Local sales tax competition and the effect on county governments' tax rates and tax bases.
Although the economics literature provides theoretical predictions for tax competition between jurisdictions, few researchers have undertaken empirical analysis of the subject. This study builds on the theoretical frameworks provided by Mintz and Tulkens (1986) and Kanbur and Keen (1993), which examine tax competition between jurisdictions where residents can freely cross borders to purchase goods subject to lower taxes. I examine local tax competition by testing whether local governments consider neighboring jurisdictions' sales tax rates when setting their own rates. I also analyze the responsiveness of sales tax rates to changes of the tax base and vice versa.
Governments compete to attract tax bases and economic activity for local businesses. Their policies can draw residents and businesses, prevent residents and businesses from exiting, or simply match the innovations of other jurisdictions (Fisher 1980). Governments use both tax rates and expenditures to compete with their neighbors (Wildasin 1988). For example, governments may keep tax rates low in hopes of attracting tax-sensitive shoppers and businesses to the area (depending on the elasticity of cross-border shopping), but they may spend relatively large sums on education to appeal to both families and businesses that depend on an educated work force.
Prior empirical research examines tax competition at the state level and the relationship between home and bordering states' tax and spending policies. Hewett and Stephenson (1983) examine the relationship between revenue and tax rates and find that Iowa's tax revenues depend both on its own rates and the tax rates of competing jurisdictions. Case et al. (1989) find that a state's spending level affects the spending levels of its neighbors, supporting the notion that jurisdictions compete using expenditures. Case (1993) examines border states and the political consequences of raising taxes. She finds that legislators are more likely to raise income taxes when neighboring states have done the same and that voters look at neighboring states to determine whether a tax increase in their own state is appropriate. Finally, Rork (2000) finds that competition between states exists for state taxes on cigarettes, motor fuel, and corporate income.
I estimate two separate equations, the local sales tax rate and sales tax base, in both the short and long run. The local sales tax rate equation represents the county's tax policy choices and the sales tax base equation represents the demand for the county's taxable sales. The study reaches two basic conclusions. First, local governments raise sales tax rates in response to their neighbors' rate increases in both the short and long run. Increases in the tax rates of adjoining counties make it economically and politically easier for local governments to raise tax rates at home. Second, the sales tax rates of both home and competing counties affect the sales tax base. Counties gain from a neighbor's rate increase as consumers cross borders to take advantage of lower rates or home county residents choose to shop more at home; however, counties lose tax base in response to their own tax rate increases.
This study contributes to the literature in a number of important ways. First, I analyze tax competition using both short-run and long-run models. Tax rate changes occur only in large, discrete intervals, which means that the actual tax rate in a county is likely to differ from the desired tax rate for some time period after a change, such as after a rate increase in a neighboring county. The long-run model investigates the responses of home tax rates to neighbors' rate increases and tax base changes once time has permitted all responses to be fully realized. The short-run adjustment model for the tax rate estimates the initial change in a county's tax rate in response to changes in the neighbor's tax rate, the tax base, and other factors. Further, the short-run model measures the length of time required for the actual sales tax rate to return to the desired rate after a change, such as in a neighbor's tax rate. Estimates of both short- and long-run equations also add to the robustness of the results.
Second, this study examines sales tax competition at the local level, while most prior studies focused primarily on tax competition between states. The study uses data for Tennessee's 95 counties for the years 1977-1993. The data permit investigation of competition between the 95 counties as well as competition between Tennessee counties and counties in the eight states bordering Tennessee. Reactions to neighboring tax rate changes should be easier to observe at the local level than at the state level because of the smaller geographic area involved. In addition, the study is better able to capture the effects that cross-border shopping has on governments' decisions to raise tax rates because of the focus on sales taxes. Finally, unlike most previous research, the study is able to directly examine the effects of a neighbor's tax rate change on the home jurisdiction's tax rate and base.
The questions raised in this paper are increasingly vital as cross-border shopping increases, including crossing physical borders or ordering via mail-order or Internet. (1) Tax competition will grow significantly, and state and local tax bases erode in response, unless Congress acts to enable sales tax collection on remote sales. (2) The results of the study have implications for the broader issue of setting tax structures in an era of remote commerce, though the time period studied is prior to the development of e-commerce.
The remainder of the paper is organized as follows. The second section discusses interjurisdictional competition and the theoretical background. The third section presents the empirical model, and the fourth section describes the data. Results are discussed in the fifth section, followed by the contributions and limitations of this research.
INTER JURISDICTIONAL COMPETITION AND THEORY
Most U.S. cities and counties rely almost entirely on property taxes and local option sales taxes for tax revenues. According to the U.S. Census Bureau, property and sales taxes for all U.S. states comprised 89.3 percent of all local government tax revenues in 1996-97, with 73 percent of the tax revenue coming from property taxes and 16.3 percent coming from sales taxes. In Tennessee, property and sales tax revenues accounted for 94 percent of 1997-98 local tax revenues, with 59 percent coming from property taxes and 35 percent from sales taxes. This study focuses on the sales tax because it is highly visible to consumers, appearing as a separate line item on virtually all retail purchases. Consequently, consumers can quickly adjust their spending behavior to take advantage of rate differences in adjacent jurisdictions. Another reason I focus on sales taxes is because prior research has indicated no statistically significant association between property tax rates and the location of retail outlets (Fox 1986).
Sales taxes are intended to be destination taxes, or taxes imposed on consumption in the jurisdiction where the good is enjoyed and not necessarily where the good is purchased. If governments could enforce destination taxes, tax competition would not exist (Kanbur and Keen 1993). However, state and local governments have a limited capacity to enforce destination taxes because of the free movement of businesses and individuals. (3) When one jurisdiction increases its taxes, consumers and producers can shift their purchases to another relatively lower-taxed jurisdiction, thereby reducing sales in the tax-increasing jurisdiction and increasing sales in the neighboring one. Elected representatives therefore face a tradeoff in setting sales tax rates because the revenue from the higher rates can come at the expense of a smaller tax base.
