Content uploaded by David J. Maddison
Author content
All content in this area was uploaded by David J. Maddison on Aug 21, 2014
Content may be subject to copyright.
A hedonic analysis of agricultural land prices
in England and Wales
David Maddison
University College London and University of East Anglia, UK
Received April 1999; ®nal version received August 2000
Summary
With the hedonic approach, farmland price dierentials are held to be indicative of
underlying productivity dierences. Data characterising over 400 separate transac-
tions in farmland in England and Wales in 1994 are analysed, and the marginal
value of particular farmland characteristics is computed. The analysis indicates that
climate, soil quality and elevation are all important characteristics, in addition to
the structural attributes of farmland. It is found that landowners are unable to
costlessly `repackage' their land and that regulated tenancies halve farm values.
Some doubt is cast on the accuracy of professional valuations performed by land
agents.
Keywords: hedonic prices, farmland values, England and Wales
JEL classi®cation: Q12, Q24
1. Introduction
The value of land derives from its use, and anything that aects the
productivity of land will be re¯ected in its purchase price. In principle, there-
fore, land prices contain information on the value that farmers place on
various characteristics of the land. Using the `hedonic' approach described
below, price dierentials associated with particular characteristics can be
used to measure the productive value of those characteristics.
The hedonic approach has already been employed by Miranowski and
Hammes (1984) to determine the implicit prices for soil characteristics in
Iowa. Brown and Barrows (1985) and Ervin and Mill (1985) have used the
hedonic technique to measure the impact of soil erosion on farmland produc-
tivity. Palmquist and Danielson (1989) have applied the technique to measure
the value of erosion control and drainage. More recently, Mendelsohn et al.
(1994), Dinar et al. (1998) and Evenson and Alves (1998) have used the hedonic
technique to measure the impact of climate on agricultural productivity in the
United States, India and Brazil, respectively.
This paper uses the hedonic technique to measure the productivity of farm-
land characteristics in England and Wales. Apart from the geographical
domain of its application, what distinguishes this analysis is the fact that it
deals with the price of individual plots of agricultural land over a large
European Review of Agricultural Economics Vol 27 (4) (2000) pp. 519±532
#Oxford University Press and Foundation for the European Review of Agricultural Economics 2000
geographical area. Previous applications have dealt with the price of indivi-
dual plots within a small geographical area and have not therefore been
able to investigate the productivity impacts of characteristics such as climate
that vary over large distances. Other applications have examined variation in
land prices over large geographical areas but have used county averages for
the returns to land along with county averages for climate, elevation and
soil quality.
The remainder of the paper is organised as follows. Section 2 summarises
the hedonic technique and explores the theoretical basis for placing restric-
tions upon the functional form of the hedonic price equation. In Section 3
the data on farm sales in England and Wales used to estimate the hedonic
price equation are discussed. In Section 4 the hedonic price equation is
estimated, and in Section 5 the values of various farmland characteristics
are computed and the results discussed. Section 6 concludes.
2. The hedonic technique
Although widely used in other ®elds, the hedonic technique may be unfamiliar
to many agricultural economists. Before proceeding to the empirical imple-
mentation of the hedonic technique it may therefore be useful to present a
brief summary of the technique.1
Dierent plots of agricultural land possess dierent characteristics. Some of
these characteristics, such as soil quality and climate, cannot be altered. These
are dubbed non-produced characteristics. Other features of the land (such as
the number of buildings) can be changed. These are termed structural attri-
butes. The price of land (P) is functionally dependent upon its characteristics
(q) both structural and non-produced. This relationship is represented by the
hedonic price equation Pq1;q2;...;qn. It is assumed that no individual is
able to in¯uence the position of the hedonic price equation. None the less, the
hedonic price schedule is determined by the interaction of many buyers and
sellers, and adjusts to eliminate excess supply and demand for each type of
land.
