Introducing time horizons: future cooperation prospects

These tenuous assumptions make it questionable whether all coalitions are actually distinguishable from one-party governments with regard to their time horizons. Coalition governments, I argue, should be even more capable than single-party governments to produce favourable fiscal policy outputs if their longer time horizon through “prospective cooperation” in the future permits them to credibly agree on debt-stabilising policies. Behind this argument are three assumptions. First, coalitions with a shared past governing experience have higher chances to govern together in the future again, because they are familiar with each other. Second, such coalitions are likely to internalise the future costs of indebtedness. Third, broad and durable coalitions can diffuse the electoral risks of contested fiscal policies more effectively than single-party governments.

In detail, the first assumption states that parties in coalition governments see a chance that they might cooperate again in the future. Evidence comes from the coalition formation literature. Parties that have regularly cooperated in the past have created high degrees of familiarity among each other (Franklin and Mackie Reference Franklin and Mackie1983). The shared experiences of past government participation reduce bargaining costs for coalition agreements and go along with reputational benefits (Warwick Reference Warwick1996, 474). Furthermore, parties prefer keeping the same partners, as they know more about each other and are better able to communicate (Bäck and Dumont Reference Bäck and Dumont2007, 487). Incumbent coalitions with high familiarity therefore have higher chances to form the next government (Martin and Stevenson Reference Martin and Stevenson2001, Reference Martin and Stevenson2010). Conversely, the chance for prospective cooperation is reduced with dissention and betraying behaviour in the past (Bäck and Dumont Reference Bäck and Dumont2007, 487; Tavits Reference Tavits2008). All these studies argue that the prospects for future cooperation are decisively influenced by the cooperation behaviour in the past. Thus, it is reasonable to assume that prospective cooperation can be assessed through past cooperation patterns. This assumption can be empirically tested by comparing varying retrospective measures of past coalition patterns.

The second assumption of the argument is that parties care about the costs of high future indebtedness levels. Rarely, governments face positive economic shocks and end up with comfortable fiscal situations so that they can sustain high budget deficits without constraints. More typically, higher debt ratios bear some undeniable long-term costs for all governments (Alesina and Drazen Reference Alesina and Drazen1991; Reinhart and Rogoff Reference Reinhart and Rogoff2009). Although this does not rule out debt as an economic policy option, it limits unrestricted indebtedness at the cost of future generations for parties expecting future participation in government. Coalitions with prospective cooperation will therefore have to budget these long-term costs and internalise them at some point, that is not shift them to external groups or to future governments. This assumption can also be tested empirically because fiscal pressure should moderate the incentive for coalitions to internalise the long-run costs of additional spending.

Although repeated participation in government and internalisation of the future costs of debt accumulation can also apply to single-party governments, the third assumption concerns a unique characteristic of coalition governments. The literature on consensus democracies (for recent empirical evidence, see Luebbert Reference Luebbert1984; Katzenstein Reference Katzenstein1985; Lijphart Reference Lijphart2012; Alexiadou Reference Alexiadou2013; Armingeon et al. Reference Armingeon, Guthmann and Weisstanner2015a) suggests that governments that are able to forge broad coalitions with opposition parties and societal actors can more easily agree on electorally contested reforms. Coalitions with good cooperation prospects resemble such broad coalitions because they can support cross-party alliances over extended periods of time. They can gather broader electoral support for a long-term view on debt developments by exploiting opportunities to diffuse blame and responsibility for unpopular policies across coalition partners (Weaver Reference Weaver1986; Powell and Whitten Reference Powell and Whitten1993). On the other hand, although coalition governments with longer time horizons can agree on the goal of pursuing fiscal stabilisation policies, the process of agreeing on – and implementing – status quo changes may still take a longer time relative to one-party governments because of the multiple veto actors involved (Tsebelis Reference Tsebelis2002; Armingeon et al. Reference Armingeon, Guthmann and Weisstanner2014). Yet, accounting for the more difficult process of reaching consensual policy positions, their advantages in including a broader spectrum of political actors into governmental decision making should enable durable coalitions from keeping debt accumulation under control even better than single-party governments.