I develop the models in this paper from the theoretical results of Mintz and Tulkens (1986) (hereafter MT) and Kanbur and Keen (1993) (hereafter KK). Their theoretical structures are based on the premise that governments will consider their neighbors' tax rates when setting their own. The models of both MT and KK examine the efficient setting of rates in the context of tax competition between autonomous fiscal authorities. Specifically, these studies examine the Nash equilibrium in the case of a two-region economy, where each region levies an origin-based commodity tax on a private good to finance a local public good. Residents of each region can purchase the taxed good either in their own region or in the neighboring region, but consumers incur a transaction cost when they travel from one region to the other.
MT show that the home government may increase rates, decrease rates, or leave them the same in reaction to an increase in a neighbor's rate. The expectation is that local governments consider their neighbor's rate when setting their own, although the direction of a particular government's response to a change in its neighbor's tax rate is not necessarily predictable. The reaction will depend on its residents' preferences for public versus private goods. Consider the case where the neighbor county raises its tax rate. The home county may respond in three possible ways. First, the home county may increase its tax rate. This might be expected in cases where home county residents have a strong preference for additional public goods. The idea is that the marginal cost of public services was lowered since nonresidents are paying more of the cost when they increase cross-border shopping; thus, the home county buys even more public services. (4) KK's model, based on a revenue-maximization model compared to MT's utility-maximizing model, find that home counties always increase tax rates in response to a neighboring county's tax rate increase. Second, the home government can leave tax rates unchanged, thereby allowing all of the revenues from greater cross-border shopping to go for additional public services (but not seeking additional public services by raising the tax rate, as in the first case). Finally, the home government can decrease tax rates. By decreasing rates, the home government can increase public services and increase the after-tax income of its residents as some of the cross-border shopping revenues are converted to private dollars.
In summary, the theory predicts a tax policy response by the home county to a neighbor's tax rate change (though the direction is indeterminate) and anticipates that the demand for taxable goods (and therefore the tax base) will be affected by a neighbor's tax rate change. Accordingly, I develop two equations. The first equation uses the home county's tax rate as the dependent variable to measure the policy response to a neighbor's tax rate change and other factors. The second equation measures the demand response by using the home county's sales tax base as the dependent variable. I develop the equations in more detail in the following section.
I base the empirical methodology on the recognition that tax policies are changed infrequently and reflect the long-term fiscal policy choices of a jurisdiction. I test the robustness of the results by using both short-run and long-run models. First, I estimate a traditional long-run model based on the levels of tax rates and bases. Second, I estimate a dynamic short-run adjustment model because local governments need time to adjust their tax rates to policy changes and bases need time to adjust to changing consumption patterns.
The long-run model includes an equation for the local sales tax rate and the local sales tax base. The model allows for the full response in the home county to a neighbor county's tax rate increase without having to account for how fast the home county reacts. I discuss the local rate equation first, then the base equation.
Local Sales Tax Rate Equation
Counties choose their sales tax rate as part of the overall funding decision, but according to the theoretical framework of MT and KK discussed above, counties must also consider the sales tax base, tax rates of neighbors, and their own residents' preferences. Thus, the tax rate decision is modeled by including three groups of variables. First, the home sales tax base and the neighbor's tax rate are included based on the MT and KK theoretical frameworks. Second, the sales tax is one component of the overall fiscal system so the quantity of services to be financed and other revenue options available to local governments can affect the local sales tax rate (Hettich and Winer 1988). Finally, several variables are included to account for residents' preferences of one tax source over another (MT 1986). The decision to change the tax rate is not instantaneous. Therefore, all of the policy variables in the equation have been lagged by one year because of the time lag between when legislators have information and when legislators can implement tax rate changes. (5) The local sales tax rate equation is as follows:
(1) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.]
[LOCAL.sub.it] = local sales tax rate for county i in year t;
[NEIGH.sub.it - 1] = weighted average sales tax rate in adjacent neighboring counties in year t - 1;
[BASE.sub.it - 1] = local sales tax base for county i in year t - 1;
[EXPEND.sub.it - 1] = expenditures in county i in year t - 1;
[PROPERTY.sub.it - 1] = property tax revenues in county i in year t - 1;
[GRANTS.sub.it - 1] = grant revenue in county i in year t - 1;
[STATE.sub.t] = state sales tax rate in year t;
[PREFERENCE.sub.it] = a vector of demographic variables included to measure preferences for sales taxes specific to county i in year t; and
COUNTY = county fixed effects.
The county's local sales tax rate is the dependent variable. See Table 1 for a summary of the local option rates by year.
As previously discussed, a home county's government may react to a neighbor's rate increase by increasing, decreasing, or leaving its own rate unchanged, depending on the residents' preferences for public versus private goods (MT). In Tennessee, each county has between three and nine neighboring counties, so including the tax rates of all neighboring jurisdictions in the model is not practical and may result in multicollinearity. For this study. I define neighbor, NEIGH, using the average sales tax rate for all adjacent neighboring counties, weighted based on population. By using a weighted average, no single county is selected as the primary competitor. (6) Prior research uses varying lags for the neighbor's tax rate. (7) The current study uses a single year lag for NEIGH.
