Technical ReportPDF Available

Local Cost for Global Benefit: The Case of Wind Turbines

Authors:

Abstract and Figures

Given the rapid expansion of wind power capacities in Germany, this paper estimates the effects of wind turbines on house prices using real estate price data from Germany’s leading online broker. Employing a hedonic price model whose specification is informed by machine learning techniques, our methodological approach provides insights into the sources of heterogeneity in treatment effects. We estimate an average treatment effect (ATE) of up to -7.1% for houses within a one-kilometer radius of a wind turbine, an effect that fades to zero at a distance of 8 to 9 km. Old houses and those in rural areas are affected the most, while home prices in urban areas are hardly affected. These results highlight that substantial local externalities are associated with wind power plants.
Content may be subject to copyright.
Local Cost for Global Benefit:
The Case of Wind Turbines
RUHR
ECONOMIC PAPERS
Manuel Frondel
Gerhard Kussel
Stephan Sommer
Colin Vance
#791
Imprint
Ruhr Economic Papers
Published by
RWI – Leibniz-Institut für Wirtschaftsforschung
Hohenzollernstr. 1-3, 45128 Essen, Germany
Ruhr-Universität Bochum (RUB), Department of Economics
Universitätsstr. 150, 44801 Bochum, Germany
Technische Universität Dortmund, Department of Economic and Social Sciences
Vogelpothsweg 87, 44227 Dortmund, Germany
Universität Duisburg-Essen, Department of Economics
Universitätsstr. 12, 45117 Essen, Germany
Editors
Prof. Dr. Thomas K. Bauer
RUB, Department of Economics, Empirical Economics
Phone: +49 (0) 234/3 22 83 41, e-mail: thomas.bauer@rub.de
Prof. Dr. Wolfgang Leininger
Technische Universität Dortmund, Department of Economic and Social Sciences
Economics – Microeconomics
Phone: +49 (0) 231/7 55- 3297, e-mail: W.Leininger@tu-dortmund.de
Prof. Dr. Volker Clausen
University of Duisburg-Essen, Department of Economics
International Economics
Phone: +49 (0) 201/1 83- 3655, e-mail: vclausen@vwl.uni-due.de
Prof. Dr. Roland Döhrn, Prof. Dr. Manuel Frondel, Prof. Dr. Jochen Kluve
RWI, Phone: +49 (0) 201/81 49-213, e-mail: presse@rwi-essen.de
Editorial Oce
Sabine Weiler
RWI, Phone: +49 (0) 201/81 49-213, e-mail: sabine.weiler@rwi-essen.de
Ruhr Economic Papers #
Responsible Editor: Manuel Frondel
All rights reserved. Essen, Germany, 2019
ISSN 1864-4872 (online) – ISBN 978-3-86788-919-3
The working papers published in the series constitute work in progress circulated to stimulate
discussion and critical comments. Views expressed represent exclusively the authors’ own opinions
and do not necessarily reflect those of the editors.
Ruhr Economic Papers #791
Manuel Frondel, Gerhard Kussel, Stephan Sommer,
and Colin Vance
Local Cost for Global Benefit:
The Case of Wind Turbines
Bibliografische Informationen
der Deutschen Nationalbibliothek
The Deutsche Nationalbibliothek lists this publication in the Deutsche National bibliografie;
detailed bibliographic data are available on the Internet at http://dnb.dnb.de
RWI is funded by the Federal Government and the federal state of North Rhine-Westphalia.
http://dx.doi.org/10.4419/86788919
ISSN 1864-4872 (online)
ISBN 978-3-86788-919-3
Manuel Frondel, Gerhard Kussel, Stephan Sommer,
and Colin Vance1
Local Cost for Global Benefit: The Case of Wind
Turbines
Abstract
Given the rapid expansion of wind power capacities in Germany, this paper estimates the eects of wind
turbines on house prices using real estate price data from Germany’s leading online broker. Employing a
hedonic price model whose specification is informed by machine learning techniques, our methodological
approach provides insights into the sources of heterogeneity in treatment eects. We estimate an average
treatment eect (ATE) of up to -7.1% for houses within a one-kilometer radius of a wind turbine, an eect
that fades to zero at a distance of 8 to 9 km. Old houses and those in rural areas are aected the most, while
home prices in urban areas are hardly aected. These results highlight that substantial local externalities
are associated with wind power plants.
JEL Classification: IQ21, D12, R31
Keywords: Wind power; hedonic price model
January 2019
1 Manuel Frondel, RWI and RUB; Gerhard Kussel, RWI and RUB; Stephan Sommer, RWI; Colin Vance, RWI and Jacobs
University, Bremen. – This work has been supported by the Collaborative Research Center “Statistical Modeling of Nonlinear
Dynamic Processes” (SFB 823) of the German Research Foundation (DFG), within Project A3, “Dynamic Technology
Modeling”. We also gratefully acknowledge financial support by the German Federal Ministry of Education and Research
(BMBF) within Kopernikus Project ENavi (grant 3SFK4B0) and Project DIPOL (grant 01LA1809C). – All correspondence to:
Gerhard Kussel, RWI, Hohenzollernstraße 1/3, 45128 Essen, Germany, e-mail: gerhard.kussel@rwi-essen.de
1 Introduction
Germany is widely seen as a global leader in efforts to mitigate climate change, hav-
ing implemented an extensive feed-in-tariff scheme for renewable energy technologies
whose aim is to contribute to the reduction of greenhouse gas (GHG) emissions by
40% in 2020 relative to 1990. Wind power is among the most promising renewable en-
ergy technologies, as it has a high generation potential with comparatively low costs.
