Sales Tax Competition among State–Local Governments: Evidence from the US Counties

ABSTRACT

This article estimates tax reaction function of local governments competing with other governments to provide the insight for tax policy decision makers. Employing a panel of county-level data in all the states from 1970 to 2006, this article analyzes sales tax competition between county governments. In addition to a static model, this article pays more attention to dynamics in tax reaction function estimated by general methods of moments containing lagged dependent variables in the reaction function. Focusing on county governments, this article finds evidence for strategic interaction, an interaction that has positive effects on the setting of tax rates in a county.

KEYWORDS:

• Local sales tax
• tax competition
• tax reaction function

Notes

^1. The theoretical works are Haufler (1996), Kanbur and Keen (1993), Lockwood (1993), Mintz and Tulkens (1986), Nielsen (2001, 2002), and Ohsawa (1999, 2003). The empirical studies are Besley and Rosen (1998); Devereux et al. (2007); Egger, Pfaffermayr, and Winner (2005a, 2005b); Jacobs et al. (2010); and Rork (2003).

^2. Since 2007, the US Census Bureau has not surveyed the finance data of each county government but reported the entire local governments. Thus, it is impossible to look into the recent tax reaction function. However, the long period of 1970–2006 in the data enables this study to empirically analyze sales tax reaction function between state and local governments.

^3. Although property tax is still the primary source in local revenue, it does not appear to affect changes in shoppers’ paying behavior and retailers’ locations (Fox, 1986).

^4. The average rates of state sales tax, local sales taxes, and combined sales tax except for Washington, DC are 5.62%, 1.79%, and 6.90% as of 2011, respectively. The share of local revenues from sales taxes is 52.2% in Louisiana, 47.7% in Arkansas, and 40.0% in Oklahoma.

^5. The data on county governments show that the means of local sales tax rate were 0.112% in 1970, 0.248% in 1980, 0.553% in 1990, 0.988% in 2000, and 1.108% in 2006.

^6. For a glimpse of this extensive literature, see Baicker (2005), Hou and Seligman (2010), Jung (2001), Luna, Bruce, and Hawkins (2007), Rork and Wagner (2008), and Zhao and Hou (2008). However, the recent studies have covered local sales tax in multistates (Afonso, 2015; Agrawal, 2014, 2015; Burge & Piper, 2012; Burge & Rogers, 2011).

^7. Use tax is therefore implemented to control the mobility. However, it is impossible for a government to effectively administer use tax (Jacobs et al., 2010), so a local government hardly imposes a use tax.

^8. In addition, local sales tax has been regarded as a dynamic policy tool that intensifies local capacity and diversifies local revenue structure effectively (Carroll, 2009; Drenkard, 2011; Sjoquist & Stoycheva, 2012). Furthermore, some studies analyzed the effects of local sales tax on budgets and reported that the increase in local sales tax rate expands local expenditure (Sjoquist, Walker, & Wallace, 2005), reduces property tax burdens (Afonso, 2014; Jung, 2001), and diversifies revenue source (Sjoquist & Stoycheva, 2012).

^9. Marginal costs for public service will decrease when nonresidents cross a border and pay more of the costs (Luna, 2004). Research had been dependent on the game of the Nash equilibrium in order to examine how much governments provide public service has challenged setting tax rate. The Leviathan behavior has pointed out that tax competition results in the excessive tax burdens by the higher tax rates in the determination of the Nash equilibrium (Apolte, 2001; Keen & Kotsogiannis, 2003; Wrede, 2000).

10. The extant studies have investigated tax competition in the two ways using statutory tax rate and average effective tax rate. Most extant studies on tax competition used statutory tax rates (Agrawal, 2012; Devereux et al., 2007; Egger et al., 2005a; Luna, 2004; Luna et al., 2007; Rork, 2003; Rork & Wagner, 2008). Some studies used average effective tax rates instead of statutory rates (Egger et al., 2005a; Jacobs et al., 2010). However, the studies using average effective tax rates focused on state level and statutory rates were used in the aforementioned studies in local level.

11. Elected representatives in each jurisdiction decide how to collect the revenue and to deliver public services with careful considerations of their voters’ preferences; this affects their probability of being reelected. Sales in the higher taxing governments reduce, and the higher taxing governments lose sales tax revenue to its neighboring lower taxing governments. The higher therefore would spend sales tax revenue on a smaller tax base, which raises a trade-off to representatives (Luna, 2004).

^12. For better reflections of local politics, this article includes all the election results for President, Governor, and the US Congress Members of Senators and House of Representatives because the election year varies across state–local governments. If a county does not have an election in a year , the county has the previous year’s value (Adams, Clark, Ezrow, & Glasgow, 2004; Levitt, 1996).

^13. If the income level becomes lower, the unemployment rate becomes higher in an economic downturn, which leads a government less inclined to raise tax rates, but more likely to expand its outlays on social security (Berry & Berry, 1992); the reverse would be true in an economic boom year.

