THREE ESSAYS ON REAL ESTATE FINANCE...eld{ real estate nance. He was not only an academic advisor, but was also a great mentor with whom I could discuss many deep concerns. He always

THREE ESSAYS ON REAL ESTATE FINANCE

Hyun-Soo Choi

A DISSERTATION

PRESENTED TO THE FACULTY

OF PRINCETON UNIVERSITY

IN CANDIDACY FOR THE DEGREE

OF DOCTOR OF PHILOSOPHY

RECOMMENDED FOR ACCEPTANCE

BY THE DEPARTMENT OF ECONOMICS

Advisor: Harrison Hong

JUNE 2012

c© Copyright by Hyun-Soo Choi, 2012.

All rights reserved.

iii

Abstract

This dissertation presents three essays on real estate finance.

In the first chapter, I test the hypothesis that anti-predatory lending laws in-

hibited the volume of mortgage lending during the recent housing-bubble period.

I use cross-state variation in the timing of law adoptions and the strictness of the

laws in a fixed-effect model to estimate the impact of these laws on mortgage vol-

ume. Consistent with my hypothesis, I find that states with stricter laws had lower

mortgage volume. I also find a larger reduction of mortgage volume in loans with

a high Loan-to-Income ratio. I test whether, by restricting mortgage volume, these

laws impacted household expenditures and find some weak evidence that the laws

reduced household expenditures.

In the second chapter, co-authored with Harrison Hong and Jose Scheinkman,

we develop a speculation-based theory of home improvements. Housing services

are produced from a mix of land and structures. Homeowners have an option to

increase their structures (i.e. make improvements) holding fixed their land. Spec-

ulative improvements arise because overconfident homeowners mistakenly believe

they can add to structures for a lower cost than a competitive construction in-

dustry. Improvements are increasing and convex in home prices. There is excessive

home remodeling in equilibrium. And the change in the recoup ratio (the ratio of

resale value of improvements to construction costs) is negatively correlated with

construction cost growth controlling for home price appreciation. We provide ev-

idence consistent with our theory using data on the costs and recoup values of

remodeling projects across U.S. cities.

In the third chapter, I test the pure consumption-motive remodeling which

tends to lead to negative relationship between remodeling expenditures and the re-

coup ratio. I find that remodeling expenditures are not significantly associated with

the remodeling recoup ratio, which rejects the null hypothesis. This indicates that

iv

investment-motive remodeling is also an important part of remodeling activities. I

also find that home price appreciation, the availability of mortgage credit for remod-

eling, per capita income, and housing supply elasticity are significant determinants

for home improvements.

v

Acknowledgments

I was able to finish my dissertation with great help from many people.

First of all, I extend my greatest gratitude to my main advisor, Harrison Hong.

Throughout my entire doctoral period, he was the one who guided and challenged

me with various topics and questions. Under his guidance, I improved my under-

standing and knowledge of behavioral finance, and developed my main research

field– real estate finance. He was not only an academic advisor, but was also a

great mentor with whom I could discuss many deep concerns. He always treated

me as a close friend and was extremely supportive on every issue.

I am also indebted to many of my colleagues at Princeton University. Wei Xiong

was my co-advisor in my early period of study. He was always there when I needed

him and supported me in my academic progress. David Sraer, Jose Scheinkman,

and Hyun Song Shin also nourished my academic thinking. I also thank Dong Beom

Choi, Eleanor Jawon Choi, Sung Mun Choi, Han Jun Kim, Yoon Soo Park, and

Woong Yong Park, –my great friends throughout my time at Princeton. I especially

thank Dong Beom Choi, with whom I spent my whole Princeton life from start to

finish. It is simply impossible to find any Princeton memory without him.

Above all, I thank to my family –my wife, (forthcoming) daughter, and par-

ents.... There are no proper words to describe their endless support and love. I owe

them so much that I cannot repay all of my debt during my lifetime, but I will

definitely try.

CONTENTS vi

Contents

Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii

Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v

1 The Impact of Anti-Predatory Lending Laws 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 The Anti-Predatory Lending Law . . . . . . . . . . . . . . . . . . . 7

1.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

1.4 The Effect on Home Mortgage Origination . . . . . . . . . . . . . . 22

1.5 The Effect on Household Expenditure . . . . . . . . . . . . . . . . . 28

1.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

1.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2 Speculating on Home Improvements 49

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

2.2 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

2.3 Comparative Statics . . . . . . . . . . . . . . . . . . . . . . . . . . 62

2.4 Empirical Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

3 The Motive of Home Improvements 90

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

3.2 Consumption vs Investment . . . . . . . . . . . . . . . . . . . . . . 94

CONTENTS vii

3.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

3.4 Time Series Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 101

3.5 Cross-sectional Analysis . . . . . . . . . . . . . . . . . . . . . . . . 103

3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

Bibliography 114

List of Figures

1.1 The MSA-level CFED index . . . . . . . . . . . . . . . . . . . . . . 44

1.2 The Mortgage Rate and Home Mortgage Origination . . . . . . . . 45

1.3 The Home Price Index and Home Mortgage Origination . . . . . . . 46

1.4 The Home Price Index by Elasticity of Housing Supply . . . . . . . 47

1.5 By Elasticity of Housing Supply . . . . . . . . . . . . . . . . . . . . 48

2.1 Average COST and RECOUP . . . . . . . . . . . . . . . . . . . . . 88

2.2 %∆Cost and %∆RECOUP . . . . . . . . . . . . . . . . . . . . . . . 89

3.1 Mortgage Credit Availability and Remodeling Expenditures . . . . 113

viii

List of Tables

1.1 Indices for the State Anti-Predatory Lending Laws . . . . . . . . . 34

1.2 The Coverage of the HMDA Data . . . . . . . . . . . . . . . . . . . 36

1.3 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

1.4 The Anti-Predatory Lending Law on Home Mortgage Origination . 39

1.5 By Loan-to-Income ratio . . . . . . . . . . . . . . . . . . . . . . . . 40

1.6 The Anti-Predatory Lending Law on Household Expenditure . . . . 42

2.1 List of Projects and Cities . . . . . . . . . . . . . . . . . . . . . . . 81


2.3 %∆RECOUP on %∆COST . . . . . . . . . . . . . . . . . . . . . . 83

2.4 The Robustness of Results . . . . . . . . . . . . . . . . . . . . . . . 84

2.5 Supply Elasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

3.1 Expenditures on Remodeling . . . . . . . . . . . . . . . . . . . . . . 107


3.3 Determinants of Remodeling Activities (Time Series) . . . . . . . . 111

3.4 Determinants of Remodeling Activities (Cross-section) . . . . . . . 112

ix

CHAPTER 1. THE IMPACT OF ANTI-PREDATORY LENDING LAWS 1

Chapter 1

The Impact of Anti-Predatory

Lending Laws

1.1. Introduction

In this paper, I test the impact of anti-predatory lending laws, examining: 1)

whether the laws inhibited mortgage volume during the recent housing-bubble pe-

riod; and 2) whether restricted mortgage volume as a result of these laws reduced

household consumption.

Predatory mortgage lending has been defined generally as a variety of unfair or

deceitful lending practices, directed at “vulnerable populations,” that often result

in serious personal losses, including bankruptcy and foreclosure. Predatory lending

practices in the home mortgage market grew with the expansion of mortgage credit.

To cope with increasing evidence of predatory lending practices, Congress en-

acted the Home Ownership and Equity Protection Act (HOEPA) in 1994. However,

HOEPA was not sufficient to curb predatory lending in the mortgage market. Con-

sequently, starting with North Carolina in 1999, many states enacted stricter laws

limiting predatory mortgage lending. As of January 2007, 31 states and the District


of Columbia had anti-predatory lending laws in effect. I use cross-state variation in

the timing of law adoptions and the strictness of these laws to test the hypotheses.

The impact of anti-predatory lending laws is a little-studied but important ques-

tion, because these laws are closely connected to the two characteristic features of

the recent housing crisis. Specifically, during the housing-bubble period, loose lend-

ing standards caused excessive mortgage lending.1 The extended credit was heavily

concentrated on populations that were the most liquidity-constrained, such as racial

minorities, the elderly, and the uneducated.2 Since anti-predatory lending laws were

designed to impose stricter lending standards on high-cost loans and the targets

of predatory lending coincided with the populations with extended credit, the role

of anti-predatory lending laws during the housing-bubble period is expected to be

significant. In fact, major settlements on abusive lending allegations in recent years

hint at the extent of predatory lending and the application of these laws (U.S.

General Accounting Office (2004)). In 2002, for example, Household International3

paid a $484 million to settle allegations that it used unfair and deceptive lending

practices. Citigroup paid a $240 million settlement in 2002 due to allegations that

subsidiaries of Citibank4 engaged in systematic and widespread abusive lending

practices.

Anti-predatory lending laws are different from traditional usury laws because

they do not impose any interest ceiling, but instead, require stricter lending stan-

dards on high-cost loans. HOEPA was the first anti-predatory lending law and

1See Dell’Ariccia, Igan, and Laevan (2009), Mian and Sufi (2009), Keys, Mukher-jee, Seru, and Vig (2010), and Demyanyk and Van Hemert (2009) regarding theloose lending standards.

2See Mian and Sufi (2009) and Rugh and Massey (2010) regarding the concen-trated mortgage lending to vulnerable populations.

3Household International, Inc. was the oldest, as well as the second largest,consumer finance company in the United States. On March 28, 2003, HSBC acquiredHousehold International, which was merged in 2005 with a subsidiary company ofHSBC that became the HSBC Finance Corp.

4Associates First Capital Corporation and Associates Corporation of NorthAmerica


the prototype for state anti-predatory lending laws. HOEPA applied to closed-

end loans secured by the borrower’s principal residence, and did not cover home

purchase loans. Once a loan met the criteria for HOEPA, lenders had to provide

more detailed disclosures to the borrower. Under HOEPA, balloon payments, neg-

ative amortization, prepayment penalties, and loan flipping were partially or fully

restricted. Government-sponsored enterprises (GSEs), such as the Federal Home

Loan Mortgage Corporation (Freddie Mac) and the Federal National Mortgage As-

sociation (Fannie Mae), were not allowed to purchase mortgage loans covered under

HOEPA.

Since HOEPA was deemed insufficient to curb abusive practices, many states

enacted laws further limiting predatory mortgage lending to fill HOEPA’s gaps.

State anti-predatory lending laws followed HOEPA’s structure but employed vary-

ing levels of strictness based on their own discretion. In addition, many of the

states included home purchase loans under the state anti-predatory lending laws.

As a result, state laws varied in the strictness of their provisions. Common HOEPA

structure makes it convenient to compare the strictness of the laws across states.


laws in a fixed-effect model to estimate the impact of these laws on total mortgage

volume. I use the anti-predatory lending (APL) law index from the Corporation

for Enterprise Development (CFED), a nonprofit organization that evaluates state

policies, to evaluate the strictness of the various state laws. I develop a time-varying

CFED Index by interacting the CFED Index with the effective dates of state APL

laws. I use data from the Home Mortgage Disclosure Act–the most comprehensive

data available on home mortgage origination–for mortgage origination.

Consistent with my hypothesis, I find that states with stricter laws had lower

mortgage volume. On average, a unit increase in the APL law index decreases the

amount of mortgage originations by 1.21%, and decreases the number of mortgage


applications by 1.74% when all other variables are held constant. The law index

ranges from 0 to 12. 0 is the pre-law level of law strictness, which equals to the

federal law. 12 is the maximum strength in the law index such as North Carolina and

New Mexico. Considering that post-law mean level of the index is 4.1, moving from

pre-law level to post-law mean level will reduce the amount of mortgage originations

by 4.9% and reduce the number of mortgage applications by 7.1%.

I also find that a unit increase in APL law index decreases the mortgage denial

rate by 0.475%.5 Moving from pre-law level to post-law mean level will reduce

the mortgage denial rate by 1.9%. The denial rate is defined as the number of

denied loans to the number of loan applications. The denial rate can decrease if the

reduction in the number of loan applications comes from the loan applications that

are more likely to be denied. When stricter laws prevent mortgage lenders from

originating imprudent loans, loan applications that are more likely to be denied

can decrease further.

Whether the reduction in mortgage volume comes solely from inhibited preda-

tory loans remains an empirical question. In principle, the anti-predatory lending

laws can also inhibit legitimate high-cost loans. Previous studies on the laws’ effect

showed that the reduction in subprime mortgage volume mostly came from high-

cost loans with predatory terms (Elliehausen, Staten, and Steinbuks (2006); Li and

Ernst (2007)). I also find that the reduction in mortgage volume becomes larger

as the Loan-to-Income ratio increases. Since the Loan-to-Income ratio is one of the

frequently used proxy measure for the riskiness of a loan, this result shows that

the laws had a bigger impact on loans with higher risk characteristics. A larger

reduction of the denial rate in high Loan-to-Income ratio loans indicate a larger

improvement in the quality of loan applications among the group.

5This coincides with the findings in previous literatures (Ho and Pennington-Cross (2006) and Bostic, Engel, McCoy, Pennington-Cross, and Wachter (2008)).


I also test whether, by restricting mortgage volume, these laws influenced house-

hold expenditures. The wealth effect of housing value is limited by the unique char-

acteristic of the housing asset–that is, people have to live somewhere. Increased

home equity can be realized mainly through mortgage refinancing. Mian and Sufi

(2011) found that money extracted through refinancing was not used for the repay-

ment of existing debts, and concluded that the money was used for “real outlays”

such as consumption or home improvements.

I use the result that the APL laws inhibited mortgage volume, which includes

mortgage refinancing, to estimate a reduced form model for the APL laws’ effect on

household expenditures. Again, I use the time-varying CFED Index for law strict-

ness and use the Consumer Expenditure Survey by the Bureau of Labor Statistics

for household expenditures. Due to the small sample size in the household expen-

ditures, statistical significance of results is limited, but I find that strict APL laws

are associated with total expenditure reductions. On average, a unit increase in

the law index reduces total expenditures by $161.9 when all other variables are

held constant. Moreover, reductions are also found in various categories of expen-

ditures, including Housing, Transportation, and Cash Contributions. On average,

a unit increase in the law index reduces annual expenditures on Housing by $53,

Transportation by $48.73, and Cash Contributions by $55.32.

This paper contributes to the policy literature regarding the effect of anti-

predatory lending laws on mortgage volume. While I focus on the laws’ effect on

overall mortgage volume using cross-state variation in the timing of law adoptions

and the strictness of the laws, other studies focused on the laws’ effect on subprime

mortgage volume using static cross-state variation in the strictness of the laws.

Starting with an evaluation of the North Carolina anti-predatory lending law, other

national-level studies followed as the number of states with such laws increased. Al-


though the results vary with data and sample periods6, there is a consensus that

subprime applications and originations decline with APL law adoption, especially

among lower-income borrowers, minority groups, and non-bank subprime lenders.

In addition, significant reductions are found in the volume of high-cost loans and

loans with predatory terms.

This paper also contributes to the literature on the wealth effect of housing

value on household consumption. Since rising home values also increase the user

cost of housing, a direct wealth effect of housing on household consumption has

been denied in previous literature.7 However, Muellbauer (2008) found that the

home equity market plays a pivotal role on the housing wealth effect, by mobiliz-

ing increased home equity. In addition, Cooper (2009) found that the wealth ef-

fect appears only among budget-constrained households. Hurst and Stafford (2004)

also found that liquidity-constrained households convert two-thirds of every dollar

from home equity extraction into consumption. This paper complements previous

studies regarding the wealth effect on household consumption, by analyzing the

effect of various state laws that seek to control mortgage volume, especially toward

predatory-targeted populations.

The rest of the paper proceeds as follows: Section 2 explains the state anti-

predatory lending laws and analyzes the various indices for the cross-state law

6For analyzing the NC law’s effect, different data sets with different sampleperiods are used in the studies. The HMDA data is used by Harvey and Nigro(2004) (1998-2000) and Burnett, Finkel, and Kaul (2004) (1997-1998, 2000-2002).Subprime loans in the HMDA data are identified using the HUD list of subprimelenders. Elliehausen and Staten (2004) use the subprime data from the American Fi-nancial Services Association (1995-2000) and Quercia, Stegman, and Davis (2004)use the Loan Performance data (1998-2002). For the national-level studies, theHMDA data is used by Ho and Pennington-Cross (2006) and Bostic et al. (2008).Elliehausen, Staten, and Steinbuks (2006) use proprietary data on subprime mort-gages by the subprime subsidiaries of eight large financial institutions from 1997 to2004. Li and Ernst (2007) use the Loan Performance data on securitized subprimeloans from 1998 to 2005.

7Case, Quigley, and Shiller (2005); Calomiris, Longhofer, and Miles (2009); Car-roll, Otsuka, and Slacalek (2011). See Peek (2010) for detailed literature review.


variation; Section 3 explains and summarizes the data; Section 4 reports the results

regarding the effect of anti-predatory lending laws on home mortgage origination;

Section 5 reports the results regarding the effects of anti-predatory lending laws

on household expenditure; Section 6 provides additional discussions; Section 7 con-

cludes.

1.2. The Anti-Predatory Lending Law

Home mortgage loans are types of loan that are collateralized by the value of

the underlying residential property. Homes account for a large part of household

wealth but are normally illiquid and indivisible assets. Mobilizing home equity

is important for households’ lifetime consumption smoothing, especially for those

who are liquidity-constrained. In fact, home mortgage loans are one of the most

commonly used financial products in the United States.

Predatory lending has been defined generally as a variety of unfair or deceitful

lending practices, aimed at “vulnerable populations”, that result in serious personal

losses, including bankruptcy and foreclosure.8 Quantifying the extent of predatory

lending is impracticable due to the absence of a precise definition of the term, since

the abusiveness of any particular loan depends on the overall context of the loan

and the borrowers. However, major settlements on abusive lending allegations in

recent years hint at the extent of predatory lending (U.S. General Accounting Of-

fice (2004)). For example, in October 2002, Household International, a subsidiary

of HSBC, paid a $484 million settlement on allegations by attorneys general in

8The formal definition of predatory lending can be found in Engel and McCoy(2002). They defined predatory lending as “a syndrome of abusive loan terms orpractices that involve one or more of the following five problems: 1) loans struc-tured to result in seriously disproportionate net harm to borrowers; 2) harmful rentseeking; 3) loans involving fraud or deceptive practices; 4) other forms of lack oftransparency in loans that are not actionable as fraud; and 5) loans that requireborrowers to waive meaningful legal redress.”


46 states that it used unfair and deceptive lending practices. In September 2002,

Citigroup paid a $240 million settlement on charges by the Federal Trade Com-

mission and private parties alleging that Associates First Capital Corporation and

Associates Corporation of North America, subsidiaries of Citibank, engaged in sys-

tematic and widespread abusive lending practices.

Predatory lending in the home mortgage market emerged after the deregulations

of 1980s. Originally, state usury law regulated home mortgage loans as a part of

consumer loans. Usury laws, which prohibited any loans with interest rates higher

than the usury ceiling, caused an insufficient credit supply in the mortgage market.

Credit shortages increased when interest rates rose significantly during late 1970s.9

Because the home mortgage market was important for households and their

secured nature was different from the other consumer loans, deregulations in the

home mortgage market continued from late 1970s into the early 1980s. As part of

the deregulation, home mortgages were exempted from state usury laws, and the

use of alternative mortgage transactions was allowed.10 Home mortgage loans with

high interest rates and alternative mortgage features such as adjustable rates or

balloon payments became possible, including predatory home mortgage lending.

Behind the deregulation, there was a firm belief in the market. If mortgage

lenders competed and households were informed enough to choose the best deal

between different products, the market would price mortgage products properly

without any legal guidance. To allow for comparison by borrowers, the Truth in

9For example, the federal funds rate was 5.54% in 1977, 7.94% in 1978. By1979 it jumped up to 11.2%, in 1980 it was 13.35%, and in 1981 it was 16.39%.http://www.federalreserve.gov/releases/h15/data.htm

10The Depository Institutions Deregulation and Monetary Control Act of 1980preempted state usury ceilings from any mortgage secured by a first lien on residen-tial property. The Alternative Mortgage Transaction Parity Act of 1982 preemptedthe state statues that restricted the use of alternative mortgage transactions, suchas adjustable rate mortgage, balloon payments and negative amortization, fromthe loans secured by residential property. These deregulations ultimately set thestage for the subprime home equity industry today. More details can be found inMansfield (2000).


Lending Act of 1968, or TILA, required disclosures about terms and the cost of

mortgage loans. However, the TILA was not mandatory and additional regulation

was needed to curb increasing predatory lending practices.

The Anti-Predatory Lending Laws: Federal and State

In 1994, the Congress enacted the Home Ownership and Equity Protection Act, or

HOEPA, amending TILA to deal with increasing predatory lending practices in the

home equity market. HOEPA was the first comprehensive anti-predatory lending

law that applied to closed-end loans secured by a borrower’s principal residence,

other than home purchase loans. HOEPA regulated “high-cost loans”, which were

loans exceeding one of the following two triggers: (1) the annual percentage rate

(APR) exceeded the yield on comparable Treasury securities plus 8%11(10%) for

first-lien (subordinate-lien) loans; or (2) total points and fees exceeded the greater

of 8% of the total amount of the mortgage or a set dollar amount ($592 for 201112).

Once a loan met the criteria for HOEPA, lenders must provide more detailed

disclosures in addition to those required by TILA. Further, some loan features were

partially or fully restricted, including balloon payments, negative amortization,

prepayment penalties, and loan flipping.13 Lenders were not permitted to originate

mortgage loans based on the value of the collateral but were required to check a

borrower’s repayment ability. The government-sponsored enterprises (GSEs), such

as the Federal Home Loan Mortgage Corporation (Freddie Mac) or the Federal Na-

11The APR trigger for first-lien was 10% in original HOEPA. But it has beenlowered to 8% in the amendment of 2002.

12Exact dollar amounts are adjusted annually, based on the Consumer PriceIndex.

13Balloon payments for loans with less than five-year terms were restricted. Refi-nancing within twelve months without the best interest of borrowers (loan flipping)and prepayment penalties beyond five years after origination were banned. Negativeamortization and due-on demand clause were fully restricted.


tional Mortgage Association (Fannie Mae), were not allowed to purchase mortgage

loans that fell under HOEPA.

However, the federal regulation was not sufficient to curb ongoing predatory

practices in mortgage markets.14 Some critics have charged that the triggers for the

high-cost loans in HOEPA were too high. Facing problems with growing predatory

mortgage lending, states started to enact anti-predatory lending laws to bolster

consumer protection from these predatory practices. Starting in North Carolina

in 1999, stricter laws limiting predatory mortgage lending were adopted. As of

January 2007, 31 states and the District of Columbia had anti-predatory lending

laws in effect.

For their own anti-predatory lending laws, states followed HOEPA’s structure

but employed varying levels of strictness based on their own discretions. As a re-

sult, the strictness of anti-predatory lending laws differs by state. For example, in

North Carolina’s anti-predatory lending law, the APR trigger remained the same

as HOEPA, but the trigger for points and fees was reduced from 8% of the total

amount to 5% for loans under $20,000. North Carolina’s anti-predatory lending

law also prohibited balloon payments for all high-cost loans, while HOEPA only

banned balloon payments for loans with less than five-year terms. Consequently,

the North Carolina anti-predatory lending law covers a larger group of mortgages

than HOEPA, with tighter restrictions.

Many of the states also included home purchase loans under the state anti-

predatory lending laws. For example, HOEPA was extended to cover home purchase

loans in Massachusetts’s anti-predatory lending law while Illinois’s anti-predatory

lending law only covered mortgage refinancing as in HOEPA.

14See the U.S. Department of Housing and Urban Development (2000) on thecontinuing predatory lending practices after HOEPA.


Indices for the State Anti-Predatory Lending Laws

To capture the cross-state variations of the law strictness on predatory mortgage

lending, quantitative law indices have been developed in earlier studies. Indices

measured the strictness of state laws relative to HOEPA. However, due to the

complexity in the law variations, creating a quantitative index of a qualitative law

is far from simple. Accordingly, previous law indices showed significant differences.

In this section, I review the indices in previous studies and examine the differences

between the indices. Three different indices by state are reported in panel A in

Table 1. The Summary statistics of the indices are reported in panel B.

In the first index, the Corporation for Enterprise Development (CFED)15 rated

the strength of state regulations on curbing predatory lending based on information

obtained from the Center for Responsible Lending (CRL).16 CFED evaluated eight

features of state anti-predatory lending (APL) laws as they existed in 2007. Those

features were: 1) the type of loans covered; 2) points and fees triggers; 3) substantive

legal protections; 4) remedies available to borrowers; 5) the regulations on loan-

flipping; 6) prepayment penalties and; 7) sound underwriting.17 The CFED Index

was the sum of the scores of the eight law features. Column (1) in Table 1.A

15The Corporation for Enterprise Development is a national nonprofit organi-zation based in Washington, DC. CFED publishes research, partners with localpractitioners to carry out demonstration projects and engages in policy advocacywork at the local, state and national levels. CFED provides state policy measureson various subjects. More details can be found at http://scorecard.cfed.org/.

