Upload
independent
View
0
Download
0
Embed Size (px)
Citation preview
The influence of Environmental policies on farmland Prices in the region Bretagne of France
Elodie Letort1 and Chalachew Temesgen2
Abstract
The region Bretagne is the top leading agricultural region in France in terms of agricultural annual turnover. In addition to urban pressure, the competition for farmland is enhanced by strong environmental regulations and incentives. In this paper, we study some determinants of farmland prices in Bretagne. Two particular points are stressed. The first one concerns the role of the different operators of the land market. The second one concerns the impacts of environmental policies and regulation on land market. For this, we estimate spatial hedonic pricing models based on individual level data. This data set gathers all transactions of farmland sales which are notified by notaries in Bretagne from 2007 to 2010. Estimation results show an increase in farmland prices in more constrained environmentally areas. It explains by the rising demand from farmers who have to meet the manure spreading regulations. Results also show the importance of spatial interaction on farmland market.
Key-words: environmental policies; hedonic price function; spatial econometric model.
1 ARAP (Association Régionale pour l’Agriculture Paysanne) – Research analyst associated with UMR SMART INRA, Rennes, France
2 UMR SMART INRA Rennes, France.
Introduction
The region Bretagne is a very agricultural area located at the north-west of France3.
Agriculture covers about 65 % of total land area. This share is higher than the French average
(53%). However the decline of farmland is faster in Bretagne, where almost 3% of the
regional farmland has been converted to non-agricultural activities between 2000 and 2010,
and only 1% in France (Agreste, 2011). The demands for land by agriculture and by non
agricultural businesses have both strong implications for the farmland sale market where both
farmers and non-farmers operate. French young farmers therefore face intense competition
from the farmers in place and non-farmers who live towns and urban areas (Levesque, Liorit,
and Pathier, 2012). Various studies report that farmers want to enlarge the operational area
size of their farm, with a farmers’ tendency to specialize their production system (Levesque et
al., 2012). Besides, non-farmers put a pressure on farmland so as to reap the prospect of
windfall gain due to future expected changes in land price over time. This capital gain is
mainly depending on the location and accessibility of land to the nearest urban growth pole
and communication center. The existing environmental regulation and agricultural zoning
policies is another factor that affects the structure of farmland demand within agriculture.
These double edge demands for farmland resource have two effects in the functioning of
farmland market in France. The supply effect refers to the declining availability of farmland
for agriculture production. The demand effect refers to the agricultural productivity of
farmland. The demand effect would naturally increase the market clearing price. However, the
trend of farmland prices in France has remained relatively stable over time as compared to
other European countries (Latruffe, 2008 and Swinnen, 2009)4. This is due to land market
regulations that involve the French government and farmers' organizations. Regional land
offices (Sociétés d’aménagement foncier et d’établissement rural or SAFER) operate on
farmland markets according to policy objectives concerning agriculture, environment or
infrastructure development. In addition, the farmland rental rates are constrained by
administrated bounds and the law limits the rights of landowners to protect farmers’ access to
3 According to the European nomenclature of regional levels (NUTS), Bretagne is one of the 22 NUTS2 regions of metropolitan France. There are 4 additional French oversea NUTS2 regions. The different levels of territorial units for the region Bretagne are presented in Annex A.
4 The historical price of farmland was increasing continuously during 1960s until 1980s and declines or stable from 1980s to 1996 and started to increases slightly the price from 1997 to 2010 ( Latruffe et al., 2008; Cavailhes et al., 2011).
farmland. These regulations often induce side payments in farmland transactions which are
not registered in their observed prices.
Principally, there are two main approaches to study the determinants of land prices. The first
approach is based on the actualized value of the farmland which is known as the Net Present
Value (NPV)5 model. The NPV model is based on theoretical and empirical developments of
the Ricardo capitalization formula. It sets the present value of the land as the discounted sum
of future expected revenues provided by farmland. This approach is often used to explain the
temporal evolution of the price of land in relation to macroeconomic variables. The NPV
farmland price model is also used to study the impact of agricultural support policies on the
farmland price. Several studies have shown that the policy agricultural supports, especially
direct payments to farmers, capitalize into the farmland price, either fully or partially,
depending on the modalities of their implementation (Floyd, 1965, Altson and James, 2001;
Guyomard et al., 2004, Cianian and Swinnen, 2006 and 2007 and Feichtinger and Salhofer
2011). The second approach relies on the construction of a hedonic price model. The hedonic
method helps to derive the marginal values of farmland characteristics which affect the
willingness to pay of buyers (Palmquist, 1989). The price of farmland should reflect its
characteristics, especially geographical location characteristics. This analysis deals with
individual plots of farmland over the defined geographical zones. The hedonic approach has
already been employed in a number of farmland market researches. For example, Miranwski
and Hammes (1984) determined the implicit price of for soil characteristics in Iowa (USA).
Palmquist and Danielson (1989) applied this technique to measure the impact of soil erosion
and drainage on farmland prices. Shi, Phipps and Colyer (1997) and Plantiga et al. (2002)
showed how parcel characteristics and urban factors influence farmland sales price in United
States. Cavailhés and Wavresky (2003) analyzed the urban influence in the southern part of
France. Attention has also been given to the buyers and sellers characteristics (King and
Sinden, 1994; Harding, Knight, and Sirman, 2003; Geerte Cotteleer, Corneils Gardbroek, and
Jan Lujit,2008). Empirically there are very few works dealing with environmental policies
and regulations.
The hedonic method is applied in this paper. Farmland characteristics include productive
values of farmland with the effects of environmental policies and regulations of farmland
market, taking into account the characteristics of buyers and sellers that may interfere with the
5 NPV models of farmland price are considered as theoretically sound and are the most cited model in farmland price literature (Alston 1986, Burt 1986, Featherstone et al.1987, Campbell and Schiller 1987 and Clark 1993).
valuation of the different farmland characteristics. We use a unique data gathering of the
individual transactions in Bretagne from 2007 to 2010.
Two particular points are stressed in this paper. The first one concerns the role of the different
operators of the land market. As farmland can be used for different purposes, we assumed that
a farmland will not be valued in the same way by all agents. For example, Facchini (1997)
defined three types of farmland values. These are the productive value, the investment value
and the consumption value. The production value of farmland is the profitability of
agricultural activities which usually depends on the marginal productivity of the land. The
investment value derives from the rental value of farmland. When an investor seeks to
enhance its savings by purchasing farmland, the farmland reflects both the current agricultural
rent and the potential future non-agricultural rental income if it is converted to urban
activities. The location and accessibility of the land will determine the non-agricultural rental
income and its probability. Finally, some individuals or agents would likely to buy farmland
only for personal recreational activities. They can use this farmland as garden or sport field
for example. This is defined as "the consumption value" of farmland that usually correspond
tonon-monetary benefits for the owner. The measurement of this consumption value is beyond
the scope of this paper, because it requires data which are not available. Therefore, this paper
only focuses on the production value and on the investment value of farmland.
The second important point of this paper concerns the environmental policies and regulation
according to the specificities of the studied region. Bretagne area is divided according to
several environmental zones. These zones are defined according to animal density and water
quality6. In these zones, farmers must respect restricted practices of fertilization and manure
management. In addition, the increase in animal density is limited or forbidden. There are
policy documents that outline the specifications of fertilization, manure spreading plan for
each farming type. For example the nitrate regulation limits the organic fertilization. Farmers
with manure excess must seek additional areas for manure spreading to maintain or increase
their herd size. This leads to increased competition between farmers which may increase the
price of the farmland. The effect of this particular environmental regulation on the price of
farmland is identified and measured in this paper.
