Do State-funded Property Tax Exemptions Actually Provide Tax Relief? Georgia s HTRG Program

Transcription

1 Article Do State-funded Property Tax Exemptions Actually Provide Tax Relief? Georgia s HTRG Program Public Finance Review 2014, Vol. 42(5) ª The Author(s) 2014 Reprints and permission: sagepub.com/journalspermissions.nav DOI: / pfr.sagepub.com Spencer T. Brien 1 and David L. Sjoquist 2 Abstract We examine the effect of a state-funded property tax homestead exemptions on the burden of property taxes. This class of exemptions is characterized by a grant from the state to local governments that is intended to replace the reduction in property tax revenue due to the exemption. The median voter model predicts that part of the homestead exemption will be used to increase expenditures. In addition, fiscal illusion could reduce the effectiveness of this type of grant in lowering the tax burden. We test these predictions for the Georgia s Homeowner s Tax Relief Grant program by separately using panels of county-level data and school system data. We find that over one-third of funds transferred to counties through this program 1 School of Public Affairs, College of Public Programs, Arizona State University, Phoenix, AZ, USA 2 Department of Economics, Andrew Young School of Policy Studies, Georgia State University, Atlanta, GA, USA Corresponding Author: Spencer T. Brien, School of Public Affairs, College of Public Programs, Arizona State University,411N.CentralAve.,Suite400,Phoenix,AZ85004,USA. sbrien@asu.edu

2 Brien and Sjoquist 609 are used to increase revenues rather than provide tax relief. We find evidence of possible fiscal illusion for school systems. Keywords tax relief, intergovernmental grants, property tax, tax incentives While the [grant program] had great motives initially to reduce the local tax burden it has not worked out that way. Governor Sonny Perdue (Salzer 2008) The burden of residential property taxes is a consistent source of voter concern, and as a result all states have implemented policies intended to provide some sort of property tax relief to homeowners. Baer (2008) reports that forty-seven states offer homeowners either a homestead exemption or a credit, thirty-four states have property tax circuit breakers, and twenty-six states allow deferrals on property taxes owed. While some of these state property tax relief policies are simply regulations that restrict the tax that local governments can levy on a given property, many tax relief programs involve actual transfers of state funds to local governments intended to replace lost property tax revenues. One specific type of property tax relief program is a state-funded homestead exemption. Duncombe and Yinger (2001) identified thirteen states with state-reimbursed property tax exemptions, though seven of these states only funded exemptions to special groups, such as residents over sixty-five years of age or military veterans. For the six states that do fund general homestead exemptions for all homeowners, Duncombe and Yinger note that state expenditures on this type of tax relief program can be substantial; for example, New York spent $3.1 billion on the School Tax Relief (STAR) program in FY2010 while California paid out $442.5 million in FY2009. While that level of spending is small relative to the size of California s budget, it is large in comparison to the $293.1 million in unrestricted local government financing proposed in for that year s budget. 1 Georgia adopted a state-funded homestead exemption, entitled the Homeowner s Tax Relief Grant (HTRG) program, in 1999, during a period of state budget surpluses and rising assessed property values and property tax payments. The governor at the time, Roy Barnes, proposed the creation of a state-funded homestead exemption that would transfer surplus state

3 610 Public Finance Review 42(5) funds to local governments as a means of offsetting rising property taxes (Tilley 1999). The HTRG initially gave all homeowners a homestead exemption worth $2,000 of assessed property value, which would provide on average a $50 tax credit. 2 The amount of the exemption changed multiple times over the life of the program. In its second year, the exemption doubled to $4,000 and then increased by $2,000 in each of the subsequent two years. It then stayed at $8,000 for the rest of the program s duration; at this level, the annual appropriation exceeded $430 million. Total spending from 1999 through 2008 was over $3.3 billion. The last year the HTRG program was funded was 2008, after which it was discontinued due to state budget shortfalls and opposition from Governor Perdue. Under the program, homeowners receive a credit on their property tax bill equal to the property tax rate times the homestead exemption. 3 The credit is nonrefundable and is net of other homestead exemptions. 4 The credit is initially financed by the local government, but the state provides a grant to each jurisdiction to offset the revenue local governments forgo as a consequence of the exemption. We explore whether Georgia s HTRG program succeeded in providing property tax relief to homeowners, or whether the program resulted in increases in pre-credit property taxes, as implied by the previously mentioned quote by Governor Purdue. There are three reasons why the latter outcome is likely. First, the increased homestead exemption reduces the tax price of local expenditures, thereby encouraging an increase in public service consumption. Second, the program gives local officials an incentive to raise their property tax rates since the size of the credit, and thus the size of the grant, depends on the local millage rate. Third, because the revenue replacement grants are paid directly to local governments, taxpayers may not be fully aware of how much tax relief they should receive. Given such fiscal illusion, it may be possible for local officials to capture the grant by raising property tax rates. The result of these factors is that some or possibly all of the state funds designated for reducing the property tax burden through a homestead exemption may instead be used to increase local spending. There are very few articles that have estimated the impact of policies similar to the state-funded homestead exemption on local fiscal behavior. The article that is closest to ours is Fisher (1988). Fisher applies the median voter model to measure the effect of Michigan s circuit breaker property tax relief program on the indirect demand for local property taxes. Fisher concludes that the credit led to a modest increase of somewhere between 1 and 8 percent in the level of local property taxes, suggesting that a portion of the tax relief was used to increase local public spending. Eom, Duncombe, Nguyen-Hoang, and Yinger