16The Center for Responsible Lending (CRL) is a nonprofit , non-partisan re-search and policy group based in Durham, North Carolina, and with offices inWashington, DC and Oakland, California. Its purpose is to educate the publicabout financial products and to advocate for policies that curb predatory lending.CRL is affiliated with the Center for Community Self-Help.

17The CFED Index was constructed by a method analogous to the method usedby Li and Ernst (2007)(LE Index). Li and Ernst (2007) ranked state APL lawsaccording to six criteria including the type of loans covered, points and fees trig-gers, substantive legal protections, and remedies available to borrowers. Two morecriteria were added to the CFED Index, in addition to the six criteria of the LEIndex. The CFED Index and the LE Index are highly correlated (the correlation is0.9). Knowing that Li and Ernst were affiliated with the CRL and the CRL pro-vides information to the CFED, the close relationship between two indices is notsurprising.


reports the CFED Index. North Carolina, New Mexico, and Massachusetts were in

the group of strict APL law states. In contrast, Arizona, California, and Nevada

were in the group of weak APL law states.

In the second index, Ho and Pennington-Cross (2006) developed a two-component

index of state APL laws as they existed in 2005. Hereafter, the index will be de-

noted as the HP Index. The HP Index consisted of two subindices: 1) the Coverage

Index and; 2) the Restriction Index. The Coverage Index measured the breadth of

law coverage, and the Restriction Index measured the strictness of law restriction.

The Coverage Index was defined as the sum of ratings in four criteria regarding

the coverage of the law. The criteria for law coverage included the type of loans

covered, APR triggers, and points and fees triggers. Each criterion was rated by

the relative strictness of the state law to HOEPA. Construction of the Restriction

Index was similar to the Coverage Index. Criteria regarding law restriction included

restrictions on prepayment penalties and balloon payments. The Full Index was the

sum of the Coverage Index and the Restriction Index. Columns (2)-(4) in Table 1.A

report the HP Index. The index covered only 25 states. In terms of the Full Index,

Illinois, Colorado, and Georgia were the strictest APL law states. And Nevada,

Maine, and Florida were the weakest APL law states. The Coverage Index and the

Restriction Index were weakly correlated. The correlation between the two indices

was 0.35 (Table 1.C). For example, Colorado was one of the strict APL law states

in terms of the Coverage Index but it was not strict in terms of the Restriction

Index. This means that the Colorado APL law covered broader types of mortgage

loans but the restrictions on the covered loans were weak.

In the third index, Bostic et al. (2008) extended the HP Index by introducing

an additional index for law enforcement. Hereafter, that index will be denoted as

the BEA Index. The three subindices of the BEA Index were the Coverage Index,

the Restriction Index, and the Enforcement Index. The Coverage Index and the


Restriction Index were analogous to the corresponding indices in the HP Index.

The Enforcement Index measured the scope of liability and the strength of available

legal actions when violations occurred in loan origination. The Enforcement Index

was defined as the sum of the ratings in two criteria regarding law enforcement

(i.e. assignee liability). The Enforcement Index increases with larger responsibility

and stronger penalties on violations. For example, if assignees of mortgages were

liable even after the exercise of due diligence, the Enforcement Index would be high.

The BEA index covered all the states. The Full Index is defined as the sum of the

three indices. Columns (5)-(8) in Table 1.A report the BEA Index. According to

the Full Index, New Mexico, West Virginia, and Massachusetts are the states with

the strictest APL laws.

In Panel C of Table 1, the correlation matrix of the indices is reported. In

summary, the correlation matrix shows some level of consistency among the indices,

but also shows significant differences among them. First, all indices are positively

correlated. The law variation roughly coincides within the indices. Second, the

correlations within the indices based on components of the law that are similar

tend to be higher than the correlations across the indices based on components

of the law that are different. For example, the Coverage Index of the HP Index is

more correlated with the Coverage Index of the BEA Index than with any other

indices. The Restriction Index of the HP Index shows the highest correlation with

the Restriction Index of the BEA Index. Third, the indices based on components of

the law that are similar are not highly correlated, considering that the indices are

intended to capture similar law variations. The correlation between the Coverage

Indices is 0.57 and the correlation between the Restriction Indices is 0.63. This

shows significant differences across law indices. Last, the CFED Index and the

Enforcement Index of the BEA Index are more correlated with the Restriction

Indices than the Coverage Indices.


There are several reasons for the differences. First, the major difference emanates

from the selection of index criteria for measuring law strictness. Note that all of the

indices have been constructed by the weighted sum of the score from the criteria.

It is relatively straightforward to rank the laws by a single criterion since the state

APL laws have followed HOEPA’s structure. For example, Utah has maintained the

points and fees trigger as in HOEPA (8% of total loan amount), but North Carolina

has lowered the trigger to 5% for loans under $20,000. Therefore North Carolina

has stricter regulations than Utah in terms of the points and fees trigger. In cases

of the restriction on prepayment penalties, Pennsylvania has maintained the five-

year periods of prepayment penalties as in HOEPA, but Massachusetts has banned

all prepayment penalties for high-cost loans. Therefore Massachusetts has stricter

regulation than Pennsylvania on prepayment penalties. However, the selection of

criteria has been an issue and has generated the divergence found in the law indices.

The CFED index evaluated eight features of the law, which were selected by the

experts at CRL. The criteria were chosen by authors for the HP Index and the BEA

Index. The Coverage Indices and the Restriction Indices included four criteria and

the Enforcement Index included two criteria.

Another reason for the differences in the law indices stems from the inclusion

of non-HOEPA-type regulations. Following HOEPA, many states introduced “com-

prehensive” APL laws to curb predatory lending. “Comprehensive APL laws” refer

to complete laws, uniquely designed to curb predatory lending in the home mort-

gage market. In column (9) of Table 1.A, effective dates of the state APL laws

are reported.18 An effective date exists only if the state adopted a comprehensive

anti-predatory lending law.19 However, there are some states that have selective

18The source of data include the Standard & Poor’s Predatory Lending Categoriesand summary reports from Butera & Andrews, a law firm.

19Having an effective date does not directly mean that the state law is strict. Somestates have enacted the state laws but the laws have been only a copycat statutes


regulations on the recurrent practices of the predatory lending, even though no com-

prehensive APL law has been enacted. These selective regulations existed in other

consumer credit laws. For example, Iowa has restrictions on prepayment penalties

in home mortgage loans, but has not enacted any comprehensive APL law.20 These

non-HOEPA-type regulations are incorporated in the CFED Index but not in the

HP Index or the BEA Index.21

Unfortunately, the law indices are not perfect and are not free from some limi-

tations in measuring state anti-predatory lending laws. First, all of the law indices

are static and evaluate state anti-predatory lending laws as they existed at a certain

point in time. The CFED Index evaluates APL laws as they existed in 2007 while

the HP Index and the BEA Index evaluates the laws as they existed in 2005. The

effective dates of the state laws are reported in column (9) of Table 1. Note that

states adopted anti-predatory lending laws during the sample period of 1998-2007.

Second, many states included home purchase loans in their own state laws, but

some did not. However, none of the law indices make that distinction. The coverage

of the state anti-predatory lending laws in terms of loan purpose is reported in

column (10) of Table 1. Among 32 states with the state APL laws, 11 states had

APL laws that only cover mortgage refinancing loans and 21 states had APL laws

that also cover home purchase loans.

The Anti-Predatory Lending Law Index for Analysis

For the state APL law index in following analysis, I mainly use the CFED Index to

show the aggregate effect of the law. The CFED Index has several advantages over

of HOEPA. Those include Florida, Kentucky, Maine, Nevada, Ohio, Oklahoma,Pennsylvania, and Utah (Ding, Quercia, and White (2009)).

20Details can be found in the chapter 535.9 of the Iowa law.21Bostic et al. (2008) also constructs additional indices on the non-HOEPA-type

regulations. The indices show low correlations with the indices in table 1. Detailsare omitted for brevity.


other indices for measuring aggregate law effect. First, in terms of closeness to the

first principal component of the law indices, the CFED Index is a representative law

index. The principal component analysis (PCA) is used to identify the core varia-

tion among different law indices in Table 1. The first principal component explains

51% of the variations within the indices. Further, the first principal component is

most highly correlated with the CFED index, with a correlation of 0.87. Second,

the CFED index is a single dimensional index. Multi-dimensional measures, such

as the BEA Index and the HP Index, are useful to understand partial law effects.

But without a reasonable weighting method, it is difficult to judge the aggregate

law effect, especially when partial effects conflict with each other. Lacking a gen-

erally accepted method for aggregating multi-dimensional measures, Bostic et al.

(2008) proposed an additive index, which is the Full Index. However, the Full Index

assumes that partial effects from the Coverage Index and the Restriction Index

have equal weights on aggregate law effect. Third, the CFED Index includes non-

HOEPA-type regulations. By ignoring non-HOEPA regulations, other indices may

underestimate the law strictness in some states. Last, the CFED Index exists for

all the states while the HP Index is only available for the half of the states.

To overcome the limitation of static law index, I develop a time-varying CFED

Index by interacting the CFED Index with the effective dates of state APL laws:

APLi · I{t>=ED}, where, the i and t index, respectively, represent MSA and year.

The effective date (ED) of the state law is reported in column (9) of Table 1. As

a result, the time-varying law index captures cross-state variation in the timing of

law adoptions as well as the strictness of the laws.

The metropolitan statistical area (MSA) as of Dec 200922 is used as a main

regional unit of interest. Considering the local characteristics of the housing market,

22See http://www.whitehouse.gov/sites/default/files/omb/assets/bulletins/b10-02.pdf.


MSA-level analysis is more appropriate than state-level analysis. For applying state-

level law indices to MSA-level analysis, state-level law indices are mapped into a

MSA-level variable. For those MSAs within a state, the state law index has been

assigned. For those MSAs across several different states, a population-weighted

state index has been assigned. There are 44 MSAs across multiple states.23 For

example, in the Chicago-Naperville-Joliet, IL-IN-WI MSA for the years 1998-2007,

on average, 91% of the population lived in Illinois, 7% lived in Indiana, and 2% lived

in Wisconsin. As a result, the law index assigned to Chicago-Naperville-Joliet, IL-

IN-WI MSA is closest to the law index of Illinois. Figure 1 reports the MSA-level

CFED Index on the map of the United States.

1.3. Data

I have collected data from several sources for the analysis including: 1) loan-level

residential mortgage origination data from the Home Mortgage Disclosure Act

(HMDA); 2) MSA-level consumer expenditure data from the Consumer Expen-

diture Survey (CES) prepared by the Bureau of Labor Statistics (BLS); 3) MSA-

level house price index (HPI) from the Federal Housing Finance Agency (FHFA);

4) MSA-level mortgage rates from the Monthly Interest Rate Survey (MIRS) by the

FHFA; 5) MSA-level elasticity of housing supply data from Saiz (2010); 6) MSA-

level unemployment rates from the BLS and; 7) MSA-level income and population

data from the Bureau of Economic Analysis (BEA).

2344 MSAs include the Boston-Cambridge-Quincy, MA-NH MSA, the Charlotte-Gastonia-Concord, NC-SC MSA, the Chicago-Naperville-Joliet, IL-IN-WI MSA,the Minneapolis-St. Paul-Bloomington, MN-WI MSA, the New York NorthernNew Jersey-Long Island, NY-NJ-PA MSA, the Philadelphia-Camden-Wilmington,PA-NJ-DE-MD MSA, and the Washington-Arlington-Alexandria, DC-VA-MD-WVMSA.


Loan Origination: The Home Mortgage Disclosure Act (HMDA)

Enacted by Congress in 1975, the Home Mortgage Disclosure Act, or HMDA, re-

quires most mortgage originators located in metropolitan areas to report basic

attributes of mortgage applications. HMDA covers a broad set of financial insti-

tutions, both depository and non-depository. Inclusion of an institution depends

on the size of its origination and the weight of residential mortgage lending in its

portfolio.24

Because HMDA covers a wide number of loans across the country, using HMDA

offers great advantage in analyzing loan origination process. HMDA is known as

the most comprehensive source of mortgage origination data, and covers almost

80% of all home loans nationwide (Avery, Brevoort, and Canner (2007b)). Table 2

reports the scope of the HMDA data by comparing it with the mortgage origination

estimate from the Mortgage Bankers Association (MBA).25 From 1998 to 2007, on

average, the HMDA data covers 84% of total originations, 94% of total originations

in the refinance market, and 76% of total originations in the home purchase market.

HMDA data includes loan characteristics such as loan type (conventional or

government-backed), loan purpose (home purchase, home improvement, and re-

financing), types of property (single-family and multifamily), loan amounts, and

location (state, county, and Census tract). HMDA data also includes borrower char-

acteristics such as income, race, and gender. In addition, HMDA data provides the

status of applications, that is, whether a loan was originated or denied.

24Any depository institution with an office in an MSA must report to HMDA if ithas originated any home purchase loans or has refinanced any home purchase loansand if it has assets above an annually adjusted threshold ($39 million for 2010).Any non-depository institution with at least 10% of its portfolio in home purchaseloans must report to HMDA if its assets exceed $ 10 million. As a result, smalllenders and lenders without any office in a MSA are not reported in the HMDAdata (Dell’Ariccia, Igan, and Laevan (2009)).

25See http://www.mbaa.org/ResearchandForecasts/ForecastsandCommentary.


I use HMDA data from 1998 to 2007. This includes the period of the recent

housing market boom, which is the year of interest for this study. To analyze the

impact of state APL laws on mortgage loan origination, I focus on conventional

single-family loans.26 Following the suggestion by Avery, Brevoort, and Canner

(2007a), I categorize home improvement loans as refinancing. Using the five-digit

Federal Information Processing Standard (FIPS) code, I aggregate loan-level data

into MSA-level data using the MSA definition as of December 2009. In total, 363

MSAs are defined.

In Table 3, I present the summary statistics of variables. Panel A in Table 3

summarizes the variables from HMDA data. The number of applications (APPL),

the amount of originations (AMTO), and denial rates (DR) are reported by loan

purpose: all loans, home purchase loans, and refinance loans. First, refinance loans

show a greater average number of applications and a greater average amount of

originations than applications and originations for home purchase loans. The av-

erage AMTO is $2.93 billion for refinance loans but $2 billion for home purchase

loans. The average APPL in refinance loans is twice as large as in home purchase

loans. Second, the standard deviations of APPL and AMTO are also larger in refi-

nance loans than in home purchase loans, implying larger variations of APPL and

AMTO across MSAs in mortgage refinancing. Last, refinance loans show a higher

average denial rate (42.74%) than home purchase loans (28.84%).

Expenditure: The Consumer Expenditure Survey (CES)

To measure household economic activity, the expenditure data from the Consumer

Expenditure Survey (CES) is used. The survey data is collected for the Bureau of

Labor Statistics by the U.S. Census Bureau to provide information on the buying

26Government-backed loans are regarded as safer mortgages and less likely to beaffected by abusive mortgage practice. Normally, multi-family loans are not coveredunder the APL laws.


habits of American consumers, including data on their expenditures, income, and

consumer unit (families and single consumers) characteristics. The data provides

households’ expenditures in detailed categories: food, housing, transportation and

alcoholic beverages.

Due to confidentiality issues, specific geographic information of sample house-

holds is missing in public microdata. However, BLS additionally reports MSA-level

expenditures for some major MSAs.27 The number of available MSAs changes from

period to period: 28 MSAs from 1998 to 2004, 24 MSAs in 2005, and 18 MSAs from

2006 onward. In spite of the small number of samples, those MSAs encompass wide

coverage and include most of the representative MSAs in the country: Atlanta ,

Baltimore, Boston, Chicago, Cleveland, Dallas-Fort Worth, Detroit, Houston, Los

Angeles, Miami, Minneapolis-St.Paul, New York, Philadelphia, Phoenix, San Diego

, San Francisco, Seattle, and Washington, D.C.

Panel C in Table 3 summarizes average consumer expenditures for 1998 to 2007.

Total annual expenditures are on the first row and detailed breakdowns by expen-

diture purpose follows. The mean of annual consumer expenditures was $47,176.

Among MSAs with a full sample, the highest average expenditure during the sam-

ple period ($59,267) was reported in San Francisco, CA, and the lowest average

expenditure ($40,727) was in Cleveland, OH. On average, expenditures for housing

(35%), transportation (18%), and food (13%) add up to two-thirds of total expen-

ditures.

Local Housing Market and Demographics

To control for the condition of local housing markets, three variables are consid-

ered. First is the house price index (HPI) from the Federal Housing Finance Agency

27See http://www.bls.gov/cex/csxmsa.htm


(FHFA).28 The HPI Index is available for all 363 MSAs. Second, the regional mort-

gage rates are from the Monthly Interest Rate Survey (MIRS) by the FHFA.29 The

survey provides annual information on interest rates, loan terms, and house prices

by state. Data for all of 51 states is available.

Lastly, the elasticities of housing supply are from Saiz (2010). Using satellite-

generated geographic data, he measured the supply elasticity of housing as a func-

tion of both the amount of developable land and the regulatory constraints.30 How-

ever, the supply elasticities are defined by the MSA definition as of 1999 and a

significant change in the MSA definition occurred in 2000. I link up 1999 MSAs

with current MSAs by hand-matching counties in MSAs. In total, I find 280 MSAs

according to the current definition where the housing supply elasticity is applicable.

In addition to the controls for the regional housing market, local demograph-

ics are considered. MSA-level unemployment rates are taken from the BLS data.

Average personal income and population of MSAs are taken from the Bureau of

Economic Analysis (BEA). Demographic variables are available for all 363 MSAs.

Panel B in Table 3 summarizes the variables of local housing market conditions

and local demographics, from 1998 to 2007. The effective interest rates on mortgage

(EFFINT) are a variable by state. Annual home price appreciation (HPIAPP),

28In 1975, the HPI was developed by the Office of Federal Housing EnterpriseOversight, or OHFEO as a regulator of Fannie Mae and Freddie Mac. In 2008,OHFEO became a part of FHFA. The HPI was called the OHFEO HPI, but nowis called as the FHFA HPI. The FHFA HPI is a weighted, repeat-sales index ofsingle-family house price. It means that the HPI measures average price changesin repeat sales or refinancings on the same properties in 363 MSAs. Informationis obtained by the mortgage transactions that have been purchased or securitizedby Fannie Mae or Freddie Mac. Since the HPI Index only includes houses withmortgages within the conforming amount limits, the index has a natural cap anddoes not account for jumbo mortgages.

29The survey was conducted by asking a sample of mortgage lenders to reportthe terms and conditions on all single-family, purchase-money, non-farm loans. Thesurvey excludes FHA-insured and VA-guaranteed loans, multifamily loans, mobilehome loans, and loans created by refinancing another mortgage. Although the dataonly incorporates home purchase loans, it is still useful to control for the leveldifferences between the regional mortgage rates.

30Data is available from Saiz’s website. http://real.wharton.upenn.edu/∼saiz/.


housing supply elasticity (Elasticity), unemployment rates (UNEMP), per capita

income (INC), and the log number of populations (logPOP) are the variables by

MSA. For the housing supply elasticity, the Los Angeles-Long Beach-Santa Ana,

CA MSA shows the lowest elasticity with 0.63 and the Pine Bluff, AR MSA shows

the highest elasticity with 12.15.

1.4. The Effect on Home Mortgage Origination

Home mortgage origination has been largely affected by housing prices, mortgage

interest rates, and housing supply elasticity. Here, the effects of these variables on

home mortgage origination have been examined by loan purpose, using the U.S.

aggregate data.

First is the effect of mortgage interest rates on the amount of originations

(AMTO) and on the denial rate (DR). When mortgage rates are low, households

can refinance existing mortgages with either lower periodic payments, or larger

cash-out amounts, or both. And households can borrow with lower interest cost

when they purchase a home. Considering that mortgage originators have supported

increased demand during the recent housing boom period (Dell’Ariccia, Igan, and

Laevan (2009)), the decline in mortgage rates will increase the AMTO and de-

crease the DR, in both refinance loans and home purchase loans. Figure 2 reports

the AMTO and the DR with the 30-year mortgage rate.31 The HMDA data is used

to report the U.S. aggregate AMTO and the U.S. average DR, by loan purpose. In

panel A of figure 2, the AMTOs in mortgage refinancing and in home purchases

are compared with the 30-year mortgage rate. AMTOs are negatively related to

the mortgage rate, both in mortgage refinancing and in home purchases, but the

31The conventional, conforming 30-year fixed-rate from Pri-mary Mortgage Market Survey by Freddie Mac is used(http://www.freddiemac.com/pmms/pmms archives.html).


negative correlation is more evident in refinancing loans. The AMTO in mortgage

refinancing rapidly increases from 2000 to 2003 and this period coincides with the

large decline of mortgage rates. After 2004, mortgage rates stopped declining and

the AMTO in mortgage refinancing dropped considerably. In panel B, the DR in

mortgage refinancing and in home purchases are compared with the 30-year mort-

gage rate. DRs are positively correlated with the mortgage rate. In both markets, a

large drop in the DR appeared with the large decline of mortgage rates from 2000

to 2003.

Second is the effect of housing price on the AMTO and on the DR. When home

prices increase, households can refinance their home loans to cash-out the increased

home equity. Although the role of home price appreciation on the demand for

home purchase loans is not obvious, the demand will increase if households expect

a momentum in home prices. Mortgage originators are more willing to originate

home mortgage when the collateral value rises. Figure 3 reports the AMTO and

the DR with the U.S. national Home Price Index (HPI) from FHFA. As before,

the HMDA data is used for the U.S. aggregate AMTO and the U.S. average DR,

by loan purpose. The HPI is compared with AMTOs in panel A and with DRs in

panel B, by loan purpose. While AMTOs show positive correlation with the HPI,

DR is negatively related to the HPI.

Third is the role of elasticity of housing supply. Historically, inelastic MSAs have

experienced precipitous increases of HPI. Given the supply constraint, when there

is a demand shock, price has reacted more sensitively in inelastic MSAs. Glaeser,

Gyourko, and Saiz (2008) show that the price run-ups of the 1980s were almost

exclusively experienced in cities where the housing supply is more inelastic. Figure

4 reports the time-series of the HPI by housing supply elasticity. MSAs are divided

into two groups: elastic MSAs and inelastic MSAs. Average HPIs of the groups are

plotted. The plot shows that HPI appreciation has been isolated to inelastic MSAs


during the recent housing boom. Larger mortgage demands in inelastic MSAs are

expected since the expected benefit from home price appreciation is high. Figure 5

reports the time series of the AMTO and the DR, by elasticity. The time-series of

the AMTO and the DR are reported, in panel A and panel B respectively, by loan

purpose and by supply elasticity. Panel A shows that the growth in the AMTO

has been concentrated in inelastic MSAs, both in the refinance market and in the

home purchase market. Panel B shows that the DR in refinancing tends to be low

in inelastic MSAs.

In summary, home price appreciation and the decline of mortgage rates tends

to promote home-based borrowing, by increasing the amount of originations and

by decreasing the denial rate. Growth in the amount of originations has been con-

centrated in inelastic areas.

The Impact of Anti-Predatory Lending Laws on Mortgage Volume

I test the hypothesis that anti-predatory lending laws inhibited mortgage volume

during the recent housing-bubble period. I use cross-state variation in the timing of

law adoptions and the strictness of the laws in a fixed-effect model to estimate the

impact of these laws on total mortgage volume. First specification is the panel re-

gression with year and MSA fixed-effects, following Jayaratne and Strahan (1996).

Yit = αi,t + β ·APLi · I{t>=ED} + εit

where, the i and t index, respectively, represent MSA and year. The years covered

by the regression are from 1998 to 2007. The denial rate (DR), the log amount

of originations (logAMTO), and the log number of applications (logAPPL), are

left-hand side variables. For the APL law variation, a time-varying CFED index

is used by interacting the CFED Index with the effective date of the state law

(APLi · I{t>=ED}). The effective date (ED) of the state law is reported in column


(9) of Table 1. For the robust level of significance, all standard errors are clustered

by MSA.

Columns (1)-(3) in Table 4 reports the regression results. The enactment of

strict APL law is associated with a significant reduction in the amount of mortgage

originations and the number of mortgage applications. On average, a unit increase

in the APL law index decreases the amount of mortgage originations by 1.21%, and

decreases the number of mortgage applications by 1.74% when all other variables

are held constant. Given that the APL law index ranges from 0 to 12, the results

represent sizable reduction in mortgage volume in states with the strictest laws.

The economic significance is calculated and reported under the APL variable. For

example, a one-standard deviation increase in the APL law index is associated with

a 0.02-standard deviation reduction in the logAMTO. (A one-standard deviation

increase in APL is associated with a 2.98 [1 SD] x 0.0121 [slope] = 0.036 reduction in

logAMTO, which is 0.036 / 1.46 = 0.02-standard deviation of DR.) A one-standard

deviation increase in the APL law index is associated with a 0.04-standard deviation

reduction in the logAMTO.

I also find that the enactment of strict APL law is associated with a significant

reduction in the DR. On average, a unit increase in APL law index decreases the

mortgage denial rate by 0.475%. A one-standard deviation increase in the APL

law index is associated with a 0.16-standard deviation reduction in the logAMTO.