The remainder of the paper is organized as follows. The second section briefly outlines the
specificities of the farmland market in Bretagne. The third section gives a brief description of
6 Following the Nitrate Directive (1991), various devices are implemented at the European level to reduce agricultural
pollution and improve water quality. These are mainly preventing nitrate content form farming practices and promoting environmentally friendly agricultural activities.
the hedonic approach and its specification is discussed. The fourth part provides the
description of data sets and variables and discusses the spatial autocorrelation problems and
related econometric solutions. We use a data set belonging to PERVAL. This data set gathers
all transactions of farmland sales which are notified by notaries. The results are presented and
discussed in the fifth section.
1. Characteristics of the farmland market in Bretagne
The region Bretagne is the top leading agricultural region in France in terms of agricultural
annual turnover. It is the first region for the production of milk, eggs, pork and poultry, and
several vegetables, such as cauliflower, artichokes and potatoes. All these various farms
compete for farmland in the region. Vegetable farmland is usually more expensive than crop
farmland and animal grazing farmland. Pigs and poultry require a minimal area for spreading
livestock manure. The region Bretagne is also a great region of mixed farming dominated by
milk production. Milk production is regulated by dairy quotas, introduced in 1994. These
quotas are not exchanged on a market, but their transfer is permitted with associated land.
Some studies found evidence that dairy quotas are capitalized into farmland prices
(Bartholomew and Boinon, 2001). This is also the case for the single farm payment, which
can be transferred between farmers since 2006 with farmland transactions.
On the other side, the competition for farmland is also intensified by strong urbanization
effects which are induced in part by the regional demographic dynamism. Non-agricultural
use of farmland is strong around major cities and peri-urban areas in Bretagne. Nearly half of
the municipalities of Bretagne belong to an urban area. Bretagne region is also bordered by
2800 km of coastline. On the Bretagne coast, the difference in land prices can sometimes vary
with a ratio of 1 to 200. These urban developments encourage investors to buy farmland in
the most coveted areas, anticipating a future conversion from agricultural to residential use. ,
Their willingness to pay is probably much higher than most farmers’ one. The impact of
competition between different land uses, particularly between agriculture and urbanization has
been studied recently by Lefebvre and Rouquette (2011) and Dachary et al. (2011). They
conducted a hedonic farmland price analysis for all regions of France between 1995 and 2010.
They showed that the demographic pressure and accessibility to urban centers are important
drivers of the farmland price.
In addition to urban pressure, the competition for farmland is enhanced by environmental
regulations and incentives in Bretagne. These policies mainly target and the water quality. In
1993, the regional authorities identified and classified “nitrate vulnerable zones” according to
the nitrate concentration of surface water. In 1996, additional measures have been
implemented in designated areas (French acronym ZES) with higher environmental pressure
from agriculture. These designated areas (ZES) have animal densities resulting in nitrogen
surplus that exceeds the ceiling of the nitrate directive (170kg of nitrogen per hectare). In
2001, the nitrate directive also motivated the creation of areas with complementary actions
(French acronym ZAC) in order to improve the quality of water used for the production of
drinkable water. Complementary actions are mainly winter coverage of arable land. Nearly
half of the NUTS47 regions in Bretagne were ZES in 2006 and nearly one-third of the water
basins are under the obligations of ZAC. Since 2002, a part of the region Bretagne is in
European litigation for non-respect of the Nitrates Directive of 1975 concerning the quality
required of surface water intended for the abstraction of drinking water. In these areas,
farmers are constraints by additional mandatory measures.
In 2009, a huge algal bloom was observed on the beaches of Bretagne. This phenomenon
enhanced the recurring debate on water quality in Bretagne and the poor effectiveness of
environmental policy measures in the agricultural sector. Nitrogen discharges associated with
animal effluent and fertilizer were reckoned as the main cause of the proliferation of the algal
bloom, the algae decomposition on beaches produces toxic gases. The death of wild boars in
one of the famous Bretagne Bay in July 2009 due to toxic gas further stimulates the public
debate and the tensions between agricultural and environmental lobbies. A national action
plan was prepared in 2010 for 8 designated water basins that correspond to the bays most
impacted by algal blooms in Côtes d'Armor and Finistère, two NUTS3 regions of Bretagne.
The action plan aims at reducing nitrate flows by 30% to 40% before 2015. This plan includes
curative and preventive measures. At the farm level, the plan requires the reporting of
nitrogen management, the shortening of manure spreading periods and fertilization limitations
according to the nitrogen balance. These measures are not more stringent than those existing
in ZES or in areas in European litigation. The plan also proposes voluntary measures,
encouraging the development of grassland based production systems.
7 The different levels of territorial units for the region Bretagne are presented in Annex A.
Few studies have examined the impact of environmental policies on the farmland price. Le
Goffe and Salanié (2005) analyzed the impact of the progressive implementation of the
"Nitrate Directive" in Bretagne from 1994 to 2000. This implementation consists in the
ceiling of the quantity of organic nitrogen per hectare. The theoretical approach of the paper
assumes that farms above the ceiling buy the right to spread manure from farms operating
below the ceiling, or buy additional land for the same purpose. Their analysis focused on in
house pig production. Their empirical investigation showed that in regions characterized by
high densities of pigs, the equivalent land rents increase by, 1 € per kg of nitrogen8. This cost
is higher than the rate of farm pollution tax (between 0.15 and 0.30 € / kg of excess nitrogen)
but much lower than the estimated cost of manure treatment in dedicated plants (€ 3 /kg of
nitrogen). They conclude that the regulation has some effect on farmland price, reflecting the
fact that pig farmers were forced to deviate from their unconstrained profit maximizing
behavior.
This kind of environmental policies and directives increases the competition between farmers
because of the limited availability of farmland. In one way or another, this would affect
prices. In addition the farmland market in France is governed by a set of laws and legal
institutions. SAFER is a public and non-profit agency which is composed of shareholders
(representatives of chambers of agriculture, trade unions, banks, representatives of councils,
management center ...). The main mission and responsibility of SAFER is to regulate the
farmland market in every French region. This agency has a mandate to improve the young
farmers’ accessibility to farmland, to help smallest farms to enlarge and to moderate land sale
price. SAFER directly operates on the farmland market either by agreement or by using its
pre-emption right9 given by law.
The status of farming tenancy is another specificity of the French farmland policy which may
go against free market mechanisms in many aspects. Tenant farmers are more protected by
law than the landlords. On the one hand, the tenant farmer has automatic right to purchase the
land he farms if his landowner wishes to sell. In our study, for example, tenant farmers are
buyers in 40% of the transactions from 1995 to 2010 in Bretagne. On the other hand, the
duration of a lease is usually nine years and automatically renewed as many times as the
8 This is considering organic nitrogen fertilization of 100 kg/ ha. 9 The friendly acquisition is where an individual decides to sell directly to the SAFER. The preemptive right is the right to
acquire property in priority to any other person when the owner expresses its willingness to sell.
tenant wants to remain on the farm. This is called "life time tenancy contract". In France, it is
very difficult for a landlord to retrieve the farmland from his tenant farmer10. This can explain
that higher sale prices are observed for "free farmland" (farmland without ongoing rental
contract at the time of sale) than for "leased farmland" (farmland with going rental contract at
the time of sale). A recent study by Lefebvre and Rouquette (2011) indicates that that this gap
has been steadily widening for the past twelve years.
