DOES IT MATTER WHOM AN AGENT SERVES?
Evidence From Recent Changes In Real Estate Agency Law
Christopher Curran and Joel Schrag*
July 16, 1998
Abstract: Recent changes in real estate law hastened the shift from a seller’s agency regime, in
which real estate agents serve the interests of sellers, to a buyer’s agency regime, in
which agents serve the interests of buyers. We develop a theoretical model that shows
that real estate prices and the time needed to sell a house can either rise or fall after a
shift from seller’s agency to buyer’s agency. Using data from the Atlanta real estate
market, we show that the shift to buyer’s agency led to a significant decline in real
estate prices in the market for relatively expensive houses, while real estate prices did
not significantly change in the market for relatively inexpensive homes. In both
markets, the average time needed to sell a house fell after the change in agency
regimes. These results are consistent with a conclusion that a shift to buyer’s agency
improves the efficiency of the search process.
*
Associate and Assistant Professors of Economics, Emory University, Atlanta, GA 30322. Curran can be
reached at econcc@emory.edu, and Schrag can be reached at jschrag@emory.edu. We wish to thank John
Curran, Robert Chirinko, Henrik Lando, Richard Arnott, two anonymous referees and seminar participants at
the 1997 meetings of the European Association of Law and Economics for their helpful comments and
suggestions. Any errors remaining are our responsibility.
1. Introduction
In 1993, Edina Realty, Inc., a Minnesota-based real estate broker, lost two class action lawsuits brought by
former clients. At issue in these cases, which together cost Edina more than $18.2 million, was the question of
whether Edina had adequately informed its clients of the nature of their agency relationships with the firm.
These cases were a wake-up call to the real estate brokerage industry in the United States, which immediately
recognized its vulnerability to large judgments in litigation arising from confusion over the nature of real estate
agency relationships. Subsequent pressure from the real estate brokerage industry served as the catalyst for legal
reforms in many individual states. Broadly speaking, these new laws clarify the duties that real estate agents owe
to buyers and sellers in each of the different possible agency relationships and thus provide some protection from
litigation to the real estate brokerage industry.
In this paper, we analyze how the real estate market responded to the changes in Georgia’s real estate law
that took effect in the wake of the lawsuits against Edina Realty. The new Georgia law clarifies real estate
agents’ duties to their clients and requires agents to disclose the full range of available agency relationships to
potential buyers. We argue below that these changes in Georgia’s law, particularly the disclosure provision,
hastened the shift from the traditional seller’s agency regime, in which all real estate agents serve the interests of
sellers, to a buyer’s agency regime, in which buyers hire agents who protect and promote their interests.
The nature of the prevailing agency relationship has an important effect on what sellers know about the
potential buyers that they meet. Under seller’s agency, a real estate agent usually has a legal obligation to report
what he knows about a potential buyer to the seller. The seller can then use this information to her advantage
during subsequent bargaining. Under buyer’s agency, on the other hand, an agent who works for a buyer has a
duty to keep what he knows about his client confidential. Therefore, a switch to buyer’s agency should curb the
flow of information from buyers to sellers and may increase the buyers’ bargaining power by depriving the
sellers of valuable information about potential buyers. But then the effect of a legal change that causes a shift
from seller’s agency to buyer’s agency may seem obvious. If buyers have more bargaining power under buyer’s
agency, they should be able to capture a larger share of the gains from trade, and the price of real estate should
fall after a legal reform that promotes buyer’s agency.
1
We believe that this simple intuition fails to recognize all of the possible effects of a change in agency
relationships in the real estate market or, more generally, in any search market. We observe that the prevailing
agency relationship in a search market effectively determines who has a property right over the information that
a buyer reveals to his agent. Under seller’s agency, the seller has a property right over this information; under
buyer’s agency, the buyer holds the property right. Of course, the Coase Theorem establishes that, in the absence
of transaction costs, the distribution of property rights affects only the distribution, and not the size, of the total
surplus generated by economic activity. But transaction costs are not zero in search markets, and, therefore, the
distribution of property rights over information may influence allocative efficiency.
In any search market in which agents operate, allocative efficiency depends on the agents’ ability to match
buyers and sellers in the way that optimally balances two competing goals: minimizing search costs and
maximizing the aggregate gains from trade. While there is no a priori reason to believe that agents approach
this tradeoff in the socially optimal way, it is reasonable to believe that allocative efficiency improves when
buyers reveal more information to their agents. If an agent is familiar with a buyer’s preferences, she can more
effectively identify matches that generate large expected gains from trade. With more information about a
buyer’s willingness to pay, the agent can also more easily identify bargaining situations in which there are no
gains from trade. The agent may be able to use this information to prevent buyers and sellers from wasting time
in fruitless haggling and thereby reduce search costs.
But a buyer’s incentive to reveal information to his agent depends on the distribution of property rights over
that information. If the seller holds the property right, the buyer faces a conflict when deciding how much
information to reveal to his agent. By revealing more information, the buyer both helps his agent to match him
with the best possible trading partner and strengthens the seller’s position during subsequent bargaining. If the
buyer holds the property right, on the other hand, the buyer faces no such tradeoff, because he does not worry
that the seller will learn what the buyer’s agent knows. We conclude that a buyer would reveal more information
to his agent in a buyer’s agency regime, possibly improving the efficiency of the matching process.
In order to analyze these issues, we develop a stylized model of the housing market in which the prevailing
agency relationship influences the flow of information from buyers to sellers. Each seller meets a stream of
buyers who have private information about their willingness to pay. Under seller’s agency, a real estate agent
2
may communicate information about a buyer’s willingness to pay to the seller. In this case, the seller’s optimal
reservation price in each meeting with a potential buyer depends on what she knows about that particular buyer’s
willingness to pay. Under buyer’s agency, on the other hand, the seller receives no information about the buyer’s
willingness to pay. Therefore, she chooses the same reservation price in each meeting with a potential buyer.
A shift from seller’s agency to buyer’s agency has competing effects on the market price of housing. On the
one hand, a switch to buyer’s agency should increase buyers’ bargaining power, putting downward pressure on
the price of housing. On the other hand, however, a switch to buyer’s agency may lead buyers to reveal more
information to their agents, improving their ability to find houses that their clients are willing and able to buy.
From the sellers’ points of view, this effect improves the distribution of buyers that they face, leading them to
increase their reservation prices and putting upward pressure on the price of housing. Thus, a shift from seller’s
agency to buyer’s agency could either increase or decrease the price of housing, depending on which of these two
effects is greater.
A shift from seller’s agency to buyer’s agency may also affect search costs by influencing the probability that
each meeting between a buyer and a seller results in a transaction. If a switch to buyer’s agency deprives sellers
of information about buyers’willingness to pay, the probability that each match results in a transaction may fall.
Intuitively, sellers who are well informed about buyer valuations behave like price-discriminating monopolists,
and they are more likely to trade in each period than sellers who have little information about buyer valuations.
On the other hand, if sellers face a better distribution of buyers after a switch to buyer’s agency, the probability
that each match results in a transaction may rise. Intuitively, sellers face a tradeoff between choosing a high
price and obtaining a quick sale. When the distribution of buyers that sellers face improves, they choose to mute
the increase in the prices that they demand in order to sell their houses faster. Given these countervailing forces,
the average time needed to sell a house could either rise or fall after a shift to buyer’s agency.
Because of the competing effects that we identify, the theoretical model cannot unambiguously predict
whether the switch to buyer’s agency will cause the price of housing and the average time needed to sell a house
to rise or fall. Therefore, we also empirically analyze how the change in Georgia’s law influenced the housing
market in Atlanta. Our empirical results generally indicate that prices fell after the change in Georgia’s law,
though the effect was far more pronounced in the market for relatively expensive homes. This finding is
3
consistent with a conclusion that the legal change— and the ensuing shift to buyer’s agency— did deprive sellers
of significant information about buyer valuations. Our empirical results also strongly indicate that the average
time needed to sell a house fell after the change in Georgia’s law. This finding is evidence that the legal change
improved efficiency by reducing search costs, and it is consistent with a conclusion that, under buyer’s agency,
real estate agents are better able to match buyers with houses that they like.
Our paper contributes to the literature on search models of the real estate market by analyzing the role of
agents who facilitate trades between buyers and sellers.1 Other papers that investigate the matchmaking role that
intermediaries play in search markets include Yavas,2 who studies whether agents can help traders avoid
coordination failures, Salant,3 who studies a seller’s decision about whether or not to use an agent, and
Rubinstein and Wolinsky,4 who analyze the factors that determine the extent to which intermediated trade
replaces search by the traders themselves. None of these papers consider how the institutions that govern the
relationship between intermediaries and traders affect the equilibrium that arises in search markets. This issue is
our main focus. Our paper also contributes to the burgeoning literature on bargaining and markets by showing
how institutions that influence the outcome of bilateral bargaining, such as agency law, have an important effect
on the equilibrium of search markets.5
In the next section we describe the nature of recent changes in real estate agency law, and trace their origin
to the large class-action lawsuits that arose in Minnesota. In section 3 we analyze our model of the real estate
market, and in section 4 we present our empirical work. Section 5 concludes.
1
See, for instance, Wheaton, William C. (1990) “Vacancy, Search, and Prices in a Housing Market Matching
Model, Journal of Political Economy 98:1270-1292. and Yavas, Abdullah. (1992) “A Simple Search and
Bargaining Model of Real Estate Markets,” AREUEA Journal 20:533-548.
2
Yavas, Abdullah. (1995) “Can Brokerage Have an Equilibrium Selection Role?” Journal of Urban Economics
37:17-37.
3
Salant, Stephen W. (1991) “For Sale by Owner: When to Use a Broker and How to Price the House,” Journal
of Real Estate Finance and Economics 4:157-173.
4
Rubinstein, Ariel and Asher Wolinsky. (1987) “Middlemen,” Quarterly Journal of Economics 102:581-593.