In Tennessee, the two primary available revenue sources for local governments are property taxes and local option sales taxes. So, given a certain level of property tax revenues, the local government depends on the sales tax for the balance of needed total revenues. (Tennessee requires a balanced budget each year.) A county with a lower sales tax base must raise the tax rate to meet its desired expenditure level, all else equal. Therefore, a decrease in the prior period sales tax base is expected to have a positive relation with the current sales tax rate. (8,9)
I include four variables in the equation to account for the quantity of services to be financed and other revenue options. First, the local rate depends on the level of desired public expenditures (MT). The expenditure variable, EXPEND, proxies for the size of government in a county. Although local governments have revenue sources other than sales taxes, a higher level of expenditures should lead to higher sales tax rates if the other revenue sources remain fixed.
Tennessee counties derive significant revenues from property taxes and intergovernmental grants. If revenue from either of these sources increases and expenditures do not, then the revenue required from the sales tax should decrease. Thus, I predict a negative relation between property tax revenues (PROPERTY) and the local sales tax rate and between grants (GRANTS) and the local sales tax rate. The total sales tax rate for a Tennessee county includes the state-level tax and the local option tax. Tennessee raised its state tax rate twice during the study's time period and has a higher state-level sales tax rate than six of its eight neighbors. (10) While county governments can control only the local option sales tax rate, state-level sales tax rates may influence the local tax rate because of changes in cross-border shopping. State rate increases can cause Tennessee residents to shop out of state, potentially causing local governments to want lower rates. On the other hand, the state imposes an upper limit on local sales tax rates, and the limit was historically linked to the state rate. Until 1992, when the state rate increased from 5.5 percent to 6 percent, Tennessee state law restricted the local option rate to one-half of the state rate. Since 1984, the maximum local option rate allowed has remained at 2.75 percent. Thus, higher state rates can be associated with higher local rates. Based on the latter effect, I predict a positive relation between the state tax rate and LOCAL. (11)
I include four variables that represent preferences for sales taxes (PREFERENCE) in the model, based on research by Murphy and Izraeli (1997). Since sales taxes are regressive, counties with higher per capita income (INCOME) levels should prefer sales tax increases to other types of tax increases. Further, an older population with many residents past peak earning years may prefer sales taxes over property taxes since elderly residents typically spend a comparatively high level of their income on untaxed services such as medical care and untaxed prescription drugs. Tennessee requires that one-half of sales tax revenues be spent for education. Therefore, counties with a relatively large percentage of children may prefer more spending for schools (and thus higher sales tax rates). The variables %OLD (defined as the percent of the county's population in the county over age 65) and %YOUNG (defined as the percent of the county's population between the ages 5 and 14) are included in the model to control for these population groups.
KK show that the largest counties should have higher tax rates because they can benefit relatively less from operating as a tax haven. Also, larger counties can attract shoppers because they offer a larger selection of retail outlets. The population variable (POP) is included to capture any competitive advantages large counties have over their smaller neighbors. I include county fixed effects (COUNTY) to account for all other unmeasured county factors that influence the local tax rate.
Local Sales Tax Base Equation
The local sales tax base equation represents a standard consumption function for taxable commodities, where demand is a function of prices, income, and tastes. Tax rates, including the home county state and local tax rates and the neighbor's tax rate, and travel costs represent price. (12) Per capita income levels represent income and four demographic variables represent taste for taxable goods. The long-run local sales tax base equation is as follows:
(2) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.]
[BASE.sub.it] = local sales tax base for county i in year t;
[LOCAL.sub.it] = local sales tax rate for county i in year t;
[NEIGH.sub.it] = weighted average sales tax rate in adjacent neighboring counties in year t;
[TRAVEL.sub.i] = proxy for transportation costs to cross-border shop;
[STATE.sub.t] = state sales tax rate in year t;
[INCOME.sub.it] = per capita income for county i in year t;
[TASTE.sub.it] = a vector of demographic variables included to measure consumption patterns specific to county i in year t; and
COUNTY = county fixed effects.
Price and tax rate changes are expected to influence the consumption relationship. The base equation tests whether the home county and neighboring county sales tax rates, LOCAL and NEIGH, respectively, have a significant effect on the home county's sales tax base. Broadly speaking, sales tax increases by the home county have three possible effects (see Hewett and Stephenson 1993). First, consumers will have less after-tax income, leading to an overall decrease in sales (income effect). Second, consumers will substitute lower-taxed or nontaxed goods for the taxed goods whose price has increased (substitution effect). Finally, consumers will substitute purchases in a low-tax region for goods purchased in their own region if the after tax-cost at home (price + tax) is higher than the after-tax cost (price + tax + travel costs) in the neighboring jurisdictions (cross-border shopping effect). Increases in the home county's tax rate will make retail goods in the home county relatively more expensive on a tax-inclusive basis, making them less desirable and leading to a decrease in the home county tax base (Mikesell 1970, 1971; Fox 1986; Mikesell and Zorn 1986; Walsh and Jones 1988; Ferris 2000). Therefore, I expect a negative relation between the home county sales tax rate and the home county sales tax base.
Similarly, if a neighbor raises its sales tax rate, the tax-inclusive cost of the good in the neighboring jurisdiction increases. Because of the potential to cross-border shop, the higher the neighbor county's relative rate, the more neighboring residents will cross the border to shop in the lower taxing county (home). In addition, home county residents may stay at home to shop because the rates are lower. Therefore, the amount of the taxed good that is purchased at home should increase, causing the home county's sales tax base to increase.
Cross-border shopping is expected to depend not only on the tax differential between two jurisdictions, but also on the cost of traveling to the lower taxed jurisdiction (MT; KK). As the costs of traveling increase, the total cost savings from shopping in a lower taxed jurisdiction decrease, raising the tax differential that must exist before cross-border shopping becomes desirable. Given a tax rate differential, an increase in travel costs should positively affect the home tax base because more of its residents will stay home and shop in the home county (Fox 1986; Walsh and Jones 1988). Because Tennessee's sales tax rates are generally higher than those of neighboring states, the overall effect of an increase in travel costs (TRAVEL) should be an increase in the sales tax base. Fox (1986) uses real automobile costs per mile to proxy for travel costs. Walsh and Jones (1988) use the distance from the county to the nearest across-border commercial district. This study proxies for travel costs by using the product of the roundtrip number of miles between the county seats and the Internal Revenue Service standard mileage rate for the respective year.