Between 2000, when feed-in-tariffs were introduced under Germany’s Renewable En-
ergy Act, and 2017, the number of onshore wind turbines roughly tripled, increasing
from 9,359 to 28,675. Over the same interval, electricity generation from wind power
increased from 9.5 to 106.6 billion kilowatthours, corresponding to a share of 18.8% of
Germany’s net electricity generation in 2017 (Data source: WindGuard).
Notwithstanding a broad-based popular acceptance of wind power, companies
planning new wind turbines frequently meet massive resistance of local communi-
ties owing to negative externalities. In addition to posing hazards for birds and bats,
the turbines make noise and affect the aesthetic appeal of the landscape by adding
movement in the form of rotation and shadow flickers, leaving a more industrialized
and less tranquil impression. Ultimately, such impacts may bear negatively on house
prices. Yet, while there is some international evidence on the effect of nearby wind
turbines on real estate prices, empirical evidence for Germany is scant.
Various methods can be availed for valuing external environmental costs, includ-
ing stated-preference surveys, as well as by investigating revealed preferences as ex-
pressed via real estate prices. We add to the latter strand of the literature by analyzing
the impact of wind turbines on the price of single-family houses. Drawing on a data
set of asking prices from more than 2.7 million houses in Germany posted between
2007 and 2015 on the site of Germany’s leading online broker, our approach employs
a hedonic pricing model whose specification is informed by the causal forest machine
leaning algorithm (?) to identify sources of heterogeneity.
1
We find an average treatment effect (ATE) of up to -7.1% for houses within a one-
kilometer radius of a wind turbine, an effect that fades to zero at a distance between
8 and 9 km. As suggested by the causal forest algorithm, additional specifications are
estimated that allow for differential effects of the wind turbines by the house’s location
and age. We find that very old houses and houses in rural areas suffer price reduc-
tions of up to 23%, probably due to stronger preferences for a pristine landscape, while
house prices in urban areas are not affected at all. Our results illustrate that while elec-
tricity generation via wind turbines may have global benefits, these are accompanied
by substantial local externalities.
The subsequent section provides a brief review of the literature on the effect of wind
turbines on real estate prices. Section 3 concisely summarizes our database, followed
by the description of our methodology in Section 4. We present and discuss our results
in Section 5. The last section closes with a summary and conclusions.
2 Findings from the Literature
Growing global energy demand and the increased awareness of anthropogenic cli-
mate change have led to an increase in wind power capacities worldwide. The rising
number of wind turbines, however, draws increasing attention to their negative exter-
nalities. Wind turbines not only endanger animals in their natural environment, no-
tably birds and bats (Arnett et al., 2008; Barclay et al., 2007; Smallwood, 2007), but also
make noise, create flicker effects, and negatively impact the scenery. Numerous stated-
preference surveys suggest that people have a positive attitude towards wind power
in general, but, at the same time, are concerned about external effects and environmen-
tal costs (Krekel and Zerrahn, 2017; Brennan and Van Rensburg, 2016; Meyerhoff et al.,
2010; Swofford and Slattery, 2010).
Such surveys, however, may be subject to measurement error, particularly when
respondents do not wish to state their true preferences. An alternative approach is to
2
make use of peoples’ revealed preferences, which are less prone to biases from strategic
responses. Real estate prices consist of peoples’ revealed willingness to pay for numer-
ous housing, locality and environmental characteristics, leading to a hedonic house
price model (Lancaster, 1966; Rosen, 1974). Adding the proximity to a wind turbine as
a feature, peoples’ valuation of corresponding externalities can be identified.
Following this basic idea, the effect of nearby wind turbines on housing prices has
been analyzed in diverse settings, yielding mixed results. Lang et al. (2014), for in-
stance, examine the impact of wind turbines on real estate prices in the U.S. state of
Rhode Island. Employing a modified difference-in-difference approach, these authors
find no effect of nearby wind turbines on real estate prices across model specifica-
tions. Similar results are obtained by Hoen et al. (2015), who model more than 50,000
real estate transactions from all over the U.S. by means of ordinary least squares and
difference-in-difference estimation. Analyzing data from densely populated commu-
nities in the U.S. state of Massachusetts, the results of Hoen and Atkinson-Palombo
(2016) also suggest that wind turbines have no effect on real estate prices. In contrast,
employing a repeat sales fixed-effects approach, Heintzelman and Tuttle (2012) find a
significantly negative effect in two of three analyzed municipalities in New York State.
