BORDER EFFECTS IN SUBURBAN LAND USE

(1)

BORDER EFFECTS IN SUBURBAN LAND USE

Benoy Jacob and Daniel McMillen

Differences in property tax rates and sales tax distribution formulas can give sub-urban municipalities strong incentives to attract commercial and industrial firms to their jurisdictions. If these land uses generate negative externalities for local residents, suburban governments may form concentrations of non-residential land use at the borders of neighboring suburbs. We test this prediction using land use data for every parcel in the Cook County suburbs of Chicago. After controlling for proximity to major streets and rail lines, both commercial and industrial parcels are significantly more likely to be located near municipal boundaries. Low-priced industrial properties are also significantly more likely to be located near municipal boundaries than in the interior of a suburb.

Keywords: land use patterns, state and local taxation, border effects JEL Codes: R14, H71

I. INTRODUCTION

T

he structure of a region’s tax system has the potential to directly affect local land

use decisions. In many jurisdictions, commercial and industrial property is either taxed or assessed at higher rates than residential property. Even if tax rates are identical across land uses, adding non-residential land can lower the average property tax burden for local residents if businesses consume less in services than they pay in taxes. In some places, non-residential establishments offer additional fiscal benefits when local jurisdictions receive a portion of the sales and income revenues generated by businesses located within their borders. However, non-residential land uses may also generate significant negative externalities for local residents. Few residential suburbs allow any but minor industrial properties in their jurisdictions, and commercial establishments can exacerbate existing traffic problems.

The idea that land use decisions are influenced by fiscal incentives is referred to as “fiscalization” in the urban planning literature. Its roots in the economics literature go Benoy Jacob: Center for Local Government Research and Training, Buechner Institute for Governance, University of Colorado-Denver, School of Public Affairs, Denver, CO, USA ([email protected]) Daniel McMillen: Department of Economics, University of Illinois, Urbana, IL, USA ([email protected])

(2)

back at least as far as Fischel (1975, 1976), although the term itself appears to have its origins in Misczynski (1986). Significant empirical studies include Chapman (1988, 2008), Lewis (2001), and Wassmer (2002). This literature places less emphasis on the location of non-residential land uses within a community than on the allocation of

businesses across communities.1_{Yet not all locations are equal if non-residential firms}

generate negative externalities for local residents. Siting commercial and industrial parks at the municipal boundary can allow suburbs to reap the fiscal benefits of non-residential land use while avoiding a substantial portion of the harmful effects. Of course, neighboring jurisdictions may not be happy with the resulting land use pattern, but the responsibility for zoning decisions most commonly lies with municipalities rather than regions in the United States.

Our objective in this paper is to test whether non-residential land uses are more likely to be located near municipal boundaries in the Cook County suburbs of Chicago. Cook County is a particularly relevant area to analyze because it has an exceptionally large number of suburban jurisdictions, which rely heavily on the property tax for revenue. Cook County has a classification system of property taxation that leads to significantly higher effective tax rates for commercial and industrial property than are incurred by homeowners. Moreover, a full percentage point of the state’s 6.25 percent sales tax rate is returned to the jurisdiction in which a sale takes place. Together, these features of the region’s tax system provide strong incentives for suburbs to attract non-residential land uses to their jurisdictions.

We estimate multinomial logit models of land use using a data set comprising every parcel within the suburban portion of Cook County. Small (6 units or fewer) residential units form the base land use, with commercial and industrial uses as alternatives. As might be expected, the probability of commercial land use is significantly higher near major streets, while industrial land use is also attracted to sites near rail lines. After controlling for access to the transportation network, both commercial and industrial parcels are significantly more likely to be located near municipal boundaries. The magnitude of this border effect is unaltered by the inclusion of fixed effects for the 91 suburbs included in our sample.

We also test whether property values exhibit discontinuities at municipal borders. If suburban jurisdictions attempt to mitigate the effect of negative externalities by placing offending land uses at its borders, then low-value property may be more likely to be located near borders than within the interior of the municipalities. Using data on the assessed values of all properties, we find that industrial property values are significantly lower near suburban city borders than in the interior. Smooth cubic spline estimates suggest that residential property values are not influenced greatly by distance to the municipal boundary, while the value of both commercial and industrial property declines dramatically near borders between suburbs.

1 _{An exception is Bowman and Pagano (2004), which considers different spatial models that are based on}

(3)

II. TAXATION IN COOK COUNTY

Local governments in Illinois rely heavily on the property tax for revenue. According to the Chicago Metropolitan Agency for Planning (CMAP), the 1,226 governments in the seven-county Chicago metropolitan area together have approximately $40 billion in annual revenue, with the property tax accounting for 30 percent of the total (CMAP, 2010). State intergovernmental revenues account for another 20 percent of local rev-enue in the region, while charges and user fees account for another 15 percent. Major intergovernmental revenues include the local portion of the state sales tax, income tax, and the motor fuel tax.

