ESSAYS IN REAL ESTATE by STEVEN STELK LEONARD - Acumen

ESSAYS IN REAL ESTATE by STEVEN STELK LEONARD - Acumen

ESSAYS IN REAL ESTATE by STEVEN STELK LEONARD V. ZUMPANO, COMMITTEE CHAIR ARTHUR W. ALLAWAY THOMAS W. DOWNS HAROLD W. ELDER KIMBERLY R. GOODWIN A DI...

376KB Sizes 0 Downloads 0 Views

Recommend Documents

THREE ESSAYS ON RISK AND INSURANCE by BOYI - Acumen
A special thanks to Dr. James Peery Cover, the chair of my committee, my advisor and my mentor. He spent countless hours

REAL ESTATE IN FINLAND
The Finnish real estate market has undergone a very strong development in the ...... buildings and flats and leasehold r

Careers in Real Estate
Aug 29, 2017 - Annually, Certified Residential Specialists earn more than triple the income of the average ... The Inter

three essays on residential real estate brokerage - UKnowledge
Three Essays on Residential Real Estate Brokerage. The first essay investigates how individual characteristics influence

three essays on residential real estate brokerage - UKnowledge
The second essay defines market concentrations of residential real estate broker- ... consent of the Dean of the Graduat

Real Estate Distressed Real Estate & Real Estate Debt - Pillsbury
assets of Donald J. Trump including an $800 million personal guarantee. .... single family residences located in Califor

Survey finds Strength in Commercial Real Estate - Upland Real Estate
Apr 27, 2006 - Survey. Several Hubbell brokers presented the results by category, and ... and warehouse activity but wea

Industrial real estate remains in demand - Barclay Street Real Estate
May 28, 2015 - 28 May 2015 Calgary Herald DAVID PARKER David Parker ... Lakhoo, president of Coastal Resources, at the A

Killington Real Estate in Review - Prestige Real Estate of Killington
Apr 1, 2009 - Nyberg reinforced his enthusiasm and optimism for the future by sharing with us the reason Powdr Corp. cho

trading in real estate as a real estate broker - Real Estate Council of
Does a person need a licence to trade in real estate? A person who trades in real estate on behalf of another and for co

ESSAYS IN REAL ESTATE

by STEVEN STELK LEONARD V. ZUMPANO, COMMITTEE CHAIR ARTHUR W. ALLAWAY THOMAS W. DOWNS HAROLD W. ELDER KIMBERLY R. GOODWIN

A DISSERTATION

Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the Department of Economics, Finance and Legal Studies in the Graduate School of The University of Alabama

TUSCALOOSA, ALABAMA

2013

Copyright Steven John Stelk 2013 ALL RIGHTS RESERVED

ABSTRACT This study exploits the recent financial crises as a unique natural experiment to examine relationships in residential real estate brokerage and real estate investment through three essays. The first essay examines the impact of agency disclosure on residential restate transactions in the post-financial crises period and extends the literature with three key findings. First, the overall proportion of buyers that report receipt of agency disclosure has not improved since previous studies were completed. Second, there is no evidence that buyers who do not report receipt of agency disclosure pay different prices for homes than buyers who do report receiving agency disclosure. Finally, there is evidence that the timing of agency disclosure matters. Among buyers that do receive agency disclosure, those receiving disclosure at a time other than the first contact with a broker are associated with 3.2% higher home prices. The results demonstrate the need for continued improvement in mandatory disclosure statutes. The second essay investigates the real estate brokerage market’s impact on home prices in both a seller’s market (2006) and a buyer’s market (2009). In both years, homes sold with brokerage assistance realized higher prices when compared to homes sold without the aid of a broker, even after controlling for selection bias in the seller’s choice to use a broker. This is the first study using a national dataset that finds evidence of price segmentation in the residential real estate market. The findings may be the result of the extreme market conditions housing market participants faced in 2006 and 2009.

ii

The third essay examines the impact of REITs on the Value-at-Risk (VaR) of a mixed asset portfolio surrounding the financial crises using a new, more accurate method of estimating VaR, conditional autoregressive value at risk (CAViaR). The more accurate VaR estimates show that adding REITs to the portfolio has no significant impact on VaR until after the financial crises begins in 2006. After 2006, adding REITs to a portfolio of stocks and bonds dramatically increases VaR. The results have significant implications for portfolio selection.

iii

DEDICATION This dissertation is dedicated to everyone who helped me through the trials and tribulations of creating this manuscript. In particular, it is dedicated my wife, children, parents, and close friends who stood by me throughout the completion of this masterpiece.

iv

LIST OF ABBREVIATIONS AND SYMBOLS

2SLS

Two-stage least squares

Adj.

Adjusted

Avg.

Average

β

Coefficient

CAViaR

Conditional autoregressive value at risk

Col.

Column

e.g.

Exempli gratia or for example

et al.

Et alii or and others

etc.

Et cetera or and others

f(.)

Function indicator

F

F-test

LAMBDA

Invers Mills ratio

max(yt-1,0)

Use the highest value of the set (yt-1, 0)

min(yt-1,0)

Use the smallest value of the set (yt-1, 0)

ln.

Natural log

OLS

Ordinary least squares

P(.)

Probability of

Prob

Probability

v

REIT

Real estate investment trust

Std. Dev.

Standard deviation

TOM

Time on market

VaR

Value-at-Risk

Z

Z-score from wilcoxon signed-rank test

*

Significant at the ten percent level

**

Significant at the five percent level

***

Significant at the one percent level

|

Given

|.|

Absolute value

+

Addition or positive sign

-

Subtraction or negative sign

%

Percentage

>

Greater than

=

Equal to



Summation

vi

ACKNOWLEDGMENTS

I would like to thank the many colleagues, friends, and faculty members who have helped me with this research project. I am most indebted to Leonard Zumpano, the chairman of this dissertation, for sharing his research expertise and wisdom regarding residential real estate brokerage. I would also like to thank all of my committee members, Arthur Allaway, Thomas Downs, Harold Elder, and Kimberly Goodwin for their invaluable input and support of both the dissertation and my academic progress. This research would not have been possible without the support of my friends, my fellow graduate students, and of course of my family who never stopped encouraging me.

vii

CONTENTS ABSTRACT ................................................................................................ ii DEDICATION ........................................................................................... iv LIST OF ABBREVIATIONS AND SYMBOLS ........................................v ACKNOWLEDGMENTS ........................................................................ vii LIST OF TABLES .......................................................................................x CHAPTER 1: INTRODUCTION ................................................................1 Investigating the Impact of Agency Disclosure on Residential Real Estate Transactions .................................................................................1 Can Real Estate Brokers Affect Home Prices Under Extreme Market Conditions? .............................................................................................2 REITs in a Mixed Asset Portfolio: An Investigation of Extreme Risks ...2 CHAPTER 2: INVESTIGATING THE IMPACT OF AGENCY DISCLOSURE ON RESIDENTIAL REAL ESTATE TRANSACTIONS .....................................................................................4 Introduction and Literature Review ..........................................................4 Data and Methodology..............................................................................9 Empirical Results ....................................................................................18 Conclusion ..............................................................................................32 References ..............................................................................................33 CHAPTER 3: CAN REAL ESTATE BROKERS AFFECT HOME PRICES UNDER EXTREME MARKET CONDITIONS? ..................35

viii

Introduction and Literature Review ........................................................35 Data and Methodology............................................................................40 Empirical Results ....................................................................................48 Conclusion ..............................................................................................64 References ..............................................................................................65 CHAPTER 4: REITS IN A MIXED ASSET PORTFOLIO: AN INVESTIGATION OF EXTREME RISKS ...........................................68 Introduction and Literature Review ........................................................68 Data and Methodology............................................................................72 Empirical Results ....................................................................................74 Conclusion ..............................................................................................81 References ..............................................................................................81

ix

LIST OF TABLES 2.1 Description of Variables ......................................................................12 2.2 Timing of Agency Disclosure ..............................................................19 2.3 Average Price by Agency Disclosure ..................................................20 2.4 First-Stage Probit Estimates on Reporting Disclosure.........................21 2.5 Second-Stage OLS Price Regressions with Selection Correction .......23 2.6 OLS Price Regressions without Selection Correction .........................28 2.7 OLS Price Regressions ........................................................................30 3.1 Description of Variables ......................................................................42 3.2 Statistics on Real Estate Broker Use ....................................................50 3.3 Summary Statistics on Time-on-Market (TOM) and Final Selling Price (All Transactions) .......................................................................51 3.4 Summary Statistics on Sales Price .......................................................51 3.5 First-Stage Probit Estimates on Choice of Broker (2006) ...................53 3.6 First-Stage Probit Estimates on Choice of Broker (2009) ...................54 3.7 Second-Stage OLS Price Regressions with Selection Correction (2006) ...................................................................................................56 3.8 Second-Stage OLS Price Regressions with Selection Correction (2009) ...................................................................................................57 3.9 2SLS Price Regressions (2006) ...........................................................62 3.10 2SLS Price Regressions (2009) .........................................................63

x

4.1 Summary Statistics...............................................................................75 4.2 Wilcoxon Signed-Rank Test ................................................................77 4.3 Comparison of VaR Means ..................................................................79

xi

`

CHAPTER 1 INTRODUCTION

This study exploits the recent financial crises as a natural experiment to examine relationships in residential real estate brokerage and real estate investment through three essays.

Investigating the Impact of Agency Disclosure on Residential Real Estate Transactions Residential real estate consumers have historically been confused about the nature of their relationship with brokers. Agency disclosure laws were enacted in all fifty states to better protect home buyers and sellers, but past studies found that real estate consumers’ reported receipt of agency disclosure did not improve following the passage of mandatory disclosure statutes. While calls for improved agency disclosure are justified, the existing literature has yet to provide empirical evidence that consumers are harmed when they do not report receipt of agency disclosure from a broker. Using data from the National Association of REALTORS® 2012 Home Buyer and Seller Survey, this study extends the literature with three key findings. First, the overall proportion of buyers that report receipt of agency disclosure has not improved since previous studies were completed, although the types of buyers that are more likely to report disclosure have changed. Second, there is no evidence that buyers who do not report receipt of agency disclosure pay different prices for homes than buyers who do report receiving agency disclosure. Finally, there is evidence that the timing of agency disclosure matters. Among buyers that do receive agency disclosure, those receiving disclosure at a time other than the first 1

contact with a broker are associated with 3.2% higher home prices. Taken together, the results demonstrate the need for continued improvement in mandatory disclosure statutes.

Can Real Estate Brokers Affect Home Prices Under Extreme Market Conditions? This study investigates the real estate brokerage market’s impact on home prices in both a seller’s market (2006) and a buyer’s market (2009). The results indicate that real estate brokers were able to influence housing prices in both years. In 2006 and 2009, homes sold with brokerage assistance realized higher prices when compared to homes sold without the aid of a broker, even after controlling for selection bias in the seller’s choice to use a broker. This is the first study using a national dataset that finds evidence of price segmentation in the residential real estate market. The findings may be the result of the extreme market conditions housing market participants faced in 2006 and 2009. Further investigation is needed to see whether the price differential persists after the housing market returns to more normal conditions.

REITs in a Mixed Asset Portfolio: An Investigation of Extreme Risks Until the recent financial crises, it was widely believed that adding REITs to a mixedasset portfolio expanded the efficient frontier and provided superior risk-adjusted returns. More recent evidence suggests that REITs may have higher volatility, Value-at-Risk (VaR), and expected shortfalls than equities in times of increased market volatility, precisely when the stabilizing properties of REITs are most desirable. This study expands on the emerging literature with two contributions. First, it examines REITs’ impact on the VaR of a portfolio of stocks and bonds over the last two decades. Second, a new, more accurate method of estimating VaR, conditional autoregressive value at risk (CAViaR), is used. The more accurate VaR estimates show that adding REITs to the portfolio has no significant impact on VaR until after the financial

2

crises begins in 2006. After 2006, adding REITs to a portfolio of stocks and bonds dramatically increases VaR. The results have significant implications for portfolio selection.

3

CHAPTER 2 INVESTIGATING THE IMPACT OF AGENCY DISCLOSURE ON RESIDENTIAL REAL ESTATE TRANSACTIONS

Introduction and Literature Review Real estate consumers have been confused about the nature of their relationship with real estate brokers for some time. In 1984 the Federal Trade Commission (FTC) released a study of the residential real estate brokerage industry that raised awareness about real estate consumers’ misconceptions regarding the role of real estate brokers. At the time, the conventional agency representation model had both the listing and selling brokers working as agents of the home seller. The FTC study found that in transactions involving two brokers, where both the listing broker and the selling broker would represent the home seller, 76.5% of sellers thought the listing broker represented their interests. Just over 74% believed that the other broker, who worked with the buyer, represented the buyer’s interests. While this is not correct, it is also not necessarily harmful to the seller. On the other hand, 74.2% of buyers believed the broker they dealt with (the selling broker) represented their interests only. This misconception could cause harm to the buyer if he revealed sensitive information to the broker, such as the maximum price he would be willing to pay for a property. Other studies around the same time reported similar confusion on the part of brokers and home buyers (Ball and Nourse (1988), and Marsh and Zumpano (1988)).

4

In response to these findings, the NAR organized an Agency Task Force to investigate the problem and recommend solutions. The Task Force recommended that every state pass mandatory disclosure laws requiring real estate brokers to disclose whose interest they represent. Today, every state has some form of mandatory disclosure statute, although states differ on how and when a broker must provide agency disclosure. Some states require written disclosure and mandate the exact text that buyers should receive, while other states leave it to the broker to write the disclosure form. A few states require only verbal disclosure. The timing of mandatory disclosure also differs between states. Some states require agency disclosure before any material discussions with a potential buyer, while others require disclosure only before the final contract is signed. Following mandatory disclosure reform, many states also created additional agency relationships with the hope of providing consumers with more appropriate representation. Now many states have selling and listing agents that represent the seller’s interest, buyer’s agents that represent only the buyer’s interest, non-agency brokers that assist in the transaction but do not act as agents for either party, and dual agents. A dual agent is a single agent that is expected to act in the interest of both the buyer and the seller in the transaction. Some researchers suggested that the differing requirements of agency disclosure and the increased number of agency relationships may not have solved the original problem of better informing and better serving real estate consumers (Miceli, Pancak, and Sirmans (2000) and Olazábal (2003)). Using the 2004 National Association of REALTORS® (NAR) Home Buyer and Seller Survey, Wiley and Zumpano (2004) confirmed that there was still a widespread problem with agency disclosure. The authors found that 40% of home buyers who used a real estate agent reported either no disclosure from their agent (20.5%), or were not sure if there was disclosure (19.5%). The remaining 60% reported that there was disclosure at some point during the purchasing process.

