The Assessment Gap: Racial Inequalities in Property Taxation

Carlos F. Avenancio-Leon Troup Howard

University of California – San Diego University of Utah

1 / 41
Research Question

Individuals face three main taxes in US: 1) sales, 2) income, 3) property

2 / 41
Research Question

Individuals face three main taxes in US: 1) sales, 2) income, 3) property

Two people, same store, same purchase ⇒ pay same amount in sales tax

2 / 41
Research Question

Individuals face three main taxes in US: 1) sales, 2) income, 3) property

Two people, same store, same purchase ⇒ pay same amount in sales tax

Average differences by personal traits would be surprising (Why? How?)

2 / 41
Research Question

Individuals face three main taxes in US: 1) sales, 2) income, 3) property

Two people, same store, same purchase ⇒ pay same amount in sales tax

Average differences by personal traits would be surprising (Why? How?)

This paper: Holding property tax rates fixed, do racial and ethnic
minorities in the US face higher property tax burden?

2 / 41
Research Question

Individuals face three main taxes in US: 1) sales, 2) income, 3) property

Two people, same store, same purchase ⇒ pay same amount in sales tax

Average differences by personal traits would be surprising (Why? How?)

This paper: Holding property tax rates fixed, do racial and ethnic
minorities in the US face higher property tax burden?
- Document: yes
- Two channels; underlying mechanisms
- Potential solution
2 / 41
Motivation: Why Study Race and Property Tax?

1 Property taxes affect essentially everyone

• Central funding source for: schools, roads, public safety, etc.
• $450-$500 billion annual total revenues

2 Large implications for household finance

• Median black / white household net worth: 13k / 139k
• Many families: home is largest asset & primary savings/leverage technology

3 Institutional discrimination and statistical/algorithmic bias

• Racial disparities illegal under federal law since 1968 (F.H.A)
• Race-blind policies vs race-neutral outcomes

3 / 41
How Can Tax Burden Vary, Holding Rates Fixed?

Key feature of property tax:

• Tax paid intended to be proportional to market value of home…
• … but tax bills are computed based on “assessment” value

4 / 41
How Can Tax Burden Vary, Holding Rates Fixed?

Key feature of property tax:

• Tax paid intended to be proportional to market value of home…
• … but tax bills are computed based on “assessment” value

Two people, i and j, subject to same tax, r:

(𝑇ℎ𝑒𝑜𝑟𝑒𝑡𝑖𝑐𝑎𝑙 𝑟 𝑀$ 𝑟 𝑀&
=
a𝑑 𝑣𝑎𝑙𝑜𝑟𝑒𝑚 𝑡𝑎𝑥) 𝑀$ 𝑀&

4 / 41
How Can Tax Burden Vary, Holding Rates Fixed?

Key feature of property tax:

• Tax paid intended to be proportional to market value of home…
• … but tax bills are computed based on “assessment” value

Two people, i and j, subject to same tax, r:

(𝑇ℎ𝑒𝑜𝑟𝑒𝑡𝑖𝑐𝑎𝑙 𝑟 𝑀$ 𝑟 𝑀& Tax Bill

=
a𝑑 𝑣𝑎𝑙𝑜𝑟𝑒𝑚 𝑡𝑎𝑥) 𝑀$ 𝑀&

4 / 41
How Can Tax Burden Vary, Holding Rates Fixed?

Key feature of property tax:

• Tax paid intended to be proportional to market value of home…
• … but tax bills are computed based on “assessment” value

Two people, i and j, subject to same tax, r:

(𝑇ℎ𝑒𝑜𝑟𝑒𝑡𝑖𝑐𝑎𝑙 𝑟 𝑀$ 𝑟 𝑀&
= Effective Tax Rate
a𝑑 𝑣𝑎𝑙𝑜𝑟𝑒𝑚 𝑡𝑎𝑥) 𝑀$ 𝑀&

4 / 41
How Can Tax Burden Vary, Holding Rates Fixed?