While the empirical literature for the U.S. predominantly detects no effect of wind
turbines on real estate prices, the available studies for European regions point to sig-
nificantly negative impacts. Using a difference-in-difference methodology, Dröes and
Koster (2016), for example, analyze Dutch house price data and estimate a small nega-
tive effect of -1.4% for wind turbines within a 2 km distance. With a similar approach
for England and Wales, Gibbons (2015) finds a price reduction of up to 6% on houses
having a wind turbine within 2 km, fading to zero at a distance of 8 to 14 km. Employ-
ing various estimators that distinguish separate effects of noise and visual pollution,
Jensen et al. (2014) obtain an effect of similar size for Denmark. These authors attribute
a 3% price reduction to visual disamenities and 3 to 7% to noise pollution, which only
affects houses in immediate proximity to a turbine.
3
To the best of our knowledge, the only available evidence for Germany is provided
by Sunak and Madlener (2015) and Sunak and Madlener (2016), who analyze the effect
of wind turbines on real estate asking prices in a small semi-urban region. Using aug-
mented spatial econometric models, Sunak and Madlener (2016) find a strong effect of
-9 to -14% for the most affected houses. Providing the first comprehensive analysis for
Germany as a whole, we add to this strand of the literature by offering insights into
the sources of treatment effect heterogeneity.
3 Data
Our primary data source is drawn from ImmobilienScout24, Germany’s leading online
real estate platform. This data includes asking prices and building characteristics for
more than 7 million residential units posted between 2007 and 2015. The focus of our
analysis is on house sales, as the effect of amenities is presumably a less important
factor for rental units. For the same reason, we exclude multi-family houses, instead
solely focusing on single-family houses.
A potential drawback of the data is that the recorded prices are asking prices, rather
than transaction prices. This would be problematic if the difference between asking
and transaction prices is correlated with real estate or locality characteristics. Several
recent studies using the ImmobilienScout24 data argue, however, that this concern is
unfounded. These include an assessment of the effect of nuclear power plant shut
downs on surrounding real estate prices by Bauer et al. (2017) and the analysis of the
effect of national borders on house prices by Micheli et al. (2019). Frondel et al. (2019),
who investigate the effect of mandatory disclosure of energy information in sales ad-
vertisements on German house prices, likewise explore this issue. They compare data
on asking prices from ImmobilienScout24 with municipal data on transaction prices
from Germany’s capital Berlin, finding that (1) the difference between the two price
series is moderate, with transaction prices being about 7% lower than asking prices,
4
and that (2) this difference remains approximately constant over time.
While our sample consists of houses that were offered between 2007 and 2015, for
estimation purposes, we pruned the data along several dimensions, excluding houses
with (i) unusual prices below e20,000 or above e2,000,000, (ii) a reported living space
of either less than 40 m2or more than 800 m2, (iii) either less than 1 or more than 20
rooms and, (iv) a lot size smaller than 20 m2or larger than 5,000 m2. As a result, our
final data set comprises 2,855,466 observations.
The summary statistics reported in Table 1 indicate that the average asking price of
the sample properties is about e274,000, the mean size is 154 m2, and the mean number
of rooms is 5.4. With about 55%, detached houses represent the majority of the sample
properties, with another 17% of the properties being semidetached. With respect to the
temporal dimension, the offers are almost equally split across the period 2007-2015.
Table 1: Descriptive Statistics of Real Estate Offers
Mean Standard Deviation Minimum Maximum
Asking price in e273,786.4 203,136.9 20,000 2,000,000
Year of construction 1979.5 36.9 1300 2016
Living space in m2153.7 60.4 40 800
Lot size in m2676.4 536.3 20 5,000
Number of rooms 5.4 1.8 1 20
Detached house 0.55 - 0 1
Semidetached house 0.17 - 0 1
Other house type 0.08 - 0 1
Terrace house 0.04 - 0 1
Mid-terrace house 0.06 - 0 1
End-terrace house 0.04 - 0 1
Bungalow 0.03 - 0 1
Villa 0.03 - 0 1
Offer year 2007 0.08 - 0 1
Offer year 2008 0.15 - 0 1
Offer year 2009 0.13 - 0 1
Offer year 2010 0.12 - 0 1
Offer year 2011 0.11 - 0 1
Offer year 2012 0.09 - 0 1
Offer year 2013 0.11 - 0 1
Offer year 2014 0.11 - 0 1
Offer year 2015 0.10 - 0 1
Number of Observations: 2,855,466
Separated descriptives for the treatment and control group are reported in Table A.1 in the appendix
In addition to the information on real estate characteristics, the data contains the
5
exact coordinates of each house. This feature allows us to merge it with other georef-
erenced data sources, such as the database RWI-GEO-GRID (Breidenbach and Eilers,
2018), which provides high-resolution socio-demographic data on the scale of a 1x1
km grid. We make use of information on purchasing power per capita, population
density, the unemployment rate, the share of foreigners, the number of buildings and
demographic structure of the grid.1
To complete the locality characteristics, we add the distance to the center of the
next city with more than 100,000 inhabitants and dummy variables for all German mu-
nicipalities. Table 2 demonstrates substantial heterogeneity in the socio-demographic
characteristics of the neighborhood. For instance, the purchasing power per capita
ranges between e5,900 and e139,000. Moreover, we observe a large diversity in the
population density, spanning as low as 1 inhabitant per km2in very rural areas to al-
most 27,000 inhabitants in highly urbanized areas.
