sample of established democracies fall in the election year, similar to Persson and Tabellini (2003) but they do not allude to any tax cuts before elections. In addition, they find that even in the case of ‘new’ democracies, political cycles, either in budget balance or expenditures, weaken significantly after the second elections (in new democratic systems) and disappear completely after the fourth elections, suggesting that longer experience in democracy plays a determinant role in inhibiting the political manipulation of at least those policy instruments which are easily observable. Shi and Svensson (2006) also find that there are significant differences between developing and developed countries regarding political cycles in budget balances. They find that the deficit increases significantly before elections in developing countries while such loose electoral fiscal policy is almost non-existent in developed countries.

Shi and Svensson (2006) suggest that institutional indicators, such as government corruption, rent-seeking activities and access to free media, can explain a large part of the differences in the size of policy cycles between developed and developing countries. While Alt and Lassen (2006) show the relevance of transparency, Brender and Drazen (2005) emphasize the lack of experience that voters have in new democracies regarding the existence of political fiscal cycles. That better institutions and more experienced, informed and rational voters lead to less politically manipulated economic policy before elections, seems to be a common view in recent research.⁴ These are features of mature democracies. Generally, the more advanced a democracy is, the better the institutions and the more experienced, informed and rational the voters are, the lower is the presence of political budget cycles.

Thus, in consolidated democracies, the incumbent has a more limited space for ‘fiscal maneuvering’ before elections, and alternative options may be considered. While some studies find increased deficits before elections, there is limited understanding as to how those deficits were incurred. Obviously, they may be caused by lower revenues from taxes and/or increased public expenditures.

In turn, tax revenues are affected by both tax rates and tax collection performance. Recent research has found that the incumbent engages in favorable taxing policy prior to elections. In the case of Germany, there is evidence of a political cycle in tax setting as the local business tax rate was found to be significantly reduced in the election years and the year prior to the elections, while it significantly increased in the year after elections (Foremny and Riedel 2014). In Italy, municipalities were found to set lower real estate tax rates when close to elections (Alesina and Paradisi 2017). Meanwhile, Ehrhart (2013) finds robust evidence of lower indirect taxes being applied by incumbents in the period just prior to elections in developing countries. Persson and Tabellini (2002, 2003) argue that taxes are cut during the election year and they suggest that this is the case even for a sub-sample of more mature democracies. However, they consider only the total revenues of central governments expressed as a percentage of GDP, for which they find a significant reduction of −0.3% of GDP in election years. They do not investigate whether standard tax rates have been cut or whether other factors, such as laxness of tax collection (administration), more tax exemptions or other preferential treatments could be behind the fall in revenues in proportion to GDP. Furthermore, total revenues include non-tax revenues, which are broadly unrelated to tax policy.

Since an increase in public expenditures and/or a decision to reduce standard tax rates prior to elections would easily draw wide public attention, the incumbents might explore alternative ways of providing fiscal incentives before elections. One alternative could be ‘relaxing’ the stance of tax collection effectiveness, either by lowering tax collection enforcement or by applying tax exemptions and preferential tax treatments. This strategy could likely have the following benefits for the incumbent:

• Direct benefit to businesses, which pay less taxes enabled by higher ‘degree of tolerance’ by the government—they could perceive the incumbent as ‘generous’ and might be tempted to support/vote for the incumbent in the upcoming elections.

• Provide stimulus to the economy before elections. Intuitively, lower tax collection could have a similar direct effect on agents’ behavior as lower tax rates.

• It represents an opportunity for ‘pork barrel’ politics—the tolerance in tax collection effectiveness could be applied at higher degree to groups of companies or regions where the electoral benefit of such strategies could be the highest (e.g. target swing voters).

The advantages of such a manipulation strategy for the incumbent is that the core of tax law and standard tax rate can remain unchanged, while decisions are taken by elected and career officials, whose motivation can vary in conjunction with elections. Elected officials and ‘politicized’ public service/administration officials can lower their general vigilance before elections, focusing more on elections than on their duties or exploit opportunities to grant preferential tax treatments without attracting much public attention; while other public administration officials may simply lower their performance before elections, opportunistically exploiting legal ambiguities regarding tax assessment and control subject to lower accountability from their elected superiors. Young et al. (2001) find that Internal Revenue Service audits adjust in conjunction to elections, being less intense in electorally sensitive districts in the USA. Tax evasion can be associated or caused by reduced audits by tax authorities—firms that can respond to looser monitoring by immediately underreporting actual sales as suggested by Skouras and Christodoulakis (2014) in the case of Greece.