5

The 1984 FTC study found that 75% of home buyers acknowledge some form of agency disclosure from the real estate agent. Wiley and Zumpano found that a smaller proportion of home buyers reported disclosure after mandatory disclosure laws were passed. After examining several state disclosure forms, Wiley and Zumpano (2004 and 2009) note that their 2004 findings are not surprising. The forms are often not clear and will be one of many that home buyers will sign at some point in the purchase process. They argue that home buyers can easily be overwhelmed by the process and may not realize they have received an agency disclosure notice. In fact, they find that first time home buyers, younger buyers, lower income buyers, and buyers with English a secondary language are all less likely to report agency disclosure. While one could argue that these types of buyers are more likely to be taken advantage of by unethical brokers, it is also possible that these buyers are less experienced and more likely to be overwhelmed by the home buying process. These buyers may not have remembered, or may have not realized, that one of the many forms signed during the process was an agency disclosure form. Although the lack of reported agency disclosure is troubling and needs to be addressed, to date no study has provided evidence that home buyers are materially harmed when they do not report receipt of agency disclosure. This study seeks close the gap by looking for evidence of consumer harm when there is a reported lack of agency disclosure. A home buyer may be harmed by a lack of agency disclosure is if he is unknowingly working with an agent of the seller and reveals sensitive information, such as the maximum price he is willing to pay. The agent has both a fiduciary duty to share this information with the seller, and a financial motivation since most real estate agents are paid a percentage of the selling price.

6

These buyers may be harmed by paying higher prices for homes when compared to similar buyers who acknowledge receiving agency disclosure, understood the agency relationships of the involved brokers, and kept sensitive information private. One might ask how buyers may not be harmed if they do not receive agency disclosure from a broker. It is not possible to determine whether the home buyer knew whom the broker represented using the NAR survey. It is only possible to know if the buyer reported receiving agency disclosure. The question from the NAR survey used in this study and the 2004 study reads, “At the time you began working with a real estate agent, did your agent ask you to sign a disclosure statement indicating who he/she represented in the transaction?” Survey respondents then pick from the following answers: “Yes, at first meeting”, “Yes, when contract was written”, “Yes, at some other time”, “No”, or “Don’t know”. If a home buyer chooses one of the “Yes” answers, it is fair to assume he or she knew whom the broker represented. If the buyer chooses either “No” or “Don’t know”, then it is not clear whether he or she knew whom the broker represented. For example, the buyer may have known that the agent represented the seller, but did not remember receiving a disclosure statement. If this is the case, then it is unlikely that the buyer revealed sensitive information to the broker. It is also possible that brokers are confused about their agency relationship with real estate consumers. Moore, Smolen, and Conway (1993) conducted a survey of brokers and home buyers across Ohio shortly after mandatory disclosure took effect in the state. They found that 30% of responding Ohio brokers believed they represented either the buyer, or both the buyer and seller in real estate transactions. At the time, all agents operating in the multiple listing service were considered agents of the seller. If the broker believes he is an agent of the buyer, then he is not likely to pass along harmful information to the seller.

7

Given that it is not possible to determine whether the home buyer knew whom the broker represented in the transaction, and that brokers themselves may not realize whom they are supposed to represent, it is possible that even if home buyers do not report receiving agency disclosure, they may not be harmed in the negotiation process when compared to buyers that do report receiving disclosure This paper advances the literature by answering four questions. First, has receipt of agency disclosure improved since previous studies were completed? Second, have the types of buyers that are less likely to report receipt of agency disclosure changed? Third, are home buyers that do not report receipt of agency disclosure harmed by paying higher prices when compared to buyers that report receipt of agency disclosure? Finally, among those buyers who receive agency disclosure, does the timing of disclosure matter? The 2012 survey results show that while the types of buyers that are more likely to report receipt of agency disclosure have changed, the overall proportion of buyers reporting receipt of disclosure has not improved. Roughly 37% of home buyers said that they did not receive an agency disclosure form at some point during the home buying process in 2012. This is close to the result that Wiley and Zumpano (2004) found using 2004 NAR survey data, and worse than the original 1984 FTC study. Despite the fact that agency disclosure is still a problem, there is little evidence that home buyers who do not report receiving agency disclosure pay higher prices for homes when compared to those that report receiving agency disclosure. The results are robust to selection bias concerns from the types of buyers that are more likely to report receipt of agency disclosure. There is, however, evidence that the timing of disclosure matters. Looking only at home buyers who reported receipt of agency disclosure, those reporting disclosure at a time other than at the first meeting with a broker are associated with 3.2% higher home prices

8

when compared to buyers that receive disclosure at first contact. This suggests that buyers are placed a disadvantage during the negotiation process when agency disclosure is not made immediately. Data and Methodology The data comes from the 2012 NAR Home Buyer and Seller Survey. The full dataset includes 8,501 responses from both home buyers and sellers throughout the United States. This study examines home buyers’ receipt of agency disclosure and the associated impact on purchase price, so only those respondents who purchased a home with the assistance of a real estate broker are considered. Narrowing the data to these home buyers and dropping observations with missing or erroneous responses leaves a maximum of 5,525 and a minimum of 5,435 observations depending the model specification. The Models Trying to model the impact of agency disclosure on purchase price presents a potential selection bias problem. The simplest model choice is to use a hedonic pricing model to control for housing and buyer characteristics and include a dummy variable for the receipt of agency disclosure. However, Wiley and Zumpano (2004) find that first time home buyers, buyers under 35, buyers 65 and over, Hispanic buyers, Asian buyers, and buyers with English as a secondary language are all less likely to report receipt of agency disclosure. If these buyers also tend to pay different prices for homes as compared to other buyers, due to systematic differences in income, for example, then an OLS price model with a dummy variable for agency disclosure will be biased. To address this issue, a two-stage treatment effect regression using a Heckman correction is used to detect and correct for selection bias.

9

The first stage uses a probit regression to estimate the probability of reporting receipt of agency disclosure given various buyer characteristics. The second stage uses OLS to model purchase price on buyer demographics, housing characteristics, and a dummy variable for the receipt of agency disclosure. It also includes the inverse Mills ratio from the first stage to test for and correct selection bias. The first stage probit regression is specified as (a full description of variables is listed in Table 2.1): DISCLOSE = f(BTW35_55k, BTW55_75k, BTW75_100k, BTW100_150k, BTW150_200k, BTW200_500k, OVER500k, UNDER35, OVER65, BLACK, ASIAN, HISP, OTHER, SINGLE, CHILD, lnEARNERS, DIS, NOEXP, ENG_SEC, BB

(2.1)

DISCLOSE is coded as 1 if the buyer chooses “Yes, at first meeting”, “Yes, when contract was written”, or “Yes, at some other time” in response to question D13: “At the time you began working with a real estate agent, did your agent ask you to sign a disclosure statement indicating who he/she represented in the transaction?”1 It is coded as 0 if the buyer answers either “No” or “Don’t know”. Income (BTW35_55k, BTW55_75k, etc.- under 35k is the omitted category) may be related to agency disclosure since higher income buyers tend to be better educated, older, and more experienced than lower income buyers, and so may be more likely to remember signing an agency disclosure form. Controls for age (UNDER35 and OVER65 – between 35 and 65 is omitted), race (BLACK, ASIAN, HISP, OTHER – white is the omitted category), first time homebuyers (NOEXP), and buyers with English as a secondary language (ENG_SEC) are included because previous research found that first time home buyers, buyers under 35, buyers 65 and over, Hispanic buyers, Asian buyers and buyers with English a secondary language are all 1

The results are qualitatively unchanged if responses of “Don’t know” are dropped from the regression analysis.

10

less likely to report receipt of agency disclosure. Buyers purchasing a home more than 50 miles from their current residence (DIS) are likely unfamiliar with the area and are working under time constraints when meeting with a broker from out of town. It is possible that they are less likely to recall receipt of agency disclosure given these additional difficulties. Family characteristics such as whether the buyer is married (SINGLE) and how many children under the age of 18 are in the home (CHILD) help capture the degree of distraction during the home buying process. Single buyers may not have the benefit of a second individual to help pay attention to and remember some aspects of the purchase process and may be less likely to recall singing a disclosure statement. Buyers with more children at home may face greater time and attention constraints and thus may less likely to recall receiving a disclosure statement. The number of income earners in the home (lnEARNERS) may help predict reported receipt of agency disclosure. It is possible that income earners are involved at some level in the home buying process. With more people involved in the process, it is more likely that one of them will recall receiving a disclosure form. The presence of a buyer broker is expected to be positively related to receipt of agency disclosure. A buyer broker likely spent time discussing and promoting the unique agency relationship with the buyer and so the buyer may be more likely to recall signing a disclosure statement. Previous research (Baryla et al. (2000)) found that buyer brokers are not associated with different selling prices as compared to other types of broker assisted transactions. The 2012 survey data show that purchasers using buyer brokers are not associated with different prices for homes as compared to other buyers, but are significantly more likely to report agency disclosure. State dummy variables are also included to control for variation in mandatory disclosure laws from state to state.

11

Table 2.1 Description of Variables Variable

Survey Question

DISCLOSE

D13

lnSP

C1

BTW35_55k

H11

BTW55_75k

H11

BTW75_100k

H11

BTW100_150k

H11

BTW100_150k

H11

BTW150_200k

H11

BTW200_500k

H11

OVER500k

H11

UNDER35

H7

OVER65

H7

ASIAN

H8

BLACK

H8

HISP

H8

Description Indicator variable equal to 1 if the buyer reported receiving agency disclosure and 0 if the buyer did not report receipt of agency disclosure, or if he was not sure if there was agency disclosure. The natural log of the purchase price. Indicator variable that equals one if buyer’s total household income was between $35,000 and $54,999 inclusive, and zero otherwise. Indicator variable that equals one if buyer’s total household income was between $55,000 and $44,999 inclusive, and zero otherwise. Indicator variable that equals one if buyer’s total household income was between $75,000 and $99,999 inclusive, and zero otherwise. Indicator variable that equals one if buyer’s total household income was between $100,000 and $149,999 inclusive, and zero otherwise. Indicator variable that equals one if buyer’s total household income was between $150,000 and $149,999 inclusive, and zero otherwise. Indicator variable that equals one if buyer’s total household income was between $150,000 and $199,999 inclusive, and zero otherwise. Indicator variable that equals one if buyer’s total household income was between $200,000 and $499,999 inclusive, and zero otherwise. Indicator variable that equals one if buyer’s total household income was $500,000 or more, and zero otherwise. Indicator variable equal to one if the buyer is 35 years old or younger. Indicator variable equal to one if the buyer is 65 years of age or older. Indicator variable that equals one if the buyer is Asian, and zero otherwise. Indicator variable that equals one if the buyer is black, and zero otherwise. Indicator variable that equals one if the buyer is Hispanic, and zero otherwise. 12

OTHER

H8

SINGLE

H1

CHILD

H2

lnEARNERS

H6

DIS

A12

NOEXP

A1

EM

A21

NH

A21

PRIM

A10

PREVOCC

A2

DETSFAM

A8

lnSF lnBED lnBATH lnHOME_AGE

A11 A11 A11 A7

RURAL

A13

URBAN

A13

BURB

A13

RESORT

A13

BB

D3

LAMBDA

NA

Indicator variable that equals one if the buyer does not self -report race as white, Asian, Hispanic, or black. White is the omitted category. Indicator variable equal to one if the buyer is single and zero if a married or unmarried couple. Number of children under the age of eighteen in the buyer’s household Natural log of the number of income earners in the home buyer’s household. Indicator variable that equals one if the buyer is purchasing a home more than 50 miles from his previous residence. Indicator variable equal to 1 if the respondent is a first-time home buyer and zero otherwise. Indicator variable that equals one if the respondent is buying a home due to an employer mandated move. Indicator variable equal to one if the move is due to a change in family situation such as marriage or divorce. Indicator variable that equals one of the home purchased is a primary residence, and zero otherwise. Indicator variable that equals one if the home was previously occupied, and zero if it was new. Indicator variable that equals one if the home is described as a detached, single family home and zero otherwise. Natural log of the home’s square footage. Natural log of the number of bedrooms in the home. Natural log of the number of bathrooms in the home. Natural log of the age of the home. Indicator variable equal to one if the location of the home is described as rural and zero otherwise. Indicator variable equal to one if the location of the home is described as urban and zero otherwise. Indicator variable equal to one if the location of the home is described as suburban and zero otherwise. Indicator variable equal to one if the location of the home is described as a resort and zero otherwise. “Small town” is the omitted category. Indicator variable the equals one if the home was purchased with the assistance of a buyer broker, and zero otherwise. Inverse Mills ratio from the first stage probit regression.

13

The second-stage OLS regression is specified as: lnSP = f(BTW35_55k, BTW55_75k, BTW75_100k, BTW100_150k, BTW150_200k, BTW200_500k, OVER500k, UNDER35, OVER65, BLACK, ASIAN, HISP, OTHER, SINGLE, CHILD, lnEARNERS, NOEXP, EM, NH, PRIM, PREVOCC, DETSFAM, lnSF, lnBED, lnBATH, lnHOME_AGE, RURAL, URBAN, BURB, RESORT, (2.2)

DISCLOSE, LAMBDA The dependent variable, lnSP, is the natural log of the final purchase price.

The first several independent variables include buyer characteristics that may influence the final purchase price. Income is expected to be positively related to purchase price. Age is also expected to be positively related to price since income tends to rise with age, at least until retirement. Race may be positively or negatively related to price if different racial groups tend to have systematic differences in income or demand for certain housing characteristics. Family characteristics may be positively or negatively related to price. Single buyers may have lower household income than married households, which would tend to associate single buyers with lower home prices. However, single buyers may have fewer non-housing related expenses than married buyers, such as childcare or multiple car payments, and may be able to use a greater portion of income to purchase housing. If this effect dominates, it would tend to associate single buyers with higher home prices. The relationship between the number of children in the home and final purchase price could also run in either direction. On one hand, households with more children may demand larger and thus more expensive homes to accommodate the family, which would lead to a positive relationship between the number of children and purchase price. On the other hand, more children in the home would lead to more child care costs (daycare, one parent staying at home rather than working, etc.), which would

14

reduce household income available for housing. This would lead to a negative relationship between the number of children and home purchase price. The number of income earners in the home (lnEARNERS) may be positively or negatively related to purchase price. More income earners in the home would tend to drive up total household income and thus be positively related to purchase price, but it could be that homes with more income earners are less likely to have one or two large breadwinners. They may have more income earners precisely because the total household income is low, which would lead to a negative relationship between the number of earners and final purchase price. NOEXP should be negatively related to price since first time home buyers tend to be younger and have lower incomes than more experienced home buyers. Employer mandated (EM) moves tend to involve a set time frame and include some relocation assistance from the employer. These buyers may be more motivated to find a home quickly and less price sensitive, which would lead to a positive relationship with purchase price. Buyers purchasing because of the creation of a new household through marriage of divorce (NH) may be likely to pay more or less for a home depending on the circumstances. The next set of independent variables includes housing characteristics that may help explain purchase price. Buyers purchasing a primary residence (PRIM) are likely to spend a greater proportion of their household income on the home when compared to a vacation or investment property, which would predict a positive sign on PRIM. Previously occupied homes tend to sell for less than new homes, so PREVOCC is expected to sign negative. DETSFAM is expected to sign positive since detached single family homes tend to sell for higher prices than similar, non-detached single family homes (condos, for example).