2. The hedonic approach
Before discussing the specification of the empirical hedonic price model, a brief theoretical
recall is presented.
The theoretical foundation of the hedonic price method was developed by Lancaster in 1966.
In his seminal work, Lancaster reckons that consumer goods are quite heterogeneous and that
comparisons between each other are difficult. Lancaster makes the assumption that
consumer’s utility does not directly derives from the consumption good, but from its
characteristics or attributes. This decomposition of any heterogeneous consumption good into
homogenous attributes facilitates the comparison between two goods. The hedonic price
method estimates the implicit price of each attribute by regressing the good price over its
attributes. In 1974, Rosen used the theoretical framework of Lancaster to analyze the
functioning of the housing market and estimates a hedonic price function with the
characteristics of houses. The hedonic price function estimates was used to measure the
implicit price of each house characteristic and to calculate the willingness to pay for its
marginal change. Following this work, several problems were identified including the
potential simultaneous choice between the house price and the quantities of certain
characteristics, or the correlation between the explanatory variables and residuals (Epple,
1987).
This method was applied to the price of farmland by Palmquist (1989) who showed how to
derive the bid function for a plot of farmland. Different plots of farmland are endowed with
different characteristics in terms of soil quality, climate, irrigation potential and infrastructure.
We assume that a person buys a particular plot for its attributes and its location, and that price
of this land plot is determined by buyers’ willingness to pay for these specific characteristics.
It is assumed that no individual is able to influence the hedonic price equation as the market 10
Presently over two third of the current farmland is cultivated by non-owner farmers.
clearing price would eliminate the excess supply and demand for each type of farmland. This
approach has been widely used in the literature to study the agricultural land price in different
countries, such as Georgia (Elad, Clifton, and Epperson 1994), United States (Bastian et al.,
2002), Northern Ireland (Patton and McErlean, 2003), France (Le Goffe and Salanié, 2005)
and Finland (Pyykkönen, 2005).
We use different sets of explanatory variables to estimate the price of the land. Variables
describe the characteristics of the land, such as the size of the plot or the soil quality. These
factors affect the productivity of the land and therefore the expected income of farmland. All
policy supports that are bound to the agricultural area or production, like the manure
spreading rights and dairy quotas, can be capitalized into the land and are included in this
model. Variables, such as the coastline proximity or the location in urban-rural fringe area,
can represent the intensity of non-agricultural activity demand for land. Our model includes
two additional sets of variables: i) variables indicating the tenancy status of the land plot (land
under tenancy contract or farmland without on-going tenancy contract) and the direct or
indirect involvement of SAFER in the transaction process; ii) variables that represents the
environmental situation of the municipality of the transacted farmland. The simpler hedonic
price function applied to individual land price observations is linear and encompasses the
preceding sets of variables. It can be written as follows:
P X Z S F E= α + β + γ + ζ + δ + η + ε (1)
P is the vector of observed prices of transacted farmland plots (unit is euro per hectare,
excluding tax and notary fees) of farmland, X is the matrix of agricultural characteristics of
the plot, Z is the matrix of its non-agricultural characteristics, S is the array of policy
instruments related to the farmland, F and E describing the institutional and environmental
situation of the plots of farmland, respectively. The stochastic error term is represented by ε .
Several functional forms can be used in hedonic studies. The functional form of the hedonic
regression equation can either be linear, semi-log, or log-log. The most common specification
is the semi-logarithmic form. Each parameter estimate directly provides the percentage of
price that depends on the corresponding characteristic. The parameters measure the relative
change of the price following a unit change of respective characteristics. We also chose the
semi-log specification for this facility of implementation11. The econometric specification is
modified to take into account the spatial dependence and the spatial autocorrelation of our
observations of transacted farmland plots.
3. Problems of spatial autocorrelation
Where data have a spatial dimension, two specific issues must be considered. These are
spatial heterogeneity and spatial autocorrelation. Bretagne municipalities are highly
heterogeneous. A part of this heterogeneity is controlled by the inclusion of certain
municipality characteristics in the set of explanatory variables: population density, location in
suburban area, coastline proximity in our case study. If spatial unobserved heterogeneity
remains, we are confronted with a problem of heteroskedasticity and/or instability of the
model parameters that vary systematically with location (Le Gallo, 2000b). Taking account of
this unobserved heterogeneity can be done by correcting a possible heteroskedasticity and /or
using standard econometric methods (random model parameters ...).
Unlike the spatial heterogeneity, the treatment of spatial autocorrelation requires specific
econometric methods. Spatial autocorrelation is defined as the correlation of a variable with
itself according to the geographical pattern of the observations. This can be a spatial
dependence between the observations of the endogenous variable, a spatial dependence
between observations of exogenous variables or a spatial dependence between the error terms.
This problem is usually caused by omitted variables which have spatial dependence. In our
case, the sale price of farmland may be affected by the value given to surrounding farmland
and by the attributes of surrounding farmland. Location factors such as the demographic
pressure and the urban geographical structure of the area are the main factors that influence
the price of farmland, apart from its production values. Spatial autocorrelation destroys the
independence of observations which is assumed in usual econometric methods like ordinary
least squares (OLS). It is therefore necessary to detect their presence.
There are strong and complex links between spatial dependence and spatial heterogeneity.
Poor model specification or omission of explanatory variables can cause heteroskedasticity
11 The Box-Cox transformation is often used for its flexibility. Three reasons motivated the choice of the log-linear form: the interpretation of the results is simpler, easier to adapt to model spatial autocorrelation, and several studies have shown that the results changed little between the two models (Le Goffe and Salanié 2005).
and can also lead to spatial autocorrelation of the error terms (Le Gallo 2000a, 2000b, 2002).
It is therefore difficult to distinguish autocorrelation effects and heterogeneity effects between
each other. Similarly, the correction of a problem linked to the spatial dimension of the data is
likely to have side effects on other potential problems. For example, the inclusion of
explanatory variables in the model to control the spatial heterogeneity is likely to reduce or
eliminate the spatial autocorrelation of errors. In addition, the autoregressive model
specification with a spatially lagged endogenous variable probably captures the influence of
omitted variables on the dependent variable and reduces the presence of spatial
autocorrelation of error terms. It is therefore difficult to detect a specific dependence effect in
the presence of different forms of spatial dependence and heterogeneity.
