Tax policy and yardstick voting in Flemish municipal elections

Abstract

Recent theoretical papers develop political agency models in which voters compare tax policy with that in neighbouring jurisdictions. In these yardstick competition models voters judge incumbents by comparing their policy with policy in neighbouring jurisdictions. This paper reports an analysis of municipal elections in Flanders during the period 1982 to 2000 and finds empirical evidence for yardstick voting. Incumbents are punished for higher tax rates. Importantly, the electoral punishment also depends on tax rates in neighbouring municipalities. Higher rates in neighbouring municipalities are favourable for the incumbents.

Acknowledgements

We thank participants at the EPCS Annual Meeting in Berlin 2004 as well as at the MICE seminar at the Vrije Universiteit Brussel for helpful comments. We especially thank John Ashworth, Benny Geys, Martin Paldam, Kurt Schmidheiny and an anonymous referee. We thank Jo Buelens for providing us with electoral data and the Research Foundation-Flanders for financial support.

Notes

¹ Among others, Ferejohn (1986), Rogoff (1990), Banks and Sundaram (1993), Besley and Case (1995), Persson and Tabellini (2000) and Revelli (2002a).

² The idea of relative performance evaluation was introduced by Lazear and Rosen (1981), Holmström (1982) and Nalebuff and Stiglitz (1983). Shleifer (1985) introduced the concept of yardstick competition. The benefits of relative performance evaluation in decentralized government were first explored by Salmon (1987).

³ An effect of spending growth on presidential elections is also found by Cuzán and Bundrick (1999).

⁴ Brender (2003) finds an impact of fiscal performance (debt and debt change) in local elections in Israel.

⁵ Ashworth and Heyndels (1997, 2000) and Heyndels and Ashworth (2003) present evidence that in Flemish municipalities politicians’ opinions on whether a tax rate is high (or low) or on whether taxes should be increased, depend on prevailing tax rates in neighbouring municipalities.

⁶ This includes the provinces however.

⁷ The six ‘national’ parties were: the ecologist Agalev, the Christian democratic CD&V, the social democratic SP.a, the liberal democratic VLD, the nationalist VU and the extreme-right Vlaams Blok. For these parties, we put the word national between quotation marks, as their political action is limited to only the Flemish Community, i.e. the Flemish speaking subset of the Belgian population. In federal elections, French-speaking ecologists, Christian democrats, … participate too. These parties are separate entities without formal links with their Flemish speaking sister parties. At the beginning of the seventies Belgian political families split up in Dutch speaking party and a French speaking party.

⁸ Note that the local property tax is levied in all 308 municipalities; the local income tax is in use in 305 municipalities.

⁹ Note that this is not a lagged dependent variable. This is only a lagged dependent variable when the previous government remained in power, i.e. then .

¹⁰ The municipality of Herstappe is considered as an outlier and is left out. The municipality, the smallest in Flanders, had only 85 inhabitants and 72 voters in 2000.

¹¹ The electoral data come from the Elections Database of the Political Science Department of the Vrije Universiteit Brussel. Data on municipal governments come from the municipal data collection ‘Gemeentelijk Zakboekje’(1985, 1994, 1999, 2002).

¹² In this we follow Buelens and Deschouwer (1997).

¹³ As we include the per capita expenditures in Flemish neighbouring municipalities in our regression, we loose an additional observation (the municipality of Voeren has no Flemish neighbours). This leaves us with 688 observations. We did not have expenditure data for Brussels and Walloon municipalities.

¹⁴ The argument is also true for divided government (Nicholson et al ., 2002).

¹⁵ We also include neighbouring municipalities in the Brussels and Walloon region, as the context is very similar. The local income tax and the local property tax are also the major local taxes and the tax bases of both taxes are defined on the federal level.

¹⁶ As we do not have expenditure data for the Brussels and Walloon region, the average per capita expenditures in neighbouring municipalities take only into account Flemish neighbours.

¹⁷ However, a study by Heyndels (1989) shows both the local income tax and the local property tax are identified as a local tax by around half of the voters. Voters were asked to name the most important taxes in their municipality. Around 61% of the voters could name a local tax. Of this group respectively 49.5% and 48.5% mentioned respectively the local property tax and the local income tax as a local tax.

¹⁸ Heyndels and Smolders (1994) analyse the presence of different types of fiscal illusion in Flemish municipalities. They find no evidence for renter illusion.

¹⁹ We include party-year dummies for the five ‘national’ parties that participated in municipal government. A dummy is one if a certain party was part of municipal government in that election.

²⁰ These year effects measure the electoral change common to all governments in a certain year. They could for example measure the electoral rise of the extremist party Vlaams Blok (which was not present in any of the municipal governments).

²¹ We had only data for house prices for the period 1990–2000. We therefore used data from 1990 for the election in 1988.

²² In these 2SLS regressions the area of the municipality and the neighbouring municipalities are dropped as instruments, as the first-stage regressions also include fixed municipality effects and the area variables are time-invariant.

²³ Moreover, our dataset is composed of only three time periods. Taking first differences would imply that we are left with only two time periods and lose a significant amount of observations. With only two time periods left, we would also not be able to present a test for second-order autocorrelation, which is necessary to test the validity of the method.

²⁴ We also leave out insignificant party-year effects. The Wald test is for the remaing party-year effects.