The denial rate is defined as the number of denied loans to the number of loan

applications. The denial rate can decrease if the reduction in the number of loan

applications comes from the loan applications that are more likely to be denied.

When stricter laws prevent mortgage lenders from originating imprudent loans, loan

applications that are more likely to be denied can decrease further.


Second specification is the panel regression with MSA-level control variables, in

addition to year and state fixed-effects.

Yit = αs,t + β ·APLi · I{t>=ED} + γ0 ·HPIAPPit + γ1 · EFFINTst

+ γ2 · Elasticityi + θ ·Xit + εit

where, the i, s, and t index, respectively, represent MSA, state, and year. As above,

the years covered by the regression are from 1998 to 2007. The denial rate (DR),

the log amount of originations (logAMTO), and the log number of applications

(logAPPL), are left-hand side variables. For the APL law variation, a time-varying

CFED index is used by interacting the CFED Index with the effective date of the

state law (APLi · I{t>=ED}). Control variables include annual home price apprecia-

tion (HPIAPP), state-level mortgage rate (EFFINT), supply elasticity (Elasticity),

income per capita (INC), unemployment rate (UNEMP), and the log numbers of

population (logPOP). For the robust level of significance, all standard errors are

clustered by MSA.

Columns (4)-(6) in Table 4 reports the regression results. As in the first spec-

ification, the enactment of strict APL law is associated with a reduction in the

DR, the AMTO and the APPL. On average, a unit increase in the APL law index

decreases the mortgage denial rate by 0.457%, decreases the amount of mortgage

originations by 0.4%, and decreases the number of mortgage applications by 1.09%

when all other variables are held constant. Given that the APL law index ranges

from 0 to 12, the results represent sizable reduction in mortgage volume in states

with the strictest laws. The economic significance is calculated and reported under

the APL variable.

Among control variables, note that HPIAPP and Elasticity are the variables

that reflect local housing market conditions. With favorable market signals, the DR


declines, the AMTO increases, and the APPL increases. Favorable market signals

are high HPIAPP and low Elasticity. For other control variables, higher INC is

associated with a decline in the DR, an increase in the AMTO, and an increase in

the APPL. High UNEMP does the opposite. logPOP is positively associated with

the AMTO and the APPL.

Whether the reduction in mortgage volume comes solely from inhibited preda-

tory loans remains an empirical question. In principle, the anti-predatory lending

laws can also inhibit legitimate high-cost loans. Previous studies on the law effect

showed that the reduction in subprime mortgage volume mostly came from high-

cost loans with predatory terms (Elliehausen, Staten, and Steinbuks (2006); Li and

Ernst (2007)).

I find that the law impacts on mortgage volume becomes larger as the Loan-

to-Income ratio (LTI) increases. Regressions are same as in Table 4 except that

dependent variables are disaggregated by LTI. Table 5 reports the fixed-effect panel

regression results by LTI. Panel A reports the regression results with MSA fixed-

effect and panel B reports the regression results with state fixed-effect. In panel A,

columns (1)-(2) report the results on the DR, columns (3)-(4) report the results

on the logAMTO, and columns (5)-(6) report the results on the logAPPL. As in

Table 4, strict APL laws are associated with a significant decrease in the DR, the

AMTO, and the APPL. However, the magnitude of reduction is larger in the high

LTI groups. On average, a unit increase in the APL law index decreases the DR

by 0.17% in the high LTI group but by 0.11% in the low LTI group, decreases the

APPL by 1.57% in the high LTI group but by 0.07% in the low LTI group, and

decreases the AMTO by 1.59% in the high LTI group but by 0.4% in the low LTI

group. The economic significance is also larger in the high LTI group.

In panel B, columns (1)-(3) report the results on the DR, columns (4)-(6) re-

port the results on the logAMTO, and columns (7)-(9) report the results on the


logAPPL. The regression results are similar as in panel A. Strict APL laws are

associated with a larger decrease in the DR, the AMTO, and the APPL and the

magnitude of reductions are larger in the high LTI groups. For testing the differ-

ences of law effects between two LTI groups, pooled regression results are reported

in columns (3), (6), and (9) in panel B of Table 5. DIFF is defined as the differences

of law effect in the high LTI group relative to the low LTI group. The reductions are

significantly larger in the high LTI group. Since that LTI is one of the frequently

used proxy measures for the riskiness of a loan, this result shows that the laws

had bigger impact on the loans with higher risk characteristics. A significant larger

reduction of the denial rate in high Loan-to-Income ratio loans indicate a larger

improvement in the quality of loan applications among the group.

1.5. The Effect on Household Expenditure

During the recent housing boom period, favorable market conditions for household

credit coexisted in the home mortgage market: low mortgage rates, high home

price appreciation, and evolutions in the mortgage market such as securitization.

U.S. household leverage increased sharply during the period and a significant share

of the rise came from the borrowings against increased home equity by existing

homeowners. (Greenspan and Kennedy (2007); Khandani, Lo, and Merton (2010)).

However, Mian and Sufi (2011) found that money extracted through refinancing

was not used for other investments or the repayment of existing debts, postulating

that the money must have been used for “real outlays” such as consumption or

home improvements.

Using the finding that anti-predatory lending laws inhibited mortgage volume,

including mortgage refinancing, I test whether household expenditures have been

affected by mortgage availability in the recent housing boom. Considering the piv-


otal role of refinancing to the housing wealth effect on household consumption

(Muellbauer (2008); Cooper (2009); Peek (2010)), significant reductions in house-

hold expenditures is expected in states with strict APL laws.

I use the result that the APL laws inhibited mortgage volume to estimate a

reduced form model for the APL laws’ effect on household expenditures. Following

fixed-effect panel regression is applied to the household expenditures by purpose.

EXPit = αs,t + β ·APLi · I{t>=ED} + γ0 ·HPIAPPit + γ1 · EFFINTst

+ γ2 · Elasticityi + θ ·Xit + εit

where the i, s, and t index, respectively, represent MSA, state and year. The years

covered by the regression are 1998 to 2007. Left-hand side variables are the an-

nual average expenditures of the consumer unit. MSA-level household expenditure

data is from the Consumer Expenditure Survey (CES). Again, the time-varying

CFED Index is used for the APL law variation (APLi · I{t>=ED}). Control vari-

ables include annual home price appreciation (HPIAPP), state-level mortgage rate

(EFFINT), supply elasticity (Elasticity), income per capita (INC), unemployment

rate (UNEMP), and the log numbers of population (logPOP). All standard errors

are clustered by MSA.

Table 6 reports the regression results. Panel A reports the regression results on

total expenditures and panel B reports the results on the expenditures by purpose.

Panel C reports the results by subdividing the expenditure purposes in panel B. For

brevity, demographic control variables are not reported. Due to the small sample

size in the household expenditures, statistical significance of results is limited but I

find that strict APL laws are associated with total expenditure reductions in panel

A. On average, a unit increase in the APL law index reduces the total expenditure by

$161.9 while all other variables are held constant. The economic significance is also


reported. A one-standard deviation increase in the APL law index is associated with

a 0.05-standard deviation decline in total expenditures. (A one-standard deviation

increase in the APL law index is associated with a 2.12 [1 SD] x 161.9 [slope x -1]

= 343.91 decrease in total expenditures, which is 343.91 / 7475.56 = 0.05-standard

deviation of DR.)

Panel B reports the results by expenditure purposes. As above, small sample

size limits statistical significance but I find negative associations in various expen-

diture purposes, including Housing, Transportation, and Cash Contributions. On

average, a unit increase in the law index reduces annual expenditures on Housing by

$53, Transportation by $48.73, and Cash Contributions by $55.32. Economic sig-

nificances are large in some expenditure purposes. Panel C reports the regression

results by subdividing the expenditure purposes in panel B. In the Transportation

category, the expenditures on Gasoline and Motor Oil and Other Vehicle Expenses

show significant decline. In the category of Housing, significant reductions are found

in Rented Dwellings and Household Operations. Although it is not statistically

significant, the largest expenditure reduction comes from Shelter, which includes

Owned Dwellings and Rented Dwellings.32

1.6. Discussion

Considering that most of the predatory loans occurred in the subprime market,

analyzing the subprime market alone would show stronger results. However, an ef-

fective distinction between subprime loans and prime loans is unattainable in the

analysis due to data issues. First, HMDA does not identify whether a loan applica-

32The definition of Owned Dwellings includes expenses for repairs and mainte-nance (including DIY), interest on mortgages, refinancing/payment charges, andproperty insurance. The definition of Rented Dwellings includes rent paid fordwellings, parking fees, maintenance, and other expenses.


tion is for subprime or prime. Second, common methods for identifying subprime

loans in HMDA are not suitable for this study.

To separate subprime loans from prime loans, two different methods have been

commonly used by researchers: 1) the rate spread data in HMDA, or 2) the Sub-

prime and Manufactured Home Lenders list by the Department of Housing and

Urban Development (HUD).33 Since the HUD list is only available up to 2005 and

the rate spread in HMDA is available from 2004, inconsistency in defining subprime

lending is inevitable. Considering that the two methods have own inherent biases34

in measuring subprime volumes and that 2004-2005 were the years of extra impor-

tance, the change of definition in 2004-2005 makes it hard to separate the subprime

market results. When using the HUD list for identifying subprime loans in HMDA,

33http://www.huduser.org/portal/datasets/manu.html.34In 2004, HMDA started to report the rate spread to the comparable-maturity

Treasury for first-lien mortgages with an APR 3% points over the Treasury bench-mark and for junior liens with an APR 5% percentage points over the benchmark.Researchers interpreted the loans with the rate spread information as a proxy forsubprime loans. However, biases exist in this definition. Bias can be seen in thefollowing: First, high interest rates do not occur solely in subprime loans. Whenthe spread of mortgage rates increases relative to the Treasuries, 3% points overTreasury benchmark might classify some part of prime loans as subprime loans.Second, in cases of adjustable rate mortgage (ARM), subprime loans might beunder-represented. Using comparable-maturity Treasuries, an ARM with a contractmaturity of 30 years will be compared with long-term Treasury security, which nor-mally has higher than short-term interest rates. But ARMs’ interest is based onshort-term interest. So ARM loans are less likely to be identified as subprime loansthan fixed rate mortgage (FRM) loans. Third, there is no consideration for regionaldifferences in the interest rate for classifying subprime loans. As a result, a loanoriginated in the region with a high average interest rate is more likely defined as asubprime loan than the same loan originated in a region with low interest rates. Fi-nally, the rate spread is reported only if a loan is originated, so we cannot computea denial rate for subprime loans.

In the HMDA data before 2004, researchers defined subprime loans using theHUD list of subprime lenders. HUD made a list of subprime lenders by identifyinglenders that specialize in subprime lending (See Mayer and Pence (2008) for moredetailed explanation). But defining subprime loans by the HUD list also generatesbias. Subprime lenders may also originate prime loans, and non-subprime lenderscan originate subprime loans. Gerardi, Shapiro, and Willen (2007) suggested thatthe latter case is a larger source of bias and subprime loans would be understatedby this method.


Gerardi, Shapiro, and Willen (2007) estimate that 26% of loans by prime lenders

were in the category of subprime loans.

1.7. Conclusion

In this paper, I test the impact of anti-predatory lending laws, examining 1) whether

the laws inhibited mortgage volume during the housing-bubble period and 2) whether

restricted mortgage volume as a result of these laws reduced household consump-

tion. The impact of anti-predatory lending laws is a little-studied but important

question, since the laws are closely connected to the main characteristic features of

the recent housing crisis.

Predatory lending practices in the home mortgage market have grown with

mortgage credit expansion. To cope with increasing evidence of predatory lend-

ing practices, Congress enacted the Home Ownership and Equity Protection Act

(HOEPA) in 1994. Since HOEPA was deemed insufficient to curb abusive practices,

many states enacted laws limiting predatory mortgage lending to fill HOEPA’s gaps.

State anti-predatory lending laws followed HOEPA’s structure but employed vary-

ing levels of strictness based on their own interests. As a result, state laws varied

in the strictness of their provisions.


laws in a fixed-effect model to estimate the impact of these laws on total mortgage

volume. I use the anti-predatory lending (APL) law index from the Corporation

for Enterprise Development (CFED), a nonprofit organization that evaluates state

policies, to evaluate the strictness of the various state laws. I develop a time-varying

CFED Index by interacting the CFED Index with the effective dates of state APL

laws. Consistent with my hypothesis, I find that states with stricter laws had lower

mortgage volume. I also find that the reduction in mortgage volume becomes larger


as the Loan-to-Income ratio increases. Since the Loan-to-Income ratio is one of the

frequently used proxy measure for the riskiness of a loan, this result shows that the

laws had a bigger impact on loans with higher risk characteristics.

I also test whether, by restricting mortgage volume, these laws influenced house-

hold expenditures. I use the result that the APL laws inhibited mortgage volume

to estimate a reduced form model for the APL laws’ effect on household expen-

ditures. Again, I use the time-varying CFED Index for law strictness and use the

Consumer Expenditure Survey by the Bureau of Labor Statistics for household ex-

penditures. Due to the small sample size in the household expenditures, statistical

significance of results is limited, but I find that strict APL laws are associated with

total expenditure reductions.


Table 1.1: Indices for the State Anti-Predatory Lending Laws

Table 1 reports the indices on the state anti-predatory lending laws. Panel A reports the indices by state. Column(1) shows the index from the Corporation for Enterprise Development (2007). Columns (2)-(4) show the indexfrom Ho and Pennington-Cross (2006). The HP Index measures the law by two separate dimensions: Coverageand Restriction. Full Index is the sum of the Coverage Index and the Restriction Index. Columns (5)-(8) show theindex from Bostic et al. (2008). The BEA Index extends the HP Index by introducing the Enforcement Index inaddition to the Coverage Index and the Restriction Index. Full Index is the sum of the three indices. In column(9), effective dates of the state anti-predatory lending laws are reported. Effective dates exist only for states withcomprehensive anti-predatory lending laws. Panel B reports summary statistics of the law indices. Panel Creports the correlation matrix between the indices.

Panel A : Indices by State

CFED HP BEA EffectiveFull Coverage Restriction Full Coverage Restriction Enforcement Date

State (1) (2) (3) (4) (5) (6) (7) (8) (9)

New Mexico 12 12.91 6.28 6.63 9.9 4.17 3.27 2.46 1/1/04North Carolina 12 5.07 1.11 3.96 6.4 1.72 3.27 1.41 7/1/00Massachusetts 10 9.68 4.13 5.55 8.43 2.15 3.82 2.46 11/7/04Rhode Island 10 0 0 0 0 12/31/06West Virginia 10 9 5.6 1.64 1.76 6/5/02New Jersey 9 6.27 3.13 3.14 7.34 2.15 2.73 2.46 11/27/03South Carolina 9 8.83 2.36 6.47 4.8 0.86 2.18 1.76 1/1/04New York 8 6.82 4.13 2.69 5.82 2.15 1.91 1.76 4/1/03Georgia 7 14.89 4.13 10.76 6.83 1.72 3 2.11 3/7/03Illinois 6 17.16 8.73 8.43 8.11 3.74 1.91 2.46 1/1/04Arkansas 5 10.06 2.73 7.33 6.56 1.72 2.73 2.11 7/16/03DC 5 14.89 10.5 4.39 7.76 3.74 1.91 2.11 5/7/02Indiana 5 7.55 2.36 5.19 6.75 1.29 3 2.46 1/1/05Minnesota 4 7.01 6.46 0.55 0 1/1/03Tennessee 4 0 0 0 0 1/1/07Texas 4 3.8 0.74 3.06 4.33 0.86 1.36 2.11 9/1/01Vermont 3 0 0 0 0Alaska 2 0 0 0 0Connecticut 2 6.93 2.73 4.2 4.88 0.86 1.91 2.11 10/1/01Florida 2 1.98 0 1.98 3.75 0 1.64 2.11 10/2/02Iowa 2 0 0 0 0Maine 2 1.47 1.47 0 3.01 0 0.55 2.46Maryland 2 10.51 5.84 4.67 3.4 1.44 0.55 1.41 10/1/02Michigan 2 5.99 5.17 0.82 0 12/23/02Idaho 1 0 0 0 0Kansas 1 0 0 0 0Kentucky 1 4.96 0.74 4.22 5.85 0.86 2.18 2.81 6/24/03Missouri 1 0 0 0 0Virginia 1 0 0 0 0 6/26/03Wisconsin 1 2.63 1.55 1.08 0 0 0 0 2/1/05Alabama 0 0 0 0 0Arizona 0 0 0 0 0California 0 7.07 5.09 1.98 4.92 2.15 1.36 1.41 7/1/02Colorado 0 16.18 12.87 3.31 4.18 0.43 1.64 2.11 6/7/03Delaware 0 0 0 0 0Hawaii 0 0 0 0 0Louisiana 0 0 0 0 0Mississippi 0 0 0 0 0Montana 0 0 0 0 0Nebraska 0 0 0 0 0 3/20/03Nevada 0 1.47 1.47 0 2.81 0 0 2.81 10/1/03New Hampshire 0 0 0 0 0 1/1/04North Dakota 0 0 0 0 0Ohio 0 2.37 1.47 0.9 3.47 0 1.36 2.11 5/24/02Oklahoma 0 4.59 0.74 3.85 4.29 0 2.18 2.11 1/1/04Oregon 0 0 0 0 0Pennsylvania 0 2.91 1.47 1.44 3.47 0 1.36 2.11 6/25/02South Dakota 0 0 0 0 0Utah 0 2.55 1.47 1.08 3.9 1.72 2.18 0 5/3/04Washington 0 0 0 0 0Wyoming 0 0 0 0 0


Table 1 (continued)

Panel B : Summary Statistics

CFED HP BEAFull Coverage Restriction Full Coverage Restriction Enforcement

State (1) (2) (3) (4) (5) (6) (7) (8)

Mean 2.80 7.34 3.49 3.85 3.00 1.00 1.00 1.00S.D. 3.64 4.85 3.23 2.67 3.19 1.62 1.17 1.10

Max 12 17.16 12.87 10.76 9.9 6.46 3.82 2.81Min 0 1.47 0 0 0 0 0 0

Panel C : Correlation Matrix

CFED HP BEAFull Coverage Restriction Full Coverage Restriction Enforcement

State (1) (2) (3) (4) (5) (6) (7) (8)

CFED 1.00 0.40 0.13 0.57 0.67 0.55 0.66 0.44

HPFull 1.00 0.86 0.78 0.64 0.69 0.40 0.21

Coverage 1.00 0.35 0.40 0.57 0.08 0.11Restriction 1.00 0.69 0.56 0.63 0.25

BEAFull 1.00 0.79 0.88 0.79

Coverage 1.00 0.49 0.30Restriction 1.00 0.78

Enforcement 1.00


Table 1.2: The Coverage of the HMDA Data

Table 2 compares the HMDA data with the mortgage origination estimate from the Mortgage Bankers Association(MBA) as a benchmark. The HMDA data includes conventional single-family loans. Benchmark estimates alsocover all home mortgages for single-family loans. Panel A reports the total originations. Panel B reports theamount of originations for home purchase loans. Panel C reports the amount of originations for refinance purposeloans. In each panel, the HMDA data is compared with the mortgage origination estimate from the MBA.

Panel A: Total Originations

HMDA Market (Estimate) CoverageYear (millions of dollars) (percent)

1998 1,292 1,656 78.001999 1,078 1,379 78.152000 882 1,139 77.402001 1,798 2,243 80.172002 2,495 2,854 87.432003 3,327 3,812 87.282004 2,470 2,773 89.082005 2,774 3,027 91.642006 2,505 2,726 91.902007 1,965 2,306 85.23

Average 2,059 2,392 84.63

Panel B: Originations for Home Purchase


1998 518 795 65.141999 589 878 67.112000 616 905 68.052001 671 960 69.942002 817 1,097 74.452003 903 1,280 70.552004 1,105 1,309 84.382005 1,325 1,512 87.622006 1,216 1,399 86.942007 922 1,140 80.84

Average 868 1,128 75.50

Panel C: Originations for Refinance


1998 774 862 89.781999 488 500 97.692000 266 234 113.582001 1,127 1,283 87.822002 1,678 1,757 95.532003 2,424 2,532 95.742004 1,366 1,463 93.362005 1,449 1,514 95.712006 1,289 1,326 97.202007 1,044 1,166 89.51

Average 1,191 1,264 95.59


Table 1.3: Summary Statistics

Table 3 reports the summary statistics of the variables. The years covered are 1998 to 2007. Panel A summarizesthe variables from the HMDA data. All variables are defined by the MSA. The number of applications (APPL),the amount of originations (AMTO), and the denial rate (DR) are reported by loan purpose: all loans, homepurchase loans and refinance loans. Panel B summarizes the variables on the local housing market conditions andlocal demographics. Effective interest rates on mortgages (EFFINT) is a variable by state. Annual home priceappreciation (HPIAPP), housing supply elasticity (Elasticity), unemployment rates (UNEMP), per capita income(INC), and the log number of populations (logPOP) are the variables by MSA. Panel C summarizes consumerexpenditures. Total annual expenditures are on the first row and detailed breakdowns by expenditure purposesfollows.

Panel A: Mortgage Origination

Obs Mean Std.Dev. Min Max Source

All Loans

APPL 3630 46412 106057 1050 1430241 HMDAAMTO ($ bils) 3630 4.93 15.76 0.035 263.1 HMDADR (%) 3630 37.28 8.66 12.76 70.3 HMDA

Refinance Loans


Home Purchase Loans


Panel B: Local Housing Market Conditions and Demographics

Obs Mean Std.Dev. Min Max Source

HPIAPP (%) 3627 0.06 0.05 -0.13 0.34 FHFAEFFINT (%) 510 6.67 0.72 5.42 8.46 FHFAElasticity 280 2.52 1.42 0.63 12.15 Saiz (2010)

UNEMP (%) 3630 0.05 0.02 0.01 0.3 BLSINC ($) 3630 29502 6425 12723 79576 BEAlogPOP 3630 12.61 1.06 10.75 16.75 BEA


Table 3 (continued)

Panel C: Consumer Expenditure

Obs Mean Std. Dev. Min Max Source

Annual Expenditure 246 47,176 7,416 32,928 70,611 BLS

Food 246 5,970 843 3,888 8,393 BLSFood at home 246 3,369 488 2,298 4,629 BLS

Cereals and bakery products 246 474 66 313 681 BLSMeats, poultry, fish, and eggs 246 848 131 538 1,182 BLS

Dairy products 246 362 59 234 522 BLSFruits and vegetables 246 602 110 384 916 BLS

Other food at home 246 1,083 208 652 1,641 BLSFood away from home 246 2,601 458 1,589 4,070 BLS

Housing 246 16,308 3,160 9,993 27,310 BLSShelter 246 10,058 2,511 4,946 19,519 BLS

Owned dwellings 246 6,637 1,647 3,080 12,526 BLSRented dwellings 246 2,812 1,004 1,412 6,432 BLS

Other lodging 246 609 221 222 1,454 BLSUtilities, fuels, and public services 246 3,004 530 1,977 4,551 BLSHousehold operations 246 910 315 383 2,871 BLSHousekeeping supplies 246 566 101 328 853 BLSHousehold furnishings and equipment 246 1,770 394 936 3,090 BLS

Transportation 246 8,326 1,366 5,108 12,596 BLSVehicle purchases (net outlay) 246 3,527 909 999 6,383 BLSGasoline and motor oil 246 1,574 524 862 3,274 BLSOther vehicle expenses 246 2,654 400 1,748 3,759 BLSPublic transportation 246 572 237 196 1,479 BLS

Alcoholic beverages 246 479 139 222 928 BLSApparel and services 246 1,981 396 954 3,137 BLSHealthcare 246 2,370 439 1,481 3,705 BLSEntertainment 246 2,355 558 1,401 4,297 BLSPersonal care products and services 246 595 113 270 932 BLSReading 246 155 49 31 312 BLSEducation 246 887 342 285 1,895 BLSTobacco products and smoking supplies 246 291 83 95 538 BLSMiscellaneous 246 885 244 240 2,090 BLSCash contributions 246 1,522 541 585 3,687 BLS

Personal insurance and pensions 246 5,053 1,260 3,026 9,408 BLSLife and other personal insurance 246 413 118 130 1,078 BLSPensions and Social Security 246 4,640 1,260 2,632 8,905 BLS

CHAPTER

1.THEIM

PACT

OFANTI-P

REDATORY

LENDIN

GLAW

S39

Table 1.4: The Impact of the Anti-Predatory Lending Law on Home Mortgage Origination

Table 4 reports the fixed-effect panel regression results with all loans. For the APL law variation, a time-varying law index is used by interacting the CFED Index with theeffective date of the state law (APLi · I{t>=ED}). The effective date (ED) of the state law is reported in column (9) of Table 1. The years covered in the regression are 1998to 2007. The log number of applications (logAPPL), the denial rate (DR), and the log amount of originations (logAMTO) are left-hand side variables. Columns (1)-(3)report the results with MSA FE and year FE. Columns (4)-(6) report the results with control variables, in addition to state FE and year FE. Control variables includeannual home price appreciation (HPIAPP), state-level mortgage rate (EFFINT), supply elasticity (Elasticity), income per capita (INC), unemployment rate (UNEMP), andthe log numbers of population (logPOP). The table reports point estimates with robust t-statistics in parentheses. ***, **, * denotes 1%, 5%, and 10% statisticalsignificance. All standard errors are clustered by MSA. Economic significance is also reported under the APL variable. For example, a one-standard deviation increase in theAPL law index is associated with a 0.16-standard deviation reduction in the DR. (A one-standard deviation increase in APL is associated with a 2.98 [1 SD] x 0.00475[slope] = 0.014 decrease in DR, which is 0.014 / 0.09 = 0.16-standard deviation of DR.)