5
See, for instance, Binmore, K.G. and M.J. Herrero. (1988) “Matching and Bargaining in Dynamic Markets,”
Review of Economic Studies 55:17-31; Samuelson, Larry. (1992) “Disagreement in Markets with Matching and
Bargaining.” Review of Economic Studies 59:177-185; and Rubinstein, Ariel and Asher Wolinsky. (1990)
“Decentralized Trading, Strategic Behaviour and the Walrasian Outcome,” Review of Economic Studies 57:6378.
4
2. Recent Changes in Real Estate Agency Law
In the United States, real estate transactions typically involve four people: a seller of real property, a buyer, a
listing agent who markets the seller’s property, and a selling agent who brings together the buyer and the seller.
In the traditional seller’s agency system, both the listing agent and the selling agent represent the interests of
sellers. The listing agent’s duty to the seller arises when these two parties execute a listing agreement that states
what actions the agent will take to market the seller’s property, such as listing the property with the Multiple
Listing Service (MLS). The listing agent typically owes the seller the fiduciary duties of good faith, loyalty,
reasonable care, diligence, disclosure, and accounting. In exchange for performing these duties the listing agent
receives a commission from the sale of the property.6
The selling agent’s duty to the seller, on the other hand, does not arise directly from a written agreement
between the selling agent and the seller. Indeed, because the MLS provides a selling agent with access to homes
represented by many different listing agents, a selling agent often never meets the owners of the homes that he or
she shows to prospective buyers. Nevertheless, prior to recent legal changes the selling agent ordinarily owed the
fiduciary duties described above to the seller in a transaction, including an obligation to share with the seller any
information that the buyer reveals to the agent. Because of then existing MLS rules, a selling agent who used the
MLS system automatically became a subagent of the listing agent and, therefore, an agent of the seller.
In recent years it has become more common for buyers to employ a buyers’agent in their search for a house,
rather than a traditional selling agent. A recent survey by the National Association of Realtors reveals that in
1993 thirty percent of buyers used a buyer’s agent.7 Under buyer’s agency the agent owes the fiduciary duties
described above to the buyer rather than the seller, giving the agent the responsibility to work to advance the
6
Lecoe thoroughly explains the different types of agency relationships. Lefcoe, George. (1993) Real Estate
Transactions. Charlottesville: Michie. Also see Federal Trade Commission. (1983) The Residential Real Estate
Brokerage Industry. Washington, DC: U.S. Government Printing Office and Brown, Ronald Benton, Joseph M.
Grohman, and Manuel R. Valcarcel. (1995) “Real Estate Brokerage: Recent Changes in Relationships and a
Proposed Cure,” Creighton Law Review 29:29-97.
7
Pryor, Thomas. (1994) “Dual Agency Disclosure Issues Unresolved,” New Jersey Lawyer, March 7, 1994: 35.
Also see Marino, Vivian. (1993) “Buyer-Broker Transactions on the Rise Across Nation,” Chicago Tribune May
16, 1993: 2R and Ryon, Ruth. (1987) “New State Laws Assist, Confuse Professionals,” Los Angeles Times,
December 20, 1987: 20.
5
buyer’s interests. By employing a buyer’s agent the buyer theoretically ensures that any information revealed to
the agent will remain confidential.
The recent proliferation of real estate agency relationships created the possibility for confusion regarding an
agent’s loyalties and duties. This confusion in turn led to litigation. As mentioned in the introduction, two
notable examples are the cases of Dismuke v. Edina Realty, Inc.8 and Bokusky v. Edina Realty, Inc.9, both of
which are Minnesota class action lawsuits. At issue in Dismuke was the question of whether Edina Realty, the
fourth largest real estate broker in the United States, had adequately informed sellers for whom it was a listing
agent of its legal status in transactions where Edina also represented the buyers. Bokusky, which was joined by
both buyers and sellers, involved the same basic question.
Although Edina’s disclosure forms satisfied
Minnesota’s statutory requirements, in both cases the court ruled that Edina did not satisfy the more stringent
common law requirements of undivided loyalty and complete disclosure. The parties eventually settled for $5.9
million in Dismuke and for $12.3 million in Bokusky.
The outcome of these two lawsuits illustrates the real estate industry’s vulnerability to large judgments
arising from disputes over real estate agency relationships. In direct response to Dismuke v. Edina and Bokusky
v. Edina, the National Association of Realtors (NAR) in November, 1993, decided to press state legislatures to
clarify the law governing real estate agency relationships. In particular, the NAR voted to ask state legislatures to
preempt the common law of agency in real estate transactions and replace it with statutory requirements that
clarify the duties and disclosure requirements owed by real estate agents to their clients. Brown, et. al. (1995)
report that 26 states have altered their real estate agency laws since 1988. Moreover, these authors report that
new agency rules were in effect for only five states by 1992;10 at least thirteen additional states changed their
laws between 1993 and 1995.11
8
Dismuke v. Edina Realty, Inc., No. 92-8716, 1993 WL 327771 (Minn. Dist. Ct., June 17, 1993).
9
Bokusky v. Edina Realty, Inc., No. 3-92 CIV 223, 1993 WL 515827 (D. Minn. August 6, 1993)
10
The five states and the year they changed the relevant laws are California (1988), Alabama (1989), Rhode
Island (1989), Alaska (1990), and New York (1991). See Brown, et. al. (1995), pp. 52-71. According to press
reports (Ryon, 1987) the catalyst for the legal changes that occurred in California was a rising tide of lawsuits
over real estate agency relationships. Thus, the reaction in California to the threat of litigation is similar to what
occurred nationally after Dismuke v. Edina and Bokusky v. Edina.
11
The thirteen states that changed their real estate agency laws in 1993 or after are: Arkansas (1994), Colorado
(1993), Connecticut (1995), Florida (1994), Georgia (1993), Illinois (1995), Maryland (1995), Michigan (1994),
6
On January 1, 1994 a new real estate agency law took effect in Georgia. This law, passed in the 1993
session of the Georgia legislature, clearly addresses the real estate industry’s concerns arising from the litigation
in Minnesota. The legislation changes Georgia’s real estate law in three main ways. First, for real estate agency
relationships the Georgia law replaces the common law of agency with a new statutory law of “limited agency.”
Under “limited agency” real estate agents are no longer considered fiduciaries of their clients, though agents still
must work to promote their clients’ interests. Second, the new law eliminates automatic subagency. Thus, a
buyers’ agent who uses the Multiple Listing Service no longer automatically becomes a subagent of the listing
agent. Third, the new law includes a disclosure provision that requires agents to inform prospective buyers about
the full range of different agency relationships that are available to them, including buyer’s agency.
Conversations with real estate agents and brokers indicate that, since the passage of the new Georgia law,
an overwhelming majority of buyers have chosen to hire a buyer’s agent.12 Moreover, according to these real
estate professionals, agents responded to the new law by changing how they conduct their business in several
important ways. First, buyers’ agents provide more information to buyers than they did when they were agents
of the sellers. In particular, buyers’ agents provide their clients with information on the selling prices of
comparable houses. This information is quite valuable to a potential buyer in the negotiation process. Second,
buyers’ agents help buyers develop negotiation strategies. Third, real estate agents no longer present an offer
directly to the seller in the presence of the listing agent. These meetings were an excellent opportunity for the
buyer’s real estate agent to reveal confidential information about his client’s willingness to pay. Instead, buyers’
agents present the offer to a listing agent alone, usually by fax. The listing agent then transmits the offer to the
seller. All of these changes serve to place greater distance between the buyer’s agent and the seller.
Minnesota (1994), Nebraska (1995), Oregon (1993), Tennessee (1994), and Wisconsin (1994). Brown, et. al. do
not offer dates for the other eight states that have changed the real estate agency law since 1988. These states are
Delaware, Indiana, Iowa, Louisiana, Maine, South Dakota, Texas and Washington. op. cit. pp. 52-72.
12
Unfortunately, real estate firms do not systematically record the nature of the agency relationships in the
transactions that they handle. Nevertheless, the brokers with whom we spoke were in broad agreement that a
large majority of buyers have chosen buyer’s agency since the Georgia law changed.
7
3. A Model of the Real Estate Market
In this section, we investigate how a change in the prevailing agency relationship from seller’s
agency to buyer’s agency affects the equilibrium that arises in the real estate market. In every period t ∈
{0, 1, 2, … }, real estate agents, whose behavior we do not explicitly model, introduce each prospective
buyer to a prospective seller and vice versa. Thus, all market participants are matched in every period that
they are in the market. We assume that the market is “thick,” in the sense that real estate agents can
match every buyer and seller with a new partner in every period they are in the market.
In order to simplify the model, we assume that sellers make take-it-or-leave-it offers that buyers either
accept or reject.13 Therefore, each seller chooses a pricing rule that identifies the price that she demands
as a function of what, if anything, she knows about the potential buyers she could meet.14 Each buyer,
meanwhile, chooses the acceptance rule that identifies which proposed transactions he will accept. In each
match, trade occurs if the buyer accepts the seller’s proposed price, at which time the pair exits the
market. If, in a particular match, the buyer rejects trade at the seller’s proposed price, the match dissolves
and both the buyer and the seller meet new partners in the next period. We analyze the model's symmetric
equilibrium in which all sellers choose the same pricing rule and all buyers choose the same acceptance
rule.
In each match between a buyer and a seller, the buyer’s valuation of the seller’s house is a random
variable.
The different buyers' valuations for the houses that their agents take them to see are
independently and identically distributed. Define vt as a representative buyer’s valuation of a seller’s
house in period t. In order to simplify the model, we assume initially that vt is distributed uniformly on
the unit interval. It might seem reasonable to assume that a buyer is privately informed about his
13
In another paper, we analyze a strategic bargaining model in which sellers and potential buyers make
alternating offers. Permitting this possibility did not change our main conclusions, because privately informed
buyers did not signal their information in equilibrium. Intuitively, high valuation buyers always had an incentive
to deviate to some offer made by lower valuation buyers in order to elicit a favorable counteroffer from the seller.