Lowering tax performance either by subsiding tax collection efforts or by introducing sporadic preferential tax treatments, is less observable in real time or attracts less public attention and scrutiny, as compared to conventional expansionary fiscal policies (e.g. a decision to increase public expenditures and/or reduce standard tax rates). Furthermore, the responsibilities and the underpinning reasons behind weaker tax performance before elections are more difficult to assess as compared to alternative conventional expansionary policies, when the electorate can clearly spot without delay ‘who proposes’, ‘who approves’, etc.

Our study aims to analyze VAT performance before elections in advanced, OECD democracies. Our main hypothesis is that tax collection effectiveness (or fiscal performance) in advanced democracies is lower before elections. In addition, we test whether the length of experience of being an advanced democracy makes a difference, expecting lower fiscal performance in younger democracies.

3. Data, Variables, and Model Specifications

Our analysis focuses on collection effectiveness of VAT before elections. VAT is a major part of the tax system in over 136 countries, raising about one-fourth of the world's tax revenue (Ebrill et al. 2002). For countries included in this study, VAT revenue represents (on average) about 37% of the total tax revenue or about 7% of GDP. While election-related performance cycles could be present across other types of budget revenue administered by incumbents, and therefore also for total budget revenue, it is difficult in practice to find adequate measures of the effectiveness for total revenue or other types of revenue. The mere ratio of total revenue to GDP is not a proper measure of tax performance, as it cannot account and differentiate for the underlying factors altering the ratio. In contrast, in the case of VAT, there exist relatively good measurement practices and accurate data for most advanced democracies, permitting a reliable assessment of its collection effectiveness. The VAT Revenue Ratio (VRR) is widely employed as the most appropriate available way to measure VAT collection effectiveness. VRR is a benchmark measure refined upon earlier similar attempts to gauge VAT performance. It is also widely known as the ‘C-efficiency ratio’ originating from the International Monetary Fund (Ebrill et al. 2001) and is currently by far the most commonly used measure found in the literature. VRR measures the gap between the actual VAT revenue collected and the revenue that would theoretically be raised if VAT were levied at the single standard rate on all final consumption assuming perfect collections enforcement. The OECD annually estimates and reports VRR for member states.

In theory, the closer the VAT system of a country is to a ‘pure’ VAT regime, that is no exemptions and preferential treatments and perfect tax collection enforcement, the closer its VRR is to 1. Referring to VRR estimations, OECD reports suggest that in many of its member countries, a considerable part of the potential VAT revenue is not collected. That is due to erosion of the tax base either by exemption or other preferential treatments, poor compliance or poor tax administration or a combination of these (OECD 2014). In the context of obvious underperformance in VAT revenue, our main hypothesis is that the VRR in advanced democracies deteriorates systematically before elections as compared to its ‘natural’ pattern. That would suggest that incumbents in advanced democracies may seek other means to increase their chances for re-election that are less observable by the typical voter. In addition, in line with previous research that finds a distinction between ‘old’ and ‘new’ democracies as regards political budget cycles’ patterns and magnitudes (e.g. Brender and Drazen 2007), we test the hypothesis of VRR decrease in both types of advanced democracies, ‘older’ and ‘younger’, and assess/compare the respective magnitudes. Figure 1 suggests that there is generally a lower VRR in election years.

To test the hypothesis of this article, we estimate the effect of elections on the collection effectiveness of VAT in a strongly balanced panel of 25 OECD countries from 1995 to 2012. An important advantage of using time observations from different entities (i.e. panel of countries in this case) is the ability to disentangle political cycles from other potentially coinciding cycles (e.g. economic cycles). Therefore, we employ as our baseline specification fixed effect (FE) regression with time FEs to control for such potentially misleading occurrences. The inclusion of time FEs was also supported by adequate statistical test as explained in details later in this section. The selection of countries is based on the ‘democracy strength’ as measured by the Polity IV index, as well as on the availability of data. The Polity IV index ranges from −10 (strongly autocratic) to +10 (strongly democratic), and we included all those OECD countries with scores of at least +8 (i.e. 8–10).⁵ All 25 countries are also categorized as countries with ‘very high human development’ by the Human Development Report (UNDP 2015), meaning not only that these countries, part of the OECD club, feature developed democracies, but that they are amongst the most economically developed countries.