15

The square footage of the home (lnSF), the number of bedrooms (lnBED), and the number of bathrooms (lnBATH) should be positively related to purchase price. lnHOME_AGE is expected to sign negative since older homes tend to come with greater expected maintenance costs and thus sell at a discount to newer homes. The location of the home is likely important in explaining price. Homes in URBAN or RESORT areas likely sell for higher prices than similar homes in RURAL or suburban (BURB) areas. If home buyers who do not report receipt of agency disclosure are systematically paying higher prices for homes, then DISCLOSE will sign positive and significant. This would be clear evidence that buyers who do not understand the agency relationship with their broker are revealing sensitive information and are being harmed during the negotiation process. If DISCLOSE is not significant, or signs negative and significant, then there is little evidence that buyers who do not report receipt of agency disclosure are being harmed by paying higher prices for homes. LAMBDA is the invers Mills ratio from the first stage. If it is significant in the second stage, there is evidence of selection bias and including LAMBA will make the second stage results unbiased and consistent. If LAMBDA is not significant, then there is no evidence of selection bias and the results from a standard OLS model are preferable. The OLS model is the same as equation (2) above, but without the inclusion of LAMBDA. State dummy variables are also included in the second stage (and standard OLS model) since prices for similar homes tend to vary by state (consider the expected price difference of a home purchased in New York or California as compared to a similar home purchased in Mississippi, for example).

16

The timing of disclosure may also be important to buyers. Some states require agency disclosure at first contact, while other states only require disclosure before a final contract is signed. If the buyer only realizes whom the broker represents after revealing sensitive information, such as the maximum price he is willing to pay for a given property, it could be just as harmful as never receiving disclosure from the broker. To see if the timing of disclosure is meaningful, the sample is limited to those buyers who received a disclosure statement at some point during the home purchase process. The three options for the timing of disclosure are then given a dummy variable in the OLS price regression. DISCLOSE_CONTRACT is coded as 1 if the buyer chose “Yes, when contract was written” to the question of when disclosure was received, and zero otherwise. Once a contract is written, the buyer has offered a price and the seller has accepted. Only receiving a disclosure statement after the price has been set is arguably just as bad as never receiving disclosure. DISCLOSE_OTHER_TIME is coded as 1 if the buyer chose “Yes, at some other time” and zero otherwise. “Yes, at first meeting” is the omitted category. If the probability that a buyer will reveal harmful information to a seller’s agent increases the longer he works with a broker without receiving agency disclosure, then buyers receiving disclosure at a time other than at first meeting may be associated with higher prices. The regression is: lnSP = f(BTW35_55k, BTW55_75k, BTW75_100k, BTW100_150k, BTW150_200k, BTW200_500k, OVER500k, UNDER35, OVER65, BLACK, ASIAN, HISP, OTHER, SINGLE, CHILD, lnEARNERS, NOEXP, EM, NH, PRIM, PREVOCC, DETSFAM, lnSF, lnBED, lnBATH, lnHOME_AGE, RURAL, URBAN, BURB, RESORT, DISCLOSE_OTHER_TIME, DISCLOSE_CONTRACT

17

(2.3)

If home buyers who receive agency disclosure sometime after the first contact with a broker are systematically paying higher prices for homes when compared to those who received disclosure at first contact, then DISCLOSE_OTHER_TIME and/or DISCLOSE_CONTRACT will sign positive and significant. This would suggest that buyers are placed at a disadvantage during the negotiation process when disclosure is made sometime after first meeting with a broker. Empirical Results Summary Statistics Despite the fact that mandatory disclosure laws have been in effect in all 50 states for several years, buyers’ reported receipt of agency disclosure has not improved. The responses to question D13 (“At the time you began working with a real estate agent, did your agent ask you to sign a disclosure statement indicating who he/she represented in the transaction?”) are reported in Table 2.2. Sixty-three percent of buyers report being asked to sign an agency disclosure statement at some point during the home buying process. The remaining 37% report that they either were not asked to sign a disclosure form (19%), or did not know if they were asked to sign a disclosure form (18%). The results match closely with Wiley and Zumpano’s 2004 findings. The proportion of home buyers reporting receipt of agency disclosure in 2012 is still lower than the 1984 FTC report where 75% of home buyers acknowledged receipt of agency disclosure. Home buyers’ rate of reported agency disclosure is still troubling, but a key question remains unanswered: Are these buyers harmed due to a lack of recognition of agency disclosure? Table 2.3 shows the average purchase price for buyers that report receipt of agency disclosure and buyers that do not report receipt of agency disclosure. For the purposes of this study, buyers that did not receive a disclosure form and those that are not sure if they received a form,

18

i.e. respondents that answered “No” or “Don’t know” to questions D13 on the 2012 survey, are included in the group that did not receive disclosure. In either case, the buyer may have revealed sensitive information because he did not realize whose interest the broker represented.2 Buyers that reported receipt of disclosure paid an average price of $265,405. Buyers that did not report discloser paid an average price of $242,802. Buyers that report not receiving disclosure pay slightly less for homes on average, but this does not correct for any buyer or housing characteristics that could impact price.

Table 2.2 Timing of Agency Disclosure Question D13: At the time you began working with a real estate agent, did your agent ask you to sign a disclosure statement indicating who he/she represented in the transaction? Number of

Percentage of

Cumulative

responses

responses

percentage

Yes, at first meeting

2,248

30.85%

30.85%

Yes, when contract was

1,491

20.46%

51.31%

846

11.61%

62.92%

No

1,406

19.29%

82.21%

Don’t know

1,296

17.79%

100.00%

written Yes, at some other time

2

All results are qualitatively unchanged when buyers that were not sure if they received agency disclosure, i.e. those that answered “Don’t know” to question D13, are dropped from the sample.

19

Table 2.3 Average Price by Agency Disclosure Receipt of Agency Disclosure

Average Purchase Price

Yes

$265,405

No

$242,802

Treatment Effect Results The results from the first stage probit regression on receipt of agency disclosure and the second stage OLS regressions on purchase price are reported in Tables 2.4 and 2.5, respectively. The variable of interest in the second stage, DISCLOSE, is not significant in explaining purchase price at any conventional level (Table 2.5). However, LAMBDA, the inverse Mills ratio from the first stage, is also not significant; suggesting that there is not a selection bias problem and OLS is the appropriate model. The OLS regression results are reported in Table 2.6 and discussed in the OLS Results section below. Prior to discussing the OLS results, the next section compares the first stage probit results with previous studies. Even though there is no evidence of selection bias, the results from the first stage probit regression (Table 2.4) are interesting. Wiley and Zumpano (2004) find that first time home buyers, buyers under 35, buyers 65 and over, Hispanic buyers, Asian buyers, and buyers with English as a secondary language are all less likely to report receipt of agency disclosure. The 2012 data show that first time home buyers and younger buyers are still less likely to report receipt of agency disclosure. However, Asian and Hispanic buyers are more likely to report receipt of agency disclosure. Perhaps better training for real estate agents (or greater fear of liability) has induced agents to expend more effort discussing agency disclosure with these buyers. 20

Table 2.4. First-Stage Probit Estimates on Reporting Disclosure. Dependent variable is DISCLOSE. Variable

Coefficient

Standard Error

P-Value

CONSTANT

-0.151

0.318

0.634

BTW35_55k

-0.014

0.072

0.842

BTW55_75k

0.141*

0.073

0.052

BTW75_100k

0.059

0.074

0.429

BTW100_150k

0.046

0.075

0.539

BTW150_200k

0.109

0.093

0.242

BTW200_500k

0.025

0.095

0.793

OVER500k

0.075

0.210

0.722

UNDER35

-0.100**

0.046

0.032

OVER65

0.050

0.070

0.472

BLACK

0.146

0.103

0.155

ASIAN

0.196**

0.101

0.053

HISP

0.170*

0.094

0.070

OTHER

-0.164

0.144

0.252

SINGLE

-0.032

0.048

0.500

CHILD

-0.008

0.018

0.656

lnEARNERS

0.223***

0.072

0.002

DIS

-0.088*

0.047

0.061

NOEXP

-0.183***

0.049

0.000

ENG_SEC

-0.104

0.109

0.340

21

BB

0.750***

State Dummies?

Yes

Observations

5,435

0.037

0.000

Chi-

9,806

squared(144) Prob > Chi-squared

22

0.000

Table 2.5. Second-Stage OLS Price Regressions with Selection Correction. Dependent variable is ln(purchase price). Variable

Coefficient

Standard Error

P-Value

CONSTANT

8.028***

0.184

0.000

BTW35_55k

0.164***

0.023

0.000

BTW55_75k

0.297***

0.023

0.000

BTW75_100k

0.451***

0.024

0.000

BTW100_150k

0.567***

0.025

0.000

BTW150_200k

0.732***

0.031

0.000

BTW200_500k

0.964***

0.032

0.000

OVER500k

1.256***

0.067

0.000

UNDER35

0.006

0.015

0.656

OVER65

0.039*

0.022

0.076

BLACK

-0.053*

0.031

0.090

ASIAN

0.111***

0.030

0.000

HISP

-0.018

0.028

0.537

OTHER

0.020

0.046

0.655

SINGLE

0.027*

0.015

0.077

CHILD

-0.002

0.006

0.724

lnEARNERS

-0.118***

0.023

0.000

NOEXP

-0.010

0.016

0.572

EM

0.034*

0.020

0.092

NH

-0.030

0.021

0.143

23

PRIM

0.287***

0.042

0.000

PREVOCC

-0.020

0.030

0.501

DETSFAM

0.070***

0.017

0.000

lnSF

0.439***

0.022

0.000

lnBED

-0.018

0.028

0.505

lnBATH

0.187***

0.021

0.000

lnHOME_AGE

-0.053***

0.008

0.000

RURAL

-0.058***

0.016

0.000

URBAN

0.080***

0.017

0.000

BURB

-0.088***

0.020

0.000

RESORT

0.121***

0.047

0.010

DISCLOSE

-0.056

0.043

0.198

LAMBDA

0.029

0.027

0.288

State Dummies?

Yes

Observations

5,435

Chi-

9,806

squared(144) Prob > Chi-squared

24

0.000

Buyers with English as a secondary language are no more or less likely to report receipt of agency disclosure. Only one income category, BTW55_75k, is significant. This income group is more likely to report receipt of agency disclosure while the other groups are no more or less likely to report disclosure. The number of income earners in the household is positively related to receipt of agency disclosure. Perhaps this is because income earners are more likely to be involved at some level in the home buying processes. More people involved in the processes may increase the chances that one of the household members recalls receipt of the disclosure statement. The presence of a buyer broker is positively related to the receipt of agency disclosure. Buyer brokers may spend time promoting the advantages of this unique agency relationship with the buyers and so the buyers may be more likely to recall signing a disclosure statement. It is encouraging that Asian and Hispanic buyers are more likely to report receipt of agency disclosure while buyers with English as a secondary language are no longer less likely to report disclosure. These changes from the 2004 study suggest improvements in agency disclosure from previous years. First time home buyers and younger buyers are still less likely to report receipt of disclosure. OLS Results Table 2.6 reports the results from the standard OLS regression. Most variables run in the expected direction. Buyer income, the square footage of the home, and the number of bathrooms are all positively associated with purchase price. Buyers over 65 years of age tend to pay higher prices for homes, perhaps reflecting the fact that Americans are retiring later and staying in their prime income earnings years longer.3 Single buyers also tend to pay higher prices for homes.

3

Increasing the number of age categories or using the natural log of age produces similar results: buyer age is positively associated with purchase price.

25

This may be because single buyers tend to spend more of their total income on housing when compared to married buyers that may have additional demands on household income (multiple car payments, childcare, etc.). The number of income earners in the household is negatively related to purchase price, possibly because homes with more income earners are less likely to have one or two large breadwinners and so have more income earners because the total household income is low. Buyers purchasing a primary residence and those buying a detached single family home tend to pay more for the property. The variable of interest, DISCLOSE, is not significant at any conventional level. After controlling for all buyer and housing characteristics available in the survey data, there is no evidence that buyers who report not receiving a disclosure form pay different prices than those that report receipt of disclosure. The result is important because it does not support the most commonly cited example of how home buyers may be harmed by a lack of agency disclosure. There is no doubt that the potential for harm exists when a buyer does not understand his agency relationship with a broker, but there is no evidence that these buyers tend to over pay when they do not acknowledge receipt of agency disclosure. Table 2.7 reports the results from a price regression that includes only buyers who report receipt of agency disclosure. Buyers that did not receive disclosure, or are not sure if they received disclosure, are dropped from the sample. The regression includes dummy variables for the timing of disclosure. Buyers reporting disclosure at a time other than the first meeting with a broker (DISCLOSE_OTHER_TIME) are associated with 3.2% higher home prices when compared to buyers that reported receipt of agency disclosure at the first meeting with a broker (the omitted category). Interestingly, buyers that report receipt of agency disclosure when the

26

contract is written (DISCLOSE_CONTRACT) are not associated with different prices when compared to those that receive disclosure at first contact. This contradictory result may be because of some confusion on the part of survey respondents. DISCLOSE_CONTRACT is coded as 1 if the buyer reported receipt of agency disclosure “…when contract was written”, but the term contract is not defined in the survey. One would assume that “contract” refers to the contract for the house, but there is no guarantee that the survey respondents interpreted it this way. If the buyer was asked to sign any kind of form, he may have assumed it was a contract with a broker and reported receiving disclosure when this “contract” was written. This may have been soon after the first meeting with the broker, sometime just before the negotiations were finished, or anytime in between. If the buyer responded “Yes, when contract was written”, but received agency disclosure soon after the first meeting, then the probability of revealing harmful information is reduced. These buyers would be more like those that received disclosure at first meeting. If enough of these respondents are in the sample, it would bias against finding a meaningful difference between those that report disclosure at the first meeting and those that report receipt of agency disclosure when the “contract” is written. It may also be that buyers who receive agency disclosure at the end of negotiations live in states that do not require disclosure until just before the contract is written. Brokers in these states may be informing buyers long before the contract signing, either out of concern for their reputation with buyers or out of fear of litigation, but only providing the signed disclosure form at signing as a matter of procedure.4

4

Dropping buyers that report disclosure receipt when the contract was written from the sample increases the magnitude and significance of the difference between buyers that report receipt of disclosure at first meeting and those that report receipt of disclosure at some other time. In untabulated results, buyers that report receipt of disclosure at a time other than the first meeting are associated with 4.2% higher home prices. The coefficient is significant at the 5% level.