I include a variable for the state-level sales tax rate (STATE), in addition to the local option sales tax rate. An increase in the Tennessee state-level sales tax rate should have a negative effect on the sales tax base for Tennessee counties because of the increase in tax-inclusive prices for all goods sold in Tennessee. Therefore, I predict a negative relation between STATE and BASE.
Consumption relationships are also expected to depend on income and residents' tastes. The purchase of taxable commodities in a jurisdiction is expected to increase with per capita income (INCOME) increases. Four variables are included to represent residents' tastes for taxable goods versus other uses of resources. First, an older populous may spend a larger percentage of its income on items not subject to sales tax (e.g. health care, services), and young families may spend more on taxable goods such as food and clothing. In addition, a variable is included to measure whether taxable consumption patterns differ between urban and rural counties, which may be the results of different shopping patterns. Finally, the unemployment rate is included because for any given level of income, unemployed individuals may be more budget-conscious than are workers. I predict that the coefficients for %YOUNG and URBAN will have a positive effect on BASE and that the coefficients for %OLD and UNEMPLOY will have a negative impact on BASE. I include county fixed effects (COUNTY) in the base equation to account for all other unmeasured county factors that influence the local sales tax base.
The sales tax base lagged one period is included in the sales tax rate equation because of the interval between when the economic activity occurs and when policymakers can respond with rate changes. However, the current sales tax base depends on the current tax rate since consumers can react quickly to tax-induced prices changes. Therefore, the tax base is also determined simultaneously with the tax rate. (13) I employ the instrumental variable approach as an alternative econometric technique to test the robustness of the results.
Short-Run Adjustment Model
Changes in equilibrium tax rates and bases may not be immediately realized, but may occur only after multiple years. The study uses a short-run adjustment model to examine the home county's initial responses to changes in the neighbor's tax rate and other factors and to examine the time period necessary for the home county to fully respond to changes. The short-run adjustment model estimates the short-run response of the home tax rate and base to changes in factors in the long-run model such as the neighbor's tax rate. Generally, I expect to find the same relationships in both the long-run and short-run equations. (14)
The short-run adjustment model includes an error correction term (ECT), accounting for the change in behavior that is expected from the extent to which each county's tax rate or base is out of equilibrium at the beginning of the period. (15) The ECT for the tax rate for county i in time period t is given by:
(3) EC[T.sup.i.sub.t] = [LOCAL.sub.it - 1] - [??] - [n summation over (k=1)] [??] [X.sup.k.sub.it - 1].
The error correction term is the difference at each point in time between the actual tax rate (the first term on the right-hand side of Equation (3)) and the desired (or predicted) tax rate (the second and third terms in Equation (3)). This is based on the recognition that counties change tax rates in discrete steps and that responses to changes in independent variables may not occur immediately. The coefficient on the ECT, [lambda], measures the rate at which disequilibrium (the right-hand side of Equation (3)) is reduced in each time period and can be used to estimate the number of periods necessary to eliminate the disequilibrium. The sign must be negative, indicating that the home tax rate and base are converging to an equilibrium. For example, the further a county's tax rate is above (below) the desired tax rate, the more the county will choose to decrease (increase) its current tax rate during the next period. I expect the coefficient for ECT to be smaller for tax rate equation than for the tax base equation, meaning policy will change more slowly than the location of sales. The general form of this short-run model for the sales tax rate and base is as follows:
(4) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.]
(5) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.]
[DELTA][LOCAL.sub.1] = the first difference of the local option sales tax rate in county i;
[DELTA][BASE.sub.i] = the first difference of the local sales tax base in county i;
[DELTA][NEIGH.sub.i] = the first difference of the neighboring county's sales tax rate for county i;
[X.sub.i] = a vector of all other variables for the local sales tax rate equilibrium model for county i;
[Z.sub.i] = a vector of all other variables for the local sales tax base equilibrium model for county i; and
EC[T.sub.i] (error correction term) = the amount of "disequilibrium" in the prior period t - 1.
The empirical model is estimated for all 95 counties in Tennessee for the 17-year period beginning with 1977. Table 2 presents descriptive statistics for variables in both equations. I obtained state and local sales tax rate information directly from Tennessee and its eight contiguous state governments. I gathered information for the local sales tax base from the Tennessee Department of Revenue, Monthly Statement of Revenue Collections, Collection Report by Counties--Local Sales Tax. In addition, I obtained Tennessee county revenue, including expenditures, property tax revenues, and federal and state grants, from County and Municipal Finances, Comptroller of the Treasury, Division of Local Finance. The sales tax base, expenditures, property tax revenues, grants, and per capita income are adjusted for inflation by dividing the nominal amount by the price index for gross domestic product, obtained from Wharton Econometric Forecasting Associates (WEFA). To reduce problems of heteroscedasticity, I divided these variables by the population for the respective county.
I obtained demographic information from a variety of sources. The U.S Department of Commerce, Bureau of Economic Analysis reports personal income annually. I obtained county population estimates, population by age, and the percent of urban population from the U.S. Department of Commerce, Bureau of the Census, Census of Population and Housing. (16) Finally, the Tennessee Department of Employment Security provided the annual unemployment rate.
I discuss the results for the long-run model first for both the sales tax rate and base equations. Only an OLS result is presented for the rate equation because all of the right-hand side variables have been lagged, precluding any requirement to consider simultaneity. Next, I present results for the short-run model. Tables 3 and 4 report the long-run results and Tables 5 and 6 report the short-run results.
