2. Fiscal deficits and executive planning horizons

From previous literature, there are at least two ways in which to formally characterize the relationship between expected executive tenure and fiscal performance. First, where incumbent governments face an exogenously given probability of being in office at time t + 1 which is less than unity, the likelihood that they will use public debt as an instrument to influence successive administrations discretionary budgeting powers is expected to increase (Alesina and Tabellini Reference Alesina and Tabellini1988; Persson and Svensson Reference Persson and Svensson1989; Devereux and Wen Reference Devereux and Wen1998; Debrun and Kumar Reference Debrun and Kumar2007). This should lead to larger fiscal deficits in years where there exists a high expected probability that the executive in office in period t, will not be in office in period t + 1. Second, it is possible that the same executive may attempt to use fiscal policy as a campaigning devise, increasing government expenditures in years where the probability of a transition is high in order to “buy back” myopic voters. These two scenarios make it difficult to separate out whether a defeated executive generated high levels of debt to constrain their successor, or, a winning executive was successful because of deficit-financed increase in political support. From a theoretical perspective, an easy solution would be to exogenously fix the probability of remaining in office in order to concentrate on the fiscal discount effect with comparative statics.

Alesina and Tabellini take on such an approach proposing a model where citizen disagreement, rather than myopia, influences fiscal policy in democracies. Two parties are assumed to choose the same levels of taxation and public consumption (private consumption-leisure tradeoff are equal under both parties) but differ with respect to preferences for the composition of public goods. The incumbent government's objective functions are time separable into an intra-period problem of choosing taxes and provision of public goods for a given deficit (static), and, an inter-temporal problem of choosing the size of the deficit (dynamic). This inter-temporal fiscal decision is influenced by the (exogenous) probability of the executive remaining in office in future periods as well as the distance between the preferences of incumbents and successor administrations. For example, a conservative incumbent who knows with a high probability that they will lose power to a liberal successor can strategically use deficit spending on their preferred composition of expenditures in order to constrain their successor's ability to provide its preferred composition. In equilibrium, policymakers set their marginal utility of leaving debt to the future equal to the time discounted expected marginal cost of inheriting that debt tomorrow. This implies that incumbents, who have low expectations of inheriting future debt, will fail to internalize the inter-temporal distortions created by running large deficits. Governments who discount the future at higher rates than the general public will run larger than optimal deficits in order to increase expenditure on their preferred bundle of public goods at time t while constraining future governments from spending on their less preferred compositions at time t + 1. Alternatively, a social planner with infinite horizons (re-appointed with probability of one), adopts a social welfare maximizing weighted average of the preferences of the citizens and balance the budget in every period (Alesina and Tabellini, Reference Alesina and Tabellini1990). From this, it should be expected that high degrees of polarization between party preferences, and low levels of probabilistic executive tenure to generate sub-optimal fiscal outcomes and relatively higher levels of equilibrium debt (ceteris paribus).

Persson and Svensson (Reference Persson and Svensson1989) take on a similar approach but assume policymakers to possess different preferences for the level, rather than the composition of, government expenditure. Finite horizon governments are confronted with a trade-off between two types of distortions: volume distortions, which occur where government consumption is higher than the optimal level preferred by that government, and; inter-temporal distortion, which occurs where, for a given level of public consumption, the time profile of taxes differs from the ex-ante optimum solution. Assuming that policymakers are forward-looking, strategic, and the volume of inherited debt has an effect on newly elected governments taxation and spending decisions, “stubborn” incumbents who put a significant amount of weight on minimizing volume distortions relative to inter-temporal distortions should be expected to borrow more than it would if it had infinite planning horizons (Persson and Svensson Reference Persson and Svensson1989). This is to say that governments who know with some non-zero probability that they will hand over power to a new administration with different preferences for levels of government consumption, will chose to leave a deficit/surplus in order to force its successor to spend less (in the case of a conservative government) or more (in the case of a liberal government). Again, the ideological distance between current and expected future incumbents, as well as the probabilistic tenure of the current executive, is both expected to generate significant distortionary effects on fiscal outcomes.