27

Table 2.6. OLS Price Regressions without Selection Correction. Dependent variable is ln(purchase price). Variable

Coefficient

Standard Error

P-Value

CONSTANT

7.986***

0.180

0.000

BTW35_55k

0.165***

0.023

0.000

BTW55_75k

0.303***

0.023

0.000

BTW75_100k

0.454***

0.024

0.000

BTW100_150k

0.572***

0.025

0.000

BTW150_200k

0.737***

0.031

0.000

BTW200_500k

0.969***

0.032

0.000

OVER500k

1.276***

0.067

0.000

UNDER35

0.004

0.015

0.785

OVER65

0.041*

0.022

0.058

BLACK

-0.052*

0.031

0.097

ASIAN

0.111***

0.030

0.000

HISP

-0.020

0.023

0.463

OTHER

0.020

0.046

0.679

SINGLE

0.029**

0.015

0.050

CHILD

-0.001

0.006

0.879

lnEARNERS

-0.108***

0.022

0.000

NOEXP

-0.010

0.015

0.536

EM

0.030

0.020

0.139

NH

-0.035*

0.021

0.094

28

PRIM

0.274***

0.041

0.000

PREVOCC

-0.021

0.030

0.484

DETSFAM

0.069***

0.017

0.000

lnSF

0.441***

0.022

0.000

lnBED

-0.019

0.028

0.498

lnBATH

0.183***

0.021

0.000

lnHOME_AGE

-0.053***

0.008

0.000

RURAL

-0.057***

0.016

0.000

URBAN

0.076***

0.017

0.000

BURB

-0.093***

0.019

0.000

RESORT

0.130***

0.047

0.006

DISCLOSE

-0.011

0.012

0.362

State Dummies?

Yes

Observations

5,525

F(79, 5445)

121.65

Adjusted R-

0.6331

Prob > F

0.000

squared

29

Table 2.7. OLS Price Regressions. Includes only buyers reporting disclosure. Disclosure at first meeting is the omitted category. Dependent variable is ln(purchase price). Variable

Coefficient

Standard Error

P-Value

CONSTANT

8.155***

0.219

0.000

BTW35_55k

0.162***

0.029

0.000

BTW55_75k

0.302***

0.029

0.000

BTW75_100k

0.445***

0.030

0.000

BTW100_150k

0.575***

0.031

0.000

BTW150_200k

0.732***

0.037

0.000

BTW200_500k

0.961***

0.040

0.000

OVER500k

1.309***

0.079

0.000

UNDER35

0.003

0.018

0.864

OVER65

0.042

0.027

0.113

BLACK

0.051

0.036

0.164

ASIAN

0.072**

0.035

0.041

HISP

-0.017

0.034

0.616

OTHER

-0.022

0.060

0.711

SINGLE

0.040**

0.018

0.027

CHILD

-0.001

0.007

0.856

lnEARNERS

-0.144***

0.027

0.000

NOEXP

-0.018

0.019

0.347

EM

0.017

0.025

0.492

30

NH

-0.054**

0.025

0.032

PRIM

0.205***

0.050

0.000

PREVOCC

-0.054

0.036

0.138

DETSFAM

0.057***

0.020

0.005

lnSF

0.438***

0.027

0.000

lnBED

0.037

0.034

0.275

lnBATH

0.157***

0.025

0.000

lnHOME_AGE

-0.045***

0.009

0.000

RURAL

-0.055***

0.020

0.006

URBAN

0.052**

0.022

0.016

BURB

-0.101***

0.023

0.000

RESORT

0.142**

0.058

0.014

0.032*

0.019

0.087

-0.010

0.016

0.518

DISCLOSE_ OTHER_TIME DISCLOSE_ CONTRACT State Dummies?

Yes

Observations

3,522

F(80, 3441)

80.97

Adjusted R-

0.6331

Prob > F

0.000

squared

31

Conclusion After a 1984 FTC study publicized the widespread misunderstanding of agency relationships in residential real estate, all fifty states passed mandatory agency disclosure laws. Many states also created additional agency relationships to better serve home buyers. Despite these attempts to better inform and serve consumers, recent studies have found that real estate consumers’ reported receipt of agency disclosure has not improved. While calls for better agency disclosure are justified, the existing literature has not presented empirical evidence that consumers are harmed when they do not report receipt of agency disclosure from a broker. The most commonly cited example of how a home buyer may be harmed by a lack of agency disclosure is if a buyer, unknowingly working with an agent of the seller, reveals sensitive information, such as the maximum price he is willing to pay. The agent then shares this information with the seller and uses it to extract the maximum possible price from the buyer. These buyers may be harmed by paying higher prices for homes when compared to similar buyers that acknowledge receiving agency disclosure, understood the agency relationships of the involved brokers, and kept sensitive information private. Using data from the National Association of REALTORS® 2012 Home Buyer and Seller Survey, this study extends the literature with three key findings. First, although the types of buyers that are more likely to report receipt of agency disclosure have changed, the overall proportion of buyers reporting receipt of disclosure has not improved since previous studies were completed. Roughly 37 percent of home buyers report not receiving a disclosure statement. Second, there is no evidence that buyers who do not report receiving agency disclosure pay different prices for homes than buyers who do report receiving agency disclosure. The results are robust to selection bias concerns from the types of buyers that have historically been more

32

likely to report receipt of agency disclosure. This suggests that on average, consumers that do not receive agency disclosure do not pay higher prices for homes when compared to similar buyers purchasing similar homes. Finally, there is evidence that the timing of disclosure matters. Looking only at home buyers that reported receipt of agency disclosure, those reporting disclosure at a time other than at the first contact with a broker, but before the contract is signed, are associated with 3.2% higher home prices when compared to buyers that receive disclosure at first contact. This suggests that buyers are placed a disadvantage during the negotiation process when agency disclosure is not made immediately. Taken together, the results demonstrate the need for continued improvement in mandatory agency disclosure statutes. References Elder, W., L. Zumpano, and E. Baryla. (2000). “Buyer Brokers: Do They Make a Difference?” Real Estate Economics 28(2), 337-362. Federal Trade Comission. (1984). “Staff Report,” The Residential Real Estate Brokerage Industry 1 and 2, Washington, DC: U.S. Government Printing Office. Ball, J. and H. Nourse. (1988). “Testing the Conventional Representation Model for Residential Real Estate Brokerage,” Journal of Real Estate Research 3, 119-131. Marsh, G. and L. Zumpano. (1988). “Agency Theory and the Changing Role of the Real Estate Broker,” Journal of Real Estate Research 3, 151-164. Micelli, T. J., K. A. Pancak, and C. F. Sirmans. (2000). “Restructuring agency relationships in the real estate brokerage industry: an economic analysis,” Journal of Real Estate Research, 20(1), 31-47. Moore, G., G. Smolen, and L. Conway. (1993). “The Effects of an Informational Disclosure Form on the Real Estate Agency Representation Model,” Journal of Real Estate Research, 7, 31-47. Olazábal, A. (2003). “Redefining Realtor Relationships and Responsibilities: The Failure of State Regulatory Responses,” Harvard Journal on Legislation, 40, 65-132. Wiley, J. and L. Zumpano. (2004). “Questioning the Effectiveness of Mandatory Agency Disclosure Statutes,” Journal of Housing Research, 15(2), 161-174.

33

Wiley, J. and L. Zumpano. (2009). “Agency Disclosure in the Real Estate Transaction and the Impact of Related State Policies,” Journal of Real Estate Research, 31(3), 265-283.

34

CHAPTER 3 CAN REAL ESTATE BROKERS AFFECT HOME PRICES UNDER EXTREME MARKET CONDITIONS?

Introduction and Literature Review The market for residential real estate, being local, has always been subject to high information and transaction costs, and, therefore, deemed to be inefficient in the sense that prices may not reflect underlying market conditions and selling time may be unduly lengthy. Such a market gives rise to middlemen (e.g. real estate brokers) who work to reduce information and transaction costs, helping to bring buyers and sellers together and, in turn, reduce economic inefficiencies. There is a sizeable literature investigating what benefits, if any, real estate brokers provide home buyers and sellers. Among other things, researchers have investigated brokers’ impact on selling price and marketing time. In many of these studies, the effects are time-dependent or use local (state or MSA level) data, with varying results depending on when the study was undertaken and what market was examined. The literature search revealed a rough consensus that broker intermediation does not necessarily result in different selling prices, but does tend to reduce time on market (TOM), when compared to a for sale by owner transaction (or FSBO). This paper extends the current literature by comparing the performance of the real estate brokerage market over different time

35

periods and different market conditions. Specifically, this paper will compare the impact of broker intermediation on the sales price of a home during a seller’s market, as represented by 2006 data, and the market conditions that prevailed in 2009, very much a buyer’s market, when prices were falling and marketing time became much longer. Using a nation-wide dataset from the National Association of REALTORS® (NAR) Home Buyer and Seller Survey, this study presents evidence that homes sold with broker intermediation realized higher sales prices when compared to FSBO transactions in both 2006 and 2009. The results are robust over a number of model specifications that control for selection bias in the choice to use a broker and for the endogeneity of sales price and time on market (TOM). This is the first study using nation-wide data from extreme boom and bust markets that has documented a significant difference in the sales price of broker-marketed versus ownermarketed homes. Some evidence suggests the results may be due to the extreme market conditions that existed in 2006 and 2009, but more data is necessary to verify that the price differential is a temporary artifact of the unique real estate markets in those years. The imperfect flow of information is a well-established characteristic of the residential real estate market. Properties are heterogeneous and buyer and seller reservation prices are private information. In such a market, real estate brokers function as middlemen to help market participants gather information. If a home seller knew all potential buyers’ reservation prices for his property, then the choice of who to sell to at what price would be obvious. Since this information is not readily available, and is costly to acquire, the home seller is faced with the choice of paying for the services of a broker or gathering the information himself. If a seller chooses to market the property without aid, he will expend both time and money advertising the house, showing the house, negotiating with potential buyers, and tending

36

to the various administrative duties associated with closing the sale. Engaging the services of broker shifts a portion of this burden to the broker, but requires the seller pay a commission, usually a fixed percentage of the final sales price. Brokerage firms often market their services to home sellers by claiming that intermediation can help find a buyer faster and negotiate a better price as compared to selling the property without the aid of a broker (Huang and Rutherford (2007)). Early theory work assumes that brokers can better match buyers and sellers, resulting in higher sales prices and lower marketing times (see Yinger (1981), Jud (1983) and Jud and Frew (1986), for example). This assumption is rooted in the fact that most brokers have access to the multiple listing service (MLS). Users of the MLS claim that the system provides a broker with more up-to-date pricing, financing, and market information as well as access to a larger pool of buyers as compared to non-MLS users. If true, then the likelihood of selling at a higher price and in a shorter time frame through the MLS reduces the net cost of hiring a broker. Rational sellers will weigh the net cost of hiring a broker against the opportunity cost of marketing the property without aid. Whether brokers, or the MLS, help sellers obtain higher sales prices and shorter marketing times is an empirical question. Early results on brokers’ price impacts were mixed, but a rough consensus has emerged that brokerage intermediation does not necessarily affect sales prices, but does tend to reduce marketing time. The mixed results on the brokerage market’s impact on sales price are not surprising. Setting aside the various econometric problems that researchers face when trying to model home sales price, it is not clear ex ante whether brokers should obtain higher prices for the homes they help market. On the one hand, a broker marketed property reduces buyer search costs and increases buyer welfare by helping the buyer find a better match (Bagnoli and Khanna (1991)

37

and Ford et al. (2005)). The reduction in search costs and the increase in welfare may lead buyers to pay more for a broker-marketed home. If true, this would allow home sellers, who often pay the brokerage commission, to pass along at least part of the commission to buyers in the form of higher sales prices (Jud and Frew (1986) and Bagnoli and Khanna (1991)). On the other hand, brokers reduce sellers’ search costs and thus a seller may be willing to accept a lower price than if he were to market the property as a FSBO. There is also evidence that the fixed commission fee structure does not align the broker’s interests with the seller’s in obtaining the highest possible sales price in a given time period. Instead, the broker is incented to sell the home as quickly as possible. This could lead the broker to pressure a seller into accepting a lower sales price, even if a higher price might be obtained with a counter offer, or waiting for another offer (Jud (1983); Zumpano and Hooks (1988); Rutherford et al. (2005); and Levitt and Syverson (2008)). In one of the first studies to empirically investigate broker choice and broker impact on sales price, Jud (1983) extends Yinger’s (1981) theoretical model and presents empirical evidence. Jud finds that the choice to use a broker is primarily fueled by seller transactions cost and that brokers do not affect home prices. In contrast, Jud and Frew (1986) compare MLS and non-MLS listed properties in Charlotte, NC in 1977 and find that MLS listed properties tend to have higher sales prices. Since all properties listed on the MLS over the sample period are broker-listed homes and most non-MLS listed properties are FSBO homes, the authors argue the result as evidence that brokers are able to obtain higher prices for the homes they list. Johnson et al. (2005) found that broker-marketed properties sold outside the MLS had higher prices than similar properties sold by brokers within the MLS. The authors argue that brokers working outside the MLS may be better able to match buyers and sellers than more

38

traditionally broker-marketed homes. Other studies found that brokers do not impact the final sales price of a home (Kamath and Yantek (1982) and Colwell et al. (1992), for example). Many of these studies do not address the issue of selection bias when choosing to list the home with a real estate broker. Jud (1983) found that transaction costs tend to drive the decision to use a broker, and an important source of transaction cost is the seller’s income. Higher income sellers incur greater costs as they take time away from work to show the home and negotiate with potential buyers. Since income and home prices are highly correlated, it could be that broker-marketed homes are systematically more expensive than FSBO properties due to the characteristics of sellers that choose to work with brokers rather than any value-added service brokers provide. Although the authors only consider home buyers, Zumpano et al. (1996) and Elder et al. (2000) find evidence of selection bias in the choice to work with a broker. Specifically, both studies find that higher income home buyers are more likely to use a broker. After using a Heckman correction to control for selection bias, neither study finds a significant difference in sales price between broker-assisted purchases and homes purchased directly from the owner in a FSBO transaction. More recently, Hendel et al. (2008) compared properties listed on the FSBOMadison.com website to MLS listed properties in Madison Wisconsin from 1998 – 2005. They find that MLS listed properties sell faster, but not for higher prices. The authors control for selection by observing the same house and the same seller over multiple transactions. Bernheim and Meer (2008) compare homes sold on the Stanford University Campus from 1980 – 2005. The homes are only available to Stanford faculty and senior staff, whom the authors argue are more homogeneous than the general population. Stanford also provides a free listing service for

39

qualified faculty and staff, so there is no need to list a property on the MLS. Some sellers still choose to engage a broker while others do not. Comparing the broker-marketed homes with FSBO properties, the authors find that brokers accelerate sales, but do not have a significant impact on selling price. In addition to having a more homogeneous sample of home sellers than other studies, the authors also use home and seller fixed effects to control for selection. The above literature review is only a sampling of the existing work on real estate brokerage intermediation. The interested reader will want to review Benjamin et al. (2000) and Zietz and Sirmans (2011) for a more comprehensive review of the literature. The types of home sellers that tend to use brokers and how brokers impact sales prices for the homes they market is well researched. This study extends the literature by testing whether the established relationships changed during one of the most dramatic residential real estate boom and bust cycles in U.S. history. Data and Methodology This study uses the National Association of REALTORS® Home Buyer and Seller Survey from 2006 and 2009. The full dataset for 2006 (2009) includes 7,548 (9,138) responses from home buyers and sellers from every state in the US. This study only considers those respondents that sold a home without the assistance of a broker at any stage, and those that used a broker without trying to sell the home themselves at any stage. Limiting the 2006 sample to these sellers and dropping observations with missing or erroneous responses leaves a minimum of 2,454 and a maximum of 2,455 observations depending on the model specification. Limiting the 2009 sample to these sellers and dropping observations with missing or erroneous responses leaves a minimum of 1,759 and a maximum of 1,735 observations depending on the model specification.

