United States
Environmental Protection
Agency
Research and Development
Air and Energy Engineering
Research Laboratory
Research Triangle Park NC 27711
EPA/600/S7-85/034 Dec. 1985
Project  Summary
Inferring Willingness to  Pay for
Housing Amenities from
Residential  Property Values
Joel L Horowitz
  The standard hedonic model of con-
sumers' choices in housing markets,
which forms the basis of much envi-
ronmental benefits analysis, fails to
take into account  three  important
characteristics of the bidding process
through which houses are sold. First,
the prices of houses are established
through sequential bidding, so sellers
do not necessarily accept the highest
bids that might be  made  for their
houses. Second, ignorance of the mar-
ket may lead potential buyers to make
bids that are either higher or lower than
the ones they would make if they had
complete information. Finally, the dis-
tribution of bids for a house is truncated
at the seller's asking price. These char-
acteristics of housing markets  imply
that buyers cannot choose houses so as
to maximize a deterministic utility func-
tion subject to a deterministic budget
constraint as required by the standard
model. This  report  describes a new
model of consumers' choices in housing
markets that incorporates the foregoing
market characteristics. The new model
has been tested econometrically against
the standard model, and the standard
model has been strongly rejected. The
new model gives estimates of buyers'
willingness to pay for housing amenities
that are significantly different from the
estimates produced by  the standard
model.
  This Project Summary was developed
by EPA's Air and Energy Engineering
Research Laboratory. Research Triangle
Park, NC. to announce key findings of
the research project that is fully docu-
mented in a separate report of the same
title (see Project Report ordering infor-
mation at back).

Introduction
  Econometric estimation of households'
willingness to pay for environmental and
other residential amenities often is based
on observations of the effects of envi-
ronmental and other factors on residential
property values. The standard estimation
procedure is based on the hypothesis that
each household occupies a house that
maximized its utility subject to a budget
constraint. It can be shown that, subject
to regularity conditions, utility maximiza-
tion occurs when the household's mar-
ginal willingness to pay for each housing
attribute (e.g., house size, air quality at
the residential location) equals the mar-
ginal price of that attribute.
  In the standard procedure, willingness-
to-pay estimation takes place in two steps.
First, a hedonic price function describing
the relation between the observed market
prices of houses and attributes such  as
size and air quality is estimated. The first
partial derivative of the hedonic price
function with respect to an attribute level
yields the marginal price function of this
attribute. The numerical marginal prices
pertaining to a particular household can
be obtained by substituting the values of
the attributes of that household's house
into the estimated price function. The
resulting marginal price of the attribute is
then treated as an observation  of that
household's marginal willingness-to-pay
for the attribute. In the second estimation
stage, the relation between marginal
willingness-to-pay and relevant  explan-
atory variables is specified up to a finite

-------
set of constant parameters. The param-
eter values are inferred from  the "ob-
served" marginal willingness-to-pay val-
ues (i.e., the estimated marginal  prices)
using  suitable econometric methods.
  This hedonic pricing approach has been
used in a variety of studies of willingness-
to-pay for environmental improvements.
However, the method has some serious
difficulties.  Three that are particularly
relevant to the research described in this
report are:

  1.  The  assumptions  underlying the
     standard hedonic model imply that
     houses are sold to the households
     that are willing to pay the highest
     prices for them (i.e., houses are sold
     to the highest bidders). However, it
     is unlikely that houses are  sold to
     the highest bidders in real housing
     markets. In  real markets, houses
     normally are sold through sequen-
     tial bidding processes. Bids must be
     accepted or rejected as  they are
     received, and a seller is not able to
     assemble all  bids that might be
     made before deciding which one to
     accept. Therefore,  the seller does
     not necessarily accept the highest
     bid that might be made. Substantial
     transactions costs tend to prevent
     the rectification  of erroneous de-
     cisions regarding the rejection and
     acceptance of bids.
  2.  Potential buyers bid for houses with
     little or  no knowledge of the bids
     made by others. Therefore,  buyers
     do not know the price  at which a
     house will be sold in advance of the
     sale, and buyers have  incomplete
     information  about the house price
     components of their budget con-
     straints.  This makes it impossible
     for buyers to choose  houses  by
     maximizing utility subject to deter-
     ministic budget constraints. Trans-
     actions costs tend to prevent buyers
     from rectifying any resulting errors
     in bidding or in choosing houses.