Treating the probability of remaining in office in the next period as exogenously given assumes away the possibility that it is partially dependent on the fiscal behavior of the incumbent, which can influence the longevity of their tenure in office (Johnsen et al., Reference Johnsen, Shepsle and Weingast1981; Ferejohn and Krehbiel, Reference Ferejohn and Krehbiel1987; Dewan and Myatt, Reference Dewan and Myatt2010). For example, policymakers may use deficit financing strategically to increase their perceived relative performance in close elections (Milesi-Ferretti and Spolaore, Reference Milesi-Ferretti and Spolaore1994) or may “bring home the bacon” by securing discretionary earmarked funds for their constituencies (Stratmann, Reference Stratmann2012). In order for this phenomenon to hold there must exist some degree of fiscal illusion, and/or a low degree of fiscal transparency, where voters do not fully internalize the inter-temporal budget constraint of the government and therefore overestimate the benefits of current expenditures relative to the future tax burden (Alt and Lassen, Reference Alt and Lassen2006). Opportunistic office-seeking politicians can take advantage of this illusion/lack of transparency by deficit financing new spending (without increasing taxation) in order to buy public support in years where the probability of an executive transition is high (Alesina and Perotti, Reference Alesina and Perotti1995). The probability of being in office in the next period, and thus the degree to which policymakers internalize future distortionary burdens, can therefore both affect, and, be affected by, fiscal outcomes. Political parties will discount the future to the extent that they believe they will no longer be in power, however, their likelihood of remaining in power can be influenced by expansionary fiscal policy. This endogeneity problem has made it difficult for researchers to measure the effects of expected tenure (outside of a fixed term limit context) on fiscal outcomes. Where incumbents use fiscal maneuvering as a strategic device to win an election, we would expect deficits to be highest in the midst of close elections where both incumbent and opposition parties stand a good chance of winning, whereas; for incumbents who form their fiscal policy based on expectations of future tenure in the executive, we would expect high deficits where the likelihood of re-election is very low.

The small number of empirical contributions thus far have either worked within the context of fixed-term limits (Zupan, Reference Zupan1990; Besley and Case Reference Besley and Case1995Reference Besley and Casea, Reference Besley and Caseb; Carey, Reference Carey1996; Alt and Rose, Reference Alt, Rose, Boix and Stokes2009) or have ignored the endogeneity of expected tenure, focusing instead on exogenously given levels of government stability or average tenure (Edwards and Tabellini, Reference Edwards and Tabellini1991; Debrun and Kumar, Reference Debrun and Kumar2007;) From a fixed term limit context, Besley and Case (Reference Besley and Case1995Reference Besley and Casea, Reference Besley and Caseb) find evidence that exogenously imposed term limits have a significant effect of fiscal policy outcomes in US gubernational elections over the 1950–1986 period. The results suggest that governors who are ineligible to stand for election in the following period are found to generate higher levels of sales taxes (7–8$ per capita on average), income taxes (9$ per capita on average) and state expenditure (15$ per capita on average). Alt and Rose (Reference Alt, Rose, Boix and Stokes2009) build on these results within the context of US elections across 45 US states between 1974 and 1999 in a principal-agent framework which considers the relationship between real fiscal outcomes and two focal areas of “incentives” and “ability”. Their results suggest that states, whose approval rating ranges between 40 and 60 percentFootnote ¹, show a statistically significant pattern of per capita spending which is estimated to be 38$ higher in election years (relative to midpoint of election cycle). Outside of a fixed term limit framework, Debrun and Kumar (Reference Debrun and Kumar2007) find evidence from a sample of 18 European Union economies over 1990–2004 period, that decreases in government stability (which they dub a “plausible proxy of the risk faced by an incumbent of being voted out”) generates negative effects on cyclically adjusted primary balances (CAPB) to the extent that very unstable governments will, on average, run 1 percent of GDP higher CAPB than very stable ones (ceteris paribus). Finally, Edwards and Tabellini (Reference Edwards and Tabellini1991) find some support for a positive relationship between fiscal deficits and the frequency of government changes for a sample of 42 developing countries between 1963 and 1988, validating the theoretical expectation that “the policymaker may wish to borrow in excess of the optimum, and let his successor ‘pay the bills’.”