Several features of the Cook County tax system provide local governments with an

incentive to attract commercial and industrial establishments to their jurisdictions.2

The Illinois constitution permits counties with more than 200,000 residents to adopt a classification system in which different types of property are assessed at different rates. Only Cook County has adopted this classification; whereas the other 101 counties assess all properties at 1/3 of market value, commercial and industrial properties are assessed at significantly higher rates than residential properties in Cook County. Prior to 2009, commercial and industrial properties were supposed to be assessed at 38 percent and 36 percent of market value, respectively. In contrast, Class 2 properties (residential build-ings with 6 or fewer units) were supposed to be assessed at 16 percent of market value. Since 2009, statutory assessment rates have been 25 percent for commercial/industrial properties and 10 percent for Class 2. Although assessment ratio studies conducted by the Illinois Department of Revenue suggest that actual assessment rates were much lower than statutory rates prior to 2009, rates have always been much higher for non-residential than for Class 2 properties. The Illinois constitution requires equal tax rates for all properties, which means that effective tax rates are much higher for properties that are assessed at higher rates.

A simple example demonstrates the magnitudes of the difference in taxes for commer-cial/industrial and Class 2 properties. If both properties have market values of $100,000, the assessed values will be $25,000 for the non-residential property and $10,000 for the Class 2 property. There are, however, several complicating factors. First, the Illinois constitution requires that the total assessed value of all properties in a county be 1/3 of market value. Since this figure is nearly mathematically impossible when the statutory rate for the vast majority of properties is under 1/3, an “equalization factor” is applied to assessed values to bring them up to the equalized rate. In 2012, the equalization fac-tor for Cook County was 2.8056. The equalized assessed value of the non-residential property is $70,140 (i.e., $100,000 × 0.25 × 2.8056), compared with $28,056 for the Class 2 property. Next, a homestead exemption applies only to residential properties. The exemption was $7,000 in 2012, which results in an “adjusted equalized value” of $21,056 for an accurately assessed Class 2 property. When the same tax rate is then applied to the representative properties, the resulting tax bill is 3.33 (i.e., 70140/21056)

(4)

times higher for the non-residential property owner. This property tax differential provides local governments with a strong incentive to attract non-residential land uses.

The structure of the sales tax system in Illinois also gives local governments an incentive to attract retail establishments. One percentage point of the state’s 6.25 per-cent sales tax is returned to the jurisdiction in which a general merchandise sale takes place. Moreover, 154 municipalities in the Chicago metropolitan area have local sales taxes (CMAP, 2014).

The incentives for locating non-residential uses near jurisdictional boundaries may differ for commercial and industrial properties. Offices and retail establishments are likely to generate significantly fewer negative externalities than industrial properties, with traffic being the greatest offender. A jurisdiction gets the full local portion of the sales tax revenue from any store within its border. Stores located at the border may have the added advantage of attracting customers from nearby jurisdictions. Industrial properties generate no sales tax revenue, but do offer potential property tax advantages. As the land use that is most likely to generate negative externalities, there is a strong incentive for industrial firms to be concentrated at a jurisdiction’s borders.

III. DATA

The main data source for the land use model is the Cook County Assessor’s Office, which provided the full assessment roll for the county for 2003. The file indicates the assessment class for every parcel in the county. Of the approximately 1.7 million parcels in Cook County, just over 1,406,816 are identified as Class 2 properties and 94,532 are listed as Class 5, which represents the combination of commercial and industrial properties. Approximately 57 percent of the Class 2 properties and 49 percent of the

Class 5 properties are in the suburbs. A shapefile indicates the location of each parcel.3

Approximately 78 percent (1,096,520) of the Class 2 properties and 89 percent of the Class 5 properties (84,455) were successfully geocoded using the parcel map. Of these, 523,087 of the residential parcels and 36,511 of the non-residential properties are located in municipalities that have at least 10 parcels in commercial and 10 parcels in industrial use. This final sample restriction excludes unincorporated areas from the analysis, along with a group of 35 suburbs that have virtually no non-residential land

use – a necessary requirement for estimating a multinomial model of land use.4_The

locations of the included and omitted municipalities are shown in Figure 1.

3 _{The shapefile is available from the Cook County web site, https://datacatalog.cookcountyil.gov/GIS-Maps/}

ccgisdata-Parcel-2012/e62c-6rz8.