Methods for testing and accounting for spatial autocorrelation were developed in the late
1970s. Since 2000, these methods have been improved and applied to various empirical
studies. In parallel new theoretical approaches, such as economic geography, have been
developed and the availability of spatial data has been increasing a lot. To test and capture the
spatial interdependence between observations, we must consider the geographical position of
the farmland. We have the municipalities in which sales occur but we do not know the exact
position of the transacted land in each municipality. So we start from the assumption that the
spatial interaction between two farmland sales depends on the distance between the
municipalities in which the farmland is located. The instrument used to represent this
interaction is the spatial weight matrix. The weight matrix enables the connection of each
observation with the others according to their relative geographical location. If y is a spatial
variable and W the weight matrix, we can measure the intensity of the overall effect of the ith
observation values in space by expression (2):
[ ]1
N
ij jij
Wy w y=
=∑ (2)
This notion of spatial lag is important because it allows us to introduce the effects of spatial
autocorrelation in the econometric models. The weight matrix can be written in different
ways. The technique often used in the literature (Patton and McEarlen 2002 Pyykkönen 2005)
consists of filling in the matrix with the inverse of the squared distance for each pair of
geographical locations to represent how municipalities are spatially connected. Without the
precise location of the farmland within its municipality, we use the municipality area to
calculate and distance between two hypothetical farmlands randomly located in the same
municipality. Above a certain distance between two municipalities we assume that the spatial
interaction is zero. The choice of this distance threshold depends on the size of the farmland
market in our studied area. By convention, the diagonal elements of the matrix are equal to 0.
These matrices are often normalized with the sum of each row set to 1.
The spatial lag autoregressive model (SAR model), characterized by the autocorrelation of the
endogenous variable is written as (3):
P WP Q= α + ρ + µ + ε
(3)
Where Q include all characteristics’ variables. This specification takes into account the
interactions that may exist between neighbors in determining the selling price of farmland.
The second term of the right hand member is the spatially lagged term. It should be treated as
an endogenous variable. OLS is not appropriate for this model since this estimator would be
biased and inefficient. The specification of the spatial error model (SEM model) with the
spatial autocorrelation of the error terms is written as (4):
W vε = λ ε + (4)
The error term is split into v which refers to the true independent homoskedastic residual
term which has mean zero and a constant variance. In this case, the OLS estimator is unbiased
but inefficient. The details of both models are developed in Lesage and Pace (2009). There is
another model that combines both a lagged endogenous variable and the spatial correlation of
error terms. It called spatial auto-correlation model (SAC model).
The spatial Durbin model (SDM model), characterized by a spatial lag of the dependent
variable and a spatial lag of the explanatory variables, is written as (5):
P WP WQ= α + ρ + µ + ε
(5)
Since the year 2000, the farmland market studies using the hedonic price approach have
focused on the potential spatial interactions between neighboring transactions. Elad et al.
(1994) segmented the land market into different local sub-markets to measure the spatial
heterogeneity. They estimated a specific hedonic price function for each sub-market. Patton
and McErlean (2003) introduced advanced spatial econometrics to estimate a hedonic price
model in Northern Ireland farmland. Their result showed that there are many spatial
interactions in this market: spatial heterogeneity and spatial dependence between the
observations of the endogenous variable. Ignoring these effects may lead to biased estimates.
These results suggest that it can be difficult for the owner to identify the value of his own
farmland characteristics and to set appropriate price. In this case, potential sellers of the
farmland set prices according to the historical sale price of nearby plots even if these plots
have different characteristics. This mimetic behavior introduced a direct influence of one’s
transaction on other neighborhood transactions.
There are various tests for spatial autocorrelation in the literature. These are based on the
Moran test and the statistical test of the Lagrange Multiplier (LM). These tests can detect the
presence of one or the other form of spatial dependence. The methodology and explanation of
these tests are largely presented in Le Gallo (2000) and Lesage and Pace (2009). In cases
where the both types of dependence exist, Anselin and Rey (1991) propose to retain the model
corresponding to the highest test statistical value. Pyykkönen (2005), Patton and McErlean
(2003) followed this rule and estimate a model with lagged endogenous variable to describe
farmland market.
Maximum likelihood (ML) is consistent for spatial models. The ML adapted approach is to
estimate part of the first-order conditions in a first step. In a second step the solutions of the
first step are introduced into the log-likelihood function. This log-likelihood function is said
"concentrated" since it depends on a fewer parameters. Much of the spatial econometrics
literature has focused on ways to avoid maximum likelihood estimation because of
computational difficulties. Patton and McErlean (2003) estimated this model using an
instrumental variable method based on the White estimator of the variance-covariance matrix
which is robust for any heteroskedasticity forms (Anselin and Bera 1998). However, it was
shown that tests for heteroskedasticity are not always reliable in the presence of spatial
autocorrelation of error terms (Anselin and Griffith, 1988). Lagged explanatory variables are
generally used as instruments (Kelejian and Robinson 1992). Pyykkönen (2005) compared an
adapted Maximum Likelihood (ML) estimator and with the preceding instrumental variable
(IV) method to estimate a model with lagged endogenous variable applied to the Finnish
farmland market. He finds that results of these two approaches are very similar.
Recently, some authors, like Lesage (2009), provide new approach to reduce computational
tasks and to construct maximum likelihood estimates in only few minutes. In this paper, we
use this approach to avoid choosing instrumental variables in case of lag dependent variable.
4. The empirical Model
All information of farmland transactions in Bretagne comes from the "PERVAL" data base.
We select the observations of farmlands with no buildings or forested land. After eliminating
farmland corresponding to the tails of the distribution of prices, a total of 5,700 observations
from 2007 to 2010 were used for the analysis. Additional variables describing the location
(the municipalities in this case) of the traded farmland are obtained from the agricultural
census of 1999 and 2008 of "INSEE" data base. The description and summary statistics of
these variables are presented in Table 1.
The price of farmland is defined in euro per hectare excluding transaction costs (trading costs)
and notary fees. The nominal price was deflated by the consumer price index (base 100 in
2005). Total size of traded farmland is used as an explanatory variable in the model. The
agronomic quality of soil is not directly observable in the database. However, "potentiality of
irrigation" was taken into account. Variables indicate whether the sold farmland have a
system of irrigation infrastructures, a drainage facility or a retention pond. The proportion of
vegetable farms in the Utilized Agricultural Area (UAA) of the municipality (where the
farmland sale is located) approximates the soil quality. An index of soil quality is build at the
NUTS5 level12. It indicates if the soil is more clay, silty clay or sandy. A dummy variable was
also built for each NUTS4. The average land rental rate is also including in the model. It
provides information on agronomic quality of soils and productivity of farmers because its
calculation is based on the average income of farmers in each NUTS3 region, and it can be
different from one area to another depending on agronomic capacity.
The geographic location of the farmland is an important factor for some buyers. We therefore
considered two variables: a variable that represents a farmland geographical proximity to the
coast and a variable that identifies whether a farmland is located in an urban area13. These
variables, associated with the population density, approximate the competition effect of
urbanization and tourism. The value of these farmland attributes would likely differ according
to the characteristics of the buyer. These variables were therefore crossed with the status of
the buyer (farmers, non-farmers, farming companies, local authorities and SAFER).
12
The different levels of territorial units for the region Bretagne are presented in Annex A.
13 A municipality is called in an urban area, if it belongs to an urban area (center offering at least 10,000 jobs and is not located in the crown of another urban center) or suburban (all municipalities in the urban area to the exclusion of the pole).
The role of the farmland market rules and policies is approximated by two variables in the
model: first, a variable represents whether a farmland was purchased by the farmer in place or
not and second, a variable indicates the involvement of local or regional associations or the
involvement of SAFER during the transaction process.