4 Brien and Sjoquist 611 (2013) explore the impact of the New York School Tax Relief Program (STAR). One of the distinguishing features of the STAR program that sets it apart from other state-funded homestead exemptions is that the amount of the exempted property value is not uniform across the state, but it is weighted using a sales price differential. Their results indicate that the STAR program resulted in an increase in the property tax rate of approximately 21.3 percent for the average New York county. 5 We employ several panel regression models to estimate the impact of Georgia s HTRG program on county and on K 12 school system property tax levies and tax rates. After controlling for other factors, we find that only a portion of the state funds dedicated for tax relief was used to reduce local property taxes and that the program is associated with higher property tax rates. The rest of the article is structured as follows: in the next section, we set forth a theoretical framework in which to analyze the HTRG program. The third section presents the empirical strategy and discusses the data employed. The fourth section presents the results; some concluding thoughts are provided in the final section. Framework Within the framework of the median voter model, and assuming that property value is given and that the marginal cost of a unit of public service is one, 6 the tax price of the median voter equals the ratio of the net assessed value of the median voter s home divided by the net aggregate assessed value. Denote that tax price by t 1 m ¼ u m=v, where v m is the net assessed value of the median voter s property and V is the net assessed value of all local property. Assume that the state mandates an additional locally funded homestead exemption, denoted by E L. The tax price is then given by t 2 m ¼ðu m E L Þ=ðV ne L Þ, where n is the number of homesteaded properties. If u m < V n, then ðu m E L Þ=ðV ne L Þ < u m V, that is, the median voter s tax price will be smaller with the homestead exemption. Of course, to finance the homestead exemption, the government will have to raise the tax rate or use other revenue. The difference between a t 1 m and t2 m reflects the change in tax price due to the locally funded homestead exemption. With a state fully funded homestead exemption, the state government compensates the locality for the loss of revenue due to the exemption by providing a grant equal to tne S, where t is the property tax rate and E S is the state-funded homestead exemption. The median voter s share of the cost of funding local services declines because the grant effectively restores the

5 612 Public Finance Review 42(5) value of the tax base removed by the exemption. 7 This results in a tax price, denoted by t 3 m, equal to ðu m E S Þ=V, noting that ðv ne S þ ne S Þ¼V: t 3 m is clearly smaller than t1 m. The median voter, facing a reduced price for local public services, will prefer a greater public service level, which in turn will require an increase in t. Note that the percentage reduction in the tax price due to a state-funded homestead exemption is equal to E S /n m. Because the value of the homestead exemption is uniform across jurisdictions, the percentage reduction in the median voter s tax price will be progressively distributed according to the median voter s assessed value. If the price elasticity of demand for public services is equal across jurisdictions, we would expect a larger increase in property taxes in jurisdictions with smaller values of n m. In addition to the effect due to changes in the tax price, the level of public expenditures could change if the structure of the reimbursement grant facilitates a form of fiscal illusion. Following Filimon, Romer, and Rosenthal (1982), who present a grant illusion model in which taxpayers base their voting on perceived aid and perceived government expenditures, we assume that the perceived property tax exemption, E S ; is a percentage of the true exemption, that is, E S ¼ð1 rþe S ; ð1þ where 0 r 1. Substituting (1) into the expression for t 3 m results in a new tax price, t 4 m, t 4 m ¼ u m ð1 rþe S ; ð2þ V which is larger than t 3 m for r > 0. Consider the extreme case in which the taxpayer is completely unaware of the exemption, that is, r ¼ 1. In this case, the taxpayer s perceived tax price will be equal to the initial pre-exemption tax price, that is, t 1 m. Thus, the median voter s preferred level of property taxes will remain unchanged when E S is set. The household s net taxable value, however, has been reduced to n m E Sm. Given r ¼ 1, local officials can raise the property tax rate from the original tax rate, t, to a higher rate, t 0, in order to maintain a constant tax levy on the median voter, as depicted in equation (3). tu m ¼ t 0 ðu m E S Þ: At this higher tax rate, t 0, total revenue will be the sum of the original property tax levy, that is, t 0 ðv ne S Þ, plus the revenue replacement grant, ð3þ