Key feature of property tax:

• Tax paid intended to be proportional to market value of home…
• … but tax bills are computed based on “assessment” value

Two people, i and j, subject to same tax, r:

(𝑇ℎ𝑒𝑜𝑟𝑒𝑡𝑖𝑐𝑎𝑙 𝑟 𝑀$ 𝑟 𝑀&
=
a𝑑 𝑣𝑎𝑙𝑜𝑟𝑒𝑚 𝑡𝑎𝑥) 𝑀$ 𝑀&

4 / 41
How Can Tax Burden Vary, Holding Rates Fixed?

Key feature of property tax:

• Tax paid intended to be proportional to market value of home…
• … but tax bills are computed based on “assessment” value

Two people, i and j, subject to same tax, r:

(𝐻𝑜𝑤 𝑡ℎ𝑒 𝑝𝑟𝑜𝑝𝑒𝑟𝑡𝑦 𝑟 𝐴$ 𝑟 𝐴&

=
𝑡𝑎𝑥 𝑎𝑐𝑡𝑢𝑎𝑙𝑙𝑦 𝑤𝑜𝑟𝑘𝑠) 𝑀$ 𝑀&

4 / 41
How Can Tax Burden Vary, Holding Rates Fixed?

Key feature of property tax:

• Tax paid intended to be proportional to market value of home…
• … but tax bills are computed based on “assessment” value

Two people, i and j, subject to same tax, r:

> 𝑟 𝐴$ 𝑟 𝐴&
: “assessment ratio” =
? 𝑀$ 𝑀&

4 / 41
How Can Tax Burden Vary, Holding Rates Fixed?

Key feature of property tax:

• Tax paid intended to be proportional to market value of home…
• … but tax bills are computed based on “assessment” value

Two people, i and j, subject to same tax, r:

> 𝑟 𝐴$ 𝑟 𝐴& >

: “assessment ratio” = Effective tax rate: 𝑓 ;𝑟
? 𝑀$ 𝑀& ?

4 / 41
How Can Tax Burden Vary, Holding Rates Fixed?

Key feature of property tax:

• Tax paid intended to be proportional to market value of home…
• … but tax bills are computed based on “assessment” value

Two people, i and j, subject to same tax, r:

𝑟 𝐴$ 𝑟 𝐴&
If: > , 𝑒𝑓𝑓𝑟𝑎𝑡𝑒$ > 𝑒𝑓𝑓𝑟𝑎𝑡𝑒&
𝑀$ 𝑀&

4 / 41
How Can Tax Burden Vary, Holding Rates Fixed?

Key feature of property tax:

• Tax paid intended to be proportional to market value of home…
• … but tax bills are computed based on “assessment” value

Two people, i and j, subject to same tax, r:

𝑟 𝐴$ 𝑟 𝐴&
If: > , 𝑒𝑓𝑓𝑟𝑎𝑡𝑒$ > 𝑒𝑓𝑓𝑟𝑎𝑡𝑒&
𝑀$ 𝑀&

Ø Within taxing jurisdiction, variation in assessment ratio is

sufficient for inequality
4 / 41
This Paper

Form taxing jurisdictions

◦ Holds fixed: intended taxation, public goods, assessment practices
◦ Challenging: local governments very spatially complex
◦ Rely on shapefiles for universe of local governments

Form assessment ratios

◦ Annual assessments for 118M homes; 53M observed transactions
◦ Restriction to arms-length, full consideration sale

Associate assessment ratios with homeowner race & ethnicity

Variation in assessment ratio ⇒ reject equitable tax null

5 / 41
Theoretical Assessment Ratio (Assessed Value / Market)

Assessment Ratio, Deviations from Mean, PA, Philadelphia

6 / 41
Philadelphia: Assessment Ratios and Demographic Heatmap

Realized Assessment Ratio (demeaned by jurisdiction), PA, Philadelphia

Quartile 4

Quartile 3

Quartile 2

Quartile 1

6 / 41
Philadelphia: Assessment Ratios and Demographic Heatmap

Realized Assessment Ratio (demeaned by jurisdiction), PA, Philadelphia Population Share: Black and Hispanic Residents, PA, Philadelphia
1.0