The central theme in all of these studies is the importance of finite horizon policymakers whose interest in optimal future fiscal outcomes is influenced by the expected probability that they will be in office in time t + 1. The theoretical contributions suggest two possibilities: (i) as this probability increases, incumbent executives are more likely to internalize the future costs of deficit financing by generating responsible fiscal outcomes (linear), and, (ii) where this probability approaches 50 percent, incumbent executives may attempt to win the election via expansionary/irresponsible fiscal policy (non-linear). Although little is known about the empirical validity of these models (Besley and Case Reference Besley and Case1995Reference Besley and Casea, Reference Besley and Caseb), the sparse evidence thus far has suggested that finite planning horizon effects do exist in the same direction as predicted in the theoretical literature, but have not pinned down whether they are based on discount factors (linear effects) and/or are used strategically to win elections (quadratic effects).

Lastly, some of the problems associated with short political horizons in fiscal policymaking have been shown to be dampened with the existence of well-defined budget rules (Von Hagen and Harden Reference Von Hagen and Harden1995; Poterba Reference Poterba1996; Alesina et al., Reference Alesina, Hausmann, Homes and Stein1999; De Haan and Volkerink Reference De Haan and Volkerink2001; Hallerberg et al., Reference Hallerberg, Strauch and Von Hagen2007, Reference Hallerberg, Strauch and von Hagen2009; IMF 2009, 2012, 2014, 2015, 2017; Wehner, Reference Wehner2010a, Reference Wehnerb, IMF 2012, IMF 2017). Along with fiscal councils and strong monitoring, these have become popular devices for counteracting deficit biases and achieving greater levels of fiscal responsibility. Based on this literature, it is possible that the existence of fiscal rules may help in reducing fiscal indiscipline associated with short executive planning horizons.

3. The data

Data on GDP growth, unemployment, inflation, and interest rates were taken from the World Banks World Development Indicators (WB-WDI) for the years 1980–2012.Footnote ² Consolidated central government deficits/surplus figures were computed from International Monetary Fund Government Finance Statistics Yearbook for the years 1980–2012.Footnote ^3, Footnote ⁴ With respect to political variables, the selection was limited by the scope of the analysis. Ideology is classified into one of three categories of left right and center (see World Bank—Database of Political Institutions (DPI) 2012, p. 6–7 for definitions) which is used to compute government polarization as the maximum distance (using the left, center, right classification) between the executive party and the three largest parties in government. Fragmentation is measured using the effective number of parties in the executive.

Theoretical findings also suggest that ideological affiliation of opposition parties should be expected to play a role in the fiscal decisions of the executive where there is a divergence between the party currently in power and the party most likely to take office should the incumbent lose (i.e. left executive and right opposition, or vice-versa). Where the executive expects that a transition of power will also be a transition of government ideology, we might expect different fiscal behavior, relative to a transition of power without a change in government ideology (Alesina and Tabellini, Reference Alesina and Tabellini1988; Persson and Svensson, Reference Persson and Svensson1989; Testa, Reference Testa2010). In order to capture this effect, a variable is computed as the difference between executive ideology and the ideology of the main opposition party (both measured on a three-point scale). An additional binary variable is then computed equaling one where there is an executive transition and the losing incumbent party is predominantly affiliated with liberal/conservative ideology while the new executive is affiliated with conservative/liberal ideology (zero otherwise). The raw data for this variable is taken from DPI and the variables are coded by the author. Data for fiscal rules (debt and balanced budget) were sourced from the IMF's Fiscal Rules Database.Footnote ⁵

