156 4.1.2 Economic voting and tax reforms – hypotheses How can we analyze the relationship in between tax policy and the voting decision? Which hypotheses should we test? The best case for an analysis of the electoral consequences of tax policies would be if we could use detailed data on the distributional consequences of tax policy on a micro-level. Then we could calculate who profits from tax policy changes, who loses and how these welfare changes affect the vote. However, we are far from disposing of these data.280 Based on the data-set we employ here we have to put the question differently. The most important variables we have are the extent of tax increases and of tax reductions and the net fiscal effects of reforms (increases minus reductions). Nonetheless, we argue that these variables allow us to test important and previously inaccessible hypotheses.281 Most importantly we can analyze whether the fiscal effects of tax reforms in the legislative period previous to the election play a role in the voting decision. Based on the main questions discussed in the literature on economic voting we can test several important hypotheses based on the fiscal effects of tax reforms: As we have only data on past policies and not on voters´ expectations at the time of the elections, our first hypothesis is concerned with the general importance of past policies (retrospective voting). Do we observe an important influence of past tax policies on the re-election probabilities of incumbent governments (and how strong is it)? A sub-hypothesis is whether voters focus on tax policy decisions (reflected by the adoption of tax reforms by the chambers of parliament) or on reforms being implemented (reflecting the effective change in the tax burden). If we find evidence for retrospective voting we can ask: What is the time horizon of tax policies in the consideration of voters? Does just tax policy in election years matter, or do voters take the policy record in the whole legislative period before an election into account? Finally, we can address the question of a grievance asymmetry: Do voters punish governments more heavily for past tax increases than they reward tax reductions? 4.2 The influence of tax reforms on elections – empirical analysis To test the derived main hypotheses on tax policy and economic voting, we discuss first how to measure the independent variables in the analysis: changes in voting decisions affected by tax policies. Then we move on to the empirical analysis of economic voting which integrates macroeconomic variables and tax policy. 280 Examples for studies that analyze the consequences of a tax reform for different income deciles are Haan and Steiner (2005) or Corneo (2003). However these studies are limited to only one income tax reform. 281 See description and discussion of the data-set in part III and IV. 157 4.2.1 The independent variable – changes in the vote On which variables should we focus to measures changes in the vote? As we analyze the effects of tax policy in case of retrospective voting, we concentrate on the election results of the incumbent governments in form of changes in the vote share. In our view this is the best available indicator to measure content or discontent of voters with economic conditions and tax policy. The time frame for our analysis is given by our data-set on tax policy which starts in 1964/65 and ends in 2004. As we have seen in the discussion of the tax policy process (see part V.3.1), the federal government plays the central role in tax policy as the states have only very limited autonomy. Therefore, we argue that the federal government is hold responsible for tax policy by the voters and focus on economic voting in federal elections.282 Eleven federal elections fell within the period covered by our data-set. We have to exclude 1965 (coming down to ten elections), as our data-set starts only in 1964/65 and we have no reliable data on the relevant period before the election in 1965. Figure 70 shows our independent variable: the change in the vote share for the incumbent governments on the y-axis together with the vote shares of the newly established governments on the x-axis.283 All governments were coalition governments headed either by the CDU/CSU or the SPD. Altogether we observe