The panel data include 450 country-year observations. A total of 118 elections took place during this period in all 25 countries taken together, or on average about 5 elections per country. The data are mainly sourced from the OECD’s statistical database, while some missing VAT standard rates for some country-year observations as well as data on elections are obtained from other public sources.⁶ We only include legislative elections for those countries with parliamentary systems whereas presidential elections for those with presidential or semi-presidential systems.⁷

The main dependent variable we utilize is the first difference of the natural logarithm of VRR ‘dlogVRR_{(c, y)}’ which is approximately the year-on-year change rate of VRR. Using the VRR’s rate of change instead of its level (or its log-level) ensures the stationarity of the dependent variable.¹⁰ Therefore it allows utilization of static panel-data models, which generally provide more reliable estimators and statistical inference than dynamic panel models, while the interpretation of estimated parameters is quite straightforward. The latter class of models often suffers from econometric drawbacks such as inconsistency due to the ‘Nickell bias’ (Nickell 1981) and thus could lead to invalid statistical inference (Gaibulloev et al. 2014). However, to check the robustness of our main results and findings we replicated the regression estimations using the original level of VRR as well as its natural logarithm as the dependent variable, employing adequate dynamic panel-data modeling (i.e. GMM dynamic panel-data models). Another auxiliary econometric feature of employing VRR’s rate of change is absence of serial correlation evidence in the error structure ‘ɛ_{(c, y)}’ of the estimated models, although as explained later, to address the presence of heteroskedasticity we run all models (i.e. baseline and alterative specifications) with robust standard errors clustered at cross-sectional level which take care of both heteroskedasticity and autocorrelation, should the later have been present.¹¹

Two types of electoral dummy (ED) variables ‘ED_{(c, y)}’ were defined, aiming to capture the effect of approaching elections on VAT collection effectiveness. The estimated parameters of these dummy variables offer empirical evidence in support of or against the hypothesis of this study, and therefore are of main interest. The first type of ED variable is the ‘usual’ one, defined similarly to the most common practices throughout the relevant literature, taking the value ‘1’ in election years and value ‘0’ for all other years. In the absence of intra-annual frequency data (i.e. monthly or quarterly frequency data) and considering the potential drawbacks that annual data could have in correctly revealing the presence of political budget cycles behavior, as pointed out by Akhmedov and Zhuravskaya (2004), the second type of ED is a ‘weighted’ dummy variable, which aims to approximate the effect of elections on VRR during the last year (four cumulative quarters) prior to elections. The ‘weighted’ dummy variable takes a value of either ‘0.25’ or ‘0.5’ or ‘0.75’ or ‘1’ in the election year and the preceding one depending on the specific quarter that elections took place within the year. For instance, if elections took place in February of year ‘Y_E’, therefore in the first quarter of year ‘Y_E’, the ED takes the value ‘0.25’ in the election year ‘Y_E’, the value ‘0.75’ in the preceding year ‘Y_E-1’ and the value ‘0’ in the other years. Or, if elections took place in October of year ‘Y_E’, therefore in the last quarter of year ‘Y_E’, the dummy takes the value ‘1’ in the election year ‘Y_E’ and ‘0’ in the other years. A formal way to define each of the EDs is the following:

• The ‘usual’ electoral dummy variables ‘ED_U’

  $EDu={1:intheElectionyearYE0:otherwise$
• The ‘weighted’ electoral dummy variables ‘ED_W’

  $EDW={(0.25*j): in the Election year YE(1−0.25*j): in the preceding year YE−1; j∈(1,2,3,4)Election quarter0:otherwise$

