Digitized by the Internet Archive
in
2011 with funding from
Boston Library Consortium IVIember Libraries
http://www.archive.org/details/testingforpriceaOOashe
MIT
3 =,oaO
UBRARIES
00770*=I16
6
working paper
department
of economics
Testing for Price Anomalies in Real
Estate Auctions
Orley Ashenfelter
David Genesove
No.
598
Jan.
1992
massachusetts
institute of
technology
50 memorial drive
Cambridge, mass. 02139
Testing for Price Anomalies in Real
Estate Auctions
Orley Ashenfelter
David Genesove
No.
598
Jan.
1992
MAT
MAR
IJBRARIES
9 1992
"
.
MIT Economics Working Paper 598
TESTING FOR ERECE ANCMALIES IN REAL ESTATE AUCnOJS
Orley Ashenfelter
and
David Genesove
Princetcai University and MIT
ABSTRACT
This p^)er r^»rts on the results of an auction sale of 83 aondcminium
apartment units in New Jersey. At the auctiOTi every unit was hammered
dcvm, but, unknown to the 2,348 registered bidders, 40% of the sales fell
throuc^. Prices in the subsequent sale of oondaninium 'onits in face to
face negotiatioTS resulted in identical units selling for 13% less than
they fetched at auction and the discount was largest for those units
haininered dcwn early in the auction.
These results are inccaTsistent with
the usual predictions from the theory of cannon value aucticxis and suggest
that uninformed bidders in this auction may have been the subject of a
"winner's curse" v*iich generated considerable profit for the seller.
The authors are eiffiliated with Princetai Iftiiversity and the Natiaial
Bureau of Econcndc Research (Ashenfelter) and MIT (Genesove)
This p^jer is a preliminary r^»rt of the results of a larger study of
condcDdnium real estate aucticxis and was prepared for presentaticsi at the
Meetings of the American Econanic Association cxi January 3, 1992 in New
Orleans, in a sessicai entitled "The Empirical Study of Auctions: In Hcaior
of William Vickrey.
TESTING FOR FRICE ANCMALIES IN PEAL ESTATE ALJCnOJS
Orley Ashenfelter and David Genesove
In this ]paper we r^xjrt the results of a study of ccndaniniim prices.
Our study cxupares the prices paid in face to face bargaining with the
prices fetched for identical condcodnium units sold at auctic^.
results are striking.
Ihe
Our findings indicate that auction prices for
identical units were 13% hi^ier than for units subsequently sold in face to
face bargaining.
Moreover, the price decline obtained by face to face
bargainers was not independent of the order in
auctioned.
v^ch
the units were
Face to face buyers achieved hi^ier discounts relative to
auction buyers
vdio
purchased early in the auction.
Taken together these
results indicate that the optimal strategy for a risk neutral ccndcndnium
biQ^er is
to make a purchase well cifter the auctic^ has begun and, ideally,
after the auction has been oorpleted!
I.
ThB Price Decline Ancnaly and the Data
We collected the data for this study at an aucticxi of 83 ccndoninium
units held near Princeton, New Jersey in
^ril of
1990.
CXir
intention was
to record the winning bids and the order in vAiich the units were sold with
an eye to detezmining vAiether condctninivim aucticxis also shew evidence of
uepartment of Eoononics, Prinoebcn IMiversity, Prinoetcn, NJ 08544
and Dqartment of Eoancmics, MIT, Cai±)ridge, MA 02139, respectively. We
thank Kevin HcLllock for exenplary research assistance and Alan Knieger and
Bruce Vanderporten for helpful ccranents.
the "price decline ancnaly" noted by Ashenfelter (1989) and McAfee and
Vincent (1991) for wine avictions.
In these auctions it has been
established that prices for identical objects are more likely to decline
is not viiat is predicted for the behavior of risk
than to increase,
viiiich
neutral bidders.