6 Brien and Sjoquist 613 that is, t 0 ne S, which yields t 0 ðv ne S Þþt 0 ne S ¼ t 0 V. In this case, the entire grant would be dedicated to expanding public expenditures rather than replacing property tax revenue. With r ¼ 1, equation (3) implies that the percentage increase in the tax rate associated with the fiscal illusion effect is given by t 0 t 1 ¼ u m ðu m E S Þ 1 ¼ E S ðu m E S Þ : One possible cause of this form of fiscal illusion is if taxpayers are attentive only to the actual tax payment that they must make and ignore the property tax rate. Officials could then raise the property tax rate when E S is set without increasing the taxpayer s tax payment relative to the pre-exemption levy, and thus capture the unobserved portion of E. Support for this cause of fiscal illusion can be found in the public choice literature dealing with perceptions of the property tax; see Ordeshook (1979) and Lankford (1986). Homeowners may also accept a degree of rational ignorance regarding the true cost of their services. Duncombe, Miner, and Ruggiero (1997) hypothesized that state aid, by lowering the tax price, reduces taxpayer s incentive to monitor the efficiency of local service production. 8 This argument is supported by Eom and Rubenstein s (2006) findings that New York s state-funded homestead exemption (STAR) induces a greater reduction in efficiency of school system service provision than would a similar change in the homeowner s share of the tax base. The objective of the HTRG was to reduce property taxes on homeowners while retaining the same property tax rate. However, the median voter model implies that the program creates incentives that are predicted to cause local governments to increase total expenditures and the property tax rate. Property taxes may fall but not by the full amount of the grant calculated at the original property tax rate. Fiscal illusion, at its extreme, implies that the state grant will result in the property tax rate increasing to the point that there is no reduction in property taxes. ð4þ Empirical Strategies and Data Empirical Strategies Our objective is to estimate the effect of the Georgia state-funded homestead exemption on property taxes. We consider the effect of the credit program on two dependent variables: (1) the log of the sum of property tax receipts and the replacement grant, denoted by T, and (2) the log of the

7 614 Public Finance Review 42(5) property tax rate, denoted by R. We approach the question in several different ways using two different types of jurisdictions: counties and school systems. In what follows, we first discuss the approach using T and then discuss the approach using R. In our first approach, we model the demand for T as a function of the local tax price, personal income, and variables that measure taste for and cost of public services. We first measure the effect of the HTRG program on T with the tax price parameter t 3 m. A coefficient on t3 m that is negative and statistically significant is consistent with the hypothesis that the HTRG program has an effect on T. Since the coefficient on t 3 m does not tell us how much the HTRG program affected T, we separate the tax price into two components in order to measure the effect of HTRG more directly. In particular, we separate t 3 m into the tax share pre-htrg exemption, that is, t 1 m ; and a measure of change in the tax price due to the HTRG program, denoted by Dt. We measure the change in the tax price as the percentage reduction in the tax price caused by the HTRG program, that is, E S /(n m ), which we denote as percentage change in tax price (PCTP). 9 In these regressions, we do include t 1 m. To explore the presence of fiscal illusion, we use t 4 m in place of t3 m.ifthe median voter s perceived tax price does not equal t 3 m, the coefficient on t3 m would be biased. Fiscal illusion is thus another reason why the coefficient on t 3 m may not measure the effect of HRTG. In general, our model is represented by the following linear form: T jt ¼ a i þ b 1 t jt þ b 2 Dt jt þ b 3 Y jt þ b 4 TT t þ X kb k Z kjt þ d j þ2 jt ; ð5þ where j represents a jurisdiction and t represents the year; T jt is the log of the sum of property tax and replacement grant revenue per capita; t jt represents the log of tax price; Dt jt is the change in the tax price due to the exemption (PCTP); Y jt is the log of per capita personal income; TT t is a time trend; Z kjt is a set of k taste and cost factors; d j are jurisdiction fixed effects; and 2 jt is a random error term. Equation (5) is estimated for each of the two different price structures: (1) t jt equals t 3 m and Dt jt is omitted; (2) t jt is t 1 m and Dt jt is E S /(n m ). Equation (5) is estimated with fixed-effect panel regression models; we correct for within-group correlation of the 2 jt term by using jurisdictionclustered standard errors and control for heteroscedasticity by estimating robust standard errors. The jurisdiction fixed-effects control for timeinvariant jurisdiction-level factors that are unobserved in our data, but that

8 Brien and Sjoquist 615 influence T jt. Since the composition of jurisdictions do not change much over the time period, the fixed-effects control for factors such as age, gender, race, and education level. We also estimate an equation that uses a first difference model of equation (5). The first difference model eliminates the d j term and focuses the analysis on changes in the price variables over time rather than cross-sectional differences. Our second approach to measuring the effect of the HTRG replaces the Dt jt variables with the actual replacement grant revenue per capita received through the HTRG program, denoted by GRT jt. If there is no effect of E S on the combined revenue T jt, then the HTRG grant would reduce property tax revenues on a dollar-per-dollar basis. If that were true, the coefficient on GRT jt would be zero. On the other hand, a positive coefficient on GRT jt would imply that there was an effect on T jt from E S. Using the revenue from the grant as an explanatory variable is problematic because it is endogenous with the dependent variable; both the level of revenue from the property tax and the size of the replacement grant are determined in part by the jurisdiction s property tax rate. The simultaneous determination of the two variables would cause the value of the grant to be correlated with the error term in the estimating equation. Because of this relationship between the two variables, we present an additional model that replaces the log of the actual amount of the grant with the log of the grant that the jurisdiction would have received if the millage rate had been maintained at the pre-htrg level, denoted by PreGRT. Since the tax rate and the grant are both endogenous, we calculate the PreGRT at the level it would have been if the millage rate was kept at the average millage rate for the three years prior to the HTRG program. Our final approach attempts to identify the effect of fiscal illusion. If we assume that voter behavior conforms to the median voter model and there is no fiscal illusion, then voter behavior should be based on the tax price given by t 3 m. If there is complete fiscal illusion, that is, r ¼ 1, then the perceived tax price would equal t 1 m. To estimate the extent of fiscal illusion, we use t4 m as the principal independent variable and estimate equation (5) using nonlinear regression techniques. We also explore the effect of the HTRG program on the property tax rate. Note first that the sum of property tax revenue and the grant can be expressed as PT þ Grant ¼ tðb ne S ÞþtnE S ¼ tb; since tne S ¼ Grant, and where PT is the property tax revenue, Grant is the grant, B is the tax base gross of E S, and the other variables are as defined ð6þ