In line with the relevant literature and common intuition, several control variables were included in the baseline and alternative specifications. Economic growth is expected to have a positive effect on tax collection as better (worse) economic conditions of economic agents, particularly private sector, increase (reduce) incentives for tax compliance. Hence, real economic growth ‘GDPg_{(c, y)}’ (year on year percentage change) is introduced as an explanatory variable. Inflation ‘Inflation_{(c, y)}’ (annual average inflation rate in percentage) is also included to control for any potential variation due to nominal conjunctures in the economy. Based on common practices of tax collection, in particular as regards VAT collection, effectiveness is generally higher for VAT levied on imported goods due to more concentrated, simplified and supervised collections systems of custom offices compared to VAT collected on domestically produced goods. Ebrill et al. (2002) also presume that it is relatively easier to collect VAT at the point of import than domestically. Therefore, the ratio of goods’ imports to GDP ‘Imports_{(c, y)}’ (in percentage of GDP) is also included as an explanatory variable. In addition, we introduced a dummy variable to control for variations related to changes of the VAT rate ‘VATrch_{(c, y)}’. When a change in a certain tax rate takes place it is generally expected that in the short run the collection effectiveness of that tax deteriorates to some extent due to the time required for both taxpayers and tax collectors to economically and administratively adapt to the new tax rate. The dummy variable introduced for this purpose takes the value ‘1’ in both the year when the change of VAT rate becomes effective and the year following, and the value ‘0’ in other years. Although only advanced democracies are considered in this analysis, still, presumed corruption should not be neglected as a possible factor affecting tax effectiveness. Hence, we controlled for corruption ‘Corruption_{(c, y)}’ by introducing as a proxy explanatory variable the rate of change of the Corruption Perception Index measured annually by Transparency International (2012).¹²

We explored several alternative specifications including simultaneously or separately different lags of the continuous explanatory variables (i.e. lags 1 and 2 of GDP growth, inflation, imports’ share and corruption perception index). None revealed any additional explanatory power or altered the findings, thus we stick to the parsimonious models with only contemporary variables.

We performed fixed effects panel estimation (FE) for all baseline models. The Hausman specification test (Hausman 1978) suggests that FE panel regression is the most appropriate model specification. Employing the FE estimator is also preferable to other estimators as it controls for unobserved country-specific FEs, which are typically present in country panel data. Indeed, one statistical concern in our panel study is that unobserved time-invariant features of countries could be correlated either with the number of elections held in each country or the dependent variable, which could than lead to biased estimation of coefficients of primary interest. Introducing country FEs in the baseline specifications ‘FE_(c)’ controls for these possible omitted variables, leaving only time variation within countries to estimate elections’ effects. However, the Hausman test concludes only at the borderline of 10% level of significance (p = 0.089). Therefore, in order to check the robustness of our results we reiterated all estimations employing random effects (RE) estimator as well as Pooled Ordinary Least Square (Pooled OLS) panel modeling.

We also include time FEs in the regression specifications. As discussed earlier, time FEs in panel data regression enable controlling for other potentially present cycles (i.e. economic cycle) occurring at synchronized waves with political cycles which could than lead to spurious results. Including time FEs is also in line with Ebrill et al. (2002) who found that time experience with VAT matters and the longer the VAT has been in place, the better the collection performance. Inclusion of time FEs is also supported by a Wald (1943) test, which rejected the null of ‘all dummies for all years are equal to zero’ at the 5% of significance (p = 0.030).

Heteroskedasticity was present in the errors’ structure of all baseline estimated regressions as indicated by the modified Wald test for group-wise heteroskedasticity in FE regression models, which is performed following Greene (2000). We employed heteroskedasticity robust standard errors clustered at cross sectional level (i.e. clustered by 25 countries) in all estimated baseline FE regressions as well as in the alternative specifications. Employing clustered robust S.Es. corrects for heteroskedastic panels and at the same time takes care for any presence of autocorrelation, which in our setting is particularly relevant for the alternative specifications.

We diagnosed also for the possibility of cross-sectional (spatial) dependence. Indeed, two out of three tests employed for this purpose rejected the null of ‘cross-sectional independence’ for all baseline estimated regressions. The Pesaran (2004) and Frees (1995, 2004) tests rejected the null of cross-sectional independence respectively at the 5% and 1% levels of significance, while a Friedman (1937) test was not rejected at conventional levels.¹³ To ensure the robustness of our results, all baseline regressions were re-estimated and tested with appropriately modified errors, which account for cross-sectional dependence. We employed Driscoll and Kraay (1998) standard errors for coefficients estimated by FEs (within) regression, which are robust to general forms of cross-sectional dependence, heteroskedasticity and autocorrelation.¹⁴

We also run an analysis of the panel of countries regarding conventional fiscal instruments, namely government expenditures and budget balance, following the same procedure. The aim is to compare the results of the fiscal performance analysis with governmental expenditure and budget balance.