  3.  The sale price of  a  house rarely
     exceeds the seller's  asking price.
     Thus, the asking price effectively
     truncates the distribution of bids. If
     ignorance of the  market or other
     factors cause  sellers to  establish
     asking prices  that are below the
     prices buyers  are  willing to  pay,
     then truncation of the distribution
     of bids  at the asking  prices will
     prevent identification of the buyers
     willing to make  the  highest bids.
     Therefore,  houses will not neces-
     sarily  be sold to these  buyers.
     Moreover,  a  process  in  which
     buyers choose houses by maximiz-
     ing utility subject to budget con-
     straints  will  not  assign  unique
     buyers to houses.
  This  report describes a model of con-
sumers' choices in housing markets that
incorporates the market characteristics
just discussed. The  new model  and  the
standard model have been estimated and
tested  econometrically  to  determine
which provides a better description of the
operation of these markets. The formula-
tion and testing of the new model and the
implications of this model for willingness-
to-pay  estimation  are summarized  in
report Chapter 1. The remaining chapters
are more detailed discussions: Chapter 2
gives detailed mathematical formulations
of the new and standard models;  Chapter
3  presents  econometric tests  of  the
models;  and  Chapter 4 presents  the
willingness-to-pay analysis.

Formulation of the Models
  In this research, the new and standard
models have been formulated as bidding
models.  Bidding models  are housing
market models in which the dependent
variable  is the sale price of a  house,
modeled as a function of attributes of the
house  and the buyer. It is assumed in
these models that sales occur as results
of a bidding process.
  The  standard model used in  this re-
search is the well-known bid rent model
of housing markets. Roughly speaking,
this model assumes that houses are sold
to the buyers who bid the most for'them,
and a house's sale price is the amount the
highest bidder is willing to pay  for that
house. Under the customary assumptions
of hedonic analysis,  the bid rent model is
equivalent to the model in which buyers
choose houses to maximize utility subject
to a budget constraint. Moreover, under
the same assumptions, willingness-to-
pay analysis based on the bid rent model
is equivalent to analysis based on a utility
maximization approach. Thus, the stand-
ard market  model can be tested using
either  the bid  rent or the utility  maximi-
zation  form. The bid rent form  is more
convenient  in the  research  discussed
here.
  In the version of the bid  rent model
used here, the probability density that a
house  with observed attributes X has sale
price p8 is:
Density (p.) = f[pjW(X)],
(1)
where W(X) denotes the buyer's willing-
ness to pay for  a  house with observed
attributes X, and f is the function giving
the probability  density that  ps  is the
highest bid made for the house.
  In the  new  model, the  probability
density that a house with observed attri-
butes X and asking price Pa has sale price
Ps is:

Density (ps) =
Q(pa|Pa)g[pa|W(X)]|P(sale)             (2)

if ps < Pa,  where g  is the  probability
density that the buyer makes a bid of ps,
Q(ps|Pa) is the probability that the seller
accepts a bid of size ps, and P(sale) is the
probability that the house is sold  at any
price. The probability that ps =Pa is:

ProbfPa) = {1 - G[Pa| W(X)]}/P(sale),    (3)

where  G  is the cumulative  probability
distribution function corresponding to g.
The probability that ps > Pa is zero. P(sale)
is given mathematically by:

P(sale) =
/£• Density(p)dp +  {1 - G[Pa|W(X)]}.   (4)

  Equations (2-4)  constitute a bidding
model  of consumers' choices in housing
markets that is analogous to the standard
model  but that is based on different
assumptions concerning the bidding pro-
cess that operates in housing markets. In
particular, Equations(2-4) incorporate the
hypotheses that sellers  process bids
sequentially, buyers do not have  certain
knowledge of the bids made by other
buyers, and the distribution of bids for a
house  is truncated at the sellers' asking
price. These hypotheses are inconsistent
with the standard  model's assumption
that houses always are sold to the buyers
willing to pay the  most for them. As a
consequence, it is possible to distinguish
empirically between the new model [i.e.,
the model of Equations (2-4)] and the
standard model and to test each against
the other empirically to determine which
better explains the available data.

