OECD-IMF WORKSHOP Real Estate Price Indexes Paris, 6-7 November 2006 - Managing hedonic housing price indexes: the French experience

 
CONTINUE READING
OECD-IMF WORKSHOP

               Real Estate Price Indexes
               Paris, 6-7 November 2006

                           Paper 12

Managing hedonic housing price indexes: the French experience

      Christian Gouriéroux (CREST and University of Toronto)
             and Anne Laferrère (INSEE and CREST)
Managing Hedonic Housing Price Indexes: the French
                  Experience
                      Christian Gouriéroux∗, Anne Laferrère†
                                    September 2006

                                         Abstract

   Despite their theoretical advantages, hedonic housing price indexes are not
so commonly used by statistical agencies or real estate professionals. Many
published indexes still rely on mean or median prices, or favor repeat sales
methods, which require less data and technicality, but are less accurate and ro-
bust. In France, as in other countries where housing sales have to be recorded
in front of a notary, complete data sets on transaction prices and charac-
teristics of dwellings are available. Such data have been centralized since
1994, and a regular computation of quarterly hedonic housing price indexes
has been done since 1998. This paper describes the institutional setting of
housing transactions in France, and the collaboration established between the
notaries and the national statistical agency (INSEE). The notaries are respon-
sible for data collection and regular computation, whereas the national agency
takes scientific liability for the method. The detailed transaction information
remain proprietary data, but desaggregated indexes are publicly and freely
available. This organisation and role assignment have proven their efficiency
and might be extended to countries with similar institutional setting.

Keywords : Housing Price Index, Hedonic Method, Pricing System.

1       Introduction
The theoretical advantage of hedonic methods for computing housing price indexes has
long been acknowledged (see e.g. Case et al., 1991). Indeed, this is the only way to
control for changes in the quality mix of dwellings, whose transaction prices are observed.
    ∗
   CREST (Centre de Recherche en Economie et Statistique ) and University of Toronto.
    †
   INSEE (Institut National de la Statistique et des Etudes Economiques) and CREST (Centre de
Recherche en Economie et Statistique ), email: anne.laferrere@insee.fr.

                                             1
For instance, other indexes based on mean observed trading prices can be biased since
the observed sales are not a representative sample of the set (portfolio, or ‘basket’) of
dwellings that one wants to follow. An index based on the median transaction price is less
sensitive to extreme observed values, but still subject to selectivity bias, as the quality of
the properties evolves over time. The hedonic approach assumes a pricing model where
a dwelling is represented by a limited number of observed characteristics, with their own
prices, whose combination (its quality mix) makes the dwelling value. The pricing model
is estimated from observed prices and characteristics of traded properties. Then the
estimated model is used to follow over time the estimated value of a chosen basket of
dwellings, even if some types of dwellings in the basket have not been traded at each date.
    Such an achievement comes at a cost, since both prices and characteristics of properties
need to be observed and recorded. This rarely happens for actual transactions. Hence
hedonic methods are often applied to valuations by chartered surveyors, or to quoted
asked prices, rather than to observed transaction prices. In countries such as the US where
residential mobility is high1 , some have turned to repeat sales methods. The repeat-sales
index is computed by comparing the fetched prices of the same dwelling at two different
points in time, and assuming that the quality mix stays exactly the same. However,
besides the need of a high turnover, there is no means to be sure that the dwelling is
identical (rehabilitation is not usually recorded, apartments can be divided or reunited2 ),
and the selection bias is still present, as the set of traded dwellings (and those with
multiple sales) can be a non-representative sample of the basket of interest.
     The high cost or even the impossibility of observing transaction prices and charac-
teristics of the traded dwellings explains why a regular computation of hedonic indexes
by statistical agencies or real estate professionals is not common3 . Many official indexes
still rely on mean or median prices, or favor repeat sales methods which are less data
demanding. In countries where the law requires housing sales to be recorded in front
of a notary, data on transaction prices and characteristics of properties can be available.
France is such a country, where the data on sales have been collected and centralized since
1994, and made possible the computation of quarterly hedonic housing price indexes since
1998. This paper describes the institutional setting of housing transactions in France, the
main indexes,and their diffusion policy (Section 2), together with the way the data are
collected, and the collaboration established between the notaries and the national statis-
tical agency (INSEE). The notaries are responsible for data collection and computation,
whereas the national agency takes the scientific liability for the hedonic method. The
database is described in Section 3, and the hedonic specifications are presented in Section
4. A by-product of the hedonic method is a valuation expert system, briefly described
in Section 5. This job organisation and role assignment for the notaries and National
Statistical Institute have proven their efficiency, and might be extended to countries with
similar institutional setting.

   1
     The annual residential mobility rate is about 17 to 18 percent in the US compared to 8 to 9 percent
in France (Long, 1991; Baccaı̈ni, 2001).
   2
     This may explain why the method is employed for single family units, which are more easily identified
by their address than apartments in a building. Some repeat-sales methods are combined with hedonic
models for observed characteristics (see for instance Quigley, 1995, or Englund, Quigley, Redfearn, 1998)
   3
     Vrancken (2004) reports seven hedonic price indexes only, for second-hand housing, in Hong-Kong,
Norway, Sweden, Switzerland and the UK, respectively.

                                                    2
2       The French institutional setting

The French institutional setting is characterized by a network of notaries (notaires, in
French) who have a monopoly in registering real estate transactions, and by a national
statistical agency.
    In France, all real estate transactions have to be registered in front of a notary who has
a monopoly. The role of a notary is to verify the existence of property rights, to draft the
legal sale contract and deed, to send the records to the Mortgage Register (Conservation
des Hypothèques)4 , and to collect the stamp duty for the government5 . A notary is both
a public officer (officier ministériel ), and a private professional6 . Thanks to this feature
of the French legislation7 , a notary has access to the transaction price, together with the
dwelling characteristics that are written on the sale contract. Moreover, each notary has
to send information on the price fetched by the property to the tax authorities, since the
sale tax is function of the price. The corresponding data are appropriate for computing
hedonic housing price indexes. They cover all sales, and thus there is no problem of sample
representativeness (see the discussion below); they provide actual transaction prices and
are not submitted to the uncertainty of a valuation process; the series are available over
a long period with regular availability and continuity thanks to a stable legislation; the
data frequency is adequate, as the notaries have to send the information and pay the tax
to the Finance Ministry within 24 hours of a sale.
    The central statistical agency, that is the National Institute of Statistics and Economic
Studies (INSEE8 ), is in charge of providing official statistics. Among other price data, it
is responsible for the retail price indexes, the industrial price indexes, or the construction
cost index. Up to the end of the 1990s, INSEE published no housing price index. The city
of Paris was an exception, as a ‘Notaires-INSEE’ quarterly index was created in 1983 for
second-hand apartments in Paris. INSEE helped defining segments and provided weights
from the Population Census; then the index was computed by the notaries as a weighted
average of transaction prices9 .
    In 1997, the Conseil Supérieur du Notariat (CSN), that is the National Union of
Notaries, decided to create a price index for dwellings located outside the Paris region,
the so-called Province. They turned to INSEE for advice. INSEE agreed to provide a
methodology, because a public service of reliable housing price indexes was missing in
France. To ensure long-term involvement of both parties, formal agreements were signed
in 1998 and 1999 between the CSN and INSEE, and in 2000 and 2002 between the CINP
and INSEE, for renovated hedonic indexes.