4 _{The omitted suburbs are Bartlett, Bensenville, Buffalo Grove, Burbank, Country Club Hills, Deerfield,}

Flossmoor, Ford Heights, Glencoe, Golf, Hanover Park, Hinsdale, Hometown, Homewood, Indian Head Park, Inverness, Kenilworth, La Grange Park, Merrionette Park, North Riverside, Olympia Fields, Orland Hills, Palos Heights, Palos Park, Park Ridge, Phoenix, Richton Park, River Forest, Riverside, Roselle, South Barrington, Western Springs, Wilmette, Winnetka, and Worth.

(5)

The underlying estimating equation for the models is,

(1) y_is= X_iβ δ+ +_s u ._is

Separate equations apply for residential, commercial, and industrial properties. For the

logit models of land use, y_is represents the underlying propensity for a parcel i in

loca-tion s to be in the given land use. The explanatory variables, X, do not vary across land

use. We present estimates with and without the fixed effects for location, d_s, which is

taken to be the municipality.

Access to the transportation network is a critical factor in the location decisions of commercial and industrial firms. Thus, we expect proximity to major streets and highways to significantly increase the probability of both commercial and industrial

land use.5_{Although highways are easily defined as roads with U.S. or state numbers,}

Figure 1

Cook County Census Places

(6)

“major” roads are less easily defined a priori. We rely on the Cook County government’s

definition of a major road to define this variable.6

A potential concern is that the location of major roads may be endogenous if one of the factors determining a road’s classification as “major” is the number of businesses located along the road. As a check for endogeneity, we use quarter section boundaries as an instrument for major roads. A quarter section is a one-half by one-half mile square comprising 640 acres. The original official governmental surveys of land in Illinois divided the state into quarter sections, and the federal government subsequently sold land to the original settlers using these units to demark the land. The grid street structure followed by much of Cook County generally follows quarter section boundaries, with larger streets occurring every ½ mile. Thus, distance from a quarter section boundary is an excellent instrument for distance from a major street.

Our final set of parcel-level variables controls for proximity to rail lines, which is likely to be a disamenity for residential landowners but an important attraction for industrial firms. Although homeowners may prefer not to live near a rail line, access to the commuter rail network (the “Metra”) is valuable. Thus, we expect that the prob-ability of commercial land use will be higher closer to Metra stations.

Descriptive statistics for all variables are presented in Table 1. The key variable of interest is proximity to the municipal boundary. The question is whether the probability of commercial and industrial land use is higher at the edge of a municipality, even after controlling for locations along major roads and other determinants of land use decisions. Table 1 also includes data on assessed values for 2003. Although assessments are not necessarily good predictors of sales prices, they are the relevant variable from a fiscal perspective because they form the property tax base. The assessments reported in Table 1 are not market values. Given assessment levels in 2003, commercial and industrial assessments are somewhere in the neighborhood of 25 percent of market value, while

Class 2 assessments are closer to 10 percent of market value.7

Figures 2–7 present kernel density estimates for the major explanatory variables. The estimates are constructed using Silverman’s (1986) reflection method to impose zero densities at negative values of distances. To provide a basis of comparison, we also present density estimates for comparable data for the City of Chicago. The Chicago data set includes 483,043 parcels, of which 439,799 (91.0 percent) are residential, 31,115 (6.4 percent) are commercial, and 12,129 (2.5 percent) are industrial.

Figure 2 shows kernel density estimates for distance from a municipality boundary. Distances can be much larger in Chicago than in the suburbs simply because the city is much larger than suburban municipalities. Whereas residential land use is somewhat more common than commercial or industrial land on the Chicago side of the municipal-ity boundary, the number of industrial parcels rises sharply near cmunicipal-ity lines in suburban

6 _{The shapefile showing major roads in Cook County is available from https://datacatalog.cookcountyil.}

gov/GIS-Maps/ccgisdata-Street-Midlines/ujuy-nfcm.

7 _{Statutory assessment rates were 38 percent, 36 percent, and 16 percent for commercial, industrial, and}

Class 2 properties in 2003. However, assessment ratio studies conducted by the Illinois Department of Revenue indicate that actual assessment rates were much lower than the statutory. The “recalibration” of rates in 2009 to 25 percent for commercial/industrial and 10 percent for Class 2 suggests that these rates are closer to the target ratios in 2003.