Taken together, the causal mechanism of the argument has started from the common observation that, stemming from their fragmentation, coalitions have a “time-inconsistency problem” (Kydland and Prescott Reference Kydland and Prescott1977), because without future perspectives some coalition partners have electoral incentives to maximise spending on their preferred policy areas and externalise their costs (Bawn and Rosenbluth Reference Bawn and Rosenbluth2006). However, from a critical examination of these assumptions about coalitions’ common pool resource and veto player problems, it follows that coalitions can overcome the time-inconsistency challenge if (a) they expect future cooperation in government, (b) they desire to avoid unreasonably high future debt levels and (c) they represent broad societal coalitions that minimise electoral risks associated with controversial reforms. In these circumstances, coalition agreements that lead to farsighted fiscal policymaking can be reached more easily because the partners are familiar with each other. Opt-out options – by leaving the coalition or sabotaging agreements with coalition partners – are associated with higher costs because the parties have to consider that their partners would not be likely to cooperate with them anymore in the future in such cases. The “walk-away value” for coalition bargaining (Lupia and Strøm Reference Lupia and Strøm2008) or the benefit of the bargaining leader for leaving the bargaining round diminishes in such situations. The expectation of prospective cooperation therefore leads to a contract-like situation that relies on reciprocal trust and reputation and that can be the source for credible long-term agreements (Kreps Reference Kreps1990; Majone Reference Majone1996). This leads to the main hypothesis of this article about the fiscally beneficial effects of prospective cooperation:

With the goal of keeping debt accumulation in check, the credibility advantages of prospective cooperation facilitate a wide range of fiscal policy agreements. Possible examples include the capacity to freeze spending levels on existing levels, maintain deficit and/or debt ratio targets, set growth and employment objectives, agree on fiscal consolidation measures by reducing expenditures or increasing taxes and committing to strict implementations of fiscal objectives. In practice, we find many instances where coalition parties openly refrain from unilaterally claiming fiscal resources for their own constituencies. For example, the Swedish centre-right coalition – the traditional opposition block to the Social Democratic party – governing between 2006 and 2014 committed itself to budget surpluses in both their 2006 and 2010 electoral manifestos (Allians för Sverige 2006; Alliansen 2010). Grand coalitions in Austria – Christian and Social Democrats share a long history of co-governing – regularly agree on detailed coalition programmes such as a 288-page document for the 2008–2013 term, which underlined the goal of a balanced budget with specific measures of fiscal consolidation (Bundeskanzleramt 2008). Another instance is Ireland’s centre-left government of 1994–1997, characterising itself as a “government of renewal” (Irish Government 1994). Fine Gael and the Labour Party had collaborated three times between 1973 and 1987, expectedly facilitating the favourable outcomes of cyclically adjusted primary surpluses of more than 4% from 1995–1997. These three examples also stress that I expect longer time perspectives to facilitate fiscal outcomes irrespective of the ideological orientation of the governing parties involved.

Are there alternative factors that condition the effectiveness of coalitions’ cooperation outlooks to contain debt dynamics? First, as fiscal policy sometimes co-varies with electoral cycles (Nordhaus Reference Nordhaus1975) and even governments with long-term cooperation prospects face electoral uncertainties (Lupia and Strøm Reference Lupia and Strøm2008; Jacobs Reference Jacobs2011), election years should temporarily suspend debt stabilisation efforts of governments seeking re-election. Second, if the assumption is valid that parties care about the costs of future debt (see above), the incentives that cooperation involves for long-sighted fiscal policymaking should be stronger when the government faces greater fiscal pressure. Third, even when governing parties agree on debt-stabilising policies, the effectiveness of fiscal outcomes may depend on the degree of centralisation in budgetary institutions – especially in the delegation of fiscal authority to strong finance ministers (Hallerberg and von Hagen Reference Hallerberg, Strauch and von Hagen1999; Hallerberg et al. Reference Hallerberg and von Hagen2009). This leads to three sub-hypotheses:

Although these factors are expected to moderate the strength of the effect of prospective cooperation, they are not necessary conditions determining whether cooperation effects take place at all. In other words, one should find evidence for lower debt growth for coalitions with long time horizons even during election years, under low fiscal pressure and under low centralisation of budgetary institutions.