    4
     There are 354 decentralized property registers.
    5
     Or the Value Added Tax in case of a new construction. Notaries also collect the capital gain tax
when applicable.
   6
     As a public officer his/her fees are regulated by law; they include a fixed part and one part roughly
proportional to the sale value.
   7
     Notaries with similar duties exist in Belgium, The Netherlands, Morocco, etc.
   8
     Institut National de la Statistique et des Etudes Economiques.
   9
     More precisely, this index was computed by the Chambre Interdépartementale des Notaires Parisiens
(CINP), that is the Parisian branch of the profession. The 72 segments were defined by crossing the
number of rooms, the date of construction and the level of comfort.

                                                    3
2.1     A quarterly INSEE monitoring
The notaries collect the data and compute the indexes at their own cost. By-products
of the index computation are sold by the notaries to finance the data collection and the
indexes updating. They go from part of the database, statistics on buyers and sellers, to
a complete valuation system of dwellings and an expertise on real estate prices10 .
    INSEE does not compute the indexes but is answerable for the index method. For
this purpose a quarterly quality control of the main indexes has been established. It
relies on information on the data gathering (time of integration in the databases, quality
controls) and on the comparison of the evolution of means prices and indexes for different
regions and categories, in order to detect a possible structural modification. The volumes
of sales, their structure by dwelling type (typically, the number of rooms) are followed
and compared to the reference stock. Zones with extreme variations of price or volume
compared to the preceding quarter or to other zones are also detected and checked for
potential errors.

2.2     The published indexes
Some 23 sub-indexes are currently published at the national level; they are publicly avail-
able and free. Thirteen sub-indexes concern the apartments: Paris, the seven départements
of the Petite and Grande Couronne of Paris11 , the towns of Lyon and Marseille, the urban
units of more than 10,000 inhabitants (city centers, and suburbs), the small urban units
and rural areas. Ten indexes are house sub-indexes, one for the Province, seven for the
départements of Ile-de-France, and two for the Rhône-Alpes and the Provence-Alpes-Côte
d’Azur regions. Various indexes at a larger geographical level are obtained by combining
appropriately the sub-indexes; they concern the Petite Couronne, the Grande Couronne,
the Ile-de-France, Rhône-Alpes and Provence-Alpes-Côte d’Azur regions, the Province,
and France, both for houses and apartments, and for the two types of housing together.
In some urban units or regions with enough sales, local indexes are also computed, but
not yet all published by INSEE (see Fig.1). They will be published in a near future12 . All
indexes can be found in the Bulletin Mensuel de Statistique (BMS), in January, April,
July, and October, which is regular publication of INSEE, now entirely electronic, as well
as on the INSEE website. Each published index is identified in the BMS by a code. They
can also be found at //http/www.indices.insee.fr (Indices et séries statistiques, Con-
struction Logement, Indices trimestriels des prix des logements anciens). In each case, in
the first week of quarter t + 1, two indexes are provided quarterly that are a provisional
index for quarter t − 1 and a revised final index for quarter t − 213 . For instance, on July 5
2006, the revised index for 2005 Q4 and the provisional index for 2006 Q1 are published.
The main indexes are presented in Table 1.
  10
     The question of the cost is not dwelt upon here. Quarterly or annual press conferences held by the
notaries of the Paris area are available at //http/www.paris.notaires.fr. For the rest of France see,
http://www.immoprix.com/.
  11
     First outer ring and more remote suburbs, respectively.
  12
     The next region to have their own index are likely Nord-Pas-de-Calais, Pays de la Loire, and Midi-
Pyrénées.
  13
     Base 100 of the indexes was fixed at the second quarter of 1994 for Paris, at the fourth quarter 1994
for the Province.

                                                    4
The institutional combination of a central statistical agency and a monopolistic net-
work of notaries was at the root of the making of French hedonic housing price indexes.
However, agreement on the need for housing price indexes was only a first step. There is
a long way from the drafting of a sale contract to the publication of the index.

3      Database
The drafting of a housing sale contract by notaries is not enough to make a reliable on
line data base. The contracts are paper documents, sometimes heavy, and they are not
written in a totally standardized way all over the country. To make a proper data base,
the information has to be normalized and coded. The operation is costly, as a deed has
many pages, and there are some 850,000 to 900,000 sales per year14 . Each of the 4,600
notaries is asked to send for key-boarding an extract or a photocopy of the sale deed, plus
some extra notes on the dwelling characteristics15 . This is done on a voluntary basis. In
the near future the sale contracts should be normalized and computerized, the process
will use electronic mail and become much cheaper. This is not yet the case, even if the
first tests for electronic contracts have been conducted in 2005.
     The data on a particular sale are integrated in the database within 2 to 3 months
from the date of the signature. The speed at which each notary sends the data is crucial
for the index quality. The best incentives to induce a notary to send his/her data are
still being experimented. Before turning to email in the future, sending reminders twice a
month by mail, including pre-filled and pre-paid envelopes seem to work best, along with
additional phone calls. In the Province, the average time between a sale and the reception
of the data is now 57 days; then, it takes 40 more days to integrate the transaction in the
database. The delay is less in the Paris region.
     The index is restricted to arm-length transactions of second-hand dwellings16 . To
enter into the index a dwelling has to be free for occupation (not rented at the date of
the sale), only used for habitation (no professional use), and has to be acquired in full
property by a private individual or by a SCI (Société civile immobilière17 ). Exceptional
homes such as single rooms (service rooms), attics, artist studios, or castles are excluded.
Those restrictions eliminate about 15 percent of transactions.
     There are two databases. The base BIEN, managed by the CINP, covers the Ile-de-
France, that is Paris and the Paris region18 . The base Perval, managed by the Perval
society for the CSN, covers the rest of France. There are 86 departements outside Ile-
de-France19 . Together they included some 9,7 million transactions at the end of 2006, 30
percent in Ile-de-France, and 70 percent in the Province. This includes all real estate sales,
  14
     Among which some 90 percent are second hand dwellings which enter the scope of the index.
  15
     The systematic data collection necessary to make an hedonic index was decided at the end of the 1970
in Paris and in the 1990s for the rest of France. In 2004, about 20 people worked on the data collection,
for the Province, and 15 for the Paris region.
  16
     Second-hand dwellings are distinguished from new dwellings from the way they pay taxes. New
dwellings are submitted to value-added tax (VAT), which is lower than stamp duty. The first sale of a
new building taking place 5 years after construction is no longer under the VAT regime, and enters into
the index.
  17
     A family civil company for real estate investment.
  18
     The departements of Seine-Saint-Denis (93), Val-de-Marne (94), Val d’Oise (95), Essonne(91), Haut-
de-Seine (92), Yvelines (78) and Seine-et-Marne(77).
  19
     Corsica and the French overseas territories are left out for the time being.

                                                   5
including for instance parking lots, new buildings, or land. As only second-hand houses
and apartments are included in the hedonic index computation, 5,4 millions observations
are used for the housing price indexes. Roughly half of them are apartments, half of them
are houses. In 2006, some 780,000 new observations were added in the databases, among
which 520,000 were second-hand sales of houses or apartments.
    The distinction between the Paris region and the Province is due to the history of
centralization in France. As the Paris region, and the city of Paris itself, concentrate a
large part of the wealth, the oldest historical database is the one collected by the CINP,
as early as 1979 for Paris and Petite Couronne, since 1995 for the Grande Couronne. The
database for the Province was created in 1990 and became operative in 1993. The making
of the indexes brought the Parisian and Province databases closer. For instance a sale of
a Parisian dwelling made by a notary of Province is now included in the Parisian database
and vice-versa.