(7)

Table 1 Descriptive Statistics

Variable Mean Deviation Minimum MaximumStandard

Commercial (24,874 observations)

Distance from Major Street 0.059 0.077 0.005 0.765

Distance from Metra Station 1.437 1.131 0.017 8.571

Distance from Chicago City Line 3.635 3.334 0.003 16.772

Distance from Highway 1.698 1.321 0.022 6.693

Distance from Rail Line 0.658 0.637 0.003 4.168

Distance from Quarter Section Boundary 0.091 0.046 0.003 0.250

Within 1/16 Mile of Municipal Line 0.202 0.402 0.000 1.000

1/16 – 1/8 Mile from Municipal Line 0.123 0.328 0.000 1.000

Within 1/16 Mile of Chicago Line 0.033 0.179 0.000 1.000

1/16 – 1/8 Mile from Chicago Line 0.014 0.116 0.000 1.000

Within 1/16 Mile of an Unincorporated Area 0.042 0.200 0.000 1.000

1/16 – 1/8 Mile from an Unincorporated Area 0.034 0.182 0.000 1.000

Assessed Value ($Thousands) 190.18 844.96 0.001 66,893

Log of Assessed Value 10.769 1.518 0.000 18.019

Industrial (11,637 observations)

Log of Assessed Value 11.279 1.517 0.693 16.156

Residential (523,087 observations)

(8)

Density 0.0 1.0 2.0 3.0 0.0 0.2 0.4 0.6 Density Suburbs Residentia l

Commercial Industrial Residentia

l

Commercial Industrial

Chicago

Distance from City Line (Miles

)

Distance from City Line (Miles

) 0.0 0.5 1. 01 .5 01 23 45 67 Density 0.0 0.2 0.4 0.0 0.2 0.4 0.6 Density Suburbs

Residential Commercial Industrial Residential Commercial Industrial

Chicago

Distance from Metra Station (Miles

)

Distance from Metra Station (Miles

) 02 46 8 01 23 4 Figur e 3 Ker nel D ensities: Distanc e fr om a M etr a S ta tion Figur e 2 Ker nel D ensities: Distanc e fr om the Near est C ity Line

(9)

Figur e 5 Ker nel D ensities: Distanc e fr om a R ail Line Figur e 4 Ker nel D ensities: Distanc e fr om the Near est H igh w ay Density 0.0 0.2 0.4 0.0 0.2 0.4 0.6 Density Suburbs

Residential Commercial Industrial Residential Commercial Industrial

Chicago

Distance from Nearest Highway (Miles

)

Distance from Nearest Highway (Miles

) 02 46 01 23 4 Density 0.0 1.0 2.0 0 1 2 3 4 Density Suburbs Residentia l Commercia l Industria l Residentia l Commercia l Industria l Chicago

Distance from Rail Line (Miles

)

Distance from Rail Line (Miles

) 01 23 4 0.0 0.5 1.0 1.5 2.0

(10)

Figur e 7 Ker nel D ensities: Distanc e fr om a Q uar ter S ec tion B oundar y Figur e 6 Ker nel D ensities: Distanc e fr om a M ajor S tr eet Density 0.0 0.4 0.8 1.2 0 20 60 100 Density Suburbs Residentia l Commercia l Industrial Residentia l Commercia l Industrial Chicago

Distance from Major Street (Miles

)

Distance from Major Street (Miles

) 0.0 0.5 1.0 1. 52 .0 2. 5 0.00 0.02 0.04 0.06 0.08 0.10 Density 0 2 4 6 8 0 2 4 6 8 10 Density Suburbs Residentia l Commercia l Industria l Residentia l Commercia l Industria l Chicago

Distance from Quarter Section Boundary (Miles

)

Distance from Quarter Section Boundary (Miles

) 0.00 0.05 0.10 0.15 0.20 0.25 0.00 0.05 0.10 0.15 0.20 0.25

(11)

jurisdictions. Other variables show less marked differences between Chicago and the suburbs. Commercial land use is relatively common near suburban Metra stations, near major streets, and near quarter section boundaries. Industrial land use is relatively more frequent near highways (particularly in Chicago) and near rail lines.

IV. MULTINOMIAL LOGIT RESULTS

In this section, we report the results of multinomial logit models explaining the prob-ability that a parcel is commercial or industrial, with Class 2 residential as the base category. The base model includes measures of proximity to a municipal boundary and proximity to a major street, along with straight-line distances between each parcel and the nearest Metra station, the Chicago city line, the nearest highway, and the nearest rail line. To allow for differences in the effects of distance to the municipal boundary if the neighboring area is part of Chicago or an unincorporated portion of Cook County, we also include measures of proximity to these areas. As any variable that is close to Chicago or an unincorporated area is also close to a municipal boundary by definition, these variables measure the difference between being close to another suburban munici-pality rather than being close to Chicago or an unincorporated area.