With respect to a continuous measure of loss probabilities, this paper relies on a modified version of the Kayser-Lindstadt method which characterizes unobservable probabilities of a swing in power between the largest and second largest party in a government during an election year as:

$$s_t = \lpar {\Delta \nu_{1t}} \rpar \eta _1^{^\ast} -\lpar {\Delta \nu_{2t}} \rpar \eta _2^{^\ast} $$

where Δv _qt measures the vote share difference for party q between time t + 1 and time t: (ν _q,t+1 − ν _qt), and, $\eta _q^* $ measures the (time constant) vote-seat elasticities.Footnote ⁶ Given that ν _q,t+1 is unknown and unobservable for each district, the authors map past swings from previous periods beginning at n = 1 through t-1, into a loss function which is used to estimates the probability of an electoral swing $(g:\lcub {s_n} \rcub _1^{t-1} \to f(s))$. The sparsity of elections over a relatively short time period, lead the author to smooth the series via a standard normal kernel density function: $K_\emptyset (s) = \phi (s)$. From this, it is possible to estimate the probability of a plurality shift as:

$$pplur_{it}^{^\ast} = \int_{-\infty} ^{-S_{inc}} {K_\emptyset (s)ds} $$

where S _inc is the proportional advantage in seats held by the incumbent party relative to the largest opposition party ($S_{inc} = s_{inc}-s_{opp}/s_{inc} + s_{opp} + \mathop \sum \nolimits_{i = 3}^{TP-2} s_i$; where TP = total parties in parliament). See Kayser and Lindstadt (Reference Kayser and Lindstadt2015) for more detailed discussion. Within this paper, feasible options for votes shares (ν _1t and ν _2t) in our broader sample of countries, were computed as: (a) aggregate vote shares of the largest governing party minus largest opposition party, or: (b) aggregate vote shares of the entire governing coalition minus the combined opposition vote shares (where this differs from (a)). Standard normal kernel density functions K _h(s) for both (a) and (b) were then estimated. From this, $pplur_{it}^* $ estimates can be easily computed given threshold values computed using S _inc, S _opp, and $\sum\nolimits_{i = 3}^{TP-2} {s_i.\;} $ To get an idea of the relationship between the two measures, Figure 1 below shows a simple scatter plot between the K–L district-level data and the aggregate estimates from our (a).

Source: Kayser-Lindstadt (2015) and authors calculations

Figure 1. Aggregate and District Level Kayser-Lindstadt Loss Probabilities

As expected, the aggregate measure of loss probabilities is skewed towards zero given the narrower domains on which the likelihoods are estimated. Effectively, this means that the district level densities will give some non-zero probability of executive turnover to very unlikely swings while the aggregate levels densities will assign these zero probabilities. There was sufficient information on all dimensions for a globally representative unbalanced panel of 83 countries spanning the 1980–2012 period. A list of variables and summary statistics is available in online Appendix B.

4. Estimation

This paper begins with a standard fixed effects model, regressing central government fiscal balances on a common set of tax smoothing and political variables, including fiscal rules and the binary event of an election. Common pool resource and veto player theory suggest that both size and fragmentation should have directional (temporal) and volatility (inter-temporal) effects on government balances (Tsebelis Reference Tsebelis2002; Chang and Tsebelis Reference Chang and Tsebelis2004; Franzese Reference Franzese and Magara2007). In the case of the common pool resource problem, we should expect an increase in the number of meaningful actors in the budget process to increase the temporal deficit bias of government as these actors fail to fully internalize the cost of their decisions. We should also expect that, as more decision makers are included in the budget process, the likelihood of a consensus move away from the status quo will decrease, making it more difficult to implement intertemporal changes in fiscal outcomes (Chang and Tsebelis Reference Chang and Tsebelis2004; Blais et al., Reference Blais2010; Franzese Reference Franzese2010).Footnote ⁷ From this, the equation for central government surplus can be specified as:Footnote ⁸

(1)$$y_{it} = \alpha _i + \delta \lpar {elec_{it}} \rpar + {\bi P}_{{\bi it}}{\bi \beta} + {\bi X}_{{\bi it}}{\bi \theta} + \varepsilon _{it}$$

where y _it is the surplus/deficit or net lending/borrowing as a percentage of GDP in country i at time t and α _i is an unobserved intercept for country i. The first term on the right-hand side is a binary indicator equaling one in years where an executive election taking place.Footnote ⁹P and X are matrices capturing all other political and macroeconomic determinants, respectively, including intertemporal effectsFootnote ¹⁰ and the existence of balanced budget and debt rules, on central government fiscal balances. δ, ρ _l, ρ _i, β, and θ are unknown parameters/vectors of parameters to be estimated.