3.1     Coverage rate
Since the data collection is made on a voluntary basis, the rate of coverage of the notary
database is not 100 percent. For instance, 71 percent of the notaries of Province sent some
data in 2006, whereas the rate is around 85 percent percent in Ile-de-France. The rate of
coverage of the total housing transactions by the notary database is not perfectly known,
because there are no other official statistics on housing transactions. The notaries from
Ile-de-France collect statistics on their activity by a special survey. The survey does not
separate housing from other real estate transactions. The overall coverage rate, computed
by dividing the number of transactions in the database by the total number of transactions
as measured by the survey on activity, was 85 percent in 2004 (89 percent for Paris, 88
percent for the first outer ring of Paris, and 78 percent for the more remote suburbs). An
indirect way to estimate the coverage rate is to use the amount of stamp duties collected in
each of the French départements, as known from the Tax authorities (Direction Générale
des Impôts). Dividing the total tax by the tax rate (4.8 percent) provides the total sale
value, for each department. By comparing it with the total sale value of the notary
database at the same geographic level, one gets a coverage rate, in value (not in number
of transactions)20 . The estimated average coverage rate in 2003 was 66 percent, that is,
83 percent in Ile-de-France and 64 percent for the rest of France. It varies from one place
to the other. It was lower than 30 percent in 12 départements, between 30 percent and 50
percent in 23, between 50 percent and 70 percent in 36 départements and over 70 percent
in 23 départements.
    Actually a 100 percent coverage rate is not necessary to compute a hedonic index. As
seen below the method is based on the valuation of a fixed basket of properties, defined
over 4 years of transactions. The structure of the basket is close to that of all dwellings,
as known from the Population census. At least for observed characteristics, there is no
  20
     The method is not perfect as the tax rate is now the same for housing and other real estate trans-
actions, which are no more separated in the tax statistics. If the coverage rate was the same for housing
and other real estate, this feature would not be a problem. However, before 1999 when the tax rates
were different for housing and other real estate, the coverage rate was higher for housing. Assuming that
the differential between the coverage rate for housing and for other real estate transactions is constant
over time at the département level, and that the share of housing among all transactions, as known from
year 1999 (when the distinction was possible), is also constant, a coverage rate can be computed in each
département (Friggit, 2003).

                                                   6
obvious bias in the properties that the notaries choose to send to the data base. Once
the reference basket is fixed, the hedonic method is immune to selection bias, that is from
the fact that the sales on given period are a non random sample of the stock of dwellings,
and that registration in the database is also potentially non random (see below). It is
however important to check that the coverage rate does not fall below a minimal level to
insure that the transactions are sufficient for an accurate estimation at a particular local
level. We turn back to this feature of the index below21 .

3.2     Characteristics of dwellings, and treatment of non-responses
The database is anonymous to comply with the French law. The precise address of the
dwelling is included but is not made public, and is not used in the index computation. The
only location characteristic is a municipality code (code commune), close to a ZIP code,
corresponding to a town or village (there are more than 36,000 communes in France),
with an added neighborhood code (code quartier) when the commune is large enough22 .
A neighborhood can for instance be one to four zones within an arrondissement in Paris.
Dwellings are separated between houses and apartments. Note that the French housing
park is divided nearly equally between houses and apartments. The way houses are
constructed differs widely; brick dominates in the North and East, stone and concrete in
the rest of France; constructions in wood are rare. A majority of houses are detached and
located outside city centers in suburbs, or in villages, except in some regions were town
houses can be found23 . The quality of apartments is linked to their date of construction. In
all cases, the location tells much on the dwelling appearance and quality, not only in terms
of neighborhood characteristics, but also in terms of building characteristics. For instance
19th century Hausmannian construction in Paris is of better quality than constructions
of the same period in other areas. This is why hedonic regressions are estimated at a
detailed local level, and why the models may include neighborhood dummy variables and
cross effects (see below).
    Besides the zone and date of the sale, the observed dwelling characteristics are the
following: surface (in square meters), time of construction (8 categories: < 1850, 1850-
1913, 1914-1947, 1948-1969, 1970-1980, 1981-1991, 1992-2000, > 2000), number of rooms
(from 1, to 5 and more), number of bathrooms (0, 1, or 2 and more), number of garages
or car parks (0, 1, or 2 and more) and for apartments, floor level ( 1st, 2nd, 3nd, 4th,
etc...), presence of a lift, existence of a service room (0, 1 , 2 or more). For houses, the
number of levels (1, 2, 3 or more), the presence of a basement and the surface of the plot
are also known. The rate of non-response varies among the explanatory variables (Table
2). In case of non-response, either the sale does not enter into the index computation (for
instance when the surface is unknown), or the characteristic is imputed from econometric
models estimated on complete data (Table 3).
  21
      The coverage rate is also important to consider when the database is used to follow the activity of the
housing market. Once the total amount of sale is known in each department (from tax data), dividing
it by the average transaction price (from the notary database) provides an estimation of the number of
sales.
   22
      Which side of the street, even or odd number, and even geocoding is also registered but not used in
the hedonic computation.
   23
      According to the French Housing survey of 2001, 58.4 percent of houses are detached, 24.3 percent
are semi-detached, and 17.3 percent are grouped.

                                                     7
4      The Hedonic Method
The basic assumption common to all hedonic price indexes is that each dwelling is defined
by the combination of a fixed number of characteristics, its quality mix, that enters
the consumer’s utility. Among all hedonic housing price indexes that we are aware of
the French index has the unique features of combining a large number of geographic
zones/strata and the quarterly estimation of so-called ‘reference stocks’ of dwellings in
each zone. This section describes those features in some details.
    Defining Zones/strata
    Dwellings (houses and apartments are separated all along) are assumed to be stratified
into zones where prices are homogeneous and price evolutions are roughly parallel. It is
important to estimate the hedonic models on homogeneous price zones, that are zones
where prices are not too different, and move in the same way over time. Since the strata
used for the publication of an index are not necessarily homogeneous, it is necessary to
cut or group then. The homogenous segments have been defined locally by interviewing
real estate experts. Then a tree analysis has been applied to aggregate similar segments.
Ideally a model will be estimated per segment and the elementary geographic zones can
represent rather small sub-markets. Practically we have been limited to a little less than
300 zones to ensure a sufficient number of sales per zone (over 400 per year). Typically,
for large cities, above 10,000 inhabitants, a zone is a city center or a city suburb; a zone
is a group of rural areas or smaller towns in less densely populated regions; it is close to
an arrondissement in Paris.
    In a given zone, the price index is defined as the ratio of the estimated value of a
reference stock of dwellings, a basket of houses, to its value at the base period of the
index. For each quarter, the value of each dwelling in the reference basket is estimated
from the prices of all observed sales by means of the hedonic econometric models that
have been estimated on the sales of the ‘estimation period’.
    Reference stock
    The principle of the hedonic method is to correct for the variations of the structure
of the sales at a particular date of observation. It is achieved by estimating the value of
a fixed stock of dwellings at each date. The index follows the price of the dwellings in
this reference stock. The reference stock is made of all sales during the period 1998-2001
in each of the 296 elementary zones/strata. It excludes sales in the extreme quantiles of
the distribution of prices per square meter. The size of the reference stock in each zone is
on average 2,800 dwellings, which represent about 1,220,000 dwellings for the whole stock
(Table 4, col.4). This feature of the hedonic method makes it immune to selection bias.24
    Hedonic pricing models
    Hedonic pricing models relate the prices (more precisely, the logarithm of the price
per square meter) to the characteristics of the dwellings. The characteristics include the
location (a neighborhood within a zone), and the quality of the dwelling itself. Each model
is estimated on a stock of transactions called estimation stock. It includes all dwelling
sales during the 1998-2001 period, except the transactions for which the number of rooms
is not known, or the estimated price was found ex-post to differ from the observed price
by more than two standard-errors. It is close, but not equal, to the reference stock defined
above (see Table 4). The econometric estimations are made separately in each elementary
  24
    Contrary to a method based on including time dummies in hedonic regressions estimated on all
recorded sales at each date.