One possible eoqjlanaticn for this behavior is that
bidders are risk averse and early bidders pay a premium to avoid the risk
associated with losing out on an item.
Wine
aix±ica-is
are ideal for
constructing a test of the '"price decline ancraaly" because identical items
are sold ccaisecutively.
Ihis permits a direct test of the hypothesis
without concern about the possibility that emitted quality characteristics
may bias the findings.
At the same time, wine is a very ^lecialized
ccmmodity and it seems desirable to examine evidence fron other markets.
The growth in the sale of relatively hcnaogenecus ccaidcodnium units by
aucticxi seems to offer an ideal opportunity for further tests of the price
decline aronaly.
Unlike different cases of the same wine,
identiccil.
hcxniever,
ccsidcininiums are not
Perhaps because the units are heterogeneous, real estate
auctions operate by a method different from the usual English aucticai.
(See Vanderporten (1991), for exanple.)
Condominium units are usually sold
in a "pooled" or "ri^t-to-choose" auction v*iere all the units are ccnobined
together.
desired.
When the bidding stops the hi^iest bidder may choose the unit
Ihe bidding is then re-started and continues until all the units
in the pool have been sold.
In a "pooled auction" it is inevitable that
any heterogeneity in the quality of the units offered in the pool will
cause the prices of the items in the pool to decline.
earlier bidders are biding the
ri^t to
In effect, the
select si;¥)erior products and it is
natxaral that they should pay higher pricses for them.
Price declines are
thus a predictable outocme of the pooled auction design and do not, by
themselves, constitute an
anccicily.
Ihe mere fact that prices may be expected to decline in a pooled
auction does not mean that a price decline ancnaly is not cLLso present.
Indeed it has sonetimes been su^ested that aucticsieers use devices like
the pooled auction to make it more difficult for buyers to perceive the
presence of ancnalies that may cast doubt on the integrity of the aucticai
process.
Ihe question is, hew much would we have e>qpected prices to
decline because of unit heterogeneity?
If prices decline more than would
be expec±ed, then we have evidence for the declining price ancnialy.
CXur eocaxxnetric
design for solving this problem is simple:
First
(1)
we determine the relationship between the auction price and the order of
Scile,
and
(2)
then we track down subsequent resales of the same ccndcminium
units in face to face bargaining.
If the price decline ancmaly was truly a
result of the auctioi mechanism, then any relationship between the auction
bid price and the order of sale (at the auction) would dis^^pear
item was resold.
viien
the
If, on the other hand, the relationship between the
auction price and the order of sale was due to an emitted quality
characteristic, then the subsequent price in face to face bargaining would
still be related to the order of sale at the auction held earlier.
In the
intermediate case, we may subtract out the relationship between quality and
the order of sale by using the relationship between these two variables
under face to face bargaining.
In practice the observed price decline may
be due to both unobserved quality differences and ancracdous price declines,
and belew we separate these two eooncmetrically.
When we began this project
years before
cill
vie
ejqjected that it vrould take several
of the candaninium units sold at auction were resold in
face to face bargaining.
Much to our surprise, v*iile checking (at the tax
assessor's office) the Scile prices for all the ccndoninium units "hainnered
down" at the auction we discovered that 37% of the units had not, in fact,
been sold at the "haMner price."
Instead, these 31 units had all been sold
at discounts fran the bid prices at
the auction.
viiicii
they had been hanmered dcwn in
In a series of subsequent interviews we learned that these
units had typically been resold to another hayer after the original auction
Scile "fell
through."
Althoui^ we were able to interview only a fanall
number of these subsequent buyers it e^jpears that
th^
typically
constituted a group of registered, but unsuccessful bidders
\it)o
were
subsequently contacted by the aviction ccrtpany to negotiate a sale in the
week following the auction.
vMch
In short, many of the identical units for
successful bids were established at the aviction were resold in face
to face bargaining a few weeks after the auction.
were three weeks apart. )
(Ihe average sale dates
This provides a remarkable opportunity to ccnpare
the prices of identiccil objects sold at virtually the identical time by two
different pricing mechanisms.