In the present model, the sellers’ offers can be interpreted as counteroffers to buyers’ initial (uninformative)
offers. Curran, Christopher and Joel Schrag. (1996) “Does it Matter Whom An Agent Serves? Evidence From
Recent Changes in Real Estate Agency Law,” mimeo, Emory University.
14
For expositional clarity, we use the feminine pronoun to refer to sellers and the masculine pronoun to refer to
buyers.
8
valuation for a particular house. In a seller’s agency regime, however, a buyer’s real estate agent can
transmit information about his willingness-to-pay to the seller. Therefore, a seller may be able to learn
about a particular buyer's valuation for her house. In order to model this possibility, we assume that,
under seller’s agency, in period t ∈ {0, 1, 2, … } the seller observes the buyer’s valuation vt with
probability z ∈ [0, 1], while with complementary probability 1 – z the seller receives no information about
vt. Under buyer’s agency, on the other hand, a buyer's real estate agent has a legal obligation to keep
confidential any information about the client's willingness to pay. Therefore, we assume that, under
buyer’s agency, the seller never observes vt.
Define B as the value of the representative buyer’s program and S as the value of the representative
seller’s program at the beginning of a period, before the traders meet their bargaining partners. Our
model is stationary through time, and therefore the values of the representative buyer's and seller's
programs do not depend on the time t ∈ {0, 1, 2, … }.15 Define δ∈ [0, 1) as the discount factor that is
common to both the buyer and the seller. Because sellers make take-it-or-leave-it offers, a buyer will
accept the first offer that yields him a surplus of at least δB, since this is the payoff that he anticipates
receiving if he rejects an offer and proceeds to the next period. Thus, a buyer purchases a house in period t
(and exits the market) if and only if vt – pt ≥ δB, where vt is the buyer’s valuation of the house he sees in
period t and pt is the price that the seller demands. For simplicity, we assume that the buyer accepts an
offer if he is indifferent between accepting it and rejecting it.
The seller’s pricing rule identifies the price that she proposes in period t as a function of what she
knows about the buyer’s valuation. Suppose that in period t the representative seller observes the buyer’s
valuation vt. If the seller wishes to trade with this buyer, she proposes a price pt = vt - δB, which leaves
the buyer just willing to trade. If the seller does not wish to trade with this buyer, she proposes a price
that the buyer will reject, i.e. pt > vt - δB. Alternatively, if the seller does not observe vt, she proposes a
price p that is independent of the buyer’s true valuation, and trade occurs in period t if vt ≥ p + δB.
15
Salant (1991) studies the effect of the non-stationarity that arises in the housing market due to buyers’
preferences for concluding a deal before the school year starts in September. See note 3.
9
The seller’s pricing rule can be summarized by the pair (v, p), where v represents the valuation of the
marginal buyer with whom the seller is just willing to trade when she observes vt, and p is the price that
the seller proposes when she cannot observe the buyer’s valuation. The seller chooses her strategy in
order to maximize the value of her program S(v,p), where:
1
(1)
S ( v, p ) =
∫
z ( x − δB)dx + (1− z ) p(1− p − δB)
v
1− δ( z v + (1− z )( p + δB ))
.
Sellers typically must pay a commission to the real estate agent, which is not reflected in (1), but we
assume that this commission is proportional to the selling price, as is typically the case. Under this
assumption the seller wishes to maximize the total expected discounted income from the sale of her
property, i.e. she wishes to maximize (1), even though she must pay a commission to the agent.
Given the seller’s pricing rule, the value of the buyer’s program B is:
z (1− v)δB + (1− z )
(2)
B=
∫
1
p + δB
( x − p)dx
1− δ
[ z v + (1− z )( p + δB)]
.
We summarize the representative seller’s optimal pricing rule in Proposition 1.16
PROPOSITION 1: (i) Suppose that the seller observes the buyer’s valuation in period t. If the buyer’s
valuation vt ≥ v*(z), where v *( z ) =
4 − δ(1− z ) − 2 4 − 2δ2 (1+ z ) − 2δ(1− z )
δ(3 + z )
, then the seller proposes (and
the buyer accepts) a price pt = vt - δB. If the buyer’s valuation vt < v*(z), the seller proposes (and the
buyer rejects) a price pt > vt - δB. (ii) Suppose that the seller does not observe the buyer’s valuation in
period t. Then the seller proposes a price p*(z), where:
p *( z ) =
(− δ2 (1− z ) 2 (1+ z ) − 2(3 + z )(− 2 +
2(1− δ)(2 + δ+ δz ) )+ δ(1+ 11z + 3 z 2 + z 3 − (1− z 2 ) 8(1− δ)(2 + δ+ δz ) ))
.
2(3 + z )2
A buyer accepts this price in period t if and only if vt ≥ p*(z) + δB.
16
The proof of Proposition 1 is located in the Appendix, along with the proofs of the other results.
10
Define P(z) as the expected price at which a representative seller sells her house, and define Q(z) as
the probability that a transaction will occur in a given meeting between a buyer and a seller. The expected
number of periods that a house will be on the market is Τ(z) = 1/Q(z). We have Proposition 2.
PROPOSITION 2: (i) P(z) = p*(z). Furthermore, the expected price at which a representative seller sells her
house is higher when sellers are more likely to observe buyers’valuations, i.e. dP/dz > 0.
(ii) Q ( z ) =1−
(2 + δ)(1− z ) − (1+ z ) 4 − 2δ2 (1+ z ) − 2δ(1− z )
δ(3+ z )
. Furthermore, a transaction is more likely to
occur in a given meeting between a buyer and a seller when sellers are more likely to observe buyers’
valuations, i.e. dQ/dz > 0. Therefore, the expected time that individual houses will remain on the market is
lower when sellers are more likely to observe buyers' valuations, i.e. dΤ/dz < 0.
Proposition 2 establishes that sellers benefit in two ways when they have more information about
buyers' valuations. First, the expected price at which they sell their houses increases. This result is not
surprising; a seller who is better informed about a buyer's willingness to pay can more effectively capture
the gains from trade. Second, sellers expect to sell their houses faster when they have more information
about buyers' valuations. To see the intuition for this result, it is helpful to recognize that a seller who can
observe buyers' valuations behaves like a price-discriminating monopolist.
A price-discriminating
monopolist typically sells more units than a monopolist who cannot observe buyers' willingness to pay,
because he can sell to low valuation buyers without losing revenue on infra-marginal units. Analogously,
a seller who is well-informed about buyers’ valuations is more likely to trade in a given period than would
be the case if she were uninformed, because she can trade with a low valuation buyer without worrying
that she is possibly losing revenue from a high valuation buyer.
The comparative static results summarized in Proposition 2 yield straightforward predictions about
the effects of a switch from seller’s agency to buyer’s agency. Because sellers receive less information
about buyer valuations under buyer’s agency— under buyer’s agency z = 0, while under seller’s agency z >
0— Proposition 2 yields the following predictions.
11
PREDICTION 1: All else equal, a switch from seller’s agency to buyer’s agency causes the expected price
of housing to fall.
PREDICTION 2: All else equal, a switch from seller’s agency to buyer’s agency causes the expected time
that houses will be on the market to rise.
These simple predictions about the effect of a move from seller’s agency to buyer’s agency rest on a key
assumption, namely that, in each match between a buyer and a seller, the distribution of the buyer’s valuation of
the seller’s house is the same under both agency regimes. But buyers may be more willing to reveal information
about their preferences under a buyer’s agency regime, since they could be confident that their agents would not
divulge this information to sellers. If this additional information helps the agents to match the buyers with
houses they like, the distribution of buyer valuations in each individual match may improve after a switch to
buyer’s agency.
In order to model this possibility, we suppose that, under buyer’s agency, a buyer’s valuation of a house that
he sees is a random variable that is distributed on the unit interval according to the probability density function
f(v;α) = (1 – α) + 2αv, α ∈ [0, 1], and the associated probability distribution function F(v;α) = (1 – α)v + αv2. A
buyer’s valuation is distributed uniformly, as before, when α = 0, while he is more likely to have a high valuation
for the house of a seller with whom he is matched when α is larger.17 Under the distribution F(v;α), a buyer’s
expected valuation in a match with a seller is v = (3+ α ) / 6 . Thus, a shift to buyer’s agency causes a buyer’s
expected valuation in each match to increase by α/6. This increase reflects an improvement in the agent’s ability
to match the buyer with houses that he likes.
Define s as the value of the seller’s program and b as the value of the buyer’s program. Because under
buyer’s agency sellers cannot observe the buyers’valuation vt, the representative seller chooses the price that she
will propose in each period, say p, in order to maximize the value of her program s(p), where:
(3)
s( p) =
p(1− F ( p + δb;α ))
.
1− δF ( p + δb;α )
Formally, if α′> α′
′
, then F(v;α′
) first-order stochastically dominates F(v;α′
′
), i.e. F(v;α′
) ≤F(v;α′
′
) ∀ v ∈ [0,
1].
17
12
Meanwhile, the value of the representative buyer’s program is given by:
1
∫
b=
p + δb
(4)
( x − p ) f ( x;α)dx
1− δF ( p + δb;α )
.
In the following Proposition, we summarize the model’s equilibrium.
PROPOSITION 3: Under buyer’s agency, there exists an equilibrium price p* ∈ [0,1]. This price satisfies:
(3.1)
− p*+
1− F ( p*+ δb;α) δp*(1− F ( p*+ δb;α ))
+
=0 .
f ( p*+ δb;α )
1− δF ( p*+ δb;α )
A buyer accepts this price in period t if and only if vt ≥ δ
b + p*.
In theory, we could use the solution to (3.1) to define a function p*(α), which would identify the price that
sellers demand as a function of α. Using this function, it would then be possible to analyze how the expected
selling price of a house and the expected selling time of a house depend on the parameter α. Because sellers
demand the same price in every period, regardless of the valuation of the buyer that they meet, the expected
selling price of a house would simply be p*(α). The probability that the buyer and seller complete a transaction
in any randomly chosen period would be q(α) = 1 – F(p*(α) + δ
b), while the expected number of periods that a
house is on the market would be τ(α) = 1/q(α).