Debt change

Following several studies, the dependent variable is the annual change (first differences) in gross general government debt ratios in percentage of gross domestic product (GDP) (Franzese Reference Franzese2002; Bäck and Lindvall Reference Bäck and Lindvall2015). Debt data are based on OECD Economic Outlook and Eurostat sources, but time series are extended through backward extrapolation, mostly in the 1960s and 1970s, using data from the International Monetary Fund (IMF) (Abbas et al. Reference Abbas, Belhocine, ElGanainy and Horton2010) and Reinhart and Rogoff (Reference Reinhart and Rogoff2009) (for details see Armingeon et al. Reference Armingeon, Isler, Knöpfel, Weisstanner and Engler2015c). Using non-extrapolated data yields substantially identical results (see the section below on robustness tests). Debt ratios correlate with measures of budget deficits, but may be less prone to “creative accounting” of governments (Alt et al. Reference Alt, Dreyer Lassen and Wehner2012). As growth and interest payments also influence debt trajectories, debt changes encompass the entirety of public finance development both on the expenditure and on the revenue side. Nonetheless, my results are robust to alternative indicators such as budget deficits or government expenditures.

Government form

The differentiation between coalition/multi-party governments and one-party governments is the first important independent variable following from the theoretical argument (coalitions coded 1; one-party governments coded 0).

Prospective cooperation

“Prospective cooperation” encompasses current governing parties’ chances for future participation in governments in the same constellation. This measure is based on the idea of familiarity through shared past governing experiences, which should enable governments to take a more credible long-term perspective on debt dynamics. It is detrimental for prospective cooperation if current coalition partners frequently governed with other parties or governed alone in the past. Such dissenting behaviour reduces coalition partners’ familiarity and mutual trust (Tavits Reference Tavits2008); there is no stable foundation for long-term cooperation based on successful past collaborative experiences, thus damaging the credibility of fiscal policy agreements.

The measure for prospective cooperation is constructed retrospectively and oriented towards past cooperation, following Bäck and Lindvall (Reference Bäck and Lindvall2015) and similar measures by Persson and Tabellini (Reference Persson and Tabellini2009) and Martin and Stevenson (Reference Martin and Stevenson2010). It is assumed that each governing party judges the chances for prospective cooperation on account of its past cooperation experiences with other governing parties. As familiarity is a dynamic concept that is developed over time, all cooperation experiences since 1960 are taken into account for all governing parties in the data set.

However, there are reasons to “discount” past events because coalitions have to compare present benefits and future benefits facing uncertainty about the future (Lupia and Strøm Reference Lupia and Strøm2008). For governments with higher discount rates, the potential for prospective cooperation should depend more strongly on recent government constellations than on patterns from decades ago. The functional form of how past government events have to be discounted is unclear a priori, however. Measuring democratic capital, Persson and Tabellini (Reference Persson and Tabellini2009, 101) are “agnostic” about the functional form and assume that democratic capital accumulates relatively slowly in democratic years but collapses rapidly if a regime becomes authoritarian. Past events are discounted with a geometrical depreciation function in this specification. For prospective cooperation, a similar intuition seems reasonable, in that familiarity is inherently developed over a longer time period than it can be disestablished. Bäck and Lindvall (Reference Bäck and Lindvall2015) follow the logic of Persson and Tabellini (Reference Persson and Tabellini2009) and set the depreciation rate at 0.95. This leads to a weight of 0.95⁰=1 for an event one year ago, 0.95¹=0.95 for two years ago, 0.95²=0.903 for three years ago, 0.95³=0.857 for four years ago, etc. Events about 15 years ago carry half the weight of past year’s events. The closer the depreciation rate is to 1, the more all past events are weighted equally. The closer it is to 0, the more weight recent events carry.

Martin and Stevenson (Reference Martin and Stevenson2010) also use discount rates for their measure of familiarity. They argue that familiarity provides insider information about coalition partners, but that this process depends on the collaboration of party leaders. As they estimate average durations of party leadership in Western Europe at about eight years, cooperation events more than eight years ago are “almost completely discounted” (Martin and Stevenson Reference Martin and Stevenson2010, 510). This would imply a depreciation rate of about 0.7 or 0.8. However, this may be too restrictive as depreciation rate events three years ago weigh only about half (0.7²=0.49). Rather, a higher rate seems appropriate. In this article, I will mainly apply a depreciation rate of 0.95 – yet re-run all analyses for 20 different depreciation rates between 0.80 and 0.99 (online appendix).