9 616 Public Finance Review 42(5) above. Recall that T is the log of the sum of PT and Grant and R is the log of the property tax rate t. Solving equation (6) for t and taking logs yields R ¼ T ln B: ð7þ To estimate equation (7), we replace T with the right-hand side of equation (5), and estimate the resulting equation (7) using a fixed-effect regression with robust clustered standard errors. Following the approach used by Holtz-Eakin and Rosen (1990), this strategy measures the substitution effect directly on the chief policy instrument that local officials employ to alter the property levy. 10 An alternative, and somewhat ad hoc, approach to estimating the effect of the HTRG program on the tax rate is to estimate the following equation R jt ¼ b 0 þ b 1 R j0 þ b 2 ðv jt V j0 Þþb 3 ðy jt Y j0 Þþb 4 PCTP jt þ b 5 X jt þ b 6 TT j þ2 jt ; ð8þ where R j0 is the log of the tax rate pre-htrg and thus captures all of the factors that determine the tax rate in year zero; PCTP jt is the percentage reduction in the tax price due to the exemption; V jt V j0 is the change in the log of the tax base per capita from year zero, and thus captures the effect of changes in the per capita tax base on the tax rate; Y jt Y j0 is the change in the log of income per capita from year zero, and thus captures changes in demand for public services (and thus the demand for property taxes) due to changes in income; X jt is the change in the log of ST for counties and of SCHAID for schools. Data We estimate the various regressions using Georgia county government data and K 12 school system data. We separately consider these two types of governments because they each have their own elected governing body and provide substantially difference functions, nor do they appear to cooperate in setting property tax rates. Considering the effects of HTRG on two different types of governments adds robustness to our results. For the county-level analysis, the property tax revenue data were obtained from the Report on Local Government Finances (RLGF) collected by the Georgia Department of Community Affairs; this is an annual report completed by local governments in Georgia that collects detailed information on revenues and expenditure data. However, not all counties complete the survey every year. For school systems, we collected information on system revenues

10 Brien and Sjoquist 617 from the Georgia Department of Education. Tax base and tax rate data are from the Georgia Department of Revenue. Of the 159 counties in Georgia, we possess complete data for 116 counties over the fifteen-year period of 1996 to 2010 and adequate partial data for the remaining forty-three counties. Georgia has 180 local school systems, of which 159 are county school systems (these encompass the entire county net of the independent school systems) and twenty-one are independent city systems. Because of missing data, we exclude the twenty-one city systems. Of the 159 county school systems, we possess full data for 120 school systems from 1996 through We also have at least seven years of data for the remaining thirty-nine districts. We use all jurisdiction-year observations when complete data are available for those jurisdictions for those years. We follow the standard assumption that the median voter in a given locality owns the median value home. Median home value at the county level is not available through either state or national sources except for the decennial census year. We approximate the median home values for the rest of the panel by extrapolating from the assessed residential parcel values. We isolate the total assessed residential property value from the rest of the total base and then divide this amount by the number of housing parcels in the county to obtain a per housing unit value. We then calculate the percentage change in this measure for each year and county of our panel and apply it to the 2000 Census Bureau median home value. We compared this measure with the median home value estimates reported in the annual American Community Surveys (ACS) for The ACS survey only reports median house values for a subset of all counties; we compared our interpolated median house value with the Census figures for twenty-nine counties. We ran separate t-tests for each of the four years on the twenty-nine jurisdictions and failed to reject the null hypothesis that the mean difference was zero for three of the four years using.05 a level; an a of.01 fails to reject for all four years. Failure to reject the null supports our use of this measure, as it indicates that its distribution tends to center on the ACS figure. We also compared our measure of the percentage increase in median home value with the Federal Housing Finance Agency (FHFA) housing price index. We used the price index for the three MSAs located in Georgia, Athens-Clarke County, Atlanta-Sandy Springs, and Augusta-Richmond County. We compare these price indexes with our average residential parcel growth factor for the larger counties in those MSAs. 11 We compare these figures from 1997 through A graphical analysis of the trends over this period revealed a strong similarity in the behavior of the two measures. Given these results, we are comfortable with the use of our interpolation

