Regional Science and Urban Economics

Transcription

1 Regional Science and Urban Economics 46 (2014) Contents lists available at ScienceDirect Regional Science and Urban Economics journal homepage: The millage rate offset and property tax revenue stability Keith R. Ihlanfeldt, Kevin Willardsen DeVoe Moore Center and Department of Economics, 150 Bellamy Building, Florida State University, Tallahassee, FL 32306, USA article info abstract Article history: Received 17 June 2013 Received in revised form 5 April 2014 Accepted 7 April 2014 Available online 24 April 2014 JEL classification: H7 H2 One of the alleged virtues of the property tax is that it produces stable revenues regardless of price movements in real estate markets. One explanation for this is that local governments adjust their millage rates to offset changes in their property tax base. Little evidence, however, exists on the strength of this millage rate offset mechanism. We hypothesize that the importance of this mechanism will vary among local governments depending upon the monopoly power that they possess. The results provide strong support for our hypothesis Elsevier B.V. All rights reserved. Keywords: Property tax revenues Millage rates Expenditures 1. Introduction One of the alleged virtues of the property tax is that it produces relatively stable revenues regardless of movements in real estate markets (Alm, 2013). In a recent special issue of this journal titled The Effect of the Housing Crisis on State and Local Governments this stability was documented, as three different papers found that the housing market crash has had little impact on property tax revenues (Alm et al., 2011; Doerner and Ihlanfeldt, 2011; Lutz et al., 2011). 1 In these papers and others that preceded them (e.g., Lutz, 2008), a key factor allegedly accounting for revenue stability is a change in the property tax rate in theoppositedirectionofthechangeinthepropertytaxbase.thismechanism is sometimes referred to as the millagerateoffset, although this makes it sound more like an engineering than an economics concept. 2 Because these studies provide little, if any, evidence that millage rate changes do, in fact, account for revenue stability, they bring to the forefront the long-standing issue of how local governments set their property tax rates. There are widely differing views on how property tax rates are set by local governments. One view, the residual view, maintains the rate is set to balance the budget once the tax base is known, other revenues are projected, and the level of expenditures is chosen. 3 In other words, the tax rate is mechanically determined and is not a choice variable. An implication of this view is that the elasticity of the tax rate with respect to the tax base is equal to one; i.e., as the tax base falls (rises) the tax rate rises (falls) so as to maintain property tax revenues. An alternative view, the strategic view, maintains that the tax rate is chosen with an eye toward maintaining and attracting mobile capital. 4 Under this view, each local government faces an investment demand curve, where the tax price of investment is the tax rate. An implication of this view is that the elasticity of the tax rate with respect to the tax base is, a priori, unknown and will likely vary across local governments depending upon the elasticity of their investment demand curve. For example, in response to a loss in tax base, a government with little monopoly power may decide to let its tax revenues decline instead of raising its tax rate and risk losing business investment. There is little conclusive evidence on which of the above two views is more correct. We can think of three approaches that might yield an answer. First, because state statutes tend to provide detailed instructions on how the budgetary process must be conducted at the local Corresponding author. Tel.: addresses: kihlanfe@fsu.edu (K.R. Ihlanfeldt), kevinwillardsen@gmail.com (K. Willardsen). 1 These papers were all presented May, 2010, at the Urban Institute's conference What the Housing Crisis Means for State and Local Governments. 2 An anonymous referee was the first to make this point. 3 Ross and Yan (2013) trace the residual view back to Netzer (1964) and provide this quote from him: the nominal rate of the tax is essentially a residual, derived by determining the level of expenditures, subtracting state and other non-property tax revenues from budget outlays, and comparing the remainder with assessed values [A]ssessed values are no more actual or determinants of property tax yield than are a number of other factors. 4 Brueckner (2003) reviews theory and evidence related to the property tax rate as a strategic tool that local governments use in competing against one another for mobile capital / 2014 Elsevier B.V. All rights reserved.