The actual calculation of the indicator is leaned on Bäck and Lindvall (Reference Bäck and Lindvall2015) and correlates strongly with their measure (r=0.84, n=1,011).Footnote ¹ In an initial step, a measure for prospective cooperation is developed for each individual governing party j in country i in the year t:

where δ is a depreciation rate (δ=0.95 in the main analyses). τ denotes all past observations that are taken into account for calculating the indicator, starting with t−1 and ending with the first time point in the data set (1960 at the earliest). By multiplication with (1−δ), the sum of all past observations of prospective cooperation is normalised to an interval between 0 and 1. The degree of past cooperation is captured by α _i,j,t−τ . If a current governing party (year t) has previously participated in government in year t−τ, there are three possibilities:

• ∙ It gets a maximum value of α=1 if it was in a coalition with the exact same partners in the past year t−τ.

• ∙ It gets a minimum value of α=0 if it was in a coalition with entirely different partners in the past year t−τ. Note that this also includes cases where a party has governed alone in the past, but now forms part of a coalition.

• ∙ It gets values between 0 and 1 if only a share of the current coalition partners have cooperated in government in the past year t−τ. The value is then the share of current governing parties that have cooperated with party j in that year.Footnote ²

For single-party governments and new governing parties, different logics apply, as a comparison of their current and past cooperation status is not straightforward.

• ∙ Single-party governments: according to the argument, the policymaking outcomes of coalitions’ past cooperation patterns need to be compared with the outcomes of parties governing alone, that is not cooperating with other parties. Foremost, therefore, the effect of single-party governments versus coalitions is captured by a separate dummy variable. For the cooperation variable, single-party governments are simply assigned a constant value (whose exact level is technically irrelevant, due to the additional dummy variable). In accordance with the view that parties can be seen as durable, stable coalitions of societal interests (Aldrich Reference Aldrich2011) with similar incentives like durable coalition governments to internalise fiscal preferences, parties governing alone in a given year get a maximum value of α=1.

• ∙ New governing parties: although they have not established any familiarity with other governing parties in the past, I assign new parties in government a maximum value of α=1. The reason is that otherwise the measure would “punish” a new party for not having participated in previous governments, even though it might be willing to collaborate and compromise in office. Its coalition partners, however, have a relative lack of familiarity with the new party, which reflects in a lower cooperation outlook for established parties. This is a compromise between the view that new parties should have low cooperation prospects due to their lack of familiarity with the established parties and the view that new parties should have high cooperation prospects because they have never “betrayed” the established parties.

These cases reveal that governing parties within one coalition may have different individual values of prospective cooperation. I argue that the foundation for durable cooperation depends on the elaborated amount of familiarity among all coalition partners, as “no coalition government is stronger than its weakest link” (Bäck and Lindvall Reference Bäck and Lindvall2015, 58). The effects of familiarity on future cooperation prospects therefore prevail for the coalition as a whole; otherwise, some parties could impede debt stabilisation agreements, lacking long-term cooperation perspectives. In a second step, the potential for prospective cooperation for a government in country i in the year t is thus the lowest value of prospective cooperation of any current governing party:

Figure 1 illustrates the development of prospective cooperation for each country, while also displaying the dynamics of debt ratios. In countries continuously ruled by one-party governments or unchanged coalitions, prospective cooperation always reaches the maximum value of 1 (Australia, Canada, Cyprus, Malta, Spain and US).Footnote ³ In the rest of the 36 democracies, prospective cooperation varies over time. For example, in the United Kingdom (UK), the conservative-liberal coalition government formed in 2010 has less potential for prospective cooperation because the Conservatives had governed alone since 1960, and the Liberals have no familiarity through shared governing experience with the Conservatives. Indeed, the Conservatives went back to governing alone after the elections in 2015.