While it is possible in theory to solve equation (3.1) for a closed-form solution, in practice the solution is
unwieldy for α > 0. Therefore, it is not feasible to investigate analytically how p*(⋅) and τ(⋅) depend on α.
Instead, we analyze the behavior of these functions numerically.
In Table 1 we present the equilibrium values of p* and τ for a range of different values of α. In these
calculations we fix the discount factor δ= .998, a value that corresponds to weekly meetings between potential
traders who have a yearly discount factor of approximately 0.9. 18
18
We used Mathematica v.3.0.1 in order to perform the calculations that we summarize in the tables. While in
the paper we present only the results that we obtain when δ= .998, we performed similar calculations for
numerous discount factors ranging from zero to .999. These results were consistent with the results that we
present here.
13
[Insert Table 1 about here.]
The table tells a consistent story. As the expected buyer valuation in each match increases— as α
increases— the equilibrium price of housing rises. This result is intuitive. Sellers naturally wish to capture part
of any increase in buyers’willingness to pay, which they do by increasing the price that they demand. But sellers
can also take advantage of an improvement in the distribution of buyer valuations to reduce the expected time
that their houses will be on the market. Figure 2 illustrates that, as α rises, a representative seller expects to sell
her house faster. Intuitively, a seller faces a tradeoff between the price that she receives and the expected time
that her house will be on the market. When she faces a better distribution of buyers— buyers who are, on
average, willing to pay more for her house— she can both increase the price that she receives and reduce the
expected time that her house will be on the market. In equilibrium, she optimally balances this tradeoff.
The results that we present in Table 1 call into question the simple predictions that follow from Proposition
2. Recall that, if the distribution of buyer valuations in each match is the same under both agency regimes, a
shift from seller’s agency to buyer’s agency causes the expected price of real estate to drop and the expected time
to sell a house to increase. But if the distribution of buyer valuations improves after a shift to buyer’s agency,
real estate prices may increase overall, and expected time on the market may fall. Intuitively, if sellers receive
little useful information under seller’s agency, possibly because buyers conceal information from their agents,
then a shift to buyer’s agency deprives sellers of little information, and the effect on price and time on the market
is small. But if, under buyer’s agency, buyers reveal more information to their agents, and the distribution of
buyer valuations consequently improves, the resulting upward pressure on real estate prices and downward
pressure on the expected time to sell a house may overwhelm the small effect of the sellers' loss of information.
We illustrate the different possible effects of a shift from seller’s agency to buyer’s agency in Table 2.19 Each
cell of the table identifies how a switch to buyer’s agency affects the expected price of housing (P) and the
expected number of days required to sell a house (τ), as a function of the parameters z and α. Recall that z is the
probability that a seller observes a buyer’s valuation under seller’s agency, while α reflects the extent to which a
switch to buyer’s agency improves a real estate agent’s ability to match buyers with houses that they like.
19
The calculations that underlie Table 2 reflect the previous assumption that the discount factor δ= .998.
14
[Insert Table 2 About Here]
Table 2 reveals that there are three different ways that a shift from seller’s agency to buyer’s agency can
affect the expected price of real estate and the expected time that a house will be on the market. First, when z is
very small, meaning that sellers receives little information about buyers’ valuations under seller’s agency, the
expected price of real estate tends to increase and the expected time to sell a house tends to decrease after a
switch to buyer’s agency. This result is intuitive. Suppose, for instance, that z = 0. Then the entire effect of a
switch to buyer’s agency arises from its effect on the distribution of buyer valuations. If agents are better able to
match buyers with houses that they like after a switch to buyer’s agency, i.e. if α > 0, real estate prices will
increase and the average time that houses spend on the market will fall.
On the other hand, if z is large, meaning that sellers are well-informed about buyers’ valuations under
seller’s agency, then a switch to buyer’s agency tends to cause the expected price of real estate to decrease and
the expected time needed to sell a house to increase. For instance, suppose that z = .50, meaning that a seller
observes a potential buyer’s valuation with probability .50. Then even a significant improvement in the
distribution of buyer valuations cannot overcome the downward pressure on the price of real estate and the
probability of trade in each match that results from sellers’loss of information about buyer valuations.
Finally, it is possible that both the expected price of real estate and the expected time that it takes to sell a
house will decline. Suppose, for instance, that z = .25. Then the expected price of real estate always falls after a
switch to buyer’s agency. On the other hand, the expected time that it takes to sell a house can either rise or fall;
it falls if α ≥ 0.5, i.e. if there is a sufficiently great improvement in the distribution of buyer valuations.
The results in Table 2 provide a framework that we can use to interpret the effects of a shift from seller’s
agency to buyer’s agency. Because we focus on how this institutional change affects two variables (the price of
real estate and the expected time needed to sell a house), both of which could either increase or decrease, there
are four different patterns of effects that we could potentially observe.
interpretation of each possible pattern of effects.
15
In Table 3 we summarize the
[Insert Table 3 About Here]
As Table 3 indicates, a finding that a shift to buyer’s agency causes both real estate prices and the time
needed to sell a house to rise would be inconsistent with, and would lead to a rejection of, the theoretical model.
To see why, recall that sellers face a tradeoff between the price that they receive and the expected time needed to
sell their houses. If market conditions change in a way that is favorable to sellers, they have an incentive to mute
the increase in the prices that they demand in order to speed the sales of their houses. In other words, sellers are
willing to trade off some increase in the price that they receive in order to obtain a faster sale. But then an
increase in the price of real estate that follows an improvement in the distribution of buyer valuations will be
accompanied by a fall in the time needed to sell a house.
On the other hand, a finding that a shift to buyer’s agency leads to a fall in the time needed to sell a house is
consistent with the model, and it would provide evidence of an improvement in the distribution of buyer
valuations in each match, i.e. evidence that α > 0. Such a finding would thus provide indirect evidence that a
shift to buyer’s agency improves the ability of real estate agents to match buyers with houses that they like. This
conclusion has potentially significant implications for welfare under different agency regimes. If a shift to
buyer’s agency were to increase the speed at which buyers and sellers make deals with each other, it would
reduce search costs and, thereby, promote efficiency.
Finally, a finding that a shift to buyer’s agency leads to a fall in real estate prices would provide evidence
that sellers do, in fact, receive valuable information about buyer valuations under the seller’s agency regime.
When a change in the agency regime deprives sellers of this information, their ability to capture the gains from
trade is diminished, and real estate prices consequently decline.
Of course, how the introduction of buyer’s agency actually affects real estate prices and the expected time
needed to sell a house is an empirical question. In the next section we explore this question by analyzing the
effect of a change in Georgia real estate law that paved the way for the adoption of buyer’s agency in that state.
4. Empirical Evidence
16
In this section, we develop a set of empirical results that help to illuminate how a shift from seller’s agency
to buyer’s agency affects both the overall efficiency of the real estate market and the distribution of the gains
from trade in this market. The theory that we developed in the previous section provides us with a framework
for using observed changes in real estate prices and the time needed to sell a house to assess how a switch to
buyer’s agency affects welfare. We now use the change in Georgia’s real estate law that occurred on January 1,
1994, as a natural experiment for examining the impact of the introduction of buyer’s agency on these variables.
As discussed in the introduction, the new law effectively introduced buyer’s agency in Georgia. Interpreted in
light of our theoretical results, our empirical results will permit us to judge whether or not the change in
Georgia’s law promoted efficiency.
There is no a priori reason to believe that all segments of the housing market would respond in the same
way to a change in the prevailing agency relationships. Indeed, there are good reasons to believe that the market
for relatively expensive high quality houses would react differently than the market for relatively inexpensive
“starter homes.” Obviously, the stakes involved in bargaining are much greater in the market for relatively
expensive houses. Also, the market for relatively inexpensive houses may be “thicker” than the market for
higher quality houses. In a very thick market there are many substitutes for each house and each potential buyer,
and, therefore, each house may sell at a well-defined “market price.” For both of these reasons, bargaining may
be more important in the market for relatively expensive houses, and, therefore, legal changes that influence
bargaining may have a greater effect on this market than on the market for less expensive houses. Our empirical
analysis lends support to this conjecture.
4.1. Econometric Model
We estimate two equations. First, we use the hedonic price model to relate a house’s sale price to (1) its
physical characteristics, (2) market conditions at the time of the contract, and (3) characteristics of the house’s
location. In particular, we assume that equation (13) describes the ith house’s real sale price at time t (where the
real price equals the ith house’s observed nominal sale price, pit, divided by the consumer price index for the
month preceding the house’s closing date, cpit:
17
 pit 
(13) ln
H + M t β′
M + L i β′
L + εit
 cpi 
= α 0 + H i β′
 t
where Hi is a vector of variables that describe the house’s size and quality, Mt is a vector of variables that
describe the market conditions at the time that the parties to the sale sign the contract, and Li is a vector of
variables that describe the characteristics of the house’s location. The nominal sales price includes the real estate
commission.20 The β’s are vectors of parameters; α0 is the intercept parameter; and εit is the residual error term.
Second, we use the hazard function to relate the time a house is on the market a similar set of variables that
describe (1) the house’s physical characteristics, (2) market conditions at the time of the contract, and (3)
characteristics of the house’s location. In particular, we estimate:
(14) ln (Tit )= α 1 + H i ϕ′
H + M t ϕ′
M + L i ϕ′
L + Pit ϕ P + ηit ,
where Tit is the length of time in days that a house was on the market on the day of closing; Pit is the percentage
of the time that the house is on the market that occurs after April 4, 1994; and ηit is the residual error term.21
In an earlier paper we use the Chow breakpoint test to establish that April 1, 1994, is the appropriate date to
assume that the new law became effective.22 In this paper we use the Chow breakpoint test to test between
20
The mean commission rate for the houses sold in 1993 (before the introduction of buyer’s agency) is 6.83%,
with a standard deviation of 0.5625 (n = 3611). The mean commission rate for houses sold after the introduction
of buyer’s agency is 6.79% with a standard deviation of 0.6166 (n = 9003). While these two means are
statistically different at the 1% level (t = -3.42), the difference is not economically significant. For instance, the
difference in the commissions on a house selling for $250,000 is -$97.75. In 1993 the total commission on the
sale of a house averaged at $17,077. Given that the commission must be split between two sets of real estate
brokers and agents, the statistically significant reduction of the commission after 1993 is not large enough to
have a plausible effect on the behavior of real estate agents.