The empirical results are reported in the following section and in the Appendix, including the results obtained from alternative specifications for robustness checking purposes.

4. Empirical Analysis

4.1 Baseline findings

We found that the annual VRR’s rate of change decreases by about 2 percentage points before elections, which is a considerable magnitude of deterioration given that the overall panel average is 0 (zero) %, or around one third of within-country standard deviation. That was captured by both the ‘usual’ and the ‘weighted’ electoral dummies. The estimated parameter of ED_U, which captures the dynamics of the VRR’s rate of change in the elections year, was negative (−0.021) and significant at 1 percent level with both types of standard errors used in the estimation (i.e. heteroskedasticity robust standard errors and Driscoll-Kraay standard errors) indicating a decrement to the VRR’s rate of change of 2.1 percentage points in election years (see Table 1, as well as model 1 in the Appendix Table A.2 for more detailed results). The estimated parameter of ED_W, which by definition approximates the impact in VRR’s rate of change during one year (i.e. four cumulative quarters) to elections, indicated a negative effect of −1.9 percentage points significant at the 5% level with both types of standard errors employed in the estimation (see Table 1, as well as model 4 in Table A.II).

Of course, there is always the suspicion that our election dummies may be detecting some other, unobserved influence. However, while this possibility may be difficult to discount completely in the context of a single-country time-series, it is much less likely in a panel context. Appendix Table A.1 shows the election years for each country during the sample period. Across the different countries, elections are held in different years and with different periodicity, which makes it hard to imagine some exactly matching but omitted country- and time-varying influence.

In terms of GDP, the VAT performance deterioration accounts roughly for 0.15%, which is rather substantial if considering that this is from VAT alone and probably such deterioration might take place in other tax and non-tax revenue items as well.¹⁵

We also run the baseline FE regressions on two sub-samples of the panel. The first sub-sample included only ‘the older’ democracies covered in the study and presumably the most experienced ones, namely 19 countries. The second sub-sample included the other six countries considered as ‘the younger’ and presumably the less experienced advanced democracies.¹⁶ The results obtained for each subset showed consistency with those of the whole panel of 25 countries. The rate of change of the VRR significantly deteriorates before elections in both subsets of democracies. However, it turns out that the magnitude of deterioration is seemingly lower in ‘the older’ democracies and higher in ‘the younger’ ones. That would suggest that despite being an advanced democracy, the democratic experience time span a country has does matter, in accordance with previous research, implying that the more experienced a democracy, the tighter the incumbents’ room for opportunistic electoral maneuver (Brender and Drazen 2007). The respective electoral dummy coefficients (ED_U and ED_W) show that the VRR’s rate of change deteriorates before elections by 1.2 to 1.3 percentage points in ‘the older’ democracies and by 2.9 to 5.1 percentage points in ‘the younger’ ones (see Table 1, as well as more detailed results in models 2, 3, 5, and 6 in Table A.2). Nevertheless, the chi²–test employed to test the equality of EDs’ coefficients in the two sub-samples could not infer that the difference in magnitudes was statistically significant at conventional levels (see chi²–test in Table 1).

The obtained baseline results cannot reject the main hypothesis of this article, suggesting that incumbents in developed democracies, generally facing stronger constraints against manipulation of those economic instruments that are easily observed by public opinion, could therefore engage opportunistically in altering the stance of other ‘indirect’ and less observable instruments under their control. Weakening tax collection effectiveness seems to be one such camouflaged approach toward the inherent inclination of incumbents to get re-elected.

Regarding the budget balance and total expenditures our findings are broadly in line with existing relevant research mentioned earlier. A deterioration of budget balance by 0.38% of GDP takes place during the election years, while there is no statistically significant change in government expenditures (for more details see Appendix Table A.3). Thus, the deterioration of the budget balance can be attributed to the poorer fiscal performance (tax collection).

4.2 Robustness of the results

The empirical results remained practically the same when alternative specifications were employed. Table 2 presents the estimated parameters of ED employed in each alternative specification.¹⁷ First, we estimated the same baseline regressions but now employing a random effects estimator (RE) and a pooled OLS estimator. The estimated coefficients of ED_U and ED_W using both methods were practically identical to those estimated earlier with the FE estimator, both in magnitude and in the level of statistical significance.