II.
The Enpiriccil Results
Table 1 shows the history of prices in the Oolcainade Pointe
condaninium develcpnent outside Princeton, New Jersey.
Rather than adjust
the prices for square footage we have s^arated and r^jorted the data for
the three different type \jnits (Arbor, Belvedere, and Cloister) in this
development.
Units within these groins are identical except for their
p
o CM
o n
u o
w o
M in
o
u
a
o
s
Q,
O
iH
9
>
O
Q
«
P
o
>1
u
o
•p
n
w
a
a
a
>
M
Pi
(M
"^
CM
n
o
in
VO
CM
i/y
H
H
H
en
r^
n
<T>
(Tl
in
CO
in
o
iH
CM
V
CM
in
i/y
r-i
i/y
>*•
«.
«k
in
H
H
i-l
</>
rH
rH
n
M
r~
CM
r-t
in
en
^
r-
v>
CTl
^
VO
VO
CO
CM
in
r-\
^
"*
vy
to-
«>
<rt-
O o
o 00
o
CM
VO
H
H
H
H
v>
vy
VO
00
00
'4'
o
o
CO
CO
O n
CTi
CM
i/y
r^
in
(j\
i/y
in
i/y
H m m
in
n CM
VO
H
(Ti
vo
VO
rH
en
'i*
^
CM
CO
i/y
CM
i/y
i/y
CQ
10
ID
0)
73
0)
•H
CO
»
u
M
CM
in
VI-
•H
•H
O
ja
03
-p
< CM
H
0)
w
C
c
o
•H
u
o o
o
O o
CQ
in
in
H
in
CO
CN
o
CM
CM
-P
3
<
o
o
CM
H
n
CM
(Tl
0)
00
00
jC
i/y
rH
W
•H
in
H
H
H
-p
o
s
CM
in
o^
H
^
in
vo
in
(0
o
o
in
CM
in
CO
CO
v>
n
H
ify
v>
i/y
Eh
tH
v>
-p
ify
H
H
O
A
-a
Id
•H
TJ
0)
•H
o
u
O
>^
0)
0^
0)
-H
0)
c
H
4i
c
o
-H
•e-t
+J
p
Id
«3
(0
•H
•H
>
0)
>H
04
c
•H
Q
0)
o
•H
H
>
>
Q
0)
0)
0)
T3
c
c
•H
-H
-p
+J
«J
(0
•H
•H
>
Q)
U
Q)
>
0)
Q
•H
Q
-H
Q
•
(U
•
04
•
•H
0^
c
-H
^
Ul
< 2
a<
•
cu
73
c
•
73
CQ
73
c
c
G
C
(C
(0
(0
rtJ
0)
+J
s w z
(U
•P
s W Z
73
73
C
c
C
c
C
(0
rO
rtJ
ro
(0
(0
0)
+J
(U
-P
CO
s W 2
s
2
P
s w
Q)
-p
c
3
O
o
-H
a
— —
m
c
0)
Si 0)
-p ><
Q)
-p
(0
(T3
n
O-P
en
01
.
c
T3
H
3
a)--'
"^
^rt
u
S
ji:
°^Q)
x:
4-)
•H
H
tfl-O
s
-§
3
^^^
ai
XI
c
bidde
•H
price
p
(bid*1.03)
igh
s
0)
J3
-P
P
0)
J3
';:3
IT3
Q)
(T3
^
^
TJ
ii
73
0)
^
,
-^^
c
<u
^
a
0)
rH
>-l
•
(-1
<*-!
n 8.1 S
4-1
T!