21
Pit equals 0 if the house was sold before the change in Georgia’s real estate law. It equals 1 if the house was
listed and sold after the change in Georgia’s real estate law. It equals the fraction of the total time that the house
was on the market that occurred after the law change for houses that were listed before the law change and sold
after the law change.
22
Theory cannot predict exactly when the legal change would first affect sale prices. It is possible that buyers
and sellers perfectly anticipated the impact of the law previous to January 1, 1994. A strong anticipation effect is
unlikely, however, because buyers do not enter the housing market often enough to be familiar with upcoming
changes in real estate law. In fact, one of the more important provisions of the new Georgia law is the
requirement that agents disclose to all buyers the nature of the different agency relationships that buyers can
choose. Because many of the houses that sold in late 1993 actually closed in January and February of 1994, it is
more likely that several months passed before buyer’s agency had an impact on housing prices. We also expect
that the traditional thinness of the housing market during the first quarter of any year would slow the speed with
which the new law would begin to affect the negotiations of buyers and sellers. The Chow breakpoint test allows
us to identify if, and when, there is a significant change in the estimates of the parameters of Equation (13)
between January 1, 1994, and June 1, 1994. Our results support the hypothesis that the parameter values of
18
estimates of equations (13) and (14) for houses selling for less and $175,00 and those selling for $175,000 or
more. The results of this test suggest that the regression parameters both for equation (13) and (14) for houses
selling for a price greater than $175,000 are significantly different from the parameters for houses selling for less
than $175,000.23 Next, we analyze the effect of this structural change by estimating the predicted change in the
prices of houses with the median characteristics observed in 13 neighborhoods both for the relatively more
expensive and the relatively less expensive houses. In all but one of these neighborhoods the predicted fall in
prices after April 1, 1994, are significantly greater (in absolute size and as a percentage of the predicted preApril 1, 1994, price) for houses priced over $175,000 than they are for houses priced under $175,000. Finally,
we find that the length of time it took to sell houses fell in the period after the change in the real estate law;
moreover, our estimates indicate that the fall in the time that houses were on the market was larger for more
expensive houses.
4.2. The Data
The housing data include 12,384 houses with a closing reported in the Atlanta Multiple Listing Service
(MLS) during the period July of 1993 through June of 1995. The data are from the residential suburbs of
northern Atlanta.24 Since Equations (13) and (14) are in reduced-form, the covariates included should be
Equation (13) are significantly different before and after April 1, 1994. See Curran and Schrag (cited in note 13),
pp. 24 and 40.
23
For Equation (13) the weighted residual sum of squares (RSS) for the regression for the less expensive houses
is 178.5147 while the RSS for the more expensive houses is 207.2962. The RSS for the model where these
parameters are restricted to being equal is 576.1335. The Chow test gives an F = 79.3962, a number substantially
greater than the critical value of 1.58 at the 1% level (the degrees of freedom are 76 and 12,232). For Equation
(14) the weighted RSS for the more expensive houses is 5,418.683 while the weighted RSS for the less expensive
house is 9,862.741. The RSS for the model where these parameters are restricted to being equal is 15,420.6. The
Chow test gives an F = 2.87, a number greater than the critical value of 1.60 at the 1% level (the degrees of
freedom are 39 and 12,296). See Johnston, Jack and John Dinardo. (1997) Econometric Methods, 4th Edition,
New York: McGraw-Hill Companies, Inc., pp. 126-130.
24
The data are from the MLS Areas numbered 13, 14, 61, 62, 81, 82, 83, 121, 131, and 132. All of these areas
are in the north section of town, mostly outside the perimeter road (I-285). They do not include any areas within
the boundaries of the city of Atlanta. The data include only houses that have one or more bedrooms and one or
more bathrooms and are at least one year old. The MLS data include 17,310 closings. We exclude houses with
the number of bedrooms or bathrooms reported as less than one because these entries are more than likely due to
an entry error. We exclude new houses (i.e., houses with a reported age of less than one year) because the new
house market is substantially different from the market for older houses. In particular, an additional party – the
builder – plays a major role in the negotiations and generally are able to demand that real estate agents receive
lower commissions than they receive in the sale of older houses. See Curran and Schrag (cited in note 13), p. 24.
19
variables that describe characteristics of the house that are exogenous either to the value of the house or to the
time it takes to sell the house. The vector Hi includes variables intended to measure the physical and quality
characteristics of the house. We include the number of bedrooms (BEDS), the number of bathrooms (BATHS),
their squares (BATHS2 and BEDS2), and their cross products (BEDBATH) as proxies for the size of the
house.25 The log of the age of the house, LAGE, measures the impact of the age of a house on the sales price.26
The last six variables – FIRE, HWAR, MULTI, SPLIT, COOL, and ZONED – are various measures of the
quality of the house. Presumably, more fireplaces, the presence of air conditioning, and the presence of zoned air
conditioning27 will add to the value of the house and reduce the selling time. The presence of a home warranty
on a house may be viewed by potential buyers as a positive or a negative attribute. A seller and buyer may use the
home warranty as a risk sharing device. In this case, the presence of a home warranty may raise the final selling
price and lower the selling time. On the other hand, the presence of a home warranty on a house for sale may be
a signal to a buyer that the seller sees potential problems with the house. Since a buyer knows that the seller has
an incentive to underinsure, the presence of a home warranty may actually lower the final sales price and
increase the selling time. The two variables measuring the number of floors in a house – MULTI and SPLIT –
are technically not measures of the quality of a house. The signs on these two variables depend on how
purchasers value these characteristics and thus reflect consumer tastes. Economic theory cannot predict how
these variables affect either the value of a house or the time it takes to sell the house.
We account for the temporal effects of market conditions on housing prices and selling times with the seven
variables included in Mt. The first two measure the impact of (1) the interest rate at the time of the signing of the
contract and (2) expectations about the future path of interest rates on sales prices and time-on-the-market using
the mortgage rate one month before the sale (INT1) and two months before the sale (INT2). This approach has
the advantage of avoiding the problems that arise with the use of an estimated variable (expected interest rates)
25
The square-footage of the houses is not available. Preliminary regressions using the RESET test for functional
form suggested by Thursby and Schmidt suggest that the functional form for bedrooms and bathrooms used in
this model is appropriate. Thursby, J. and P. Schmidt. (1977) “Some Properties of Tests for Specification Error
in a Linear Regression Model,” Journal of the American Statistical Association 72:635-41.
26
LAGE is equal to Log(Age + 1).
27
Homes with “zoned” air conditioning are multi-story houses with air conditioning provided by a separate unit
for each floor (or for at least two of the floors).
20
as an explanatory variable in the estimation of Equation (13). However, because this method mixes the effects of
current interest rates and expectations about future interest rates, we cannot isolate the effect on sales price or
and time-on-the-market of changes in either of these two variables. We use unemployment rate (UNEM1 and
UNEM2) and the wage levels (WAGE1 and WAGE2) in the Atlanta area for the two months previous to the
month of the sale to measure the effect of aggregate demand on the sales price and time-on-the-market. Finally,
we use a dummy variable (SUMMER) equal to one if the sale closed during the months of June, July, and
August. These months are traditionally periods of increased activity in the market because of the desire of many
families to move during the summer vacation when the move will have a minimally disruptive effect on their
children.
We use two sets of variables to measure the effect of the house’s location on the sales price and time-on-themarket. The first set includes six variables that characterize the census tract in which the house is located: the
population density of the census tract, the logarithm of the median income of the census tract, the percentage of
the occupied homes that are owner-occupied, the percentage of the population over age 25 with a bachelor’s
degree or higher, the mean time to work, and the percentage of the population that is white. These data are from
the 1990 Census. We expect that the lower population densities, higher median incomes, higher percentages of
owner-occupied homes, higher percentages of college educated adults, shorter commuting times, and a higher
percentage of whites will tend to raise final sale prices and lower the time a house is on the market.
In order to measure neighborhood effects that lie beyond the census tract but encompass an area less than
the county, we include in our location variables a set of dummy variables for census tracts that the Bureau of
Census groups together into “neighborhoods.” Finally, we include a dummy variable for the county in which the
house is located. In order to avoid multicollinearity, we omit dummy variables for at least one “neighborhood” in
each county and the dummy variable for Gwinnett county.
4.3. Regression Estimation
Estimation of Equation (13) using ordinary least squares suggests the presence of heteroskedasticity in the
model. Use of the White’s heteroskedasticity-consistent estimates does not remove this heteroskedasticity. Thus,
we adjust for the presence of heteroskedasticity by estimating Equation (13) using weighted-least squares, where
21
the weights are determined via the method suggested by Harvey (1976).28 Moreover, we include the dummy
variable SA to allow for differences in the slopes and the intercept terms of the regressions for the two periods.
The dummy variable SA is equal to one if the house closing occurs before April 1, 1994, and zero otherwise. We
eliminate all of these interactive terms that are not significant at the 15 percent level in order to improve the
precision of our estimates.
Table 4 reports the results of the two weighted-least-squares estimations of Equation (13) for the two subsamples of the data. Columns (1) and (3) report estimations of Equation (13) with all of the non-interactive
variables included while Columns (2) and (4) report estimations of Equation (13) with only the non-interactive
variables that are statistically significant at the 15 percent level included. We use the regression results in
Columns (2) and (4) to calculate the predicted price changes reported in Table 5; we get substantially the same
results using the parameters reported in Columns (1) and (3) to calculate the price changes.