We estimate four versions of the model. The first variation includes fixed effects for each of the 91 municipalities represented in the sample. Next, we substitute an instru-mental variable for distance from a major street. The instruinstru-mental variable is constructed

by regressing distance from a major street on distance from the nearest quarter section.8

Third, we estimate a version of the model that includes the instrument for distance from a major street, and finally, we also include municipality fixed effects.

The results are shown in Table 2. Consistent with the results of the kernel density functions, the probabilities of both commercial and industrial land use are significantly

higher for lots that are close to a major street, near Chicago, and close to rail lines.9_The

estimated coefficients for distance from a Metra station are significantly positive across all four specifications for industrial land use. The probability of commercial land use is estimated to decline with distance from a Metra station in the specifications that include municipality fixed effects. Including fixed effects is important in this case because some suburbs are not served by commuter train lines, yet they still have commercial firms. The estimated coefficients for distance from a highway exhibit a similar pattern: after controlling for municipality fixed effects, the probability that a parcel is commercial rather than residential is higher near major highways.

Our primary focus is the set of variables representing proximity to a municipal border. The results imply that the probability that a parcel is commercial or industrial rather than residential is significantly higher within 1/16 of a mile of a municipal boundary across all four specifications. The probability of industrial land use is significantly

8 _{The R}2_{for this regression is 0.137.}

9 _{The consistency of multinomial logit estimates requires that the independence of irrelevant alternatives}

(IIA) assumption be met. Hausman-McFadden (1984) tests reject the assumption. However, simple binary logit estimates, which remain consistent if IIA is violated, produce nearly identical coefficient and standard error estimates. The source of the rejection of IIA is due to the large sample size rather than a substantive difference in the estimated coefficient.

(12)

Table 2 M ultinomial L og it R esults Variable Base Base, Fixed Ef fects IV IV , Fixed Ef fects Commer cial Par cels

Distance from Major Street

–17.721*** –18.257*** –7.708*** –44.204*** (0.146) (0.149) (0.183) (0.525)

Distance from Metra Station

0.049*** –0.096*** 0.026*** –0.171*** (0.007) (0.015) (0.007) (0.014)

Distance from Chicago City Line

–0.013*** –0.095*** –0.001 –0.100*** (0.003) (0.009) (0.003) (0.009)

Distance from Highway

0.010 –0.072*** 0.023*** –0.081*** (0.006) (0.01 1) (0.006) (0.01 1)

Distance from Rail Line

–0.391*** –0.359*** –0.440*** –0.413*** (0.014) (0.019) (0.014) (0.019) W

ithin 1/16 Mile of Municipal Line

0.288*** 0.286*** 0.753*** 0.562*** (0.021) (0.022) (0.021) (0.022)

1/16 – 1/8 Mile from Municipal Line

–0.281*** –0.297*** –0.155*** –0.101*** (0.025) (0.026) (0.025) (0.026) W

ithin 1/16 Mile of Chicago Line

0.471*** 0.383*** 0.670*** 0.475*** (0.045) (0.050) (0.044) (0.048)

1/16 – 1/8 Mile from Chicago Line

0.359*** 0.257*** 0.394*** 0.262*** (0.062) (0.065) (0.061) (0.064) W

ithin 1/16 Mile of Unincorporated

Area –0.1 15** –0.085* –0.151*** –0.105** (0.039) (0.041) (0.038) (0.040)

1/16 – 1/8 Mile from Unincorporated

Area 0.125** 0.198*** 0.150*** 0.149*** (0.043) (0.045) (0.043) (0.044)

(13)

Table 2, C on tinued Variable Base Base, Fixed Ef fects IV IV , Fixed Ef fects Industrial Par cels

Distance from Major Street

–1.030*** –2.352*** 5.930*** –6.994*** (0.089) (0.095) (0.243) (0.726)

Distance from Metra Station

0.271*** 0.488*** 0.256*** 0.479*** (0.007) (0.022) (0.007) (0.022)

Distance from Chicago City Line

–0.037*** –0.409*** –0.054*** –0.404*** (0.004) (0.016) (0.004) (0.016)

Distance from Highway

–0.097*** –0.259*** –0.138*** –0.302*** (0.009) (0.019) (0.009) (0.019)

Distance from Rail Line

–1.980*** –2.302*** –2.023*** –2.297*** (0.030) (0.036) (0.030) (0.036) W

ithin 1/16 Mile of Municipal Line

0.548*** 0.592*** 0.657*** 0.628*** (0.033) (0.035) (0.033) (0.036)

1/16 – 1/8 Mile from Municipal Line

0.202*** 0.210*** 0.268*** 0.257*** (0.032) (0.034) (0.032) (0.034) W

ithin 1/16 Mile of Chicago Line

0.195* –0.058 0.1 19 –0.033 (0.080) (0.089) (0.080) (0.089)