Q)
0)
3
+J
X
c
0)
Ul
Q)
v^
dJ
v^
rO
01
(0
q
-p
^ -H
4J
en
•H
(U
.
rH T3
u
0)
u d
3 H
x:
X
OJ
T3
C
nj
0)
T3
nJ
<u
>i
0)
4-1
(0
T3
^
c
3
(U
3
r-H
(U
nl
0)
•H
>^
8
B
P
JJ
-^^
3m
9h
'"3
0)
"5
<w
C O
s
0)
^^
W
o +J
n
^^ w-^
J5.H C
+J ^ 3
(U
Q)
"-•
o
(a
?*
01
-P
-,
T3
*
Qj
a
Q)<u
C
13
V4
x: Q)
4j X!
-P
e
0)
0)
H
(U
4-1
4J
•H
+J
E-t
0)
c ^
> C
j3
01
-dill's
«3
4-1
Ul
0)
-D
Q)
>4/
r-l
rn
_ 3
01
(T3
4J
0)
0)
u
0)
JC!
^5S"?^
>H-^
<0
m
(0
o-u
>i3 "
+J
T3
M-l
C W
x: '^
^^
0)
-p
IH
(0
J3
cQ *-
<u
^
C
O
C W-H
(1)
Bo
«: -p
XS
+J
u
Q)
.-"^-^
fM r)
n
3 >
«J O
location, vAiich may differ becaxjse of the floor the unit is on or because
of the view.
As the table indicates, during the period Octcter 1987
throu^ the fall of 1989 about
cneHiialf of the 252 units in this
developnent were sold at prices very close to the list prices at vAiich the
units were originally put
cai
1989 until the auctioi in
^ril
the market.
In the period frcm the Fall of
1990 the market for these units
deteriorated, and another 10% of the units were sold at prices vAiich
averaged seme 10% to 15% Icuer than the list prices.
At the auction the
bid prices averaged sane 20% to 30% below the origix^al list prices.
Finally, about 12% of the units were resold after the auction at an average
discount of 13% frcm the bid prices, or a discount of seme 30% to 40% frcm
the original list prices.
It is our irnpressicn that the overall movement
in these prices is a fair reflection of general movements in real estate
prices for properties of this kind.
The data frcm the auction sale in
are plotted in Figure 1 in panels A-C.
v^ch v^
are primarily interested
Every vmit was "hammered down," and
the bid prices (indicated with the symbol "+") at the auctiai are displayed
in the order in v*iich we recorded them.
(In fact, the Cloister and
Belvedere units were sold in two s^arate "pools" and the second pools
began with units 25 and 17; this accounts for the two ^ikes in panels A
and C in the Figure.)
these bid prices.
Uhits actually sold at the auction were sold at
However, 31 of the units hammered dcMn were not, in
fact, sold at the bid prices.
Also displayed in Figure 1 are the prices
(indicated with the symbol "0") for the units
auction.
vMch
were resold aifter the
As the Figure indicates, the units that were resold in face to
face bargaining fetched prices substantially below their bid prices at the
o orice
+ Dia
o price
+ Did
130000 -
120000
-
+
*
120000
+t^
-
+
HOOOO
-
100000
-
+
+ +++
llOOOO
+
\-*
-
.
o
100000
f
-t-
-f
+
+
+
+ + +
"*"o
oo
-
o
o
90000
-
°°°
oo
90000
o
-
o
10
5
Q
80000
25
20
15
°
5
10
-t-
1
10000
lOQOOO
25
20
15
o
30
35
Oraer
Order
A:
o
-
35
30
o
Cloisters
B:
Belvederes
price
Qid
-1
-
3
.05
-
+
90000
-
eooQO
-
70000
-
+
+
t-
+
+
+
+
+
o
o
o
o
o
15
-15
Order
C:
Figure
1:
ArDors
-10
-5
5
10
15
Order Residual
0:
Stanaaroized Price Discount
vs.
Order
Bid and Prices vs. Order, by Type of Unit
STara"
aix:tion.