40

The Models Modeling a real estate broker’s impact on purchase price seems a straightforward process at first blush. The first studies used hedonic pricing models to control for the impact of housing characteristics (e.g. square footage, number of bedroom and bathrooms, age of the house, etc.) and buyer characteristics (e.g. age, race, sex, income, etc.). One could simply include a dummy variable for the presence of a real estate broker in the regression, or estimate separate equations for broker assisted purchases and non-broker assisted purchases. The problem with these approaches is two-fold. First, several studies have documented the endogeneity of sales price and time-on-market (Miller (1978), Kang and Gardner (1989), Glower et al (1998), Anglin et al. (2003), for example), which can make a pricing model inconsistent if not addressed. Second, there may be selection bias in the choice to use a real estate broker which, if present, must be corrected to isolate the broker’s impact on purchase price. This paper offers two model specifications to address these issues: a treatment effect model to identify and correct for selection bias in the choice to list a home with a broker, and a two-stage least squares model to control for the endogeneity TOM and sales price. Two-stage Treatment Effect The first specification is a two-stage treatment effect regression with a Heckman correction. The first stage uses a probit model to estimate the seller’s choice to use a broker. Variables used to explain broker choice capture the seller’s demographics, opportunity costs, motivation, and relationship with the buyer (AQUAINT). See Table 3.1 for a full variable list with descriptions.

41

Table 3.1 Description of Variables Variable

2006 Survey Question

2009 Survey Question

RE

71

E19

lnSP

64

E10

BTW35_85k

100

H8

OVER85k

100

H8

BTW35_50

95

H4

OVER50

95

H4

WHITE

97

H5

SINGLE

92

H1

CHILD

93

H2

EARNERS

94

H3

URGENT

66

E13

FINDIF

63

E9

AQUAINT

68

E18

EM

63

E9

Description

Dummy variable equal to one if the seller used a broker and zero if the home was sold without a broker’s assistance. Natural log of the final selling price. Indicator variable that equals one if buyer’s total household income was between $35,000 and $84,999 inclusive, and zero otherwise. Indicator variable that equals one if buyer’s total household income was over $85,000. Under $35,000 is the omitted category. Indicator variable equal to one if the buyer is between 35 age 50 years of age (inclusive) and zero otherwise. Indicator variable equal to one of the buyer is over 50 years of age and zero otherwise. Indicator variable that equals one if the buyer is white, and zero otherwise. Indicator variable equal to one if the seller is single and zero if a married or unmarried couple. Number of children under the age of eighteen in the seller’s household Natural log of the number of income earners in the home buyer’s household. Indicator variable that equals one if the seller reported needing to sell the home “very urgently” or “somewhat urgently”, and zero otherwise. Indicator variable that equals one if the seller sold the home due to financial difficulty. Indicator variable that equals one if the home was sold to an acquaintance of the seller and zero otherwise. Indicator variable that equals one if the respondent is selling a home due to an employer mandated move.

42

NH

63

E9

SF or lnSF

60

E6

DETSFAM

59

E7

RURAL

61

E7

URBAN

61

E7

BURB

61

E7

RESORT

61

E7

REDUCE

67

E14

lnTOM

69

E15

LAMBDA

NA

NA

Indicator variable equal to one if the move is due to a change in family situation such as marriage or divorce. Indicator variables for each square footage category (2006) or natural log of the home’s square footage (2009). Indicator variable that equals one if the home is described as a detached, single family home and zero otherwise. Indicator variable equal to one if the location of the home is described as rural and zero otherwise. Indicator variable equal to one if the location of the home is described as urban and zero otherwise. Indicator variable equal to one if the location of the home is described as suburban and zero otherwise. Indicator variable equal to one if the location of the home is described as a resort and zero otherwise. “Small town” is the omitted category. Number of times the asking price was reduced before the home sold. Natural log of the number of weeks the home was on the market before it sold. Inverse Mills ratio from the first stage probit regression.

The first stage probit regression is specified as: RE = f(BTW35_85k, OVER85k, BTW35_50, OVER50, WHITE, SINGLE, CHILD, URGENT, FINDIF, EM, NH, AQUAINT)

(3.1)

RE is coded as one if the home was sold with broker assistance (without the seller trying to sell the home as a FSBO at any stage), and zero if the home was sold without the assistance of a broker at any stage. Seller demographics include income (BTW35_85k and OVER85k), age (BTW35_50, OVER50), and race (WHITE). Previous research has shown that higher income buyers are more 43

likely to utilize the services of a broker due to higher opportunity costs. The other demographic characteristics are included to control for the possible impact on broker choice, but the expected direction of the impacts are not clear. Single heads of household (SINGLE) could find it more challenging to take on the time and financial burden of selling a home without assistance as compared to married or unmarried couples. Similarly, the number of children under 18 in the household (CHILD) is included to capture the time constraints of the seller. Sellers with many young children may find it more challenging to dedicate the time necessary to sell a home without assistance and may be more likely to hire a broker. It is accepted that broker-marketed homes sell faster than FSBO properties, so sellers motivated to complete a transaction quickly may be more likely to engage a broker. URGENT captures the seller’s reported need for urgency in the home sale. Employer mandated moves (EM) tend to involve a set time frame, relocation assistance from the employer, and moving some non-trivial distance. These sellers may be more motivated and less price sensitive, thus making them more willing to pay a brokerage commission to sell the home sooner. On the one hand, those selling due to a change in family situation such as marriage or divorce (NH) or due to financial difficulty (FINDIF) may be more motivated to sell quickly and thus hire a broker. On the other hand, these buyers may not be willing pay a brokerage commission due to financial constraints and could be less likely to work with a broker. Since the seller’s primary purpose of engaging a broker is to help find a buyer, those that sell a home to an acquaintance or family member (AQUIANT) should be less likely to utilize a broker.

44

In the second stage, the natural log of the final selling price (lnSP) is modeled as a function of seller income, motivation to sell, relationship with the buyer, and housing characteristics. It also includes LAMBDA, the inverse Mills ratio from the first stage. Including it in the second stage allows one to test for selection bias and allows OLS to give consistent estimates if selection bias is detected. See Table 3.1 for a full list of variables with descriptions. The price regression is specified as: lnSP = f(RE, BTW35_85k, OVER85k, URGENT, FINDIF, AQUAINT, DETSFAM, SF, RURAL, URBAN, SUBURBAN, RESORT, lnTOM, LAMBDA)

(3.2)

Income is expected to be positively related to price. Those needing to sell urgently may be more likely to take price concessions in order to reduce marketing time, which would predict a negative relationship between URGENT and sale price. FINDIF may be positively or negatively related to price. On the one hand, those needing to sell a home due to financial difficulty may be motivated to take price concessions in order to sell the home quickly. On the other hand, homeowners facing financial difficulty may be willing to wait for a higher offer in order to cover as much of the remaining mortgage obligation as possible, while those that do not receive a high enough offer price opt for foreclosure. Only those sellers that received offers high enough to accept are observed in the sample, which may bias the coefficient on FINDIF upwards and give a positive relationship with final sales price. Selling a home to an acquaintance or family member is likely cheaper for the seller when compared to actively marketing the home, whether through a broker or not, thus AQUAINT should be negatively related to final purchase price. The number of housing characteristics included in both the 2006 and 2009 surveys include the type of property (condominium, detached single-family home, etc.), the square footage of the

45

home, and a description of the home’s location (rural, urban, resort, etc.). While a more detailed description of the home would be preferable in modeling purchase price, square footage and location are two of the largest determinates of home prices. Six variables capture the available housing characteristics, DETSFAM, SF, RURAL, URBAN, SUBURBAN, and RESORT. Detached single family homes tend to sell for more than other type of properties, so DETSFAM should be positively related to price, as should square footage (SF).5 Homes in urban and resort areas are expected to sell for higher prices than similar homes in rural or suburban areas. Small town is the omitted category. lnTOM is the natural log of the number of weeks the home was on the market before it sold. Past studies have found that sales price and TOM are jointly determined, but the direction of the relationship is not necessarily predictable. On the one hand, higher priced homes tend to stay on the market longer because the pool of potential buyers is smaller, which makes it more challenging to find a match. In this case, TOM and price would be positively correlated. On the other hand, properties that stay on the market for a long period of time may become stigmatized and require price reductions to induce a purchase. If this effect dominates, then TOM and price would appear negatively correlated.6 If LAMDA is significant in the second stage, there is evidence of selection bias and including LAMBA will make the second stage results unbiased and consistent. If LAMBDA is not significant, then there is no evidence of selection bias and the results from a standard OLS model are preferable.

5

Square footage is a categorical variable in the 2006 data and is a discrete integer in the 2009 data. The 2006 regressions use dummy variables for each square footage category and the 2009 regressions use lnSF (the natural log of square footage). 6 The treatment effect and OLS models were run without lnTOM and the results were qualitatively unhanged.

46

The presence of selection bias, or lack thereof, may imply a number of things. One possibility is that selection bias is present and brokers are associated with higher prices even after correcting for the bias. This would imply two separate housing markets, one for broker assisted sales and one for FSBO transactions. The price premium associated with broker marketed homes in this case would represent the predisposition of those selling more expensive homes to utilize a broker. It would also imply that competitive pressure from FSBO properties was not enough put an upper bound on broker-marketed home prices. A second possibility is that there is no evidence of selection bias, but brokers are still associated with higher home prices. This would imply that the higher home prices of broker marketed homes are a result of a value added service from the broker that the seller was able to pass along to the buyer in the form of higher prices. The final possibility is that brokers are not associated with different selling prices when compared to FSBO transactions. This would imply that competitive pressure from owner-marketed properties keeps an upper bound on the prices of broker-marketed homes and prevents two separate markets from developing. Two-stage Least Squares (2SLS) The final specification is a two-stage least squares model to correct for the endogeneity of TOM and sales price. In the first stage, lnTOM is estimated with the same explanatory variables from the structural price equation with additional instrumental variables. The generated regressor

is then included in the second stage price model.

47

Specifically, the first stage is: lnTOM = f(RE_S BTW35_85k, OVER85k, URGENT, FINDIF, AQUAINT, DETSFAM, SF, RURAL, URBAN, SUBURBAN, RESORT, EM, NH, REDUCE),

(3.4)

where EM, NH, and REDUCE are the instrumental variables. All three variables are highly correlated to lnTOM, but not correlated with the final sales price. Although one would expect a negative relationship between the number of price reductions (REDUCE) and sales price, it may be that those homes needing multiple price reductions were over-priced to begin with and so the final sales price is closer to the price of similar homes. Intuitively, those selling as the result of an employer mandated move would be motivated to sell quickly, but would not do so at the expense of taking a reduced price on the home since they would likely have financial assistance from the employer to relocate. Those selling due to the creation of a new household due to marriage or divorce may also be motivated to sell quickly, but there is no clear direction of the impact on sales price. The second stage is then: lnSP = f(RE, BTW35_85k, OVER85k, URGENT, FINDIF, AQUAINT, DETSFAM, SF, RURAL, URBAN, SUBURBAN, RESORT, where

,

(3.5)

is the generated regressor from the first stage.

Empirical Results Summary Statistics Despite the different market conditions between 2006 and 2009, the proportion of home sellers using real estate brokers remained relatively unchanged. Table 3.2 reports that, in both

48

years, just over 80 percent of sellers that completed the NAR survey utilized a broker. This is similar to past studies that find between 75 and 85 percent of all home sales transactions involve a real estate broker. Table 3.3 highlights a few of the key differences between 2006 and 2009. Average timeon-market (TOM) increased from 13.14 weeks to 17.67 weeks, while average sales price fell from $280,390 to $238,639. As one would expect, home stayed on the market longer and sold for less on average in 2009 when compared to the height of the residential real estate bubble in 2006. Table 3.4 reports the average and median selling prices for homes sold through a broker and homes sold in FSBO transactions. In 2006, the average selling price of a home sold through a broker was $283,431. The average sales price for FSBO transaction was $224,750. In 2009, the average price for a broker-assisted sale was $245,948 while the average price for a FSBO sale was $195,421. While the differences are suggestive, more rigorous methods are needed to detriment if they are meaningful.

49

Table 3.2. Statistics on Real Estate Broker Use How did you sell this 2006

2009

home? Count

%

Count

%

2,966

80.5

2,603

82.3

143

3.9

82

2.6

30

0.8

36

1.1

403

10.9

290

9.2

43

1.2

46

1.5

100

2.7

106

3.4

100

3,163

100

Sold it using a real estate agent/broker First tried to sell it myself, but then used an agent Sold it to a home buying company Sold it without ever using a real estate agent/broker First listed with an agent, but then sold it myself Other Total

3,685

50

Table 3.3. Summary Statistics on Time-on-Market (TOM) and Final Selling Price (All Transactions)

Avg. TOM in weeks (Std. dev.) Avg. selling price (Std. Dev.)

2006

2009

13.14

17.67

(34.81)

(20.98)

$280,390

$238,639

(307,015)

(200,293)

Table 3.4. Summary Statistics on Sales Price Broker/Agent

All FSBO

2006

2009

2006

2009

Average

$283,431

$245,948

$224,570

$195,421

Median

$212,750

$195,000

$168,800

$137,000

51

Treatment Effect Models The results of the first stage probit model on broker choice are reported in Table 3.5 (2006) and Table 3.6 (2009). The results paint an interesting picture of the housing market in these years. As expected, those that sold a home to an acquaintance were less likely to use a broker in both years. This is the only variable that matches sign and significance between the two years. In 2006, only sellers over 50 years of age and those needing to sell urgently were more likely to use a broker. Income is not significantly related to broker choice. It may be that in this extreme seller’s market, opportunity costs, at least in the form of income, were not an important consideration for sellers when choosing how to market a home. In many areas across the country demand was strong enough that homes were selling at or above asking price with short marketing times. Perhaps even high income sellers were more willing to attempt selling a home without broker assistance under these market conditions. Buyers over 50 may have been less comfortable using technology to market a home without assistance and so were more likely to engage a broker. In 2009, only higher income sellers and those between 35 and 50 years of age were more likely to work with a broker. Marketing times were longer and prices were less certain in 2009, which would tend to lead those with higher opportunity costs to work with a broker. Income is an obvious proxy for opportunity costs, but age may be as well. It is possible that those between 35 and 50 years of age are in a more demanding time in their careers and have a more difficult time taking off of work to attend to the various administrative tasks of marketing a home.