2 168 K.R. Ihlanfeldt, K. Willardsen / Regional Science and Urban Economics 46 (2014) level; a reading of these statutes might shed some light on how tax rates are determined. Second, a survey of local officials could be conducted on how rates are set within their jurisdiction. Finally, because the two views have clear testable implications regarding the magnitude of the elasticity of the tax rate with respect to the tax base, this parameter could be estimated given the appropriate data. In sampling the statutes of various states we discovered that it is not uncommon to find prima facie support for the residual view. For example, Illinois statutes (50 ILCS 330/3 Ch.85, par.803) state that the budget must be adopted before the tax rate. However, the statutes of other states, like Florida, specify that the tax rate be adopted first (s , F.S.). This suggests that whether the residual or strategic view of the tax rate is correct varies across states. This diversity receives emphasis in Huddleston's (2005) guide to the budgetary process: The property tax is often considered to be a residual tax, that is, property tax collections are determined by the amount of revenue needed to balance the local budget after all other revenues have been taken into account. While this notion of the property tax is generally true, it is also commonly true that total property tax collections or the property tax rate are determined first, and then local spending is adjusted as needed to produce a balanced budget (page 29). While we did not survey local officials regarding their budgetary process, we did consult with both academics and practitioners knowledgeable about local government budgeting. They emphasized that a distinction must be drawn between procedures and processes. While a state statute might require either the tax rate or budget be adopted first, this is a procedural requirement that may have little bearing on the actual determination of budgets and rates. There was also an unwillingness to accept the residual view, principally because the tax rate has political repercussions. As Huddleston concludes, Which ever approach to property taxes is taken, it is universally true that the annual bane of many elected local officialsisannouncement of the property tax rate for the upcoming year. Property tax rates matter, both to elected officials and taxpayers and voters! (Page 29.) 5 While the residual view can be criticized for its naivety concerning political economy, this does not mean that the strategic view by default is correct. If anything, it suggests a third view of rate determination, where the rate is set with an eye toward avoiding voter wrath and keeping incumbents in office. Nevertheless, the political criticism of the residual view dispels the notion that the tax rate is the outcome of an arithmetic calculation in favor of it being a choice variable, which may be affected by both political and capital flight concerns. In this paper we take the third approach toward resolving the issue of whether the residual view or the strategic view of the property tax rate is more correct. Using unique panel databases for cities and counties in Florida covering the years 1995 to 2011, we estimate elasticities of the millage rate and expenditures with respect to the property tax base. We seek to answer the following questions: What is the elasticity of the millage rate with respect to the property tax base? Is this elasticity the same for upward and downward movements in the property tax base? Does the elasticity vary among different types of governments with varying amounts of monopoly power? Is there a correspondence between the millage rate and expenditure elasticities with respect to the property tax base? Answering these questions is timely in light of the fact that since the housing market crashed and the Great Recession hit (circa 2007); many local governments have experienced unprecedented declines in their property tax base. In the next section Section (2) we review the literature. Section 3 provides background information on taxing and budgeting at the local 5 Mikesell (2011, p. 494) also holds this view: A government may, of course, see the computed rate, worry about the consequences, and revise the amount of levy it chooses to raise. level in Florida. In Section 4 we develop the hypotheses. The data and estimating equations are described in Sections 5 and 6, respectively. Descriptive statistics are presented in Section 7. The results from estimating the millage rate and expenditure models are presented in Sections 8 and 9. Additional results for principal cities appear in Section 10. Conclusions are stated in Section Literature review The relevant literature consists of three types of studies. First, there are tax competition studies that view the tax rate as a strategic policy tool in the competition among local governments for mobile capital. Second, there are studies that have tested the residual view of the property tax rate. Third, there are studies that have estimated the elasticity of the millage rate with respect to the property tax base. The tax competition literature is reviewed by Brueckner (2003). He first reviews the theory of tax competition by developing his own jurisdictional objective function. Maximization requires that the jurisdiction take into account the flight of capital caused by an increase in its tax rate, which moderates the incentive to raise the rate. He then reviews eight studies, all of which provide at least some support for the prediction that tax rates are set with an eye toward possible capital flight. 6 Studies that empirically test the residual view of the property tax rate are reviewed by Ross and Yan (2013). After reviewing a half dozen or so studies, they conclude that the literature is mixed on whether the tax rate is a residual in the budgeting of local governments. They maintain that the results of all previous studies suffer from omitted variable bias. The omitted variable is housing wealth, which is correlated with the tax base, because a substantial portion of the properties that make up the base are owner-occupied residential units. As the base expands, the residual tax rate change will be biased if greater housing wealth increases the demand for public services, causing a rise in the tax rate. To overcome this bias, they use annual data on Indiana counties from 2000 to 2008 to regress the change in the tax levy (not the tax rate) on the tax base, controlling for changes in housing values. The latter variable is included to capture the wealth effect, but it is crudely measured as the change in the mean selling price computed from a 3 to 4% sample of homes that sold in each year. In addition to their criticism of prior studies for omitting housing wealth, the authors maintain that these studies ignore the endogeneity of the property tax base: It is also conceivable in some instances that property assessments are conducted in some endogenous manner, particularly if there is no cyclical basis for reappraisals (p. 13). 7 Because the state puts each county on a fixed reassessment schedule that varies across counties in 1984, the statemandated reassessments are taken as exogenous to the levy and included in their model. The results show a statistically significant wealth effect, but it is described as a qualitatively small effect (p. 22). The effect of the tax base on the levy is statistically significant but small in magnitude. The authors conclude, At best, it seems there is some amount of fiscal illusion from assessed value growth that is statistically significant, but it is substantively small and thus generally consistent with the revenue neutral adjustments in the property tax rate predicted by the residual view (p. 24). A limitation of Ross and Yan is that they only investigated 6 In line with this evidence is other evidence showing that business investment is highly responsive to tax differentials across jurisdictions, especially those located within the same metropolitan area. Bartik (1991) concludes from his review of the literature that the elasticity of business investment with respect to the property tax rate lies between 0.10 and 0.60 inter-regionally and 1.0 and 3.0 intra-regionally. 7 In Florida there is no reason to suspect that property assessments are conducted in some endogenous manner. The state mandates that property values be updated every year on January 1, each county uses a Computer Assisted Mass Appraisal vendor to obtain these values, and theflorida Department of Revenue uses various checks to make sure that assessments are in line with market values.