Table 2: Descriptive Statistics of Locality Characteristics in 1x1 km Grids
Mean Standard Deviation Minimum Maximum
Purchasing power per capita (in e) 21,464.8 4,104.9 5,916.7 139,391.0
Total inhabitants per km21,837.5 1,707.4 1.0 26,947.0
Unemployment rate (in %) 5.97 3.94 0.01 39.98
Foreigners (in %) 6.66 5.74 0.01 100.00
Number of buildings 434.8 306.2 1.0 2830.0
Share of inhabitants aged 0-20 19.52 2.66 0.18 38.67
Share of inhabitants aged 20-35 16.20 2.95 0.40 47.50
Share of inhabitants aged 35-45 14.44 2.16 0.35 42.22
Share of inhabitants aged 45-55 16.46 1.87 0.54 40.80
Share of inhabitants aged 55-65 12.70 1.86 0.27 35.44
Share of inhabitants aged 65+ 20.68 1.86 3.17 97.96
Distance to city center (in km) 24.52 20.27 0.03 146.18
Distance to next wind turbine (in km) 8.43 6.28 0.02 54.83
Number of Observations: 2,855,466
Separated descriptives for the treatment and control group are reported in Table A.1 in the appendix
Finally, we obtained geo-referenced data on wind turbines in Germany from the Re-
1The data is gathered by the commercial data provider Micromarketing-Systeme und Consult GmbH
(microm) and is aggregated from more than one billion individual data points from various sources. Raw
data are collected from companies acting in data intensive environments such as Creditreform and CEG
Consumer Reporting, as well as from official institutions such as the Federal Office for Motor Traffic, the
Statistical Offices of the Federation and the Federal States, and the Federal Employment Agency. Since
RWI-GEO-GRID is only available for the years 2005 and 2009-2015, we interpolate the information for
the years 2006-2008.
6
newable Energy Installations Core Data of the Federal Network Agency (BNetzA) and
several regional authorities. The central register provided by BNetzA was introduced
in August 2014. Hence, all information on the wind turbines that were installed after
this date is retrieved from BnetzA, while all prior information is collected from federal
state authorities. Both data sets are compatible and commonly encompass the con-
struction year and the exact position of all wind turbines in Germany, but we dropped
2,373 observations due to missing information on the construction year.
At the beginning of our study period, 13,574 wind turbines were installed in 2007,
mostly in the northeast owing to better wind conditions(Figure 1). This is the most pro-
pitious area for wind turbines, as average wind speeds are significantly higher com-
pared to other regions in Germany. By the end of 2015, after stronger incentives in
the form of higher feed-in tariffs for electricity produced from wind power were intro-
duced, 7,883 wind turbines were additionally installed, also in less windy areas, such
as the southeast of Germany.
While the mean distance of sample houses to the next wind turbine is about 8.4
km (Table 2) and the median amounts to 6.6 km, Figure 2 illustrates a great deal of
heterogeneity: 8.9% of the properties have a wind turbine within a 2 km distance,
whereas 0.6% are located more than 30 km away from a wind turbine.
7
Figure 1: Position of Wind Turbines
8
Figure 2: Distance to Wind Turbines
4 Methodology
To identify the impact of wind turbines on the prices of nearby houses, we estimate the
following hedonic price model by Ordinary Least Squares (OLS):
log(pi) = distanceT
iα+xT
iβ+mg+τ
t+εi, (1)
where log(pi)is the natural logarithm of the asking price of house iand distance is a
set of distance bands indicating whether the house is within the radius of 1, 2,...9 km
distance to a wind turbine. xcomprises house and locality characteristics, αand β
are corresponding coefficient vectors, mgand τ
tare fixed-effects for municipality gand
time t, and εiis an error term that is independent and identically distributed. Our main
focus is on coefficient vector α, which measures the average treatment effect for houses
within 1, 2,...9 km distance to a wind turbine.
As we observe the asking price for a property either in presence (Yi1) or in absence
(Yi0) of a wind turbine, but not in both states, we face the well-known evaluation prob-
9
lem (Holland 1986). Following the idea of Rubin’s (1974) potential-outcome model:
Yi=
Yi0i f Wi=0
Yi1i f Wi=1,
(2)
where Wis a binary indicator that equals unity when house iis in the range of a wind
turbine and zero otherwise, the average treatment effect (ATE) is given by ATE :=
E(Y(1)|W=1)E(Y(0|W=0)). Accordingly, if the assignment of the treatment W
were to be randomized, a situation that is implausible in observational studies, the
causal effect of the treatment can easily be estimated by a simple comparison of mean
outcomes. Yet, in our empirical example, it seems likely that wind turbines are more
frequently placed in less wealthy neighborhoods, as land prices are lower and residents
have less resources to oppose construction. At the same time, house prices in those
areas are probably lower as well.
To identify the causal effect of the treatment, we need to assume unconfounded-
ness, i.e. that all determinants affecting the probability of the treatment (having a wind
turbine nearby) and the outcome (house price) are captured by our covariates (X):
Wi(Yi0,Yi1)|X. (3)
While this assumption is critical, it is not testable. Nevertheless, below we provide a
supplementary analysis to increase the credibility of our estimates that is a based on a
placebo-regression approach.