DR logAMTO logAPPL DR logAMTO logAPPLVariables (1) (2) (3) (4) (5) (6)

APL -0.00475*** -0.0121*** -0.0174*** -0.00457*** -0.00399 -0.0109***(-5.629) (-2.835) (-5.302) (-6.014) (-1.158) (-3.783)

Economic Significance -0.16 -0.02 -0.04 -0.17 -0.01 -0.03

HPIAPP -0.344*** 2.749*** 1.624***(-15.46) (17.25) (15.07)

EFFINT -0.00363 0.338*** 0.235***(-0.409) (5.414) (5.007)

Elasticity 0.0101*** -0.0690*** -0.0232***(3.600) (-4.844) (-2.669)

INC -3.89e-06*** 4.67e-05*** 8.24e-06***(-6.697) (12.51) (3.770)

logPOP 0.0129*** 1.010*** 1.016***(5.445) (54.99) (78.63)

UNEMP 0.830*** -5.407*** -2.152***(4.550) (-6.586) (-4.358)

Constant 0.428*** 11.78*** 8.401*** 0.282*** -2.418*** -4.950***(135.2) (823.0) (744.6) (3.847) (-4.892) (-13.07)

Observations 3,630 3,630 3,630 2,800 2,800 2,800R-squared 0.822 0.977 0.981 0.738 0.968 0.977Year FE Yes Yes Yes Yes Yes YesRegional FE MSA MSA MSA State State State

CHAPTER

1.THEIM

PACT

OFANTI-P

REDATORY

LENDIN

GLAW

S40

Table 1.5: By Loan-to-Income ratio

Table 5 reports the fixed-effect panel regression results by the Loan-to-Income (LTI) ratio. Dependent variables are the denial rate (DR), the log amount of originations(logAMTO), and the log number of applications (logAPPL). Variables are divided into two groups by LTI ratio: high LTI group and low LTI group. Panel A reports theregression results with MSA fixed-effect and panel B reports the results with state fixed-effect and MSA-level control variables. In panel A, columns (1)-(2) report the resultson DR, columns (3)-(4) report the results on logAMTO, and columns (5)-(6) report the results on logAPPL. In panel B, columns (1)-(3) report the results on DR, columns(4)-(6) report the results on logAMTO, and columns (7)-(9) report the results on logAPPL. For testing the differences of law effects between two LTI groups, pooledregression results are reported in panel B. DIFF is defined as the difference of law effect in the high LTI group relative to the low LTI group. For the APL law variation, atime-varying law index is used by interacting the CFED Index with the effective date of the state law (APLi · I{t>=ED}). The effective date (ED) of the state law isreported in column (9) of Table 1. Year fixed-effect is included. Control variables include annual home price appreciation (HPIAPP), state-level mortgage rate (EFFINT),supply elasticity (Elasticity), income per capita (INC), unemployment rate (UNEMP), and the log numbers of population (logPOP). The years covered in the regression are1998 to 2007. The table reports point estimates with robust t-statistics in parentheses. ***, **, * denotes 1%, 5%, and 10% statistical significance. All standard errors areclustered by MSA.

Panel A : With MSA Fixed-Effect

DR logAMTO logAPPL

Variables High LTI Low LTI High LTI Low LTI High LTI Low LTI(1) (2) (3) (4) (5) (6)

APL -0.00170** -0.00113* -0.0159*** -0.00406 -0.0157*** -0.000700(-2.230) (-1.855) (-2.735) (-1.088) (-3.741) (-0.272)

Economic Significance -0.04 -0.04 -0.03 -0.01 -0.04 0.00

Constant 0.400*** 0.439*** 10.18*** 10.44*** 6.192*** 7.475***(135.4) (189.3) (579.0) (772.2) (441.4) (747.5)

Observations 3,630 3,630 3,630 3,630 3,630 3,630R-squared 0.862 0.852 0.965 0.979 0.971 0.985Year FE Yes Yes Yes Yes Yes YesMSA FE Yes Yes Yes Yes Yes Yes

CHAPTER

1.THEIM

PACT

OFANTI-P

REDATORY

LENDIN

GLAW

S41

Table 5 (continued)

Panel B : With State Fixed-Effect & MSA-level Control Variables

DR logAMTO logAPPL

Variables High LTI Low LTI Pooled High LTI Low LTI Pooled High LTI Low LTI Pooled(1) (2) (3) (4) (5) (6) (7) (8) (9)

APL -0.00522*** -0.00293*** -0.00293*** -0.00790* -0.000434 -0.000434 -0.0169*** -0.00458* -0.00458*(-6.049) (-4.240) (-4.240) (-1.904) (-0.167) (-0.167) (-4.381) (-1.827) (-1.827)

Economic Significance -0.17 -0.11 -0.02 0.00 -0.04 -0.01

DIFF -0.00229*** -0.00746** -0.0123***(-4.079) (-2.537) (-4.057)

HPIAPP -0.389*** -0.254*** 3.030*** 2.110*** 1.969*** 1.265***(-15.84) (-11.03) (17.10) (13.59) (15.08) (10.69)

Elasticity 0.0110*** 0.00958*** -0.0809*** -0.0485*** -0.0416*** -0.00205(3.729) (3.161) (-4.870) (-3.949) (-3.893) (-0.198)

EFFINT -0.0119 0.00472 0.365*** 0.238*** 0.270*** 0.112***(-1.190) (0.526) (4.674) (6.792) (3.997) (3.416)

UNEMP 0.963*** 0.626*** -5.863*** -3.602*** -2.581*** -1.391**(4.277) (4.297) (-6.009) (-5.139) (-4.737) (-2.204)

logPOP 0.00707*** 0.0193*** 1.036*** 0.952*** 1.032*** 1.004***(2.731) (7.715) (50.22) (53.00) (69.54) (67.59)

INC -4.21e-06*** -2.98e-06*** 5.04e-05*** 3.03e-05*** 1.45e-05*** -3.76e-06(-6.369) (-5.331) (12.26) (7.646) (5.947) (-1.416)

Constant 0.397*** 0.160** -3.332*** -2.046*** -6.117*** -4.554***(4.693) (2.221) (-5.448) (-6.754) (-11.85) (-14.27)

Observations 2,800 2,800 5,600 2,800 2,800 5,600 2,800 2,800 5,600R-squared 0.717 0.720 0.719 0.961 0.970 0.968 0.967 0.973 0.970Year FE Yes Yes Yes Yes Yes Yes Yes Yes YesState FE Yes Yes Yes Yes Yes Yes Yes Yes Yes


Table 1.6: The Effect of the Anti-Predatory Lending Law on Household Expenditure

Table 6 reports the fixed-effect panel regression results on household expenditures. Household expenditures bypurpose are left-hand side variables. Panel A reports the regression results on total expenditures and panel Breports the results on the expenditures by purpose. Panel C reports the results by subdividing the expenditurepurposes in panel B. The household expenditure data is from the Consumer Expenditure Survey (CES). For theAPL law variation, a time-varying law index is used by interacting the CFED Index with the effective date of thestate law (APLi · I{t>=ED}). The effective date (ED) of the state law is reported in column (9) of Table 1. StateFE and year FE are used. Control variables include annual home price appreciation (HPIAPP), state-levelmortgage rate (EFFINT), supply elasticity (Elasticity), income per capita (INC), the unemployment rate(UNEMP), and the log numbers of population (logPOP). For brevity, demographic control variables are notreported. The years covered in the regression are 1998 to 2007. The number of observations is 232. The tablereports point estimates with robust t-statistics in parentheses. ***, **, * denotes 1%, 5%, and 10% statisticalsignificance. All standard errors are clustered by MSA. Economic significance is also reported. A one-standarddeviation increase in the APL law index is associated with a 0.05-standard deviation decline in totalexpenditures. (A one-standard deviation increase in the APL law index is associated with a 2.12 [1 SD] x 161.9[slope x -1] = 343.91 decrease in total expenditures, which is 343.91 / 7475.56 = 0.05-standard deviation of DR.)

APL Econ Sig HPIAPP EFFINT ElasticityPanel A : Total Expenditure

Average annual expenditures -161.9 -0.05 4,122 -2,292 1,237*(-0.999) (0.535) (-1.100) (2.021)

Panel B : Expenditure by Purpose

Food -3.919 -0.01 1,099 9.448 258.6(-0.112) (0.671) (0.0195) (1.424)

Alcoholic beverages -5.941 -0.09 37.29 -3.777 22.46(-0.841) (0.171) (-0.0483) (0.629)

Housing -53.00 -0.03 -111.0 -741.7 659.0***(-1.255) (-0.0621) (-1.135) (3.301)

Apparel and services 5.434 0.03 724.8 126.6 -13.03(0.346) (1.149) (0.507) (-0.133)

Transportation -48.73 -0.08 4,236 -938.0 252.3**(-0.853) (1.386) (-1.217) (2.484)

Healthcare -13.84 -0.07 -106.2 -128.6 26.25(-0.713) (-0.140) (-0.853) (0.505)

Entertainment 9.877 0.04 -225.8 -88.68 19.31(0.472) (-0.532) (-0.417) (0.314)

Personal care products and services -2.997 -0.06 189.5 85.18* 0.386(-1.058) (0.978) (1.813) (0.0124)

Reading -1.034 -0.05 40.85 12.42 7.318**(-0.628) (0.566) (0.583) (2.077)

Education 6.747 0.04 -111.8 -197.6 -0.290(0.619) (-0.193) (-1.502) (-0.00650)

Tobacco products and smoking supplies -1.840 -0.05 92.91 -27.44 1.295(-0.582) (0.787) (-0.837) (0.118)

Miscellaneous -13.44 -0.12 -159.5 -48.14 -56.08(-0.991) (-0.464) (-0.461) (-1.364)

Cash contributions -55.32** -0.21 -850.9 60.80 -34.09(-2.557) (-0.750) (0.219) (-0.133)

Personal insurance and pensions 16.12 0.03 -730.8 -412.5 93.49(0.588) (-0.751) (-0.872) (0.778)


Table 6 (continued)

Panel C : Expenditure by Detailed Purpose

APL Econ Sig HPIAPP EFFINT ElasticityFood

Food at home 0.865 0.00 639.4 -72.18 -7.132(0.0403) (0.709) (-0.245) (-0.0781)

Cereals and bakery products 3.868 0.13 132.4 -17.95 -1.074(1.164) (1.009) (-0.353) (-0.0634)

Meats, poultry, fish, and eggs -6.254 -0.10 111.9 -87.52 -70.71***(-0.999) (0.483) (-1.081) (-3.715)

Dairy products 1.196 0.04 156.4 8.732 14.59(0.507) (1.541) (0.400) (1.172)

Fruits and vegetables -0.0112 0.00 50.63 43.45 -21.24(-0.00293) (0.295) (0.843) (-1.134)

Other food at home 2.129 0.02 188.5 -18.45 71.37*(0.286) (0.609) (-0.172) (1.935)

Food away from home -4.838 -0.02 460.4 81.50 265.5**(-0.224) (0.532) (0.343) (2.208)

Housing

Shelter -47.30 -0.04 -1,079 165.3 489.2***(-1.440) (-0.778) (0.346) (3.178)

Owned dwellings -11.38 -0.01 -810.3 -416.1 119.4(-0.382) (-0.582) (-0.976) (0.551)

Rented dwellings -46.99* -0.10 -35.15 510.8** 422.3***(-1.838) (-0.0357) (2.636) (3.012)

Other lodging 11.09 0.11 -234.3 70.55 -52.65(1.099) (-0.515) (0.517) (-1.565)

Utilities, fuels, and public services 12.23 0.05 60.42 -253.3*** -55.26(0.767) (0.138) (-2.992) (-0.446)

Household operations -17.17* -0.11 -735.2 -294.5 82.68(-1.792) (-1.250) (-1.436) (1.003)

Housekeeping supplies 0.118 0.00 327.8 -8.530 41.57*(0.0298) (1.297) (-0.110) (1.942)

Household furnishings and equipment -0.869 0.00 1,316** -349.6* 101.0**(-0.0595) (2.777) (-1.729) (2.692)

Transportation

Vehicle purchases (net outlay) -41.92 -0.10 3,094 -1,096* 132.7(-0.982) (1.415) (-1.733) (0.911)

Gasoline and motor oil -14.23** -0.06 493.0* 34.10 104.0**(-2.102) (1.864) (0.410) (2.391)

Other vehicle expenses 12.90 0.07 571.1 69.47 67.46(0.568) (0.792) (0.358) (1.062)

Public transportation -5.489 -0.05 77.03 53.88 -52.16**(-0.939) (0.256) (0.888) (-2.301)

Personal insurance and pensions

Life and other personal insurance -1.731 -0.03 32.71 -23.29 -71.47(-0.398) (0.154) (-0.355) (-1.696)

Pensions and Social Security 17.82 0.03 -762.7 -389.2 165.0(0.697) (-0.838) (-0.898) (1.320)


Figure 1.1: The MSA-level CFED index

Figure 1 reports the MSA-level CFED index on a map of the United States. For applying the state-level lawindices to the MSA-level analysis, the state-level law indices are mapped into a MSA-level variable. First, forthose MSAs within a state, the state law index has been assigned. Second, for those MSAs across several differentstates, a population-weighted state index has been assigned.

The CFED Index

0.0 - 0.6

0.7 - 3.1

3.2 - 6.0

6.1 - 9.0

9.1 - 12.0


Figure 1.2: The Mortgage Rate and Home Mortgage Origination

Figure 2 reports the amount of originations (AMTO) and the denial rate (DR) with the 30-year mortgage rate.Using the HMDA data, the U.S. aggregate AMTO and the U.S. average DR are reported by loan purpose. Theconventional, conforming 30-year fixed-rate from the Primary Mortgage Market Survey by Freddie Mac is used. Inpanel A, the AMTO in mortgage refinancing and in home purchases are compared with the 30-year mortgage rate.In panel B, the DR in mortgage refinancing and in home purchases are compared with the 30-year mortgage rate.

66

.57

7.5

8M

TG

ra

te (

%)

05

00

10

00

15

00

20

00

AM

TO

($

bils

)

1998 2000 2002 2004 2006

AMTO (RF) 30y MTG rate

Refinance

66

.57

7.5

8M

TG

ra

te (

%)

05

00

10

00

15

00

20

00

AM

TO

($

bils

)

1998 2000 2002 2004 2006

AMTO (HP) 30y MTG rate

Home Purchase

Panel A: The Amount of Originations6

6.5

77

.58

MT

G r

ate

(%

)

20

30

40

50

DR

(%

)

1998 2000 2002 2004 2006

DR (RF) 30y MTG rate

Refinance

66

.57

7.5

8M

TG

ra

te (

%)

20

30

40

50

DR

(%

)

1998 2000 2002 2004 2006

DR (HP) 30y MTG rate

Home Purchase

Panel B: The Denial Rate


Figure 1.3: The Home Price Index and Home Mortgage Origination

Figure 3 reports the amount of originations (AMTO) and the denial rate (DR) with the Home Price Index. Usingthe HMDA data, the U.S. aggregate AMTO and the U.S. average DR are reported by loan purpose. The U.S.national HPI from FHFA is used. In panel A, the AMTO in mortgage refinancing and in home purchases arecompared with the HPI. In panel B, the DR in mortgage refinancing and in home purchases are compared withthe HPI.

20

02

50

30

03

50

40

0H

PI

05

00

10

00

15

00

20

00

AM

TO

($

bils

)

1998 2000 2002 2004 2006

AMTO (RF) HPI

Refinance

20

02

50

30

03

50

40

0H

PI

05

00

10

00

15

00

20

00

AM

TO

($

bils

)

1998 2000 2002 2004 2006

AMTO (HP) HPI

Home Purchase

Panel A: The Amount of Originations2

00

25

03

00

35

04

00

HP

I

20

30

40

50

DR

(%

)

1998 2000 2002 2004 2006

DR (RF) HPI

Refinance

20

02

50

30

03

50

40

0H

PI

20

30

40

50

DR

(%

)

1998 2000 2002 2004 2006

DR (HP) HPI

Home Purchase



Figure 1.4: The Home Price Index by Elasticity of Housing Supply

Figure 4 reports time-series of the HPI by elasticity of the housing supply. By the elasticity of housing supplyfrom Saiz (2010), MSAs are divided into the elastic MSAs and the inelastic MSAs. The average HPIs of thegroups are plotted. The base year of the average HPI is 1980.

01

23

4

HP

I

1975 1980 1985 1990 1995 2000 2005 2010

Inelastic Elastic


Figure 1.5: By Elasticity of Housing Supply

Figure 5 reports a time series of the amount of originations (AMTO) and the denial rate (DR) by elasticity of thehousing supply. By the elasticity of housing supply from Saiz (2010), MSAs are divided into the elastic MSAs andthe inelastic MSAs. Panel A reports a time-series of the AMTO by the supply elasticity and the loan purpose.Panel B reports a time-series of the DR by the supply elasticity and the loan purpose.

05

00

10

00

15

00

20

00

AM

TO

($

bils

)

1998 2000 2002 2004 2006

Inelastic Elastic

Refinance

05

00

10

00

15

00

20

00

AM

TO

($

bils

)

1998 2000 2002 2004 2006

Inelastic Elastic

Home Purchase

Panel A: The Amount of Originations

20

30

40

50

DR

(%

)

1998 2000 2002 2004 2006

Inelastic Elastic

Refinance

20

30

40

50

DR

(%

)

1998 2000 2002 2004 2006

Inelastic Elastic

Home Purchase


CHAPTER 2. SPECULATING ON HOME IMPROVEMENTS 49

Chapter 2

Speculating on Home

Improvements

2.1. Introduction

We seek to develop a theory of home improvements—a little-studied but impor-

tant economic activity. While the significance of new home constructions for eco-

nomic growth during the housing bubble years of 2003-2007 is well-documented,

the contributions of home remodeling expenditures, though less heralded, are no

less impressive. The Joint Center for Housing Studies of Harvard University reports

that home improvement expenditures on, for instance, a new bathroom or a new

deck, jumped from around 1% of GDP ($229 billion) in 2003 to 2% of GDP ($326

billion) in 2007.1 Spending on remodeling projects then dropped precipitously after

2007 with falling home prices, thereby contributing to the Great Recession of 2008.

These figures suggest that home remodeling is an important industry for the U.S.

1These figures include professional remodeling projects and do-it-yourself (DIY)jobs. The purchase of raw materials from companies like Home Depot for DIY jobsare accounted for in these GDP figures but not the opportunity costs of DIY labor.


economy and the pro-cyclicality of these expenditures contributes to business cycle

fluctuations.

A consumption-based model is a natural way to rationalize remodeling activity.

For example, rising wealth in the form of home equity leads to demand for better

housing and hence for home improvements. This theory requires an additional and

eminently reasonable assumption of moving costs.2 Otherwise, homeowners would

simply move to another nicer home as opposed to remodeling an existing one.

Another sufficient assumption besides moving costs is that homeowners like to do

remodeling as a form of pleasure. There is little doubt that this consumption motive

is there for many households who undertake improvements.

However, there are a number of other stylized patterns regarding remodeling

which suggest that this consumption-based theory may not be the only important

economic force behind home improvements and their correlation with home prices.

First, home improvements are more likely to be undertaken by sellers or households

planning to move (see Joint Center for Housing Studies of Harvard University). Sec-

ond, remodeling activity picks up with rising home prices perhaps in part because it

is easier to sell homes during these times. In other words, remodeling activity picks

up when moving costs are low as opposed to being high in the time series. Third,

there is substantial anecdotal evidence that homeowners think of improvements as

an investment in the same way they think about the purchase of a home. For in-

stance, “fix it and flip it” is a phrase often associated with real estate investing in

which it is thought that the completion of a few choice remodeling projects will

add significant value to the price of a home.3 Fourth, the prevailing professional

2Moving costs could include, beyond the obvious hassle of packing and unpack-ing, preferred schools for kids or property taxes which might bias households towardstaying in the same home. For instance, such a tax bias is prominent in the stateof California.

3Google search ”fix it and flip it” and many housing sites discuss this phe-nomenon.


view, which is also backed by data, is that home remodeling typically are not prof-

itable, as the value-added of the improvements is often less than the construction

cost. Thus homeowners undertake major renovation projects with the mistaken be-

lief that improving the place will result in big profits but they often end up not

realizing these gains.

This speculative motive appears particularly widespread during the recent hous-

ing bubble years as rapid price appreciation and the easing of financial constraints

allowed homeowners to not only consume but to potentially speculate in housing

through such remodeling activities.4 Indeed, McQueen (Apr 24, 2010) reports that

the forces driving home improvements during the previous housing-boom decade

could not be more different than the ones driving home improvements after the

collapse of home prices: “Back then, people wanted to renovate their places so that

they could trade up to bigger homes, or because their home equity was soaring

and they wanted to reinvest some of the spoils. Now, the opposite is happening:

Many people who bought during the boom years are accepting the reality that they

won’t soon be swapping up for a sybaritic spread. Their mortgages may remain

above water, but after years of falling home prices, their equity is so low that the

transaction costs of buying a new house would leave little for a down payment.”

As such, we pursue in this paper a speculation-based theory of home improve-

ments. Our model has the following features. A unit of housing services is given by

a Cobb-Douglas production function of land and structures with constant returns

to scale. There are upward sloping supply curves for the two factors. There is one-

hoss-shay depreciation of homes and hence demand for new homes each period. A

competitive home-building industry determines the optimal mix of land and struc-

4The recent boom in consumption or home improvements had little to do withincome growth and more to do with the easing of financial constraints (from low in-terest rates and easy lending standards). See Melzer (2010) for more micro-evidenceon the link between home prices and remodeling.


tures. Homeowners have an option to build additional structures. Our simple model

assumes that they can only remodel existing homes and that there are no moving

frictions or leverage constraints.5

The key assumption we make is that overconfident homeowners mistakenly be-

lieve they can build quality additions for a lower construction cost than a competi-

tive building industry. Their overconfidence is modeled as them believing that they

can build new structures at old construction prices. This overconfidence assump-

tion is consistent with evidence which suggests that homeowners under-estimate

the cost of improvements and hence over-estimate the returns to this activity. This

speculative motive is akin to sub-optimal speculation by overconfident retail in-

vestors in the context of stock markets and the poor performance of overconfident

small business owners.6 We take as exogenous home prices, which impute rents and

anticipated capital gains, and consider how an unexpected change in the steady

state price of homes affects equilibrium investments in land and structures while

allowing the factor prices to adjust.

We derive three key results. The first result is that a larger growth in home prices

induces more home improvement activity. An increase in home prices naturally

increases the optimal amount of structures in a given plot of land. This effect is

partially moderated by an increase in the cost of structures. However overconfident

owners believe they can produce more valuable homes at old structure prices, and

thus ignore the increase in the cost of structures. This generates the exaggerated

5Our effects are stronger if we allow for additional home purchase and suchfrictions or constraints. But we abstract away from moving frictions or financialconstraints since the evidence during the recent boom (discussed above) indicatesthat moving costs are likely low and even very rich households without constraintsdid a lot of improvements.

6There is now plentiful evidence on retail trading behavior which indicates thatthe typical retail investor trades on noise and loses money to trading cost as aresult of this noise trading (see, e.g., Barber and Odean (2001) and Odean (1999)).The poor performance of small business owners is documented by Moskowitz andVissing-Jorgensen (2002), who suggest that either overconfidence or tax evasion areplausible explanations.


pro-cyclical pattern in remodeling expenditures with home prices. Moreover, the

growth of structures that results from home improvement is a homogeneous function

of degree larger than one of home price increases. We perform a simple calibration

that shows that the kind of mistakes we attribute to homeowners can explain in

part the boom in improvements during the recent housing bubble in contrast to

the relatively moderate increase over the previous decade, when prices grew just as

much in total, but over a longer period. However, this calibration has to be taken

with a grain of salt since credit also became much easier during the housing bubble

years.

The second result is that there is excessive investment in improvements by

homeowners. When home prices rise, because owners are overconfident about their

ability to build structures at old prices, they build more structures. But in equi-

librium, their demand for structures along with that of the competitive building

industry drives up the cost of structures and their investment in home improve-

ments turns out to be excessive. This result is not testable since we cannot measure

the counter-factual of the right level of home improvements taking into account the

right structures cost. But it is valuable as a welfare result and providing a gauge

of when such welfare losses are likely to be high—welfare losses are high when con-

struction cost have grown substantially relative to home prices. During the housing

bubble years, as we show below, construction cost for home improvements went up

a lot and hence our analysis here suggests that the welfare losses can be potentially

substantial to the extent our speculative motive is present in remodeling decisions.