seven reelections (1972, 1976, 1980, 1987, 1990, 1994 and 2002) and three changes of the federal government (1969, 1983 and 1998) within the period studied. After the grand coalition of CDU/CSU and SPD from 1966 to 1969, the vote share of the CDU/CSU was slightly reduced, while the SPD expanded its vote share strongly. A SPD/FDP coalition came into power but had only a total vote share of 48%.284 In the following election in 1972 the SPD/FDP government was re-elected as the SPD as well as its coalition partner, the FDP, gained strongly. Both parties expanded their common vote share by nearly 6 percentage points while the CDU/CSU lost. In 1976 the SPD lost strongly and her coalition partner, the FDP, lost slightly, but they nonetheless remained in government. In 1980 the vote share of the SPD stagnated, but the FDP won strongly leading to another re-election of the government. In 1983 (after the constructive vote of no-confidence in 1982) there was a landslide victory for the CDU/CSU which formed a coalition with the FDP 282 It would be interesting to analyze further, whether the influence of fiscal effects of reforms on the vote does as well depend on the variable divided and undivided government (see part V.3). However, we have to postpone this question to a later analysis. 283 To explain the graph: for example in 1972 the SPD/FDP government was re-elected and increased its vote share by 5.7 percentage points and disposed of 54.2% of the total votes. Note that there is no direct link in between the vote share of the new government and the vote share change for the (not re-elected) incumbent government in case of changes of government. 284 The coalition had nonetheless a majority of 12 parliamentary seats. This resulted partly from the far-right National-Democratic Party, the NPD, which received 4.6% of votes but no parliamentary mandates as it remained slightly below the threshold of 5% of votes. 158 and the GREENS made it into parliament for the first time. In 1987 the CDU lost strongly, but their coalition partner, the FDP, gained strongly leading to a re-election of the CDU/CSU and FDP government with a smaller total vote-share. In 1990 the CDU/CSU lost slightly but their coalition partner, the FDP, gained strongly and the CDU/CSU and FDP government remained in power. In 1994 both CDU/CSU and FDP lost but were re-elected. In 1998 strong losses of the CDU/CSU led to a government change from CDU/CSU and FDP to SPD/GREENS. In 2002 there was almost no change in the vote shares leading to a re-election of the SPD/GREENS government. Source: Federal election statistics. VOTE SHARE CHANGES AND VOTE SHARES GERMANY, 1969-2002 = change of federal government 1969 1972 1976 1980 1983 1987 1990 1994 1998 2002 -10 -8 -6 -4 -2 0 2 4 6 41 43 45 47 49 51 53 55 57 VOTE SHARE FEDERAL GOVERNMENT (%) CHANGE IN VOTE SHARE (percentage points) Figure 70: Vote share changes and vote shares of federal governments 1965-2004 Of all the elections 1969 was special because of two reasons. First, the government changed during the previous legislative period from CDU/FDP (1965-1966) to a grand coalition of CDU/CSU and SPD (1966-1969). Therefore, it is not clear whether the policy in the previous legislative term was related equally to the CDU/CSU and the SPD or more to the CDU/CSU (which had governed from 1965 to 1966 together with the FDP). Second, the situation for the voters after the grand coalition was different than usual. In 1969 it was harder to punish the government directly for bad policies as it was impossible to vote both coalition partners (CDU/CSU and SPD) out of office. It was almost certain that one of the governing parties would be the leading partner of the next coalition government. Therefore, we 159 should expect that retrospective voting might have been less strong at this election. We come back to this point later in the analysis. 