11 618 Public Finance Review 42(5) of median home value and use it as the n it term in each of the tax price specifications. If the proportion of residents living in homes is greater than 50 percent, then it is presumed that the median voter is also a homeowner. The 2010 five-year ACS from the Bureau of the Census reports that all but three Georgia counties had owner occupancy rates greater than 50 percent. The three exceptions, Chattahoochee, Clarke, and Dougherty Counties, had owner-occupied rates of 31.3 percent, 45.9 percent, and 48.5 percent, respectively. Thus, our assumption that the median voter is a homeowner is violated in only three cases. Georgia provided a number of other types of tax relief programs during the period of our panel that we take into consideration. A locally funded $2,000 homestead exemption is provided by the state to homeowners for all state, county, and school property taxes; several counties and school districts have larger homestead exemptions. Our measure of the net property tax base used to calculate t 1 m and t3 m take all local homestead exemptions into account. We use the assessed value net of homestead exemptions in the numerator of the price ratios. This follows from our analysis of a nonfunded homestead exemption depicted in t 2 m. Beginning in 1976, county governments began adopting, via referendum, a 1 percent local option sales tax (LOST), with the revenue being shared with municipal governments located in the county. The LOST revenue was designated for property tax relief; however, Jung (2001) estimates that at best only 28 percent of LOST revenue is used for property tax relief. All but 5 of the 159 Georgia counties have adopted a LOST, but there have been no LOST adoptions since the HTRG program was implemented. Since no new LOSTs were adopted during the period of our panel, no adjustment is necessary beyond our control using LOST revenue per capita. In 2009, Georgia passed a law which froze property tax assessments for 2009 through Our price measures are generated from property tax digest data and therefore control for this policy. Finally, there is a property tax cap of twenty mills that applies to most county school districts. (County property tax millage rates can be set without voter approval.) The limitation does not appear to have a significant impact on school district property tax rates; in 2000, this cap was not binding on any school systems, while in 2007 it was binding for four school systems. We use several variables to control for other determinants of the indirect demand for property taxes and the property tax rate. Log of county-level per capita personal income, denoted PCI, and log of county level total employment per capita, denoted by EMPL, come from the Bureau of Economic

12 Brien and Sjoquist 619 Analysis. We include population density, denoted by DEN (obtained from the Census Bureau), under the expectation that more concentrated populations will have a higher per capita demand for public services and thus higher property taxes. We include the percentage of county wages derived from agriculture, denoted by AGR, and from the service sector, denoted by SER, in order to measure any impact that the local economy would have on fiscal structure; these data came from the Quarterly Census of Employment and Wages collected by the Bureau of Labor Statistics. For counties, we also include the log of total sales tax receipts per capita, denoted by ST, for the jurisdiction since local governments that have local option sales taxes could use increases in sales tax revenue to reduce property tax levies. For school systems, we include the log of state school aid per capita, denoted SCHAID, since greater school aid could be used to reduce property taxes. The sales tax data for counties were obtained from the DCA survey, while the data for schools came from the Georgia Department of Education. The level equations all contain a time trend variable denoted TT. We also include the share of the property tax base that consists of residential property, denoted RES, obtained from the Georgia Department of Revenue. Summary statistics are displayed in table 1. All dollar values are in real terms using the CPI ( ¼ 100). Econometric Issues A potential problem is that the HTRG credit could be capitalized in the market price of residential property, thus introducing endogeneity into our causal analysis. We do not believe this to be a serious issue for our analysis. First, the frequency with which the exemption changed over the first few years of the program was too rapid to be both capitalized into the market value and then incorporated into the assessed value. Assessed values are determined as of January 1 for the property taxes collected in the fall of that year, while the value of the HRTG homestead exemption for that year is not set until April. Thus, it is not possible that the assessed value reflected the HRTG credit for that year. Furthermore, the results we present in the following section are robust to limiting the panel to the pre-htrg years and the first four years of the program where the exemption changes each year. Second, even if capitalization occurred, the effect on taxes would be very small and would lead to an underestimate of the effect of HTRG on property taxes. We are therefore confident that the capitalization of HRTG is not a significant issue. Of course, if the HRTG credit was in fact capitalized, then the results are subject to endogeneity.

13 620 Public Finance Review 42(5) Table 1. Descriptive Statistics, Levels Logs Variable Mean SD Mean SD County property tax and grant revenue per capita School property tax and grant revenue per capita County property tax millage rate School property tax millage rate Ln county pre-exemption tax price, t 1 m Ln school pre-exemption tax price, t 1 m Ln county post-exemption tax price, t 3 m Ln county post-exemption tax price, t 3 m Percentage change in price from HTRG (PCTP) Percentage change in price from HTRG (PCTP) a County HTRG grant per capita a Counterfactual county HTRG grant per capita a School HTRG grant per capita Counterfactual school HTRG grant per capita Ln county sales tax per capita, ST Ln state aid to school per capita, SCHAID Population density (100 per sq. mile), DEN Ln county net tax base per capita, V Ln school net tax base per capita, V County change in net base per capita from pre-htrg, (V t V 0 ) School change in net base per capita from pre-htrg, (V t V 0 ) Per capita income 13, , Change in income per capita from pre HTRG, (PCI t PCI 0 ) Percentage of gross base classified as residential, RES Percentage of wages in agricultural sector, AGR Percentage of wages in service sector, SER Employment per capita Note: HTRG ¼ Homeowner s Tax Relief Grant. a Statistics are computed over (the duration of the HTRG).