52

Table 3.5. First-Stage Probit Estimates on Choice of Broker. Dependent variable is RE. 2006 Variable CONSTANT

Coefficient 1.231***

Standard Error

P-Value

0.251

0.000

BTW35_85k

-0.131

0.180

0.468

OVER85k

-0.010

0.185

0.956

BTW35_50

0.126

0.094

0.181

OVER50

0.354**

0.122

0.044

WHITE

-0.115

0.130

0.375

SINGLE

0.174

0.120

0.147

CHILD

0.052

0.043

0.228

lnEARNERS

0.048

0.128

0.709

URGENT

0.288***

0.083

0.001

FINDIF

0.144

0.287

0.616

AQUAINT

-1.334***

0.121

0.000

EM

0.185

0.142

0.192

NH

-0.225**

0.114

0.048

Observations

2,464

Chi-

1,208

squared(27) Prob > Chi-squared

53

0.000

Table 3.6. First-Stage Probit Estimates on Choice of Broker. Dependent variable is RE. 2009 Variable

Coefficient

Standard Error

P-Value

CONSTANT

0.870***

0.278

0.002

BTW35_85k

0.274

0.191

0.152

OVER85k

0.480***

0.193

0.013

BTW35_50

0.245*

0.128

0.055

OVER50

0.195

0.138

0.158

WHITE

0.177

0.156

0.256

SINGLE

0.001

0.148

0.992

CHILD

0.014

0.053

0.796

lnEARNERS

-0.047

0.164

0.775

URGENT

0.098

0.109

0.367

FINDIF

0.325

0.215

0.131

AQUAINT

-1.654***

0.139

0.000

EM

0.014

0.132

0.916

NH

-0.059

0.144

0.682

Observations

1,771

Chi-

1,027

squared(18) Prob > Chi-squared

54

0.000

The second stage OLS results are reported in Table 3.7 (2006) and Table 3.8 (2009). LAMBDA is negative and significant at the one-percent level in the second stage in both years, suggesting strong evidence of selection bias. Even after controlling for selection bias, RE is positive and significant at the 1% level in both years. In both the seller’s market of 2006 and the buyer’s market of 2009, real estate brokers were associated with higher sales prices even after correcting for selection bias in the choice to work with a broker. This suggests two separate real estate markets: One for broker-marketed properties and one for FSBO properties. Although this is the first study using nation-wide data that finds evidence brokers are able to influence prices, the result is not surprising given the extreme market conditions in 2006 and 2009. Residential real estate market participants in 2006 faced tight housing supply, strong housing demand, and relatively easy financing. News stories from the time discussed, “…tales of waiting lists for unbuilt condos and bidding wars over humdrum three-bedroom colonials.” (CNN/Money, “Welcome to the dead zone”, May 5, 2006). In such a market home sellers may have a great deal of leverage over buyers and could more easily pass along broker commission costs in the form of higher prices. As an anecdotal example, consider a potential buyer in early 2006 who found a desirable home for sale. Assume that the buyer faced the extreme seller’s market conditions that prevailed over much of the country. If the buyer wanted that particular home, he had little choice but to pay the asking price, or even bid higher than the asking price. Otherwise he risked being outbid by another buyer.

55

Table 3.7. Second-Stage OLS Price Regressions with Selection Correction. Dependent variable is ln(purchase price). 2006 Variable

Coefficient

Standard Error

P-value

CONSTANT

10.275***

0.333

0.000

RE

0.098***

0.031

0.002

BTW35_85k

0.191***

0.056

0.001

OVER85k

0.456***

0.056

0.000

URGENT

-0.058**

0.031

0.056

FINDIF

0.246***

0.084

0.003

AQUAINT

0.113

0.125

0.368

DETSFAM

0.088***

0.031

0.005

RURAL

0.074

0.047

0.112

URBAN

0.221***

0.042

0.000

SUBURB

0.101***

0.036

0.004

RESORT

0.432***

0.104

0.000

lnTOM

-0.057***

0.010

0.000

LAMBDA

-0.368**

0.156

0.018

Observations

2,464

Chi-

1,208

squared(55) Prob > Chi-squared Square footage categories omitted to save space.

56

0.000

Table 3.8. Second-Stage OLS Price Regressions with Selection Correction. Dependent variable is ln(purchase price). 2009 Variable

Coefficient

Standard Error

P-value

CONSTANT

5.647***

0.368

0.000

RE

0.092***

0.318

0.004

BTW35_85k

0.268***

0.064

0.000

OVER85k

0.499***

0.065

0.000

URGENT

-0.050*

0.028

0.074

FINDIF

0.127**

0.050

0.011

AQUAINT

0.114

0.156

0.465

DETSFAM

0.147***

0.035

0.000

lnSF

0.737***

0.034

0.000

RURAL

-0.062*

0.036

0.082

URBAN

0.090**

0.036

0.014

SUBURB

0.004

0.042

0.930

RESORT

0.541***

0.139

0.000

lnTOM

-0.007

0.010

0.500

LAMBDA

-0.360**

0.156

0.021

Observations

1,771

Chi-

1,027

squared(18) Prob > Chi-squared

57

0.000

This anecdote does not address the issue of why there is price separation between brokerassisted and FSBO prices. Why were sellers not assisted by brokers unable capture a similar price premium? Although brokers are typically agents of the seller, they also assist buyers in the home search. On the one hand, brokers reduce buyer search costs and increase buyer welfare by granting access to a larger supply of homes through the MLS, helping negotiate an offer price, and helping arrange financing. This may induce buyers to pay higher prices for homes. On the other hand, brokers also reduce seller search costs and may induce sellers accept lower prices for homes (Jud and Frew (1986); Bagnoli and Khanna (1991); and Ford et al. (2005)). What if market conditions were such that sellers did not have to accept lower prices? What if supply was tight enough and demand was high enough that home sellers regularly entertained offers at or above the asking price? Buyers will still be incented to pay higher prices for a broker’s services, but sellers will have no incentive to accept lower prices. Instead, they can pass along the broker’s commission costs to the buyer in the form of higher prices. In other words, there may not have been enough competitive pressure from FSBO properties to place an upper bound on broker-marketed homes in 2006. Home buyers and sellers faced a much different market in 2009. Home prices had fallen significantly, interest rates were at historic lows, but credit was tight. Only well qualified buyers (good credit score, 20% down payment, etc.) could get mortgage financing, and those buyers were facing a volatile housing market. They could no longer be certain that home prices would increase after purchase as was the case over the previous decade. In fact, many markets continued to experience a downward trend in home values through 2011. Likewise, home sellers faced the daunting task of listing a home in an oversupplied market with weak demand and volatile prices. Even though the absolute information advantage brokers have today is less than

58

it was ten years ago thanks to the internet, any information advantage is valuable when facing such uncertainty. In such a market, some buyers and sellers may be more willing to pay for a broker’s market expertise to help insure that they are getting a fair price for a home, especially those with higher opportunity costs. In this case, sellers do not have the leverage over buyers that would allow them to pass along commission costs, but if buyers are willing to pay a premium for a broker’s market expertise, then there will still be a price premium associated with broker-assisted transactions. It may also be that the extreme seller’s market not only made certain buyers more likely to work with brokers, but also led brokers to be more selective in the types of homes they agreed to market. With marketing time increasing, home prices falling, and fewer homes being sold, brokers may have focused their efforts on homes that were the most likely to sell. If these homes tend to be higher priced, then brokers may be associated with higher prices not only because of a value added service they provide, but also because of the types of homes they choose to market. Most of the other variables have the expected sign in both years. Selling a home to an acquaintance (AQUAINT) tends to result in a lower sales price. The coefficient on FINDIF is positive and significant in the OLS regressions (and in every other model specification). Sellers that report wanting to sell the home mainly due to financial difficulty tend to receive higher prices, all else being equal. Although the average sales price is nearly $100,000 lower for buyers reporting financial difficulty in both years, the average TOM is six weeks higher in both years. Perhaps homeowners facing financial difficulty are willing to wait for a higher offer in order to cover as much of the remaining mortgage obligation as possible, while those that do not receive a high enough offer price opt for foreclosure. Only those sellers that received offers high enough to accept are observed in the sample, which may bias the coefficient on FINDIF upwards.

59

The square footage of the home is reported in the following categories in the 2006 survey: The lowest category is 1 to 500 square feet. The next highest category is 501 to 1000 square feet (SF501_1000), the next highest is 1,001 to 1,500 (SF1001_1500), and so on in 500 square foot increments. The final category listed in the survey is over 5,000 square feet (SFOVER5000). The model is estimated with each category as a dummy variable. The lowest category, 1 to 500 square feet, is omitted. The coefficient grows in magnitude and gains significance with each larger square footage category, as expected. Square footage is reported as an integer in the 2009 survey (lnSF) and is positively related to sales price. Homes in urban and resort areas tend to have higher selling prices than homes in suburban, rural, or small town areas. 2SLS Models Another potential weakness of using OLS to model home prices is the endogeneity of sales price and TOM. There are multiple ways that sales price and TOM may be simultaneous determined. All other things being equal, homes with higher prices have a smaller pool of potential buyers and may stay on the market longer than less expensive homes with a larger pool of potential buyers. This would lead to a positive relationship between sales price and TOM. The relationship could also run the opposite direction. For example, an impatient seller may choose to list a house below fair market value to attract a buyer faster, while a patient seller may choose to list a house above market value and wait for a buyer that meets his reservation price. It may also be true that homes that remain unsold for long periods become stigmatized and require price reductions to sell. These examples would lead to a negative relationship between sales price and TOM. Which effect dominates is an empirical question, but it is clear that price and TOM might be simultaneously determined. If so, the results from OLS are biased. Testing for endogeneity 60

revealed that the error term from the TOM model (equation 4) is significant in the structural price model (equation 5) at the one percent level in 2006, but is not significant in 2009. This suggests that price and TOM were simultaneously determined in 2006 (or there is an omitted variable problem) and a two-stage least squares model is appropriate. In 2009 there is no evidence of endogeneity, indicating that OLS is appropriate. The results from the second stage of the 2SLS models are reported in Table 3.9 (2006) and Table 3.10 (2009). As with all other model specifications, the coefficient on RE is positive and significant, indicating that brokers were associated with higher sales prices.

61

Table 3.9. 2SLS Price Regressions. lnTOM instrumented with EM, NH, and REDUCE. Dependent variable is ln(purchase price). 2006 Variable

Coefficient

Standard Error

P-value

CONSTANT

10.835***

0.189

0.000

RE

0.245***

0.044

0.000

BTW35_85k

0.129***

0.048

0.008

OVER85k

0.393***

0.049

0.000

URGENT

-0.051**

0.026

0.049

FINDIF

0.253***

0.075

0.001

AQUAINT

-0.123**

0.054

0.024

DETSFAM

0.100***

0.030

0.001

RURAL

0.049

0.046

0.289

URBAN

0.228***

0.041

0.000

SUBURB

0.101***

0.035

0.004

RESORT

0.454***

0.093

0.000

-0.010

0.022

0.657

Observations

2,655

F(22, 2,632)

52.59

Adj. R-squared

0.3013

Prob > F

0.000

62

Table 3.10. 2SLS Price Regressions. lnTOM instrumented with EM, NH, and REDUCE. Dependent variable is ln(purchase price). 2009 Variable CONSTANT

5.984***

0.217

0.000

RE

0.222***

0.046

0.000

BTW35_85k

0.239***

0.047

0.000

OVER85k

0.482***

0.047

0.000

URGENT

-0.039*

0.023

0.087

FINDIF

0.188***

0.040

0.000

AQUAINT

-0.117*

0.052

0.025

DETSFAM

0.124***

0.030

0.000

lnSF

0.773***

0.030

0.000

RURAL

-0.071**

0.031

0.022

URBAN

0.106***

0.032

0.001

SUBURB

-0.015

0.037

0.693

RESORT

0.364***

0.096

0.000

0.013

0.05

0.383

Observations

2,316

F(13, 2,302)

97.47

Adj. R-squared

0.3500

Prob > F

0.000

63

Conclusion This study exploits a unique national dataset collected by the Nation Association of REALTORS® to compare the performance of the real estate brokerage market over different time periods and in different market conditions with the goal of determining to what extent earlier findings are time-dependent. In both the seller’s market of 2006 and the buyer’s market of 2009, brokers had an upward influence the price of housing, even after controlling for selection bias in the choice to use a broker and the endogeneity of sales price and TOM. The results stand in contrast with previous research using nation-wide data. Some evidence suggests that the extreme market conditions allowed brokers to influence prices in each year. Two thousand and six was such a strong seller’s market due to tight housing supply, strong demand, and relatively easy financing that home sellers could more easily have passed along brokerage commission costs to buyers in the form of higher prices. The presence of selection bias in 2006 indicates that even after controlling for the types of buyers that tend to work with brokers, broker-marketed properties still commanded a price premium over FSBO properties. This suggests two separate real estate markets in 2006: One for broker-marketed properties and one for FSBO properties. Competitive pressure from FSBO homes was not enough to keep home sellers from passing along broker commission costs to buyers in the form of higher prices. Housing market participants in 2009 were faced with an abundant supply of homes for sale, falling prices, weak demand, and strict underwriting standards for loans. In such an uncertain market, all classes of buyers and sellers may be more willing to pay for a broker’s market expertise to help insure that they are getting a fair price for a home. There is no evidence of selection bias in 2009, suggesting that brokers were associated with higher home prices not

64

because of the type of sellers they tend to work with (higher income, for example), but rather because they provided some value-added service that all classes of market participants were willing to pay for in the form of higher prices. If market conditions are driving the results, then the price differential between brokermarketed properties and owner-marketed properties should mitigate after the housing market returns to normal. Future research is needed to determine if the differential continues once home prices and housing supply stabilize. References Anglin, P. M., R. Rutherford, T. M. Springer (2003). “The trade-off between the selling price of residential properties and time-on-the-market: The impact of price setting,” The Journal of Real Estate Finance and Economics, 26(1), 95-111. Bagnoli, M. and N. Khanna (1991). “Buyers’ and Sellers’ Agents in the Housing Market,” Journal of Real Estate Finance and Economics, 4, 147-156. Baryla, E.A., L. Zumpano, and W. Elder (2000). “An Investigation of Buyer Search in the Residential Real Estate Market Under Different Market Conditions,” Journal of Real Estate Research 20, 1/2, 75-91. Bernheim, D. and J. Meer (2008). “How Much Value do Real Estate Brokers Add: A Case Study,” National Bureau of Economic Research. Benjamin, J. D., G.D. Jud, and G.S. Sirmans (2000). “What Do We Know About Real Estate Brokerage?” Journal of Real Estate Research 20(1), 5-30. Colwell, P.F., D.P. Lauschke, and A. Yavas (1992). The value of real estate marketing systems: Theory and evidence.” University of Illinios (mimeograph). Elder, W., L. Zumpano, and E. Baryla (2000). “Buyer Brokers: Do They Make a Difference?” Real Estate Economics 28(2), 337-362. Ford, S. F., R.C. Rutherford, A. Yavas (2005). “The effects of the internet on marketing residential real estate,” Journal of Housing Economics. 14, 92-108. Glower, M., D. R. Haurin, and P. H. Hendershott (1998). “Selling time and selling price: The influence of seller motivation,” Real estate economics, 26(4), 719-740.