The third result concerns the recoup ratio (the ratio of resale value of improve-

ments to construction costs). Since homeowners have decreasing returns to scale

technology in home building and there is no entry, they can earn profits when

home price appreciation is high, controlling for construction cost growth. But their

overconfidence leads to losses when construction cost growth is high controlling for


home price appreciation. Hence, the recoup ratio increases in home price growth

controlling for construction cost growth and decreases with construction cost growth

controlling for home price appreciation.

This third prediction is particularly interesting because it is testable and cuts

against the improvements-as-consumption theory. In the consumption theory, de-

mand for amenities would drive up construction costs but presumably the greater

demand for amenities would also drive up the resale value or value added of these

improvements. In other words, the value added from improvements implicit in the

home price increase ought to keep up with construction costs.

We conduct a test of this third prediction using data on construction costs and

resale value of home improvement projects across the U.S. The data comes from

Realtor Magazine which collects surveys of construction cost and resale value for a

basket of home remodeling projects. In the cross-section, we find that cities with

more construction cost growth have actually witnessed a more dramatic fall in re-

coup ratio during the period of the housing bubble period of 2003 to 2009. This

relationship is statistically and economically significant. A one standard deviation

increase in the percentage change in construction cost results in a fall of the per-

centage change of recoup value that is roughly 46% of a standard deviation of the

percentage change in recoup value over this period. This relationship need not be

hardwired since resale values could grow by more. We conduct a series of robustness

checks to verify that this relationship is robust.

Our paper is most closely related to Montgomery (1992) who develops a partial

equilibrium model of the consumption-motive and the trade-off between improve-

ments and moving in the presence of moving costs for households. This paper then

estimates a demand function for home improvement expenditures of households.

Our focus is on the speculative motive and the equilibrium implications and our

empirical work focuses on the returns to improvement projects.


Our theory and empirical analysis contributes to the burgeoning literature of

household finance (see Campbell (2006) and Barber and Odean (2011) for surveys).

Much of this literature has focused on the financial decisions of households such as

stock market participation or financial products such as types of mortgages that are

used and whether the appropriate decisions are made. There is a consensus that at

least poor and uneducated households do not make appropriate decisions. But this

literature has entirely ignored home improvement decisions and whether the right

level of remodeling is undertaken. These economic decisions as we pointed out above

account for an enormous industry that extends to even wealthy households, with

consequences for the macro-economy. Our paper suggests that there are excessive

expenditures on home remodeling.

Our paper also contributes to the real estate economics literature and in par-

ticular the study of housing bubbles and the response of supply to housing price

appreciation. A recent interesting paper by Glaeser, Gyourko, and Saiz (2008) ar-

gues that home prices went up more in low supply elasticity states (such as New

York and California where land is limited and zoning and development rules are

stricter) in which supply could not quickly adjust to the rising home prices. In high

supply elasticity states, home price appreciation was less but the distortions to sup-

ply were potentially more as lots of homes got built. Rather than analyzing new

constructions, we consider home improvements, which is much less explored in the

literature, and how it is affected by the housing bubble. Our data on the returns

and costs of improvement projects, however, is unique in providing a more detailed

look at the excesses associated with remodeling engendered by the housing boom.

Our paper also contributes to a burgeoning literature on the effect of asset

prices or bubbles on real investments. This literature examines whether capital

investments by firms are influenced are stock price bubbles, with the dot-com bubble

being the primary object of focus (see, e.g., Polk and Sapienza (2008), Gilchrist,


Himmelberg, and Huberman (2005), and Farhi and Panageas (2004)). However,

getting a measure of excessive investment is difficult in this context. Excessive

investment is measurable in our context and we show that asset price bubbles can

affect real investments by homeowners.

We present our model in Section 2 and derive the key comparative statics results

in Section 3. We discuss the empirical work regarding home improvement projects

in Section 4. We conclude in Section 5.

2.2. Model

2.2.1. Set-up

Our model has three dates t = 0, 1, 2. We assume that the interest rate is zero. At

t = 0, competitive (price-taking) home builders make new homes from two inputs:

land L0 and structure (i.e. construction) S0. The units of housing H0 is given by a

Cobb-Douglas production function with constant returns to scale:

H0 = Lα0S1−α0 , (2.1)

where α is the share of land in housing. Let p0 be the expected value of the home

at t = 1, which includes both rents and capital gains associated with one unit of

housing (i.e. we are assuming infinitely elastic demand at p0). Let wL,0 be the price

of land and wS,0 be the price of structure (including the cost of labor and materials)

at t = 0. Builders maximize profits per house by choosing the optimal mix of land

and structure, taking as given the price of homes and the costs of the two inputs:

MaxS0,L0

(p0L

α0S

1−α0 − wL,0L0 − wS,0S0

). (2.2)


We assume that a δ fraction of the houses vanish each period (i.e. one-hoss-

shay depreciation). Given the (unique and optimal) choice of inputs S0 and L0 for

each house, the total demand for land and structures are given by δL0 and δS0,

respectively. We assume upward-sloping supply curves for land and structures of

the form:

wL,0 = (δL0)γ, (2.3)

wS,0 = (δS0)β, (2.4)

where γ and β capture the sensitivity of the prices of land and structures to demand

for them, respectively. We then assume that there is free-entry, which enforces a

zero-profit condition for this competitive home-building industry.

Homeowners live one-period and re-sell their homes the next period to new

homeowners. We assume that there is no population growth. At t = 1, existing

homeowners have the option to do home improvements before selling to future

homeowners at t = 2. Now let p1 be the present value of rents and capital gains

associated with house (i.e. the expected price at t = 2). And let wL,1 and wS,1 be

the prices of land and structure at t = 1, respectively.

Existing homeowners then maximize

MaxX1

(p1L

α0 (X1 + S0)1−α − wS,0X1

), (2.5)

subject to the constraint that X1 ≥ 0. X1 is the addition to structure above the

initial amount of structure S0. All homeowners have the same L0 because we imagine

p0 having been the price for a while and we are modeling an unexpected shock to

price in which p0 jumps to p1. Notice that homeowners take wS,0 as the price of


their structure in their optimization problem. The interpretation for this modeling

choice is that homeowners are overconfident that they can build quality additions

or structures at the old prices or for a price lower than the competitive building

industry.

There are no moving costs. As such, we are not ruling out homeowners buying

new homes. Rather, we show that they would optimally want to improve their old

one instead. We are assuming that the natural technology for the typical homeowner

is that they cannot expand their land or the size of their lots. They can make

improvements by building extra structures on their lots. If homeowners could adjust

also the amount of land, they would add even more structures (Le Chatelier’s

Principle). In other words, our effects would be bigger.

Professional home builders are also active at time 1 to build new homes. Let

wL,1 be the price of land and wS,1 be the price of structure. The builders again

maximize profits taking as given the price of homes and the costs of inputs.

MaxS1,L1

(p1L

α1S

1−α1 − wL,1L1 − wS,1S1

). (2.6)

Again, we assume upward sloping supply curves for land and structure of the form:

wL,1 = (δL1)γ, (2.7)

wS,1 = (δS1 + (1− δ)X1)β. (2.8)


Notice that there are only 1−δ old houses left, and so the total demand for structure

at t = 1 is δS1 + (1 − δ)X1.7 We then impose the same competitive zero profit

condition as in time 0.

2.2.2. Equilibrium at t = 0

We now solve for the equilibrium at t = 0. The first-order conditions from the home

builders’ maximization problem with a Cobb-Douglas production function yield the

optimal proportions of land and structure as a function of the relative factor prices

wS,0wL,0

and factor shares α1−α :

L0 =wS,0wL,0

α

1− αS0. (2.9)

The relationship is the usual one: the ratio of land to structure is higher the higher

is the relative cost of structure to land and the higher is the share of land in value

relative to structure.

Imposing zero-profit for homebuilders yields:

p0

wαL,0(

α

1− α)α(1− α) = w1−α

S,0 . (2.10)

And adding the two supply functions then gives four equations in four unknowns

for L0, S0, wS,0, wL,0.

To solve for these four quantities, we first plug the two supply functions into

the first-order condition of the homebuilders to give

wL,0 = w1+ββ

γ1+γ

S,0 (α

1− α)

γ1+γ . (2.11)

7Since δL0 land would become vacant, we could alternatively have written wL,1 =max{0, (δ(L1 − L0)γ)}. However this only leads to minor algebraic changes.


Intuitively, the bigger is β, the more inelastic the supply curve of construction, the

higher the wage of construction relative to wage of land.

We then plug wL,0 into the zero-profit condition to give

wS,0 = [p0(α

1− α)

α1+γ (1− α)]

1

(1−α)+α( 1+ββ

γ1+γ ) . (2.12)

We will get wL,0 as a result. Then we can use the supply function for wS,0 to obtain

the equilibrium amount of structure as

S0 = δ−1[p0(α

1− α)

α1+γ (1− α)]

1

(1−α)+α( 1+ββ

γ1+γ )

1β

, (2.13)

which will then allow us to obtain L0. Note L0 will also be proportional to δ−1. We

will focus on the equilibrium S0.

2.2.3. Equilibrium at t = 1

To solve for the equilibrium at t = 1, we first turn to the home improvement expen-

ditures of existing homeowners at t = 1. Since homeowners believe they can add

structure at the previous price wS,0, the first-order condition of the home improve-

ment optimization problem is given by

p1Lα0 (1− α)(X1 + S0)−α − wS,0 = 0, (2.14)

or

(X1 + S0)−α =wS,0p1

1

1− α1

Lα0, (2.15)

assuming X1 ≥ 0. Notice that if p1 = p0, then X1 = 0 since at the old home price

level of p0, S0 was the optimal amount of structure.


The first-order condition of home builders at t = 1 is given by

L1 =wS,1wL,1

α

1− αS1. (2.16)

And imposing the zero profit condition of homebuilders yields

p1

wαL,1(

α

1− α)α(1− α) = w1−α

S,1 . (2.17)

We then plug in the two supply functions into the first-order condition of the

homebuilder to write wL,1 as function of wS,1 and X1:

w1+γγ

L,1 = wS,1(α

1− α)[w

1β

S,1 − (1− δ)X1]. (2.18)

Note that S1 = δ−1[w1β

S,1−(1−δ)X1] and so δ−1 cancels from the first-order condition.

Then plug wL,1 into the zero-profit condition for equation in wS,1 to get

p1{wS,1α

1− α[w

1β

S,1 − (1− δ)X1]}−α( γ1+γ

)(α

1− α)α(1− α)− w1−α

S,1 = 0. (2.19)

Let x = (1− δ)X1 and w = wS,1 and write equation (2.19) as F (x,w) = 0.

If p1 > p0, X1 > 0. To show that wS,1 > wS,0 amounts to showing that dw/dx >

0. By the implicit function theorem, dw/dx = −Fx/Fw. It is easy to show that that

Fx > 0 and Fw < 0 and hence dw/dx > 0. Hence, we have the following result:

Proposition 1 If p1 > p0, the equilibrium cost of structure at t = 1, wS,1, is greater

than wS,0, the equilibrium cost of structure at t = 0.

2.2.4. Extensions of Model

Before we derive the comparative statics results that will be the basis of our em-

pirical work, it is worth mentioning that we can extend our model in home im-


provements. In our current set-up, we have assumed that home improvers mistake

costs of improvements. They think they can do the improvements at the time-0 cost

(wS,0) when it fact the time-1 cost (wS,1) will prevail. And as we have just shown,

in equilibrium, the time-1 cost will be higher than the time-0 cost, thereby leading

to a potentially costly mistake.

We could have also modeled this overconfidence in remodeling by additionally

assuming the home improvers lose a fraction of the structures that they build (λ)

and that this loss fraction (λ) of structures increases the higher is the cost growth.

This is a way of modeling that with higher cost growth, improvers are overconfident

about the amount of structures that actually get built.

Either modeling choice yields the same economics, which is that a mistake either

in terms of costs or of the amount of structures that get built will result in a lower

recoup value, defined as the ratio of resale value or value-added of improvements

to construction costs of improvements. Hence, recoup value and its relationship to

construction costs are a natural focus of our analysis since they potentially measures

the mistakes of home improvers. But the economic magnitudes of our model would

be larger if we added this additional channel for overconfidence.

2.3. Comparative Statics

In this section, we take the equilibrium results from the previous section to derive

some comparative statics.

2.3.1. Improvements Are a Homogenous Function of De-

gree Bigger than One of Home Price Appreciation

The first result concerns the relationship between the growth of structures that

result from home improvement and home price appreciation. Recall that we assume


p1 > p0 and that as a result X1 > 0 in equilibrium. The following result relates

the growth of structures that result from home improvement X1+S0

S0to home price

appreciation p1p0

.

Proposition 2 The growth of structures that results from home improvement is a

homogeneous function of degree 1α> 1 of home price increases, that is:

X1 + S0

S0

=

(p1

p0

) 1α

. (2.20)

To see this, note that from the profit maximization of the professional home

builders at t = 0

p0Lα0 (1− α)S−α0 − wS,0 = 0. (2.21)

Because S0 was optimal and so we have

S−α0 =wS,0p0

1

1− α1

Lα0. (2.22)

The result follows from the expression for the X1 that solves the homeowners’

problem. The sensitivity of home improvement to price appreciation depends on

the share of land in the value of houses. If, for instance, α = 2/3 (i.e. land is two-

thirds of the value of homes), than the degree of homogeneity is 3/2. As α decreases,

the degree of homogeneity (and hence the convexity of the relationship) increases.

The first implication of the proposition is that improvements increase with in-

creases in home prices. An increase in home prices increases the optimal amount

of structures in a given plot of land. Normally, this effect would be moderated by

the increase in the cost of structures. However, overconfident owners perceive an

ability to produce more valuable homes at old structure prices, and thus ignore the


increase in the cost of structures. This is consistent with the pro-cyclical pattern in

remodeling expenditures with home prices.

The second implication of the proposition is the homogeneity of degree 1α> 1 of

this relationship. This result is more special and depends on the Cobb-Douglas pro-

duction functions. However, more general production functions would still imply a

similar relationship between growth in structures resulting from home improvement

and price appreciation. Because homeowners (wrongly) assume that the price of

structures is constant, a doubling of home prices induces an increase in the amount

of structures so that, holding land constant, the marginal product of structures in

the production of housing services is halved. For many of the common production

functions, such an increase requires more than doubling the amount of structures

because the marginal product of structures falls slowly.

A simple calibration shows that the economic forces we highlight here can ex-

plain in part the boom in improvements during the recent housing bubble in contrast

to the relatively moderate increase over the previous decade. If overconfident home-

owners base their calculations on prices that prevailed at a fixed lag, a large price

change in a given year engenders a bigger response in remodeling investments than

do a series of small increases across many years. In fact, the cumulative house price

increase between the ten-year period of 1993 to 2003 is roughly equal to that of the

five-year period of 2003 to 2007 - both are around 100%. This means that the home

price appreciation per year during the latter 5 year period is roughly double that

of the per year appreciation in the earlier decade. Remodeling expenditures grew

only modestly during the earlier ten-year period while they nearly doubled during

the five-year period of 2003 to 2007 (rising from 1% of GDP to 2% to GDP as we

discussed in the introduction). Of course, this calibration is necessarily rough and

should be taken with a grain of salt as many other factors also changed during this


period. Notably, the expansion of easy credit no doubt contributed to the explosive

growth as well.

2.3.2. Excessive Remodeling

The second result we derive concerns the over-investment in structures by existing

homeowners. Notice that we have shown that if p1 > p0 than wS,1 > wS,0. The next

proposition relates the growth on cost of construction to excessive investments.

Proposition 3 Let X̂1 be the investment of the homeowner at the right equilibrium

price wS,1 (as opposed to the old construction cost of wS,0). Then,

X1 + S0

X̂1 + S0

=

(wS,1wS,0

) 1α

, (2.23)

that is, there is is excessive investment in structure at t = 1 and the percentage of

excessive investments is a a function of the growth in the price of structures that is

homogenous of degree 1α> 1.

Equation (2.23) follows immediately from Equation (2.15), the first-order condition

for the homeowner’s home improvement problem. In particular, investments are

more excessive if in equilibrium the price of construction increased more between

time 0 and time 1.

This result is intuitive and follows directly from the overconfidence of the home-

owners. Since they get the structure price wrong, they end up over-investing in home

improvements. This finding matches conventional wisdom that homeowners when

they undertake remodeling projects usually under-estimate construction costs. This

result is not testable since we cannot measure the counter-factual of the right level

of home improvements taking into account the right structure cost. But it is valu-

able as a welfare result and providing a gauge of when such welfare losses are likely


to be high—i.e. in instances or regions where construction costs have grown a lot.

Otherwise, the mistakes would be minor and not consequential. During the housing

bubble years, as we show below, construction costs for home improvements went

up a lot and hence our analysis here suggests that the welfare losses are likely to

have been substantial.

2.3.3. Recoup Value

The third result we derive, which is testable and that is the focus of our empirical

work below, concerns recoup value, defined as the ratio of resale value or value-added

to construction costs of improvements. Note that we could derive similar results for

resale values using the alternative set-up to our model. But recoup value is the most

natural and captures mistakes in estimated resale value or in construction costs.

To begin with, recall that homeowners choose X1 to solve

MaxX1

(p1L

α0 (X1 + S0)1−α − wS,0X1

).

Homeowner’s gains from the added investment are thus

p1Lα0 (X1 + S0)1−α − p1L

α0S

1−α0 .

The recoup value is the ratio of gains to cost that is:

R :=p1L

α0 (X1 + S0)1−α − p1L

α0S

1−α0

wS,1X1

.

Notice that in computing recoup value we use the actual costs paid by home-

owners. In principle the recoup value can be greater than or less than one. On the

one hand homeowners face decreasing returns, and if they made no mistakes the

recoup value would exceed one. On the other hand, they underestimate the cost


of construction services what leads to ex-post losses. A lower bound for the recoup

value iswS,0wS,1

< 1.

To derive an upper bound, notice that since for any concave function f , f(y)−

f(x) ≤ f ′(x)(y − x) we have:

R ≤ p1(1− α)Lα0S−α0 X1

wS,1X1

=p1

p0

p0(1− α)Lα0S−α0

wS,1=p1

p0

wS,0wS,1

,

where the last equality comes from the first order condition at time zero.

Since X1 = X1(p1), the recoup value R = R(p1, wS,1). Changes in p1 also move

wS,1, but other parameters also affect wS,1, such as the elasticities of supply of land

and construction services. Hence it makes sense to examine the change in R holding

p1 fixed. It is straightforward to check that

∂R

∂wS,1< 0.

In our model, prices have been at a steady state p0 and then change to p1. The

cost of structures stay constant at wS,0 and then move to wS,1. Hence until the price

change homeowners make no mistakes concerning the cost of structures but have

no opportunity to invest. The recoup value prior to the price change is 1. Thus

R = R(p1, wS,1)− 1, can be thought as the change in recoup value. The result here

predicts that the change in recoup value is negatively related to the increase in

the cost of construction, holding the price change in housing services constant. A

higher p1 leads to a higher w1 but may also benefit recoup values. For this reason

the change in recoup value is not necessarily negatively related to an increase in

the cost of construction unless we control for the change in the price of housing

services.


As such we can formally state the following which is the focus of our empirical

work.

Proposition 4 The change in recoup value is negatively correlated with construc-

tion cost growth controlling for home price appreciation and positively correlated

with home price appreciation controlling for construction cost growth.

2.4. Empirical Work

Our empirical analysis centers on testing our theory’s prediction regarding the

change in the recoup value of remodeling projects being negatively correlated with

construction cost growth controlling for the percentage change in home prices. We

test this prediction using data across US cities and regressing the percentage change

in recoup value of projects in each city on the percentage change in construction

cost of these projects and the percentage change in home prices.

2.4.1. Data

Our data on home improvement projects comes from the Cost vs. Value Report

provided by Realtor Magazine, which is the official monthly magazine of the Na-

tional Association of Realtors. It provides information regarding costs and resale

value (value added to home) for a range of major home improvement projects across

many cities starting in 1998 and ending in 2009. (Note that magazine did not con-

duct a survey in 2006 and hence data is missing for this year.) The projects and

cities are listed in Table 1. Panel A lists the projects covered ranging from attic

bedroom remodel to deck addition. For some of the projects on this list, starting in

2002, the magazine separated the projects into a mid-scale and an up-scale version

that differ in terms of square footage and use of high-end finishes. These projects

are labeled with an a mid- and up-scale in parenthesis. The core projects covered


have increased slightly from 12 in 1998 to 14 now. So the set of projects has re-

mained relatively stable over time. These projects represent major remodeling as

opposed to small ones such as replacing a door which would more appropriately

be labeled as maintenance. Panel B lists the cities surveyed. The magazine started

with 60 cities in 1998 and increased the coverage to 80 cities in 2009. The list of

cities includes the 35 metro markets listed as the top remodeling markets by the

Harvard Joint Center for Housing Studies.

For each remodeling project in each city, the database reports for each year

its cost, resale value and recoup value, which is simply the resale value divided

by the cost. The cost and resale figures in the annual reports are compiled sepa-

rately. The cost figures come from HomeTech Information Systems, a remodeling

estimating software company in Bethesda, Maryland. HomeTech collects current

cost information quarterly from thousands of contractors nationwide. The project

costs are based on estimates for hypothetical projects made by the firm based on

their information of the real costs of projects actually undertaken. The figures in-

clude markup and are adjusted to account for pricing variations in different parts

of the country. The construction cost figures include labor, material, subtrades, and

contractor overhead and profit. The cost data in the early years of the sample were

split among various firms but for most of the sample since has been completed by

HomeTech Information Systems.8 HomeTech is a leading provider of construction

cost estimates for various projects across the country.

It is interesting to consider what is driving differences in home improvement

costs across different cities. We gathered anecdotal evidence on this issue by reading

material provided by HomeTech on their website and various sources in the internet

8For 1998, cost estimates for projects were from three publishers of constructioncost estimating guides and software: Craftsman Books, Carlsbad, CA; HomeTechInformation Systems, Bethesda, MD; and R.S. Means, Kingston, MA. For 1999,R.S. Means, Kingston, MA was the only source for the cost estimates.


media regarding home remodeling. Interestingly, wages in different cities accounts

for only a fraction of the differences in costs. There are also variations in lumber

and materials prices in different cities depending on factors such as the number of

lumber suppliers in that city according to HomeTech. Moreover, a bigger source of

cost inflation that is often reported for the high-end improvement projects that are

at the center for our data set is that there is a trade-off in terms of the speed of

construction and cost. For instance, consider the cost of an Italian marble bathroom.

The biggest cost is procuring the Italian marble and depending on search time the

cost differences can be substantial. In cities in which people are in a rush to build,

busy contractors are less picky about how much such quality items cost. Instead of

searching for an extra month to find the same quality at a lower price, the contractor

uses the most expensive one. This is the big cost difference across cities and over

time between building during the bubble years versus the non-bubble years.9 So

one should think of variations in remodeling costs from this database across cities

as driven by these local supply factors in addition to any wage differences. We look

at this issue of wages in residential construction more closely and relate it to our

model in the robustness sub-section below.

Resale values (or value-added) of these home improvements are based on the

professional judgment of members of the National Association of Realtor (NAR).

To obtain these judgments, the magazine sent out surveys containing customized

cost-to-construct data for each city, as well as information on median house prices,

via e-mail to appraisers, sales agents, and brokers. The magazine gives instructions

in the survey for the respondents not to make judgments about the motivation of

the homeowner in either the decision to undertake the remodeling project or to

9See for instance coverage on this issue at the follow-ing site: http://moneywatch.bnet.com/saving-money/article/five-home-renovations-that-pay-off/353008/. See also the description inhttp://www.remodeling.hw.net/remodeling/cost-vs-value-report-2006.aspx.


sell the house. The real estate professionals then responded with dollar figures for

each remodeling project that represent the value the completed project would add

to the selling price of the house under current market conditions. The surveys are

obtained from hundreds of real estate professionals and aggregated to provide the

resale values in the magazine.

We construct for each city by year a COST variable which is simply the equal

weighted average of the costs of the projects in that city. We construct a similar

variable RESALE which is the average of the resale values for the projects in that

city. We divided RESALE by COST to obtain RECOUP, which corresponds to

recoup value in our theoretical analysis and is our main dependent variable of

interest. Our home price index for each city obtained from the Federal Housing

Finance Agency is HPI. Data for HPI is defined at the level of the Metropolitan

Statistical Area. So we use as the home price for each city the price of the MSA

that it belongs to.

Our analysis largely focuses on the three variables of COST, RECOUP, and HPI.

Indeed, our theory can be completely summarized by these variables. To test our

theory, we really do not need to look at any other variables. But for completeness of

analysis, we include demographic variables in our robustness checks. They should

have no material impact on the relationships we are predicting. As such, we will

not have many comments regarding the demographic variables. The relationships

of these demographic variables to our variables of interest in all likelihood are

determined by complicated equilibrium effects.

The demographic variables are as follows. INC is income per capita in the city.

