Political budget cycles in U.S. municipalities

Specifically, the excess volatility caused by the political budget cycle reduces welfare by making it more difficult for constituents to smooth their levels of consumption over time.

As discussed below, Levitt (1997) finds that cities increase police hiring while Krebs (2008) finds evidence of an increase in intergovernmental revenue in election years. However, no other study details how local governments manipulate total expenditures, taxes, and intergovernmental revenue, as well as their disaggregated components. In this sense, our paper adds to our understanding of local political budget cycles by describing the phenomenon in significantly more detail.

Levitt (1997) exploits electoral cycles in police hiring to identify the effect of police on crime, but it does not examine political budget cycles per se.

Drazen and Eslava (2010) examine electoral cycles in the composition of taxes and spending in a developing country context.

See Shi and Svensson (2003) and Drazen (2000) for a review of the literature.

Using data from the U.S., Tufte (1978) finds evidence of a pre-election increase in social security payments and veterans’ benefits. Alesina (1988), using US data over the years 1961–1985 finds an increase in net transfers over GNP in election years. Alesina et al. (1992), Alesina et al. (1997) examine data from thirteen OECD countries over years 1961–1993 and find that the government budget deficit is 0.6 % of GDP higher in election years. There has also been substantial work examining the prevalence of country-level political budget cycles in developing countries. According to Schuknecht (1996), developing countries have greater potential to manipulate fiscal policy, as checks and balances are weaker and the incumbent has more power over fiscal policy. Schuknecht examines 35 developing countries over years 1970–1992 and finds that the deficit is increased in election years, but decreased the year following an election. Using data from 17 Latin American countries over years 1947–1982, Ames (1987) finds that government expenditures increased by 6.3 % in election years, but decreased by 7.6 % in the year after elections. Block (2000) used data from 44 sub-Saharan African countries over years 1980–1995 and finds that the deficit, as a share of GDP, increases by 1.0–2.6 % in election years and decreases by 1.5 % in the year after an election.

Veiga and Veiga (2007) and Foucault et al. (2008) find evidence of increased pre-election expenditures at the municipal level in Portugal and France, respectively. Drazen and Eslava (2010), using municipal election data from Colombia, find evidence of changes in the composition of expenditures in election years.

Poterba and Rueben (1995) define the following 11 states as those that do not have effective limits: Connecticut, Georgia, Hawaii, Maine, Maryland, New Hampshire, South Carolina, Tennessee, Vermont, Virginia, and Wisconsin.

Smart et al. (2011), using data on German local governments, find evidence that changing from a council-manager system to a mayor-council system leads to increased local government expenditures. Since the sample used in our analysis is limited to city-years in which a mayor-council system is used, we do not expect that this “switching forms of government” effect is influencing our results.

Other descriptions of this dataset can be found in Brooks et al. (2011) and Brooks and Phillips (2010).

To examine year-to-year growth rates in spending we need consecutive years of fiscal data, so we forgo the use of 1967 data.

If an election takes place in a city-year, but we do not directly observe this election, it will be falsely recorded as a non-election year. If we then compare outcomes for election years and non-election years, the difference in means between these two groups would be biased downwards.

Elections were held in 27.5 % of city-year observations in 2003–2004, but only 18.5 % of city-year observations in 2005–2006.

Elections were held in 24.8 % of city-year observations with population less than 50,000, and 29.3 % of city-year observations with population greater than or equal to 50,000. Hoover, AL; Miramar, FL; Mount Vernon, IN; Wells, NV; and Newark, OH, are all examples of cities with population less than 50,000 reporting only one election in years 1970–2004. This could be due to underreporting of elections, as we found Hoover, AL, held elections every 4 years from 1980 through 2008 (Velasco 2000; U.S. States News 2008).

For instance, no elections are reported for Berkeley, CA, before 1998, but Shirley Dean was first elected Mayor of Berkeley in 1994. See http://​www.​berkeley.​edu/​news/​berkeleyan/​1998/​1209/​dean.​html. Similarly, no mayoral elections are reported for Las Vegas, NV, after 1999, but Oscar Goodman was reelected twice after first becoming mayor in 1999. See http://​www.​deseretnews.​com/​article/​700124524/​Vegas-Mayor-Goodman-backs-wife-in-city-election.​html.

When testing for manipulation of education spending in election years we further restrict our sample to cities with dependent school districts. Limiting our sample to cities with dependent school districts decreases our sample size from 6394 city-year observations to 1345 city-year observations. Additionally, as there were large year-to-year changes in per capita education expenditures in some cities, we delete potential education spending outliers by dropping city-year observations with a year-to-year change in per capita education expenditures greater than 100 %. This leaves us with data on education expenditures for 1218 city-year observations.

We do this because our analysis is intended to describe the typical mayor or typical city, rather than the expected effect for the typical city resident. As a robustness check, we also tested all specifications with observations weighted by population, such that the results represent the average person rather than the average city. All results were very similar in magnitude, direction, and statistical significance, suggesting that there is little heterogeneity of effects across city-year population sizes.

By average we mean log-average, or geometric mean.

The specification of Finkelstein (2009) was also notably similar, as she regressed the first difference in the log state minimum tolls on an indicator for whether state elections are held.

In addition, we analyzed several other categories of spending, employment, and revenues, but in the interests of space we here report only results with substantial statistical or economic significance. The rest of the results are available upon request.

The only exception is in the parks and recreation spending category. In Column 3 of Table 4, we observe a 1.9 % increase in parks and recreation spending with a p value of 0.11. In Column 3 of Table 13, when state fixed effects are included, the point estimate increases to 3.0 % with a p value of 0.09, marginally statistically significant at the 10 % level. Tables 14 and 15 provide similar results for disaggregated employment categories and revenue sources, respectively.

Relatedly, Morgan and Watson (1995) show that mayoral strength is not a significant predictor of municipal expenditures.