Palmquist (1989) shows how to derive the farmer's bid function for a plot of
land. This bid function will be a function of the characteristics of the land, the
price of inputs and outputs as well as a vector of farmer-speci®c skills. In
equilibrium, the partial derivative with respect to the characteristics of the
land of the farmer's bid function must equal the derivative of the hedonic
price equation. Otherwise, the farmer could increase pro®ts by using land
with more or less of particular characteristics. In addition, the zero-pro®t
constraint means that the farmer's total bid for a plot of land must equal
the rental price.
If all farmers had identical skills, then the functional form of the hedonic
price equation would correspond to the (common) bid function for each
characteristic. If, on the other hand, the productive skill of each farmer
1 For a more extensive theoretical discussion of the hedonic technique as applied to farmland, see
Palmquist (1989).
520 David Maddison
diers, then a farmer's marginal willingness to pay for particular characteris-
tics can be inferred from the locational choice of the farmer. In this case,
identi®cation of the bid function itself requires the analyst to observe how a
given farmer's locational choice changes in dierent markets. This is
seldom attempted as it would typically require information from dierent
time periods as well as characteristics of the farmers themselves (see Palmquist
(1989) for more details).
Estimating additional bene®ts arising from changes in the characteristics of
farmland is easy or not depending upon whether these changes aect few or
many sites. If changes aect only a few plots, the owners of these plots would
enjoy a capital gain whose magnitude is easy to calculate given the hedonic
price schedule. If a change in the characteristics of land simultaneously aects
many plots, then the hedonic price schedule may itself move. Even in this case,
it is still possible to approximate the ensuing change in economic welfare and
in some cases the resulting measures are exact (see Palmquist (1989) for more
details). All welfare measures depend, however, on the assumption of
unchanged input and output prices and are therefore partial equilibrium in
nature.
Underpinning the hedonic technique is the assumption that farmers are
perfectly informed about the characteristics of dierent tracts of land.
Given that the costs of acquiring information concerning particular tracts
of land are low, relative to the costs of purchasing land, this seems plausible.
It also assumes that there is a uni®ed market for land. This too is plausible
enough given that most agricultural crops are traded on national markets
and this would tend to integrate land markets. It should be noted that the
hedonic model is normally developed in terms of prices prevailing in the
rental market. In practice, however, there are typically better data on sales
rather than on the rental market. In this case, the approach must be modi®ed
to account for the ¯ow of future bene®ts associated with the uses and charac-
teristics of the land. Typically, this is achieved by annuitising the observed
price dierentials.
The question of what constitutes the appropriate functional form of the
hedonic land price schedule is typically viewed as being solely an empirical
question. Parsons (1990) shows, however, that there are in fact theoretical
reasons for preferring particular functional forms. Parsons argues that the
price of non-produced characteristics such as climate increases proportio-
nately with farm size, that the hedonic price function should be additively
separable in terms of the structural attributes of the land, and that the price
per acre should be independent of the quantity of land.
The logic of these conditions is obvious. For example, if the price per acre2
were found to depend on the number of acres or the price of non-produced
characteristics were not proportional to plot size then it would mean that
pro®ts could be earned from either dividing or combining plots of land.
Likewise, the price function should be additively separable in structural
2 1 hectare 2.47 acres (approximately).
A hedonic analysis of agricultural land prices 521
characteristics, as it is always possible to buy an undeveloped plot and build
on it at a cost that is (usually) independent of either the size of the plot or the
non-produced characteristics.
The assumptions underpinning these theoretical restrictions are nothing
more than the assumption of perfect competition in the market for land
and attributes as well as zero transaction costs (so that the number of sales
does not itself enter as an argument in the seller's pro®t function). If, however,
transaction and bargaining costs form a signi®cant cost, then land cannot be
costlessly repackaged, and plots of dierent sizes that are otherwise identical
may well be sold for dierent unit costs. It seems, therefore, preferable that
the values attached to particular characteristics should be inferred from a
model whose functional form is based solely on a goodness-of-®t criterion.
Certainly, this is the route most other researchers have followed.