In order to facilitate the interpretation of these parameter estimates, we use the two sets of parameter
estimates reported in Table 4 to compute the predicted real sales price before and after April 1, 1994, for each of
the thirteen neighborhoods in our sample. In computing these estimates we use the median values for the
independent variables. Let Xmi be the vector of median values of the matrix – that is, let,
[
]
X mi = 1| H 0mi | M 0m | L0mi , where i is an index indicating whether the house sold for more or less than
[
]
$175,000.29 Then, ln( pˆki ) = X mi βˆki , where p̂ ki is the predicted real sales price, βˆki = αˆ0 |βˆHi |βˆMi |βˆLi , and k
= 1, 2 is an index representing the pre- and post-April 1, 1994, periods. Using this split of the data gives us 13
pairs of estimated real sale prices for each of the two price groups.
28
(
)
z′
α
Harvey assumes that the model to be estimated is yi ~ N X′
. We can estimate the unknown
iβ, e
∃ ) on zi, where u∃i are the square of the residuals
parameters α by applying OLS to a regression of the log( u
from the estimation of Equation (1). Harvey shows that the estimator is consistent if we subtract 1.2704 from the
intercept term. See Harvey (1976). The variables we include in zi are Hi, LPDEN, LINCOME, and PBD. Table 6
reports the results of the estimation of this regression. Harvey, A. C. (1976) “Estimating Regression Models with
Multiplicative Heteroscedasticity,” Econometrica 44:461-465.
2
i
29
2
In order to maintain comparability in the estimates for houses prices above $175,000 and those priced below
$175,000, we use the median values of the time-dependent variables INT1, INT2, WAGE1, WAGE2, UNEM1,
UNEM2, and SUMMER for the whole sample.
22
We report the results of these estimations in Table 5 along with their standard errors.30 The estimates in
Table 5 tell a fairly consistent story. The price changes for houses selling for under $175,000 are neither
statistically nor economically significant. Moreover, for houses selling at a price of $175,000 or more, all but one
of the neighborhoods exhibit a decrease in price. The decreases for ten of these twelve neighborhoods are
significantly different from zero; all of the decreases are economically significant. The one neighborhood
showing an increase in the sales price for houses priced over $175,000 — Buford — has only ten observations in
the sample, suggesting that the market for expensive houses in that neighborhood is too thin to draw any
conclusions.
The estimation of Equation (14) uses much the same methodology as does the estimation of Equation (13).
We adjust for heteroskedasticity via the methodology suggested by Harvey (1976).31 Moreover, we include in
Equation (14) the variable Pit, the percentage of total time that the house is on the market that occurs after April
1, 1994. Pit has the value of zero for houses listed and sold previous to April 1, 1994, and a value of one for
houses both listed and sold after April 1, 1994. The coefficient for this variable allows us to measure the impact
of the introduction of buyer’s agency on the length of time that a house is one the market. A negative value for
this parameter implies that houses tended to sell faster, everything else held constant, after the introduction of
buyer’s agency. Table 6 reports the values of this coefficient (1) when the estimation includes all houses,
(2) when the estimation includes only houses with a real selling price greater than or equal to $175,000, and
30
We calculate the standard errors of the predicted price differences from the interactive terms reported in the
right-hand-side of Table 9. We write Equation (13), dropping subscripts, as: ln ( p )= Xβ+ (X⋅Y93)βd + ε , where
Ω
ε~ N (0,Ω ) and Ω =  11
Ω 21
Ω 12 
 . Ω 22 is the variance-covariance matrix associated with the interactive
Ω 22 
X ′Ω X 

variables. Clearly, ln ( pˆt =93 )− ln ( pˆt > 93 )= X mβˆd ~ N X m βd , m 22 m  , where n is the sample size. We
n


calculate the standard errors of the predicted differences in the logarithms of the prices for the two period from
X′
m Ω̂ 22 X m
.
n
31
The evidence for heteroskedasticity in the estimation of Equation (14) is weaker than it is for Equation (13).
The variables used in the second stage of the Harvey method are the interest rate lagged one month, the
unemployment rate lagged two months, and the percent of the time a house is on the market that occurred after
the introductions of buyer’s agency. See note 28 for a brief description of Harvey’s methodology.
23
(3) when the estimation includes only houses with a real selling price less than $175,000.32 All three values are
negative and highly significant, suggesting that houses spent significantly less time on the market after the
change in Georgia’s law.
Our theoretical model provides a framework that we can use to interpret these results. Consider the effect of
the change in Georgia’s real estate law on the market for relatively inexpensive houses. In this market, we find
that the legal reform caused both real estate prices and the time that houses were on the market to fall, though
the price effect was statistically insignificant. Using Table 3 to interpret these results, the slight fall in real estate
prices leads us to conclude that the change in Georgia’s law may have deprived sellers of at least some
information about potential buyers’ valuations. The fall in the time that houses were on the market leads us to
conclude that, on average, buyers had higher valuations for the houses that their agents showed them after the
legal change became effective. This finding is indirect evidence that the change in Georgia’s law improved the
ability of real estate agents to match buyers with houses that they like.
Now consider the effect of the change in Georgia’s real estate law on the market for relatively expensive
houses. In this market, we again find that the legal reform caused both real estate prices and the time that houses
were on the market to fall. Furthermore, the price effect was quite large in this market. We interpret these
results as evidence both that the legal change deprived sellers of a significant amount of information about the
buyers’ valuations and that real estate agents more effectively matched buyers with houses that they liked after
the legal reform.
5. Discussion and Conclusions
Our empirical findings strongly suggest that a buyer’s decision to use a buyer’s agent improves his relative
bargaining power in negotiations with sellers, particularly in the market for relatively expensive homes. It is not
surprising that a shift to buyer’s agency— and sellers’ subsequent loss of information about buyer valuations—
has the greatest impact on the “high-end” real estate market. Because relatively expensive houses are more
32
Table 6 reports the estimates of the coefficient estimates for Pit for reasons of brevity. The full regression
results are available from the authors.
24
highly differentiated than less expensive houses, information about the buyer’s preferences is probably more
valuable to sellers in the high-end market. Of course, a shift to buyer’s agency may also influence bargaining in
ways that are outside the scope of our model. For example, buyer’s agents may play an active role in the
negotiations between buyers and sellers and, because they are skillful negotiators, may be able to obtain more of
the gains from trade for their clients than buyers would be able to extract if they negotiated on their own behalf.
This skill is more valuable in the market for relatively expensive homes, and therefore it is sensible to believe
that agents who possess special bargaining skills would primarily serve that market.
Our empirical results help to identify how a shift from seller’s agency to buyer’s agency affects welfare.
Consider the finding that the change in Georgia’s real estate law caused the average time needed to sell a house
to fall. This result suggests that a shift to buyer’s agency lowers search costs in the real estate market. Clearly,
such an effect would be welfare improving. Interpreted through the lens of our model, this finding is also
evidence that a shift to buyer’s agency improves the ability of real estate agents to match buyers with houses that
they like and, thereby, increases the expected gains from trade in any meeting between a buyer and a seller. This
effect would also be welfare improving.
Our empirical results provide evidence that some Georgia residents — those who owned relatively
expensive homes — suffered a financial loss when the new agency law took effect. Nevertheless, the conclusion
that sellers were hurt by the change in real estate law must be tempered by a recognition that many sellers are
also buyers; a homeowner who sells his or her house typically buys another. Therefore, while some sellers’
houses apparently lost value after the new law took effect, those sellers who buy another house in Georgia
presumably benefit from the drop in prices brought about by buyer’s agency. Sellers who purchased a relatively
inexpensive house before the change in the law and now wish to “trade up” to a relatively expensive house
should particularly benefit from the differential effect of the legal change in the different segments of the real
estate market. 33
It may seem that real estate agents suffered a loss when the new agency law took effect. Because agents’
commissions are typically a fixed percentage of the sale price, anything that reduces real estate prices should also
33
On the other hand, older homeowners frequently sell their large, relatively expensive houses and “trade down”
after their children leave home in order to extract some of the wealth they hold in the house. These individuals
suffered a real loss in wealth due to the fall in the price of relatively expensive houses.
25
reduce commissions and possibly, in the long run, the number of real estate agents. This conclusion, however, is
not consistent with lobbying by the National Association of Realtors and various state-level real estate
associations to change real estate agency law in ways that favored the expansion of buyer’s agency. Clearly, real
estate agents believed that these legal changes would benefit them. The most plausible explanation for this belief
is that the real estate industry viewed legal reform as the best way to reduce its vulnerability to large class action
lawsuits such as Dismuke v. Edina Realty. Thus, the specter of costly litigation over real estate agents’ duties to
buyers and sellers created a demand for new laws that clarify these duties, even at the cost of smaller
commissions. This explanation is appealing because it helps to explain why these legal changes did not occur
sooner. If real estate prices, and hence commissions, are indeed lower under buyer’s agency, then it clearly
would not have been in the interest of the real estate industry to lobby for these changes until it became apparent
that new laws were needed to reduce the threat of litigation.
This paper has attempted to illuminate how the prevailing agency relationship affects the price of housing
and the expected time that is needed to sell a house. Further research is needed to explore the robustness of our
conclusions that moving to a buyer’s agency system leads to a significant drop in the price of relatively expensive
homes but leads to little change in the price of relatively inexpensive houses, while reducing the time needed to
sell houses in both markets. Ideally, researchers would study the effect of similar legal changes in other states.
Our theoretical and empirical results indicate that the nature of the relationship between intermediaries and
buyers and sellers has an important effect on the equilibrium that arises in search markets. This conclusion
raises the obvious question of what factors determine the nature of the socially optimal relationship between
intermediaries and buyers and sellers. In order to answer this question, it is necessary to study more carefully
how the different possible relationships between real estate agents and their clients affect allocative efficiency in
the real estate market. In particular, it would be helpful to know, both theoretically and, if possible, empirically,
how buyer’s agency and seller’s agency relationships influence both the aggregate gains from trade and
aggregate search costs in the real estate market. This is an important question for future research.