0.8

0.6
Quartile 4

Quartile 3

Quartile 2

Quartile 1
0.4

0.2

0.0

6 / 41
Cook County, IL: Assessment Ratios and Demographics

Realized Assessment Ratio (demeaned by jurisdiction), IL, Cook

Quartile 4

Quartile 3

Quartile 2

Quartile 1

7 / 41
Cook County, IL: Assessment Ratios and Demographics

Realized Assessment Ratio (demeaned by jurisdiction), IL, Cook Population Share: Black and Hispanic Residents, IL, Cook
1.0

0.8

0.6
Quartile 4

Quartile 3

Quartile 2

Quartile 1
0.4

0.2

0.0

7 / 41
Preview of Findings

Assessment gap: 10-13% higher tax burden for black and Hispanic homeowners
◦ Cannot be Tiebout sorting along preferences for public goods
◦ $300-$390 annually for median minority homeowner
◦ At 90th percentile: approx $800 annually

Two channels:
◦ 6%-7%: neighborhood attributes and racial sorting (spatial / between)
I Assessments insufficiently responsive to highly local characteristics

◦ 5%-6%: individual homeowner (not spatial / within)

I Racial differential in appeals behavior/outcomes

Small-geography Home Price Indices are potential policy fix

◦ Simple algorithm, using public data, fixes ˜70% of total inequality
8 / 41
Contribution to the Literature
Black-white wealth gap
◦ Spatial sorting: Cutler and Glaeser 1997, Card and Rothstein 2007, Charles and Guryan
2008, Ananat 2011, Chetty et al 2014, Chetty et al 2019
◦ Here: public finance channel; highly persistent; wealth rather than wages

Racial and ethnic differences in outcomes

◦ Housing markets: Charles and Hurst 2002, Bayer et al 2007, Card et al 2008, Bayer et al
2017, Atuahene 2018, Atuahene and Berry 2019
◦ Here: national differences in tax burdens; non-market setting

Bias in algorithms
◦ Machine learning and lending: Bartlett et al 2018, Fuster et al 2018, Kleinberg et al 2018
◦ Here: race-blind policies will exacerbate discriminatory outcomes

9 / 41
1 Motivation

2 Setting and Institutional Details

3 Results
Data and Estimating Equation
Assessment Gap
Spatial Decomposition
Evidence on Appeals Mechanism
Extensions

4 Policy Approach

5 Conclusion
10 / 41
Property Taxes Central for Local Governments
Property taxes are local taxes that provide the largest source of money local gov-
ernments use to pay for schools, streets, roads, police, fire protection and many other
services” -Texas State Comptroller

Total Property Tax Receipts (Billions of 2018$) Average General Revenue Breakdown, Local Units
500

Property Taxes (56%)

400
300
200

Other Taxes (7%)

100

User Fees (38%) 10 / 41

Residential Property Taxes Are Ad Valorem

2018 Georgia Code, Title 48, Chapter 5: Ad Valorem Taxation of Property:

Except as otherwise provided in this Code section, taxable tangible property shall
be assessed at 40 percent of its fair market value and shall be taxed [...] according to
40 percent of the property’s fair market value.

“Fair market value of property” means the amount a knowledgeable buyer would
pay for the property and a willing seller would accept for the property at an arm’s
length, bona fide sale.