14 Brien and Sjoquist 621 A second potential concern is that variations in the quality of assessments will be associated with variations in the tax price as well as the tax rate. We reviewed the annual sales assessment ratio for each county and find no meaningful change in the ratio for the years when the HTRG was in place as compared to other years. Nor is there any change in the pattern across years for the years in which the HTRG was in place. Thus, we don t believe that endogeneity is a major issue. Results Property Taxes and Replacement Grant Revenues Per Capita Table 2 contains the estimation results of our fixed-effects model of T in levels. Columns 1 and 2 are the results using t 3 m, that is, the log of the post-exemption tax price, for counties and for school systems, respectively. The coefficients on t 3 m are negative, as expected, for both the counties and school systems, and are very similar in magnitude. The implication is that the reduction in the tax price due to the HTRG program increased property taxes inclusive of the HTRG credit. Given that the HTRG program reduced the tax price by an average of 22 percent, the coefficients on t 3 m imply that the HTRG program increased county (school system) property taxes inclusive of the HTRG credit by about 3.9 percent (4.0 percent). At its peak, the HTRG provided an allowable exclusion of up to $8,000 in assessed value. In that year, the average percentage change in tax price due to the exemption was 31.2 percent, which yields an estimated 5.5 percent increase in property taxes inclusive of the HTRG credit for counties and 5.7 percent for school systems. The coefficients on PCI, that is, the log of per capita income, are positive, as would be expected, and statistically significant, and are less than one. The income elasticity is larger for schools. The coefficient on TT, the time trend, is positive and statistically significant in both regressions. With the exception of the residential share (RES) and agricultural share (AGR), the other control variables are generally not statistically significant. The coefficients on the tax price variables are robust to which control variables are included in the regressions. The rest of table 2 contains the results when we separate t 3 m into t1 m and the measure of the effect of the HTRG program on the tax price, PCTP. But for comparison purposes, we first report in columns 3 and 4 regressions using t 1 m but not the variable measuring the change in the tax price. We can think of this as a naive model, since t 1 m ignores the effect of E S on the tax price. As with t 3 m, the coefficients on t1 m are negative and statistically

15 Table 2. Basic Regression Results Using Tax Price-Dependent Variable: T, the Log of the Sum of Property Tax Revenue and HRTG per Capita. Variables (1) County: postexemption tax price (2) School: post-exemption tax price (3) County: preexemption tax price (4) School: preexemption tax price (5) County: PCTP (6) School: PCTP t 1 m ( 0.959) t 3 m 0.179*** ( 5.876) 0.184*** ( 5.120) 0.200** ( 2.273) ( 1.229) PCTP 0.249*** ( 5.342) 0.213** ( 2.395) 0.229*** ( 5.937) PCI 0.268** (2.241) DEN (0.153) EMPL (0.956) ST ( 0.471) 0.515*** (3.466) 0.051*** (4.083) (0.939) SCHAID (1.100) SER ( 0.769) AGR (0.575) (0.763) 0.004** ( 2.180) 0.427*** (3.602) (0.283) (0.911) ( 0.414) ( 0.795) (0.640) 0.651*** (4.879) 0.054*** (4.454) (0.793) (1.400) (0.673) 0.004** ( 2.008) 0.257** (2.136) (0.150) (1.153) ( 0.596) ( 0.681) (0.501) 0.496*** (3.322) 0.050*** (3.995) (0.915) (1.137) (0.772) 0.004** ( 2.091) 622

16 RES 0.004** (2.188) TT 0.031*** (12.952) Constant (1.620) 0.006*** (3.398) 0.020*** (9.420) ( 0.102) (0.478) 0.033*** (13.619) (0.388) 0.005** (2.247) 0.021*** (10.235) ( 1.089) (1.145) 0.031*** (12.918) Observations 2,320 2,325 2,320 2,325 2,320 2,325 R Residual sum squares 2.004* (1.723) *** (2.706) 0.020*** (9.714) (0.015) Note: PCI ¼ per capita personal income; DEN ¼ density; EMPL ¼ employment per capita; HTRG ¼ Homeowner s Tax Relief Grant; PCTP ¼ percentage change in tax price; ST ¼ sales tax; SER ¼ service sector; AGR ¼ agriculture; RES ¼ residential; TT ¼ time trend. Fixed-effects regression is with robust clustered standard errors. t-statistics values are shown within parentheses. *p <.1, **p <.05, ***p <

17 624 Public Finance Review 42(5) Table 3. Regression Using HTRG Grant per Capita. Variables (1) County (2) School (3) County (4) School t 1 m 0.148* ( 1.941) 0.266*** ( 3.086) ( 1.518) 0.254*** ( 3.307) GRT 0.036*** (7.503) 0.033*** (9.662) Pre-GRT 0.046*** (7.670) 0.028*** (7.912) PCI (1.254) 0.357** (2.377) 0.290* (1.912) 0.285** (2.149) DEN ( 0.131) 0.040*** (3.249) (0.575) 0.043*** (2.790) EMPL (0.977) (0.803) (0.037) (1.123) ST ( 0.702) 0.030** ( 2.243) SCHAID (1.359) (0.447) SER ( 0.542) (0.930) ( 0.534) (0.490) AGR (0.708) 0.004* ( 1.784) (0.820) 0.003* ( 1.907) RES (1.294) 0.007*** (2.939) (0.992) 0.006*** (2.971) TT 0.032*** (13.026) 0.021*** (9.985) 0.035*** (11.995) 0.020*** (9.205) Constant 2.945** (2.456) (0.902) (1.112) (1.626) Observations 2,320 2,325 1,844 1,836 R Residual sum squares Note: PCI ¼ per capita personal income; DEN ¼ density; EMPL ¼ employment per capita; HTRG ¼ Homeowner s Tax Relief Grant; ST ¼ sales tax; SER ¼ service sector; AGR ¼ agriculture; RES ¼ residential; TT ¼ time trend. Fixed-effects regression is with robust clustered standard errors. t-statistics values are shown within parentheses. *p <.1, **p <.05, ***p <.01. significant for both types of jurisdictions. The coefficients on t 1 m are larger in absolute value than the coefficients on t 3 m, particularly for schools; this result is not surprising since the value of t 3 m is smaller than the value of for all jurisdictions, which should yield a larger coefficient. t 1 m