                                               8
geographical zone.
       The model is the following :
                                         3
                                         X                 4
                                                           X                 K
                                                                             X
                   Log pi = Log p0 +           αa Ya,i +         θt Tt,i +         βk Xk,i + i       (1)
                                         a=1               t=1               k=1

where pi denotes the price per m2 of dwelling i, Ya,i is a dummy variable for the year of
sale of dwelling i, Tt,i a dummy for the quarter of sale of dwelling i and Xk,i , k = 1, . . . , K,
are continuous or dummy variables computed from the dwelling characteristics. They can
include nonlinear or interaction effects. For instance the presence of an elevator is crossed
with floor level. The coefficients of the model characterize the prices of the characteristics
levels, which together define a reference dwelling, the price of which is p0 25 .
    The variables X include the number of rooms, the floor, the number of levels, the av-
erage size of rooms, the presence of a service room, a parking, a terrace, a balcony, a base-
ment, or a garden, the number of bathrooms, the period of construction, the condition26 .
Some estimated models include also a neighborhood dummy, and, in some specifications,
the number of rooms is crossed with the neighborhood dummies. The lot size is included
for houses. Remember that each model is estimated in a particular zone/strata, and thus
all variables are de facto interacted with the zone. The choice of the explanatory variables
including interactions has been done by an automatic classification and robustified, by a
reduced rank analysis [see Gouriéroux, Jasiak (2006), chapter on multiple scores].
    Two examples of hedonic models are reported in this paper. A first one for houses
in the outskirts of Paris (Seine et Marne), a second one for houses in Dijon, a city of
Burgundy (Tables 5 and 6). The dependent variable is the logarithm of the price per
square meter (for apartment) or the total price (for houses) in Euros. The goodness of fit
quality of the hedonic regressions as measured by the determination coefficient R2 , varies
between 0.18 and 0.70 for apartments, and between 0.50 and 0.80 for houses. The number
of observations ranges from 1,721 to 19,342. For individual cross-section data, values of
R2 in the range of 0.25-0.40 for 1000 to 3000 observations and around 20 variables are
considered good. This is what is obtained in most zones.

       Current value of the reference dwelling

    The same type of model is used at the current period τ , with the same reference
dwelling of price p0,τ . The price per square meter of dwelling j sold in period τ is written
as27 .
                                                           K
                                                           X
                          Log (pj,τ ) = Log (p0,τ ) +            βk,τ Xk,j,τ + j,τ .
                                                           k=1
  25
     The reference dwelling is one of a precise quarter and year of sale. The value of a dwelling with the
same characteristics, but sold at a different time is computed from p0 by multiplying by the corresponding
quarter and year parameters exp θt and exp αt .
  26
     In Ile-de-France the variables ‘fair condition’ and ‘terrace or balcony’ are not known.
  27
     The evolution of the price of the reference dwelling is the core of the index construction. For this
reason it must include seasonal and cycle effects. This is why the quarter and year parameters are not
in the current period model, while they were introduced in the first model because the estimation was
made over more than one quarter. The price for a dwelling of quarter (a, t) would be: Log (p0,a,t ) =
Log p0 + αa + θt .

                                                    9
The period τ is chosen according to the type of index. More precisely, the index for a
quarter t is computed over all arm-length transactions of a period τ ending with quarter
t28 .
      Let us now explain how the price of the reference dwelling is computed from data on
current sales. Let us assume that the βk,τ coefficients are known, and denote pej,τ the price
that would fetch dwelling j with the characteristics of the reference dwelling, then:
                                                              K
                                                              X
                                    pj,τ ) = Log (pj,τ ) −
                               Log (e                               βk,τ Xk,j,τ .
                                                              k=1
   p̃j,τ defines the ‘reference dwelling equivalent price’ of dwelling j, τ . The model can be
rewritten as :

                                       Log (e
                                            pj,τ ) = Log (p0,τ ) + j,τ .
   Hence, if the βk,τ coefficients were known, the logarithm of the price of the reference
dwelling Log (p0,τ ) would be estimated as the mean of all estimated prices:
                                                        Jτ
                                                    1 X
                                       Log(b
                                           p0,τ ) =        Log(e
                                                               pj,τ ),
                                                    Jτ j=1
where Jτ is the number of transactions at period τ .
   In practice, the hedonic models are found to be very stable over time, and it is assumed
that the relationship between the characteristics and the price of a house is fixed, in a given
zone, for a period of up to around five years29 . This allows to replace the βk,τ coefficients
by the βbk estimated over the reference period. It simplifies the quarterly computation, of
hedonic prices as they involve no further econometric estimation:
                                              K
                                              X                                pj,τ
                   pj,τ ) ' Log (pj,τ ) −
              Log (e                                β̂k Xk,j,τ = Log [       PK                ].
                                              k=1                        exp( k=1 β̂k Xk,j,τ )
    Then, the log of the price per square meter of the reference dwelling in period τ ,
is estimated by a geometric mean of the ‘reference dwelling equivalent prices’ of the Jτ
dwellings sold in period τ :
                                             Jτ                  Jτ
                                         1 X               1    Y
                           Log pb0,τ   =        Log pej,τ = Log( pej,τ ),
                                         Jτ j=1            Jτ   j=1

  28
     Up to the end of 2003, the Parisian index was computed on a six-month basis, hence τ = [t − 1; t];
indexes for the Province were annual, τ = [t − 3; t]. From 2004 on they are all pure quarterly indexes,
τ = t, which makes them more reactive and allows to study seasonal price variations. However, quarterly
indexes at a more local level remain semestral or annual to ensure a sufficient number of transactions in
the zone. Monthly indexes are currently tested for Paris.
  29
                                                                          P3               P4
     The models assume that the time effect is captured by the term a=1 αa Ya,i + t=1 θt Tt,i and that
the coefficients βbk are time invariant during the years following the estimation period. The time invariance
assumption was checked. It was verified that the difference between the estimated value of dwellings with
characteristic Xk and their actual sale price, that is the residual ui , satisfies the stochastic assumption of
the model, and does not include an unobserved deterministic component. The time evolution of the mean
of the residuals in some zones was computed for each of the coefficients βbk , k = 2 . . . , K. They were
found stable over time. After a maximum of 5 years, they are checked and changed updated if necessary.
This has been done in 2002-2003, with no major effect on the index profile.