Ihere are three inportant findings of our research that are readily
First, the bid prices typiccilly declined as the
observable in Figure
1.
aucticai progressed.
Second, the prices of the items resold after the
auction shew little or no relationship with the order in viiich the units
were sold in the auction.
Note also that the units that were resold are
scattered randcmly throu^ the auction.
of Figure
1,
Finally, as indicated in Panel D
the price discount for subsequent resales was largest for
those units sold earliest in the auction.
In sum, these data provide
strong evidence that the early auction bt^ers peiid a premium for the vmits
they purchased that does not reflect their hi^ier quality, at least as
judged by resale prices.
More formal tests of these hypotheses are contained in Table
Column
2.
of the table r^xsrts a regression of the logarithm of the bid
(1)
price on dummy variables indicating the unit type and the numerical order
in viiich the vmit was sold.
The data indicate that the bid price declines
about .27% per unit sold, or about 10% from the beginning of the auction to
the end.
Column
(2)
reports the bid price regressicxi for those units
actually sold at the auction, v4iile column
(3)
r^xarts the bid price
regression for those units that were not sold at the auction.
These two
regressions hctve virtually identical coefficients (P=.3 for a test of their
difference)
,
which indicates that there is no ccnnectic^ between the
bidding behavior of the auction participants and the reascxi the sale was
not finalized.
(In fact, a probit function fit to e3q)lain the probability
that a sale falls throu^ indicates no statistically significant
relationship with the condcminium type or the order of sale.)
,
^-^
,„0m^
VO
(N
CO
03
T3
•
•
-H
VO
t^
rH
*
o
^-^
o
n
o
•
•
00
OD
(N
O
O
CO
O
O
O
^-v
CN
^
CO
H
rH
•
*
0)
o
•H
'S'
,-»
Pu
CM
O
T3
o
CM
CN
..-V
.-^
o
O
(Tl
O
O
O
O
O
VD
CO
CM
'3'
•H
in
n
CQ
o
H
to
ca
M
o
0)
P
o
•H
u
Id
h^
o
in
<-«
CM
O
cr>
H
fVJ
O
CM
H
CM
O
O
O
O
O
O
O
n
CO
n
o
o
o
en
CO
n
•
•
o
(3
«
A
O
—
•i-l
•P
r^
U3
u
3
C
o
W
•H
O
1-1
«
CM
O
w
in
CM
+J
0)
U
3
rH
<:
(0
+J
C
W D
n
-H
CN
CM
O
w
•
-H
(M
^^
.
I
^i^
«-«
rH
O
r^
CM
CM
•
•
•
•
^—
r)
rH
•
CO
rH
^-^
I
O
CM
O
O
cn
o
o
•
•
CO
I
H
O
>
(N
O
O
•
•
^-^
in
rH
^-1
>-'
I
in
o
o
o
VO
CO
•
•
—
in
0)
(0
a>
QQ
X!
+J
C
0)
v^
c
o
•H
^-^
^0m^
W
00
rH +J +J
T3
•H
< 3 C
< E3
•
H
O
CM
•
•
'O
rH
O
^—^
"a"
H
O
•
•
t^
CM
O
O
a
"i*
O
O
o
Hc
00
•
03
cn
u
o
u
u
4J
..
C
<0
(U r-{
p
c
-o X>
(C
C (0
Q) -rH
atH
4J
0)
nj
Q>
u
-H
^
Ul
0)
0)
T3
u
0)
ja
rH
CO
c
0)
u
<
>
0)
CQ
0)
U
0)
rH
•H
g
>H
(0
a
0)
-0
Vh
O
04
(S
p<
w
Vh
(0
T3
c
(0
-p
w
Column
(4)
of Table 2 cxDntains the regression of the logarithm of the
ac±ual price of the resold units on the order in which the same units were
hammered dcwn at the auction.