Table 2. Descriptive statistics of variables
Variables Unit Average Standard
devation
Data base
Prices of the sold farmland plot €/ha 4016 2974 Perval 2004-2010
Agricultural Factors
Area of the plot ha 7.21 11.40 Perval 2004-2010
Possibility of irrigation yes=1/no=0 0.01 - Perval 2004-2010
Vegetable share in UAA /ha SAU commune 0.05 0.14 RGA 2010
Soil quality - -
Clay soils yes=1/no=0 0.38 - Build variable
Silty clay soils yes=1/no=0 0.30 - Build variable
Sandy soils yes=1/no=0 0.01 - Build variable
Other soil quality yes=1/no=0 0.31 - Build variable
Non-Agricultural factors
Urban zone yes=1/no=0 0.52 -
Coastline proximity yes=1/no=0 0.24 -
Farmland policy factors
Land was leased by the purchaser yes=1/no=0 39.66 - Perval 2004-2010
Land acquired by a farmer yes=1/no=0 0.61 -
Land acquired by societies yes=1/no=0 0.09 -
Land acquired by non farmer individuals yes=1/no=0 0.26 - Perval 2004-2010
Land acquired by the SAFER yes=1/no=0 0.03 - Perval 2004-2010
Land acquired by local authorities yes=1/no=0 0.01 - Perval 2004-2010
Land rental rate €/ha 147.99 24.14 Build variable
Agricultural Policy factors
Density of dairy cows /ha SAU commune 0.46 0.16 RGA 2010
Single farm payments in Ille-et-Vilaine €/ha 379.82 55.79 Build variable
Single farm payments in Côtes d’Armor €/ha 375.36 31.75 Build variable
Environmental Policy factors
Porcine nitrogen load Kg N/ha SAU commune 115.53 43.29 RGA 2000
ZES yes=1/yes=0 0.60 - Build variable
Green algae areas (since 2010) yes=1/no=0 0.001 - Build variable
Contentious areas yes=1/no=0 0.10 - Build variable
Agricultural policy factors must be taken into account to explain farmland prices. Milk quotas
or payment entitlements transferred with farmland will likely have a positive impact on land
prices. The density of dairy cows at municipality’s level is included to approximate the
probability that the exchanged land is associated with a milk quota. The average single farm
payment entitlements at the NUTS4 regional level are available only for 2 NUTS3 (Ille-et-
Vilaine and Côtes d’Armor)14.
Concerning environmental policy factors, the farmland demand for manure spreading was
measured by the nitrogen pressure indicator at the municipality (NUTS5) level which was
designed by Le Goffe and Salanié (2005). Data on the nitrogen load due to total animal
production were directly calculated from the 2000 agricultural census of French Agriculture
Ministry data base15. We also crossed this variable with a variable indicating the location in
ZES, where total organic load exceeds the available land for manure spreading. The square of
this variable was included in the model as well to grasp any non linear effect. It was expected
that the demand for farmland differs in environmentally sensitive areas. Some dummies
variables (Green algae area) indicate the location in one of the bays targeted by the national
plan against algal blooms, and the location in one of the area concerned by environnemental
litigation.
The choice of the model is based on statistical tests. As a first step, the model is estimated by
OLS. From the results obtained, tests based on Lagrange multiplier and likelihood ratio, were
performed to detect the presence of autocorrelation. Another test is performed from the SAR
model to confirm the presence of spatial error dependence. All these statistical tests, presented
in Table 2, make evidence of spatial correlation in residuals and in dependent variable.
Table 2. Statistical tests for spatial auto-correlation
Test Model Value Probability Chi-squared
LR test OLS 347.129 0.000 6.635
LM test OLS 646.08 0.000 17.611
LM test SAR
They have different implications for coefficient estimates in case of misspecification. Table 3
describes properties of coefficient estimates obtained by the four models (SEM, SAR, SAC
and SDM model) in cases of misspecification. For cases where the true data generating
14
These informations are provided by local authorities of the NUTS3 regions of Ille-et-Vilaine and Côtes d’Armor.
15 Information available in the RGA 2010 does not allow us to calculate the animal nitrogen load; this is why we calculate this variable from the RGA 2000. In addition this 2000 variable can be assumed exogenous in our regression without any problem.
process (DGP) has spatial dependence in disturbances and includes spatial lag dependence,
the SAR, SAC and SDM models will produce unbiased coefficient estimates. The SDM
model is the only model that will produce unbiased coefficient estimates under all four data
generating process (James and Lesage 2009).
Table 3. Properties of coefficient estimates depending on the true DGP
True DGP SEM model SAR model SAC model SDM model
SEM - biased/ inefficient biased/inefficient unbiased/inefficient
SAR unbiased/inefficient - unbiased/inefficient unbiased/inefficient
SAC biased/inefficient unbiased/efficient - unbiased/inefficient
SDM biased/inefficient biased/ inefficient biased/ efficient -
Three estimates are then performed: the SAR model defined by equation 3, the SAC model
(without the specification of two different weight matrix), and the SDM model defined by
equation 5. All spatial models are estimated using the method of maximum likelihood and
codes developing by Lesage on Matlab and available on his website16.
We set to 0 the weighting matrix elements if the distance between the two corresponding
municipalities is greater than 10 km. Above this distance, we assume no spatial interaction
between the endogenous variables. Various estimates were performed using different
threshold distances. Our choice was based on R² and log-likelihood. However it must be
stressed that the results appear to be insensitive to the choice of this threshold distance
between 10 and 30 km.
The estimated coefficients from the three methods are shown in Table 4. Pace and Lesage
(2006) distinguish direct effects and indirect effects. In the cases where the model contain
spatial lags of explanatory or dependent variables, the interpretation of parameters are more
complicated. In fact, a change in the explanatory variable for an observation can potentially
affect the dependent variable in all other observations. The average direct impact represents
the average response of the dependent variable according to independent variables. This is
similar to typical regression coefficient interpretations. It measures the effects of change in
16
http://www.spatial-econometrics.com/
the ith observation of Q on iP . In contrast, the average indirect impacts measure the effects
of changes in the ith observation of Q on jP for j i≠ . The average total impacts measure
how changes in a single observation influence all observations.
For continuous variables, parameters were multiplied by 100 to provide the percentage change
in price due to a unit increase in the quantity of the attribute. Similarly, for binary variables,
parameters were also multiplied by 100 to provide the percentage change in price due to the
introduction of the characteristics relative to the baseline.
5. Results
To compare the three spatial models, we use some quality criteria. According to the log-
likelihood value in Table 4, the SDM model is the best one. Another way to compare models
is to consider R² criteria. The SAC model has the highest R² value and the largest number of
parameters statistically different from 0. Given the results, it is difficult to decide which is the
best model. Results will be based mainly on those obtained by SAC model which provide
more detailed information with 23 significant estimates. Table 5 presents average direct
effects estimated from the three spatial models. Table 6 presents average indirect (spillover)
effects. Average total effects are obtained by adding direct and indirect effects.
Table 4. Comparison of model selection criteria
SAR model SAC model SDM model
Log-likelihood -1830.19 -1820.5 -1786.4
R² criteria 0.272 0.326 0.300
Significant parameters 21 23 15
Parameter ρ 0.608*** 0.738*** 0.452***
Parameter λ - -0.462*** -
Coefficient estimates and inference of some parameters highly depend on model used. Results
obtained by SAC and SAR model are relatively similar. The main difference between these
two models is that the indirect effects are more important from the SAC model, because of the
higher parameter ρ (associated to the lag dependent variable). Results obtained by SDM
model are more different, especially concerning indirect effects. In this model, the indirect
effects are relied not only on lagged dependent variable but also on lagged explanatory
variables. Indirect effects show how changes in a single observation influence all
observations. The SDM model allows the spillover from a change in each explanatory
variable to differ, as opposed to the SAR and SAC cases which have a common multiplier for
each variable.