The results from three specifications are given below in Table 1. The first and second, build on past theoretical and empirical findings from equation (1). The third specification includes three additional indicators to test the Persson and Svensson (Reference Persson and Svensson1989) hypothesis that ideological differences between the incumbent and opposition party will lead incumbents to “restrain” future policy options. The first two indicators are binary variables representing right and left wing executives (relative to center). The third indicator measures the ideological distance between the executive branch of government and the largest opposition party on a three-point scale (left/center/right). For example, where the executive is predominantly right (left) wing and opposition is predominantly left (right) wing, this indicator will take on a value of −3 (3), while a left executive–left opposition or right executive–right opposition combination would produce a value of zero.

Table 1. Within Country Determinants of Central Government Balances (1980–2012) (Fixed effects with cluster-robust standard errors in parenthesis)

*p < 0.10; **p < 0.05; ***p < 0.01

Consistent with past findings, the evidence in Table 1 suggests that levels of central government fiscal balances during the 1980–2012 period are largely determined by fiscal inertia, political, and economic circumstances. As would be expected in Barro's tax smoothing framework, unemployment and debt service costs significantly reduce fiscal surplus's/increase deficits, while GDP growth has a robust and significant positive effect on fiscal balances. The number of parties appears to have significant temporal and intertemporal effects on fiscal balances. The negative temporal effects confirm past results regarding the “Law of 1/N” where an increase in the number of meaningful actors (parties) in the budget process leads to larger deficit bias (common pool resource problem). This finding supports the findings of DeHaan and Volkerink (Reference De Haan and Volkerink2001) who find that the effective number of parties and number of cabinet ministers generate significant positive effects on government debt accumulation in a sample of 22 OECD countries for the earlier 1971–1996 period. The positive intertemporal effects (interaction with lagged fiscal balances) suggest that an increase in meaningful actors in the budget process (veto players) decreases rates of adjustment or ability to implement large reforms (expansionary or contractionary). These results are consistent with past theoretical expectations and findings from Franzese (Reference Franzese and Magara2007, Reference Franzese2010) and Blais et al., (Reference Blais2010), where single-party governments are free to pursue high magnitude expansionary or contractionary policy whereas multiparty governments face more pressure to converge on more watered down changes in fiscal balances. There is also evidence to support Persson and Svensson (Reference Persson and Svensson1989) where conservative incumbents are expected to run lower deficits which tend to increase where there exists a liberal opposition, and; liberal governments expected to run higher deficits which tend to decrease given a conservative opposition. Column 3 of Table 1 suggests that right-leaning executives will tend to run 0.73 percent lower deficits, and left-wing governments will tend to run 0.01 percent higher deficits (both relative to center governments); however, where right/left executives are confronted by left/right leaning oppositions, these deficits will increase/decrease by 0.22 × 3 = 0.66 percent. For example, if we assume that the central leaning government tend to run a balanced budget, we would expect a right-wing government, who believes that their most likely successor is of a similar ideology, will tend to run a 0.73 percent surplus, while a right-wing government who believes that their successor is of a left-leaning ideology will tend to run a 0.07 percent (0.73–0.66) surplus. Lastly, it appears that countries can dampen the effects of irresponsible fiscal policy by creating binding constraints in the form of a balanced budget rule.Footnote ¹¹ The magnitude of this effect on fiscal balances (between 0.71 and 0.91 percent of GDP) is more than sufficient to offset the indiscipline resulting from election year effects (between −0.37 and −0.40 percent of GDP).

