The political economy of the German Länder deficits: weak governments meet strong finance ministers

Abstract

We analyse the deficits of the German Länder (regional states) for the period 1960 to 2005 and test a number of hypotheses derived from the literature on the political economy of public deficits. We find evidence for the weak government hypothesis, that is, coalition governments issue significantly more debt than single party governments – a result that is typically explained by the common pool problem. As our data suggest, this result crucially hinges on the position or strength of the finance minister within coalition governments. We find that coalition governments with a strong finance minister are – in terms of borrowing – not significantly different from single party governments. In addition, we find (weak) evidence for an opportunistic political business cycle. As borrowing is significantly lower in pre-election years it appears that German voters favour fiscal discipline. There is no evidence for partisan behaviour, so party ideology seems to play a negligible role.

Acknowledgements

We are indebted to Tom Cusack, Benny Geys, Kai A. Konrad, Lars-Hendrik Röller, Friedrich Schneider, Helmut Seitz, Viktor Steiner, and the participants of several conferences and seminars. We thank Alexandra Günther and Walter Becker for providing the GDP data for regional states and Julia Gerstung and Anna Kossendey for excellent research assistance. Remaining errors are ours.

Notes

¹ Table A1 in the Appendix offers a more detailed (but still partial) review of the empirical literature related to our study.

² We consider a finance minister to be strong when he or she has the same party affiliation as the prime minister.

³ We ran our empirical models on shorter time periods, including those used in Seitz (2000) and Galli and Rossi (2002) and largely lose significance. In other words, the differences in results between our article and the other two are mainly due to the much longer time period that we consider.

⁴ In our empirical model, state fixed effects account for that fundamental difference as well as for other time-invariant state characteristics.

⁵ For the same reason, the other two papers analysing fiscal variables of the German states, namely, Seitz (2000) and Galli and Rossi (2002), also concentrate on data at the state level.

⁶ In 2007, the PDS merged with the ‘Alternative Choice Employment and Social Justice’ (Wahlalternative Arbeit und Soziale Gerechtigkeit (WASG)). The latter largely consisted of disappointed former social democrats and union members. The so-formed new party, The Left (DIE LINKE), is about to gain influence in West German states.

⁷ To some extent, Berlin is an exception as the SPD currently forms a coalition with the DIE LINKE. Note, however, that today's Berlin is not a former Western Land.

⁸ For an explanation of the variables, see Table A2 in the Appendix.

⁹ Of course, one may argue that if the election is early in the year, expenditures should raise in the pre-election year. We discuss this in detail in Subsection ‘Political opportunism’ of Section V.

¹⁰ Note that we concentrate on public deficits. It may well be that there are no partisan effects in borrowing but in the structure of public spending (see, for instance, Drazen and Eslava, 2005).

¹¹ An alternative theory that explains higher deficits for coalition governments is offered by Alesina and Drazen (1991) and Alesina and Perotti (1994, pp. 22–29): consider a permanent fiscal shock. Coalition partners will then fight about the allocation of the fiscal burden to the respective constituencies. This situation is well modelled by the ‘war of attrition’. In general, delayed adjustment to the fiscal shock will obtain, allowing debt to accumulate.

¹² Note that real debt growth is simply real deficit over real debt.

¹³ Definition of all variables can be found in the Appendix in Table A2.

¹⁴ Elections are not uniformly distributed over time. This is why year dummies are correlated with the variables ELECTION and PREELEC. While the ELECTION coefficient is never statistically different from zero we lose significance for PREELEC in a model with time fixed effects. All other results reported below are robust to this alternative specification.

¹⁵ One may argue that debt issue is influenced by the possibility of a federal bailout so that the bailout variables are endogenous. Since debt started to accumulate in the 1970s in most states, about 20 years prior to the first and so far only bailout, it is hard to imagine that borrowing incentives were influenced by the possibility of a bailout. Moreover, there is no rule or directive specifying when the federal government has to step in. There is, thus, no reason to believe that the corresponding variables are endogenous.

¹⁶ The Hausman test suggests that the random effects model is consistent. Note, however, that the test requires that the fixed effects estimator is consistent. As this is clearly violated in a dynamic model (see below) we follow our intuitive argument and use fixed effects.

¹⁷ He also showed, however, that the bias approaches zero as T tends to infinity. Since T is relatively large in our study (T = 46), the bias is likely to be moderate. Note that although T is much smaller in Seitz (2000, T = 21) and Galli and Rossi (2002, T = 21) both studies use the LSDV estimator.

¹⁸ We consider the regressors summarized in w _it as strictly exogenous so that variables themselves and all their lags are valid instruments. Furthermore, note that the AB estimator takes first-order autocorrelation of into account. Thus, neither consistency nor efficiency is affected by first-order autocorrelation. But second-order correlation implies inconsistency (Arellano and Bond, 1991, pp. 281–282).

¹⁹ The estimates with BB as initial estimator have slightly higher SEs. Apart from that, results remain unchanged. The complete estimates for the BB and AH estimators are available upon request.

²⁰ Deficit data are taken from the Statistisches Bundesamt (Federal Statistical Office) (2005). Data for the GDP was provided by the Statistical Office of Baden Württemberg and the Federal Statistical Office. All nominal numbers were deflated by the consumer price index for all households obtained from the Statistisches Bundesamt (2006). Our main results do not change qualitatively when using nominal values instead. Election dates were taken from Forschungsgruppe Wahlen (Election Research Team) (2007).

²¹ Some coefficients change substantially which suggests that a regression without political variables suffers from omitted variable bias.

²² Note that the 1994 bailout variable and EQUAL are highly correlated; they only differ in 1994. Even if we drop one of them we do not gain significance of the other. Although correlation is dampened under the alternative codes of the bailout variable it remains insignificant.

²³ This differs from Galli and Rossi (2002) who found significantly positive election year effects but no pre-election year effects.

²⁴ Although we will eventually adopt Model 2 one should note that evidence for political opportunism is not overwhelming. The p-value is 9.4% and only two of five specifications (Models 2–6) show significance at the 10% level.

²⁵ So suppose the cut-off is 31 January, then an election in, say, January 1982, would have 1981 as election year and 1980 as pre-election year. An election in February 1982 would have 1982 as election year and 1981 as pre-election year.

²⁶ This robustness is not too surprising since there are only very few elections early in a year (8% of all elections were held in January or February).

²⁷ Results are independent of the alternative adopted.

²⁸ Again, in years of government changes, we use the inaugural date of the new government and 1 July as the cut-off date to assign a value to COAL.

²⁹ There never was a coalition government with more than three parties.

³⁰74% of coalition governments have a prime minister and a finance minister belonging to the same party.