11 / 41
Local Property Tax Overview

• Local Choice (political) Policy Rate: 5%

• Unobserved Target Assessment Ratio: 40%

12 / 41
Local Property Tax Overview

• Local Choice (political) Policy Rate: 5%

• Unobserved Target Assessment Ratio: 40%

Home 1 Value: $200k Home 2 Value: $400k

Accurate Assessments Assessment: $80k $85k Assessment: $160k $150k

Tax Bill: 5% of $80k:$85k: Tax Bill: 5% of $160k:150k:

$4,000 $4,250 $8,000 $7,500

12 / 41
Local Property Tax Overview

• Local Choice (political) Policy Rate: 5%

• Unobserved Target Assessment Ratio: 40%

Home 1 Value: $200k Home 2 Value: $400k

Accurate Assessments Assessment: $80k $85k Assessment: $160k $150k

Tax Bill: 5% of $80k:$85k: Tax Bill: 5% of $160k:150k:

$4,000 $4,250 $8,000 $7,500

Assessment Ratio: .425 Assessment Ratio: .475

Effective Rate: 2%15% Effective Rate: 2%75%

Constant Assessment Ratios: Necessary Condition for Equitable Tax Burden

12 / 41
Assessment Ratio is Object of Interest

• Local Choice (political) Policy Rate: 5%

• Unobserved Target Assessment Ratio: 40%

Home 1 Value: $200k Home 2 Value: $400k

Realized Assessments Assessment: $80k $85k Assessment: $160k $150k

Tax Bill: 5% of $80k $85k: Tax Bill: 5% of $160k 150k:

$4,000 $4,250 $8,000 $7,500

Assessment Ratio: .425 Assessment Ratio: .375

Effective Rate: 2.125% Effective Rate: 1.875%

12 / 41
Assessment Ratio is Object of Interest

• Local Choice (political) Policy Rate: 5%

• Unobserved Target Assessment Ratio: 40%

Home 1 Value: $200k Home 2 Value: $400k

Realized Assessments Assessment: $80k $85k Assessment: $160k $150k

Tax Bill: 5% of $80k $85k: Tax Bill: 5% of $160k 150k:

$4,000 $4,250 $8,000 $7,500

Assessment Ratio: .425 Assessment Ratio: .375

Effective Rate: 2.125% Effective Rate: 1.875%

Variation in the Assessment Ratio: sufficient for unequal tax burden

12 / 41
Two Major Empirical Challenges

Challenge #1:
◦ Must hold fixed intended level of taxation and public goods
◦ 75,000 potential taxing entities; annual changes
◦ Extremely complex spatial overlay of local governments
• Tax Code Areas (TCAs) fail to capture provision of public goods for nontaxing local districts.

Challenge #2:
◦ Must also hold target assessment ratio fixed (unobserved)
◦ “Natural” benchmark of 1-to-1 is less common
◦ Target may change annually by legislation

Realized Jurisdiction ARs

13 / 41
“Taxing Jurisdiction”: Precise Definition

County
Jurisdiction 1

City School
District
Jurisd. 3
Jurisdiction 2 Jurisdiction 4

“Jurisdiction”:
“Jurisdiction”: Geography served by unique network of overlapping gvts
Region touched by a unique network of overlapping governments
Further Theoretical Example Real-World Example

14 / 41
“Taxing Jurisdiction”: Precise Definition

County
Jurisdiction 1

City School
District
Jurisd. 3
Jurisdiction 2 Jurisdiction 4

“Jurisdiction”:
“Jurisdiction”: Geography served by unique network of overlapping gvts
Region touched by a unique network of overlapping governments
Further Theoretical Example Real-World Example

14 / 41
Outline

1 Motivation

2 Setting and Institutional Details

3 Results
Data and Estimating Equation
Assessment Gap
Spatial Decomposition
Evidence on Appeals Mechanism
Extensions

4 Policy Approach

5 Conclusion

14 / 41
Data Sources

1 Panel data: 118M properties, 2003-2016; annual assessments; all transactions (53M);
longitude & latitude; home attributes (ATTOM)
I California: results in paper. Here: results from 49 states.