                                                       10
or:

                                                   Jτ
                                                              ! J1
                                                   Y               τ

                                         pb0,τ =      pej,τ              .
                                                   j=1

   Current value of the reference stock

    Once the value of the reference dwelling has been estimated, the estimated value of
any dwelling of the reference stock can be computed, and, by aggregation, the value of
the stock itself. The computations are made per zone. For this reason, let us re-introduce
the index s of the zone. The value of dwelling i of the reference stock of zone s in the
current period τ is estimated from its characteristics Xk,i,s , which are time invariant, by
definition of the reference stock. The approached value is:
                                                            K
                                                            X
                        pb∗ i,s,τ = exp(Log pb0,s,τ +              β̂k,s Xk,i,s )Ai,s ,
                                                            k=1

where Ai,s is the surface of dwelling i, s.
   The estimated current value of the Ns dwellings of the reference stock of zone s is
obtained by summation:
                                                   Ns
                                                   X
                                          W
                                          cs,τ =          pb∗ i,s,τ .
                                                    i=1

   In the same way, the value of the reference stock is estimated at the base period of
the index, denoted t = 0,. We get:
                                  Ns
                                  X                           K
                                                              X
                       W
                       cs,0 =           exp(Log pb0,s,0 +               βbk,s Xk,i,s )Ai,s .
                                  i=1                          k=1

   Quarterly computation of the index

   The elementary index for zone s measures the evolution of the value of the reference
stock of that zone s. It is given by:
                                 PNs
                                               b0,s,τ + K
                                                        P b
                          W       i=1 exp(Log p          k=1 βk,s Xk,i,s )Ai,s
                          cs,τ
               It/0 (s) =      = PNs                    PK b                   .
                          W
                          cs,0        exp(Log pb0,s,0 +
                                         i=1                 βk,s Xk,i,s )Ai,s
                                                                             k=1
   The index of zone s can also be written as:

                            It/0 (s) = exp(Log pb0,s,τ − Log pb0,s ),
and involves only the evolution of the price of the reference dwelling. The computation of
the index at date t does not require the computation of the implicit value of each dwelling
of the reference stock; the coefficients Log pb0,s,τ are obtained by:
                                        Jτ                  K
                                    1 X                    X
                      Log pb0,τ   =        Log (pj,s,τ ) −     β̂k,s .X k,s,τ ,
                                    Jτ j=1                 k=1

                                                   11
where X k,s,τ is the mean of the Xk,j,τ variables for the Jτ transactions of the current
period in zone s.
    Aggregate indexes
    Most elementary indexes per zone/stratum are not published. They are aggregated
at larger geographical levels. For instance, the index for the ‘Province’ measures the
evolution of the value of the whole reference stock of Province. This index can be written
as
                                                   P c
                                              W
                                              cτ      Ws,τ
                                     It/0   =    = Ps      ,
                                              W
                                              c0
                                                    s W
                                                      cs,0
where the summation is made on the zones of the Province. It can be interpreted as a
mean of the elementary indexes per zone, weighted by the total sales value in the zone in
the reference stock:
                                                 !
                                   X      W
                                          cs,0
                            It/0 =       P c       It/0 (s).
                                            Ws,0
                                            s     s

    Practically, the weights of some indexes are corrected by a parameter δs for zones
where the notary database is deemed to be non exhaustive30 .
    The main Notaires-INSEE indexes are presented in Fig.2 (for apartments) and Fig.3
(for houses), together with their rate of increase. There are strong seasonal effects, espe-
cially for houses, which may be linked to residential mobility of families and the school
year calendar. INSEE also publishes seasonally adjusted housing and apartment price
indexes.
   To summarize, the process involves several steps. The first four steps are done once
and for all, and only updated every five years:

step 1 Define zones (strata), where the price evolution is assumed to be homogeneous;

step 2 Define a hedonic pricing model, that is introduce correction coefficients for quality
     effects, for each zone;

step 3 Estimate the correction coefficients from an estimation stock of dwellings in each
     zone;

step 4 Compute the value of a reference stock at the base date for each zone;
       The three following steps are repeated every quarter.

step 5 Compute the value of the reference stock, from data on all current period sales
     per zone;

step 6 Compute the price index as the evolution of the value of the reference stock
     between base and current date;

step 7 Publish indexes and sub-indexes by aggregation of local zone indexes.
  30
    Corrected weights are fixed and estimated from stamp duty returns and correspond to the value of
the reference stock in each zones.

                                                 12
Note that the quarterly computation of the index involves no econometrics. This
feature makes it most attractive as the word hedonic is sometimes perceived by statistical
agencies as synonymous of sophisticated and time consuming.

4.1    The model updating
As the index is based on the valuation of a fixed basket of dwellings (the reference stock),
the question arises of the updating of the basket. This is done every four or five years.
We have to check for the stability of the models, their specification, and the local baskets
themselves since the dwelling park is constantly evolving over time with new construction,
destruction and rehabilitation. The zones/strata themselves may have to be redefined, or
at least checked, as population is moving and being redistributed over the territory.
    The first revision took place in 2003. Some zones were redefined; the period of refer-
ence, hence the basket, was updated (going from 1992-1996 or 1994-1996 to 1998-2001);
the specifications were only marginally changed. This updating has no major effect on
the indexes.

5     Expert System
The construction of an hedonic price index is based on an econometric pricing model,
which explains how the price of an appartment or house depends on its characteristics.
The estimated model can be used to predict the price of any mix of characteristics, as
done in the construction of the hedonic price index itself. By using the information on the
estimated variance of the error term and the estimated variance of the beta coefficients,
the model can also be used to get a 95 % prediction interval for any mix characteristics,
that are a minimal and a maximal.
    This approach has been followed for building a pricing expert system, which is available
on line, and is one of the source for financing the construction of the indexes.
    Such an expert system can be used for different purposes as a source of information
on market prices before a transaction, or as a source for checking ex-post it a transaction
price is compatible with the market, for instance to detect a possible fraud to tax payment.

    Such an expert system is also required for the implementation of the new regulations
in Finance (Basel II), or Insurance (Solvency II). Indeed, the banks, credit institutions
and insurance compagnies have a significant part of their portfolio directly invested in real
estates, or indirectly since real estate is the standard collateral for mortgages or firm loans.
In the current regulations the value of this portfolio has to be computed and updated very
frequently. A pricing expert system is the natural tool for computing the values of real
estate portfolios.

6     Conclusion
Thanks to the conjunction of sales data, good will and accurate methodology, reliable
housing price indexes now exist for France. These three elements are necessary and it is
important that they persist in the long run. The data should go on being collected, that
is the notaries have to settle on a durable way of funding them. The tax authorities are

                                              13
unifying and computerizing the real estate sale documents. A side effect will likely be a
reduced cost for data gathering and a better quality of data. But the information needed
for the hedonic models, and not requested by tax authorities, has to be provided for the
index and the hundreds of small notary practices have to be motivated. This leads to the
second element, good will. It is fuelled by information about the use of the indexes. To
the notaries, they should become a trademark, and the valuation system linked with the
indexes should prove useful and a mean to make the enterprise profitable. On the INSEE
and academic side, and for the general public, the mere existence of reliable indexes and
of all the related informations, has begun to fuel new types of studies. As prices can be
compared both in space and over time, they can be introduced in microeconomic models
of agents decisions, and provide more reliable guidelines to public and individual choices.
As housing and more generally real estate prices and consumer prices evolution can differ
widely, it is of primary importance for economic policy to make use of both. As for
methodology, its unique feature is the use of the valuation of reference parks at a detailed
geographic level.
    Finally, the assumption of time stability of the model implies that there is no further
econometric estimation in the given period of computation of the index, which saves time
and cost. This feature of the hedonic method makes it attractive for government agencies.
The first updating of the period base, the reference stock and the model specification just
took place. It was rather easy to perform and without major effect on the index profile,
which comforts the long term relevance of the hedonic methodology.