3 K.R. Ihlanfeldt, K. Willardsen / Regional Science and Urban Economics 46 (2014) whether the levy is affected by upward movements of the property tax base. Finally, two papers have preceded us in providing estimates of the elasticity of the property tax rate with respect to the property tax base. Using data on Georgia school districts for the years 1997 to 2009, Alm et al. (2011) regress the percentage change in the millage rate on the percentage change in the per capita property tax base, the percentage change in population, the percentage change in per capita income and a set of year dummy variables. Their results suggest a millage rate offset of about 30%. In a preliminary exploration of our panels, which at the time ended in 2009, Ihlanfeldt (2011) investigated the average city's and county's immediate, first year tax rate response to a change in its property tax base. The short-run elasticities estimated are.097 and.212 for cities and counties, respectively. The present study goes well beyond earlier studies in exploring the relationship between changes in the tax base and changes in the tax rate. We estimate the elasticity of the millage rate with respect to the tax base for both upward and downward movements in the property tax base. We also allow governments to respond to a change in their tax base over a number of years. Finally, we do not assume that the millage rate elasticity is uniform across local governments but instead posit and test that it varies by type of government, where the typology is designed to capture differences in monopoly power. 3. The Florida background To properly interpret the results obtained from the estimation of our millage rate and expenditure models, in this section we provide some background information on Florida's property tax system and the budget process at the local level The property tax in Florida In Florida the property tax is the most important revenue source of counties, accounting on average for 31% of total revenues in Other revenue sources with sizable shares of total revenues are service charges (21%) and general government taxes (10%). 8 The property tax is also an important source of revenue for cities, accounting on average for 20% of their total revenue in However, it is not the revenue source providing the largest share of total revenue. Service charges hold this distinction, generating 30% of total revenue in Consistent with counties, the third most important revenue source of cities is general government taxes (12%). Concerning the administration of the tax, Florida's Constitution requires property appraisers to establish the property tax base for their county annually. In doing so, they determine the just, or market, value of each parcel of property as of January 1 of each year. Market values are predicted in all of Florida's 67 counties using Computer Assisted Mass Appraisal systems (CAMA). These CAMAs use all three appraisal methods (comparable sales, replacement value, and the income approach) in valuing properties. Upon obtaining market values, the appraiser applies all valid exemptions and assessment limitations to determine each property's taxable value. Florida has exemptions for widows and widowers, the disabled, the blind, persons 65 and older, and veterans. However, these are all small relative to its homestead exemption. Until 2008, this exemption reduced assessed value by $25,000 for taxpayers residing permanently in their homes. In 2008, the State Constitution was amended by referendum to increase the homestead exemption to $50,000. Florida has an assessment limitation called Save Our Homes. SOH limits the annual increase in the assessed value of homesteaded homes to 3% or the percent change in the CPI, whichever is less. The accumulated difference 8 General government taxes include a wide variety of non-ad valorem taxes: local business taxes and franchise fees, communication service taxes, local option fund taxes, and utility service taxes. between the homeowner's assessed value and the market value is the SOH benefit. Before 2008 the SOH benefit was lost when the homeowner moved, creating a lock-in effect (Ihlanfeldt, 2011). With the passage of the amendment that also raised the homestead exemption in 2008, the SOH benefit became portable, meaning that it could be transferred by homestead owners from their old homestead to a new homestead. Property tax rates (a.k.a. millage rates) are set every year by local governments. These rates are subject to a, ten thousandths or 10 mill cap, but the cap can be exceeded by a local referendum. 9 Millage rates for each jurisdiction are uniform across all property types Local government budgeting Local government budgeting in Florida tends to follow the same steps as Huddleston (2005) outlines for cities and counties generally in the U.S. During the first quarter of each year Florida jurisdictions develop targets or budget guidelines based on a financial forecast. These are communicated to the departments and agencies requesting funding. These requests are submitted in March or April to the budget manager. Over the next three months work sessions are held with the elected representatives of the jurisdiction (i.e., the city council or county commission). The outcome of these work sessions is a tentative budget, including a millage rate policy, proposed program enhancements or reductions, and revenue adjustments. On July 1, the county property appraiser informs jurisdictions of their property tax base. By the end of July the tentative budget has been modified by the jurisdictional administrator in order to be in accordance with the jurisdiction's tax base and strategic plan. The modified budget is the proposed budget that is publicly announced in August. During the months of August and September two public hearings are held to give citizens the opportunity to provide input on the proposed budget to the county commission (city council). At the second hearing, a millage rate and budget are approved by the commission (council). Our description of the budget process suggests that neither the budget nor the millage rate is a residual in the budget process. The implication is that they are jointly determined through a series of budget negotiations. 