The potentiality of irrigation represents approximately 3% of the total transaction farmland
sales in the data. Le Goffe and Salanié (2005) showed that irrigated plot is 75% more
expensive than non-irrigated plot and a drained plot is 25% more expensive than non-drained
farmland. These results are closed to those obtained in American studies (Bastian et al, 2002).
Depending on the specifications, a farmland associated with irrigation or draining capacity is
42% to 58% more expensive than other farmland. The official rental rate, fixed by local
authorities, has a positive impact on farmland prices, indicating an increase of prices due to a
better agronomic quality of land. Similarly, clay and silty clay soils are a little more
expensive.
In our sample, only 50% of sold farmland was leased farmland i.e. there was ongoing tenant
contract during the time of sale of farmland. As expected, nearly 90% of farmland was
purchased by tenant farmers who rented the land before the sale. A French farmland market
study from 1997 to 2010 (Lefebvre and Rouquette 2011) showed that leased farmlands were
sold 15% cheaper than free farmland. This effect was partly due to the French legal status of
agricultural tenancy which gives an automatic priority to the tenant to buy the land he farms
and thus reduces the competitive mechanisms. Our results confirm that the land sold to its
former tenant farmer is cheaper (11% cheaper if we consider only directs effects, and between
28 and 40% cheaper with the spillover).
Lefebvre and Rouquette (2011) also showed that the price of free farmland is growing faster
than the leased farmland. They assume that a growing part of non-farmer individuals buy free
farmland at higher prices than farmers. A descriptive analysis of our data shows that about
60% of buyers were farmers. Almost 30% of buyers were individuals who did not declare
themselves as farmers. Surprisingly, about 95% of them live in the same municipality (or less
than 5km far) of the plot of farmland they purchased. Moreover, results show that non-
farmers buy at lower price than farmers (about 10% cheaper if we consider only directs
effects, and between 35 and 40% cheaper with the spillover), on average. Most of these
individual non-farmers buy farmland for rental income. The cash rental income of farmland is
usually considered as risk free and safe investment. The lower willingness to pay of non-
farmers reflects that the cash rental is set below the marginal value of the farmland in farming
activities (Dupraz and Temesgen 2012). Non-farmers can also purchase farmland for their
own use and to develop various personal activities. Home or shared gardens are in full growth
in the region Bretagne, mainly in the suburban areas. Finally, some individuals also bought
farmland near the coast or in urban areas at higher price for speculation, expecting a
conversion into residential or industrial use. On average, a non-farmer buy a land located in
an urban area about 15% more expensive than a farmer (and between 40 and 65% more
expensive if we consider indirect effect). According to the results obtained by SDM model, a
non-farmer is willing to pay this type of land 69% more than a farmer.
Local authorities and SAFER also intervene in farmland market. They purchased farmland
generally located in a suburban area to store and lend to farmers who wish to produce for
local markets of direct sales or to settle in organic farming. The price of this farmland is
generally high because of their suburban location. They are forced to pay higher price to
encourage the landowners to sell rather than waiting for a change of destination of their land.
Farmer societies operated 10% of transactions of total farmland sales in 2010 against only 5%
in 1995. Besides the division of labor, administrative facilities and diversification, farmers
create societies mainly for tax advantages and transmission facilities. They can transfer land
in form of society’s shares outside the usual regulated land market. They probably have a
strong interest (upgrading, isolation ...) to acquire additional farmland and they are able to pay
more (payment facility ...).
Concerning impacts of agricultural policies, it can be assumed that farmland values will
increase with the associated production entitlements. Bartholomew and Boinon (2001) show
that the dairy quota rent is twice the price of milk. The impact of dairy cows’ density on
prices differs greatly depending on the econometric specification. According to the SDM
model, the price of farmland increases by 1600 € per hectare with one additional dairy cow
per hectare. Assuming an average milk yield of 6000 kg per cow, this corresponds to an
average value of 0.46€ per kg of milk quota17. As expected, the CAP payment entitlements
have also positive and significant effects on farmland prices.
17 This calculation is realized in assuming an average density of 0.59 cows per hectare, which occur in 50% of transactions.
Table 5. Coefficient estimates by SAR, SAC and SDM model – Average direct effects
Variables SAR model SAC model SDM Model
Agricultural Factors Q Q Q WQ Area of the plot -0.0006 -0.0008 -0.0008 0.005
Possibility of irrigation 0.151*** 0.162*** 0.159*** -0.379***
Vegetable share in UAA 0.329*** 0.4619*** -0.152 0.877***
Soil quality (reference: other soil quality)
Clay soils 0.024 0.034** 0.022 0.027
Silty clay soils 0.031 0.037** 0.0049 0.057
Sandy soils -0.009 -0.0002 -0.037 -0.053
Land rental rate 0.0026*** 0.0029*** 0.0022 0.0034
Farmland Policy factors
Farmland’s buyer (reference: farmer)
Land was leased by the purchaser -0.110*** -0.107*** -0.105 -0.094
Land acquired by the SAFER 0.054 0.064 0.076*** 0.045
Land acquired by local authorities 0.645*** 0.634*** 0.629*** 0.100
Land acquired by non farmer individuals -0.107*** -0.104*** -0.101*** -0.126
Land acquired by societies 0.097** 0.095** 0.099 -0.014
Non-Agricultural factors
Urban zone
Constant (reference: farmer buyer) -0.029* -0.016 -0.023 -0.017
Land acquired by non farmer 0.148*** 0.152*** 0.155*** -0.077
Land acquired by societies 0.104 -0.050 -0.047 0.122
Land acquired by the SAFER 0.158** 0.154** 0.135*** -0.181
Land acquired by local authorities -0.050 0.09 0.095 0.629
Coastline proximity
Constant (reference: farmer buyer) 0.056** 0.077** -0.006 -0.0003
Land acquired by non farmer -0.012 -0.020 -0.036*** 0.424***
Land acquired by societies 0.051 0.064 0.067 -0.187
Land acquired by the SAFER -0.014 -0.034 -0.013 1.545***
Land acquired by local authorities -1.037*** -1.01*** -1.03*** -0.767
Agricultural Policy factors
Density of dairy cows 0.069 0.094* 0.059 0.166
Single farm payments in Ille-et-Vilaine 0.0012*** 0.0012*** <0.0001 0.003
Single farm payments in Côtes d’Armor 0.0007*** 0.0005*** 0.0003 0.0016**
Environmental Policy factors
Total nitrogen load
Constant -0.00039 -0.00037* -0.0007 -0.0001
Squared total nitrogen load 0.000002*** 0.000002*** <0.0001 <0.0001
ZES 0.00035*** 0.00035*** 0.0009*** -0.0007***
Green algae areas (since 2010)
GAA 1 -0.0079 0.0033 0.067*** -0.191***
GAA 2 -0.128** -0.142*** -0.176*** 0.067
Contentious areas (since 2010)