65

Huang, B., R. Rutherford (2007). “Who You Going to Call? Performance of Realtors and Non-realtors in a MLS Setting,” Journal of Real Estate and Financial Economics, 35, 77-93. Hendel, I., A. Nevo, and F. Ortalo-Magne (2007). “The relative performance of real estate marketing platforms: MLS versus FSBOMadison.com” (No. w13360). National Bureau of Economic Research. Johnson, K.H., T.H. Springer, and C.M. Brockman (2005). “Price Effects of NonTraditionally Broker-Marketed Properties,” The Journal of Real Estate Finance and Economics 31, 3, 331-343. Jud, G.D. (1983). “Real Estate Brokers and the Market for Residential Housing,” AREUEA Journal 11, 69-82. Jud, G.D., and J. Frew (1986). “Real Estate Brokers, Housing Prices, and the Demand for Residential Housing,” Urban Studies 23, 21-31. Kamath, R. and K. Yantek (1982). “The influence of brokerage commissions on prices of single-family homes,” Appraisal Journal 50, 63–70. Kang, H. B. and M. J. Gardner (1989). “Selling price and marketing time in the residential real estate market,” Journal of Real Estate Research, 4(1), 21-35. Levitt, S. D., and C. Syverson (2008). “Market distortions when agents are better informed: The value of information in real estate transactions,” The Review of Economics and Statistics 90(4), 599-611. Miller, N. G. (1978). “Time on the market and selling price” Real Estate Economics, 6(2), 164-174. Rutherford, R.C., T.M. Springer, and A. Yavas (2005). “Conflicts Between Principals and Agents: Evidence from Residential Brokerage,” Journal of Financial Economics 76, 627-665. Tully, Shawn (2006, May 5). “Welcome to the dead zone,” CNNMoney. Retrieved from http://money.cnn.com/2006/05/03/news/economy/realestateguide_fortune Yinger, J. (1981). “A search model of real estate broker behavior,” American Economic Review 71, 591–605. Zietz, E. and G. Sirmans (2011). “Real Estate Brokerage in the New Millennium, Journal of Real Estate Literature 19, 5-40.

66

Zumpano, L., W. Elder, and E. Baryla (1996). “Buying a House and the Decision to Use a Real Estate Broker,” Journal of Real Estate Finance and Economics 13, 169181. Zumpano, L., and D. Hooks (1988). “The real estate brokerage market: A critical reevaluation,” AREUEA Journal, 16, 1–16.

67

CHAPTER 4 REITS IN A MIXED ASSET PORTFOLIO: AN INVESTIGATION OF EXTREME RISKS

Introduction and Literature Review Over the last decade the academic literature has consistently suggested that adding real estate to a mixed-asset portfolio offers diversification benefits and yields superior risk adjusted returns (e.g. Ibbotson and Siegel (1984), Goetzmann and Ibbotson (1990), Norman et al. (1995), and Quan and Titman (1999)). REITs have become an attractive way to add real estate exposure to a portfolio because of their historically high returns and high dividend payouts, but evidence of the link between REIT returns and physical real estate returns has been mixed. Many studies have found that REITs do offer diversification benefits similar to physical real estate (e.g. Giliberto (1990), Ghosh et al. (1996), Glascock et al. (2004), Lee and Stevenson (2005)), while others have found evidence that REIT returns can be highly correlated to stock returns and may not offer the same diversification benefits of physical real estate (e.g. Chan et al. (1990), Gyourkoand Nelling (1996), Glascock et al. (2000), Hung et al. (2008)). In many of these studies, the results changed based on the type of REIT examined and the time period under consideration. The recent financial crisis led to a resurgence of studies examining the diversification benefits of REITs in times of financial distress. The recent popularity of REIT research is not surprising given that the latest financial crises was rooted in real estate and that REITs have become increasingly important in the last twenty years. The National Association of Real Estate 68

Investment Trusts (NAREIT) reports that the market capitalization of U.S. REITs grew from $11.7 billion in 1989 to over $389 billion in 2012. Despite the recent surge in REIT research, few studies have investigated the extreme downside risk of REITs. One measure of downside risk, Value-at-Risk (VaR), has become the industry standard measure of a portfolio’s market risk used by investors and regulators. Regulators and investors view effective risk management against extreme events as essential to successful investing and financial system stability. For example, the Basle Committee on Banking Supervision requires banks to set aside sufficient risk capital to cover expected losses on the banks’ trading portfolio over a ten day holding period with 99% confidence. VaR has been comprehensively studied in the stock market (e.g. Bao et al. (2006), Kuester et al. (2006), and Manganelli and Engle (2004)), but only a handful of recent paper have applied VaR to REITs. In one such recent paper, Anderson, Boney, and Guirguis (2012) investigate the impact of switching regimes and monetary shocks on REITs. They find that unexpected monetary shocks impact REITs about twice as much as the general equities market under high-variance regimes. In another study, Zhou and Anderson (2011b) use three different VaR methods (nonparametric, parametric, and semiparametric) in nine international REIT markets to see which method is most accurate. They then ask whether the optimal VaR method for REITs is the same for stocks. The authors find that there is no universally adequate method to model the extreme risks of REIT returns across all nine countries. They also find that the optimal method for estimating extreme risks could be different between stocks and bonds. The final, and most concerning result is that the extreme risks for REITs are generally higher than those of stock markets, and the timing of extreme market movements between REITs and stock indices is almost perfectly in-sync. The authors conclude that diversification benefits of REITs are not present when needed most.

69

In another recent paper, Zhou and Anderson (2011a) use quantile regression and find that REITs exhibit strong herding behavior. Herding is described as the tendency for investors to follow the actions of others rather than their own beliefs. Zhou and Anderson find that the distribution of returns significantly tightens in market downturns as traders start moving in the same direction. Liow (2008) examines how the 1997-98 Asian financial crises influenced the VaR dynamics in selected international REIT markets. He found that extreme risks in Asian real estate returns are greater than those in European and North American markets. Furthermore, he finds that REIT returns are riskier than the corresponding broader stock indices in each of the seven counties considered. The recent papers do not question the long-run diversification benefits of REITs, but they do suggest that REITs may be harmful to a mixed-asset portfolio during times of market distress. Although the previous studies are suggestive, no direct test has yet been done to explore what effects, if any, REITs have on the extreme risks of a mixed-asset portfolio. This paper attempts to close that knowledge gap with two contributions. First, it will compare the VaR of a series of mixed-asset portfolios with and without REITs to test whether REITs increase the extreme downside risk of an otherwise identical portfolio. Second, this paper uses a new, more accurate method to calculate VaR: conditional autoregressive value at risk (CAViaR), developed by Engle and Manganelli (2004). CAViaR offers some technical advantages over other VaR estimation methods. Traditional VaR methods model the whole return distribution and then assume that the same distribution applies to the extreme tails. The herding evidence in Zhou and Anderson (2011a) suggests that REIT returns cluster and that the process governing returns in the extreme tails is different than elsewhere in the distribution. Assuming that the process is the same across the entire return distribution can

70

lead to inaccurate estimates of VaR, which by definition is concerned with the extreme left tail of the distribution. Lu et al. (2009) present evidence that traditional VaR methods are in fact problematic for REITs. The authors employ five popular methods to calculate the VaR of twelve REIT portfolios and evaluate the accuracy of each method. They find that each method performs differently at different confidence levels and no method dominates the others. Zhou and Anderson (2011b) found similar evidence applying three different popular VaR methods to REIT portfolios in nine different countries. The empirical fact that financial returns tend to cluster means that they are autocorrelated, suggesting an autoregressive process is appropriate to model returns. Since VaR is tightly linked to the standard deviation of returns, it must also be autocorrelated. CAViaR specifies the evolution of the quantiles over time with an autoregressive process that directly models each quantile (see Koenker and Bassett (1978) for a discussion of quantile regression). Modeling each quantile directly avoids the problem of imposing the full return distribution on every quantile. CAViaR also avoids another source of VaR model miss-specification since it does not require any distribution assumptions for the error terms of returns. Many traditional VaR methods require the error terms of returns be iid, if not normal. Engle and Manganelli (2004) find that CAViaR leads to highly precise VaR estimates for stocks. So far, only one study has applied CAViaR to REITs. Zhou (2011) finds that CAViaR leads to highly precise VaR estimates at both the 95% and 99% confidence levels. This paper extends the emerging literature with two key findings. First, adding REITs to a mixed-asset portfolio does not have a significant impact on the average daily return or VaR of a portfolio before 2006. After 2006, however, adding REITs to a portfolio of stocks and bonds significantly increases VaR. For a portfolio that already contains REITs, adding additional

71

weight to REITs further increases the VaR of the portfolio. Second, REITs have a greater impact than small cap stocks on the VaR of an otherwise identical portfolio after 2005. Although REITs and small cap stocks may share common drivers, REITs had greater extreme risks during the recent financial crises. The rest of the paper is organized a follows: the following section outlines the econometric method and data. Section three presents the results and section four concludes. Data and Methodology Estimating VaR Value-at-risk (VaR) is defined as the maximum expected loss of a portfolio over a given holding period at a specified confidence level. VaR has become popular with risk managers and regulators due to its simplicity. It summarizes in a single number the potential change in the value of a portfolio resulting from a severe downside risk exposure. VaR is the θ th -quantile of the conditional probability distribution of the financial returns: (4.1)

P (Y t < −V t I t −1 ) = θ

where Yt is the return of interest, Vt is the VaR, It −1 is the information set up to time t-1. In practice, θ is set to be a small percentage like 5% or 1% to represent extreme downside risk. The following is a brief description of the CAViaR model. Please refer to Engle and Manganelli (2004) for more details. A generic CAViaR model takes the following form: q

r

f t ( β ) = β 0 + ∑ β i f t −i ( β ) + ∑ β j l ( X t − j ) i =1

(4.2)

j =1

where ft (β) ≡ ft (Xt−1, β) denotes the time t θ th -quantile of the financial return distribution formed at time t-1, β is the model parameters, Xt is a vector of time t observable variables, and l

72

is a function of a finite number of lagged values of Xt . Autoregressive terms β i f t −i ( β ) , i = 1,..., q enter the specification. This is inspired by the empirical fact that volatilities of

financial returns tend to cluster over time. VaR, which is tightly linked to the standard deviation of the return distribution, is expected to exhibit similar behavior. The role of the term l ( X t − j ) is to link f t ( β ) to observable variables that belong to the information set. A natural candidate for X t − j is lagged returns. Given the generic form (2), Engle and Manganelli (2004) outlined four

specific specifications. Three of the specifications impose a symmetric VaR impact from positive and negative shocks. The herding evidence in REITs during down markets from Zhou and Anderson (2011) suggests that VaR responds differently to negative shocks than it does to positive shocks. Only one CAViaR specification allows positive and negative shocks to impact VaR asymmetrically: Asymmetric slope CAViaR: f t ( β ) = β1 + β 2 f t −1 ( β ) + β 3 ( y t −1 ) + + β 4 ( y t −1 ) −

(4.3)

where ( y t −1 ) + = max( y t −1 ,0 ) , and ( y t −1 ) − = − min( y t −1 ,0 ) Asymmetric slope CAViaR allows the current VaR estimate to vary based on information from the lagged VaR (β2), past positive return shocks (β3), and past negative return shocks (β4). Zhou (2011) finds that asymmetric slope CAViaR led to the most precise VaR estimates at both the 95% and 99% confidence levels. No other VaR method has been found to perform consistently well at differing confidence intervals. Data This study creates five portfolios using daily U.S. stock, bond, and REIT returns from January 1, 1989 to December 31, 2010. Stock returns are from CRSP’s value-weighted index

73

return on the NYSE, AMEX, and Nasdaq. Bond returns are taken from Barclays Aggregated U.S. Bond index return, and the REIT returns are from the CRSP/ZIMAN value-weighted return index. There are 5,547 observations for each of the five portfolios created. The baseline portfolio, labeled StockBond, is composed of 60% equity and 40% bonds. The next two portfolios remove equal weights from stocks and bonds and replace them with REITs. The first, labeled REIT10, removes 5% equities, 5% bonds, and adds 10% REITs. The new portfolio balance is 55% equities, 35% bonds, and 10% REITs. The next portfolio, REIT20, removes 10% stock and 10% bonds from the baseline portfolio and adds 20% REITs for a balance of 45% equities, 35% bonds and 20% REITs. For the final two portfolios, the weight in REITs is replaced with small cap stocks7 to test if these stocks have similar effects on the VaR of an otherwise identical portfolio. Empirical Results Summary Statistics Summary statistics are reported in Table 4.1. The average daily returns for each of the five portfolios reported in the first five rows are similar, ranging from a low of 0.0375% for the StockBond portfolio, to a high of 0.0398% on the REIT20 portfolio. The median daily returns are significantly higher than the average returns, suggesting a left-skewed distribution. The negative skewness results confirm that the return distributions are left-skewed. The large kurtosis figures indicate that they are also fat-tailed. Such a distribution means that risk managers and regulators would benefit from a more precise extreme risk measurement like CAViaR.

7

Small cap stocks are as defined by the French website

74

VARIABLES

Table 4.1: Summary Statistics (1) (2) (3) (4) (5) Mean Median SD Skewness Kurtosis

StockBond 0.0375 0.0612 0.684 -0.145 11.46 REIT10 0.0386 0.0626 0.734 -0.161 14.14 REIT20 0.0398 0.0611 0.801 -0.121 17.43 SmallCap10 0.0385 0.0704 0.726 -0.230 11.55 SmallCap20 0.0396 0.0777 0.772 -0.296 11.57 VaRStockBond 1.539 1.348 0.680 3.231 19.61 VaRREIT10 1.586 1.348 0.838 3.626 22.16 VaRREIT20 1.644 1.358 1.008 3.853 22.97 VaRSmallCap10 1.633 1.407 0.808 3.228 19.34 VaRSmallCap20 1.738 1.488 0.889 3.184 18.58 VaRStockBond_05 0.995 0.854 0.503 3.159 18.87 VaRREIT10_05 1.022 0.860 0.584 3.546 21.89 VaRREIT20_05 1.056 0.857 0.701 3.822 23.42 VaRSmallCap10_05 1.056 0.898 0.558 3.156 18.66 VaRSmallCap20_05 1.134 0.965 0.602 3.146 18.28 This table reports the summary statistics of variables from 1/3/1989 to 12/31/2010. StockBond refers the daily return of portfolio with 60% stock and 40% bonds; REIT10 refers the daily return of portfolio with 55% stock, 35% bonds and 10% REITs; REIT20 refers the daily return of portfolio with 50% stock, 30% bonds and 20% REITs; SmallCap10 refers the daily return of portfolio with 55% stock, 35% bonds and 10% small cap. stocks; SmallCap20 refers the daily return of portfolio with 50% stock, 30% bonds and 20% small cap. stocks; The prefix VaR represents the value at risk of each variable calculated under 99% cutoff; The suffix _05 represents the value at risk of each variable calculated under 95% cutoff. VaR is reported as a percentage of portfolio value. Skewness and Kurtosis test shows the distributions for all variables distributions differ from normality at 1% significance level.