A final estimation issue concerns the possible existence of interaction terms that
capture differential magnitudes in the effects of wind turbines. In this regard, it is con-
ceivable that the effect of proximity to a wind turbine is dependent on other features
of the house and of the surrounding landscape. It stands to reason, for example, that
houses located in densely settled areas would be affected differently by wind turbines
than those surrounded by pristine landscapes.
10
While theory can provide some guidance in identifying such sources of heterogene-
ity, the attempt to specify a complete set of interactions risks embarking on an iterative
search for results that are, even if statistically significant, purely spurious (Assmann
et al., 2000; Cook et al., 2004). Building on work by Athey and Imbens (2016), Wa-
ger and Athey (2018) develop a nonparametric machine learning algorithm to address
this challenge. In essence, their approach draws on asymptotic normality theory to
enable statistical inference using a forest-based method to generate predictions that
are asymptotically unbiased. The method, which we implement using an R package
provided by the authors, is akin to an adaptive nearest neighbor method, producing
estimates of the conditional average treatment effect. We employ the method as an
exploratory tool, using it to identify sources of heterogeneity in the estimation of treat-
ment effects that we incorporate in the specification of Equation 1.
5 Empirical Results
Table 3 presents OLS estimates from a specification of hedonic price model 1 that
excludes treatment heterogeneity, while Figure 3 allows visualization of the corre-
sponding estimates of each distance band and its confidence interval. The figure il-
lustrates that the average treatment effects are statistically and economically signif-
icant for houses that are within a distance of up to 8 km to a wind turbine. Un-
surprisingly, the strongest effect is found for houses in the smallest radius of a one-
kilometer distance, where the presence of turbines reduces house prices by 7.1% (=
100[ex p(0.0735)1]). In addition to impairing the scenery, wind turbines in such
close proximity create audible noise and flicker effects. Although the treatment effects
abate with distance, they remain statistically significant up to a radius between 7 to 8
kilometers, where noise should be irrelevant (Gibbons, 2015).
The coefficients on the remaining covariates are all statistically significant and ex-
hibit the expected signs, albeit the effect sizes are small in many cases. Given the
11
Table 3: OLS Regression Results of Equation 1
Coefficients Standard Errors
Wind turbine within
1 km distance -0.0735∗∗ (0.00763)
1 to 2 km distance -0.0615∗∗ (0.00424)
2 to 3 km distance -0.0560∗∗ (0.00399)
3 to 4 km distance -0.0441∗∗ (0.00381)
4 to 5 km distance -0.0416∗∗ (0.00384)
5 to 6 km distance -0.0294∗∗ (0.00394)
6 to 7 km distance -0.0253∗∗ (0.00393)
7 to 8 km distance -0.0139∗∗ (0.00413)
8 to 9 km distance -0.000786 (0.00427)
Housing characteristics:
Year of construction 0.00453∗∗ (0.0000479)
Living space (in m2) 0.00410∗∗ (0.0000298)
Lot size (in 100 m2) 0.0122∗∗ (0.00221)
Number of rooms -0.0104(0.000801)
Detached house 0.0260∗∗ (0.000993)
Semidetached house -0.0601∗∗ (0.000576)
Terrace house -0.153∗∗ (0.00114)
Mid-terrace house -0.149∗∗ (0.000895)
End-terrace house -0.0838∗∗ (0.00108)
Bungalow 0.0383∗∗ (0.00114)
Villa 0.214∗∗ (0.00122)
Locality characteristics:
Purchasing power per capita (in 1,000 e) 0.0382∗∗ (0.00106)
Total inhabitants (in 1,000) 0.031(0.00173)
Unemployment rate (in %) -0.00452∗∗ (0.0000849)
Foreigners (in %) 0.00407∗∗ (0.000595)
Number of buildings -0.00002(0.00000810)
Share of inhabitants aged 0-20 -0.0135∗∗ (0.000905)
Share of inhabitants aged 20-35 0.00619∗∗ (0.000594)
Share of inhabitants aged 35-45 0.00507∗∗ (0.000981)
Share of inhabitants aged 45-55 -0.0148∗∗ (0.000745)
Share of inhabitants aged 55-65 -0.0102∗∗ (0.000824)
Distance to city center (in km) -0.00420∗∗ (0.000208)
Year dummies Yes
Municipality dummies Yes
Number of Observations: 2,855,466
R20.711
Note: ∗∗ and indicate statistical significance at the 1% and 5% level,respectively;
standard errors are clustered at the GEO-Grid level.
log-linear specification of the model, most of these estimates can be interpreted as the
percentage change in the house price given a unit change in the explanatory variable.
We see, for example, that each square meter increase in living space increases the house
asking price by 0.4%, while each additional kilometer from the nearest city center de-
creases the price by about the same amount.
12
Figure 3: Effects of Wind Turbines on logged House Prices
Note: Standard errors are clustered at the GEO-Grid level.
5.1 Heterogeneity in Treatment Effects
In principle, any of the control variables could be a source of heterogeneity in the treat-
ment effects. To identify such sources, we plotted the results obtained from the causal
forest algorithm applied to the covariates included in Equation 1. For this purpose, we
collapsed the nine treatment dummies into a single dummy equaling unity if a wind
turbine is within a distance of 8 kilometers of the home and zero otherwise. For the
overwhelming majority of covariates, we find no significant mediating effect on the
treatment dummy. Two exceptions are the distance to the next city center and the year
of construction, both of which exacerbate the effect of proximity to a wind turbine.