18 Brien and Sjoquist 625 Since the proportional change in the tax price due to the HTRG program, PCTP, is measured as ( E S /n m ), we expect the coefficient to be negative if HTRG increases T. For counties and school systems, the coefficients on PCTP are negative and statistically significant (columns 5 and 6 of table 2). Given the coefficient for counties of 0.249, the mean value of PCTP of 0.22 implies that T (property taxes inclusive of the HTRG credit) increased by about 5.4 percent due to the HTRG. 12 The estimated effects of HRTG on property taxes using PCTP are consistent with the estimates using the coefficients on t 3 m. The coefficients on the control variables are similar in columns 1 and 5 and in columns 2 and The next set of results uses the actual replacement grant amount to control for the impact of the HTRG on local fiscal policy. Table 3 contains two estimates each for the counties and the school systems. The first two columns use the log of the actual replacement grant per capita, that is, GRT, as the key variable, while the dependent variable remains the log of the combined property tax and replacement grant revenue per capita, T. Note that the actual replacement grant revenue is a component of the dependent variable and is the key explanatory variable in these equations. If there is no response to the HRTG, then the coefficient on the grant revenue should be zero because the grant would simply be replacing the property tax revenue and the overall revenue level would be held constant. The coefficient estimates for GRT are positive and statistically significant for both the counties and the schools, indicating that a 10 percent increase in replacement grant revenue increases combined property and replacement grant revenue by about 0.3 percent for counties and for school systems. However, there is the potential for endogeneity to bias this estimate because the amount of grant revenue is determined by the current year s millage rate, and thus is endogenous with property tax revenue. Therefore, in columns 3 and 4 of table 3 we use an alternative measure, PreGRT, which is the current year s HTRG homestead exemption multiplied by the pre- HTRG property tax rate, which measures what the grant would have been had the jurisdiction retained the same millage rate. The coefficients on PreGRT are statistically significant, but slightly smaller in magnitude than those in columns 1 and 2. These findings provide additional supporting evidence that the localities did exhibit a price response to the HTRG. We also estimated regressions in which we replaced PreGRT with the percentage change in net taxable assessed value due to E S, that is, (ne S )/V, but we do not include these results in a table. Although this approach lacks the theoretical connection between the median voter s preferences and the public choice mechanism, it reflects the magnitude of the overall exemption

19 626 Public Finance Review 42(5) Table 4. Regressions for Log of Tax Rate. Variables (1) County (2) School t 1 m ( 1.390) ( 0.466) PCTP 0.185*** ( 4.101) ( 0.490) V 0.629*** ( ) 0.310*** ( 8.583) PCI (0.704) 0.145* (1.809) DEN 0.044* (1.715) (1.135) ST ( 1.590) SCHAID ( 1.196) EMPL ( 0.142) (0.295) SER ( 0.349) ( 0.017) AGR (0.787) ( 1.238) RES (0.159) ( 0.787) TT 0.023*** (9.750) 0.012*** (6.642) Constant 7.128*** (5.808) 4.180*** (5.476) Observations 2,335 2,314 R RSS Note: PCI ¼ per capita personal income; DEN ¼ density; EMPL ¼ employment per capita; ST ¼ sales tax; SER ¼ service sector; AGR ¼ agriculture; RES ¼ residential; SCHAID ¼ log of state aid to school districts per capita; TT ¼ time trend. Fixed-effects regression is with robust clustered standard errors. t-statistics values are shown within parentheses. *p <.1, **p <.05, ***p <.01. relative to the tax base. Assuming that the HTRG program does affect property taxes, we would expect that the larger the reduction in net assessed value due to the HTRG program, the larger the increase in property taxes inclusive of the HTRG credit. Consistent with this supposition, we find that the coefficients on the percentage change in the net tax base are positive and statistically significant. Property Tax Rate An alternative approach to measuring the effect of the HTRG program is to examine changes in the property tax rate. The property tax rate, as one of the determinants of the size of the replacement grant, is independently controlled by the local governments and would be the primary fiscal tool used to capture the property tax relief funds. Table 4 contains the results of applying the framework implied by equation (7). The coefficients for PCTP are negative for both counties and schools, indicating that property tax rates