^14. Business cycle through unemployment rate influences the behavior that a county government sets its tax rate.

^15. Where people live, work, and purchase goods might differ from each other; however, the amount of employment proxies for the unemployment level in a county. After reviewing the data, the number is greater than the total population in some county governments. Hence, this article uses the number as a proxy of unemployment level in a county, and the number is converted to the natural logarithm values.

^16. They determine the fiscal capacity in a county, influence the demands and types of public service, and the boundary of public service delivery.

^17. Compared to a sales tax, a property tax is inelastic but stable over economic cycle, and grants are uncertain for local governments due to the limited fiscal capacity in the upper levels. Local governments make much effort to diversify their revenue sources, and a local sales tax is an alternative revenue source under the statutory restrictions.

^18. Because any change in fiscal policy affects local fiscal conditions in the current time and its future periods, fiscal conditions are equivalently a transfer function of the current and past changes until other changes are made in the jurisdiction and its neighboring jurisdiction.

^19. According to Moran I test, the spatial lag term correlates with error terms, so this study corrects spatial error dependence following Kapoor, Kelejian, and Prucha (2007).

^20. According to Esteller-Moré and Solé-Ollé (2001), intra-jurisdictional competition raises externalities because of the co-occupancy of the same base of taxation, and this competition between state and local governments affects each other. Although this article employs a panel data set of county governments in all the US states, the primary research targets are the counties in the 10 states that are not threatened by intra-jurisdictional competition.

^21. They are Mississippi, Nebraska, South Dakota, and Vermont.

^22. According to state statutes, the 12 states include Alaska, Florida, Georgia, Hawaii, Idaho, Nevada, North Carolina, Ohio, Pennsylvania, South Carolina, Virginia, Wisconsin, and Wyoming. The time period in this article is 1970–2006, and it is hard to capture the dynamics in Alaska because municipalities were also allowed to levy sales taxes before 2005; Hawaii has allowed local sales taxes for local governments since 2007. Cities in Virginia are independent jurisdictions from counties and the city sales tax is fixed at the uniform 1% rate; tax competition is not observed because Virginia administers local sales taxes. However, the authorized local sales taxes in Georgia, Idaho, and Pennsylvania are considered to be local (option) sales taxes that some specific municipalities and school/special districts can levy; vertical tax competition of local sales taxes is not observed in those states. Hence, this article focuses on 10 states not threatened by the endogeneity of vertical competition (intra-jurisdictional competition) by excluding Virginia and Alaska but including the three states of Georgia, Idaho, and Pennsylvania.

^23. The 10 states in the final sample include Florida, Georgia, Idaho, Nevada, North Carolina, Ohio, Pennsylvania, South Carolina, Wisconsin, and Wyoming.

^24. However, this study focuses on the 10 states excluding the states that have not authorized local sales tax for counties and vertical tax competition challenges those states.

^25. The description shows that the ratio of revenue to expenditure in Carbon, WY, was 3079.76 in 1991. Although Wyoming has unique revenue structures that heavily depend on oil and gas revenues, this study treats that observation as an extreme outlier and excludes Carbon from the sample.

^26. For the year 2000s, the average tax rate includes the period of 2001–2006.

^27. Some county governments are not included because they do not have their historical date of local sales tax rates (i.e., most counties in Louisiana lost the data after Katrina and the counties in Missouri are not available before 1990).

^28. The Hausman test reports that the analyses with fixed effects are more appropriate. However, this study provides both results with random and fixed effects.

^29. An instrument in the GMM estimator must be independent of unobserved heterogeneity and any error terms. The instrument can be validated by an overidentifying restrictions test when the number of included endogenous variables is less than the number of instruments. The Sargan overidentification test fails to reject the null hypothesis of no overidentified instruments under the conventional significance level of 0.05. It confirms the validity of the instruments in this study (Bhargava, 1991). Furthermore, the GMM estimation should pass the Arellano–Bond test for autocorrelation. The error terms in this study are AR(1) but do not follow AR(2), which verifies that the error terms in GMM estimation have no autocorrelation in their idiosyncratic errors before differenced (Roodman, 2009). Last, the GMM estimation should test the exogeneity of the proposed instruments and fail to reject the null hypothesis. Based on those test results for the validity of the instruments, this study concludes that the instrument is valid for the GMM estimation as shown in Table 4.

^30. The extant empirical studies of state sales tax have shown that the estimators of tax reaction function are −0.24 and −0.16 weighted by contiguity and population, respectively, in state sales tax (Rork, 2003), 0.16 in local sales tax in Tennessee (Luna, 2004), and 0.96 in the average effective rate of state sales tax (Jacobs et al., 2010).

^31. Since the main concern of this study is dynamics in tax reaction function, the static model does not include the lagged dependent variable of sales tax rate following Jacobs et al. (2010).