Local Sales Tax Rate Equation
The coefficient for NEIGH is positive and highly significant, indicating that Tennessee local governments raise sales tax rates in response to tax increases by their neighbors. (17) Local governments may find that a neighbor's rate increase makes it politically easier to raise tax rates at home. Also, the results are consistent with the notion that Tennessee residents have a strong preference for public goods, as described by MT. The coefficient on NEIGH in the rate equation, which is less than 1, indicates that the average Tennessee county increases its rate less than that of its neighbor. The increasing reaction function with a coefficient below 1 is consistent with the findings of both KK's and MT's models, and is also consistent with the empirical results of Case (1993).
The coefficient for BASE is positive and therefore contrary to expectations, but it is not statistically significant. One possible explanation is that as the sales tax base declines, local governments look to other sources for additional revenues such as increasing property taxes or obtaining additional grants. Alternatively, local governments can reduce spending levels. The signs for expenditures, grants, and property tax revenues are all in the predicted direction and the coefficients for these variables are all statistically significant at a p-value < .001. These results, combined with the insignificant sign for BASE, support the notion that local governments must find a balance between increasing expenditure levels and tax rates. Furthermore, as other major sources of revenues decline, local governments must increase sales tax rates to fund a desired level of services to their residents.
The sign for STATE is positive and highly significant, supporting the idea that when the state raises its sales tax rate, local governments that previously were taxing at the statutory maximum will increase their rates. Finally, three of the four preference variables are statistically significant in the long run. (See Table 3.)
Local Sales Tax Base Equation
Table 4 presents the results for the local sales tax base equilibrium model. Because the OLS and 2SLS regression results are very similar, I discuss them together. The signs on the tax rate coefficients are consistent with the theoretical expectations. For both econometric techniques, the coefficient for LOCAL is negative, but it is only statistically significant using 2SLS. The OLS result suggests that, over time, consumers will settle back to their previous shopping patterns. However, the 2SLS results show that the local sales tax rate has a negative, significant effect on the local sales tax base, even in the long run. This indicates that a tax increase makes local goods more expensive and leads residents to crossborder shop in neighboring jurisdictions.
The coefficient for NEIGH is positive and highly significant for both OLS and 2SLS. The results indicate that a 1 percent increase in the neighbor's tax rate will cause the home county's tax base to increase by approximately $2.20-$3.77 per resident. In addition, an increase in the state sales tax rate has a significant negative effect on the sales tax base (though only significant in the OLS equation). This finding is important for state policymakers, as an increase in the state sales tax rate appears to be encouraging various types of cross-border shopping (e.g., crossing state lines, mail-order).
Contrary to expectations, TRAVEL is negative and significant for both OLS and 2SLS, signaling that shoppers may not recognize the travel costs associated with cross-border shopping. Furthermore, prior research has shown that travel costs are difficult to measure accurately, and the variable TRAVEL may not be adequately capturing all relevant costs (e.g. time opportunity costs). As expected, the coefficient for INCOME is positive and highly significant, indicating that residents purchase more taxable commodities with higher levels of income. Finally, all of the coefficients for the taste variables are statistically significant for the OLS model, and three of four of the variables are significant for the 2SLS model.
Short-Run Adjustment Model Local Sales Tax Rate Equation
Table 5 reports the short-run results for the local sales tax rate model. The results are generally consistent with the long-run model, evidencing the robustness of the findings to the specification. In the short run, [DELTA]NEIGH has a positive and significant effect on changes in the local tax rate. As expected, the short-run coefficient for [DELTA]NEIGH is smaller than the long-run NEIGH coefficient, indicating that the home county only responds partly to a neighbor's tax rate increase in the current period.
A change in the short-term base has a strong and significant effect on a change in the local sales tax rate, as expected. In other words, if the sales tax base declines, then governments raise rates in the short run to make up the revenue losses. One possible explanation for the difference between the long-run and short-run results is that a cyclical decline in the base, which might occur during a recession, is an important determinant of rate increases during the cycle but does not influence the long-term choice of rates.
Consistent with the long-run results, the sign on [DELTA]STATE is positive. The coefficients for [DELTA]PROPERTY and [DELTA]GRANTS both have the expected negative signs, but only [DELTA]PROPERTY is statistically significant, indicating that short-run sales tax rate increases are influenced by decreases in property tax revenues. The sign on [DELTA]EXPEND is contrary to expectations, but is not statistically significant. These results show that current period changes in expenditure levels do not lead to current period tax rate increases. In the long run, however, higher spending drives higher tax rates, as described above.
As expected, the coefficient for ECT is negative and highly significant. The adjustment parameter, [lambda], is the asymptotic response of the county's actual tax rate to the difference between the actual and desired rate at the beginning of the year. The result shows that counties are slow to respond to changes in the conditions they face, such as a rate increase by a neighboring county. It indicates that 23.1 percent of the disequilibrium in a county's tax rate is eliminated in year one and 73.1 percent is eliminated by year five. (18)
Local Sales Tax Base Equation
Table 6 presents the short-run results for the sales tax base model. A change in the local sales tax rate has a negative and significant effect on the sales tax base in the short run, which means that there is an initial increase in cross-border shopping and/or a decrease in consumption of taxable goods. When the home county raises its sales tax rate, the coefficient is smaller than the long-run 2SLS coefficient, indicating that it takes some time for consumers to fully respond to the rate increases. The sign on [DELTA]NEIGH is positive and highly significant, and again with a smaller coefficient than in the long-run model. This result provides further evidence that cross-border shopping into the home county increases in response to tax rate hikes by neighbors.