3. The dataset
The empirical analysis draws on data on land transactions in England and
Wales taken from Farmland Market (1994). This journal contains a county-
by-county record of over 400 transactions in farmland during the ®rst 6
months of 1994.3The journal records the location, acreage and whether the
property sold has vacant possession, along with other important details
concerning dwellings and other buildings included in the sale. Some of this
farmland was sold by public auction although 63 per cent of the transactions
were by private treaty. Whenever a farm was sold by private treaty, the only
available indicator of its worth was the professional valuation, whereas when
a farm was sold by auction the actual sale price was available.
The Ordnance Survey Gazetteer of Great Britain (Ordnance Survey, 1992)
was used to determine the grid reference location of the individual properties
from the given address either to a named farm or to the nearest named settle-
ment. It was possible to include a set of land quality variables in the dataset by
indirectly utilising the 5-km grid square Agricultural Land Classi®cation of
England and Wales (Ministry of Agriculture, Fisheries and Food, 1988).
This classi®es land into one of seven grades according to the extent to
which its physical characteristics impose long-term limitations on agricultural
use. This grading system does not, it is claimed, necessarily re¯ect the current
economic value of the land, although it is almost invariably taken that grade 1
land is the best, as there are few limitations to its use. Grade 7 is either urban
land or, as is the case here, land that was not surveyed. The principal physical
factors included in the grading system are those relating to the site (gradient,
micro-relief and ¯ood risk) and soil (texture, structure, depth and stoniness).
Although the soil quality data have the form of a single variable taking the
integer values 1±7, the information is incorporated into our statistical
analysis by means of six dummy variables.
3 Agents voluntarily submit details of sales to the journal and it is generally considered that the land
prices in Farmland Market exceed those given in of®cial statistics.
522 David Maddison
Climate variables are calculated on a 10-km grid square basis.4A number of
summertime (April±September) and wintertime (October±March) 30-year
climate averages (1961±1990) are included. Also recorded are the population
density of the county in which the farm is located, the average elevation of the
farm and a dummy variable indicating whether the farm is situated in Wales
or England. The purpose of this dummy variable is to test for the geographi-
cal segmentation of the market for farmland. The information currently
contained in the dataset is described in Table 1, and a statistical summary
is given in the Appendix (Table A1).
Table 1. De®nition of variables in the dataset
Variable De®nition
PRICE/ACRE Price per acre (£)
ACRES Number of acres
PRIVATE Dummy variable; 1 if the property was sold by private treaty, 0
otherwise
BEDROOMS Number of bedrooms in dwellings where speci®ed
COTTAGES Dwellings for which the number of bedrooms is not speci®ed
MILK Milk quota oered with the property (1000 l)
POSSESS Dummy variable; 1 if the property has vacant possession, 0
otherwise
POPDEN Persons per square kilometre measured at the county level
SOILjDummy variable; 1 if land is ocially classi®ed as grade j,0
otherwise; j1;...;6
FD_SUMMER 30-year average number of frost days April± September
FD_WINTER 30-year average number of frost days October± March
TEMP_SUMMER 30-year average temperature April± September (8C)
TEMP_WINTER 30-year average temperature October± March (8C)
WIND_SUMMER 30-year average wind speed April±September (m/s)
WIND_WINTER 30-year average wind speed October±March (m/s)
PRECIP_SUMMER 30-year average precipitation April± September (mm)
PRECIP_WINTER 30-year average precipitation October± March (mm)
SUN_SUMMER 30-year average hours of sunshine April± September
SUN_WINTER 30-year average hours of sunshine October± March
REH_SUMMER 30-year average relative humidity April±September (percentage)
REH_WINTER 30-year average relative humidity October±March (percentage)
ALT Average elevation above sea level (m)
WALES Dummy variable; 1 for farms located in Wales, 0 otherwise
Source: see text.
4 These were provided by the Climate Research Unit of the University of East Anglia under the
auspices of the TIGER initiative.
A hedonic analysis of agricultural land prices 523
The relevance and expected impact on land values of some of the recorded
variables may need further explanation. First, many of the farms or land were
sold together with other assets of worth. Virtually all farmsteads have large
farmhouses attached to them and often labourers' cottages or holiday homes.