CA 1 0.148*** 0.189*** 0.273*** -0.052***
CA 2 -0.030 -0.043 -0.107 0.028
***, **, * implies that coefficient estimates are statistically significant at respectively 10%, 5% and 1% level.
Table 6. Coefficient estimates by SAR, SAC and SDM model – Average indirect effects
Variables SAR model SAC model SDM Model
Agricultural Factors WY WY WQ et WY Area of the plot -0.0013 -0.0017 0.0088*
Possibility of irrigation 0.256*** 0.416** -0.542
Vegetable share in UAA 0.704*** 0.891*** 1.465***
Soil quality (reference: other soil quality)
Clay soils 0.053** 0.066* 0.065
Silty clay soils 0.057** 0.084* 0.105
Sandy soils -0.0006 -0.024 -0.130
Land rental rate 0.0044*** 0.0070*** 0.0078**
Farmland Policy factors
Farmland’s buyer (reference: farmer)
Land was leased by the purchaser -0.166*** -0.305*** -0.251**
Land acquired by the SAFER 0.102 0.149 0.088
Land acquired by local authorities 0.978*** 1.782*** 0.724
Land acquired by non farmer individuals -0.159*** -0.296*** -0.289
Land acquired by societies 0.149** 0.269** 0.054
Non-Agricultural factors
Urban zone
Constant (reference: farmer buyer) -0.026 -0.083 -0.049
Land acquired by non farmer 0.235*** 0.411*** -0.026
Land acquired by societies -0.078 -0.137 0.181
Land acquired by the SAFER 0.237** 0.438** -0.125
Land acquired by local authorities 0.157 0.290 1.179
Coastline proximity
Constant (reference: farmer buyer) 0.118** 0.154** -0.002
Land acquired by non farmer -0.034 -0.035 0.723**
Land acquired by societies 0.099 0.14 -0.271
Land acquired by the SAFER -0.047 -0.045 2.810**
Land acquired by local authorities -1.56*** -2.868*** -2.182
Agricultural Policy factors
Density of dairy cows 0.145* 0.186 0.340*
Single farm payments in Ille-et-Vilaine 0.0019*** 0.0033*** 0.0064***
Single farm payments in Côtes d’Armor 0.0011** <0.0001 0.0032**
Environmental Policy factors
Total nitrogen load
Constant -0.0005 -0.0011 -0.0008
Squared total nitrogen load <0.0001 -0.354*** <0.0001
ZES 0.00076*** 0.00035*** -0.0005
Green algae areas (since 2010)
GAA 1 0.0047 -0.022 -0.296
GAA 2 -0.222*** -0.354*** -0.018
Contentious areas (since 2010)
CA 1 0.287*** 0.407*** 0.134
CA 2 -0.065 -0.084 -0.042
***, **, * implies that coefficient estimates are statistically significant at respectively 10%, 5% and 1% level.
The results for the organic nitrogen load were also very dependent on the type of model used
for estimation. Le Goffe and Salanié (2005)18 found an increase in the price of 4.4€ per kg of
nitrogen of porcine origin. They interpreted this increase in farmland prices by the rising
demand from farmers who have to meet the manure spreading regulations. This also shows
the intensification of pig production in Bretagne. In the regulation, municipalities were
categorized as highly loaded over 50 kg per hectare of organic nitrogen associated to pig
production. In sales data, 95% of transactions belong to a municipality with more than 50 kg
of nitrogen per hectare. The average amount of animal nitrogen is about 115 kg per hectare in
the region Bretagne. Our results obtained by SAC model show that farmland price increase by
2.48 € per kg of additional nitrogen per hectare. The inclusion of a quadratic term in our mode
allowed us to differentiate the impact of nitrogen loading according to its level. We noted that
when the nitrogen pressure increases, the price of farmland increases less quickly. Similar
result was also observed by Le Goffe and Salanié (2005). They interpreted this as the result of
regulatory measures implemented in ZES that make farmland available for manure spreading
specially in small farms (treatment obligation imposed on surpluses of larger farms and
limiting trade by establishing ceilings for manure spreading).
In 2010, 115 farmland plots were exchanged within 8 water basins concerned by a huge algal
bloom, mainly on the north coast of the region Bretagne. First, eight binary variables are
included in the model, equaling 1 when the land belongs to one of these water basins and 0
otherwise. We find that their effects on prices are different depending on the water basins. We
then grouped water basins with a positive impact on land price in a binary variable (called
GAA1), and grouped water basins with a negative impact on land prices in another one
(called GAA2). So we kept only two binary variables in models to analyze impacts of these
environmental areas. On the studied period, 584 farmland plots were exchanged within areas
concerned by an environmental litigation. In the same way as above, we kept only two binary
variables. The variable CA1 contains areas in litigation with a potential positive impact on
prices, and CA2 areas in litigation with a potential negative impact on prices. Most of
farmland located in bay concerned by green algae suffers a decline in prices (between 40%
and 48% taken into account spillovers). A farmer knows he will have to meet additional
constraints or can anticipate future constraints. In addition, the purchased farmland may not
18 We consider that the amount of nitrogen is different depending on the animal (pig, hog, sow), which has not been taken into account in the calculations made by Le Goffe and Salanié (2005). The calculation is made on all animals and not only on porks.
be directly usable because it requires environmental improvements. This may explain that
farmland price is lower. Moreover, farmers already located in these areas, have not yet been
forced to change their agricultural practices, in contrast to areas in litigation, where
environmental constraints are already imposed to farmers. In the most areas concerned by
European litigation, we observe a high increase in the farmland prices (between 15% and 27%
from direct effects and between 40% and 56% from total effects). This may explain by the
need for farmers to buy additional plots to spread the manure and thus comply with
environmental constraints.
Conclusion
Results and tests proved the existence of spatial interaction on farmland market in Bretagne.
Although it is difficult to interpret the effects due to the spatial data, there is likely a
"spillover" effect between farmland sales. Sellers are likely influenced by transactions that
have occurred in their neighborhood. This could be intensified by lack or asymmetry of
information. Consequently they rely on sales information available near to the location of
farmland. It would be good to better understand differences observed in results obtained by
the three models. This might help to better define spatial effects and to better analyze their
impact on the results, especially on environmental factors.
Nevertheless, we observe significant increase of farmland prices in more constrained
environmentally areas such as ZES and areas affected by European litigation. This may
indicate some effectiveness of environmental policies or at least the awareness of farmers to
environmental problems. In 2012, the European Commission has decided to refer France to
the European Court of Justice for failure to comply with the Nitrates Directive in 1991 that
involves 39 water basins. All these water basins are located in Bretagne.
This notion of effectiveness of environmental policies arises especially as government tends
towards simplification and relaxation of regulations and environmental constraints. Since
2011, a decree has extended the total farmland area taken into account for calculating the
surface usable for manure spreading. According to a French environmental association Eau et
Rivière, this will increase the amount of nitrogen applied to the soil by 20%. In addition, the
government planned to remove the ZES and ZAC in 2013. These recent evolutions, and the
future ones, are intended to less constrain farmers, in encouraging them to modernize and
better control themselves their nitrogen load.
References
Agreste Primeur 2011. Recensement agricole 2010, premières tendances, n° 266, 4 pages.
Agreste, 2011. Statistique Agricole, Agreste Primeur. Ministry of Agriculture, Numero 265, August
2011.
Alston, J. M., 1986. An analysis of growth of U.S. farmland prices, 1963-1982. American Journal of
Agricultural Economics, 76:1-9.
Alston, J.M. and J.S. James 2002. The Incidence of Agricultural Policy, in B.L. Gardner and G.C.