Similar to the rate equation, the coefficient for ECT is negative and significant in the base equation. As expected, [lambda] is larger for short-run base model, indicating that bases adjust more rapidly than tax rates. In other words, political bodies respond slower in the setting of tax rates than consumers do in changing their shopping behavior. The reduction in the disequilibrium is 29.4 percent in year one and 82.4 percent by year five.
SUMMARY, LIMITATIONS, AND EXTENSIONS
This study examines how local jurisdictions select sales tax rates when their choice may be affected by the amount of cross-border shopping and bow their tax bases respond to the rate changes. I find that the sales tax rates of the home and competing counties affect the sales tax base because people cross-border shop to take advantage of lower tax rates. In addition, the results show that local governments consider sales tax rates of neighboring counties when setting their own rates, both in the short and long run.
Another important finding is that an increase in the state-level sales tax rate is associated with a reduction in the sales tax base for Tennessee counties in the long run. Declines in the state sales tax base place pressure on the state to raise rates, and the resulting reduction in the base caused by higher state rates puts additional pressure on local governments to raise rates. The sales tax base in Tennessee has eroded over time relative to the economy, with untaxed services and prescription drugs comprising an increasing proportion of economic activity. (19) Furthermore, Tennessee's relatively higher sales taxes and the emergence of mail-order shopping (and now Internet sales) may have increased cross-border shopping and caused further erosion in the sales tax base.
This research has a number of limitations. Because Tennessee relies heavily on sales taxes and has no individual income taxes, further research should be done to see whether similar results are found in states with different predominating tax structures (e.g., income tax, property tax). A second limitation is that the model may not adequately capture all factors that may affect changes in the sales tax base. For example, politicians can legislate other policies that have a direct influence on the sales tax base (e.g., zoning and tax administration laws), or commuting patterns may affect the sales tax base. Similarly, tax rate increases may impose an important political component that the model cannot adequately capture for such a large sample (Case 1993; Merriman 2000). A final limitation is that the model does not try to capture the government's motivations. Theory predicts that a government's decision to raise or lower rates depends on whether the locality's goal is to maximize revenue or maximize utility of its citizens. Because these assumptions are not observable, further research is warranted to understand governments' motivations.
TABLE 1 Tennessee Local Option Sales Tax Rates Summary of 95 Counties 1977-1993 Rates Year 0.00 .75 1.00 1.50 1.75 2.00 2.25 2.50 2.75 1977 4 0 18 57 8 2 6 0 0 1978 3 0 13 56 7 6 10 0 0 1979 3 0 11 53 7 8 13 0 0 1980 2 0 9 54 7 9 14 0 0 1981 1 0 8 46 7 10 23 0 0 1982 1 0 7 41 7 11 28 0 0 1983 1 1 4 34 7 12 37 0 0 1984 0 1 3 25 5 12 49 0 0 1985 0 1 2 22 5 12 53 0 0 1986 0 1 2 21 5 13 53 0 0 1987 0 0 1 18 5 14 54 2 1 1988 0 0 1 17 5 13 54 3 2 1989 0 0 1 14 5 12 55 5 3 1990 0 0 1 12 5 12 55 5 5 1991 0 0 1 11 5 12 55 5 6 1992 0 0 1 10 5 12 54 7 6 1993 0 0 1 9 5 12 54 8 6 TABLE 2 (b) Change Variables: [DELTA]BASE = change in local sales tax base; [DELTA]EXPEND = change in per capita local expenditures; [DELTA]GRANTS = change in per capita grant revenues; [DELTA]INCOME = change in per capita income; [DELTA]LOCAL = change in local option sales tax rate; [DELTA]NEIGH = change in the neighbor's sales tax rate; [DELTA]%OLD = change in percent of population over age 65; [DELTA]%YOUNG = change in percent of population between ages 5-14; [DELTA]POP = change in population; [DELTA]PROPERTY = change in per capita property tax revenues; [DELTA]STATE = change in state tax rate; [DELTA]TRAVEL = travel cost between the home county and the neighboring county; [DELTA]UNEMPLOY = change in percent of population that is unemployed; and [DELTA]URBAN = change in percent of county that is urban. TABLE 3 Equation (1): OLS Regression of Local Sales Tax Rate Long-Run Model (n = 1,520) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.] Independent Predicted Parameter Variables (a) Sign (t-statistic) NEIGH ? 0.16 (7.10) *** BASE - .00 (1.18) EXPEND + .03 (3.27) *** PROPERTY - -.08 (-2.94) *** GRANTS - -.10 (-5.36) *** STATE + .17 (6.01) *** Preference Variables INCOME + -.00 (-.94) %OLD + .02 (1.87) * %YOUNG + -.02 (-1.91) * POP + .00 (2.22) ** ***, **, * Represent significance at p < .01, .05, and .10, respectively. [R.sup.2] = .61 (a) Variables: LOCAL = local option sales tax rate; NEIGH = neighbor's sales tax rate; BASE = per capita local sales tax base adjusted for inflation; EXPEND = per capita local expenditures adjusted for inflation; PROPERTY = per capita property tax revenues adjusted for inflation; GRANTS = per capita grant revenues adjusted for inflation; STATE = state tax rate; INCOME = per capita income adjusted for inflation; %OLD = percent of population over age 65; %YOUNG = percent of population between ages 5-14; POP = population/ 1000; and COUNTY = dummy variables representing each Tennessee county. TABLE 4 Equation (2): OLS and 2SLS Regression of Local Sales Tax Base Long-Run Model (n = 1,615) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.] Independent Predicted Parameter (t-statistic) Variables (a) Sign OLS 2SLS LOCAL - -.34 -12.84 (-.60) (-3.03) *** NEIGH + 2.20 3.77 (3.36) *** (4.05) *** TRAVEL + -1.36 -.89 (-3.62) *** (-2.01) ** STATE - -2.08 -.32 (-2.70) *** (-.30) INCOME + .22 .23 (9.32) *** (8.24) *** Taste Variables %OLD - 2.36 2.88 (9.30) *** (8.33) *** %YOUNG + 1.35 1.37 (4.61) *** (4.02) *** URBAN + .60 .77 (8.26) *** (8.00) *** UNEMPLOY - -.25 -.04 (-4.45) *** (-.46) ***, **, * Represent significance at p < .01. .05, and .10, respectively. [R.sup.2] = .90 (a) Variables: BASE = local sales tax base; LOCAL = local option sales tax rate; NEIGH = neighbor's sales tax rate; TRAVEL = travel cost between the home county and the neighboring county; STATE = state tax rate; INCOME = per capita income adjusted for inflation; %OLD = percent of population over age 65; %YOUNG = percent of population between ages 5-14; URBAN = percent of county that is urban; UNEMPLOY = percent of population that is unemployed; and COUNTY = dummy variables representing each Tennessee county. TABLE 5 Equation (1): OLS Regression of Change in Local Sales Tax Rate Short-Run Model (n = 1,425) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.] Independent Predicted Parameter Variables (a) Sign (t-statistic) NEIGH ? .04 (2.37) ** BASE - -0.01 (-13.91) *** EXPEND + -.00 (-.12) PROPERTY - -.05 (-2.06) ** GRANTS - -.00 (.27) STATE + .08 (4.34) *** Preference Variables INCOME + .00 (.82) %OLD + .02 (2.45) ** %YOUNG + .00 (.32) POP + -.01 (-2.79) *** ECT - -.23 (-13.45) *** ***, **, * Represent significance at p < .01, .05, and .10, respectively. [R.sup.2] = .29 (a) Variables: [DELTA]LOCAL = change in local sales tax rate; [DELTA]NEIGH = change in the neighbors sales tax rate; [DELTA]BASE = change in per capita local option sales tax base; [DELTA]EXPEND = change in the per capita local expenditures; [DELTA]PROPERTY = change in per capita property tax revenues; [DELTA]GRANTS = change in per capita grant revenues; [DELTA]STATE = change in the state tax rate; [DELTA]INCOME = change in per capita income; [DELTA]%OLD = change in percent of population over age 65; [DELTA]%YOUNG = change in percent of population between the ages 5-14; [DELTA]POP = change in population: ECT = amount of disequilibrium in the prior period t-1; and COUNTY = dummy variables representing each Tennessee county. TABLE 6 Equation (2): OLS Regression of Change in Local Sales Tax Base Short-Run Model (n = 1,520) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.] Independent Predicted Parameter Variables (a) Sign (t-statistic) [DELTA]LOCAL - -9.44 (-15.76) *** [DELTA]NEIGH + 1.64 (3.96) *** TRAVEL + -.01 (-.11) [DELTA]STATE - .88 (1.88) * [DELTA]INCOME + .05 (2.37) ** Taste Variables [DELTA]%OLD - 1.92 (8.38) *** [DELTA]%YOUNG + 2.16 (6.46) *** [DELTA]URBAN + .20 (.82) [DELTA]UNEMPLOY - -.20 (-4.63) *** ECT - -.29 (-16.98) *** ***, **, * Represent significance at p < .01, .05, and .10, respectively. [R.sup.2] = .39 (a) Variables: [DELTA]BASE = change in local sales tax base; [DELTA]LOCAL = change in local option sales tax rate; [DELTA]NEIGH = change in the neighbor's sales tax rate; TRAVEL = travel cost between the home county and the neighboring county; [DELTA]STATE = change in the state tax rate; [DELTA]INCOME = change in per capita income; [DELTA]%OLD = change in percent of population over age 65; [DELTA]%YOUNG = change in percent of population between ages 5-14; [DELTA]URBAN = change in percent of county that is urban; [DELTA]UNEMPLOY = change in percent of population that is unemployed; ECT = amount of disequilibrium in the prior period t-1; and COUNTY = dummy variables representing each Tennessee county.
This paper is based on my dissertation at the University of Tennessee. I am indebted to my committee, Ken Anderson, Bruce Behn, Dan Murphy, and especially Bill Fox for their guidance. This research has also benefited from Ron Fisher and the 2000 National Tax Association's annual conference participants, Editors Fran Ayres and John Robinson, and two anonymous reviewers. I also thank Raquel Alexander, Dan Ivancevich, Sue Porter, Joanne Rockness, Tim Rupert, Rebecca Sawyer, and the JATA editorial consultant for their editorial assistance.
(1) Forrester Research estimates total online retail purchases of $1.91 trillion by 2003.
(2) In 2003, e-commerce is estimated to cause about $10.8 billion in additional tax revenue losses nationwide (Bruce and Fox 2000).
(3) The use tax is intended to enforce destination taxation. Unfortunately, use tax laws are routinely ignored, particularly by individuals, and are virtually impossible to enforce. Further, use taxes are not levied at the local level in many states. Therefore, without border controls or honest reporting of purchases outside one's jurisdiction, administration of the retail sales/use tax on a destination principle is very difficult.
(4) The home county will lose some revenues from less cross-border shopping if it raises its rate but overall will gain more from its own residents (KK 1993).
(5) Of course, for planning purposes a growth factor could be applied to the variables, but this will not affect the estimated coefficients if all local governments assume the same growth rate.
(6) Other definitions of neighbor are tested and discussed in the results section.
(7) Case (1993) uses a two-year lag. Hewett and Stephenson (1983) find that three years provided the best results. I test various lags of NEIGH ranging from 1 to 4 and note no significant differences.
(8) State law determines the statutory definition of the tax base, and the base is the same for all Tennessee counties. The Tennessee Department of Revenue reports the local sales tax revenues and local option sales tax rates, but not the base. For this study, I estimate the base by dividing local sales tax revenue for each county by its respective sales tax rate.