Although theydo not contribute tothe productivity of the land, these structures
add to the asset value of the farm and hence to its purchase price. Therefore, it is
necessary to control for them in the regression analysis. Usually the number of
bedrooms is recorded only for the main farmhouse. Hence one can distinguish
between dwellings of known and unknown size. By contrast, farm buildings are
not consistently recorded in the source and are consequently excluded from the
dataset; this represents an unfortunate closure on the information available.
Milk quotas permit a farmer to sell up to a given volume of milk without
paying the milk superlevy. Milk quota is thus of considerable value to dairy
farmers and is sometimes sold along with the farm.5Farm equipment and live-
stock are, by contrast, usually sold separately. Nowadays in England and Wales
only a very few farms are sold as tenanted property. Tenanted property is highly
regulated in England and Wales and it seems likely that the regulations govern-
ing tenanted property reduce the value ofsuch property in the eyes of an institu-
tional investor.6The dataset thus records whether a property has vacant
possession or is sold as a regulated tenancy.
4. Empirical analysis and results
A hedonic price equation was speci®ed taking price per acre as the dependent
variable to reduce possible problems associated with heteroscedasticity.
Initially, the set of regressors included the linear and quadratic values of all
the independent non-binary variables.7This allows the marginal eect of
changes in the level of the variables to dier. Furthermore, as the equation
combines both professional valuations and actual sale price data the equation
was augmented by a set of additional regressors multiplied by the dummy
variable PRIVATE indicating whether the professional valuation or the
actual sale price was used.8
5 Two observations included in the dataset merit comment. These are observations for which the
number of bedrooms is very large (43 bedrooms) and the price per acre is very high (£32,500 per
acre). The ®rst observation is of a farm with a very large house attached to it and several labourers'
cottages. The number of bedrooms is recorded for each of these labourers' cottages so they are
coded as extra bedrooms rather than mere cottages. The second is for a very small farm
(11 acres) but with a large farmhouse and a relatively large amount of milk quota attached to it
(300,000 l). This pushes up the price per acre. There is nothing else that is unusual about either
of these properties.
6 Most tenants and their successors have security of tenure for up to three generations.
7 It should be noted that the variables describing the structural attributes of land were divided by the
number of acres. This is because the dependent variable is in terms of price per acre and, in prin-
ciple, structural attributes should enter the hedonic price function additively (see Parsons, 1990).
Dividing the structural attributes by the number of acres does indeed substantially improve the
®t of the hedonic price equation.
8 In the dataset the number of farms sold by private treaty is 254 and the number of farms sold by
auction is 149. Two farms were broken up and sold partly by auction and partly by private treaty.
524 David Maddison
A linear functional form has an important defect in that the marginal eect
of characteristics is independent of the level of any other characteristic, imply-
ing, for example, that the value of climate is independent of soil quality. A
commonly chosen procedure, therefore, is to conduct a limited Box±Cox
search in which the linear model, the semilog model, the log±linear model
and the inverse semilog model are compared. Using the method described
by Maddala (1977), it was found that the semilog model was more likely to
have generated the observed data. It was, however, observed that the
RESET test for functional form was still not passed and after some experi-
mentation it was determined that a term in acres cubed needed to be added
to the regression equation.
For the purposes of presentation the equation was re-estimated dropping
the statistically insigni®cant higher-order terms and the statistically
insigni®cant interaction terms involving the dummy variable PRIVATE.9
The resulting equation is shown in Table 2. The model succeeds in explaining
65 per cent of the variation in the log of PRICE/ACRE. Tests for functional
form and heteroscedasticity are not signi®cant. The test for non-normality
is, however, statistically signi®cant, indicating that, whereas the OLS esti-
mator is still consistent, the reported t-statistics have only asymptotic
justi®cation.10
5. Discussion
Holding the characteristics of farmland at their sample means and varying
plot size results in the relationship between price per acre and farm size
described by Figure 1. On the face of it, this implies that there are potential
pro®ts to be made from repackaging parcels of land into smaller or larger
plots. An alternative explanation is that this is a re¯ection of signi®cant
transaction costs. Transaction costs might prevent a farmer from either
dividing his or her land or combining his or her land with land from someone
else to form a plot with a higher value per acre. Figure 1 suggests that these
transaction costs must be substantial. The most valuable plots of farmland
appear to be in excess of 1000 acres in size and this may be indicative of
the minimum ecient scale for agricultural production.