We then altered the dependent variable. We regressed the level of the VRR and the natural logarithm of the VVR, which as explained earlier, seem to be non-stationary panels by some of the tests applied, employing appropriate Arellano-Bond dynamic panel data modeling. The latter is basically a GMM estimator developed first by Arellano and Bond (1991) to consistently estimate dynamic panel models, appropriate for specifications with non-stationary dependent variables. The battery of explanatory variables in the dynamic settings remains the same as in the baseline specifications, except for the additional lag (1) of the dependent variable as well as related instrumental variables.¹⁸ Robust standard errors were used in all estimations. The estimated parameters of ED_U, which is of principal interest, were negative and significant at 1% level of significance in both cases (‘VRR’ and ‘logVRR’). Whereas the coefficients of ED_W,despite being negative, their statistical significance was weaker than ED_U. Actually, when the log of VRR was employed as the dependent variable, it was not significant at any conventional level. The magnitude of coefficients was also very similar to the baseline specifications (see Table 2).

One should note that the comparison of coefficients between specifications having the VRR’s rate of change as the dependent variable (i.e. baseline FE; alternative RE, Pooled OLS) and those alternative dynamic specifications regressing the level or log-level of VRR (i.e. GMM) is not straightforward and some simple computation is needed. For instance, the ED_U’s coefficient in the dynamic model that regresses the level of VRR results negative (−0.0104), which infers that the level of VRR decreases by 0.0104 in the elections year. Given that the overall panel average of VRR level is 0.5458, it implies a pre-electoral deterioration of VRR level by 1.9% (−1.9% = −0.0104 ÷ 0.5458 × 100), which is similar with the respective coefficients in static specifications.

We assume that all elections dates are exogenously determined, meaning not influenced at all by any of the factors which simultaneously could influence tax collection effectiveness (the dependent variable). However, sharing the same concern raised in previous research work regarding the validity of this assumption and the potential endogeneity of the timing of elections, we constructed another electoral dummy variable which excludes snap (early) elections ‘ED_{U(no_snap_elections)}’. There were 12 cases of snap elections called by the incumbents earlier than foreseen by the constitution.¹⁹ We run all baseline and alternative specifications employing the new electoral dummy ‘ED_{U(no_snap_elections)}’ and obtained results that were almost identical with those when ‘ED_U’ was employed (see Table 2).

We further scrutinized the robustness of the results by checking whether such deterioration in VAT collection effectiveness was present also in other years during the electoral term, which if true would have weakened our case for not rejecting the hypothesis of this study. Therefore, in the same way we did with ED—we defined two other non-electoral dummies, one for the year succeeding the elections year ‘NED_U(+1)’ and the other modeling the year preceding the election year ‘NED_U(-1)’. Hence the reference time (year) is still ‘elections’ (elections year) for all countries. We estimated all regressions now incorporating each of the two non-electoral dummies instead of EDs and found no statistical evidence of VRR deterioration in any of these two other years of the electoral term (see Table 2). Interestingly, the dynamic models regressing the original level of the VRR and its log revealed an improvement of the VRR in the year succeeding the elections year by around 1.1 percentage points statistically significant at 5% level. If anything, this result tends to strengthen the evidence in favor of the hypothesized presence of electoral cycles in VAT collection effectiveness in developed democracies.

To rule out any contaminating effect on our findings due to possible data inaccuracy in the earlier periods we performed the whole empirical analysis only on the last decade of (2002–2012), totaling 286 observations. The results obtained on the sub-period panel are consistent with the previous results on the entire dataset. The magnitude of VRR deterioration before elections was even stronger during the last ten years than for the entire period of the panel. In 2002–2012, the rate of change of VRR deteriorated by around 3 percentage points before elections and the coefficients were significant at the 1% level in all specifications. The intensified deterioration was also present for each sub-sample of ‘older’ and ‘younger’ democracies, respectively deteriorating by around 2 and 3.7 percentage points both significant at the 5% level.²⁰ Interestingly, this last robustness check is also intuitively in line with the notion that the more experience the average voters have with the elections-related political economy, the more the incumbents may shift the opportunistic focus to less easily perceivable instruments under their control, such as tax collection effectiveness.

5. Conclusion

Incumbents may engage in expansionary (fiscal) policies before elections to improve the likelihood of re-election. However, in consolidated democracies, the incumbent has a more limited space for ‘fiscal maneuvering’ prior to elections, since voters are more experienced and informed about the incumbent attempts to manipulate the economy before elections and the longer run implications of such behavior.