4.2.2 Empirical analysis of the influence of tax policy on voting The literature on economic voting identifies a strong influence of the macroeconomic variables real GDP growth (which is linked to changes in unemployment) and inflation on the election results. What role did these variables play in the federal elections in Germany from 1969 to 2004? We have tested the relationship between these variables and vote changes and found a strong correlation between real GDP growth and vote changes while there were only weak correlations between vote share changes and the other macroeconomic variables. Figure 71 demonstrates the correlation of average GDP growth per year in the legislative periods before elections and vote changes from 1965 to 2004 graphically. We see that governments increased their vote share whenever average annual real GDP growth in the legislative period before elections was higher than 2.5% while the governments´ vote shares were reduced in all other cases. N=10 -12 -10 -8 -6 -4 -2 0 2 4 6 8 -1 -0,5 0 0,5 1 1,5 2 2,5 3 3,5 4 4,5 AVERAGE REAL GDP GROWTH p.a. V O TE C H A N G E (% P O IN TS ) 1990 1980 1972 2002 1987 1976 19941998 1983 1969 AVERAGE PER LEGISLATIVE PERIOD % THE MACROECONOMY AND VOTE CHANGES 1965-2004 Own calculations based on: Federal Ministry of Finance (2004)/Federal Statistical Office (2007). Federal Election statistics (2007 ). Figure 71: Effects of real GDP growth on votes changes (legislative periods) 160 What do we find if we analyze the effects of macroeconomic developments and tax policy on vote changes in federal elections together? Based on the hypotheses we derived (see part V.4.1.2) we should test whether changes of the relevant variables in the year before the elections or the average development in the whole legislative period were decisive. Furthermore, we should check whether tax policy (reflected by adopted tax reforms) or direct changes of the tax burden (reflected by implemented tax reforms) were crucial. Finally, we should distinguish in between tax reform data adjusted for temporary measures and including temporary measures.285 We have estimated several empirical specifications with different combinations of the relevant variables. Our dependent variable is the vote share change of incumbent governments in percentage points. To make tax reforms as one independent variable in different years comparable, we have divided the fiscal effects of reforms by nominal GDP. To cover the influence of macroeconomic developments on electoral results we have integrated the real GDP growth rate as independent variable in the analysis. As we dispose only of a small sample of ten observations, the results of a regression analysis are especially vulnerable to outliers. Therefore, we test for a statistical relationship within our small sample but review the bivariate relationships in between vote share changes, GDP growth, and tax reforms graphically as well (see Figure 71, Figure 72 and Figure 73). We test for a relationship in between our variables by the following equations using OLS: ln tt10 ??? +?+=? GDPVotet GDPln tt2t10 ???? +?+?+=? TVotet With: 0? , x? Constant, coefficients t election years ?Votet Vote share change incumbent government (percentage points) ?lnGDPt Real GDP growth rate (in the election year or average annual growth rate in the legislative period preceding the elections) ?Tt Fiscal effects of tax reforms/GDP (in the election year or average annual effects in the legislative period preceding the elections) t? Stochastic error term 285 See for a discussion part III.3. 161 Constant Real GDP growth rate R2 (adjusted) Dependant Variable: Explanatory Variables (election years): DW Notes: t-statistics of the estimated parameters in parantheses. * significant at the 10% level; ** significant at the 5% level; *** significant at the 1% level Observations TAX REFORMS AND ELECTIONS 1965-2003 FISCAL EFFECTS PER YEAR (ELECTION YEARS+) Fiscal effects tax reform (implemented)# Fiscal effects tax reform (implemented)# Adjusted for temporal measures Fiscal effects tax reform (adopted)# Fiscal effects tax reform (adopted)# Adjusted for temporal measures Vote Change (%points) 0.816567 (1.418266) -4.079868 (-1.832424) 0,10 2.75 10 (OLS) A Vote Change (%points) 0.43 1.73 -4.064399* (-2.300913) 0.913797* (1.992626) (OLS) -5.038** (-2.390) 10 B Vote Change (%points) 0.40 2.38 -3.934783* (-2.1615577) 0.778987 (1.654858) (OLS) -7.983* (-2.232) 10 C Vote Change (%points) 0.25 1.77 10 -2.534* (-1.596) -5.093968* (-2.386260) 1.056947* (1.928389) (OLS) D Vote Change (%points) 0.14 2.44 -2.985 (-1.155) -5.226577* (-2.180709) 0.963844 (1.666600) (OLS) 10 E +1965, 1969, 1972, 1976, 1980, 1982 (election