Specifically, as seen from Figures A4 and A5 in the appendix, there are rapid increases
in the magnitude of the treatment effect for a distance to the next city center of more
than 10 kilometers and construction years before 1950.
Based on these results, we construct an urban indicator that equals unity if the
13
house offer is within a 10 km2radius around the next city center and zero otherwise,
as well as an age indicator for houses that were built before 1950, and interact both indi-
cators with the treatment dummies in Equation 1. Figure 4 illustrates the differences in
the treatment effects between urban and rural areas. While there are substantial treat-
ment effects in rural areas, the effect on prices of houses close to urban environments
is considerably weaker and statistically insignificant at any conventional level.
Figure 4: Effects of Wind Turbines on logged House Prices in Rural and Urban Areas
Note: Standard errors are clustered at the GEO-Grid level. Coefficient estimates are reported in Table
A.3 in the appendix.
Contrasting with Dröes and Koster (2016), who find a stronger effect in urban en-
vironments, our finding seems intuitive given two potential explanations: First, to the
extent that the urban landscape is already developed, the sight of a wind turbine might
not change the overall impression of the landscape. Second, a more urban environment
has a higher density of buildings that conceal the view of the wind turbine (Sunak and
Madlener, 2015). This second explanation, however, does not apply to our data, as
we control for the density of buildings. Hence, we conclude that the nativeness of the
2According to this definition 24,99% of the observations are located in urban areas. Results of robust-
ness checks with alternative definitions (5 km, 9,19%; 20 km, 53,13%; and 50 km, 88,40%) are reported in
the appendix (Figures A1 - A3).
14
landscape and the corresponding preferences of the residents seem to determine the
effect size.
With respect to the age of buildings, Figure 5 shows a remarkable effect of up to -
23% on the prices of houses built before 1950, whereas newer buildings are affected to a
much lower extent. This effect may also be explained by preferences for a preindustrial
impression of the building and the surrounding landscape.
Figure 5: Effects of a Wind Turbine on logged House Prices of Old Buildings and
Newer Houses
Note: Standard errors are clustered at the GEO-Grid level. Coefficient estimates are reported in Table
A.4 in the appendix.
5.2 Unconfoundedness
As discussed in Section 4, our results can only be interpreted as causal if the uncon-
foundedness assumption holds, i.e. all determinants affecting the probability of the
treatment – a nearby wind turbine – and the outcome – the house price – are cap-
tured by our covariates. To probe this assumption, we begin by estimating Equation
1 using three sets of control variables: First, Equation 1 is estimated without any local
controls using only house characteristics and time fixed effects; second, we addition-
15
ally include community fixed effects and, third, Equation 1 is estimated using all con-
trol variables. Figure 6 illustrates that the treatment effects shrink significantly when
county fixed effects and the detailed RWI-GEO-GRID information are added. (Coeffi-
cient estimates are reported in Table A.2 in the appendix.) Apparently, when controls
for locality characteristics are excluded in the estimation, the effects of wind turbines
are overestimated, reflecting the fact that windmills tend to be installed in low-price
regions.
Figure 6: Effects of Wind Turbines on logged House Prices
0 to 1 km
1 to 2 km
2 to 3 km
3 to 4 km
4 to 5 km
5 to 6 km
6 to 7 km
7 to 8 km
8 to 9 km
-0.4 -0.3 -0.2 -0.1 0
No regional controls County fixed effects only
Full controls
Note: Standard errors are clustered at the GEO-Grid level.
To provide further evidence that the unconfoundedness assumption holds, we ran
placebo regressions. In a first step, we drop from the estimation sample all “treated”
houses, that is, those which had a wind turbine within 9 km when they were offered,
instead focusing on houses where there was no turbine when they were offered, but
where a turbine was constructed in the following years. We then estimate Equation 1
replacing the treatment dummies distance with placebo treatment dummies indicating
the future presence of a wind turbine in a distance up to 9 km. To exclude anticipation
effects, in addition to dropping actually treated houses, we also drop observations
16
Figure 7: Placebo Regression Results: Effects of Future Wind Turbines on logged House
Prices
0 to 1 km
1 to 2 km
2 to 3 km
3 to 4 km
4 to 5 km
5 to 6 km
6 to 7 km
7 to 8 km
8 to 9 km
-0.2 -0.15 -0.1 -0.05 0 0.05
No regional controls County fixed effects only
Full controls
Note: Standard errors are clustered at the GEO-Grid level.
where a wind turbine was constructed within the two years following a house sale
offer. Hence, there should be no treatment effect and the estimated coefficients can be
interpreted as the selection effect of wind turbines in specific localities.
The resulting treatment effects are presented in Figure 7, while the OLS coefficient
estimates are reported in Table A.5 in the appendix. The negative and significant coeffi-
cients on the treatment dummies in the "no local control setting" support the presump-
tion that the placement of wind turbines is negatively correlated with surrounding
house prices. However, the effects do not follow the decreasing pattern observed in
our baseline estimation.