Figure 1 Prospective cooperation and debt ratios, 1960–2013 Note: Debt ratio of Bulgaria in 1993 (292%) reduced to 200% for graphical reasons. Source: Own calculations based on Armingeon et al. (Reference Armingeon, Isler, Knöpfel and Weisstanner2015b, Reference Armingeon, Isler, Knöpfel, Weisstanner and Engler2015c).

In Switzerland, another system of generally stable coalition patterns, with the same parties forming the government between 1959 and 2011, there is a surprising dip in 1995. This is because in our data source, the SVP was re-coded to a right-populist party since 1995, and in effect became a “new” party (in an ideological sense). Thus, the expectations for the “old” parties for long-term cooperation with the new party are reduced. However, the indicator drops only moderately because three of the four parties can still count on a high amount of familiarity.

Austria is an illustrative case to see the intuition of the indicator. Until 1982, after a long series of grand coalitions between 1947 and 1966, the SPÖ and ÖVP governed alone. In 1983, prospective cooperation decreases below 0.5 after the SPÖ for the first time entered a coalition with the FPÖ. The socialists thus have a high potential for “betrayal” against its partner because they had governed alone or with the ÖVP for such a long time. During the four-year government term, prospective cooperation slowly increases again. However, it slightly decreases again in 1987 with the next grand coalition between SPÖ and ÖVP, because the Social Democrats now have a betrayal potential against the Christian Democrats, too. Until 2000, prospective cooperation steadily increases under consecutive grand coalitions. It then falls again with the first coalition between ÖVP and FPÖ in 2000.

Japan is another demonstrative example. Until 1983, the Liberal Democrats governed alone. Between 1983 and 1986 and between 1994 and 2009, however, they had a series of small coalition partners. These have no familiarity because of the long-standing hegemony of the Liberal Democratic Party (LDP) and could not anticipate being a long-term part of a coalition government. Finally, the new accession countries of the EU in Central and Eastern Europe start from comparatively high levels of prospective cooperation because their party systems are dominated by new parties (Powell and Tucker Reference Powell and Tucker2014). However, there are also clear downward trends in countries like Estonia, Latvia and Slovenia because of unstable coalition patterns in the 2000s. Such downward trends, as well as increasing volatility over time, are observable on average in all countries. This can possibly be explained by party system changes, but in any case the cross-national variation remains considerable.Footnote ⁴

Looking at bivariate associations reveals negative but rather weak correlations between country-level averages of prospective cooperation and debt changes: r=−0.07 (n=36, p=0.69) for all countries and r=−0.17 (n=30, p=0.38), excluding countries with permanent maximum values of prospective cooperation due to one-party governments. Correlation coefficients within countries are negative in 21 countries (significant on the 90% level in Belgium, Croatia, Estonia, France, Greece, Iceland, Japan, Portugal, Romania, Slovenia, Sweden and UK) and insignificantly positive in nine countries (p>0.1 in all countries). Average debt changes confirm the theoretical expectations. One-party governments (average annual debt change 0.85% of GDP, n=532) and coalition governments with high values for prospective cooperation (debt change 0.80% of GDP, n=412, prospective cooperation>0.78) both have substantially lower debt increases than coalitions with low values for prospective cooperation (debt change 1.80% of GDP, n=418).

Control variables

These bivariate correlations should be scrutinised in a multivariate model controlling for alternative influences on debt changes. The variables presented here allow a replication of the models of Bäck and Lindvall (Reference Bäck and Lindvall2015). Their models in turn are based on the comprehensive empirical study by Franzese (Reference Franzese2002, Chapter 3). Control variables are from Armingeon et al. (Reference Armingeon, Isler, Knöpfel, Weisstanner and Engler2015c) and are described in detail in the online appendix. First, a number of economic control variables are introduced. The lagged values of debt changes and levels control for persistence of debt trajectories (see also a statistical motivation below). In addition, higher debt growth should be immediately caused by higher unemployment levels and changes, lower GDP growth. Fiscal pressure, which Hypothesis 3 expects to increase the effects of prospective cooperation, is operationalised as the leverage of GDP growth minus inflation minus nominal interest rates on lagged debt levels. The openness of the economy is not included in the main models as it proved insignificant.