in March 1983), 1986 (election in January 1987), 1990, 1994, 1998, 2002;. # as a percentage of nominal GDP Table 11: Effects of GDP growth and fiscal effects of tax reforms on votes share changes (election years) We start by reviewing the results for the influence of GDP growth and fiscal effects of tax reforms in the election years.286 Our results applying OLS regressions are displayed in Table 11. Real GDP growth alone is not significant to explain vote share changes and is only significant on a 10% level if combined with the fiscal effects of tax reforms. The coefficient has the expected sign so that increases in real GDP growth are correlated with vote share increases. With respect to tax reforms only unadjusted fiscal effects of adopted reforms are significant at a 5% level (if adjusted for temporary measures only at a 10% level). The coefficients point in the expected direction with tax burden increases influencing the change in the vote share negatively. With respect to the explanatory impact, specification B – based on real GDP growth and fiscal effects of adopted reforms – performs best. Specification B is able to explain 43% of the vote share changes. How do our results change if we analyze average annual values of the relevant variables in the whole legislative period preceding the election? Table 12 shows that 286 We have analyzed the influence of fiscal effects of tax reforms and GDP growth in the election years when elections took place after September (as in 1969, 1972, 1976, 1980, 1990, 1994, 1998 and 2002) and in the year before elections if the elections took place before March (in 1982 (election in March 1983) and 1986 (election in January 1987)). 162 here the results are far stronger. Real GDP growth is significant on a 1% level for all specifications. Average real GDP growth is positively related with vote changes of the incumbent government with a one percentage point higher real GDP growth (per year of the legislative period) leading to an increase of the vote share of around 3 percentage points. Fiscal effects of adopted reforms are significant on a 5% level while fiscal effects of implemented tax reforms are significant even at a 1% level. The coefficients show the expected signs for all specifications (tax burden increases have a negative influence on vote shares). The effect is strongest in case of implemented tax reforms (specification D). Here an average annual tax increase of 1% of GDP leads to a decrease of the vote in the following elections of 7 percentage points.287 Constant Real GDP growth rate/ R2 (adjusted) Dependant Variables: Expl. Var. (Average LP): DW Notes: t-statistics of the estimated parameters in parentheses. * significant at the 10% level; ** significant at the 5% level; *** significant at the 1% level Observations TAX REFORMS AND ELECTIONS 1965-2004 AVERAGE FISCAL EFFECTS PER YEAR (LEGISLATIVE PERIOD [LP]) Vote Change (%points) 0,64 1.77 10 (OLS) 2.932*** (4.113) -8.771*** (-4.494) A Vote Change (%points) 0.81 1.97 -9.354*** (-6.530) (OLS) -5.597** (-2.856) 10 2.847*** (5.489) B Vote Change (%points) 0.74 2.18 -9.561*** (-5.664) 2.992*** (4.976) (OLS) -4.667* (-2.066) 10 C Vote Change (%points) 0.85 1.77 10 -7.166*** (-3.543) -9.865*** (-7.672) 3.034*** (6.640) (OLS) D Vote Change (%points) 0.77 1.83 -6.119** (-2.423) -10.112*** (-6.184) 3.191*** (5.578) (OLS) 10 E Own calculations based on: Federal Ministry of Finance (2004), Federal Statistical Office (2007). # as a percentage of nominal GDP Fiscal effects tax reform (implemented)# Fiscal effects tax reform (implemented)# Adjusted for temporal measures Fiscal effects tax reform (adopted)# Fiscal effects tax reform (adopted)# Adjusted for temporal measures Table 12: Effects of GDP growth and fiscal effects of tax reforms on vote share changes (legislative periods) Specification D – the combination of average annual fiscal effects of implemented tax reforms and average annual GDP growth in the legislative period before the elections can explain 85% of the vote share changes of governments compared to an explanation of 64% of vote changes by average GDP growth alone. 