If the declining bid prices reflected in the
"order" coefficient in column
(3)
were a result of \jncbserved quality
differences only, then the coefficient of the order variable in column
should be the same as it is in column
"order" variable in column
(4)
(3)
.
(4)
In fact, the coefficient of the
is only -.06% per sale and it is not
significantly different frcm zero, indicating that no more than 25% of the
price decline is due to unobserved quality differences.
case, the regressicai in column
(5)
As must be the
indicates that the bid price/sold price
discount is greater for the units hammered dcwn earliest in the aucticxi.
Finally, the regression in column
the regression in column
(5)
,
,
vAiich is
an unrestricted versicai of
iixiicates (as must be the case since the
order of sale is uncorrelated with
correlated with the order in
(6)
vMch
seile price)
that the bid price is
the vmit was hammered down even after
controlling for the price at vAiich the \init was ultimately sold.
Ill .
Inpl ications
Ihe data in this paper indicate that 37% of the condoniniums hananered
dcwn in the aviction we attended were not, in fact, sold.
Ihese units were
sold a few weeks later in face to face bargaining, but at considerably
Icwer prices than the identical units fetched at the auction.
Iforeover,
the discounts these resold units fetched were greater for those units
hammered dcwn earlier in the auction, confirndng the "declining price
anomaly."
We think these enpirical results should surprise most econcndsts.
8
After all, more than 14,000 pecple visited the 83 imits on
2,348 pecple registered as bidders.
and fully
seile,
Moreover, the coTdoninium units
cai
sale, vAiich are located on U.S. Hi^iway 1 behind a shopping mall,
constitute a small fraction of the thousands of similar housing units
located in the same area.
Finally, many of these units were purchased by
investors for the purpose of rescile and those bought for occupancy will no
doubt be sold by their owners within a few years.
In short, the bidders in
this "ccOTnon value" auction (vAoere it is not known, at the time of the
auction, the price an item will fetch in its subsequent use or scile) should
have been expected to pay prices similar to the prices the units v^suld
fetch vdien they were resold.
The
hi^
prices that aucti<^i bidders paid for
these condominiums opens up the possibility that they were the subject of a
"winner's curse" (vAiere buyers construct point estimates of the value of a
property and
feiil
to shade their bids in anticipation of estimation error,
as in Kagel and Levin (1986)
.)
More fundamentally, these data indicate that the market mechanism had
a substantial effect on the prices at vdiioh these condominiums were sold
and they raise some
de^
questions about the strategies and infonnation the
bidders brought to this auction.
After the fact, it is obvious that the
optimal strategy for a bidder would have been to tie vp a ccxTdominium with
a
hi^
bid, and then renege on its sale and renegotiate a
evidence from this aucticxi indicates that few, if
strategy.
arr/,
loKi^er
price.
Ihe
btyers followed this
We suspect that most buyers did not engage in this strategy
because they were not aware of it.
Ihis suggests that the auction seller
has a considerable informational advantage over the buyers in a real estate
auction, and that this advantage may have been put to good use in recent
years to generate auctioneer profits at the expense of uninfonned buyers.
REFERENCES
Ashenfelter, Orley, "How Auctions Work for Wine and Art, " Journal nf
Econcmic Perspectives
.
Sunmer 1989, 2r 23-36.
Kagel, John and Levin, Dan, '"The Winner's Curse and Public Infonnaticai in
Cannon Value Auctions,"
The American Econcmic Review
.
December 1986,
76, 894-920.
McAfee, R. Preston and Vincent, Daniel, "Ihe Aftemocxi Effect," CMSIMS
Discussion P^jer No. 961, Northwestern Uhiversity, Evanstcsi, Illinois,
October 1991.
Vanderporten, Bruce, "Timing of Bids at Pooled Real Estate Aucticxis," The
Tntimal of Real Estate Finance and Eoonanics . forthocming, 1992.
10
J J c
u
Date Due f-^^-^z-
MI, LIBRARIES DUPL
3
^OflO
OO??^!?