Originally, many hedonic studies were conducted using professional
valuations rather than sale price data.11 Here the combination of actual
market data and professional valuations makes it possible to test whether the
slopes and intercept of the model dier depending on which of the two price
measures is employed. Judging by the signi®cance of the variables PRIVATE
and PRIVATE BEDROOMS/ACRES, it seems that professional valuations
9 In the case of the variable ALT, the linear term was insigni®cant and so was dropped instead.
10 More speci®cally, the distribution of the OLS residuals is symmetric but has a kurtosis less than that
expected for the normal distribution.
11 The majority of hedonic studies use opinion-based data rather than actual sale price data. For a
recent hedonic land price study using only professional valuations coupled with an attempt to
assess their adequacy, see Roka and Palmquist (1997).
A hedonic analysis of agricultural land prices 525
Table 2. The hedonic price equation
Variable Coecient t-statistic
Intercept 8.862 2.90
ACRES ÿ0:169 10ÿ3ÿ4.73
ACRES20:254 10ÿ53.95
ACRES3ÿ0:807 10ÿ9ÿ3.16
PRIVATE ÿ0.152 ÿ2.89
PRIVATE BEDROOMS/ACRE 5.769 3.84
COTTAGES/ACRE 34.251 4.59
(COTTAGES/ACRE)2ÿ195.284 ÿ4.59
BEDROOMS/ACRE 12.152 8.23
(BEDROOMS/ACRE)2ÿ44.453 ÿ6.17
MILK/ACRE 0.109 5.21
POPDEN 0:301 10ÿ32.11
POSSESS 0.722 4.69
SOIL1 0.066 0.45
SOIL2 0.167 1.92
SOIL3 0.108 1.52
SOIL4 0:588 10ÿ20.07
SOIL5 ÿ0.179 ÿ1.36
SOIL6 0.156 1.69
FD_SUMMER ÿ0.046 ÿ1.69
FD_WINTER 0.020 2.36
TEMP_SUMMER ÿ0.494 ÿ2.30
TEMP_WINTER 0.194 1.20
PRECIP_SUMMER ÿ0:509 10ÿ3ÿ0.53
PRECIP_WINTER 0:636 10ÿ31.12
SUN_SUMMER 0:469 10ÿ30.25
SUN_WINTER ÿ0.170 ÿ0.44
REH_SUMMER ÿ0.105 ÿ2.23
REH_WINTER 0.043 1.03
WIND_SUMMER 0.499 1.51
WIND_WINTER ÿ0.334 ÿ1.88
ALT2ÿ0:686 10ÿ5ÿ3.17
WALES 0.018 0.13
Number of observations 405
Adjusted R-squared 0.620
Fisher test for zero slopes F32;372 21:62
Breusch Pagan test for heteroscedasticity 2
32 36:10
Ramsey's RESET(2) test for functional form t371 1:58
Jarque Bera test for normality 2
250:43
Dependent variable log (PRICE/ACRE)
Signi®cant at the 1 per cent level.
526 David Maddison
generally underestimated the value of farmland but overestimated the value
attached to additional living accommodation.
Turning to the characteristics of the farmland itself, as one would expect,
structural attributes are important determinants of the price per acre. Milk
production quota per acre, cottages per acre and the number of bedrooms
per acre are all highly signi®cant. The implicit valuations of an additional
bedroom, an additional cottage and additional quota for 1000 l of milk are
shown in Table 3. All the implicit prices in Table 3 re¯ect the valuations of
purchasers rather than professional valuation experts. These valuations,
moreover, refer to a farmer who has chosen to locate at a site corresponding
to the sample averages. These implicit prices are not annuitised.