1/16 – 1/8 Mile from Chicago Line

0.428*** 0.243** 0.306*** 0.280*** (0.071) (0.079) (0.072) (0.079) W

ithin 1/16 Mile of Unincorporated

Area 0.003 –0.348*** –0.081 –0.299*** (0.059) (0.065) (0.060) (0.065)

1/16 – 1/8 Mile from Unincorporated

Area 0.368*** 0.050 0.288*** 0.1 10* (0.050) (0.055) (0.050) (0.055) Notes: The base category is residential land use. The sample comprises 559,598 parcels from the 91 municipalities that have at least 10 observations in each land use category . The IV estimator replaces distance from a major street with the predicted values from a regression on distance from the nearest quarter section boundary .

(14)

higher than the probability of either commercial or residential land use in these areas. The probability of commercial land use then declines relative to the other two uses in the next 1/16 mile ring around a municipal line, but the probability of industrial land use remains high. These results are consistent across the four specifications. This pat-tern suggests that suburban jurisdictions attempt to reduce their residents’ exposure to any negative externalities associated with industry by zoning land near municipal boundaries for industrial use.

The probability that a parcel is commercial is significantly higher in a site next to the Chicago border than a comparable parcel next to another suburban jurisdiction. In contrast, sites near incorporated areas, though still more likely to be commercial than residential, have relatively lower probabilities of commercial use than comparable locations near other suburban jurisdictions or Chicago. These results suggest that at least a portion of the tendency for commercial land to be at municipal boundaries is a result of businesses locating near populous areas. Since unincorporated areas tend to have low population density, they are not as attractive for commercial firms as sites near other suburbs or Chicago. Nonetheless, the probability of commercial land use at the boundaries between suburbs and unincorporated areas is significantly higher than the probability of residential land use. Moreover, the decline in the probability of com-mercial land use at the boundaries of unincorporated areas is combined with a significant increase in the commercial probability in the next zone 1/16 – 1/8 of a mile from such border. Similarly, the probability that a parcel is industrial rises significantly in areas 1/16 – 1/8 of a mile from either Chicago or an incorporated area.

V. PROPERTY VALUES

Although the fiscal advantages of commercial and industrial properties give suburbs an incentive to allow non-residential properties in the community, they do not have direct implications for where the properties should be located. Since property taxes are directly related to property value, any location that produces higher non-residential property values has the potential to reduce the tax burden of local residents. Not all non-residential property is high-priced. If relatively low-priced properties are also those that generate negative externalities, then we should expect to see a greater concentration of low-priced non-residential land near municipal boundaries.

To test this prediction, we merge data on assessed values in 2003 with our land use data set. Assessments are the relevant measure of value for this analysis because they are the primary determinant of the property tax. Moreover, assessed values are avail-able for every property in the data set, which is a distinct advantage over transactions prices since only a fraction of single-family homes sell in a given year, and sales of commercial and industrial properties occur even less frequently. Of course, infrequent sales may also produce relatively inaccurate estimates of market value. However, there does not appear to be any reason why properties at suburban boundaries should be assessed less accurately than properties in interior areas, which is the key assumption for our analysis of the within-suburb variation in assessed values.

The dependent variable is the natural log of the official assessed value for January 1, 2003. Given Cook County’s three-year assessment cycle, assessments for properties

(15)

in the northern suburbs will actually date from January 1, 2001 while assessments in the southern suburbs will date from January 1, 2002. Municipality fixed effects will

control for nominal differences in assessment levels across the two regions.10_Although

the level of the assessment does not influence coefficients other than the intercepts when the regressions are estimated with a logarithmic dependent variable, it is worth emphasizing that assessed values are significantly lower than would be implied by statutory assessment rates for all three uses, but particularly so for Class 2 properties. To compare the estimates graphically, we normalize the predicted partial effects of distance from the municipal boundary such that the predicted values for all three land uses have the same mean — the unweighted mean of log assessed value across all properties in the data set. Since the sample is dominated by Class 2 properties, this average will be relatively low. The point of normalization is to facilitate the direct comparison of the spatial variation in the predicted values across the three land uses.