2 Shapefiles: a) cities, towns, school districts, b) special/utility districts,

c) custom shapefiles for any issuer of public debt. (Atlas Muni Data)

3 Loan-level reported race & ethnicity for mortgage origination (HMDA)

4 Demographic info from ACS; tract and block group shapefiles from US Census

Race and Ethnicity

15 / 41
Timing Details

𝐴"#$ 𝐴" 𝐴"%$ 𝐴"%& 𝐴"%'

𝑴𝒕

t-2 t-1 year t t+1 t+2 t+3

16 / 41
Timing Details

Race/ethnicity carried forward

from mortgage origination

𝐴"#$ 𝐴" 𝐴"%$ 𝐴"%& 𝐴"%'

𝑴𝒕

t-2 t-1 year t t+1 t+2 t+3

16 / 41
Timing Details

Race/ethnicity carried forward

from mortgage origination

𝐴"#$ 𝐴" 𝐴"%$ 𝐴"%& 𝐴"%'

𝑴𝒕 𝑴𝒕%𝟐

t-2 t-1 year t t+1 t+2 t+3

16 / 41
Timing Details

Race/ethnicity carried forward

from mortgage origination

𝐴"#$ 𝐴" 𝐴"%$ 𝐴"%& 𝐴"%'

𝑴𝒕 𝑴𝒕%𝟐

t-2 t-1 year t t+1 t+2 t+3

+,-.
in final dataset
/,-.
(with seller race/ethnicity)

16 / 41
Timing Details

Race/ethnicity carried forward

from mortgage origination

(only for refinance transactions)

𝐴"#$ 𝐴" 𝐴"%$ 𝐴"%& 𝐴"%'

𝑴𝒕 𝑴𝒕%𝟐

t-2 t-1 year t t+1 t+2 t+3

+,-.
in final dataset
/,-.
(with seller race/ethnicity)

16 / 41
Estimating Equation

◦ Equitable tax null: β = 0

◦ Omitted group in all regressions: white, non-Hispanic residents

: property, : jurisdiction, : year, race: race or ethnicity

Equitable Null Derivation

17 / 41
Estimating Equation

A
ln( Mijt ) = γjt + β raceijt + εijt
ijt

◦ Equitable tax null: β = 0

◦ Omitted group in all regressions: white, non-Hispanic residents

i: property, j: jurisdiction, t: year, race: race or ethnicity

Equitable Null Derivation

18 / 41
Group Means: Legal Grounding

A
ln( Mijt ) = γjt + β raceijt + εijt
ijt

“Disparate impact” is legal standard by which courts evaluate discrimination claims

Federal Law, 24 CFR S100.500(a):

“[a] practice has a discriminatory effect where it actually or predictably results in
a disparate impact on a group of persons[...] because of race, color, religion, sex,
handicap, familial status, or national origin”

US Supreme Court (2015): in housing, sufficient for discrimination

19 / 41
Overall Assessment Gap

log(Assessment) - log(Market)

(1) (2)

Black Mortgage Holder 0.1266∗∗∗

(0.0150)

Black or Hispanic Mortgage Holder 0.0984∗∗∗

(0.0106)

Fixed Effects Jurisd-Year Jurisd-Year

Other Controls N N
No. Clusters 37723 37723
Observations 6,987,915 6,987,915
R2 0.8798 0.8798

Note: ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01

Median minority homeowner: 207k home and 1.4% tax: $300 - $390 annually

Fiscal Cycle Robustness

20 / 41
State Breakdown

Assessment Gap, by State

0.4
●
Assmt Gap, Black Residents
0.3

●
●
●
●
0.2

● ●

●
●
● ●
● ●
0.1

● ● ● ● ●
● ● ●
●
● ● ● ● ●
● ● ● ●
● ●
● ● ●
● ● ●
0.0

●
●
●
−0.1

MO

WV
WA
MN

MD

NM
OH

MA

CO

OR
NC
OK

GA

ND
NH
SC

DE
AR
NE

NY

SD
NV
TN

KY

CT

KS
AZ

TX

VT
VA
AL

LA

WI

NJ
PA

FL
MI
RI

HI

IN
IA
IL

State Breakdown, Black or Hispanic California Results

21 / 41
Outline

1 Motivation

2 Setting and Institutional Details

3 Results
Data and Estimating Equation
Assessment Gap
Spatial Decomposition
Evidence on Appeals Mechanism
Extensions

4 Policy Approach

5 Conclusion

21 / 41
Decomposing Assessment Gap

Roadmap:
1 Distinguish: within-neighborhood inequality vs between-neighborhood inequality
2 Neighborhood Composition: between-variation in assessment ratio
3 Homeowner Effect: within-variation in assessment ratio