4. Hypotheses Consider first how a local government might respond to a loss in its property tax base. To balance its budget it has four options: it can raise the millage rate, cut current expenditures, cut capital expenditures and draw down reserves. 10 These options are not mutually exclusive and governments might adopt a combination of responses to a loss in tax base. However, between the options of raising the millage rate and cutting current expenditures there are reasons to expect that local governments will favor the former over the latter option. First, taxpayers may be more concerned with the size of their payment than the rate they pay. 11 Payments are the product of the millage rate and the assessed value of the home. If the millage rate is raised to offset lower assessed values (i.e., a lower tax base), the payments remain the same and voter wrath is avoided. Second, as noted by Wolman (1983), cuts in current expenditures will bring immediate negative feedback from the special interest groups affected by the expenditure cuts. Hence, the 9 Only a small percentage of cities and counties are at the cap in our panels. The results obtained from estimating our millage rate models are insensitive to the inclusion or exclusion of city (county)/year observations at the cap; hence, we included all observations in our models. 10 It could also increase its level of indebtedness by borrowing more but this option is seldom used by local governments in Florida. Borrowing is reserved for the financing of capital projects. 11 Many taxpayers may not even be aware of the rate that they pay because this requires that they read and understand the Truth-in-Millage Notice that they receive inthe year before their taxes are owed.

4 170 K.R. Ihlanfeldt, K. Willardsen / Regional Science and Urban Economics 46 (2014) self-interested local government official will prefer an increase in the millage rate over a reduction in current expenditures to avoid being voted out of office at the next election. Because of the political repercussions resulting from a cut in current expenditures, if expenditures are to be cut, we expect that the cuts will be in capital expenditures. These expenditures can be postponed in the short-run without having much of an effect on the delivery of public services. Hence, in addition to relying on an increase in its millage rate, a local government will choose to reduce its capital expenditures in response to a loss in tax base. However, a limitation on both of these options, especially an increase in the millage rate, arises from local officials being economic development minded. They realize that a higher tax rate adversely affects their ability to retain and attract business investments. As noted above, empirical evidence supports this realization for it shows that business investment is responsive to property tax differentials across jurisdictions, especially those located within the same metropolitan area. The extent to which a local government can raise its millage rate without having a dire effect on its level of business activity depends on the monopoly power it wields. The monopoly power possessed by a local government is high (low) if the government faces an inelastic (elastic) demand curve of business investment, where the tax price of investing in the community is the rate of property taxation. A key determinant of the elasticity of demand is the extent to which businesses consider alternative locations outside the jurisdiction as substitutes. If outside locations are considered poor substitutes, the local government will face an inelastic demand curve for business investment and will be able to raise its millage rate without fear of capital flight. To investigate the effect of a local government's monopoly power with respect to the decision to raise its millage rates, we divided local governments into four groups with varying amounts of monopoly power: principal cities, other cities within the metropolitan area, rural cities, and counties. 12 Local governments in the second group have the least amount of monopoly power. On average, there are more than a dozen of these cities within each of Florida's metropolitan areas. Competition is keen among them with each one being a close substitute for the others as a location for business investment. 13 While the other three groups all have unique features that enable them to hold and attract business investment, principal cities may have the greatest amount of monopoly power. The monopoly power of these cities stems from the unique agglomeration economies they provide that cannot be found within suburban areas. 14 Rosenthal and Strange's (2006) review of the literature on agglomeration economies highlights their strong locational pull on business activity. Also, as discussed below, the relatively large size of principal cities contributes to their monopoly power. Rural cities in Florida are distinguished in two respects. First, a majority are county seats and, regardless of this distinction, most possess natural resources that are inputs into the state's large tourism industry. Business activity that benefits from proximity to a county's government offices or that is tied to a natural resource is less likely to relocate in response to a higher millage rate. A final factor affecting the degree of monopoly power possessed by a local government is its physical size. Consider the average firm located within the boundaries of a local government. Assume that its choice set is limited to an area within a certain radius of its current location (perhaps, to retain its current workforce or customer base or to benefit 12 Principal cities are defined by the U.S. Census Bureau as the city within a metropolitan area with the largest population. 