Moving to the non-produced characteristics of farmland, the coecient on
the variable describing population density of the county in which the farm is
located is statistically signi®cant. This may provide support for the hypothesis
that distance to market is an important characteristic of farmland. Alterna-
tively, it may be that farmland is being bought on the assumption that
permission will be granted for the construction of houses and that population
density serves as a proxy for the potential pro®ts from realising such
ambitions. It might even be that population density captures the eect of
excluded amenities such as distance to the local school or the availability of
o-farm work for members of the farmer's family.
The coecient on the variable indicating vacant possession is both positive
and signi®cant, illustrating that regulations governing tenancies hold rents
beneath market values and reduce the value of the property to investors.
The same property sold with vacant possession fetches more than twice the
price of tenanted property.
The land quality grading system appears to play an uncertain role. Given
the way that the grades are commonly presented in a sales prospectus one
might have anticipated that the coecients on these variables would decline
Figure 1. The eect of plot size on price per acre.
A hedonic analysis of agricultural land prices 527
as land quality moves from grade 1 to grade 6. In fact, however, whereas the
value of land declines as one moves from grades 2 to 5, grade 1 farmland is not
the one valued most highly of all. The most valuable land appears to be grade
2 farmland, closely followed by grade 6. This highlights the fact that the grad-
ing system uses physical rather than economic criterion with which to classify
land (see above). Only two of the land quality variables are signi®cant at the
10 per cent level of signi®cance. However, what these tests check for is
whether the implicit value of land grades 1±6 is statistically dierent from
that of land grade 7. By contrast, the hypothesis that the implicit values of
Table 3. The implicit price of farmland characteristics (evaluated at sample means)
Characteristic Price
Bedrooms £24,175.96/bedroom
Cottages £82,154.90/cottage
Milk quotas £247.60/1000 l
Characteristic Price per acre
Population density £0.68/person/km2
Vacant possession (relative to tenanted) £1,181.54
Soil grade 1 (relative to soil grade 7) £142.24
Soil grade 2 (relative to soil grade 7) £375.44
Soil grade 3 (relative to soil grade 7) £236.95
Soil grade 4 (relative to soil grade 7) £12.23
Soil grade 5 (relative to soil grade 7) ÿ£341.79
Soil grade 6 (relative to soil grade 7) £351.79
Frost days (summer) ÿ£100.88/frost day
Frost days (winter) £46.40/frost day
Temperature (summer) ÿ£883.78/8C
Temperature (winter) £485.45/8C
Wind speed (summer) £1,466.98/m/s
Wind speed (winter) ÿ£642.62/m/s
Precipitation (summer) ÿ£1.15/mm
Precipitation (winter) £1.44/mm
Sunshine (summer) £1.06/h
Sunshine (winter) ÿ£3.84/h
Relative humidity (summer) ÿ£226.67/percentage point
Relative humidity (winter) £98.72/percentage point
Altitude ÿ£3.68/m
Wales £41.27
Source: see text.
It should be noted that in the ®rst half of 1994 £1 ECU 1.309 (Central Statistical Oce, 1995).
528 David Maddison
land graded 2 and land graded 5 are equal can be rejected at the 1 per cent
level of signi®cance.12
Three of the climate variables appear to have a statistically signi®cant
impact on farm prices at the 5 per cent signi®cance level (the number of
frost days in wintertime, summertime temperatures and relative humidity
during the summer). The fact that an increase in the number of frost days
during winter increases land values can be explained by the fact that a cold
snap during wintertime kills pests and vermin and is to the bene®t of agricul-
tural production.13 Similarly, high relative humidity may have a detrimental
impact on farmland prices insofar as it encourages diseases such as mildew.
Average wind speed during the winter months and the number of frost
days during the summer are statistically signi®cant at the 10 per cent level.