287 On average our data-set shows tax burden reductions of 0.13% of GDP per year. 163 What do we conclude from these results? First, the inclusion of our data on fiscal effects of tax reforms strongly increases our ability to explain vote share changes of incumbent governments. Second, we found that average annual values of GDP growth and fiscal effects of tax reform throughout the whole legislative period preceding the federal elections performed far better than the values of the same variables only in election years. This shows that voters cannot be fooled by policy changes only in the election year but react to economic developments and tax policies throughout the legislative period preceding the elections. Third, we found that fiscal effects of implemented reforms perform slightly better in our regressions. The adjusted R2 values are slightly higher and the coefficients are larger. Voters did react stronger to changes in their tax burden (reflected in fiscal effects of implemented reforms) than to tax policy (as reflected in the fiscal effects of adopted tax reforms) although the difference is not too large. Our results are similar, independent of whether we adjust our data for the temporary measures in effect around 1970 or not (compare column D and E in table 12). Compared to other approaches in the literature, which can explain around 30% of the vote change based on economic voting, our result of an explanation of up to 85% of the vote share change just by GDP growth and fiscal effects of tax reforms are exceptionally strong. In our view this underlines the importance of tax reforms for elections. -10 -8 -6 -4 -2 0 2 4 6 8 -0,6% -0,4% -0,2% 0,0% 0,2% 0,4% 0,6% FISCAL EFFECTS OF IMPLEMENTED TAX REFORMS/GDP p.a. V O TE C H A N G E (% P O IN TS ) N=10 1990 1980 1972 2002 1987 1976 1994 1998 1983 1969 GRAND COALITION (EFFECT ON CDU) AVERAGE PER LEGISLATIVE PERIOD GRAND COALITION (EFFECT ON SPD AND CDU) 1969 Adjusted for temporary measures (1967-1975). TAX REFORMS AND VOTE CHANGES 1965-2004 AVERAGE FISCAL EFFECTS OF TAX REFORMS Data sources: Federal Ministry of Finance (2004), Federal Election statistics (2007 ). Figure 72: Effects of tax reforms on vote share changes 1965-2004. 164 To check the reliability of the results derived in our statistical analysis, Figure 72 illustrates the bivariate relationship in between fiscal effects of implemented tax reforms and vote changes graphically. Figure 73 compares the effects of changes in election years and average changes in the whole legislative period before elections. 52Gerrit B. Koester – The economics of tax reforms Deutsche Bundesbank; November 26th 2007 THE MACROECONOMY AND VOTE CHANGES 1965-2004 -12 -10 -8 -6 -4 -2 0 2 4 6 8 -1 -0,5 0 0,5 1 1,5 2 2,5 3 3,5 4 4,5 AVERAGE REAL GDP GROWTH p.a. V O TE C H A N G E (% P O IN TS ) FEDERAL ELECTIONS N=10 1990 1980 1972 2002 1987 1976 1994 1998 1983 1969 (LEGISLATIVE PERIOD) % Own calculations based on : Federal Ministry of Finance (2004)/Federal Statistical Office(2007). 53Gerrit B. Koester – The economics of tax reforms Deutsche Bundesbank; November 26th 2007 THE MACROECONOMY AND VOTE CHANGES 1965-2004 FEDERAL ELECTIONS N=10 1990 1980 1972 2002 1987 1976 1994 1998 1983 1969 (ELECTION YEAR) % Own calculations based on : Federal Ministry of Finance (2004)/Federal Statistical Office(2007). -10 -8 -6 -4 -2 0 2 4 6 8 -2 -1 0 1 2 3 4 5 6 7 8 AVERAGE REAL GDP GROWTH p.a. V O TE C H A N G E (% P O IN TS ) 54Gerrit B. Koester – The economics of tax reforms Deutsche Bundesbank; November 26th 2007 TAX REFORMS AND VOTE CHANGES 1965-2004 AVERAGE FISCAL EFFECTS OF TAX REFORMS (IMPL.) -10 -8 -6 -4 -2 0 2 4 6 8 -0,6% -0,4% -0,2% 0,0% 0,2% 0,4% 0,6% FISCAL EFFECTS TAX REFORMS/GDP p.a. V O T E C H A N G E (% P O IN T S ) FEDERAL ELECTIONS N=10 1990 1980 1972 2002 1987 1976 1994 1998 1983 1969 GRAND COALITION (EFFECT ON CDU) (LEGISLATIVE PERIOD) GRAND COALITION (EFFECT ON SPD AND CDU) 1969 Fiscal effects/GDP Own calculations based on : Federal Ministry of Finance (2004)/Federal Statistical Office(2007). 