Data
  The  new and standard  models were
tested  and compared empirically using
data describing a sample of 1196 houses
sold in Baltimore City and County, Mary-
land, during  1978. The data  were as-
sembled from  records of  the Central
Maryland Multiple  Listing Service, Balti-
more City and  County tax records, the
U.S. Census  Bureau, the  Maryland
Bureau of Air Quality, and the results of
previous investigations of school quality

-------
 and crime levels in the Baltimore area.
 The data include the sale price,  asking
 price, and annual property tax liability of
 each sampled  house  as well  as  49
 attributes of the house and its neighbor-
 hood. Environmental quality is character-
 ized in  these data by three air pollution
 variables: annual geometric mean partic-
 ulate concentration, annual average N02
 concentration,  and annual   maximum
 hourly Os concentration.

 Results
   The new and standard models were
 both estimated (i.e., fit to the  data) using
 standard econometric methods.  These
 methods make it  possible  to test the
 models against one another by comparing
 how well each model fits the data. If the fit
 of one model is significantly worse than
 the fit of the other, then it is possible to
 conclude with a high degree of confidence
 that the worse fitting model is incorrect.
   Figures 1  and 2 showthe results of this
 comparison process.  In Figure  1, the
 observed sale prices  of 50 randomly
 selected houses are plotted against the
 predicted sale  prices obtained from the
 new model. In Figure 2, the observed sale
 prices of the same 50 houses are plotted
 against the predicted sale prices obtained
 from the  standard model. The predicted
 sale prices were computed as expected
 values conditional on attribute levels and,
 in the case of the new model,  asking
 prices.  Figure  1  shows that the new
 model fits the observations well: there is
 no indication  of serious errors  in the
 model. Figure 2 shows that the standard
 model fits the observations much more
 poorly than does the new model.
   The differences in fit shown in Figures
 1  and 2 are reflected in the root-mean-
 square (RMS) errors of the two models'
 predictions of'houses' sale prices. For the
 1196 houses in the full data set, the new
 model has  a  RMS prediction error  of
 $2,830; whereas, the standard model has
 a RMS prediction error of $11,360.
  These results suggest strongly that the
 standard model is erroneous. It is possible
 to perform a formal statistical comparison
 of the new  and standard models. This
 comparison leads to the conclusion that,
 if  the  standard model  is  correct, the
 observed goodness-of-fit  results are
 events whose probabilities of occurrence
 are less than 0.0001. Thus, it  is virtually
 certain that the standard model  is er-
 roneous and that the new model provides
 a better explanation of the operation of
the Baltimore housing market than does
the standard model. Since the bid rent
 model is equivalent to the utility  maxi-
       70
   S   6
   0.
   •S
    
-------
  •S
  
-------

-------
    Joel L Horowitz is with the University of Iowa, Iowa City, I A 52242.
    J. David Mobley is the EPA Project Officer (see below).
    The complete report, entitled "Inferring Willingness to Pay for Housing Amenities
      from Residential Property  Values," (Order No. PB 86-103 082/AS; Cost:
      $ 11.95, subject to change) will be available only from:
            National Technical Information Service
            5285 Port Royal Road
            Springfield, VA 22161
            Telephone: 703-487-4650
    The EPA Project Officer can be contacted at:
            Air and Energy Engineering Research Laboratory
            U.S. Environmental Protection Agency
            Research Triangle Park, NC 27.711
United States
Environmental Protection
Agency
Center for Environmental Research
Information
Cincinnati OH 45268
Official Business
Penalty for Private Use $300

EPA/600/S7-85/034
        0000329   PS

        U  S  EKVIR  PROTECTION  AGENCY
        RiGION  5 LIBRARY
        230  S  DEARBORN  STREET
        CHICAGO               IL   60604

-------