The average elevation of the farmland appears to exert a negative and
highly signi®cant eect on farmland prices. This is a ®nding that also emerges
from other hedonic studies. The explanation usually oered is that higher
elevations imply a greater diurnal variation in temperatures and that this is
detrimental to agricultural production. Whether a farm is located in Wales
or England does not appear to be important, although one could argue
that with only 12 Welsh farms this hardly constitutes a compelling test for
geographical segmentation of the market for farmland.
Finally, it is important to remember that the implicit prices in Table 3 re¯ect
only the willingness to pay of the `typical' farmer for marginal changes in the
level of climate variables. The value to society of marginal changes in the
level of these characteristics may be very dierent. This is because the price
of agricultural output (and therefore land values) is arti®cially in¯ated and
distorted by the Common Agricultural Policy. It is also possible that direct pay-
ments to particular farms become capitalised into land values. In principle,
these payments are like characteristics of the land and should be treated as
such by including them in the hedonic price equation (see, e.g. Barnard et al.,
1997). Unfortunately, the dataset used in this analysis does not include details
of direct payments to farms. The implication of this is that some of the charac-
teristics identi®ed by the analysis as being important may not be inherently
productive at all, except insofar as they tend to attract direct payments.14
6. Conclusions
This paper has employed the hedonic technique to calculate the value of
marginal changes in the characteristics of farmland in England and Wales.
12 Using the variance±covariance matrix of the parameter estimates and calculating the standard
error of the parameter SOIL2 ÿSOIL5. Dividing the parameter estimate by the standard error
results in a t-statistic of 2.63.
13 There is a lugubrious old saying in Eastern England: `A warm winter makes for a fat graveyard'.
14 An example of this is the support given to hill farmers. The UK has a long tradition of giving speci®c
aid to hill farmers, initially on a national basis, and, since 1975, under EU legislation. Eligible areas
cover all the main hill and upland areas in the north, west and south west of the UK, and account
for about half of the total agricultural land area (12 per cent in England). It is likely that these
payments conceal the true extent to which greater elevation inhibits agricultural activity.
A hedonic analysis of agricultural land prices 529
The results uphold the ®ndings of earlier analyses conducted in the United
States. As one would expect, structural attributes are very important
determinants of farmland prices. Population density is also important,
although it is not clear whether this is because distance to market matters
or is a result of other factors. Implicit values for soil quality and climate
are embedded in farmland prices and there is evidence that measures of
climatic extremes such as frost days can improve the ®t of the hedonic price
regression. Average elevation signi®cantly reduces farm values. The paper
also demonstrates that professional valuations provide a biased indicator of
the sale price because surveyors place too much value on the number of
bedrooms and not enough on the other characteristics. Finally, the analysis
indicates that tenanted farms are sold for signi®cantly less than similar
farmsteads with vacant possession and that farmland cannot be costlessly
repackaged in the sense that the size of a plot exerts a signi®cant in¯uence
on the price per acre. Future work might usefully examine the extent to
which direct payments to farmers are capitalised into land prices.
Acknowledgements
The Centre for Social and Economic Research on the Global Environment (CSERGE) is a
designated research centre of the UK Economic and Social Research Council (ESRC). The
author would like to acknowledge helpful comments made on an earlier version of this paper
by David Pearce, Kim Swales, Peter Pearson and four anonymous referees. Particular thanks
are due to Ana Oliveira for help with the collection of the data. The usual disclaimer applies.
References
Barnard, C., Whittaker, G., Westenbarger, D. and Ahearn, M. (1997). Evidence of
capitalisation of direct government payments into US cropland values. American
Journal of Agricultural Economics 79: 1642± 1650.
Brown, K. and Barrows, R. (1985). The impact of soil conservation investments on land
prices. American Journal of Agricultural Economics 67: 943±947.
Central Statistical Oce (1995). Economic Trends, 495. London: HMSO.
Dinar, A., Mendelsohn, R., Evenson, R., Parikh, J., Sanghi, A., Kumar, K., McKinsey, J.
and Lonergan, S. (1998). Measuring the Impact of Climate Change on Indian Agriculture.