13 According to Bartik (1991), it is the high substitutability among jurisdictions within the same metropolitan area that explains the high intra-regional elasticities of business investment with respect to the property tax rate that have been reported in the literature. 14 Ihlanfeldt (1995) reviews the evidence that documents the uniqueness of a central city's agglomeration economies. Central cities preceded their renaming as principal cities by the Census Bureau. from agglomeration economies). As the size of its government expands, the firm has fewer possible locations within its choice set that lie outside the boundaries of its current jurisdiction that could serve as a substitute location. In addition, if we assume that the cost of relocation rises with distance, the moving costs of the firm will increase with the size of its local government. Both of these factors fewer substitutes and greater moving costs suggest that larger places possess greater monopoly power than smaller places. Among our four groups, counties are the largest in size, followed by principal cities. This distinction provides both with monopoly power. Based upon our monopoly power arguments, we hypothesize that the millage rate offset to a loss in tax base will be greatest for principal cities and least for cities within metropolitan areas that are not principal cities. The magnitudes of the offsets chosen by rural cities and counties will lie in between. While our millage rate offset predictions are based on differences in the degree of monopoly power across jurisdictions, there are other factors that could give rise to these same predictions. For example, as noted above, local governments may rely more on an increase in millage rates in response to a loss in tax base if cutting expenditures is an unattractive option. The latter may be true for a variety of reasons, including the desires of elected officials to maximize spending through the agenda process (Romer and Rosenthal, 1978), the number or strength of local interest groups (Mueller and Murrell, 1986) or the perceived need to maintain expenditures to meet the needs of a dependent population. These factors could play out in a variety of ways, leading to the same predictions as based on our monopoly power arguments. But one example, the millage rate offset may be greatest for principal cities, because they have the largest dependent populations. As a second example, the effect of interest groups on maintaining expenditures may be felt more strongly in more populated places, such as principal cities and counties. This might lead to the prediction that, in comparison to suburban cities, these places will favor an increase in the millage rate over a cut in expenditures. Thus far we have advanced predictions regarding how local governments adjust their millage rates in response to a loss in tax base. Predicting how they will respond to an increase in tax base is less clear. One possibility is that they may act in a symmetric fashion, with the elasticity of the millage rate with respect to a loss in tax base equaling the elasticity of the millage rate to a gain in tax base. This is plausible because; when the tax base rises after a fall, the government may wish to return to the budget it had before the loss in tax base. Moreover, if the goal is to stabilize property tax revenues, symmetry in behavior is expected. An asymmetric response, however, is suggested based on a government's interest in economic development. For example, consider principal cities, which we have hypothesized face an inelastic demand curve for business investment. While an increase in the millage rate may not induce capital flight, neither will a decrease in the millage rate attract much investment. Hence, these governments may choose not to reduce their millage rate as their tax base expands, and instead use the additional revenues to add to their reserve fund or to increase their expenditures. 5. Data This study relies on two primary sources of data: the Florida Department of Financial Services (FDFS) City and County Annual Financial Reports (AFR) and the Florida Department of Revenue (FDOR) Annual County Property Tax Rolls. The FDFS data come from budgetary data that each city and county is required to submit after each fiscal year (s , F.S.). In these audited reports, local governments are required by the FDFS to report total current and capital expenditures. The FDFS data cover the years The FDOR data set comprises county tax rolls from each of Florida's 67 county property appraisers. Annual preparation is required by statute (s , F.S.) and supervised by the FDOR (s , F.S.). Tax

5 K.R. Ihlanfeldt, K. Willardsen / Regional Science and Urban Economics 46 (2014) rolls are collected for the purpose of monitoring the performance of county tax assessors. Spanning the same years as the FDFS data (1995 to 2011), these rolls align with the FDFS data by year and across jurisdictions. Obtained from the rolls is the millage rate for each panel year for each county and city. Within each jurisdiction the millage rate is the same for all property, regardless of its use. Also obtained is the taxable value of the jurisdiction's property tax base. The taxable value of a property is the assessed value minus all exemptions; that is, it is the value that is multiplied by the millage rate to determine the property tax bill. Our principal cities are those defined by the U.S. Census Bureau. Florida has 20 metro areas and therefore 20 principal cities. Cities that fall into our second group incorporated places outside principal cities but within metropolitan areas with at least 1000 people number 264 in Florida in the year Our third group of cities is incorporated places located outside metropolitan areas with at least 1000 people. In 2011 there