55Gerrit B. Koester – The economics of tax reforms Deutsche Bundesbank; November 26th 2007 TAX REFORMS AND VOTE CHANGES 1965-2004 AVERAGE FISCAL EFFECTS OF TAX REFORMS (IMPL.) -10 -8 -6 -4 -2 0 2 4 6 8 -1,5% -1,0% -0,5% 0,0% 0,5% 1,0% FISCAL EFFECTS TAX REFORMS/GDP p.a. V O TE C H A N G E (% P O IN TS ) FEDERAL ELECTIONS N=10 1990 1980 1972 2002 1987 1976 1994 1998 1983 1969 (ELECTION YEARS) Fiscal effects/GDP Own calculations based on : Federal Ministry of Finance (2004)/Federal Statistical Office(2007). Figure 73: Effects of GDP growth and tax reforms in election years and legislative periods on vote changes288 Three additional observations from a graphical inspection are especially important. First, the graphs support the negative correlation of tax burden changes throughout the legislative period and changes in the vote share which we found based on our statistical analysis. In direct comparison we see that the effects of average changes in the whole legislative period before the elections were far stronger than the effects of changes just in election years (see Figure 73). Second, we see that average annual tax burden increases via tax reforms led in all legislative periods to a decrease in the vote share – with exception of 1969 (see Figure 72). 1969, however, is a special case 288 Please note that the observation of small net tax burden increases in the election years 1969 and 1976 do not contradict the hypothesis of opportunistic scheduling of tax reforms. In the legislative periods before these elections annual tax burden increases were by far larger on average than before in the election years. 165 as the voters were unable to really punish the grand coalition government as both major parties governed together. Furthermore, the change in the vote share of the incumbent government – the grand coalition – increased in 1969 only if we take both coalition partners (CDU/CSU and SPD) into account. If we focus on the CDU/CSU – which was probably held responsible for tax policy as the minister of finance was a member if the CDU/CSU during the grand coalition and the CDU/CSU governed with the FDP for the first year of the legislative period – the vote share decreased in 1969 as well. Third, we see that the relationship of overall tax burden reductions and changes in the vote share was less strong than the relationship in between tax burden increases and changes in the vote share (see Figure 72). In 1983 for example, a reduction of the tax burden went along with a very strong reduction in the vote share. This can be seen as pointing towards the importance of a grievance asymmetry: voters punished tax increases more heavily than they rewarded tax burden reductions. 4.3 Summary What can we conclude based on the findings of our empirical analyses ? First, we found a strong influence of real GDP growth and of tax reforms on vote share changes. The total fiscal effects of tax reforms – which we are able to measure based on our new data-set – were strongly correlated with vote share changes. A combination of real GDP growth and the fiscal effects of tax reforms can explain up to 85% of the vote share changes of the incumbent governments in the ten federal elections analyzed here. Thereby implemented reforms had a slightly stronger influence on vote changes than adopted reforms. Taken together our main results can be seen as a strong indication for the importance of retrospective voting and especially for the influence of tax policy on the voting decision. Second, we found that the average effects of tax policy and macroeconomic development within the whole legislative period before the elections was very strong while there was only a very weak relationship in between changes in the election year and vote share changes. This indicates that voters couldn´t be easily manipulated by policy changes before the elections but took the policy record of a whole legislative period into account. Finally, we found some indication that a grievance asymmetry mattered with respect to tax policy as well. As we have seen, the link between tax burden increases and vote share changes was stronger than the link in between tax burden reductions and vote share increases. However, this finding is less well founded than the first two and would call for further investigation not pursued here. Compared to the ability of other approaches in the literature, which can explain around 30% of the vote change based on economic voting, our result of an explanation of up to 85% of vote share changes just by GDP growth and fiscal effects of tax reforms is exceptionally strong. This speaks as well in favor of an especially important role of tax reforms for electoral outcomes.