World Bank Technical Paper No. 402. Washington, DC: The World Bank.
Ervin, D. and Mill, J. (1985). Agricultural land markets and soil erosion: policy relevance
and conceptual issues. American Journal of Agricultural Economics 67: 938± 942.
Evenson, R. and Alves, D. (1998). Technology, climate change, productivity and land use in
Brazilian agriculture. Planejemento e Politicas Publicas 18: 223±258.
Maddala, G. (1977). Econometrics. Singapore: McGraw±Hill.
Mendelsohn, R., Nordhaus, W. and Shaw, D. (1994). The impact of global warming on
agriculture: a ricardian analysis. American Economic Review 84: 753±771.
Ministry of Agriculture, Fisheries and Food (1988). Agricultural Land Classi®cation of
England and WalesÐ Revised Guidelines and Criteria for Grading the Quality of Agricul-
tural Land. London: HMSO.
Miranowski, J. and Hammes, B. (1984). Implicit prices of soil characteristics for farmland
in Iowa. American Journal of Agricultural Economics 66: 745±749.
530 David Maddison
Ordnance Survey (1992). The Ordnance Survey Gazetteer of Great Britain. London:
Macmillan.
Palmquist, R. (1989). Land as a dierentiated factor of production: a hedonic model and its
implications for welfare measurement. Land Economics 65: 23±28.
Palmquist, R. and Danielson, L. (1989). A hedonic study of the eects of erosion control and
drainage on farmland values. American Journal of Agricultural Economics 71: 55±62.
Parsons, G. (1990). Hedonic prices and public goods: an argument for weighting locational
attributes in hedonic regressions by lot size. Journal of Urban Economics 27: 308±321.
Roka, F. and Palmquist, R. (1997). Examining the use of national databases in a hedonic
analysis of regional farmland values. American Journal of Agricultural Economics 75:
1651± 1656.
A hedonic analysis of agricultural land prices 531
Appendix
Table A1. Statistical summary of variables used
Variable Mean SD Minimum Maximum
PRICE/ACRE 2,641.55 2,345.91 176.06 32,545.46
ACRES 130.07 207.36 7.00 2,090.00
PRIVATE 0.63 0.48 0.00 1.00
BEDROOMS 2.27 3.78 0.00 43.00
COTTAGES 0.25 1.34 0.00 16.00
MILK 15.79 84.41 0.00 738.44
POSSESS 0.98 0.13 0.00 1.00
POPDEN 261.719 167.34 24.00 1,033.00
SOIL1 0.02 0.15 0.00 1.00
SOIL2 0.13 0.34 0.00 1.00
SOIL3 0.45 0.50 0.00 1.00
SOIL4 0.18 0.39 0.00 1.00
SOIL5 0.03 0.17 0.00 1.00
SOIL6 0.10 0.30 0.00 1.00
FD_SUMMER 19.66 4.18 7.10 29.90
FD_WINTER 82.91 13.57 42.90 111.20
TEMP_SUMMER 12.71 0.68 10.07 13.90
TEMP_WINTER 5.71 0.73 3.17 8.08
WIND_SUMMER 4.40 0.28 3.92 5.40
WIND_WINTER 5.20 0.58 4.22 7.15
PRECIP_SUMMER 387.94 91.65 274.10 754.30
PRECIP_WINTER 495.88 177.53 262.70 1,143.80
SUN_SUMMER 1,031.62 76.14 861.90 1,224.00
SUN_WINTER 431.65 34.61 342.50 517.70
REH_SUMMER 82.32 2.09 77.52 87.30
REH_WINTER 89.63 1.56 85.22 94.07
ALT 117.91 75.38 31.00 359.00
WALES 0.03 0.17 0.00 1.00
Source: see text.
Number of observations was 405.
Corresponding author: David Maddison, Centre for Social and Economic Research on the
Global Environment, University College London, and University of East Anglia. E-mail:
d.maddison@ucl.ac.uk
532 David Maddison