Table 3 presents the results of regressions of log assessed value on the same vari-ables used to explain land use in the multinomial logit models (although we omit the

Table 3

Log Assessed Value Regression Results

Variable ResidentialClass 2 Commercial Industrial

Within 1/16 Mile of Municipal Line 0.040*** –0.011 –0.227***

(0.003) (0.022) (0.037)

1/16 – 1/8 Mile from Municipal Line 0.033*** 0.183*** 0.081*

(0.003) (0.026) (0.033)

Distance from Major Street 0.099*** –0.120*** –0.069

(0.003) (0.023) (0.041)

Distance from Metra Station –0.012*** 0.039* 0.301***

(0.002) (0.018) (0.027)

Distance from Chicago City Line 0.036*** 0.124*** –0.017

(0.001) (0.011) (0.019)

Distance from Highway –0.026*** –0.057*** –0.043

(0.001) (0.013) (0.023)

Distance from Rail Line 0.103*** 0.450*** –0.032

(0.002) (0.022) (0.035)

R2 _0.339 _0.307 _0.300

Observations 523,087 24,874 11,637

Notes: Standard errors are in parentheses. Each regression also includes fixed effects for 90 municipali-ties. Asterisks denote significance at the 1% (***), 5% (**), and 10% (*) levels.

10 _{Although we are not analyzing Chicago properties here, it should be noted that the official origination date}

(16)

variables indicating proximity to the Chicago line and proximity to an unincorporated area because they had proved insignificant after controlling for proximity to a generic municipal line). The regressions also include controls for municipal fixed effects. The results are markedly different across the three land uses. Whereas the value of Class 2 properties is estimated to be higher near the city line than in interior regions, industrial

assessments are much lower near the suburban borders.11

This simple dummy variable specification significantly understates the extent to which non-residential values decline near suburban borders. To allow for more nonlinearity, we use cubic spline functions to model the effects of distance from the municipal line. Following Suits, Mason, and Chan (1978), the smooth cubic spline for a variable x adds a set of variables to a base cubic specification,

(2) =λ +λ +λ +

∑

γ − = x x x x x x D ( ) _k( _k) , k K k 1 2 2 3 3 3 1

where x₁ ...x_K is a series of equally spaced values of x (the “knots”), and D_k is a set of

dummy variables indicating that x ≥ x_k. We use the Schwarz information criterion (SC)

to choose the optimal number of knots, and to determine whether a spline function is

preferred to a simple linear, quadratic, or cubic specification.12_{The results indicate that}

a quadratic function is sufficient for the distance from a municipal line variable for Class 2 properties. The value of K that minimizes SC is 10 for commercial properties and 4 for industrial parcels.

The predictions from the spline functions are shown in Figure 8. Since each function is normalized to have the same mean value, it is the slopes that are directly comparable rather than the levels. What is striking is the way the non-residential functions drop significantly near the municipal borders. Within a short distance of the border — less than 1/5 mile for commercial properties and about 1/3 mile for industrial properties, assessed values decline dramatically as distance to the municipal line declines. A potential anomaly — the rise in residential values near the border — is minor relative to the sharp decline in non-residential values.

Data limitations prevent us from making definitive statements regarding a causal effect of borders on property values. For example, it is possible that industrial land may be located at borders because industry is land intensive and thus is drawn to border areas

where land happens to be inexpensive.13_{Moreover, it should be kept in mind that our}

data set includes assessments rather than actual sales. Our limited set of explanatory

11 _{The endogeneity of land use to property values is a potential concern. Studies such as McMillen and}

McDonald (1991) address the problem by adding a selection bias correction as an explanatory variable in the property value equation. However, the models are not well identified when there are no variables in the first-stage logit model that are not also included in the property value regression.

12 _{The Schwarz (1978) information criterion is}_SC₌_{2 log( ˆ )}_σ ₊_Jx_{log( ) / , where J is the number of total}_{n n}

number of explanatory variables in the regression and σˆ is the maximum likelihood estimate of the regres-sion standard error.

13 _{Although we have detailed data on structural characteristics and lot sizes for residential parcels, we do not}

(17)

variables may cause bias in the results if omitted variables are correlated with proximity to borders, and there is likely to be selection bias associated with the types of properties that choose border locations. Nonetheless, the sharp decline in non-residential values at borders is striking. Though not conclusive as to causality, the results do suggest that the lowest-value property is concentrated within a relatively small geographic area near city lines.

VI. CONCLUSION

Models of parcel-level land use within metropolitan areas are far less common than models of agricultural or urban fringe land use (see Brady and Irwin (2011) for a review). Our study is unique both in the level of detail in the geographic coverage of the large data set and in its focus on the fiscal incentives provided by the property and sales tax. Illinois’ heavy reliance on the property tax and the sales tax for local revenues provides strong incentives for suburbs to attract commercial and industrial firms to their

9.2 9.4 9.6 9.8 10. 0 Log of Assessed Va lue

Distance from City Line (Miles)

Residential (Class 2) Commercial Industrial 0.0 0.2 0.4 0.6 0.8 Figure 8

(18)

jurisdictions. These incentives are particularly powerful in suburban Chicago because Cook County’s classification system results in effective tax rates on commercial and industrial property that are perhaps triple the rates on residential property. If non-residential properties have negative externalities for homeowners, suburban jurisdic-tions may have an incentive to form concentrajurisdic-tions of non-residential properties near the suburb’s edge.