“Neighborhood”: US Census tract or block group (much smaller than jurisdiction)

22 / 41
Homeowner Effect

Goal: Hold constant all spatial & geographic factors

Ideal experiment: Adjacent homes; homeowners of different race/ethnicity

Feasible: Condition on successively smaller geographies; show stable estimates

23 / 41
Homeowner Effect

Assessment Ratios Relative to White Residents

0.18
0.16
0.14
0.12
Percent

0.1
0.08
0.06
0.04
0.02
0
Black Residents Black or Hispanic Residents
Jurisdiction (Baseline) Tract Block Group

Baseline Regression

24 / 41
Homeowner Effect

Assessment Ratios Relative to White Residents

0.18
0.16
0.14
0.12
Percent

0.1
0.08
0.06
0.04
0.02
0
Black Residents Black or Hispanic Residents
Jurisdiction (Baseline) Tract Block Group

Baseline Regression Tract Regression

24 / 41
Homeowner Effect

Assessment Ratios Relative to White Residents

0.18
0.16
0.14
0.12
Percent

0.1
0.08
0.06
0.04
0.02
0
Black Residents Black or Hispanic Residents
Jurisdiction (Baseline) Tract Block Group

Baseline Regression Tract Regression Block Group Regression

24 / 41
Homeowner Effect

Assessment Ratios Relative to White Residents

0.18
0.16
0.14
0.12
Percent

0.1
0.08
0.06
0.04 5-6%: Within
0.02 (“Homeowner”)
0
Black Residents Black or Hispanic Residents
Jurisdiction (Baseline) Tract Block Group

Baseline Regression Tract Regression Block Group Regression

24 / 41
Homeowner Effect

Assessment Ratios Relative to White Residents

0.18
0.16
0.14
0.12
Percent

0.1 6-7%: Between

0.08 (“Neighborhood”)
0.06
0.04 5-6%: Within
0.02 (“Homeowner”)
0
Black Residents Black or Hispanic Residents
Jurisdiction (Baseline) Tract Block Group

Baseline Regression Tract Regression Block Group Regression

24 / 41
Neighborhood Composition

Spatial sorting by race in US is well-known

◦ Ananat (2011), Cutler and Glaeser (1997); many others

Result: neighborhood attributes faced by average resident varies by race

Characteristics are capitalized differently in market prices vs assessments

Generates spatial variation in tax burden that correlates with race

25 / 41
Sample Differences

Weighted Average Tract−Level Attributes, by Race (Demeaned)

0.3
White Residents
Black or Hispanic Residents

0.2
0.1
0.0
−0.1

Minority Share Unemployment SNAP Homeowner % GINI

More Variables Baseline Regression Evidence More Regression Evidence

26 / 41
Implied Hedonic Prices

“Automated Valuation Models”: some form of hedonic regression

Estimate two hedonic models: 1) LHS = Market, 2) LHS = Assessment

Vicjt = γjt + ΘV Xicjt + β V Wcjt + εicjt

Goal: compare ΘA , β A with ΘM , β M

V : assessment or market; i: home, c: tract, j: jurisdiction

t: time, Xicjt : home attributes, Wcjt : local attributes

27 / 41
 Estimated Hedonic Prices
−0.01 0.00 0.01 0.02 0.03 0.04
Bla
ck
/H
isp
an
ic
Sh
are
SN
Relative Hedonic Prices