were 55 rural cities in Florida. Counties form our last group of local governments. There are 67 counties in Florida. Some city/year observations are lost because for some cities an AFR is not available for all years. This reduced the number of city/year observations for principal cities from 280 to 268, for other metropolitan cities 15, 16 from 3696 to 3437, and for rural cities from 728 to 654. In addition to the unbalanced panels used to estimate the millage rate and expenditure models, we formed balanced panels where a complete set of data are available for an individual jurisdiction for four selected years: 1995 (the first year of the panel), 2000 (a middle panel year), 2007 (the year of the housing market crash), and 2011 (the last year of the panel). The balanced panels are used below to follow the same set of jurisdictions over time to produce mean statistics describing their budgets. 6. Estimated models To estimate the elasticity of the millage rate with respect to the property tax base, first-differenced models are estimated. These models have the advantage of controlling for time-invariant unobservable heterogeneity across jurisdictions. 17 The specification of the model also allows the tax base to have asymmetric up and down effects on the millage rate 18 : ΔY i;t ¼ γ t þ X 2 ΔB h¼0 i;t hβ t h þ X 2 ΔP h¼0 i;t hδ t h þ X 2 þ X 2 h¼0 UB t hω t h þ X 2 h¼0 UB t h ΔB i;t h η t h þ ε i;t ; h¼0 ΔI i;t hθ t h where Y i,t is the natural log of the millage rate for jurisdiction i in year t and γ t is year fixed effects. B is the log of the property tax base and is 15 AFRs are missing for some years due to four factors: 1) some cities incorporated after 1995, 2) other citiesdisbanded their charter or were annexed after 1995, 3) some cities did not submit an AFR and are considered non-compliant, and 4) in some cases a city's AFR could not be reconciled with its audited financial statement and only those reconciled appear in the data that the FDFS makes public. 16 Florida has some very small cities. Some examples in 2011 include Bay Lake (pop. = 6), Weeki Wachee (pop. = 12), and Marineland (pop. = 16). There are numerous others of somewhat larger size. These small cities are likely to budget differently than more representative cities and therefore any city/year observation with a population of less than 1000 people is excluded. This reduced the number of observations in the city panel by about 8%. 17 Instead of first-differencing the data, an anonymous referee suggested the data be left in levels and city (county) fixed effects be added to control for time-invariant unobserved heterogeneity across places. We choose not to follow this approach for two reasons. First, after first-differencing we could not reject the null hypothesis of zero serial correlation in the idiosyncratic errors. Wooldridge (2000, p. 447) shows that unless the errors display substantial negative correlation, first-differencing is more efficient than using level data with jurisdictional fixed effects. Second, accounting for lagged adjustment in a levels model is cumbersome in comparison to differencing the data and including lagged changes as independent variables. 18 For each of the four groups of local governments between 34% and 52% of the movements in the tax base over the course of the 17 year panel are down. Between 32% and 68% of the down movements occurred after the housing market crashed. ð1þ expressed in nominal, non-per capita terms. P and I are the logs of population and per capita income, respectively. These variables are included as control variables because they vary over time and register changes in the overall demand for public services. 19,20 U B equals 1 if the change in the tax base is positive and zero otherwise. The estimate β gives the effect of a decline in the tax base on Y, while the estimate η gives the difference between up and down effects of a change in the tax base. 21 Note that for each explanatory variable, the current value and two lagged values are included. 22 Lags in the tax base are included because local governments may adjust their millage rate incrementally over a number of years in response to a change in their tax base. 23 Lagged values of the demand variables are included because it takes varying amounts of time for changes in these variables to affect public services. For example, it may take more than one year for jurisdictions to raise their public services to meet an increase in articulated demand resulting from an increase in income. Delays may also result from the time needed by citizens to articulate changes in their demands. For example, this may occur as the result of an increase in income occurring in a non-election year. The summation of the estimated coefficients on the current and lagged variables gives the Long Run Propensity (LRP). The LRP represents the change in the dependent variable that would occur within the jurisdiction if there were a permanent change in the independent variable; i.e., it registers the full adjustment in the dependent variable to a change in the independent variable and can be interpreted as the long run elasticity. As such, it is the elasticity that is the focus of our analysis. The dependent variables in the estimated expenditure models are the natural logs of current and capital expenditures. The independent variables are identical to those in Eq. (1), except that the first of the three property tax variables is measured with a one-year lag, and as a result one additional year of observations is lost in comparison to the millage rate models. The first tax base measurement is for t 1because 19 Annual city and county populations are estimates provided by the Bureau of Economic and Business Research (BEBR) at the University of Florida. They are Florida's official estimates and are used to allocate state revenue sharing funds to counties and cities. BEBR constructs its population estimates using the housing unit method, in which estimates of population change are derived from estimates of changes in occupied housing units. A detailed description of BEBR's methodology can be found at default/files/spr_8_0,pdf. County per capita income is obtained from the U.S. Department of Commerce, Bureau of Economic Analysis, Regional Economic Information System: Because annual per capita income is unavailable for cities, they are assigned the income of their county. Because the data are first-differenced and city and county income are expected to be highly correlated, the use of county income as a proxy for city income is unlikely to have much of an influence on our results. 