20 Brien and Sjoquist 627 Table 5. Property Tax Rates Dependent Variable: R t. Variables (1) County (2) School R *** (14.129) 0.684*** (9.730) PCTP 0.736*** ( 7.325) 0.171** ( 2.458) Log V t Log V ( 0.048) 0.015*** ( 3.554) ST t ST ( 1.557) SCHAID t SCHAID ( 0.309) PCI t PCI ** ( 2.192) ( 1.285) TT 0.022*** (7.988) 0.009*** (5.236) Constant (1.539) 0.469*** (2.877) Observations 1,701 1,681 R Note: PCI ¼ per capita personal income; ST ¼ sales tax; TT ¼ time trend. GLM regressions are with robust clustered standard errors. z-statistics values are shown within parentheses. *p <.1, **p <.05, ***p <.01. increased in response to the HTRG program s reduction in tax prices, but are statistically significant only for the counties. The PCTP elasticity estimate suggests that county property tax rates increased by about 4.1 percent in response to the average 22 percent reduction in the tax price due to the exemption. Similar results are obtained if we use PreGRT to measure HTRG or estimate first difference regressions. The coefficient estimate for the log of the net property tax base per capita, V, is significant and negative for both the counties and the school districts, revealing the not unexpected result that jurisdictions with higher tax bases per capita tend to have lower property tax rates. Interestingly, the magnitude of V s effect in school districts is roughly half the size of the effect for counties, a difference that is statistically significant. This result suggests that as the property tax base increases, school districts capture a larger share of revenue that would be generated by the increase in the base with no change in the tax rate than do counties. Table 5 presents the results based on equation (8), which is the more ad hoc approach. Note that R 0, V 0, and so on, are the average of values (the pre-htrg years in our panel) and that the regressions are estimated using data for The coefficients on PCTP are negative as expected and statistically significant for both counties and schools. The estimated coefficients on PCTP indicate that the average change in the tax price of 0.22 was associated with an increase in the tax rate of 16.1 percent for

21 628 Public Finance Review 42(5) the counties and 3.7 percent for the school districts. Similar results are obtained if we use PreGRT to measure the HTRG. Furthermore, the results are robust to excluding the other variables in the regression, or using the percentage change in the tax rate as the dependent variable and dropping R 0 as a dependent variable. The coefficients on the other variables are what one expects, except for the negative coefficients on income, and are generally statistically significant. Estimating the Share of Grant Used to Increase Property Taxes Our analyses of property tax revenues and millage rates both provide supportive evidence that local governments used a portion of the replacement grant to increase revenue per capita rather than lowering the property tax burden by the total extent of the state grant. To further assess the impact of the HTRG, we used the estimated coefficients to extrapolate the share of the grant that was used for tax relief. We first calculate the predicted value 14 of the aggregate property and replacement grant revenue per capita, designated PRED 1. We then compute a second set of predicted values, denoted by PRED 0, generated using the same coefficients, except that we replace the estimate of the homestead exemption s effect on the tax price with the null hypothesis value of zero. This quantity represents the predicted sum of property tax and replacement grant revenue if there had been no substitution effect associated with the state-funded exemption. We transform these predicted values into real dollar amounts, multiply by county population, and aggregate across all jurisdictions for counties and school systems, respectively. We calculate the share of the grant used to increase local revenues from these predicted values. This is done by taking the difference between the transformed values of PRED 1 and PRED 0 and then dividing this difference by the total receipts from the grant. We interpret this as the share of the grant used to increase local spending rather than reduce property taxes. When we calculate this share in aggregate across all local governments from 1999 through 2008, we find that share of the grant used to increase local spending was 33.2 percent for counties and 33.3 percent for school districts. These results are shown in table 6. Considering the effect by individual years, the estimates ranged between 29 percent and42percentforcountiesandbetween28percentand44percentfor the school systems. These estimates appear slightly larger than the actual elasticity estimates for PCTP ( and for counties and schools, respectively). Those elasticities, however, are relative to

22 Brien and Sjoquist 629 Table 6. Estimating the Percentage of HTRG Used for Property Tax. Counties Schools Aggregate predicted revenue $12,322,572,328 $21,631,150,042 Null hypothesis predicted revenue 11,964,671,099 21,045,151,562 Difference (¼ estimated property tax 357,901, ,998,480 increase) Total HTRG payments to counties 1,076,812,779 1,758,840,342 Percentage of HTRG used to increase spending 33.24% 33.32% Note: HTRG ¼ homeowner s tax relief grant. All dollar amounts are aggregated from 1999 to Predicted values generated from the estimates in columns 5 and 6 of table 4. the per capita values of the dependent variable, while these ratios are in normalized dollar amounts. Fiscal Illusion Based on Filimon, Romer, and Rosenthal (1982), we modeled fiscal illusion as a misperception of the median voter regarding the actual value of the HTRG credit, as expressed in equation (2). If (1 r) ¼ 1 in equation (2), there is no fiscal illusion, while if (1 r) ¼ 0, the median voter acts as if he is unaware of the HTRG credit. To estimate the extent of fiscal illusion, we use a nonlinear regression technique to estimate equation (5) using the log of t 4 m to measure the tax price; we include fixed effects and report robust clustered standard errors. Using the results from the regression (which are not reported here), the estimated value of (1 r) is for counties. The 95 percent confidence interval for the county estimate of (1 r) ranges from 0.20 to Thus, we can reject complete fiscal illusion for counties, that is, that (1 r) ¼ 0, but the size of the confidential interval suggests that there is a substantial probability that there is partial fiscal illusion. Similarly, the estimated value of (1 r) for the school districts was 0.73 with a 95 percent confidence interval of 0.23 to Thus, we can reject the complete fiscal illusion hypothesis that (1 r) ¼ 0, but are unable to reject the perfect information hypothesis in which (1 r) ¼ 1. Our estimates do indicate, however, that a substantial portion of the confidence intervals of our estimates fall within the range between zero and one, which would be consistent with partial fiscal illusion. Under partial fiscal illusion, the response to the HTRG program would exceed the price response of a fully informed median voter. The proportion of the confidence interval falling in