This finding still persists after adding municipality fixed effects, although the mag-
nitude of many of the coefficients is lower. But, after including controls for our de-
tailed small scale neighborhood characteristics from RWI-GEO-GRID, all coefficient
estimates are not statistically different from zero. Also, there is no discernible pattern
to the coefficient estimates, as they straddle both sides of zero. Hence, we are confident
17
that we have captured all the factors influencing the placement of wind turbines that
are associated with house prices via our detailed small-scale neighborhood data.
6 Summary and Conclusions
Wind power is among the most promising renewable energy technologies, as its high
electricity generation potential is accompanied by relatively low generation cost. Yet,
there is also increasing international evidence that wind turbines cause persistent neg-
ative externalities: In addition to posing hazards for birds and bats, turbines make
noise and affect the aesthetic appeal of the landscape. Ultimately, these impacts may
bear negatively on house prices. Despite the rapid expansion of wind power capacities
in recent decades, though, empirical evidence on the effect of nearby wind turbines on
real estate prices is scant for Germany.
Using asking prices from Germany’s leading online broker and a hedonic pricing
model coupled with a machine leaning algorithm, we fill this gap by analyzing the
effect of wind turbines on prices of surrounding single-family houses. Accounting for
detailed property and locality characteristics, we estimate an average treatment effect
of up to 7.1% for houses within 1 km distance to the next wind turbine, an effect that
fades out at a distance between 8 and 9 km.
Identifying the most important interaction terms by a machine leaning algorithm,
we add to the literature by estimating heterogeneous treatment effects: While the
prices of houses close to urban environments are not affected by nearby windmills,
houses in rural areas suffer from remarkable devaluation. This effect is even more pro-
nounced for old buildings built prior to 1949, whose asking prices decrease by up to
23%.
Our findings can be explained by differences in the appearance of the landscape
and preferences of the local population. While the urban population is accustomed to
living in an industrialized and dynamic environment, inhabitants of rural areas may
18
lose the impression of pristine nature and tranquility when noise, rotation, and shadow
flickers appear. Altogether, our results illustrate that while electricity generation via
wind turbines may have global benefits, these are accompanied by substantial local
externalities and environmental costs, primarily borne by rural communities close to
wind turbines.
19
References
Arnett, E. B., Brown, W. K., Erickson, W. P., Fiedler, J. K., Hamilton, B. L., Henry, T. H.,
Jain, A., Johnson, G. D., Kerns, J., Koford, R. R., et al., 2008. Patterns of Bat Fatalities
at Wind Energy Facilities in North America. Journal of Wildlife Management 72 (1),
61–78.
Assmann, S. F., Pocock, S. J., Enos, L. E., Kasten, L. E., 2000. Subgroup Analysis and
Other (Mis) Uses of Baseline Data in Clinical Trials. The Lancet 355 (9209), 1064–1069.
Athey, S., Imbens, G., 2016. Recursive Partitioning For Heterogeneous Causal Effects.
Proceedings of the National Academy of Sciences 113 (27), 7353–7360.
Barclay, R. M., Baerwald, E., Gruver, J., 2007. Variation in Bat and Bird Fatalities at Wind
Energy Facilities: Assessing the Effects of Rotor Size and Tower Height. Canadian
Journal of Zoology 85 (3), 381–387.
Bauer, T. K., Braun, S. T., Kvasnicka, M., 2017. Nuclear Power Plant Closures and Lo-
cal Housing Values: Evidence from Fukushima and the German Housing Market.
Journal of Urban Economics 99, 94–106.
Breidenbach, P., Eilers, L., 2018. RWI-GEO-GRID: Socio-economic Data on Grid Level.
Journal of Economics and Statistics 238 (6), 609–616.
Brennan, N., Van Rensburg, T. M., 2016. Wind Farm Externalities and Public Prefer-
ences for Community Consultation in Ireland: A Discrete Choice Experiments Ap-
proach. Energy Policy 94, 355–365.
Cook, D. I., Gebski, V. J., Keech, A. C., 2004. Subgroup Analysis in Clinical Trials. Med-
ical Journal of Australia 180 (6), 289–292.
Dröes, M. I., Koster, H. R., 2016. Renewable Energy and Negative Externalities: The
Effect of Wind Turbines on House Prices. Journal of Urban Economics 96, 121–141.
20
Frondel, M., Gerster, A., Vance, C., 2019. The Power of Mandatory Quality Disclosure:
Evidence from the German Housing Market. Journal of the Association of Environmen-
tal and Ressource Economists, revised.
Gibbons, S., 2015. Gone With the Wind: Valuing the Visual Impacts of Wind Turbines
Through House Prices. Journal of Environmental Economics and Management 72, 177–
196.
Heintzelman, M. D., Tuttle, C. M., 2012. Values in the Wind: A Hedonic Analysis of
Wind Power Facilities. Land Economics 88 (3), 571–588.
Hoen, B., Atkinson-Palombo, C., 2016. Wind Turbines, Amenities and Disamenitites:
Astudy of Home Value Impacts in Densely Populated Massachusetts. Journal of Real
Estate Research 38 (4), 473–504.