20 Following Ross and Yan (2013), in preliminary estimation we included a variable in both the millage rate and expenditure models to register changes in housing wealth within the jurisdiction. The variable, an improvement over the mean sales price used by Ross and Yan, is the change in a repeat sales housing price index that we computed for each county and city using sales price data (the two most recent sale prices on each property) from the FDOR tax rolls. In none of the equations estimated for the four groups of local governments is the wealth variable statistically significant with the theoretically correct sign. Moreover, its inclusion had almost no effect on the magnitudes of estimated elasticities. Hence, our results fail to support Ross and Yan's contention that excluding a housing wealth variable biases the estimated impact that changes in the tax base have on millage rates and expenditures. 21 An anonymous referee suggested that jurisdictions with a larger share of their tax base in commercial (as opposed to residential) property may choose to increase their millage rate by more in response to a loss in tax base, because commercial property lowers the tax price to the median voter. We question this reasoning because an increase in the millage rate is applied equally to commercial and residential property. Nevertheless, we estimated Eq. (1) including interactions between the property tax base variables and the percentage of the tax base accounted for by commercial property. The estimated coefficients on the interaction variables are all close to zero and statistically insignificant. 22 To determine whether the lagged values of the tax base should be included, we estimated Eq. (1) and two alternative models one with just one lag and the other with no lags. For each model we computed Akaike's Information Criterion (AIC). For both the millage rate and expenditure models estimated for all four groups of governments, the AIC is minimized with both lags, indicating that Eq. (1) is the preferred model. 23 Wolman (1983) argues that rates are raised incrementally over time to reduce their visibility and thereby produce less political backlash.

6 172 K.R. Ihlanfeldt, K. Willardsen / Regional Science and Urban Economics 46 (2014) Table 1 Means for selected years Principal cities (n = 18) Millage rate Property tax base 52,462 65, ,735 92,045 Current expenditures Capital expenditures Other metropolitan cities (n = 203) Millage rate Property tax base 67,810 73, , ,868 Current expenditures Capital expenditures Rural cities (n = 43) Millage rate Property tax base 25,226 29,254 58,260 39,827 Current expenditures Capital expenditures Counties (n = 67) Millage rate Property tax base 39,294 44, ,273 Current expenditures Capital expenditures expenditures are reported for the fiscal year, which is identified as the ending calendar year A preliminary look at the data Table 1 reports descriptive statistics for each of the four groups of local governments obtained from the balanced four year panels. Reported are the millage rate, property tax base, current expenditures, and capital expenditures. Except for the millage rate (expressed as taxes owed per $1000 of taxable value), all other variables are expressed in real per capita terms (2011 dollars). Millage rate changes are mixed across the groups from 1995 to 2000, but rates for all groups fell between 2000 and 2007 and increased between 2007 and These changes are consistent with the idea that local governments adjust their millage rates to stabilize property tax revenues because property tax bases increased across the board from 2000 to 2007 and then fell from 2007 to The upward and downward swings in tax bases have been dramatic. The housing market boom of the early 2000s raised real per capita tax bases within a range of 92 (principal cities) to 101 (counties) percent between 2000 and The crash in the housing market, along with the onset of the Great Recession, brought percentage declines in real per capita tax bases ranging from 27 (principal cities) to 38 (counties) between 2007 and While all local governments in Florida experienced up and down swings in their tax base over the years 2000 to 2011, the magnitudes of the upward and downward movements varied considerably from place to place with the swings being considerably larger in south than in north Florida. The factors accounting for this heterogeneity are explored by Ihlanfeldt and Mayock (2014). Their results suggest that the key explanatory factor is inter-area differences in the supply price elasticity of housing. As theory predicts, the magnitudes of the up and down swings in housing prices are found to be greater where supply is less elastic. They find that supply is less elastic where a local government's land use regulation is more stringent, which is true of jurisdictions in south Florida in comparison to those in north Florida. Of particular interest in the table are the contrasting trends between current and capital expenditures. Both current and capital expenditures experienced strong growth between 2000 and 2007 on average 33 and 41%, respectively. After the crash and Great Recession hit in 2007, capital expenditures plummeted on average by 41%, while current expenditures fell by only 1.0%. 