POP is population in the city. We do not have data for INC and POP for 2009 yet

and so in our analyses involving these variables we can only look at changes in INC

and POP up to 2008 instead. These two latter variables are from the Bureau of


Economic Analysis. UNEMP is the unemployment rate in a city, which is provided

by the Bureau of Labor Statistics.

Table 2 provides the summary statistics for our main variables of interest. In

Panel A, we report the time series average (across years) of the cross-sectional

(across cities for each year) means and standard deviations for the following vari-

ables. The mean COST of a typical project in a typical city is $39,189 with a

standard deviation of $4193. As these numbers attest, the projects are major re-

modeling undertakings and most of them have roughly the same average value. The

mean RESALE is $29428 with a standard deviation of $7959. The recoup value is

0.78 with a standard deviation of 0.18. The recoup value is on average less than one

and there is a reasonable difference in recoup values across cities. The mean HPI is

162 with a standard deviation of 26. The per capita income (INC) is $35,437 with

a standard deviation of $6319. The population of a typical city in our sample is

around 1.2 million people with a standard deviation of around 1.5 million people.

The unemployment rate (UE) is around 6% with a standard deviation of 2%.

Panel B provides the time-series average of the cross-sectional correlations of

these variables of interest. The key things to observe from this panel are the fol-

lowing. COST and RESALE are positively correlated with a correlation coefficient

of 0.61 and COST and RECOUP are also positively correlated with a coefficient of

0.23. In other words, cities with higher COST also have higher RESALE and RE-

COUP values. This positive correlation is in levels and captures potentially many

unobservables. Hence, COST and RECOUP need not be mechanically hardwired

to be negatively correlated since RESALE also adjusts. Our theory does not have

much to say about levels since it is geared to understanding the implications of

how a change in home prices stimulates home improvements. When we look at

changes in these variables, the comment regarding the hard-wiring of COST and

RECOUP in levels applies to changes—that is, the correlation of changes in these


two quantities need not be hard-wired to have a particular relationship. Moreover,

the correlation of HPI with COST, RESALE and RECOUP are all positive. Hence,

cities with higher home prices captured by HPI also have higher COST, RESALE

and RECOUP values.

2.4.2. Results

Our theory is concerned with the relationship between changes in COST and

changes in RECOUP associated with a change in home prices. To this extent,

our empirical design is centered around looking at how COST and RECOUP val-

ues changed during the bubble period of 2003-2009. Figure 1 plots the time series

for the average COST and RECOUP across cities in each year. COST has gone up

substantially from a low of around $20000 in 1998 to around $60000 in 2009. These

are nominal levels and reflect the inflation in wages, material costs and potentially

other types of zoning and regulatory costs. RECOUP starts at 0.8 in 1998 and

rose to a high of 0.9 in 2005 and then decreases with the collapse of housing prices

starting in 2007 and into 2009. The year 2003 or 2004 is generally viewed as the

beginning of the housing bubble period. To the extent this is the case, these figures

suggest that in the aggregate, RESALE values did not keep up with COST during

the housing bubble period of 2003 to 2009, which is consistent with our theory. Of

course, any strong inference here is impossible given the limited time series.

Hence we will try to test our prediction by looking at the correlation of changes

in recoup values with changes in construction costs across cities. We now turn to

our first set of results, reported in Table 3, concerning the relationship between

changes in recoup value, %∆RECOUP and changes in cost %∆COST. We focus

our analysis here on the changes in these variables during the period of 2003 to

2009. This is our baseline sample period. We will consider robustness checks below

for 1998 to 2009 and 2003 to 2007.


Panel A reports the summary statistics for changes in our variables of interest

over this period. Construction cost grew by about 57% with a standard deviation

of 7%. Recoup values fell on average by 14.6% with a standard deviation of 21.2%.

Home prices rose on average by about 20.8% with a standard deviation of 15.9%.

Income and population also expanded and so did unemployment since we have 2009

in our sample. The income and population numbers are calculated from 2003 to

2008 since we do not yet have data for 2009 for these two variables.

Panel B reports the correlation of these variables over this period. The key

take-aways are the following. The percentage changes in COST and RECOUP are

negatively correlated across cities. Higher HPI is correlated with higher RECOUP.

These correlations are consistent with our theory. Home improvers gain from their

improvements when home prices are higher and lose when construction costs growth

are higher.

Panel C reports the results for our main regression of interest: the percentage

change in recoup value on percentage change in cost. In column (1), we report

the simple univariate regression with only the percentage change in cost on the

right hand side. The regression has 35 cities as observations. The coefficient of

interest is -1.4 with a t-statistic of 3.5. A standard deviation of %∆COST is 0.07.

So a one standard deviation increase in %∆COST is associated with an increase

in %∆RECOUP of 0.098 (-1.4*0.07=0.098) which is about a 46% of a standard

deviation of %∆RECOUP. The economic significance is clearly very sizeable. The

corresponding plot of this regression is shown in Figure 2(a) with the fitted line

of this regression plotted against the scattered observations. It clearly shows a

prominent negatively sloping relationship between these two variables of interest

that does not seem to be driven by outliers.

In column (2), we add in %∆HPI which comes with a coefficient of 0.293 with a

marginally significant t-statistic of 1.6. The sign of this coefficient is consistent with


the model. The coefficient in front of %∆COST is largely unchanged. In column

(3), we add in all the other demographic variables as controls. We find that the co-

efficient in front of %∆COST is now -1.174 with a t-statistic of -2.95. The economic

and statistical significance is cut down a bit but is nonetheless quite sizeable. The

coefficient in front of home price change is now 0.563 and the statistical significance

is higher with a t-statistic of around 3. This specification represents the strongest

evidence of our model in which both of our variables of interest are economic and

statistically significant as predicted. Of the control variables, only the percentage

change income comes in with a statistically significant effect but interestingly, it

attracts a counter-intuitive negative sign. Our model is silent on what to expect

from this change in income variable and this variable probably is better interpreted

in a consumption-based model. It appears this percentage change in per capita in-

come variable is correlated with the percentage change in home prices variable and

its addition to the regression specification boosts the explanatory power of home

price changes. The main take-away for us is that our two variables of interest, cost

growth and home price change both attract the predicted coefficients even with the

inclusion with a number of demographic variables.

These empirical findings are not only consistent with our model but are eco-

nomically large. The large negative correlation between the change in recoup value

and change in cost over this period also captures the fact that resale values are

also lower in the high cost growth cities, which is consistent with the extended

version of our model that the quality of improvements can also suffer as a result

of the speculative bubble and not simply that the cost is higher than estimated.

We are cautious to make too much of this point since our model is so stylized but

suffice it to say that the qualitative predictions are consistent with our model and

its extension.


In Table 4, we consider a number of robustness exercises. In Panel A, we look at

the results when we just consider midrange projects. The results are similar to that

in Table 4. In Panel B, we consider only upscale projects and the results are similar.

In Panel C, we briefly summarize our analysis when we look at the results project

by project. Only the coefficient for %∆COST for each of the project is reported,

though the regressions are run with home price change as a control. The coefficients

for all the projects have the right sign and only one is not significant. In Panel D,

we focus on the period 2003 to 2007. We consider this exercise to see if the collapse

of home prices during the years of 2008 and 2009 are driving all our results. Even

if this were the case, it would not necessarily be inconsistent with our model. It

turns out, however, that the results are largely similar. In Panel E, we consider

what happens when we look at changes over the entire sample period of 1998 to

2009. The results are somewhat weaker as one might expect since much of the

boom in home improvements happens from 2003 to 2007. But they are nonetheless

statistically and economically significant.

In Table 5, we take a more in depth look at which cities have experienced

the most construction cost growth and hence the worst drops in recoup values. In

Panel A, we sort cities by those with the least to the most construction cost growth

and report their percentage increase. We also report the housing supply elasticity

index for each of these cities from Saiz (2010). This index measures the ease of

response of supply to home price increases. This analysis is motivated in part to

see if there is a connection of home improvements and housing supply elasticity.

We do not have a strong sense ex ante of what to expect. It might be that there

are more improvements in cities with easier supply response, so that improvements

and professional supply response from the construction sector are one and the same.

Alternatively, it might be that home improvements are more likely in cities where

housing supply cannot respond as strongly and improvements are a substitute. We


find a negative correlation between supply elasticity and construction cost growth

of improvements suggesting perhaps that the latter scenario is true. In Panel B, we

add in this housing supply elasticity index into our baseline regressions to see how

our baseline results are affected. We find little change in the coefficients. In other

words, it appears that there is more home improvements in low housing supply

elasticity states which had higher price growth but the recoup values did not keep

up with the construction cost growth of improvements. This finding supports the

overconfidence and excessive improvements in bubble states rendition of our theory.

2.4.3. Robustness Checks and Alternative Interpretations

In this section, we discuss a number of robustness checks we have done but which

we do not report for brevity. First, we have chosen to focus on recoup values in

Table 3 which is the most natural quantity that encompasses both mistakes in

resale value or in construction costs. We could have also focused on resale values

and redone Table 3 with resale values and obtained similar results. In other words,

our results are robust to the choice of recoup or resale values, very much in line

with the economics of our model and the alternative set-up discussion.

Second, several caveats are warranted for the data. The methodology used to

compiled this database was changed slightly in 2007. In 2007, the magazine gave

clearer guidelines on the projects. They argue that this led to more accurate con-

struction costs estimates. In 2009, the Home Valuation Code of Conduct (HVCC)

took effect on May 1, 2009. The effect is to try to bring more fairness into appraisals

by having the appraiser of a home be randomly chosen from a pool. To the extent

that the respondents of the resale surveys are influenced by observed appraisals,

this might have had an effect on their responses. We dealt with this issue using var-

ious beginning and end dates for the regression sample and found roughly similar

results. The recoup value results are robust to these perturbations—some of these


results are presented in the earlier robustness check table. We also gather residential

construction wage data for counties from the Bureau of Labor Statistics and find

that wages in this industry are correlated with the costs of the improvements in

the city and that changes in these costs and changes in these wages over time are

also correlated. This provides a check as to the accuracy of the cost data, though

of course, other factors like zoning or regulatory rules and the other local supply

factors discussed above which vary across cities might also impact substantially

improvement costs.

However, our cost data is better than wage growth to test our theory for a

few reasons. The first is that wage growth in the residential construction sector is

correlated with wage growth in the city. Hence, wage growth also behaves like an

income effect in the city, and a consumption theory predicts that cities with higher

wage or income growth actually should have higher recoup values since households

will want more amenities. We find that this type of an effect is not very significant,

however, in the data. Hence, our change in cost variable is robust to the addition

of a wage growth variable and suggests that the part of the cost variable that is

picking up supply factors like lumber or the cost of Italian marble seems to be,

consistent with our priors and discussion above, the dominant variation influencing

our regressions.

There is an alternative interpretation for all results, which is that households in

the high cost growth cities are so impatient to consume improvements that they are

willing to live with lower quality improvements and hence the lower recoup values

in high construction cost cities. While difficult to rule out, it strikes us as awkward

that the higher demand for amenities in a city would not translate into high resale

values for those amenities in that city. This hypothesis also seems at odds with the

data that improvements are often done by movers who intend on selling them.


2.5. Conclusion

Home improvements are an important economic activity. Such activity accounts

from around 1 to 2% of GDP each year and this is not even accounting for the

opportunity cost of time of many homeowners who tend to under-take these efforts

by themselves. This tremendously important economic activity has been neglected

by researchers. We believe that this topic deserves the same amount of attention

that economists have given to consumption and investment decisions in financial

markets. Our paper provides a first attempt to model the economics of home im-

provements. We develop a theory of home improvements in which overconfident

homeowners mistakenly believe that they can build quality additions at a lower

cost than a competitive home-building industry. This theory yields a number of

results including convexity of improvement expenditures to home price changes,

over-investment in home remodeling, and smaller growth in recoup values in areas

with higher construction cost growth. We test this last prediction using data on

home improvement costs and resale values across US cities. Consistent with our

theory, we find that cities with higher construction cost growth experience lower

recoup value growth during the recent housing bubble period of 2003-2009. Our

analysis suggests that welfare losses associated with excessive home remodeling are

potentially large.


Appendix

Proof of Proposition 1

Note that the first order condition from home improvement decision at t=1 is

(X1 + S0)−α =wS,0p1

1

1− α1

Lα0

and the first order condition from profitmaximization at t=0 is

S−α0 =wS,0p0

1

1− α1

Lα0

If p1 > p0, X1 > 0. If p1 = p0, X1 = 0.

Recall equation (12) and (19).

p0{α

1− αw

1+ββ

S,0 }−α( γ

1+γ)(

α

1− α)α(1− α)− w1−α

S,0 = 0

p1{wS,1α

1− α[w

1β

S,1 − (1− δ)X1]}−α( γ1+γ

)(α

1− α)α(1− α)− w1−α

S,1 = 0

If p1 = p0, X1 = 0 and wS,1 = wS,0. To prove wS,1 > wS,0 when p1 > p0, it is

sufficient to prove that dwS,1/dx > 0.

Let x = (1− δ)X1 and w = wS,1. We can call this implicit function F (x,w) = 0.

By the implicit function theorem, dw/dx = −Fx/Fw. It is easy to show that that

Fx > 0 and Fw < 0 and hence dw/dx > 0 which proves the theorem.


Table 2.1: List of Projects and Cities

The Cost vs. Value Report provides cost, resale value and recoup ratio by project and city. Panel A gives a list of

projects that are covered at various point in time. Projects without sufficient data were dropped. Panel B gives a

list of cities that are covered at various point of time. Note that Virginia Beach, VA was reported as Norfolk, VA

before 2007.

Panel A : List of Projects

Attic Bedroom Remodel Basement RemodelBathroom Addition Bathroom RemodelDeck Addition Family Room AdditionMajor Kitchen Remodel Master Suite AdditionMinor Kitchen Remodel Roofing ReplacementSiding Replacement Sunroom AdditionTwo-Story Addition Window Replacement

Panel B : List of Cities

Albany,NY Dayton,OH Manchester,NH Richmond,VAAlbuquerque,NM Denver,CO Memphis,TN Riverside,CAAllentown,PA Des Moines,IA Miami,FL Rochester,NYAnchorage,AK Detroit,MI Milwaukee,WI Sacramento,CAAtlanta,GA El Paso,TX Minneapolis,MN Salt Lake City,UTAustin,TX Fargo,ND Nashua,NH San Antonio,TXBaltimore,MD Garden City,NY Nassau-Suffolk,NY San Diego,CABillings,MT Grand Rapids,MI New Haven,CT San Francisco,CABirmingham,AL Harrisburg,PA New Orleans,LA Seattle,WABoise,ID Hartford,CT New York,NY Sioux Falls,SDBoston,MA Honolulu,HI Oakland,CA Spokane,WABuffalo,NY Houston,TX Oklahoma City,OK Springfield,MABurlington,VT Indianapolis,IN Omaha,NE St. Louis,MOCharleston,SC Jackson,MS Orange City,CA Tampa,FLCharleston,WV Jacksonville,FL Orlando,FL Tulsa,OKCharlotte,NC Kansas City,MO Passaic,NJ Ventura,CAChicago,IL Knoxville,TN Philadelphia,PA Virginia Beach,VACincinnati,OH Lancaster,PA Phoenix,AZ Washington,DCCleveland,OH Las Vegas,NV Pittsburgh,PA Westchester,NYColorado Springs,CO Little Rock,AR Portland,ME Wichita,KSColumbia,SC Los Angeles,CA Portland,OR Wilmington,DEColumbus,OH Louisville,KY Providence,RI Worcester,MADallas,TX Madison,WI Raleigh,NC



Panel A reports the time series average of cross-sectional means and standard deviations for the variables of

interest. Panel B reports the time series average of correlation matrix for the variables of interest. COST and

RESALE are the equal-weighted measures across different projects by year and city. RECOUP is the recoup ratio

by year and city defined as RESALE/COST. HPI is the house price index by Federal Housing Finance Agency.

HPI is defined at the level of Metropolitan Statistical Area(MSA). HPI of MSA is used when the city belongs to

the MSA. INC is income per capita and POP is population size by year and city. INC and POP for 2009 are

excluded due to the data availability from Bureau of Economic Analysis. UNEMP is the unemployment rate by

year and city from Bureau of Labor Statistics.

Panel A : Means and standard deviations

Variable Mean Std.dev.

COST 39189.3 4193.0RESALE 29428.2 7958.7RECOUP 0.78 0.18HPI 162 26.6INC 35436.9 6319.4POP 1181105 1506280UNEMP 0.06 0.02

Panel B : Correlation Matrix

COST RESALE RECOUP HPI INC POP UNEMP

COST 1RESALE 0.61 1RECOUP 0.23 0.88 1HPI 0.29 0.50 0.45 1INC 0.40 0.52 0.40 0.41 1POP 0.33 0.29 0.18 0.24 0.11 1UNEMP 0.28 -0.01 -0.13 -0.11 -0.24 0.20 1


Table 2.3: %∆RECOUP on %∆COST

Panel A reports the mean and standard deviation of percentage change(%∆) in variables from 2003 to 2009.

Panel B reports the correlation matrix of %∆ in variables from 2003 to 2009. Panel C reports regressions of

%∆RECOUP on %∆COST from 2003 to 2009. In column (1), univariate regression of %∆RECOUP on

%∆COST is reported. In column (2), regression of %∆RECOUP on %∆COST with %∆HPI controlled is

reported. Column (3) reports the regression of %∆RECOUP on %∆COST with %∆HPI and %∆ of demographic

variables controlled. The table reports point estimates with heteroskedasticity-consistent t-statistics in

parentheses. ***, **, * denotes 1%, 5%, and 10% statistical significance.

Panel A : Means and standard deviationsMean Std.dev.

%∆COST 0.572 0.071%∆RECOUP -0.146 0.212%∆HPI 0.208 0.159%∆INC 0.245 0.078%∆POP 0.034 0.077%∆UNEMP 0.424 0.285

Panel B : Correlations%∆COST %∆RECOUP %∆HPI %∆INC %∆POP %∆UNEMP

%∆COST 1%∆RECOUP -0.47 1%∆HPI -0.04 0.24 1%∆INC 0.26 -0.18 0.51 1%∆POP -0.31 0.10 0.06 -0.42 1%∆UNEMP 0.34 -0.20 -0.29 -0.24 -0.16 1

Panel C : %∆RECOUP on %∆COST(1) (2) (3)

Variables %∆RECOUP

%∆COST -1.414*** -1.385*** -1.174***(-3.540) (-3.370) (-2.951)

%∆HPI 0.293 0.563***(1.602) (3.015)

%∆INC -1.109***(-2.978)

%∆POP -0.635(-1.621)

%∆UNEMP -0.0572(-0.567)

Constant 0.663*** 0.586** 0.726***(2.858) (2.665) (2.855)

Observations 35 35 35R2 0.225 0.274 0.354


Table 2.4: Robustness

For certain projects in the list, projects are broken down by midrange and upscale level. Panel A reports

regressions of %∆RECOUP on %∆COST within midrange projects. The dependent variable is %∆RECOUP, the

percentage change from 2003 to 2009 in RECOUP which is equal-weighted measure across different midrange

projects by year and city. In column (1), %∆RECOUP is regressed on %∆COST, the percentage change from

2003 to 2009 in COST which is equal-weighted measure across different midrange projects by year and city. In

column (2), %∆RECOUP is regressed on %∆COST with %∆HPI controlled. Panel B reports regressions of

%∆RECOUP on %∆COST within upscale projects. The dependent variable is %∆RECOUP, the percentage

change from 2003 to 2009 in RECOUP which is equal-weighted measure across different upscale projects by year

and city. In column (1), %∆RECOUP is regressed on %∆COST, the percentage change from 2003 to 2009 in

COST which is equal-weighted measure across different upscale projects by year and city. In column (2),

%∆RECOUP is regressed on %∆COST with %∆HPI controlled. Panel C reports the regression of %∆RECOUP

on %∆COST for individual project with %∆HPI controlled. The dependent variable is %∆RECOUP, the

percentage change from 2003 to 2009 in RECOUP of the project by year and city. %∆COST is the percentage

change from 2003 to 2009 in COST of the project by year and city. Panel D and E report the regressions of

%∆RECOUP on %∆COST for different time horizons, from 2003 to 2007 and from 1998 to 2009. The table

reports point estimates with heteroskedasticity-consistent t-statistics in parentheses. Regressions with different

year combinations are reported in different columns. ***, **, * denotes 1%, 5%, and 10% statistical significance.

Panel A : By Midrange Projects(1) (2)

Variable %∆RECOUP

%∆COST -1.292*** -1.324***(-4.083) (-3.776)

%∆HPI 0.308(1.598)

Constant 0.573*** 0.525***(3.337) (3.079)

Observations 35 35R2 0.184 0.238

Panel B : By Upscale Projects(1) (2)

Variable %∆RECOUP

%∆COST -2.139*** -2.034***(-4.139) (-4.053)

%∆HPI 0.278*(1.756)

Constant 0.503*** 0.412**(3.075) (2.733)



Table 4 (continued)

Panel C : By Individual Project

Project Coefficient on %∆COST t-stats

Basement Remodel (M) -1.152*** -2.753Bathroom Addition (M) -0.528*** -4.518Bathroom Addition (U) -0.848*** -4.671Bathroom Remodel (M) -1.324*** -3.372Bathroom Remodel (U) -0.808** -2.393Family Room Addition (M) -0.923*** -2.756Major Kitchen Remodel (M) -2.344** -2.651Major Kitchen Remodel (U) -1.227 -1.111Master Suite Addition (M) -0.893** -2.255Master Suite Addition (U) -1.891*** -4.234Window Replacement (M) -2.081*** -3.365Window Replacement (U) -1.920** -2.688

Panel D : From 2003 to 2007(1) (2)

Variable %∆RECOUP

%∆COST -1.687*** -1.758***(-2.868) (-2.851)

%∆HPI 0.0780(0.562)

Constant 0.706** 0.711**(2.690) (2.701)


Panel E : From 1998 to 2009(1) (2)

Variable %∆RECOUP

%∆COST -0.469** -0.444**(-2.597) (-2.657)

%∆HPI 0.162(1.624)

Constant 0.755** 0.602*(2.232) (1.870)



Table 2.5: Supply Elasticity

Panel A reports the supply elasticity from Saiz(2010) and %∆COST from 2003 to 2009 across different cities.

The correlation between supply elasticity and %∆COST is reported with heteroskedasticity-consistent t-statistics

in parentheses. Panel B reports regressions of %∆RECOUP on %∆COST, %∆HPI, and Supply Elasticity from

2003 to 2009. The table reports point estimates with heteroskedasticity-consistent t-statistics in parentheses. ***,

**, * denotes 1%, 5%, and 10% statistical significance.

Panel A : Supply Elasticity and %∆COST

City Supply Elasticity %∆COST

Dallas, TX 2.18 0.46Phoenix, AZ 1.61 0.47Houston, TX 2.30 0.47Atlanta, GA 2.55 0.48Cincinnati, OH 2.46 0.50Cleveland, OH 1.02 0.50San Antonio, TX 2.98 0.50Salt Lake City, UT 0.75 0.51Columbus, OH 2.71 0.51Denver, CO 1.53 0.52Indianapolis, IN 4.00 0.53Pittsburgh, PA 1.20 0.53St. Louis, MO 2.36 0.54Seattle, WA 0.88 0.54Virginia Beach, VA 0.82 0.56Buffalo, NY 1.83 0.56Milwaukee, WI 1.03 0.57Miami, FL 0.60 0.57Orlando, FL 1.12 0.57Portland, OR 1.07 0.57Philadelphia, PA 1.65 0.58San Francisco, CA 0.66 0.59San Diego, CA 0.67 0.60New Orleans, LA 0.81 0.60Detroit, MI 1.24 0.61Tampa, FL 1.00 0.62Providence, RI 1.61 0.62Washington, DC 1.61 0.62Minneapolis, MN 1.45 0.65Boston, MA 0.86 0.65Los Angeles, CA 0.63 0.65Chicago, IL 0.81 0.67Kansas City, MO 3.19 0.69New York, NY 0.76 0.74

Correlation -0.377 (-1.97)


Table 5 (continued)

Panel B : %∆RECOUP on Supply Elasticity

Variables %∆RECOUP

%∆COST -1.707***

(-4.241)

%∆HPI 0.385*

(2.002)

Supply Elasticity -0.00167

(-0.0490)

Constant 0.741***

(2.766)

Observations 34

R2 0.375


Figure 2.1: Average COST and RECOUP

The figure plots COST and RECOUP averaged across cities for various years. COST and RESALE are theequal-weighted measures across different projects by year and city and RECOUP is the recoup ratio by year andcity defined as RESALE/COST.

20000

30000

40000

50000

60000

CO

ST

1998 2000 2002 2004 2006 2008year

.7.7

5.8

.85

.9R

EC

OU

P

1998 2000 2002 2004 2006 2008year


Figure 2.2: %∆Cost and %∆RECOUP

The figure plots %∆RECOUP and %∆COST across cities. %∆RECOUP is the percentage change in RECOUPfrom 2003 to 2009. %∆COST is the percentage change in COST from 2003 to 2009.