Rausser (eds), Handbook of Agricultural Economics, Vol. 2B, Amsterdam: Elsevier, pp. 1689-1749.
Anselin L. and N. Lorenzo-Garcia, 2008, Errors in variables and spatial effects in hedonic house
price models of ambient air quality, paper presented at Shape seminar.
Barthélemy et Boinon, 2001, La gestion des quotas laitiers dans les pays membres de l’Union
Européenne : objectifs marchands versus objectifs non marchands, WP INRA sciences sociales.
Bastian C.T., McLeod D.M., Germino M.J., Reiners W.A. and B.J. Blasko, 2002,
Environmental amenities and agricultural land values: a hedonic model using geographic
information systems data, Ecological Economics, 40, pp 337-349.
Burt, O. 1986. Econometric Modelling of the Capitalization Formula for Farmland Prices.
American Journal of Agricultural Economics, 68:10-26.
Campbell, J.Y. and Schiller, R.J. 1987. Cointegration and Tests of Present Value Models.
Journal of Political Economy, 95:1062-1088.
Cavailhès J., Richard A. and Taverdet N. 1996. Des rentes classiques aux options de rente.
Une analyse de l’évolution du prix des terres, en France , Revue Économique, vol. 47, n° 4,
pp. 963-981.
Cavailhès J., Mesrine A. and C. Rouquette 2012. Le foncier agricole : une ressource sous
tensions. Economie and Statistiques (INSEE), no. 444-445
Ciaian, P., d’ A. Kancs and J.F.M. Swinnen 2010. EU Land Markets and the Common
Agricultural Policy, CEPS, Brussels.
Ciaian, P., d’Artis Kancs and Jerzy Michalek 2011. SPS capitalization in to land value
General Propensity Score Evidences from EU. LICOS centre for Institutions and Economic
Performance, Waaistraat 6-mailbox 3511 3000, Leuven, Belgium, LICOS Discussion Paper
Series, 293/2011.
Clark, J.S., Klein, K.K. & Thompson, S.J. 1993b. Are Subsidies Capitalized into Land
Values? Some Time Series Evidence form Saskatchewan. Canadian Journal of Agricultural
Economics, 41:155-168.
Cotteleer, Geerte, and Cornelis Gardebroek (2008). Market Power in a GIS-Based Hedonic
Price Model of Local Farmland Markets. Land Economics Volume. 84 (4): 573–592
Dachary J., Gashet F., Lyser S., Pouyanne G. and S. Virol, 2011, L’impact de la littoralisation
sur les marchés fonciers : une approche comparative des côtes basque et charentaise, Economie et
Statistiques, n°444-445.
Dupraz P. and C. Tesmegen, 2012, Farmland rental rate and marginal return to land: a French
FADN perspective, paper presented at the 86the annual conference of the agricultural economics
society, University of Warwick, United Kingdom.
Epple D., 1987, Hedonic prices and implicit markets : estimating demand and supply functions
for differentiated products, Journal of Political Economy, vol. 95, n°1, pp 59-80.
Facchini F., 1997, Politique agricole en France et prix de la terre, Politiques et Mangement
Public, vol. 15 n°4.
Featherstone, A.M. & Baker, T.G. 1987. An Examination of Farm Sector Real
Asset Dynamics. American Journal of Agricultural Economics, 69:532-546.
Feichtinger P. and K. Salhofer, 2011, The valuation of agricultural land and the influence of
government payments, WP Factor Markets.
Floyd, J.E. 1965. The effects of farm supports on the returns to land and labor in agriculture.
Journal of Political Economy, 73: 148-158.
Guyomard, H., Le Mouël, C. & Gohin, A. 2004. Impacts of alternative agricultural income
support schemes on multiple policy goals. European Review of Agricultural Economics,
31:125-148.
Hallahan C., Estimating spatial regression models with SAS/IML.
Jayet H., 2001, Econométrie et données spatiales, Une introduction à la pratique, Cahiers
d’Economie et Sociologie Rurales, n°58-59, pp 106-129.
Harding, J. P., J. R. Knight, and C. F. Sirmans 2003. Estimating Bargaining Effects in Hedonic
Models: Evidence from the Housing Market. Real Estate Economics 31, no. 4601- 622.
Lefebvre L and C. Rouquette, 2011, Les prix du foncier agricole sous la pression de
l’urbanisation, Economie et Statistiques, n°444-445.
Lefebvre L and C. Rouquette, 2011, Une dynamique différente selon le statut locatif, Economie
Agreste : la statistique agricole, n°265.
Le Gallo J., 2002, Econométrie spatiale : l’autocorrélation spatiale dans les modèles de régression
linéaire, Economie et Prévision, vol. 4, n°155, pp 139-157.
Le Gallo J., 2000a, Econométrie spatiale. 1. Autocorrélation spatiale, WP n°2000-05, Lille.
Le Gallo J., 2000b, Econométrie spatiale. 2. Hétérogénéité spatiale, WP n°2000-05, Lille.
Levesque R., Liorit D. and G. Pathier 2012. Les marchés fonciers ruraux régionaux entre
dynamiques des exploitations agricoles et logiques urbaines report for Economie and
Statistiques(INSEE), no. 444-445
Latruffe, L., Desjeux, Y., Guyomard, H., Le Mouël, C., Piet, L. 2008. Study on the Functioning of
Land Markets in the EU Member States under the Influence of Measures Applied under the Common
Agricultural Policy – Report of France for the Centre for European Policy Studies (CEPS), Brussels,
Belgium. 22 July. 32p.
Le Goffe P. and J. Salanié, 2005, Le droit d’épandage a-t-il un prix? Mesure sur le marché
foncier, Cahiers d’Economie et de Sociologie Rurale, n°77.
Lancaster K.J., 1996, A new approach to consumer theory, Journal of Political Economy, vol.74,
pp 132-157.
Pace R. and J. Lesage. 2009. Introduction to spatial econometrics. Chapman and Hall Book.
Palmquist, R. 1989. Land as a differentiated factor of production: A hedonic model and its
implications for welfare measurement, Land Economics, Vol. 65, pp. 23–28.
Palmquist, R. and Danielson, L. 1989. A hedonic study of the effects of erosion control and
drainage on farmland values, American Journal of Agricultural Economics, Vol. 71, pp.55–
62.
Patton M. and S. McEarlen, 2002, Spatial effects within the agricultural land market in northern
Ireland, Journal of Agricultural Economics, vol. 54, n°1, pp 35-54.
Pyykkönen P., 2005, Spatial analysis of factors affecting Finnish farmland prices, paper presented
at the 99th seminar of the EAAE, Copenhagen, Denmark.
Kossowski T. and J. Hauke, 2011, The method of computing the log-jacobian of the variable
transformation for spatial models – test and comments, WP.
Rosen S., 1974, Hedonic prices and implicit markets: product differentiation in pure competition,
Journal of Political Economy, vol. 82, n°1, pp 35-88.
Smirnov O. and L. Anselin, 2001, Fast maximum likelihood estimation of very large spatial
autoregressive models: a characteristic polynomial approach, Computational Statistics and Data
Analysis, 35, pp 301-319.
Swinnen, J., Ciaian, P. and Kancs, A. 2010. Study on the Functioning of Land Markets in the
EU Member States under the Influence of Measures Applied under the Common Agricultural
Policy. Report to the European Commission, Centre for European Policy Studies, Brussels,
2008.