26
APPENDIX
Proof of Proposition 1: (i) The seller’s optimal pricing rule, say (v*,p*), satisfies the following
necessary and sufficient conditions:
(1.1)
2
∂S ( − v*+ δB)(1− δ( z v*+ (1− z )( p*+ δB))) + δ( z (0.5− δB − 0.5v* + v*δB ) + (1− z ) p*(1− p*− δB ))
=
=0
∂v
(1− δ( z v*+ (1− z )( p*+ δB))) 2
2
∂S (1− 2 p*− δB )(1− δ( z v*+ (1− z )( p*+ δB ))) + δ( z(0.5− δB − 0.5v* + v*δB) + (1− z ) p*(1− p*− δB))
(1.2)
=
=0 .
∂p
(1− δ( z v*+ (1− z )( p*+ δB )))2
Rearranging (1.1) and substituting:
v* = δ(B + S).
(1.3)
Rearranging (1.2) and substituting:
p* + δB = 0.5[1 + δ(B + S)].
(1.4)
Define w* = p* + δB. Using the definitions of B and S and imposing a symmetric equilibrium, we
have:
1
(1.5)
B+ S =
1
∫
∫
z xdx + (1− z ) xdx
z (1− v*2 ) + (1− z )(1− w*2 )
1− δ( z v*+ (1− z )w*) 2(1− δ( z v*+ (1− z ) w*))
v*
w*
=
Using (1.3) – (1.5), we solve for w* and v*:
(1.6)
w*=
(1.7)
v* =
2 + δ(1+ z ) −
4 − 2δ2 (1+ z ) − 2δ(1− z )
δ(3+ z )
4 − δ(1− z ) − 2 4 − 2δ2 (1+ z ) − 2δ(1− z )
δ(3+ z )
,
.
Together with the discussion in the text, (1.7) establishes the result.
z (1− v )δB + (1− z )
(ii) Solving B =
(1.8) B =
1
∫
p + δB
( x − p)dx
1− δ
[ z v + (1− z )( p + δB)]
for B, we obtain
1− δ( z + p*(1− z )) − 1− δ2 − 2δ(1− δ)( z + p*(1− z ))
(1− z )δ2
.
Using (1.6), (1.8), and the fact that w* = p* + δB, we can solve p*:
(1.9)
p *( z ) =
(− δ2 (1− z )2 (1+ z ) − 2(3 + z )(− 2 +
2(1− δ)(2 + δ+ δz ) )+ δ(1+ 11z + 3 z 2 + z 3 − (1− z 2 ) 8(1− δ)(2 + δ+ δ
z ) ))
.
2
2(3 + z )
Together with the discussion in the text, (1.9) establishes the result.
QED
Proof of Proposition 2:
1
z
(i) The seller’s expected price P( z ) =
obtain P( z ) =
∫( x − δB)dx + (1− z) p*(1− p*− δB) .
v*
z (1− v*)+ (1− z )(1− p*− δB)
Using (1.3) and (1.4), we
z(1− δB − δS )0.5(1+ δS − δB) + (1− z )0.5(1+ δS − δB)0.5(1− δB − δS )
= 0.5(1+ δS − δB ) = p*( z ) .
z (1− δB − δS ) + (1− z )0.5(1− δB − δS )
Straightforward (but tedious) differentiation establishes that dp*/dz = dP/dz > 0 when z ∈ [0, 1].
(ii) The probability that a transaction occurs in a match is T ( z ) = z (1− v*( z )) + (1− z )(1− p*( z ) − δB) . Defining,
w* = p + δ
B and using (1.6) and (1.7), T ( z ) =1−
(2 + δ)(1− z ) − (1+ z ) 4 − 2δ2 (1+ z ) − 2δ(1− z )
Straightforward differentiation establishes that dT/dz > 0 when z ∈ [0, 1].
δ(3 + z )
.
QED
Proof of Proposition 3: Define w* = p* + δ
B. Adding and subtracting δ
B from the left-hand side of (3.1),
we obtain:
1
∫
1− F ( w*;α ) δw* xf ( x;α )dx
+
=0 .
(3.2) − w*+
f ( w*;α)
1− δF ( w*;α )
We now show that there exists a w* that satisfies (3.2).
1
∫
δ xf ( x;α )dx
1− F ( w;α )
. It is straightforward to show that I(w) is continuous. Because
+ w
f ( w;α )
1− δF ( w;α)
Define I ( w ) = − w +
1
∫
I(0)=1/(1-α) + δ xf ( x;α)dx >0 and I(1) = -1, by the Intermediate Value Theorem there exists at least one
0
solution w* ∈ [0,1] to (3.2).
Rearranging equation (3.1), which is derived from the first-order condition for the seller’s maximization
problem, and substituting w* = p* + δ
B, we have p* =
consequently exists.
(1− F ( w*;α)(1− δF ( w*;α )
. Because w* exists, p*
(1− δ) f ( w*,α )
QED
TABLE 1: Expected Sales Price, Expected Number of Periods On The Market, δ= .998; f(vt; α) = (1-α) + 2αvt,
vt ∈ [0, 1]
α
Expected Price
Expected Time On Market
(# of periods)
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
0.45
0.50
0.55
0.60
0.65
0.70
0.75
0.80
0.85
0.90
0.95
1.00
.654920
.655436
.655903
.656328
.656715
.657069
.657395
.657695
.657968
.658229
.658469
.658692
.658899
.659096
.659280
.659453
.659615
.659769
.659914
.660052
.660183
28.38
27.71
27.09
26.50
25.95
25.43
24.94
24.48
24.04
23.62
23.22
22.84
22.48
22.13
21.80
21.49
21.18
20.89
20.61
20.34
20.08
TABLE 2: Effect of a Shift from Seller’s Agency to Buyer’s Agency on Expected Price and Expected Time on
the Market
Probability Seller Observes Buyer Valuation (z)
α
0
.05
.10
.15
.20
25
.30
.35
.40
.45
.50
0
.05
.10
.15
.20
.25
.30
.35
.40
.45
.50
.55
.60
.65
.70
.75
.80
.85
.90
.95
1
n.a.
P↑,τ↓
P↑,τ↓
P↑,τ↓
P↑,τ↓
P↑,τ↓
P↑,τ↓
P↑,τ↓
P↑,τ↓
P↑,τ↓
P↑,τ↓
P↑,τ↓
P↑,τ↓
P↑,τ↓
P↑,τ↓
P↑,τ↓
P↑,τ↓
P↑,τ↓
P↑,τ↓
P↑,τ↓
P↑,τ↓
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↑
P↓,τ↑
P↓,τ↑
S,S
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↓
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↓
P↓,τ↓
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
P↓,τ↑
Legend:
(P↑,τ↓) = Expected price rises and expected time on the market falls after a shift from seller’s
agency to buyer’s agency.
(P↓,τ↓) =
Expected price and expected time on the market both fall after a shift from seller’s
agency to buyer’s agency.
(P↓,τ↑) = Expected price falls and expected time on the market rises after a shift from seller’s
agency to buyer’s agency.
TABLE 3: Interpretation of Different Possible Empirical Findings
Expected Time Needed to Sell a
House Increases (τ↑)
Real Estate Prices Rise (P↑)
Real Estate Prices Fall (P↓)
Inconsistent with Model
Sellers Receive Some Information
Under Seller’s Agency,
Distribution of Buyer Valuations
May or May Not Be Better Under
Buyer’s Agency.
Expected Time Needed to Sell a
House Decreases (τ↓)
Sellers Receive Little Information
Under Seller’s Agency,
Distribution of Buyer Valuations is
Better Under Buyer’s Agency (α >
0).
Sellers Receive Some Information
Under Seller’s Agency,
Distribution of Buyer Valuations is
Better Under Buyer’s Agency (α >
0).
Table 4. Hedonic Price Estimates. (The dependent variable is the log of the house’s sale price divided by
the consumer price index for the month of the closing. The numbers in parentheses are t-ratios.)