The evidence is mixed about the occurrence of elections-related fiscal deficit cycles, and there is limited understanding about how these deficit cycles occur. Obviously, higher deficit may be caused by lower tax revenue, and/or increased public expenditure. Since expansion of public expenditure and/or a decision to lower the main standard tax rates before elections would easily draw wide public attention and possibly backfire, the incumbent might explore alternative ways of providing fiscal incentives before elections, less noticeable by the typical voter. One alternative could be ‘relaxing’ the stance of tax collection effectiveness, through a higher ‘degree of tolerance’ by the incumbent (i.e. through the tax authorities), additional tax exemptions or preferential treatments or a combination of these. Tactically, that can be achieved through lower/weaker tax inspection or enforcement. Fall of tax revenue performance, whether deliberate or not, is less observable in real time by the voters, as compared to traditional expansionary policies. Furthermore, the responsibilities and the underpinning reasons behind weaker tax performance before elections are more difficult to assess.

We found evidence that VAT revenue performance significantly deteriorates before elections in 25 OECD advanced democracies. VAT collection effectiveness measured by the VRR rate of change deteriorates by around 2 percentage points in election years when the full panel sample average is 0%. In terms of GDP, that amounts to roughly 0.15%, which is rather considerable given that this arises from VAT alone. This evidence is robust to different econometric specifications, different time span experiences as a democracy, and during different periods of time. It turns out that the magnitude of deterioration is seemingly lower in ‘the older’ democracies and higher in ‘the younger’ ones, as expected. These differences can be explained by voters’ experience and the quality of institutional framework. Indeed, previous research has shown that various factors, such as the level of development, voters’ experience, institutional quality, age and level of democracy, electoral rules and form of government, transparency of the political process, the presence of checks and balances, and fiscal rules can affect the occurrence and intensity of fiscal policy manipulation for electoral purposes (Brender and Drazen 2005; Shi and Svensson 2006; de Haan and Klomp 2013).

Regarding budget balance and expenditures, our findings confirm a deterioration of budget balance by 0.38% of GDP during the elections year, while there is no statistically significant change in government expenditures. Thus, it can be concluded that increased deficit is caused by weaker fiscal performance.

It is rational to expect that voters react to both policies as well as promises delivered before elections. Indeed, empirical research has shown that voters respond to electoral promises, and that is more likely when promises are regarded as credible (Elinder et al. 2015), which is found to be the case for OECD countries (Mansergh and Thomson 2007). Whether the government's promises are credible could play an important role in to what extent voters react to electoral promises and to what extent they react to implemented policies. We can assume that, in the case of younger democracies, which are characterized by weaker institutional framework and shorter political parties' lifespan, voters are more likely to pay attention to pre-election policies results rather than to promises, when compared to consolidated democracies with more consolidated institutions including political parties. Consequently, incumbents in younger democracies may both have higher degrees of freedom (enabled by a weaker institutional framework including checks and balances and poorer voters’ awareness), as well as may be under more pressure to deliver tangible results before elections.

Repetto (2018) found that mayors react to more informed voters by reducing spending manipulation in the case of Italy. Thus, higher awareness/information of voters about incumbent economic policy manipulation in conjunction to elections, is likely to result in adjustment of voters’ behavior (e.g. voters may punish instead of rewarding the incumbent for such policy manipulation), which in turn, should result in lower pre-election economic policy cycles. In the case of fiscal performance, which is not as easily observable as the expansionary public expenditure policies, the role of research is crucial to identify/assess fiscal performance patterns in conjunction to elections, and feed that information into the mass media.

This article provides evidence that fiscal performance, namely tax revenues collection, can be influenced by the incumbent in conjunction to elections. It documents fiscal performance election cycles and provides one plausible explanation, that of intentional incumbent engagement to relax tax collection for electoral purposes. However, our aim is not to be conclusive about the only or main mechanism or explanatory motivation behind such behavior or pattern. The findings could be viewed as an initial exploration in need of further empirical research on the manipulation possibilities of economic policy instruments that may not be so easily observed by the typical voter, in addition to the classical ones that have been studied in the past.

Footnotes

Acknowledgement

We are thankful to Ardian Harri, Evan Scott Thomas, Geoff Pugh, Panu Poutvaara and unknown reviewers for their comments.

References

Appendix