24 For example, the 2000 fiscal year begins July 1, 1999, and ends June 30, In summary, the statistics reported in Table 1 for the four groups of local governments suggest that they all have responded to recent fiscal stresses in the same manner; namely, they have raised millage rates, severely cut capital expenditures, and maintained current expenditures. While the statistics presented in Table 1 describe how selected characteristics of local government budgets have changed over time, they do not provide ceteris paribus evidence on the impacts that changes in the property tax base have on millage rates and expenditures. To isolate these effects, we estimated the millage rate and expenditure models. Results are presented in the following two sections. 8. Results from estimating the millage rate models The results from estimating the millage rate models for the four groups of local governments are reported in Table 2. Reportedare an estimated elasticity for each of the three property tax base variables (t 1, t 2, and t 3) and the estimated LRP elasticity, which is the sum of the three tax base elasticities. 25,26 Because the LRP elasticity registers the long-run or full adjustment effect our focus will be on these estimates. Accompanying Table 2 is Table 3, which reports p-values for tests of differences in estimated LRPs between groups. The estimated LRP elasticities of the millage rate with respect to a loss in tax base (top panel of Table 2) are consistent with our expectations: The smallest and largest elasticities are for other metropolitan cities (.136) and principal cities (.670), while the elasticities for rural cities (.508) and counties (.393) fall in between. Note that the LRP elasticity estimated for other metropolitan cities is not only small, it is also only marginally significantly different from zero at the 10% level. The p-values reported in Table 3 indicate that the estimated LRP elasticity for other metropolitan cities is significantly different from each of the LRP elasticities estimated for the other three groups of government. Among these groups only the difference in the estimated LRP elasticities 25 The results obtained with the control variables (current and lagged values of population and income per capita) are not reported in Table 2 in the interest of keeping it to a reasonable length. For both the millage rate and expenditure models, the controls generally have the expected positive effects. 26 Also reported in Table 2 are estimated standard errors robust to heteroskedasticity. The standard errors could also have been made robust to serial correlation by clustering the errors at the jurisdictional level. However, results from conducting Wooldridge's test for autocorrelation in panel data (xtserial in Stata) indicated an absence of serial correlation in our first-differenced models. Drukker (2003) reports that the Wooldridge test has good size and power properties in reasonably sized samples.

7 K.R. Ihlanfeldt, K. Willardsen / Regional Science and Urban Economics 46 (2014) Table 2 Estimated elasticities of the millage rate with respect to the property tax base. a Principal cities Other metropolitan cities Rural cities Counties Base down t (.180) (.059) (.142) (.093) t (.176) (.061) (.090) (.090) t (.128) (.068) (.065) (.095) LRP.670,.136,.508,.393, (.149) (.081) (.162) (.104) Base up t (.106) (.038) (.088) (.060) t (.039) (.011) (.053) (.020) t (.015) (.013) (.050) (.014) LRP.320,.148,.214,.285, (.113) (.040) (.107) (.066) Observations Robust standard errors are in parentheses. An underline indicates that the down LRP is significantly different from the up LRP at the 10% level or better. a Control variables are changes in per capita income (t 1, t 2, and t 3) and changes in population (t 1, t 2, and t 3). Indicates statistical significance at the 10% level. Indicates statistical significance at the 5% level. Indicates statistical significance at the 1% level. Indicates that the estimated elasticity is significantly different from 1 at the 1% level. Table 3 p-values from tests of difference in estimated millage rate elasticities between groups (null hypothesis is difference = 0). a Principal cities Other metropolitan cities Rural cities Counties Down elasticities Principal cities Difference p-value Other metropolitan cities Difference p-value Rural cities Difference p-value Counties Difference p-value Up elasticities Principal cities Difference p-value Other metropolitan cities Difference p-value Rural cities Difference p-value Counties Difference p-value a Difference is equal to the elasticity for the group identified at the top of table minus the group identified under the Down elasticities and Up elasticities headings. between principal cities and counties has a low p-value (.121), which is statistically significant at the 10% level by a one-tailed test. 27 Regarding the estimated millage rate elasticities with respect to an increase in the tax base (bottom panel of Table 2), except for principal cities, none of the differences between the up and down LRP elasticities is significant. This suggests that these groups react symmetrically to upward and downward movements in their property tax base. Principal cities, on the other hand, do not decrease their millage rate as much in response to an upward movement as they increase the rate in response to a downward movement. The p-values in Table 3 show that only the up LRP elasticity difference between counties and other metropolitan cities is statistically significant. 28 With respect to the residual view of the property tax rate, note that all eight of the estimated LRP elasticities reported in Table 2 are 27 A one-tailed test is appropriate in this case because we have posited that the principal cities' elasticity is larger than the counties' elasticity. 28 For the up elasticities a two-tailed test of significance is appropriate because there are no relative size expectations.