While there appears to be a significant correlation between election years and central government fiscal balances, the planning horizons literature suggests that the “true” effect on fiscal performance is generated through the expected probability of remaining in office in year t + 1 rather than a simple binary indicator for whether an election or transition took place. It should be expected that, where an executive transition does take place with an a priori expected low likelihood of occurrence, these governments will be more likely to internalize the costs of future debt and generate lower deficits than incumbents who perceive a high a priori likelihood of a transition. In such cases, the estimates from (1) will be biased upwards as they assume that all transitions were fully known to those who left office. Past findings of significant election year effects (Debrun and Kumar, Reference Debrun and Kumar2007; Wehner, Reference Wehner2010a, Reference Wehnerb) may be a reflection of this probability, but like the binary transition indicator, is also measured with error. The inability to observe a continuous likelihood of executive transition, along with the potential for expected planning horizons to be incorporated into fiscal policy for a variety of reasons (see theoretical discussion), makes it likely that our binary election indicator in equation (1) may not generate accurate results.

Based on the theoretical discussion in Section 2, the unobserved continuous likelihood of remaining in office should have a significant impact on the degree to which current policymakers internalize future burdens of irresponsible fiscal policy. From this, the error term in equation (1) will be correlated with this unobserved likelihood. To make this point clear, we can assume the “true” equation for central government fiscal balances is:

(2)$$y_{it} = \alpha _i + \lambda \lpar {\,ptrans_{it}^{^\ast}} \rpar + {\bi P}_{{\bi it}}{\bi \beta} + {\bi X}_{{\bi it}}{\bi \theta} + \varepsilon _{it}$$

where $ptrans_{it}^* \sim [0,\,1]$ is the “true” continuous unobserved probability of an executive transition in country i at time t. Because$\; ptrans_{it}^* $ is a latent variable, equation (1) estimates the effect of executive planning horizons with some uncertainty $(ptrans_{it}^* = elec_{it} + \upsilon _{it})$.

Substituting this back into equation (2) gives,

(3)$$y_{it} = \alpha _i + \lambda (elec_{it} + \nu _{it}) + {\bi P}_{{\bi it}}{\bi \beta} + {\bi X}_{{\bi it}}{\bi \theta} + \varepsilon _{it}$$

suggesting that $cov\lsqb {elec_{it,\; \;} \lpar {\varepsilon_{it}-\lambda \nu_{it}} \rpar } \rsqb = -\lambda \sigma _\nu ^2 \ne 0$ which makes the parameter estimates from (1) (Table 1) inconsistent (Greene 2008; Wooldridge 2002).

In our sample of 83 countries over the 1980–2012 period, about half (51 percent) of all elections, result in an executive transition. From this, we would expect that the parameter estimate δ in (1) (and Table 1) to underestimate the true effect λ from (2), as the former includes incidence where the likelihood of an executive transition is very low, giving the incumbent no incentive to create a future burden than they themselves will have to bear. With the additional possibility that governments may use fiscal policy as a device to increase their popularity in a closely contested election, there may also exist, a non-linear relationship between transition probabilities and fiscal performance. Prior to addressing this problem, however, it is important to define ptrans_.*

To get a more accurate estimate for $ptrans_{it}^* $, we can begin with the work of Kayser and Lindstadt (Reference Kayser and Lindstadt2015). As noted in the data section, the authors measure electoral risk as a function of, (i) the expected variability in party level vote share, and (ii) the district level seat-vote elasticities (which are one-for-one transformations in PR systems). The advantage of this approach is the ability to easily estimate the likelihood of electoral defeat/victory which “enjoys exogeneity from potential policy-related dependent variables” (Kayser and Lindstadt, Reference Kayser and Lindstadt2015). It is important to note, however, that the Kayser-Lindstadt (KL) measure is one of plurality swings, rather than executive transitions, but the two will likely be correlated. For example, as noted in the paper, “In a two-party system, losing plurality status in parliament almost inevitably means leaving government, while in multi-party (or non-democratic) systems with coalition governments, this is not necessarily the case.”Footnote ¹² So, while Kayser-Lindstadt are fundamentally interested in “the expected probability that a plurality party in parliament losses it's seats plurality in the next election”, this paper is interested in the expected probability of a change in the executive branch of government at any point in time during their tenure (even if an election does not occur). Computing KL for a broader sample of countries, therefore, requires us to assume that all executives form expectations about their future tenure based, to some degree, on the degree of formal political opposition they face (vote shares). The advantage of this approach is the ability to examine executive transitions in a more dynamic environment which includes both election and nonelection years from a globally representative sample of countries.