Second, election years (binary indicator, 1=election year) should temporarily lower the effects of prospective cooperation, according to Hypothesis 2. Third, to disentangle cooperation effects from government stability (Bejar et al. Reference Bejar, Mukherjee and Moore2011), I include a measure for government duration based on the average duration of past government in days (applying a depreciation rate of 0.95).

Fourth, the cabinet share of left-wing parties is included as a control variable. The expected effects are somewhat ambiguous and may depend on macroeconomic conditions (Cusack Reference Cusack1999). Fifth, Edin and Ohlsson (Reference Edin and Ohlsson1991) associate minority governments (1=minority in parliament) with higher deficits, but this has been contested (de Haan and Sturm Reference de Haan and Sturm1997). Sixth, according to Hypothesis 4, the impact of prospective cooperation may be conditional on sufficiently centralised budget institutions under the control of strong finance ministers (Hallerberg and von Hagen Reference Hallerberg, Strauch and von Hagen1999). Thus, I include the “delegation index” by Hallerberg et al. (Reference Hallerberg and von Hagen2009, 74) in some models. Data are available for 15 European countries and three time points only. I assign the value from 1991 backwards until 1960 and the value from 2000/2004 until 2012. Values between 1991 and 2000/2004 are linearly interpolated.

Finally, constitutional veto players need to be taken into account (Tsebelis Reference Tsebelis2002). I include indices for presidentialism (0=parliamentarism, 1/2/3=semi-presidentialism/hybrid systems, 4=presidentialism), federalism (0=no, 1=weak, 2=strong) and bicameralism (1=unicameralism, 2=weak bicameralism, 3=medium-strength bicameralism, 4=strong bicameralism).

TSCS analysis

ADL models enable one to identify the temporal impacts of governments’ cooperation patterns. An ADL model is a “general model” in the sense that it makes no restrictive a priori assumptions about the dynamic associations between X and Y variables (De Boef and Keele Reference De Boef and Keele2008; Beck and Katz Reference Beck and Katz2011). The model is specified as follows:

where ΔDebtRatio is the annual change of general government debt in percentage of GDP and X is a vector of independent variables. β ₀ estimates the immediate short-term effect and β ₁ the effect of previous levels of X on debt changes. Including the lagged dependent variable α ₁ allows the calculation of long-run multipliers (LRM) $${\rm LRM \,{\equals}\,}{{\beta _{0} {\plus}\beta _{1} } \over {1{\minus}\alpha _{1} }}$$ , which estimate the total long-run effect of an impact variable X . ε _t is an idiosyncratic error term.

The estimation of the ADL with ordinary least squares regression depends on four assumptions, which tests show to be consistent. First, throughout the whole sample, the dependent variable seems to be stationary and not containing unit roots.Footnote ⁵ Second, Lagrange multiplier tests find no significant serial correlation among the residuals.Footnote ⁶ This avoids endogeneity bias risks by introducing the lagged dependent variable (Keele and Kelly Reference Keele and Kelly2006). Third, spatial correlation among the error terms and panel heteroscedasticityFootnote ⁷ require the use of “panel-corrected” standard errors (Beck and Katz Reference Beck and Katz1995). Fourth, tests hint that the data do not suffer from unobserved heterogeneity between countries.Footnote ⁸ Therefore, no country fixed effects need to be included (the results explaining variation within coalition governments remain robust in fixed-effects models, however). To control for common exogenous shocks and time trends, I include a year counter, and its squared and cubic transformations. Results are substantially identical using year fixed effects.

Government-centred analysis

TSCS data have a number of disadvantages. Their focus on annual short-term changes ignores that governing occurs with implementation lags. Moreover, government output is to some degree driven by electoral cycles, which naturally includes governmental changes in the middle of the calendar year. An alternative to TSCS data is therefore to set the unit of observation at the cabinet level, so that each observation captures the relevant average characteristics of one government.Footnote ⁹ As cabinet duration varies greatly, I control for the number of days governments were in office. Models are estimated with ordinary least squares (OLS) regression and panel-corrected standard errors. The main advantage of using this cabinet-centred data structure is a more realistic focus of the government as a main actor shaping policy (Boix Reference Boix1997; Obinger et al. Reference Obinger, Schmitt and Zohlnhöfer2014).