                                            14
Reference

    Baccaı̈ni, B. 2001, Les migrations internes en France de 1990 à 1999: l’appel de l’Ouest,
Économie et Statistique, 344, 39-79.
   Beauvois, M, David, A., Dubujet, F., Friggit, J., Gouriéroux, C. , Laferrère, A., Mas-
sonnet, S. and E. Vrancken, 2006, Les indices de prix des logements anciens, version 2
des modèles hédoniques, INSEE Méthode, 111, 151 p..
    Case, B., Pollakowski, H. and S. Wachter, 1991, On Choosing Among House Price
Index Methodologies, Journal of the American Real Estate and Urban Economic Associ-
ation, 19(3), 286-307.
    David, A., Dubujet, F., Gouriéroux, C. and A. Laferrère, 2002, Les indices de prix des
logements anciens, INSEE Méthode, 98, 119 p..
    Englund, P., Quigley, J., and C. Redfearn, 1998, Improved Price Indexes for Real
Estate : Measuring the Course of Swedish Housing Prices, Journal of Urban Economics,
44, 171-196.

   Gouriéroux, C., and J. Jasiak, 2006, Econometrics of Individual Risks : Credit, Insur-
ance and Marketing, Princeton University Press.

   Friggit, J. 2003, Taux de couverture des bases notariales, Note from the Conseil
Général des Ponts et Chaussées, January 7.

    Long, L.H. 1991, Residential Mobility Differences Among Developped Countries, In-
ternational Regional Science Review, 14, 133-147.
   Quigley, J. 1995, A Simple Hybrid Model for Estimating Real Estate Price Indexes,
Journal of Housing Economics, 4, 1-12.
   Vrancken, E., 2004, Foreign house price indices, CINP (Chambre Interdépartementale
des Notaires de Paris), working paper, Paris.

                                             15
Table 1. The official indexes

  Web site      Paper publication Type of index
   code               code

 081767865           00000 00       France
 086909774           00000 10       France, apartments
 086909673           00000 20       France, houses
 086937763           14000 20       Ile-de-France
 086910582           14000 10       Ile-de-France, apartments
 086911592           13000 20       Ile-de-France, houses
 086937662           15000 10       Ile-de-France (Paris excluded), apartments
 067517858           11000 10       Paris, apartments
 086909875           15010 10       Seine et Marne, apartments
 086909976           15020 10       Yvelines, apartments
 086910077           15030 10       Essonne, apartments
 086910178           15040 10       Haut-de-Seine, apartments
 086910279           15050 10       Seine Saint Denis, apartments
 086910380           15060 10       Val de Marne, apartments
 086910481           15070 10       Val d'Oise, apartments
 080557385           12000 10       Petite Couronne, apartments
 085102847           13000 10       Grande Couronne, apartments
 086910683           15010 20       Seine et Marne, houses
 086910784           15020 20       Yvelines, houses
 086910885           15030 20       Essonne, houses
 086910986           15040 20       Haut-de-Seine, houses
 086911087           15050 20       Seine Saint Denis, houses
 086911188           15060 20       Val de Marne, houses
 086911289           15070 20       Val d'Oise, houses
 086911390           15080 20       Petite Couronne, houses
 086911491           12000 20       Grande Couronne, houses
 080557486           20000 00       Province
 080557587           20000 10       Province, apartments
 067517959           20000 20       Province, houses
 067518060           21000 10       Urban units > 10,000 inhabitants, apartments
 067518161           21100 10       Urban units > 10,000 inhabitants, city center, apartments
 067518262           21200 10       Urban units > 10,000 inhabitants, suburbs, apartments
 080557688           22000 10       Urban units < 10,000 inhabitants and rural areas, apartments
 087986777           30000 00       Provence-Alpes-Côte d'Azur
 087986878           30000 10       Provence-Alpes-Côte d'Azur, apartments
 087986979           30000 20       Provence-Alpes-Côte d'Azur, houses
 087987080           31000 10       Marseilles, urban unit, apartments
 087987181           40000 00       Rhône-Alpes
 087987282           40000 10       Rhône-Alpes, apartments
 087987383           40000 20       Rhône-Alpes, houses
 087987484           41000 10       Lyons, urban unit, apartments

NB : Petite Couronne: the first circle of outskirts of Paris (Haut-de-Seine, Seine-Saint-Denis, Val-de-
Marne). Grande Couronne: the rest of Ile-de-France, further from Paris (Essonne, Seine-et-Marne,
Yvelines, Val d'Oise). Province: all other départements of metropolitan France, except Corsica.

                                                                                                          16
Table 2. Rate of non-response (percent)

      Zone            Surface Number Time of Number of Number of Level or                Lift
                              of rooms construction garages bathrooms number of
                                                   parking lots         levels
Province
   House               40.2     6.3         27.4       40.6       11.5       8.6          -
   Apartment            9.0     1.8         25.1       56.2       7.0        4.6         58.9
Ile de France
   House               46.3     0.1         51.3        -          -         0.2          -
    Petite Couronne        66         0.2       74.0          -          -         0.2          -
 Grande Couronne         38.5         0.1       42.4          -          -         0.2        -
  Apartment            12.4     1.2         16.7        -          -          -          61.1
              Paris       3.4         2.1       10.6    -          -          -           66.4
   Petite couronne       22.7         0.7       21.8    -          -          -           64.1
 Grande couronne         10.7         0.6      18.8     -          -          -           47.8

NB : See note of table 1. The rates are computed on the reference stock.
In Ile-de-France, for garages, bathrooms non-responses are mixed up with `no bathroom' or `no
garage'.

                                                                                                    17
Table 3. Treatment of non-responses

   Type of non-response             Action                Method, if imputation
           Price                   rejected
Surface and number of rooms        rejected
          Surface                  imputed                     econometric

      Number of rooms              rejected
                                  (Province)

                                imputed (Ile-de-    imputed from the surface; rejected
                                   France)         from the estimation park, included in
                                                              reference park
 Type (house or apartment)         rejected

            Lift                   imputed                       Yes
           Level                   imputed                   Ground Floor
        Bathroom                   imputed                   No bathroom
    Garage, parking lot            imputed               No garage, no parking
    Time of construction                              'Non-response' is a category
       Type of buyer               imputed               Private individual or SCI
        Occupation                 imputed                      Not rented
        Destination                imputed               Habitation, full property
 Surface of plot (for houses)      rejected

                                                                                           18
Table 4. Number of strata, neighborhoods, and size of reference and estimation parks

Index                    Number of   Number of   Size of reference park Size of estimation
                           strata  neighborhoods                               park
   Ile-de-France (total)     62         230             382 111              342 947
        Apartments           55         205
                                                        262 102              277 120
          Houses              7          25             120 009              65 827
      Province (total)      234        1 125            837 552              848 286
        Apartments           88         509             431 326              431 713
UU > 10 000 inhab.           74         410             356 133              203 276
city center                  57         297             276 395              87 283
suburbs                      17         113              79 738              115 993
UU< 10 000 inhab, rural      14          99              75 193              228 437
        Houses               146          616               406226                416 573
         Total               296          1355             1 219 663             1 191 233
NB: For houses in Ile-de-France, properties for which size is imputed are excluded from the estimation
park; they are not excluded in the Province.