The disadvantage that comes with broadening the population/sample is not having the manpower to compile district level vote-seat data which is required to estimate swing sensitivity in multiparty systems (Linzer Reference Linzer2012; Kayser and Lindstadt Reference Kayser and Lindstadt2015). One of the most pronounced challenges is the relatively small domains on which aggregate kernel densities fall.Footnote ¹³ The implications of this are narrower estimates of loss probabilities, where a larger proportion of elections will be predicted to effectively have a zero probability of a plurality shift (which may be true). If we assume that expected plurality shift likelihoods are good exogenous representatives of executive transitions $(ptrans_{it}^* \approx pplur_{it}^{} )$, their effect on central government balances can be directly estimated in two ways:Footnote ¹⁴

(4a)$$y_{it} = \alpha _i + \lambda \lpar {\,pplur_{it}} \rpar + {\bi P}_{{\bi it}}{\bi \beta} + {\bi X}_{{\bi it}}{\bi \theta} + \varepsilon _{it}$$

(4b)$$y_{it} = \alpha _i + \lambda _1\lpar {\,pplur_{it}} \rpar + \lambda _2\lpar {\,pplur_{it}} \rpar ^2+{\bi P}_{{\bi it}}{\bi \beta} + {\bi X}_{{\bi it}}{\bi \theta} + \varepsilon _{it}$$

The first specification (4a) includes a linear effect of vote swings which assumes that incumbents base fiscal decisions based on their discount factors associated with expectations about remaining in office in the future. The second specification (4b) allows for the possibility that incumbent will use fiscal policy as an instrument to increase their levels of support in tightly competitive situations. In this second case, we would expect deficit spending to be highest where $pplur_{it}^* $ is close to 50 percent.

Table 2 below shows estimates for equations (4a) and (4b) with the KL district level data for sample of 21 countries over the 1980–2008 period (columns 1 and 2), and versions (a) and (b) of $pplur_{it}^{} $ (discussed above) computed from aggregate-level data in our sample of 79 countriesFootnote ¹⁵ over the 1980–2012 period (columns 3–6). Columns 1, 3, and 5 show the linear effects of loss probabilities on fiscal balances from (4a), while; columns 2, 4, and 6, show the quadratic effects from (4b).

Table 2. Central Government Balances and Loss-Likelihoods (1980–2012) (Fixed effects with cluster-robust standard errors in parenthesis)

*p < 0.10; **p < 0.05; ***p < 0.01

These results are relatively consistent with our baseline findings in Table 1 where fiscal inertia, economic circumstances, veto players, and ideological polarization appear to be strong and robust predictors of central government fiscal balances. The first two columns show results using the KL data for a sample of 21 countries. With respect to loss-likelihoods, the results here are relatively uninformative which could be due to the smaller sample sizer or difference in measure measurement discussed above. In order to isolate the reason for these insignificant results, we re-run the regressions from columns 1 and 2, replacing the original KL data with the modified KL data for the same sample of countries. As can be seen in online Appendix A, the results are mixed suggesting that there could actually be some merit in working with narrower densities from aggregate data.Footnote ¹⁶ The loss likelihood results for our aggregate KL measure appears to have a statistically significant linear relationship with fiscal outcomes (central government balances) in our full sample of 79 countries providing evidence of a discount effect from executives who form fiscal policy based on expectations of remaining in office. The magnitude of this linear effect ranges between 0.64 and 1.05 percent higher deficits (as a percent of GDP) when comparing an executive who fully expects to remain in office relative to an executive who fully expects to lose it (ceteris paribus). These results suggest that the continuous unobserved likelihood of expected executive planning horizons $(ptrans_{it}^* )$ are significantly correlated with fiscal balances and differ in magnitude from those estimated in (1) where this probability was measured with error $(elec_{it} = ptrans_{it}^* + \upsilon _{it})$. There is also some evidence of a quadratic effect from our full sample of countries where deficits tend to be highest where the expected likelihood of a plurality swing is close to 50 percent. We can see this by setting the first derivative of y with respect to pplur from column (4) in Table 2 equal to zero,