AP
Ow
ne
rP
erc
en
Un t
em
plo
ym
en
Me t
dia
nI
nc
om
e
GI
NI

Ba
thr
oo
ms

Fir
e pla
ce

Ye
a rB
uil
t

Po
o l
Implied Elasticity of Assessment Ratio to 1 SD Shift

Pa
ti o
Sq
ua
re
Fe
e t
28 / 41
 Estimated Hedonic Prices
−0.01 0.00 0.01 0.02 0.03 0.04
Bla
ck
/H
isp
an
ic
Sh
are
SN
Relative Hedonic Prices

AP
Ow
ne
rP
erc
en
Un t
e mp
loy
me
nt
Me
dia
nI
nc
om
e
GI
NI

Ba
thr
oo
ms

Fir
e pla
ce

Ye
a rB
uil
t

Po
o l
Implied Elasticity of Assessment Ratio to 1 SD Shift

Pa
ti o
Sq
ua
re
Fe
e t
28 / 41
Spatial Variation in Tax Burden Correlated with Race

log(Assessment) - log(Market)

(1) (2)

Black Mortgage Holder 0.079∗∗∗

(0.004)

Black Share 0.299∗∗∗

(0.046)

Black or Hispanic Mortgage Holder 0.067∗∗∗

(0.003)

Black or Hispanic Share 0.277∗∗∗

(0.042)

Jurisd-Year FE Y Y
Other Controls N N
No. Clusters 37679 37679
Observations 6,944,439 6,944,439
R2 0.881 0.881

Note: ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01

29 / 41
Taking Stock

Overall assessment gap: 10-13%

Between variation: 6-7%

◦ Assessors underweight neighborhood attributes in projecting market prices
◦ Tactically: hedonic F.E. or rule-of-thumb growth for too large an area

Within variation: 5-6%

◦ So far unexplained
◦ Hypothesis: racial differential in appeals behavior/outcomes

30 / 41
Outline

1 Motivation

2 Setting and Institutional Details

3 Results
Data and Estimating Equation
Assessment Gap
Spatial Decomposition
Evidence on Appeals Mechanism
Extensions

4 Policy Approach

5 Conclusion

30 / 41
Mechanism for Homeowner Effect

Extensive social science literature:

◦ Minority residents may be less trusting of public officials
◦ May perceive institutions are not designed to serve them

Assessment Appeals:
◦ Almost always process for appealing assessment
◦ Obtained administrative micro-data from 2nd largest county

31 / 41
Cook County, IL

Population: 5M; Homes: 1.9M

◦ Appeals, 2003-2015: 3.5M

Usual to hire tax attorney - perception: connections matter

Antiquated data/tech & low staffing: “assessment by appeal”

Additional info:
1 Appeal filed
2 Win / loss
3 Amount of reduction

IL Homeowner Effect

32 / 41
Results: Appeals in Cook County

Dependent Variable:

Appeal Win Appeal Reduction

(1) (2) (3)

Black or Hispanic Mortgage Holder −0.982∗∗∗ −1.993∗∗∗ −0.258∗∗∗

(0.068) (0.245) (0.074)

Baseline Rate 14.6 67.4 12.0

Fixed Effects BG-Year BG-Year BG-Year
No. Clusters 3954 3933 3893
Observations 4,076,655 694,553 476,368
R2 0.383 0.415 0.443

Note: ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01

Notes: 1) linear probability model, 2) coefficients are (%)

Black Homeowners

33 / 41
Consistent with National Data

Racial differential in appeals ⇒ different assessment trajectories by race

Test by exploiting changes of racial ownership within properties across time

(Note: no market prices; only instance today)

∆log (Aict ) = αi + γct + β race/ethnicityict + εict

34 / 41
Results: Diff in Diff around Racial Ownership

Assessments
Growth Levels

(1) (2) (3) (4)

Black Mortgage Holder 0.0711∗ 0.2917∗∗∗

(0.0386) (0.0415)

Black or Hispanic Mortgage Holder 0.4103∗∗∗ 0.7923∗∗∗

(0.0255) (0.0274)

Fixed Effects Two-Way Two-Way Two-Way Two-Way

No. Clusters 12268641 12268641 12268641 12268641
Observations 54,970,191 54,970,191 54,970,191 54,970,191
R2 0.6925 0.6925 0.9910 0.9910