−.6

−.4

−.2

0.2

.4

%∆

RE

CO

UP

.4 .5 .6 .7 .8

%∆COST

−.5

0.5

1

%∆

RE

SA

LE

.4 .5 .6 .7 .8

%∆COST

CHAPTER 3. THE MOTIVE OF HOME IMPROVEMENTS 90

Chapter 3

The Motive of Home

Improvements

3.1. Introduction

Are home improvements just consumption? Many homeowners remodel their houses

for a better quality of life – consumption-motive remodeling. However, homeowners

also take on remodeling projects as an investment, expecting higher resale value

from remodeling. In this paper, I test whether the consumption-motive explanation

of home improvements outweighs the investment-motive explanation.

The distinction between consumption-motive and investment-motive remodeling

is not straightforward. Homeowners may improve their homes as consumption but

may also maintain an optional value of remodeling as an investment. That is, they

still can make a profit upon resale. Similarly, homeowners may improve their homes

as an investment but they also enjoy a better quality of living before they resell their

houses. In the worst case, they can simply consume their remodeled houses. Any

test on remodeling motive cannot be free from this entangled nature of remodeling

purposes.


However, there is a fundamental difference between consumption-motive re-

modeling and investment-motive remodeling. The correlation between remodeling

expenditures and the recoup ratio of remodeling projects will be different based on

remodeling purposes. Consumption-motive remodeling is closely related to the spe-

cific individual preferences. Normally, the costs for personal taste are less valued in

resale price. Therefore, more consumption in remodeling is more likely to lower the

recoup ratio. That is, consumption-motive remodeling will be negatively correlated

with recoup ratio. In contrast, recoup ratio is a primary concern for investment, and

investment-motive remodeling will be positively related to recoup ratio. Using these

opposite correlations between the recoup ratio of remodeling projects and remod-

eling expenditures, I test the null hypothesis that consumption-motive remodeling

comprises the majority of remodeling activity. The alternative hypothesis will be

the importance of investment-motive remodeling in overall remodeling market.

I collect various data sets on home improvements from various sources. I use the

Home Mortgage Disclosure Act (HMDA) data to measure mortgage credit available

for remodeling. Remodeling expenditures are from the American Housing Survey

(AHS), the Survey of Residential Alterations and Repairs (SORAR), and the Joint

Center for Housing Studies (JCHS). The recoup ratio of remodeling projects is from

Realtor Magazine published by the National Association of Realtors.

I report empirical results of time-series and cross-sectional analysis. Time series

analysis shows the following results. First, remodeling expenditures are positively

associated with home price appreciation (HPIAPP). In terms of economic signif-

icance, a one-standard deviation increase in HPIAPP is associated with a 0.22-

standard deviation increase in remodeling expenditures.

Second, remodeling expenditures are positively correlated to the mortgage credit

available for home improvement projects. The number of mortgage applications for

remodeling projects (APPL) and the amounts of the mortgage originations for


remodeling projects (AMTO) show positive association with remodeling expendi-

tures. A $1 billion increase in AMTO increased remodeling expenditures by $566

million. Economic significance is also large. For instance, a one-standard deviation

increase in AMTO is associated with a 0.55-standard deviation increase in remod-

eling expenditures. This indicates the importance of mortgage credit availability

on remodeling expenditures. When more money is available due to increased mort-

gage credit, homeowners increase remodeling expenditures. Note that the increase

in remodeling expenditures may come from either consumption-motive remodeling

or investment-motive remodeling.

Third, and the most significant is the result that remodeling expenditures are not

significantly associated with the recoup ratio of remodeling projects (RECOUP).

If the consumption-motive were the only reason for remodeling, we would expect

to observe a significant negative relationship between RECOUP and remodeling

expenditures. Insignificant results can be interpreted as a mixture of two remodel-

ing purposes: positive correlation from investment-motive remodeling and negative

correlation from consumption-motive remodeling. This result rejects the null hy-

pothesis on the dominant role of consumption-motive remodeling.

Cross-sectional analysis shows similar results. HPIAPP is positively associated

with remodeling expenditures. AMTO and APPL are also positively correlated with

the number of remodeling projects as well as remodeling expenditures. For instance,

a 1% increase in AMTO increased 1% of remodeling expenditures. RECOUP also

shows similar results to the time series analysis. RECOUP shows an insignificant

relationship with the number of remodeling projects to remodeling expenditures. As

I explained, this finding supports the important role of investment-motive remodel-

ing. Otherwise, we would expect to see a significant negative relationship between

recoup ratio and remodeling expenditures.


In addition to the other results, housing supply elasticity shows a negative as-

sociation with the number of remodeling projects and remodeling expenditures.

This indicates that more remodeling expenditures existed in areas with an inelastic

housing supply. The elasticity of housing supply can be interpreted in two differ-

ent ways. First, low elasticity means that it is hard to build a new home in the

area. Since new homes can’t be built easily, it is normal to have high demand for

remodeling in those areas. Second, low elasticity also means that home prices will

appreciate more when there is any positive demand shock. For remodeling expen-

ditures as an investment, a low elasticity area gives better chance to make profit.

Consequently, the significant results of housing supply elasticity incorporate both

remodeling purposes.

This paper is closely related to Choi, Hong, and Scheinkman (2012) who devel-

oped a speculative-based theory of home improvements. This paper studied home

improvements as an investment and found empirical evidence of speculation in the

remodeling market during the recent housing bubble period. However, Choi, Hong,

and Scheinkman (2012) didn’t deal with the potential impact of mortgage credit

expansion on the remodeling market. Since increased mortgage credit availability

can increase consumption-motive remodeling as well as investment-motive remod-

eling, the additional credit channel during housing bubble period may limit the

conclusion in Choi, Hong, and Scheinkman (2012). The focus of this paper is the

importance of investment-motive remodeling during the housing-bubble period.

This paper contributes to the literature on the remodeling market. Montgomery

(1992) developed a partial equilibrium model of the consumption-motive and the

trade-off between improvements and moving in the presence of moving costs for

households. The paper then estimated a demand function for home improvement

expenditures of households. This paper focuses on the difference in consumption-

motive remodeling and investment-motive remodeling. Helms (2003) found some


determinants of remodeling likelihood, such as the characteristics of a building and

its neighborhood. Gyourko and Saiz (2004) found that the ratio between marginal

cost and the values of marginal investments is an important determinant of remod-

eling. While these papers found micro-level determinants of remodeling, our paper

focuses on macro-level determinants.

This paper also contributes to the literature regarding the role of mortgage

credit expansion during the recent housing-bubble period. The impact of mortgage

credit expansion on housing prices has been studied in various ways.1 However,

none of the studies focused on the remodeling market in connection with mortgage

credit expansion.

The rest of the paper proceeds as follows: Section 2 explains main question of

the paper; Section 3 explains and summarizes the data; Section 4 reports time-series

analysis; Section 5 reports cross-sectional analysis; Section 6 concludes.

3.2. Consumption vs Investment

Homeowners are able to make a profit from home improvement projects, especially

when home prices appreciate. In theory, the value of a property can be divided into

the value of the land and the value of the structure. When home prices appreciate,

the increased value in newly built home comes from both parts. However, the in-

creased in value in existing homes comes mostly from the rise in land value. Given

the land size, the marginal value of additional structure rises as the value of the

property rises. Additional structures will yield higher value-added than the cost of

the addition. Thus, it is profitable to remodel existing homes when home prices

appreciate, and remodeling as an investment makes sense.

1See Khandani, Lo, and Merton (2010), Demyanyk and Van Hemert (2009),Mian and Sufi (2009), Glaeser, Gyourko, and Saiz (2008), Pence (2006).


Choi, Hong, and Scheinkman (2012) studied home improvements as an invest-

ment and found some evidence of speculation in the remodeling market during the

recent housing-bubble period. If the local supply of inputs, such as construction

materials and labor, are inelastic in the short term, higher remodeling activity will

increase construction costs. Using the Realtor Magazine data, they found a negative

correlation between the change in remodeling costs and the change in the recoup

ratio from the remodeling projects. In other words, value added from remodeling

projects could not keep up with cost increases. If speculative investment in remod-

eling projects was the main reason for the cost increase, the result would indicate

financial loss due to competition in speculation.

However, Choi, Hong, and Scheinkman (2012) didn’t deal with the potential

impact of mortgage credit expansion on the remodeling market. This was mainly

due to the simplicity of their model. Home price appreciation enabled homeowners

to borrow more against increased home equity. Homeowners might want to remodel

their houses with home equity credit if lack of funds was the reason for remodeling it

earlier. The increased budget would then promote consumption-motive remodeling

projects as well as investment-motive remodeling projects. Therefore, it is possible

that consumption-motive remodeling from additional financing was responsible for

some part of the cost increase during recent housing-bubble period. As figure 1

shows, the rise in remodeling expenditures was accompanied by the rise in mortgage

credit and this additional credit channel may limit the conclusion in Choi, Hong,

and Scheinkman (2012).

The distinction between consumption-motive and investment-motive remodeling

is not straightforward. Homeowners may improve their homes as consumption but

may also maintain an optional value of remodeling as an investment. That is, they

still can make a profit upon resale. Similarly, homeowners may improve their homes

as an investment but they also enjoy a better quality of living before they resell their


houses. In the worst case, they can simply consume their remodeled houses. Any

test on home improvements cannot be free from this entangled nature of remodeling

purposes.

The question then becomes whether consumption-motive remodeling outweighs

investment-motive remodeling. To test whether the consumption-motive home im-

provements were the dominant driver of remodeling activities during the period,

I characterize the determinants that would show different predictions on the two

remodeling expenditure purposes. By testing how these determinants are associated

with remodeling activities, I find the scenario that the data supports.

I use the recoup ratio of the remodeling projects, which has different implications

by remodeling purpose. Consumption-motive remodeling is more related to personal

preference and less related to the recoup ratio from remodeling projects. With extra

mortgage credit available, homeowners can increase consumption-motive remodel-

ing but additional remodeling is more likely to lower the recoup ratio. I expect

consumption-motive remodeling to be negatively correlated with the recoup ratio.

In contrast, investment-motive remodeling should be more positively related to

the recoup ratio than consumption-motive remodeling. I expect investment-motive

remodeling to be positively correlated with the recoup ratio.

3.3. Data

I have collected data from several sources for the analysis including: 1) loan-level

mortgage credit data for home improvements from the Home Mortgage Disclo-

sure Act (HMDA); 2) household-level remodeling expenditures from the American

Housing Survey (AHS); 3) national-level remodeling expenditures from the Survey

of Residential Alterations and Repairs (SORAR) by the Bureau of Labor Statistics

(BLS); 4) national-level remodeling expenditures from the Joint Center for Hous-


ing Studies of Harvard University; 5) MSA-level house price index (HPI) from the

Federal Housing Finance Agency (FHFA); 6) state-level mortgage rates from the

Monthly Interest Rate Survey (MIRS) by the FHFA; 7) MSA-level elasticity of

housing supply data from Saiz (2010); 8) MSA-level unemployment rates from the

BLS; 9) MSA-level income and population data from the Bureau of Economic Anal-

ysis (BEA); and 10) the recoup ratio of various remodeling projects from Realtor

Magazine published by the National Association of Realtors.

3.3.1. Mortgage Credit for Home Improvements

I use the mortgage credit data for home improvement projects from the Home

Mortgage Disclosure Act (HMDA). Enacted by Congress in 1975, HMDA requires

most mortgage originators located in metropolitan areas to report basic attributes

of mortgage applications. Because HMDA covers a wide number of loans across the

country, using HMDA data offers great advantage in analyzing the loan origination

process. HMDA is known as the most comprehensive source of mortgage origination

data. Choi (2012) reported that HMDA covered almost 84% of total originations

from 1998 to 2007.

HMDA data includes loan characteristics such as loan type (conventional or

government-backed), loan purpose (home purchase, home improvement, and re-

financing), type of property (single-family and multi-family), loan amount, and

location (state, county, and Census tract). In addition, HMDA data provides the

status of applications, that is, whether a loan was originated or denied. Using loan

purpose information, the availability of mortgage credit for home improvements can

be identified separately.

I use HMDA data from 1998 to 2009. This includes the years of the recent

housing market boom, which is the period of interest for this study. Using the five-

digit Federal Information Processing Standard (FIPS) code, I aggregate loan-level


data into MSA-level data using the MSA definition as of December 2009. In total,

363 MSAs are defined.

3.3.2. Households’ expenditures on Home Improvements

Collecting high quality, detailed remodeling data continues to be a challenge for

the industry. Quality data at the national level is limited, and data on remodel-

ing spending at the state and local level is not routinely collected. The Survey of

Residential Alterations and Repairs (SORAR) and the American Housing Survey

(AHS) were the sources of remodeling expenses at the local level, but SORAR was

discontinued in 2007. Currently, the American Housing Survey (AHS) remains one

of the best data sources at the local level.

To measure remodeling activity, I use households’ expenditures on home im-

provements from AHS. AHS is sponsored by the Department of Housing and Urban

Development (HUD) and is conducted by the U.S. Census Bureau. AHS provides

current information on a wide range of housing subjects, including households’

expenditures on home improvements.

Since the AHS is released once every two years, I use national AHS data for

1999, 2001, 2003, 2005, 2007, and 2009. National AHS data includes the Standard

Metropolitan Statistical Area (SMSA)2 information for each household. I link the

SMSA definition to the 2009 MSA definition. In total, AHS data is available for

102 MSAs.

For measuring national-level remodeling expenditures, I use SORAR data from

1994 to 2007. SORAR provides the expenditures on Maintenance and Repairs and

the expenditures on Home Improvements separately. Maintenance and Repairs ex-

penditures represent current costs for incidental maintenance and repairs that keep

2In 1959, the term SMSA replaced SMA for the official metropolitan areas de-fined by the then Bureau of the Budget. The term SMSA was used until MSAs,CMSAs, and PMSAs were introduced in 1983.


a property in ordinary working condition, rather than additional investment in the

property. Home Improvements expenditures are capital expenditures, which add to

the value or useful life of a property. I also use the remodeling expenditure estimates

at the national level from the Joint Center for Housing Studies (JCHS) of Harvard

University. Note that SORAR includes owner-occupied and rental property, but

JCHS data only includes owner-occupied property.

In Table 1, Panel A reports national level expenditures on remodeling from

various data sources and Panel B reports the correlations among them. For the

SORAR estimates, I report the remodeling expenditures on Home Improvements.

Amounts of expenditures differ by data sets due to the different scopes of measures,

i.e. whether the data includes rental property. But all of them are highly correlated,

which means that time-series variations are in common.

3.3.3. Remodeling Cost, Resale Value, and Recoup Ratio

Data on the recoup ratio comes from the Cost vs Value Report provided by Realtor

Magazine, which is the official magazine of the National Association of Realtors.

It provides information regarding costs and resale value for a range of major home

improvement projects across many cities starting in 1998 and ending in 2009. Choi,

Hong, and Scheinkman (2012) used the data to test the speculative-based theory

of home improvements. For this paper, I construct an equal-weighted recoup ratio

across various remodeling projects. The recoup ratios for 91 cities are available from

1998 to 2009, with the exception of 2006.

3.3.4. Local Housing Market and Demographics

To control for the condition of local housing markets, three variables are considered.

First is the house price index (HPI) from the Federal Housing Finance Agency


(FHFA).3 The HPI Index is available for all 363 MSAs from 1998 to 2009. Regional

mortgage rates from the Monthly Interest Rate Survey (MIRS) by the FHFA is the

second variable considered.4 The survey provides annual information on interest

rates, loan terms, and house prices by state. Data for all 51 states is available from

1998 to 2009.

The third variable is the elasticity of housing supply from Saiz (2010). Using

satellite-generated geographic data, he measured the supply elasticity of housing as

a function of both the amount of developable land and regulatory constraints.5 How-

ever, supply elasticity is defined by the MSA definition as of 1999 and a significant

change in the MSA definition occurred in 2000. I link up 1999 MSAs with current

MSAs by hand-matching counties in MSAs. In total, I find 280 MSAs according to

the current definition where the housing supply elasticity is applicable.

In addition to the controls for the regional housing markets, local demograph-

ics are considered. MSA-level unemployment rates are taken from the BLS data.

Average personal income and population of MSAs are taken from the Bureau of

Economic Analysis (BEA). Demographic variables are available for all 363 MSAs

from 1998 to 2009.

3In 1975, the HPI was developed by the Office of Federal Housing EnterpriseOversight, or OHFEO as a regulator of Fannie Mae and Freddie Mac. In 2008,OHFEO became a part of FHFA. The HPI was called the OHFEO HPI, but now iscalled as the FHFA HPI. The FHFA HPI is a weighted, repeat-sales index of single-family house price. It means that the HPI measures average price changes in repeatsales or refinancings on the same properties in 363 MSAs. Information is obtainedby mortgage transactions that have been purchased or securitized by Fannie Maeor Freddie Mac. Since the HPI Index only includes houses with mortgages withinthe conforming amount limits, the index has a natural cap and does not accountfor jumbo mortgages.

4The survey was conducted by asking a sample of mortgage lenders to reportthe terms and conditions on all single-family, purchase-money, non-farm loans. Thesurvey excludes FHA-insured and VA-guaranteed loans, multi-family loans, mobilehome loans, and loans created by refinancing another mortgage. Although the dataonly incorporates home purchase loans, it is still useful to control for the leveldifferences between the regional mortgage rates.

5Data is available from Saiz’s website. http://real.wharton.upenn.edu/∼saiz/.


3.3.5. Summary Statistics

Table 2 reports the summary statistics of the variables that have been discussed so

far. There are three panels in the table: panel A covers quarterly national variables,

panel B covers annual national variables, and panel C covers annual MSA-level

variables. Each panel reports the correlation matrix among the variables.

In panel A, I summarize quarterly national variables. Variables include remod-

eling expenditures for Home Improvements from SORAR (C50), total remodeling

expenditures from JCHS (JCHS), home price appreciation (HPIAPP), the 30-year

mortgage rate (30yMORTINT), per capita income (INC), and the unemployment

rate (UNEMP).

Panel B summarizes annual national variables. Variables include C50A, C50MR,

C50HI, JCHS, the number of mortgage applications for home improvements (APPL),

the amounts of mortgage originations for home improvements (AMTO), the recoup

ratio of average remodeling projects (RECOUP), HPIAPP, 30yMORTINT, INC,

UNEMP, and the log number of population (logPOP).

Panel C summarizes annual MSA-level variables. Variables include the log num-

ber of home improvement projects from the AHS (logAHSn), the log amount of

home improvement projects from the AHS (logAHSm), the log number of mortgage

applications for home improvements (logAPPL), the log amount of mortgage origi-

nations for home improvements (logAMTO), RECOUP, HPIAPP, 30yMORTINT,

INC, UNEMP, logPOP, and housing supply elasticity (ELAST).

3.4. Time Series Analysis

In this section, I study the determinants of remodeling expenditures through time-

series analysis. I use the remodeling expenditures from SORAR, and Table 3 reports

the regression results.


First, remodeling expenditures are positively associated with home price appre-

ciation (HPIAPP). Column (1) reports the quarterly regression result with HPI-

APP, 30yMORTINT, INC, UNEMP and quarter dummy variables. The column

shows that HPIAPP is positively associated with remodeling expenditures. Col-

umn (2) reports the yearly regression result with the same control variables as in

column (1). Again, HPIAPP is positively correlated with remodeling expenditures.

Annual regression also shows that a one-standard deviation increase in HPIAPP

is associated with a 0.22-standard deviation increase in remodeling expenditures.

(A one-standard deviation increase in HPIAPP is associated with a 4.95 [1 SD] x

1.311 [slope] = 6.49 increase in remodeling expenditures, which is 6.49 / 30.02 =

0.22-standard deviation.)

Second, remodeling expenditures are positively correlated to the availability of

mortgage credit for home improvement projects. Columns (2)-(3) report the re-

gression results with the number of mortgage applications for home improvement

projects (APPL) and the amounts of the mortgage originations for home improve-

ment projects (AMTO). Control variables include HPIAPP, 30yMORTINT, INC,

and UNEMP. Note that APPL and AMTO show a positive association with re-

modeling expenditures. A $1 billion increase in mortgage originations results in

a $566 million increase in remodeling expenditures. Economic significance is also

large. For instance, a one-standard deviation increase in AMTO is associated with

a 0.55-standard deviation increase in remodeling expenditures. This indicates the

importance of mortgage credit availability on remodeling expenditures. When more

budgets are available due to increased mortgage credit, homeowners increase remod-

eling expenditures. Note that the increase in remodeling expenditures may come

from either consumption-motive remodeling or investment-motive remodeling.

Third, remodeling expenditures are not significantly associated with the re-

coup ratio of remodeling projects. Column (5) shows the regression result with


the recoup ratio of remodeling projects (RECOUP). If consumption-motive were

the only reason for remodeling, we would expect to observe a significant negative

relationship between RECOUP and remodeling expenditures. Insignificant results

can be interpreted as a mixture of two remodeling purposes: positive correlation

from investment-motive remodeling and negative correlation from consumption-

motive remodeling. This result rejects the null hypothesis on the dominant role of

consumption-motive remodeling.

Remodeling activity is positively correlated with per capita income (INC). How-

ever, remodeling expenditures did not show any significant relationship with the

30-year mortgage rate (30yMORTINT) or the unemployment rate (UNEMP).

3.5. Cross-sectional Analysis

In this section, I search for the determinants of remodeling expenditures through

cross-sectional analysis. I use the number of remodeling projects and the amounts of

remodeling expenditures from the AHS. Table 4 reports the panel regression results

with state and year fixed-effect. Columns (1)-(4) report the regression results on the

log numbers of remodeling projects. Columns (5)-(8) report the regression results

on the log amounts of remodeling expenditures.

The results from previous time-series analysis remain unchanged in cross-sectional

analysis. HPIAPP is positively associated with remodeling expenditures. AMTO

and APPL are also positively correlated with the number of remodeling projects

as well as remodeling expenditures. The magnitude is even larger: A 1% increase

in AMTO produces a 1% increase in remodeling expenditures. Per capita income

is positively associated with remodeling expenditures. The unemployment rate and

the 30-year mortgage rate are insignificant.


RECOUP also shows similar results to the time series analysis. RECOUP shows

an insignificant relationship between the number of remodeling projects to the

remodeling expenditures. As I explained, this finding supports the significant role

of investment-motive remodeling in the overall market. Otherwise, we would expect

to see a significant negative relationship between the recoup ratio and remodeling

expenditures.

Housing supply elasticity shows a negative association with the number of re-

modeling projects and remodeling expenditures. This indicates that more remod-

eling expenditures existed in areas with inelastic housing supply. The elasticity of

housing supply can be interpreted in two different ways. First, low elasticity means

that it is hard to build a new home in the area. Since new homes can’t be built

easily, it is normal to have high demand for remodeling in those areas. Second,

low elasticity also means that home prices will appreciate more when there is any

positive demand shock. For the remodeling expenditures as an investment, low elas-

ticity area gives a better chance to make a profit. Therefore, the significant results

of housing supply elasticity incorporate both of remodeling purposes.

3.6. Conclusion

In this paper, I test whether the consumption-motive explanation of home im-

provements outweighs the investment-motive explanation. Although the distinction

between consumption-motive and investment-motive remodeling is not straightfor-

ward, there is a fundamental difference between consumption-motive remodeling

and investment-motive remodeling. The correlation between remodeling expendi-

tures and the recoup ratio of remodeling projects will be different by remodeling

purposes. That is, consumption-motive remodeling will be negatively correlated


with recoup ratio. In contrast, recoup ratio is a primary concern for investment,

and investment-motive remodeling will be positively related to recoup ratio.

Using these opposite correlations between the recoup ratio of remodeling projects

and remodeling expenditures, I test the null hypothesis that consumption-motive

remodeling comprises the majority of remodeling activities. Alternative hypothe-

sis will be the importance of investment-motive remodeling in overall remodeling

market.

Time series analysis shows following results. First, remodeling expenditures are

positively associated with home price appreciation (HPIAPP). Second, remodeling

expenditures are positively correlated to the mortgage credit available for home

improvement projects. This indicates the importance of mortgage credit availability

on remodeling expenditures. Note that the increase in remodeling expenditures may

come from either consumption-motive remodeling or investment-motive remodeling.

Most significant is the third result: remodeling expenditures are not signifi-

cantly associated with the recoup ratio of remodeling projects (RECOUP). If the

consumption-motive were the only reason for remodeling, we would expect to ob-

serve a significant negative relationship between RECOUP and remodeling expen-

ditures. Insignificant results can be interpreted as a mixture of two remodeling pur-

poses: positive correlation from investment-motive remodeling and negative corre-

lation from consumption-motive remodeling. This result rejects the null hypothesis

on the dominant role of consumption-motive remodeling.

Cross-sectional analysis shows similar results. HPIAPP is positively associated

with remodeling expenditures. AMTO and APPL are also positively correlated with

the number of remodeling projects as well as remodeling expenditures. RECOUP

shows an insignificant relationship with the number of remodeling projects to re-

modeling expenditures. This finding supports the important role of investment-

motive remodeling. Otherwise, we would expect to see a significant negative re-


lationship between recoup ratio and remodeling expenditures. In addition, hous-

ing supply elasticity shows a negative association with the number of remodeling

projects and remodeling expenditures.