(9) The frequent propensity to raise sales tax rates during the recession and weak economic environment of 1980 through 1984 is strong anecdotal evidence of this relationship.
(10) Kentucky imposes the same 6 percent state tax rate as Tennessee, and Mississippi imposes a 7 percent state tax rate; however, neither state has a local option tax. The other six neighboring states have lower state-level sales tax rates.
(11) A zero or negative coefficient for STATE is also possible if the state and local governments are competing for the same sales tax base and sales tax revenue.
(12) The net of tax price of goods is assumed to be constant across counties and therefore is not included on the right-hand side of the equation. See Besley and Rosen (1999) for further discussion of this assumption.
(13) The Hausman test was used and shows that simultaneity exits between LOCAL and BASE, LOCAL and EXPEND, LOCAL and GRANTS, and LOCAL and PROPERTY. However, no simultaneity exists when lagged values are used, so it is not necessary to estimate simultaneity for the LOCAL equation
(14) I expect the coefficients to be smaller for the short-run model because the full effect of the changes will not be fully realized. Because it is very difficult to establish precise statistical tests of the differences between the long-run and short-run coefficients, I only describe the differences.
(15) The development of error correction models is discussed in Charemza and Deadman (1997) and Patterson (2000).
(16) Because census data are only collected every ten years, some population data are interpolated.
(17) Two other definitions of neighbor are also tested. First, NEIGH is defined as the adjacent county with the lowest total sales tax rate and the shortest travel distance. Also, NEIGH is defined using only the neighbor's local option sales tax rate. The model performs best when neighbor is defined using a weighted average.
(18) [D.sub.t] = 1 - [((1 + [lambda]).sup.t]) where [D.sub.t] equals the amount of disequilibrium that has been eliminated at time t. See Davis and Palumbo (2001).
(19) States' failure to tax services has led to one-eighth of the increase in the service sector during the 1982-1992 period (Merriman and Skidmore 2000).
Besley, T. J., and H. S. Rosen. 1999. Sales taxes and prices: An empirical analysis National Tax Journal (June): 157-178.
Bruce, D., and W. F. Fox. 2000. E-commerce in the context of declining state sales tax bases. National Tax Journal (December): 1373-1388.
Case, A., I. R. Hines, and H. S. Rosen. 1989. Copycatting: Fiscal policies of states and their neighbors. National Bureau of Economic Research Working Paper No. 3032. Cambridge, MA: NBER.
--. 1993. Interstate tax competition after TRA86. Journal of Policy Analysis and Management (Winter): 136-148.
Charemza, W. W., and D. F. Deadman. 1997. New Directions in Econometric Practice. New York, NY: Edward Elgar.
Davis, M., and M. Palumbo. 2001. A primer on the economics and time series econometrics of wealth effects. Finance and Economics Discussion Series. Washington, D.C.: Board of Governors of the Federal Reserve System.
Ferris, J. S. 2000. The determinants of cross-border shopping: Implications for tax revenues and institutional change. National Tax Journal (December): 801-824.
Fisher, R. C. 1980. Local sales taxes: Tax rate differentials, sales loss, and revenue estimation Public Finance Quarterly (April): 171-188.
Fox, W. F. 1986. Tax structure and the location of economic activity along state borders. National Tax Journal (December): 387-401.
Hettich, W., and S. L. Winer. 1988. Economic and political foundations of tax structure. American Economic Review: 701-712.
Hewett, R. S., and S. C. Stephenson. 1983. State tax revenues under competition. National Tax Journal (March): 95-101.
Kanbur, R., and M. Keen. 1993. Jeux sans fontieres: Tax competition and tax coordination when countries differ in size. The American Economic Review (September): 877-892.
Merriman, D. 2000. When do states legislate tax changes? Working paper, presented at the 2000 National Tax Association Annual Meeting, Santa Fe, NM.
--, and M. Skidmore. 2000. Did distortionary sales taxation contribute to the growth of the service sector? National Tax Journal (March): 125-142.
Mikesell, J. L. 1970. Central cities and the sales tax rate differentials: The border city problem National Tax Journal (June): 206-214.
--. 1971. Sales taxation and the border county problem. Quarterly Review of Economics and Business (Spring): 23-29.
--, and K. Zorn. 1986. Impact of the sales tax rate on its base: Evidence from a small town. Public Finance Quarterly (July): 329-338.
Mintz, J., and H. Tulkens. 1986. Commodity tax competition between member states of a federation: Equilibrium and efficiency. Journal of Public Economics (March): 133-172.
Murphy, K. J., and O. Izraeli. 1997. Interstate differences in per capita state and local revenues and the neighboring state effect. Review of Regional Studies (Fall): 101-121.
Patterson, K. D. 2000. An Introduction to Applied Econometrics: A Time Series Approach. New York, NY: St. Martin's Press.
Rork, J. C. 2000. Coveting thy neighbors' tax rates. Working paper, presented at the 2000 National Tax Association Annual Meeting, Santa Fe, NM.
Walsh, M. J., and J. D. Jones. 1988. More evidence on the "border tax" effect: The case of West Virginia. National Tax Journal (June): 261-265.
Wildasin, D. E. 1988. Nash equilibria in models of fiscal competition. Journal of Public Economics (March): 229-240.
LeAnn Luna is an Assistant Professor at The University of North Carolina at Wilmington.
Submitted: March 2001
Accepted: April 2003
|Printer friendly Cite/link Email Feedback|
|Publication:||Journal of the American Taxation Association|
|Date:||Mar 22, 2004|
|Previous Article:||Does representation matter in IRS office audits?|
|Next Article:||The influence of biased tax research memoranda on supervisors' initial judgments in the review process.|