Note: ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01

Notes: coefficients are (%)

35 / 41
Outline

1 Motivation

2 Setting and Institutional Details

3 Results
Data and Estimating Equation
Assessment Gap
Spatial Decomposition
Evidence on Appeals Mechanism
Extensions

4 Policy Approach

5 Conclusion

35 / 41
Extensions & Robustness
1 Assessment gap by year Annual Estimates

2 Role of market prices Market Prices

3 Ruling out pure income effect Income

4 Ruling out pure property price effect Price Controls

5 Pass-through of assessment ratio to taxes paid Taxes Paid

6 Assessment gap distribution: county-level estimates County Estimates

7 Sample split by racial animus Animus

8 Sample split by county-level home price growth By County HPI

9 Sample split by county-level minority population County Minority Share

10 Effect of homeowner tenure Time Since Sale

11 Simple ratios instead of log(assessment ratio) Simple Ratios

36 / 41
Outline

1 Motivation

2 Setting and Institutional Details

3 Results
Data and Estimating Equation
Assessment Gap
Spatial Decomposition
Evidence on Appeals Mechanism
Extensions

4 Policy Approach

5 Conclusion

36 / 41
Algorithm for Equitable Assessments

Neighborhood composition drives at least half of distortion

Feasible to construct assessments that reflect spatial attributes?

37 / 41
Algorithm for Equitable Assessments

Neighborhood composition drives at least half of distortion

Feasible to construct assessments that reflect spatial attributes?

𝐻𝐻𝐻𝐻𝐼𝐼𝑧𝑧𝑧 𝐻𝐻𝐻𝐻𝐼𝐼𝑧𝑧𝑧
𝐴𝐴̂ 𝑖𝑖𝑖𝑖𝑖 ≡ 𝑀𝑀𝑖𝑖𝑖𝑖𝑖 𝐴𝐴̂ 𝑖𝑖𝑖𝑖𝑖 = 𝐴𝐴̂ 𝑖𝑖𝑖𝑖𝑖 𝐴𝐴̂ 𝑖𝑖𝑖𝑖𝑖 = 𝐴𝐴̂ 𝑖𝑖𝑖𝑖𝑖
𝐻𝐻𝐻𝐻𝐼𝐼𝑧𝑧𝑧 𝐻𝐻𝐻𝐻𝐼𝐼𝑧𝑧𝑧

First sale: Grow by zip-code HPI to

Set correct assessment produce subsequent assessments

37 / 41
Algorithm for Equitable Assessments

Test: compare inequality with realized assessments vs synthetic assessments

𝐻𝐻𝐻𝐻𝐼𝐼𝑧𝑧𝑧𝑧
𝐴𝐴̂ 𝑖𝑖𝑖𝑖𝑖 ≡ 𝑀𝑀𝑖𝑖𝑖𝑖𝑖 𝐴𝐴̂ 𝑖𝑖𝑖𝑖𝑖𝑖 = 𝐴𝐴̂ 𝑖𝑖𝑖𝑖𝑖
𝐻𝐻𝐻𝐻𝐼𝐼𝑧𝑧𝑧

𝐴𝐴�𝑖𝑖𝑖𝑖𝑖𝑖
First sale: 2nd sale, form :
𝑀𝑀𝑖𝑖𝑖𝑖𝑖𝑖
Set correct assessment Re-estimate assessment gap

38 / 41
Results: Using Zip-Code Level HPIs

15
Realized
Synthetic
Assessment Gap (percent)
10
5
0
−5

Black Homeowners Black or Hispanic Homeowners

Underlying Regressions

39 / 41
Conclusion

1 10-13% higher property tax burden for black and/or Hispanic residents

2 Geographic channel and a homeowner channel:

• Assessments insufficiently sensitive to local attributes
• Racial differentials in appeals behavior and outcomes

3 Inequality can be significantly reduced by linking assessments to local-HPIs

40 / 41
Thank you!

41 / 41