                                                                                                   19
Table 5. Example of hedonic model: Houses in Seine-et-Marne

Variables                          Coefficient   Standard-error   P-value
constant                            10.963           0.009         0.000
Surface in square meters             0.005           0.000         0.000
Plot in hectares                     0.163           0.000         0.000
Year 1998                                          Reference
Year 1999                             0.056          0.004        0.000
Year 2000                             0.123          0.005        0.000
Year 2001                             0.191          0.005        0.000
Quarter 1                            -0.049          0.005        0.000
Quarter 2                            -0.020          0.004        0.000
Quarter 3                             0.012          0.004        0.006
Quarter 4                                          Reference
Neighborhood 1: Meaux                              Reference
Neighborhood 2: Melun                 0.087          0.005        0.000
Neighborhood 3: Provins              -0.189          0.006        0.000
Neighborhood 4: Fontainebleau         0.053          0.005        0.000
Neighborhood 5; Torcy                0.211           0.005        0.000
Built before 1913                    -0.074          0.006        0.000
1914 – 1947                           0.020          0.006        0.002
1948 – 1969                           0.047          0.006        0.000
1970 – 1980                           0.034          0.005        0.000
Built after 1980                     0.028           0.005        0.000
Date unknown                                       Reference
0 bathroom                           -0.305          0.008        0.000
1 bathroom                                         Reference
2 bathrooms                          0.077           0.004        0.000
3 bathrooms or more                  0.140           0.012        0.000
0 garage                             -0.070          0.004        0.000
1 garage                                           Reference
2 garages or more                    0.063           0.006        0.000
1 level                              0.032           0.004        0.000
2 levels                                           Reference
3 levels or more                     -0.016          0.006        0.008
3 rooms or less                      -0.107          0.005        0.000
4 rooms                                            Reference
5 rooms                              0.026           0.004        0.000
6 rooms                              0.058           0.006        0.000
7 rooms or more                      0.054           0.007        0.000
Number of observations              18,697
R square                             0.696

                                                                            20
Table 6. Example of hedonic model: Appartments in the city of Dijon

Variables                                     Coefficient Standard-error P-value
Constant                                        8.792          0.034      0.000
Year 1998                                                  reference
Year 1999                                       0.005          0.010      0.640
Year 2000                                       0.018          0.010      0.070
Year 2001                                       0.066          0.010      0.000
Quarter 1                                      -0.013          0.010      0.218
Quarter 2                                      -0.001          0.009      0.897
Quarter 3                                       0.019          0.009      0.045
Quarter 4                                                  reference
Neighborhood
1: Historical center                                      reference
2: Victor Hugo-Montchapet                       -0.015        0.029      0.603
3: Clémenceau-30 Octobre-Voltaire               -0.099        0.032      0.002
4: Wilson-Auxonne-Parc                          -0.023        0.034      0.502
5: Facultés                                     -0.108        0.034      0.001
6: Arsenal-Castel-Moulins                       -0.115        0.030      0.000
7: Hôpital général-Bourroche-Montagne           -0.077        0.030      0.010
8: Fontaine’Ouche-Gare Berbisey                 -0.447        0.035      0.000
Built before 1850                                0.154        0.026      0.000
1850-1913                                       0.074         0.020      0.000
1914 – 1947                                      0.074        0.018      0.000
1948 – 1969                                               reference
1970 – 1980                                     0.044         0.010      0.000
1981-1991                                       0.154         0.011      0.000
1992-2010                                       0.242         0.017      0.000
Date unknown                                     0.066        0.013      0.000
0 bathroom                                      -0.069        0.043      0.114
1 bathroom                                                reference
2 bathrooms or more                             0.017         0.016      0.307
0 garage                                                  reference
1 garage                                        0.070         0.010      0.000
2 garages or more                               0.159         0.016      0.000
Ground floor                                              reference
1st floor                                       0.036         0.010      0.001
2d floor                                        0.061         0.011      0.000
3d floor                                        0.040         0.011      0.000
4th floor and more no lift                      0.037         0.011      0.001
4th floor and more with lift                    -0.054        0.023      0.020
1 room                                          0.141         0.061      0.020
2 rooms                                         0.017         0.041      0.677
3 rooms                                                   reference
4 rooms                                         0.040         0.037      0.280
5 rooms or more                                 0.073         0.038      0.053
Surface/room of 1-room apartments 30 m²        -0.071        0.023      0.002
Surface/room of 2-room 24 m²                   0.001         0.016      0.954
Surface/room of 3-room
Surface/room of 3-room 18-22 m²               reference
Surface/room of 3-room >22 m²        0.029        0.014   0.035
Surface/room of >= 4-room = 4-room 17-21 m²            reference
Surface/room of >= 4-room >21 m²     0.043        0.013   0.001
Fair                                          reference
Some rehabilitation                  -0.080       0.011   0.000
To be renovated                      -0.182       0.023   0.000
unknown quality                      -0.014       0.009   0.129
Presence of a cellar unknown         0.139        0.069   0.043
no cellar                                     reference
1 cellar or more                     -0.005       0.011   0.677
No terrace, no balcony                        reference
terrace or balcony                    0.042       0.008   0.000
1-room in neighborhood 2             -0.095       0.067   0.152
2 rooms in neighborhood 2             0.033       0.045   0.468
4 rooms in neighborhood 2            -0.036       0.040   0.368
5 rooms or more in neighborhood 2    -0.004       0.042   0.918
1-room in neighborhood 3             -0.115       0.071   0.106
2 rooms in neighborhood 3             0.052       0.048   0.277
4 rooms in neighborhood 3            -0.003       0.045   0.941
5 rooms or more in neighborhood 3    -0.111       0.049   0.024
1-room in neighborhood 4             -0.024       0.073   0.741
2 rooms in neighborhood 4            -0.036       0.051   0.484
4 rooms in neighborhood 4             0.048       0.049   0.330
5 rooms or more in neighborhood 4    0.028        0.052   0.581
1-room in neighborhood 5              0.039       0.067   0.561
2 rooms in neighborhood 5             0.075       0.050   0.131
4 rooms in neighborhood 5            -0.059       0.046   0.197
5 roomsor more in neighborhood 5     -0.046       0.053   0.394
1-room in neighborhood 6             -0.110       0.065   0.088
2 rooms in neighborhood 6            -0.020       0.046   0.658
4 rooms in neighborhood 6            -0.068       0.041   0.092
5 rooms or more in neighborhood 6    -0.128       0.045   0.005
1-room in neighborhood 7             -0.147       0.066   0.027
2 rooms in neighborhood 7            -0.009       0.046   0.852
4 rooms in neighborhood 7            -0.090       0.040   0.025
5 rooms or more in neighborhood 7    -0.103       0.051   0.041
1-room in neighborhood 8              0.240       0.077   0.002
2 rooms in neighborhood 8             0.070       0.052   0.179
4 rooms in neighborhood 8            -0.163       0.046   0.000
5 rooms or more in neighborhood 8    -0.086       0.052   0.097
Number of observations                2 215
R square                               0.63

                                                                  22
260

                         240

                         220                                              Lyon
                                                                          Marseille
                         200                                              Toulouse
                                                                          Rennes
   Index (2000 Q4=100)

                         180
                                                                          Strasbourg
                         160

                         140

                         120

                         100

                          80

                          60
                           -4

                                  -2

                                         -4

                                                -2

                                                       -4

                                                              -2

                                                                     -4

                                                                            -2

                                                                                   -4

                                                                                          -2

                                                                                                 -4

                                                                                                        -2

                                                                                                               -4

                                                                                                                      -2

                                                                                                                             -4

                                                                                                                                    -2

                                                                                                                                           -4

                                                                                                                                                  -2

                                                                                                                                                         -4

                                                                                                                                                                -2

                                                                                                                                                                       -4

                                                                                                                                                                              -2

                                                                                                                                                                                     -4
                           94

                                  95

                                         95

                                                96

                                                       96

                                                              97

                                                                     97

                                                                            98

                                                                                   98

                                                                                          99

                                                                                                 99

                                                                                                        00

                                                                                                               00

                                                                                                                      01

                                                                                                                             01

                                                                                                                                    02

                                                                                                                                           02

                                                                                                                                                  03

                                                                                                                                                         03

                                                                                                                                                                04

                                                                                                                                                                       04

                                                                                                                                                                              05

                                                                                                                                                                                     05
                         19

                                19

                                       19

                                              19

                                                     19

                                                            19

                                                                   19

                                                                          19

                                                                                 19

                                                                                        19

                                                                                               19

                                                                                                      20

                                                                                                             20

                                                                                                                    20

                                                                                                                           20

                                                                                                                                  20

                                                                                                                                         20

                                                                                                                                                20

                                                                                                                                                       20

                                                                                                                                                              20

                                                                                                                                                                     20

                                                                                                                                                                            20

                                                                                                                                                                                   20
                                                                                                             Date

                         260

                         240

                         220
                                                                                         Nice
                         200                                                             Nantes
                          180                                                            Bordeaux