The VaR for each portfolio is found using a 95% and a 99% confidence interval. In each case, the portfolios containing only stocks and bonds have the lowest average VaR. Adding either REITs or small cap stocks to the baseline portfolio increases VaR using both a 95% and a 99% cutoff. Adding more weight to REITs or small caps further increases VaR.

75

The initial results reinforce the emerging literature on the extreme downside risks of REITs and highlight the need for a more precise VaR estimate. The next section presents a more rigorous empirical analysis. Regression Results Given the skewness and kurtosis of the return distributions, a Wilcoxon signed-rank test is used to compare the average return and VaR of the different portfolios. Table 4.2 reports the Z score and p value for paired variables on the daily return and VaR at the 99% and 95% cutoff over the entire sample period. The first noteworthy result is that the average daily returns are not significantly different between the baseline portfolio and the two REIT portfolios. The only significant VaR difference is between the baseline portfolio and the REIT20 portfolio at the 99% cutoff. Despite the suggestive differences in the summary statistics, adding REITs to the baseline StockBond portfolio does not significantly alter the average return or VaR over the entire sample period. The story changes when comparing the portfolios containing small cap stocks. The difference in average daily return and VaR between the small cap portfolios and each of the other portfolios is significant in all but a few cases. While small cap stocks and REITs may share common drivers, it appears that each impacts the returns and VaR of a portfolio in a different way. While adding REITs to a mixed-asset portfolio may not change the VaR when looking at the entire sample period, the emerging literature offers considerable evidence that REITs may increase VaR in times of financial distress. To see if this is the case, the sample is split in two using the last trading day of 2005 as a break point. Although the point is chosen somewhat arbitrarily, experimentation using different points does not qualitatively change the results.

76

Table 4.2: Wilcoxon Signed-Rank Test Daily Return 99% cutoff 95% cutoff Z Prob > |z| Z Prob > |z| Z Prob > z| StockBond REIT10 -0.689 0.4909 0.507 0.6121 -1.162 0.2451 StockBond REIT20 -0.689 0.4909 -2.691 0.0071 1.002 0.3165 StockBond SmallCap10 -3.362 0.0008 -48.46 0.0000 -56.994 0.0000 StockBond SmallCap20 -3.362 0.0008 -59.207 0.0000 -63.224 0.0000 REIT10 REIT20 -0.689 0.4909 -6.469 0.0000 0.52 0.6033 REIT10 SmallCap10 -2.054 0.0400 -34.411 0.0000 -43.557 0.0000 REIT10 SmallCap20 -3.196 0.0014 -59.031 0.0000 -62.061 0.0000 REIT20 SmallCap10 -0.588 0.5562 -14.608 0.0000 -22.137 0.0000 REIT20 SmallCap20 -2.043 0.0410 -37.922 0.0000 -45.37 0.0000 SmallCap10 SmallCap20 3.362 0.0008 -62.399 0.0000 -64.21 0.0000 This table presents the Wilcoxon signed-rank test Z value and p value for paired variables in terms of daily return, VaR at 99% cutoff and VaR at 95% cutoff. The data sample spans from 1/3/1989 to 12/31/2010. StockBond refers the daily return of portfolio with 60% stock and 40% bonds; REIT10 refers the daily return of portfolio with 55% stock, 35% bonds and 10% REITs; REIT20 refers the daily return of portfolio with 50% stock, 30% bonds and 20% REITs; SmallCap10 refers the daily return of portfolio with 55% stock, 35% bonds and 10% small cap. stocks; SmallCap20 refers the daily return of portfolio with 50% stock, 30% bonds and 20% small cap. stocks.

77

Panel A of Table 4.3 reports the differences between portfolio VaR means within each sub-period. For brevity the discussion is focused on the VaR under the 99% cutoff. The conclusions drawn from the 95% cutoff are similar. The results are striking. Before 2006, the VaR of REIT10 (REIT20) is smaller than that of StockBond by 2.59 (4.88) bases points. However, after 2005 the VaR of REIT10 (REIT20) is 29.87 (63.24) bases points larger than that of StockBond. In fact, the two REIT portfolios have the highest average VaR when compared to every portfolio in the second period, with the exception of REIT10 versus SmallCap20. Panel B of Table 4.3 reports the difference in VaRs of the portfolios between the pre2006 and the post-2005 time periods using both the 99% and 95% cutoff. Again the discussion is focused on the results from the 99% cutoff. The results from the 95% cutoff are qualitatively similar. Not surprisingly, VaR increased for every portfolio in the post-2005 time period. The baseline portfolio experienced the smallest increase in VaR with a 39 basis point rise from the pre-2006 period to the post-2005 period. The two portfolios with REIT exposure experienced the largest increase in VaR. The VaR of REIT10 increased 71 basis points, from 1.42% to 2.14%. The difference is even greater as the weight in REITs increases. The VaR of REIT20 increased 107 basis points, from 1.40% to 2.47%. The SmallCap portfolios also experienced larger increases in VaR than the baseline portfolio, but the differences are not as dramatic as the REIT portfolios. All differences are significant at the one percent level Taken together, the results do not dispute the long-run benefits of REITs, but they do raise questions about the role of REITs in a mixed-asset portfolio in times of financial crises. In fact, prior to 2006, adding REITs to a portfolio of stocks and bonds does not appear to significantly alter the average daily return, but may slightly reduce VaR. It is not until 2006 and after that adding REITs to a portfolio of stocks and bonds significantly increases VaR.

78

Panel A

VaRStockBond VaRStockBond VaRStockBond VaRStockBond VaRREIT10 VaRREIT10 VaRREIT10 VaRREIT20 VaRREIT20 VaRSmallCap10

VaRREIT10 VaRREIT20 VaRSmallCap10 VaRSmallCap20 VaRREIT20 VaRSmallCap10 VaRSmallCap20 VaRSmallCap10 VaRSmallCap20 VaRSmallCap20

VaRStockBond VaRStockBond VaRStockBond VaRStockBond VaRREIT10 VaRREIT10 VaRREIT10 VaRREIT20 VaRREIT20 VaRSmallCap10

VaRREIT10 VaRREIT20 VaRSmallCap10 VaRSmallCap20 VaRREIT20 VaRSmallCap10 VaRSmallCap20 VaRSmallCap10 VaRSmallCap20 VaRSmallCap20

Table 4.3: Comparison of VaR Means 99% cutoff 1/3/1989-12/30/2005 1/3/2006-12/31/2010 1st Col. 2nd Col. 1st Col. 2nd Col. Mean Mean Difference Mean Mean Difference 1.4507 1.4248 0.0259 1.8377 2.1364 -0.2987 1.4507 1.4018 0.0488 1.8377 2.4702 -0.6324 1.4507 1.5095 -0.0588 1.8377 2.0531 -0.2153 1.4507 1.5817 -0.1311 1.8377 2.2683 -0.4305 1.4248 1.4018 0.0230 2.1364 2.4702 -0.3338 1.4248 1.5095 -0.0847 2.1364 2.0531 0.0834 1.4248 1.5817 -0.1569 2.1364 2.2683 -0.1319 1.4018 1.5095 -0.1077 2.4702 2.0531 0.4171 1.4018 1.5817 -0.1799 2.4702 2.2683 0.2019 1.5095 1.5817 -0.0723 2.0531 2.2683 -0.2152 95% cutoff 1/3/1989-12/30/2005 1/3/2006-12/31/2010 1st Col. 2nd Col. 1st Col. 2nd Col. Mean Mean Difference Mean Mean Difference 0.9292 0.9144 0.0148 1.2177 1.3887 -0.1710 0.9292 0.8958 0.0333 1.2177 1.6017 -0.3841 0.9292 0.9715 -0.0424 1.2177 1.3450 -0.1274 0.9292 1.0294 -0.1003 1.2177 1.4903 -0.2726 0.9144 0.8958 0.0185 1.3887 1.6017 -0.2131 0.9144 0.9715 -0.0571 1.3887 1.3450 0.0436 0.9144 1.0294 -0.1150 1.3887 1.4903 -0.1016 0.8958 0.9715 -0.0757 1.6017 1.3450 0.2567 0.8958 1.0294 -0.1336 1.6017 1.4903 0.1114 0.9715 1.0294 -0.0579 1.3450 1.4903 -0.1453

79

Panel B

99% cutoff 95% cutoff 1/3/19891/3/20061/3/19891/3/200612/30/2005 12/31/2010 Difference 12/30/2005 12/31/2010 Difference VaRStockBond 1.4507 1.8377 0.3871 0.9292 1.2177 0.2885 VaRREIT10 1.4248 2.1364 0.7116 0.9144 1.3887 0.4743 VaRREIT20 1.4018 2.4702 1.0684 0.8958 1.6017 0.7059 VaRSmallCap10 1.5095 2.0531 0.5436 0.9715 1.3450 0.3735 VaRSmallCap20 1.5817 2.2683 0.6865 1.0294 1.4903 0.4609 This table compares the mean VaR in different sample periods. StockBond refers the daily return of portfolio with 60% stock and 40% bonds; REIT10 refers the daily return of portfolio with 55% stock, 35% bonds and 10% REITs; REIT20 refers the daily return of portfolio with 50% stock, 30% bonds and 20% REITs; SmallCap10 refers the daily return of portfolio with 55% stock, 35% bonds and 10% small cap. stocks; SmallCap20 refers the daily return of portfolio with 50% stock, 30% bonds and 20% small cap. stocks; The prefix VaR represents the value at risk of each variable. VaR is reported as a percentage of portfolio value

80

Conclusion Conventional thinking suggests that adding real estate exposure to a mixed-asset portfolio should expand the efficient frontier and offer superior risk-adjusted returns. Over the last twenty years, REITs had become an increasingly popular way for investors to add real estate exposure. After the recent financial crisis the literature has reexamined the diversification benefits of REITs in times of financial distress. Researchers have found that during these times, REIT returns tend to be highly correlated with equity returns and often experience greater extreme risks. The findings cast doubt on the ability of REITs to stabilize portfolio returns in volatile markets. The results from this study confirm what the emerging literature suggests. Prior to 2006, adding REITs to a portfolio of stock and bonds does not have a significant impact on average daily returns, but the addition may reduce the portfolio VaR. After 2005 on the other hand, adding REITs to the same portfolio significantly increases VaR. Far from helping stabilize portfolio returns, REITs actually increase the extreme risks of the portfolio in the post-financial crises period. References Anderson, R. I., V. Boney, and H. Guirguis (2012). “The impact of switching regimes and monetary shocks: An empirical analysis of REITs.” Journal of Real Estate Research, 34(2), 157-181. Anderson, R., J. Clayton, and G. MacKinnon (2005). “REIT Returns and Pricing: The Small Cap Value Factor.” Forthcoming, Journal of Property Research 22(4), 267-286. Bao Y.,T-H Lee, and B. Saltoglu. (2006) Evaluating predictive performance of Value-at-Risk models in emerging markets: a reality check, Journal of Forecasting, 25, 101-28. Chan, K. C., P. H. Hendershott, and A. B. Sanders (1990). “Risk and return on real estate: evidence from equity REITs.” Real Estate Economics, 18(4), 431-452. Engle, R. F. and S. Manganelli (2004). "CAViaR: Conditional autoregressive value at risk by regression quantiles." Journal of Business & Economic Statistics 20(4), 367-381. 81

Glascock, J. L., D. Michayluk, and K. Neuhauser (2004). “The riskiness of REITs surrounding the October 1997 stock market decline.” The Journal of Real Estate Finance and Economics, 28(4), 339-354. Glascock, J. L., C. Lu, and R. W. So (2000). “Further evidence on the integration of REIT, bond, and stock returns.” The Journal of Real Estate Finance and Economics, 20(2), 177-194. Giliberto, M. S. (1990). “Equity real estate investment trusts and real estate returns.” Journal of Real Estate Research, 5(2), 259-263. Goetzmann, W. N. and R. G. Ibbotson (1990). “The performance of real estate as an asset class.” Journal of Applied Corporate Finance, 3(1), 65-76. Ghosh, C., M. Miles, and C. F. Sirmans (1996). “Are REITS stocks?.” Real Estate Finance, 13, 46-53. Gyourko, J. and E. Nelling (1996). “Systematic risk and diversification in the equity REIT market.” Real Estate Economics, 24(4), 493-515. Hung, K., Z. Onayev, and C. C. Tu (2008). “Time-Varying Diversification Effect of Real Estate in Institutional Portfolios: When Alternative Assets Are Considered.” Journal of Real Estate Portfolio Management, 14(4), 241-262. Ibbotson, R. G. and L. B. Siegel (1984). “Real estate returns: a comparison with other investments.” Real Estate Economics, 12(3), 219-242. Koenker, R. and G. Bassett Jr (1978). "Regression quantiles." Econometrica: journal of the Econometric Society 33-50. Kuester, K., S. Mittnik, and M. Paolella. (2006) Value-at-Risk prediction: a comparison of alternative strategies, Journal of Financial Econometrics, 4, 53-89. Lee, S. and S. Stevenson (2005). “The case for REITs in the mixed-asset portfolio in the short and long run.” Journal of Real Estate Portfolio Management, 11(1), 55-80. Liow, K. H. (2008). “Extreme Returns and Value at Risk in International Securitized Real Estate Markets.” Journal of Property Investment & Finance 26 (5), 418-446 Lu, C. S. C. Wu, and L. C. Ho (2009). “Applying VaR to REITs: A comparison of alternative methods.” Review of Financial Economics 18, 97-102. Manganelli, S., and R.F. Engle. (2004) A comparison of Value-at-Risk models in finance, in Risk Measures for the 21st Century (Ed.) G. Szegö, Chichester, UK: Wiley, pp. 123-44.

82

Norman, E. J., S. G. Sirmans, and J. D. Benjamin (1995). “The historical environment of real estate returns. Journal of Real Estate Portfolio Management.” 1(1), 1-24. Quan, D. C. and S. Titman. (1999). “ Do Real Estate Prices and Stock Prices Move Together? An International Analysis.” Real Estate Economics, 27: 183–207. Zhou, J. (2011). "Downside Risk Spillover among Global Real Estate Securities Markets." Journal of Real Estate Portfolio Management 17.3, 255-270. Zhou, J. and R. I. Anderson (2011a). "An Emiprical Investigation of Herding Behavior in the U.S. REIT Market." The Journal of Real Estate Finance and Economics 1-26. Zhou, J. and R. I. Anderson (2011b). "Extreme Risk Measures for International REIT Markets." Forethcoming, The Journal of Real Estate Finance and Economics 1-19.

83