Table 3.1: Expenditures on Remodeling

Table 1 reports national level expenditures on remodeling from various sources. Panel A reports the amount ofexpenditures on remodeling. The first column shows remodeling expenditures from the Joint Center for HousingStudies (JCHS) of Harvard University. The second column shows the remodeling expenditures from the AmericanHousing Survey (AHS). The biennial AHS data is available for 1999, 2001, 2003, 2005, 2007, and 2009. The lastcolumn shows the remodeling expenditures from the Survey of Residential Alterations and Repairs (SORAR) bythe Bureau of Labor Statistics. SORAR data was discontinued after 2007. SORAR provides the expenditures forMaintenance and Repairs and the expenditures for Home Improvements separately. The remodeling expendituresfor Home Improvements are reported. All the numbers are in billion dollars. Panel B reports the correlationmatrix. For JCHS and SORAR, quarterly expenditures data is used for computing correlation.

Panel A: Annual Expenditures ($ billions)

Year JCHS AHS SORAR (C50)

1998 70.9 91.71999 71.6 82.1 100.52000 80.7 110.72001 82.2 93.0 110.32002 94.0 125.92003 97.1 97.8 132.82004 110.2 147.92005 123.8 150.2 161.72006 141.9 174.82007 143.8 161.0 171.62008 124.72009 114.5 131.5

Panel B: Correlations

JCHS AHS SORAR (C50)

JCHS 1AHS 0.9533 1

SORAR 0.9493 0.9684 1

CHAPTER

3.THEMOTIV

EOFHOMEIM

PROVEMENTS

108


Table 2 reports the summary statistics of the variables. Panel A summarizes quarterly national variables. Variables include the remodeling expenditures for HomeImprovements from SORAR (C50), total remodeling expenditures from JCHS (JCHS), home price appreciation (HPIAPP), the 30-year mortgage rate (30yMORTINT), percapita income (INC), and the unemployment rate (UNEMP). Panel B summarizes annual national variables. Variables include C50, JCHS, the number of mortgageapplications for home improvements (APPL), the amounts of mortgage originations for home improvements (AMTO), the recoup ratio of average remodeling projects(RECOUP) from the Realtor Magazine, HPIAPP, 30yMORTINT, INC, UNEMP, and the log number of population (logPOP). Panel C summarizes annual MSA-levelvariables. Variables include the log number of home improvement projects from the AHS (logAHSn), the log amount of home improvement projects from AHS (logAHSm),the log number of mortgage applications for home improvements (logAPPL), the log amounts of mortgage originations for home improvements (logAMTO), RECOUP,HPIAPP, 30yMORTINT, INC, UNEMP, logPOP, and housing supply elasticity (ELAST).

Panel A: Quarterly National Data

A.1: Sumary Statistics

Obs Mean Std. Dev. Min Max Sample Period Unit Source

C50 168 60.91 49.06 5.40 180.30 1966-2007 $bil SORARJCHS 68 98.19 26.98 56.70 146.20 1995-2011 $bil LIRAHPIAPP 147 1.15 1.27 -2.90 4.65 1975-2011 % FHFA30yMORTINT 163 8.79 2.90 4.01 17.73 1971-2011 % FHFAINC 208 4691 3958 406 13159 1966-2011 $ BEAUNEMP 208 6.05 1.61 3.40 10.70 1966-2011 % BLS

A.2: Correlation Matrix

C50 JCHS HPIAPP 30yMORTINT INC UNEMP

C50 1JCHS 0.9597 1HPIAPP -0.1232 -0.2583 130yMORTINT -0.5863 -0.6613 0.178 1INC 0.9853 0.8888 -0.4481 -0.676 1UNEMP -0.2356 0.2575 -0.2566 0.2948 0.1239 1

CHAPTER

3.THEMOTIV

EOFHOMEIM

PROVEMENTS

109

Table 2: Summary Statistics (Con’t)

Panel B: Annual National Data

B.1: Sumary Statistics

Obs Mean Std. Dev. Min Max Sample Period Unit Source

C50 10 132.81 30.02 91.71 174.82 1998-2007 $bil SORARJCHS 12 104.60 25.76 70.85 143.83 1998-2009 $bil LIRAAMTO 12 41.86 29.27 17.30 92.57 1998-2009 $bil HMDAAPPL 12 1.39 0.37 0.57 1.84 1998-2009 $mil HMDARECOUP 12 77.69 6.85 68.18 88.28 1998-2009 % RMHPIAPP 12 4.71 4.95 -4.72 11.26 1998-2009 % FHFA30yMORTINT 12 6.44 0.82 5.04 8.06 1998-2009 % FHFAUNEMP 12 5.25 1.42 3.89 9.19 1998-2009 % BLSlogPOP 12 19.32 0.04 19.26 19.38 1998-2009 log(#) BEAINC 12 35749 4733 28754 42649 1998-2009 $ BEA

B.2: Correlation Matrix

C50 JCHS AMTO APPL RECOUP HPIAPP 30yMORTINT UNEMP logPOP INC

C50 1JCHS 0.9877 1

AMTO 0.9099 0.8108 1APPL 0.4482 0.0758 0.5346 1RECOUP 0.1132 -0.1620 0.2050 0.4107 1

HPIAPP 0.1415 -0.2227 0.3011 0.7356 0.8328 130yMORTINT -0.6605 -0.5947 -0.3892 0.4802 0.0377 0.3225 1UNEMP 0.3029 0.246 -0.0769 -0.8431 -0.2655 -0.572 -0.7987 1logPOP 0.9819 0.8823 0.5378 -0.3734 -0.3298 -0.5194 -0.7732 0.6221 1INC 0.9721 0.9106 0.5585 -0.2815 -0.4025 -0.5487 -0.6602 0.4916 0.9806 1

CHAPTER

3.THEMOTIV

EOFHOMEIM

PROVEMENTS

110

Table 2: Summary Statistics (Con’t)

Panel C: Annual MSA-level Data

C.1: Sumary Statistics

Obs Mean Std. Dev. Min Max Unit Sample Period Source

logAHSn 612 12.18 1.20 8.71 15.12 log(#) 99, 01, 03, 05, 07, 09 AHSlogAHSm 612 19.80 1.51 13.20 23.84 log($) 99, 01, 03, 05, 07, 09 AHSlogAPPL 4356 7.36 1.18 4.13 11.57 log(#) 1998-2009 HMDAlogAMTO 4356 17.20 1.40 13.86 23.20 log($) 1998-2009 HMDARECOUP 642 77 18 21 157 % 1998-2005,2007-2009 RM

HPIAPP 4353 4.43 6.29 -37.44 34.33 % 1998-2009 FHFA30yMORTINT 4356 6.49 0.77 4.89 8.46 % 1998-2009 FHFAELAST 280 2.52 1.42 0.63 12.15 Saiz(2010)UNEMP 4356 5.37 2.36 1.23 30.06 % 1998-2009 BLSINC 4356 30684 6985 12723 80899 $ 1998-2009 BEAlogPOP 4356 12.63 1.06 10.75 16.76 log(#) 1998-2009 BEA

C.2: Correlation Matrix

logAHSn logAHSm logAPPL logAMTO RECOUP HPIAPP 30yMORTINT ELAST UNEMP INC logPOP

logAHSn 1logAHSm 0.9183 1

logAPPL 0.8125 0.7292 1logAMTO 0.7459 0.7676 0.8794 1RECOUP 0.0545 0.1320 0.1590 0.2419 1

HPIAPP 0.0274 0.0108 0.1959 0.1808 0.3442 130yMORTINT 0.0008 -0.0974 0.0999 -0.1834 0.0068 0.0882 1ELAST -0.3143 -0.3571 -0.4018 -0.4915 0.1622 -0.1067 0.0405 1UNEMP -0.1031 -0.0593 -0.1741 -0.0850 0.1299 -0.3012 -0.4461 -0.0436 1INC 0.3947 0.5151 0.3352 0.6033 0.0242 -0.0820 -0.3844 -0.2869 -0.0459 1logPOP 0.9020 0.8494 0.9105 0.8501 0.1367 0.0333 -0.0409 -0.4363 -0.0312 0.4095 1


Table 3.3: Determinants of Remodeling Activities (Time Series)

Table 3 reports the regression results of remodeling expenditures on home price appreciation (HPIAPP), thenumber of mortgage applications (APPL), the amounts of mortgage originations (AMTO), average recoup ratiofor remodeling projects (RECOUP), the 30-year mortgage rate (30yMORTINT), per capita income (INC), andthe unemployment rate (UNEMP). Dependent variables are the remodeling expenditures for Home Improvementsfrom the Survey of Residential Alterations and Repairs (SORAR). Column (1) reports the result with HPIAPP,30yMORTINT, and other control variables at quarterly level. Quarter fixed-effect is also included. Column (2)reports the results as in column (1) at yearly level. Column (3) reports the results with APPL, column (4)reports the results with AMTO, and column (5) reports the results with RECOUP. The sample period is from1975 to 2007. The table reports point estimates with heteroskedasticity robust t-statistics in parentheses. ***, **,* denotes 1%, 5%, and 10% statistical significance.

Remodeling Expenditures (SORAR)Variables (1) (2) (3) (4) (5)

APPL 52.18*(2.322)

Econ Sig 0.64

AMTO 0.566*(2.696)

Econ Sig 0.55

RECOUP -0.154(-0.326)

Econ Sig -0.04

HPIAPP 2.463*** 1.311* -0.324 -1.094 1.523(3.104) (2.280) (-0.410) (-1.190) (1.404)

Econ Sig 0.06 0.22 -0.05 -0.18 0.25

30yMORTINT 0.0947 -5.342 6.521 8.914 -5.296(0.321) (-1.718) (1.331) (1.383) (-1.433)

INC 0.0152*** 0.00663*** 0.00579*** 0.00377** 0.00662***(32.99) (16.40) (11.04) (3.685) (15.35)

UNEMP 0.494 -0.593 21.76* 14.77* -0.0563(0.665) (-0.226) (2.426) (2.360) (-0.0201)

Constant -16.43** -66.33 -291.5** -144.7** -57.83(-2.100) (-1.699) (-3.081) (-2.822) (-1.184)

Observations 131 10 10 10 10R-squared 0.961 0.977 0.992 0.994 0.978Frequency Quarter Year Year Year Year


Table 3.4: Determinants of Remodeling Activities (Cross-section)

Table 4 reports the panel regression results of remodeling activities on home price appreciation (HPIAPP), thelog number of mortgage applications (logAPPL), the log amounts of mortgage originations (logAMTO), averagerecoup ratio for remodeling projects (RECOUP), the 30-year mortgage rate (30yMORTINT), housing supplyelasticity (ELAST), per capita income (INC), and the unemployment rate (UNEMP). Dependent variables arethe number of remodeling projects and the remodeling expenditures from the American Housing Survey (AHS).Columns (1)-(4) show the results for the log number of home improvement projects. Columns (5)-(8) show theresults for the log amounts of home improvements projects. The sample period is from 1998 to 2009. State andyear fixed-effect are included. The table reports point estimates with robust t-statistics in parentheses. ***, **, *denotes 1%, 5%, and 10% statistical significance. All standard errors are clustered by MSA.

The log Numbers of Remodeling Projects The log Amounts of Remodeling Projects

Variables (1) (2) (3) (4) (5) (6) (7) (8)

logAPPL 1.089*** 1.199***(18.30) (16.15)

Econ Sig 1.07 0.94

logAMTO 0.902*** 0.996***(14.08) (13.09)

Econ Sig 1.05 0.92

RECOUP 0.00326 0.00521(0.947) (1.444)

Econ Sig 0.05 0.06

HPIAPP 0.0111 -0.0167*** -0.0131*** 0.0148** 0.0173* -0.0134* -0.00951 -0.00231(1.559) (-4.436) (-2.913) (2.092) (1.897) (-1.826) (-1.285) (-0.246)

Econ Sig 0.06 -0.09 -0.07 0.08 0.07 -0.06 -0.04 -0.01

30yMORTINT 0.0443 0.00337 -0.462* -0.117 0.322 0.277 -0.237 0.541(0.243) (0.0176) (-1.758) (-0.320) (1.144) (0.999) (-0.722) (0.973)

ELAST -0.768*** -0.168** -0.233** -0.670*** -0.852*** -0.191** -0.261** -0.762***(-4.784) (-2.403) (-2.571) (-3.157) (-4.670) (-2.280) (-2.492) (-4.070)

UNEMP 0.0556 0.0287 0.0612* 0.131** 0.0409 0.0113 0.0470 -0.0302(0.725) (1.208) (1.823) (2.380) (0.483) (0.392) (1.276) (-0.384)

INC 8.02e-05*** 1.03e-05 -1.22e-05 5.01e-05** 0.000114*** 3.69e-05*** 1.18e-05 6.90e-05**(3.475) (0.920) (-1.019) (2.158) (4.260) (2.755) (0.824) (2.544)

Constant 11.11*** 3.044** 0.503 12.57*** 15.70*** 6.821*** 3.988 15.50***(5.950) (2.043) (0.261) (4.167) (6.049) (2.909) (1.443) (3.607)

Observations 594 594 594 190 594 594 594 190R-squared 0.493 0.854 0.779 0.765 0.506 0.782 0.726 0.744State FE Yes Yes Yes Yes Yes Yes Yes YesYear FE Yes Yes Yes Yes Yes Yes Yes Yes


Figure 3.1: Mortgage Credit Availability and Remodeling Expenditures

Figure 1 plots mortgage credit availability with remodeling expenditures. Remodeling expenditures are from theJoint Center for Housing Studies (JCHS) of Harvard University and from the American Housing Survey (AHS).Mortgage credit for home improvements are from the Home Mortgage Disclosure Act (HMDA): the number ofmortgage applications (APPL) and the amounts of mortgage originations (AMTO). The first column plots theremodeling expenditures from JCHS with APPL and AMTO. The second column plots the remodelingexpenditures from AHS with APPL and AMTO.

60

80

10

01

20

14

0R

em

od

elin

g E

xp

en

ditu

re (

$ b

ils),

JC

HS

.51

1.5

2M

ort

ga

ge

Ap

plic

atio

n f

or

Ho

me

Im

pro

ve

me

nt

(mils

)

1998 2000 2002 2004 2006 2008year

Mortgage Application & Remodeling Expenditures (JCHS)

60

80

10

01

20

14

0R

em

od

elin

g E

xp

en

ditu

re (

$ b

ils),

JC

HS

20

40

60

80

10

0M

ort

ga

ge

Orig

ina

tio

n f

or

Ho

me

Im

pro

ve

me

nt

($ b

ils)

1998 2000 2002 2004 2006 2008year

Mortgage Origination & Remodeling Expenditures (JCHS)

80

10

01

20

14

01

60

Re

mo

de

ling

Exp

en

ditu

re (

$ b

ils),

AH

S

.51

1.5

2M

ort

ga

ge

Ap

plic

atio

n f

or

Ho

me

Im

pro

ve

me

nt

(mils

)

1998 2000 2002 2004 2006 2008year

Mortgage Application & Remodeling Expenditures (AHS)

80

10

01

20

14

01

60

Re

mo

de

ling

Exp

en

ditu

re (

$ b

ils),

AH

S

20

40

60

80

10

0M

ort

ga

ge

Orig

ina

tio

n f

or

Ho

me

Im

pro

ve

me

nt

($ b

ils)

1998 2000 2002 2004 2006 2008year

Mortgage Origination & Remodeling Expenditures (AHS)

Bibliography

Avery, Robert B, Kenneth P Brevoort, and Glenn B Canner. 2007a. “Opportunities

and Issues in using HMDA Data.” Journal of Real Estate Research 29 (4):351–

379.

———. 2007b. “The 2006 HMDA Data.” Federal Reserve Bulletin :A73–A109.

Barber, Brad M. and Terrance Odean. 2001. “Boys Will Be Boys: Gender, Overcon-

fidence, And Common Stock Investment.” The Quarterly Journal of Economics

116 (1):261–292.

———. 2011. “Behavior of Individual Investors.” Handbook of the Economics of

Finance, eds Harris and Constantinides .

Bostic, Raphael W., Kathleen C. Engel, Patricia A. McCoy, Anthony Pennington-

Cross, and Susan M. Wachter. 2008. “State and local anti-predatory lending

laws: The effect of legal enforcement mechanisms.” Journal of Economics and

Business 60 (1-2):47–66.

Burnett, Kimberly, Meryl Finkel, and Bulbul Kaul. 2004. “Mortgage Lending in

North Carolina After the Anti-Predatory Lending Law: Final Report.” Abt As-

sociates Inc. Report .

Calomiris, Charles W., Stanley D. Longhofer, and William Miles. 2009. “The

(Mythical?) Housing Wealth Effect.” NBER Working Paper No. 15075.

114

BIBLIOGRAPHY 115

Campbell, John Y. 2006. “Household finance.” Journal of Finance 61 (4):1553–

1604.

Carroll, Christopher D., Misuzu Otsuka, and Jiri Slacalek. 2011. “How Large Are

Housing and Financial Wealth Effects? A New Approach.” Journal of Money,

Credit, and Banking 43 (1):55–79.

Case, Karl E., John M. Quigley, and Robert J. Shiller. 2005. “Comparing Wealth

Effects: The Stock Market versus the Housing Market.” Advances in Macroeco-

nomics 5 (1):1–34.

Choi, Hyun-Soo. 2012. “The Impact of Anti-Predatory Lending Laws on Mortgage

Volume.” Working Paper .

Choi, Hyun-Soo, Harrison Hong, and Jose Scheinkman. 2012. “Home Improve-

ments.” Working Paper .

Cooper, Daniel. 2009. “Impending U.S. Spending Bust? The Role of Housing

Wealth as Borrowing Collateral.” Federal Reserve Bank of Boston Public Policy

Discussion Paper Series No. 09-7.

Dell’Ariccia, Giovanni, Deniz Igan, and Luc Laevan. 2009. “Credit Booms and

Lending Standards: Evidence from the Subprime Mortgage Market.” IMF Work-

ing Paper No. 08/106.

Demyanyk, Yuliya and Otto Van Hemert. 2009. “Understanding the Subprime

Mortgage Crisis.” Review of Financial Studies 24 (6):1848–1880.

Ding, Lei, Roberto G. Quercia, and Alan M. White. 2009. “State Anti-Predatory

Lending Laws: Impact and Federal Preemption Phase I Desciptive Analysis.”

Working Paper, Center for Community Capital, The University of North Car-

olina at Chapel Hill .

BIBLIOGRAPHY 116

Elliehausen, Gregory and Michael E Staten. 2004. “Regulation of Subprime Mort-

gage Products: An Analysis of North Carolina’s Predatory Lending Law.” Jour-

nal of Real Estate Finance and Economics 29 (4):411–433.

Elliehausen, Gregory, Michael E Staten, and Jevgenijs Steinbuks. 2006. “The Ef-

fects of State Predatory Lending Laws on the Availability of Subprime Mortgage

Credit.” Credit Research Center Monograph #38 .

Engel, Kathleen C and Patricia A McCoy. 2002. “A Tale of Three Markets: The

Law and Economics of Predatory Lending.” Texas Law Review 80 (6):1255–1367.

Farhi, Emmanuel and Stavros Panageas. 2004. “The real effects of stock market

mispricing at the aggregate: Theory and Empirical Evidence.” mimeo. M.I.T.

and the Wharton School .

Gerardi, Kristopher S, Adam Hale Shapiro, and Paul S Willen. 2007. “Subprime

Outcomes: Risky Mortgages, Homeownership Experiences, and Foreclosures.”

Federal Reserve Bank of Boston Working Papers Series No. 07-15.

Gilchrist, Simon, Charles P. Himmelberg, and Gur Huberman. 2005. “Do stock

price bubbles influence corporate investment?” Journal of Monetary Economics

52 (4):805–827.

Glaeser, Edward L, Joseph Gyourko, and Albert Saiz. 2008. “Housing Supply and

Housing Bubbles.” Journal of Urban Economics 64 (2):198–217.

Greenspan, Alan and James Kennedy. 2007. “Sources and Uses of Equity Extracted

from Homes.” Finance and Economics Discussion Series, Divisions of Research

& Statistics and Monetary Affairs :1–49.

Gyourko, Joseph and Albert Saiz. 2004. “Reinvestment in the housing stock: the

role of construction costs and the supply side.” Journal of Urban Economics .

BIBLIOGRAPHY 117

Harvey, Keith D and Peter J Nigro. 2004. “Do Predatory Lending Laws Influence

Mortgage Lending? An Analysis of the North Carolina Predatory Lending Law.”

Journal of Real Estate Finance and Economics 29 (4):435–456.

Helms, Andrew C. 2003. “Understanding gentrification: an empirical analysis of

the determinants of urban housing renovation.” Journal of Urban Economics .

Ho, Giang and Anthony Pennington-Cross. 2006. “The Impact of Local Predatory

Lending Laws on the Flow of Subprime Credit.” Journal of Urban Economics

60 (2):210–228.

Hurst, Erik and Frank Stafford. 2004. “Home Is Where the Equity Is: Mortgage Re-

financing and Household Consumption.” Journal of Money, Credit, and Banking

36 (6):985–1014.

Jayaratne, Jith and Philip E Strahan. 1996. “The Finance-Growth Nexus: Evi-

dence From Bank Branch Deregulation.” The Quarterly Journal of Economics

111 (3):639–670.

Joint Center for Housing Studies. 2009. “The Remodeling Market in Transition:

Improving America’s Housing 2009.” Available on the center’s website.

Keys, Benjamin J., Tanmoy Mukherjee, Amit Seru, and Vikrant Vig. 2010. “Did

Securitization Lead to Lax Screening? Evidence from Subprime Loans.” The

Quarterly Journal of Economics 125 (1):307–362.

Khandani, Amir E., Andrew W. Lo, and Robert C. Merton. 2010. “Systematic Risk

and the Refinancing Rachet Effect.” HBS Working Paper No. 10-023.

Li, Wei and Keith S. Ernst. 2007. “Do state predatory lending laws work? A panel

analysis of market reforms.” Housing Policy Debate 18 (2):347–391.

BIBLIOGRAPHY 118

Mansfield, Cathy Lesser. 2000. “The Road to Subprime ‘HEL’ was Paved with Good

Congressional Intentions: Usury Deregulation and the Subprime Home Equity

Market.” South Carolina Law Review 51:473–587.

Mayer, Chris and Karen Pence. 2008. “Subprime Mortgages: What, Where, and to

Whom?” NBER Working Paper No. 14083.

McQueen, M.P. Apr 24, 2010. “The New Rules of Remodeling.” The Wall Street

Journal .

Melzer, Brian T. 2010. “Mortgage Debt Overhang: Reduced Investment by Home-

owners with Negative Equity.” Working Paper .

Mian, Atif and Amir Sufi. 2009. “The Consequences of Mortgage Credit Expansion:

Evidence from the U.S. Mortgage Default Crisis.” The Quarterly Journal of

Economics 124 (4):1449–1496.

———. 2011. “Housing Prices, Home Equity-Based Borrowings, and the U.S.

Household Leverage Crisis.” American Economic Review forthcoming.

Montgomery, Claire. 1992. “Explaining home improvement in the context of house-

hold investment in residential housing.” Journal of Urban Economics 32 (3):326–

350.

Moskowitz, Tobias J. and Annette Vissing-Jorgensen. 2002. “The returns to en-

trepreneurial investment: A private equity premium puzzle?” The American

Economic Review 92 (4):745–778.

Muellbauer, John N. 2008. “Housing, Credit and Consumer Expenditure.” CEPR

Discussion Paper No. 6782.

Odean, Terrance. 1999. “Do investors trade too much?” American Economic Review

89 (5):1279–1298.

BIBLIOGRAPHY 119

Peek, Joe. 2010. “Household Reaction to the Financial Crisis: Scared or Scarred?”

Special Report, Research Institute for Housing America :1–46.

Pence, Karen. 2006. “Foreclosing on Opportunity: State Laws and Mortgage

Credit.” The Review of Economics and Statistics :1–6.

Polk, Christopher and Paola Sapienza. 2008. “The stock market and corporate

investment: A test of catering theory.” Review of Financial Studies 22 (1):187–

217.

Quercia, Roberto G., Michael A. Stegman, and Walter R. Davis. 2004. “Assessing

the impact of North Carolina’s predatory lending law.” Housing Policy Debate

15 (3):1–30.

Rugh, Jacob S. and Douglas S. Massey. 2010. “Racial Segregation and the American

Foreclosure Crisis.” American Sociological Review 75 (5):629–651.

Saiz, Albert. 2010. “The Geographic Determinants of Housing Supply.” The Quar-

terly Journal of Economics 125 (3):1253–1296.

U.S. Department of Housing and Urban Development. 2000. “Curbing Predatory

Home Mortgage Lending.” HUD-Treasury Report .

U.S. General Accounting Office. 2004. “Consumer Protection: Federal and State

Agencies Face Challenges in Combating Predatory Lending.” Report to the Spe-

cial Committee on Aging, U.S. Senate .

Documents

THREE ESSAYS ON REAL ESTATE FINANCE...eld{ real estate nance. He was not only an academic advisor, but was also a great mentor with whom I could discuss many deep concerns. He always