Explanatory Variables
Intercept
Log of Age of House plus 1 (LAGE)
Number of Bathrooms (BATHS)
Number of Bathrooms Squared (BATHS2)
Number of Bedrooms (BEDS)
Number of Bedrooms Squared (BEDS2)
Beds*Baths (BEDBATH)
Fireplace (FIRE)
Home Warranty (HWAR)
Multistory House (MULTI)
Split Level House (SPLIT)
Is Air Conditioned (COOL)
Air Conditioning is Zoned (ZONED)
Interest Rate one month previous to sale (INT1)
Interest Rate two months previous to sale (INT2)
Unemployment rate one month previous to sale
(UNEM1)
Unemployment rate two months previous to sale
(UNEM2)
Wage rate one month previous to sale (WAGE1)
Wage rate two months previous to sale
(WAGE2)
SUMMER (=1 if house sold in June, July, or
August)
Log of Population Density in the Census Tract
(LPDEN)
Log of median income in the Census Tract
(LINCOME)
Houses Selling for less
than $175,000
(1)
(2)
12.52193
12.45986
(27.35)
(33.15)
-0.05513
-0.05578
(-12.40)
(-12.78)
0.46671
0.47771
(15.22)
(16.45)
-0.06612
-0.07102
(-8.73)
(-13.21)
0.32664
0.32985
(11.22)
(11.78)
-0.03061
-0.03240
(-7.18)
(-8.86)
-0.00437
—
(-0.45)
0.12718
0.12690
(11.60)
(11.61)
0.00919
0.00933
(1.99)
(2.03)
0.00787
—
(1.53)
-0.05985
-0.06427
(-11.85)
(-14.65)
0.03904
0.03711
(3.64)
(3.48)
0.11054
0.11228
(20.04)
(20.99)
0.00172
—
(0.12)
0.00503
—
(0.49)
-0.00192
—
(-0.48)
-0.00711
-0.00879
(-2.02)
(-2.81)
-0.00004
—
(-0.09)
-0.00022
—
(-0.34)
0.01468
0.01393
(2.76)
(3.21)
-0.10129
-0.10151
(-17.47)
(-18.96)
-0.15660
-0.15768
(-4.22)
(-4.27)
Houses Selling for
$175,000 or more
(3)
(4)
11.28079
9.48342
(9.87)
(19.58)
-0.09094
-0.08558
(-8.06)
(-5.71)
0.05047
—
(1.07)
0.03602
0.03653
(5.73)
(5.99)
-0.04117
—
(-0.73)
0.02366
0.01481
(3.26)
(3.68)
-0.04381
-0.03315
(-3.81)
(-3.37)
-0.01236
—
(-0.30)
-0.03900
-0.04251
(-2.96)
(-2.64)
0.07230
0.06137
(1.97)
(2.14)
0.04545
—
(1.10)
-0.01830
—
(-0.61)
0.05625
0.05538
(5.32)
(4.93)
-0.12746
-0.15565
(-1.51)
(-2.03)
0.07002
0.06567
(1.75)
(1.73)
-0.00337
—
(-0.44)
0.00949
—
(1.08)
-0.00146
—
(-0.89)
-0.00052
—
(-0.44)
-0.00078
—
(-0.04)
-0.03951
—
(-1.16)
0.19632
0.25752
(2.69)
(4.95)
Table 4. (continued)
Explanatory Variables
Percent of homes that are owner occupied
(PHOWNO)
Percent of persons over 25 who have a
bachelor’s degree (PBD)
Mean travel time to work of persons in Census
Tract (MTTW)
Percent of people living in Census Tract who are
white (PWHITE)
COBB (= 1 if house is located in Cobb County)
DEKALB (= 1 if house is located in Dekalb
County)
FULTON (= 1 if house is located in Fulton
County)
DCM (=1 if house is located in the Marietta
Census Tract)
DCEC (=1 if house is located in the East Cobb
Census Tract)
DCSS (=1 if house is located in the Sandy
Springs Census Tract)
DCR (=1 if house is located in the Roswell
Census Tract)
DCRA (=1 if house is located in the Roswell/
Sandy Springs Census Tract)
DCFC (= 1 if house is located in the Fulton
County Census Tract)
DCN (= 1 if house is located in the Norcross
Census Tract)
DCB (= 1 if house is located in the Buford
Census Tract)
DCD (= 1 if house is located in the Duluth
Census Tract)
SA (= 1 if house closing occurred before 4/1/94)
SA*LAGE
SA*FIRE
SA*HWAR
SA*MULTI
SA*ZONED
Houses Selling for less
than $175,000
(1)
(2)
0.00094
0.00098
(2.62)
(3.16)
0.00996
0.00999
(17.50)
(18.29)
-0.01364
-0.01343
(-12.28)
(-12.15)
0.00022
—
(0.23)
0.12156
0.12135
(6.20)
(6.24)
0.21411
0.21389
(9.96)
(9.97)
0.17654
0.17656
(8.87)
(8.95)
0.00748
0.00887
(0.77)
(0.91)
0.04408
0.04432
(6.82)
(6.89)
0.10060
0.09909
(4.08)
(4.06)
-0.05101
-0.05063
(-4.40)
(-4.37)
-0.16315
-0.16262
(-18.41)
(-18.41)
-0.26962
-0.27075
(-19.67)
(-21.83)
0.03987
0.04080
(2.11)
(2.16)
-0.01980
-0.01858
(-0.98)
(-0.93)
0.00594
0.00786
(0.31)
(0.42)
-0.71660
-0.55680
(-1.75)
(-2.20)
-0.01010
-0.01030
(-1.88)
(-1.92)
-0.03947
-0.04006
(-2.18)
(-2.21)
-0.01410
-0.01441
(-1.93)
(-1.98)
—
—
0.01539
(1.91)
0.01561
(1.95)
Houses Selling for
$175,000 or more
(3)
(4)
-0.00488
-0.00569
(-4.92)
(-11.52)
0.00257
—
(0.96)
-0.01119
-0.01093
(-4.51)
(-2.44)
0.01290
0.01703
(2.63)
(4.91)
-0.06399
-0.10852
(-1.39)
(-2.34)
-0.10819
-0.14572
(-1.72)
(-1.40)
-0.03559
-0.06755
(-0.61)
(-0.77)
-0.05139
-0.06135
(-1.69)
(-1.41)
-0.02068
-0.01504
(-0.87)
(-0.48)
0.15493
0.17514
(4.79)
(6.98)
-0.11186
-0.10540
(-5.29)
(-2.96)
-0.11244
-0.08645
(-5.92)
(-2.42)
-0.05969
0.00683
(-1.14)
(0.21)
-0.19901
-0.22325
(-3.02)
(-2.28)
-0.04922
-0.09198
(-0.41)
(-0.75)
-0.21969
-0.24816
(-3.78)
(-3.17)
-1.53243
-1.46776
(-1.83)
(-2.18)
—
—
—
—
—
—
-0.08054
(-2.73)
—
-0.08102
(-2.63)
—
Table 4 (continued)
Explanatory Variables
SA*INT1
SA*WAGE2
Houses Selling for less
than $175,000
(1)
(2)
—
—
0.00133
(1.55)
—
0.00097
(1.88)
—
SA*LPDEN
—
—
SA*PHOWNO
—
—
SA*PHOWNO
-0.00067
(-2.37)
0.00209
(1.93)
—
-0.00067
(-2.49)
0.00215
(2.27)
—
SA*DEKALB
—
—
SA*FULTON
—
—
SA*DCFC
—
—
SA*DCN
—
—
SA*DCB
—
—
SA*DCD
—
—
Sample size
R2
8,110
0.6672
Adjusted R2
S. E. of Regression
SA*SUMMER
SA*PWHITE
SA*COBB
Houses Selling for
$175,000 or more
(3)
(4)
0.23799
0.26781
(2.69)
(3.40)
—
—
-0.07103
(-1.89)
-0.03176
(-1.62)
0.00171
(2.59)
—
-0.05025
(-2.33)
-0.05129
(-3.27)
0.00183
(2.67)
—
8,110
0.6669
-0.00647
(-1.89)
0.46856
(7.27)
0.44750
(6.68)
0.45063
(7.10)
-0.08771
(-2.40)
0.49330
(6.96)
0.69921
(2.02)
0.53382
(5.25)
4,274
0.3590
-0.00845
(-2.41)
0.50053
(6.87)
0.48366
(6.31)
0.46755
(6.93)
-0.10499
(-3.01)
0.51525
(6.38)
0.70495
(2.03)
0.54541
(5.92)
4,274
0.3367
0.6653
0.6653
0.3512
0.3306
0.1540
0.1540
0.2706
0.2749
Table 5. Predicted Changes in the Real Sales Price for 13 Census Tract Areas by Price Range.
Part I. Houses with Real Selling Prices of $175,000 or more
Predicted Price Level:
Census Tract Area
Marietta
East Cobb
Other Cobb County
Sandy Springs
Roswell
Roswell/Atlanta
Fulton County CT
Other Fulton County
Dekalb County
Norcross
Buford
DCD
Other Gwinnett County
Observations
in Sample
79
746
180
308
319
699
487
800
314
281
14
37
10
Before 4/94
$224,316
$278,502
$235,307
$314,540
$286,138
$289,889
$288,427
$302,891
$259,579
$274,493
$361,840
$272,654
$198,347
After 4/94
$199,644
$241,570
$203,094
$274,357
$263,548
$246,599
$260,292
$272,714
$231,018
$231,604
$240,149
$208,427
$254,198
Change
-$24,672
-$36,932
-$32,212
-$40,183
-$22,590
-$43,291
-$28,135
-$30,176
-$28,561
-$42,889
-$121,691
-$64,227
$55,851
Percent
Change
-11.0%
-13.3%
-13.7%
-12.8%
-7.9%
-14.9%
-9.8%
-10.0%
-11.0%
-15.6%
-33.6%
-23.6%
28.2%
t-ratio
-1.99
-2.45
-3.34
-2.05
-1.41
-2.72
-1.67
-1.84
-2.63
-3.50
-1.18
-2.83
2.75
Change
-$701
-$524
-$853
$327
-$2,975
-$2,548
-$1,652
-$489
$534
-$284
-$2,118
-$3,147
-$1,534
Percent
Change
-0.66%
-0.38%
-0.81%
0.23%
-2.51%
-2.05%
-1.21%
-0.35%
0.36%
-0.29%
-2.27%
-2.92%
-1.47%
t-ratio
-0.36
-0.20
-0.42
0.12
-1.30
-1.07
-0.57
-0.18
0.18
-0.15
-1.15
-1.48
-0.74
Part II. Houses with Real Selling Prices less than $175,000.
Predicted Price Level:
Census Tract Area
Marietta
East Cobb
Other Cobb County
Sandy Springs
Roswell
Roswell/Atlanta
Fulton County CT
Other Fulton County
Dekalb County
Norcross
Buford
DCD
Other Gwinnett County
Observations
in Sample
651
1,231
1,631
58
291
1,266
436
642
360
789
215
411
129
Before 4/94
$105,569
$136,752
$104,632
$141,940
$118,736
$124,246
$136,661
$140,804
$150,122
$97,202
$93,183
$107,691
$104,509
After 4/94
$104,868
$136,228
$103,778
$142,267
$115,761
$121,698
$135,009
$140,315
$150,656
$96,918
$91,065
$104,543
$102,975
Table 6. The Results of the Estimation of Equation (14).
Explanatory Variable
Percentage of listing time that
occurred after 4/1/94
Whole Sample
-1.959493
(-41.00)
Houses with Real Sale
Prices ≥ $175,000
-2.205609
(-27.07)
Houses with Real Sale
Prices < $175,000
-1.785733
(-28.19)
R2
F-statistic
Probability level for F-statistic
Number of observations
0.145644
160.6664
0.000000
12,380
0.191967
55.99073
0.000000
4,273
0.115769
91.9177
0.000000
8,107