$$\displaystyle{{\partial y} \over {\partial pplur}} = -3.1 + 6.46\lpar {\,pplur} \rpar = 0\Rightarrow pplur^{^\ast} = \displaystyle{{3.1} \over {6.46}}\approx 0.48$$

suggesting that deficits will reach their highest point when the likelihood of a plurality shift is around 48 percent. Figure 2 shows the predicted relationship between the unobserved probability of an executive transition and fiscal performance measured as central governments surplus/deficit as a percentage of GDP from the third and fourth columns of Table 3 (using linear and quadratic loss probabilities). Both functional forms predict a statistically significant correlation between the likelihood of remaining in office and fiscal balances with somewhat greater robustness coming from the linear estimates.

Source: author's calculation from Table 2

Figure 2. Fiscal Performance and Loss Likelihoods (Linear and Quadratic)

Table 3. Dampening Effects of Fiscal Rules? (Fixed effects with cluster-robust standard errors in parenthesis)

*p < 0.10; **p < 0.05; ***p < 0.01

Last, it appears that the existence of a balanced budget rule can help to offset negative effects from distortionary temporal and intertemporal effects on fiscal outcomes. The magnitude of this independent effect ranges between 0.66 and 0.77 percent of GDP, which is nearly sufficient to offset the distortionary effects from an incumbent who knows that they will lose office with a probability near one (estimated in our linear specifications as ranging between 0.64 and 1.05). With respect to a dampening effect on the political business cycle which has been found in past research on US states (see Rose Reference Rose2006), we run a separate set of both linear and quadratic equations with interactions with the existence of balanced budget and debt rules.Footnote ¹⁷

The results in Table 3 suggest that, while the existence of a balanced budget rule continues to have an independent effect on fiscal performance, there is not a great deal of evidence suggesting an interactive dampening effect from either balanced budget or debt rules.

Relating these results back to our theoretical discussion from Section 2, it appears that there are several political/institutional factors which influence budgetary outcomes. First, an increase in the number of meaningful actors in the budget process (parties in government) will lead to a larger deficit bias, along with a dampening effect any expansionary/contractionary initiatives. These results are generally consistent with the theoretical expectations from past literature (Franzese, Reference Franzese and Magara2007; Reference Franzese2010; Blais et al. Reference Blais2010). Second, executive and opposition ideology both seem to have an impact on fiscal outcomes to the extent that surplus generating conservative executives will reduce these surpluses when confronted by liberal opposition (vice versa for liberal executives confronted by conservative opposition). These, previously untested, results are consistent with the theoretical expectations from Persson and Svensson (Reference Persson and Svensson1989). Third, incumbent executives will take into account the continuous expected likelihood of remaining in office when making current fiscal decisions to the extent that, executives who fully expect to remain in office will tend to run lower deficits/higher surpluses that executives who expect to compete for power. Relating this back to the theoretical discussion form Section 2, there is some evidence to support incumbents as strategic actors who will use deficit financing to influence tight elections, but may return to more responsible fiscal planning where they are certain to lose power (right-hand side of Figure 2). The more robust linear effects from columns 3 and 5, however, suggest that incumbent discount factors linearly determine their level of fiscal responsibility, to the extent that the most responsible fiscal outcomes will come from those who are certain they will always be in power.