Our data set allows us to test this prediction for suburban Cook County. After control-ling for access to the transportation network, we find that parcels near municipal borders are much more likely to be in commercial or industrial use. Moreover, the assessed values of properties near municipal borders tend to be much lower for non-residential properties relative to the interior of a municipality. The results suggest that borders have a significant influence on the pattern of land use in the Chicago area.

DISCLOSURES

The authors have no direct financial arrangements that might give rise to conflicts of interest with respect to the research reported in this paper. In the past three years, McMillen has received at least $10,000 in financial support from the Federal Reserve Bank of Chicago, the Lincoln Institute of Land Policy, O’Keefe, Lyons, and Hynes, LLC, and Applied Real Estate Analysis.

REFERENCES

Bowman, Ann, and Michael Pagano, 2004. Terra Incognita: Vacant Land and Urban Strategies. Georgetown University Press, Washington, DC.

Brady, Michael, and Elena Irwin, 2011. “Accounting for Spatial Effects in Economic Models of Land Use: Recent Developments and Challenges Ahead.” Environmental and Resource Econom-ics 48 (3), 487–509.

Chapman, Jeffrey, 1988. “Land Use Planning and the Local Budget: A Model of Their Inter-relationships.” Public Administration Review 48 (4), 800–806.

Chapman, Jeffery, 2008. “The Fiscalization of Land Use: The Increasing Role of Innovative Revenue Raising Instruments to Finance Public Infrastructure.” Public Works Policy and Man-agement 12 (4), 551–567.

Chicago Metropolitan Agency for Planning, 2010. “State and Local Taxation: Existing Conditions and Issues of Significance in Metropolitan Chicago.” Chicago Metropolitan Agency for Planning, Chicago IL, http://www.cmap.illinois.gov/documents/10180/35654/ FY10_0075+TAX+SNAPSHOT.pdf/baca0410-f127-4f73-b0b3-16d2fa319fde.

Chicago Metropolitan Agency for Planning, 2014. “Fiscal and Economic Impact Analysis of Local Development Decisions.” Chicago Metropolitan Agency for Planning, Chicago IL, https://www. cmap.illinois.gov/documents/10180/82875/Fiscal+Econ+Impacts+Dev+FINAL.pdf/6fc7ed1c-aba7-4d6a-a057-8d251aa7fbdc.

(19)

The Civic Federation, 2010. “The Cook County Property Assessment Process: A Primer on As-sessment, Classification, Equalization and Property Tax Exemptions.” The Civic Federation, Chi-cago, IL, http://www.civicfed.org/sites/default/files/100405_CookCountyAssessmentPrimer.pdf. Fischel, William A., 1975. “Fiscal and Environmental Considerations in the Location of Firms in Suburban Communities.” In Mills, Edwin S., and Wallace E. Oates (eds.), Studies in Fiscal Zoning and Land Use Controls, 119–174. D.C. Heath and Company, Lexington, MA.

Fischel, William A., 1976. “An Evaluation of Proposals for Metropolitan Sharing of Commercial and Industrial Tax Base.” Journal of Urban Economics 3 (3), 253–263.

Hausman, Jerry, and Daniel McFadden, 1984. “A Specification Test for the Multinomial Logit Model.” Econometrica 52 (5), 1219–1240.

Lewis, Paul G., 2001. “Retail Politics: Local Sales Taxes and the Fiscalization of Land Use.” Economic Development Quarterly 15 (1), 21–35.

McMillen, Daniel P., and John F. McDonald, 1991. “Urban Land Value Functions with Endog-enous Zoning.” Journal of Urban Economics 29 (1), 14–27.

Misczynski, Dean J., 1986. “The Fiscalization of Land Use.” California Policy Choices 3, 127–50. Schwarz, Gideon, 1978. “Estimating the Dimension of a Model.” The Annals of Statistics 6 (2), 461–464.

Silverman, Bernard W., 1986. Density Estimation for Statistics and Data Analysis. Chapman and Hall, New York, NY.

Suits, Daniel B., Andrew Mason, and Louis Chan, 1978. “Spline Functions Fitted by Standard Regression Methods.” Review of Economics and Statistics 60 (1), 132–139.

Wassmer, Robert W., 2002. “Fiscalisation of Land Use, Urban Growth Boundaries and Non- Central Retail Sprawl in the Western United States.” Urban Studies 39 (8), 1307–1327.

(20)