                          160
                                                                                         Lille

                          140

                          120

                          100

                          80

                          60

                                                                                                         D at e

Figure 1. The indexes for apartments in 9 cities (villes-centres)

                                                                                                                                                                                          23
Figure 2. The main Notaires-INSEE indexes : apartments

    220

    200
                                             Paris
    180
                                             Petite Couronne
    160                                      Grande Couronne

    140                                      Province

    120

    100

    80

    60
      -3

            -2

                   -1

                          -4

                                 -3

                                         -2

                                                -1

                                                      -4

                                                             -3

                                                                    -2

                                                                           -1

                                                                                  -4

                                                                                         -3

                                                                                                -2

                                                                                                       -1

                                                                                                              -4

                                                                                                                     -3

                                                                                                                            -2

                                                                                                                                   -1

                                                                                                                                          -4
     91

            92

                   93

                          93

                                 94

                                        95

                                               96

                                                      96

                                                             97

                                                                    98

                                                                           99

                                                                                  99

                                                                                         00

                                                                                                01

                                                                                                       02

                                                                                                              02

                                                                                                                     03

                                                                                                                            04

                                                                                                                                   05

                                                                                                                                          05
   19

          19

                 19

                        19

                               19

                                      19

                                             19

                                                    19

                                                           19

                                                                  19

                                                                         19

                                                                                19

                                                                                       20

                                                                                              20

                                                                                                     20

                                                                                                            20

                                                                                                                   20

                                                                                                                          20

                                                                                                                                 20

                                                                                                                                        20
Index, base 100 in 2000 Q4.

     8

     7                                Paris
     6
                                      Petite Couronne
     5
                                      Grande Couronne
     4
                                      Province
     3

     2

     1

     0
      -3

            -2

                   -1

                          -4

                                 -3

                                        -2

                                               19 1
                                                    -4

                                                    -3

                                                    -2

                                                    -1

                                                    -4

                                                    -3

                                                    -2

                                                    -1

                                                    -4

                                                    -3

                                                    -2

                                                    -1

                                                    -4

     -1
                                                    -
     91

            92

                   93

                          93

                                 94

                                        95

                                               96

                                                 96

                                                 97

                                                 98

                                                 99

                                                 99

                                                 00

                                                 01

                                                 02

                                                 02

                                                 03

                                                 04

                                                 05

                                                 05
   19

          19

                 19

                        19

                               19

                                      19

                                             19

                                               19

                                               19

                                               19

                                               19

                                               20

                                               20

                                               20

                                               20

                                               20

                                               20

                                               20

                                               20

     -2

     -3

     -4

     -5

     -6

     -7

     -8

Rate of increase of the index in percent.

                                                                                                                                               24
19

                                                 -8
                                                      -7
                                                            -6
                                                                 -5
                                                                      -4
                                                                           -3
                                                                                -2
                                                                                          -1
                                                                                                    0
                                                                                                        1
                                                                                                            2
                                                                                                                3
                                                                                                                    4
                                                                                                                        5
                                                                                                                            6
                                                                                                                                7
                                                                                                                                    8
                                                                                          91                                                                          19
                                                                                               -3                                                                       91

                                                                                                                                                                                60
                                                                                                                                                                                     80
                                                                                                                                                                                          100
                                                                                                                                                                                                120
                                                                                                                                                                                                      140
                                                                                                                                                                                                                  160
                                                                                                                                                                                                                                   180
                                                                                                                                                                                                                                         200
                                                                                     19                                                                                    -
                                                                                          92
                                                                                               -1                                                                     19 3
                                                                                     19                                                                                 92
                                                                                          92
                                                                                     19
                                                                                               -3                                                                          -
                                                                                          93                                                                          19 2
                                                                                               -1                                                                       93
                                                                                     19                                                                                    -
                                                                                          93
                                                                                               -3                                                                     19 1
                                                                                     19                                                                                 93
                                                                                          94
                                                                                     19
                                                                                               -1                                                                          -
                                                                                          94
                                                                                                                                                                      19 4
                                                                                               -3                                                                       94

                                                           Ile-de-
                                                                                     19                                                                                    -

                                                           France
                                                                                          95
                                                                                               -1                                                                     19 3

                                                           Province
                                                                                     19                                                                                 95
                                                                                          95
                                                                                               -3                                                                          -

                                                                                                                                        Index, base 100 in 2000 Q4.
                                                                                     19                                                                               19 2
                                                                                          96
                                                                                               -1                                                                       96
                                                                                     19                                                                                    -
                                                                                          96
                                                                                               -3                                                                     19 1
                                                                                     19                                                                                 96
                                                                                          97
                                                                                     19
                                                                                               -1                                                                          -
                                                                                          97
                                                                                                                                                                      19 4
                                                                                               -3                                                                       97
                                                                                     19                                                                                    -
                                                                                          98

     Rate of increase of the index in percent.
                                                                                               -1                                                                     19 3
                                                                                     19                                                                                 98
                                                                                          98
                                                                                     19
                                                                                               -3                                                                          -
                                                                                          99
                                                                                                                                                                      19 2
                                                                                                                                                                                                       Province

                                                                                               -1                                                                       99
                                                                                     19                                                                                    -
                                                                                          99
                                                                                               -3                                                                     19 1
                                                                                     20
                                                                                                                                                                                                                   Ile-de-France

                                                                                          00                                                                            99
                                                                                     20
                                                                                               -1                                                                          -
                                                                                          00
                                                                                                                                                                      20 4
                                                                                               -3                                                                       00
                                                                                     20                                                                                    -
                                                                                          01
                                                                                                                                                                                                                                               Figure 3. The main Notaires-INSEE indexes : houses

                                                                                               -1                                                                     20 3
                                                                                     20                                                                                 01
                                                                                          01
                                                                                     20
                                                                                               -3                                                                          -
                                                                                          02
                                                                                                                                                                      20 2
                                                                                               -1                                                                       02
                                                                                     20                                                                                    -
                                                                                          02
                                                                                               -3                                                                     20 1
                                                                                     20                                                                                 02
                                                                                          03
                                                                                     20
                                                                                               -1                                                                          -
                                                                                          03
                                                                                                                                                                      20 4
                                                                                               -3                                                                       03
                                                                                     20                                                                                    -
                                                                                          04
                                                                                               -1                                                                     20 3
                                                                                     20                                                                                 04
                                                                                          04
                                                                                               -3                                                                          -
                                                                                     20                                                                               20 2
                                                                                          05
                                                                                               -1                                                                       05
                                                                                     20                                                                                    -
                                                                                          05
                                                                                               -3                                                                     20 1
                                                                                     20                                                                                 05
                                                                                          06
                                                                                               -1                                                                          -4

25
You can also read