Hoen, B., Brown, J. P., Jackson, T., Thayer, M. A., Wiser, R., Cappers, P., 2015. Spatial
Hedonic Analysis of the Effects of US Wind Energy Facilities on Surrounding Prop-
erty Values. Journal of Real Estate Finance and Economics 51 (1), 22–51.
Jensen, C. U., Panduro, T. E., Lundhede, T. H., 2014. The Vindication of Don Quixote:
The Impact of Noise and Visual Pollution from Wind Turbines. Land Economics 90 (4),
668–682.
Krekel, C., Zerrahn, A., 2017. Does the Presence of Wind Turbines Have Negative Ex-
ternalities for People in their Surroundings? Evidence From Well-Being Data. Journal
of Environmental Economics and Management 82, 221–238.
Lancaster, K. J., 1966. A New approach to Consumer Theory. Journal of Political Economy
74 (2), 132–157.
Lang, C., Opaluch, J. J., Sfinarolakis, G., 2014. The Windy City: Property Value Impacts
of Wind Turbines in an Urban Setting. Energy Economics 44, 413–421.
21
Meyerhoff, J., Ohl, C., Hartje, V., 2010. Landscape Externalities from Onshore Wind
Power. Energy Policy 38 (1), 82–92.
Micheli, M., Rouwendal, J., Dekkers, J., 2019. Border Effects in House Prices. Real Estate
Economics forthcoming.
Rosen, S., 1974. Hedonic Prices and Implicit Markets: Product Differentiation in Pure
Competition. Journal of Political Economy 82 (1), 34–55.
Rubin, D. B., 1974. Estimating Causal Effects of Treatments in Randomized and Non-
randomized Studies. Journal of Educational Psychology 66 (5), 688.
Smallwood, K. S., 2007. Estimating Wind Turbine-Caused Bird Mortality. Journal of
Wildlife Management 71 (8), 2781–2791.
Sunak, Y., Madlener, R., 2015. The Impact of Wind Farms on Property Values: A Locally
Weighted Hedonic Pricing Model. Papers in Regional Science 96 (2), 423–444.
Sunak, Y., Madlener, R., 2016. The Impact of Wind Farm Visibility on Property Values:
A Spatial Difference-in-Differences Analysis. Energy Economics 55, 79–91.
Swofford, J., Slattery, M., 2010. Public Attitudes of Wind Energy in Texas: Local Com-
munities in Close Proximity to Wind Farms and their Effect on Decision-Making.
Energy Policy 38 (5), 2508–2519.
Wager, S., Athey, S., 2018. Estimation and Inference of Heterogeneous Treatment Effects
Using Random Forests. Journal of the American Statistical Association 113 (523), 1228–
1242.
22
Appendix
Table A.1: Descriptive Statistics of the Treatment and Control Group
Treatment Group Control Group
Mean Standard Deviation Mean Standard Deviation
Housing characteristics:
Asking price in e241,025.21 164,986.92 331,201.15 246,372.81
Year of construction 1979.24 37.19 1980.00 36.44
Living space in m2151.30 57.97 157.87 64.32
Lot size in m2704.85 557.34 626.62 493.40
Number of rooms 5.38 1.76 5.54 1.79
Detached house 0.60 - 0.54 -
Semidetached house 0.16 - 0.19 -
Other house type 0.06 - 0.07 -
Terrace house 0.03 - 0.04 -
Mid-terrace house 0.06 - 0.06 -
End-terrace house 0.03 - 0.04 -
Bungalow 0.03 - 0.02 -
Villa 0.03 - 0.04 -
Locality characteristics:
Purchasing power per capita (in e) 20,808.36 3,588.27 22,615.24 4,662.33
Total inhabitants per km21,684.74 1,588.06 2,105.22 1,868.65
Unemployment rate (in %) 6.39 3.96 5.23 3.78
Foreigners (in %) 6.02 5.43 7.79 6.08
Number of buildings 415.89 302.62 468.01 309.49
Share of inhabitants aged 0-20 19.61 2.72 19.37 2.57
Share of inhabitants aged 20-35 16.08 2.93 16.42 2.98
Share of inhabitants aged 35-45 14.29 2.12 14.71 2.20
Share of inhabitants aged 45-55 16.62 1.89 16.19 1.80
Share of inhabitants aged 55-65 12.76 1.88 12.59 1.84
Share of inhabitants aged 65+ 20.64 1.86 20.72 1.87
Distance to city center (in km) 25.12 19.55 23.90 21.71
Number of Observations: 1,037,399 1,818,067
Note: Treatment group includes all houses with a wind turbine in 9 km, control group those
further away than 9 km from the next turbine.
Figure A1: Effects of a Wind Turbine on logged House Prices of Rural and Urban
Houses (5km Radius)
Note: Standard errors are clustered at the GEO-Grid level.
Figure A2: Effects of a Wind Turbine on logged House Prices of Rural and Urban
Houses (20km Radius)
Note: Standard errors are clustered at the GEO-Grid level.
Table A.2: OLS Estimation Results of Equation 1 with Various Sets of Control Variables
No Regional Controls Local Fixed Effects Only Full Controls
Coefficients Standard Errors Coefficients Standard Errors Coefficients Standard Errors
Wind turbine within
1 km distance -0.395∗∗ (0.0115) -0.148∗∗ (0.00881) -0.0735(0.00763)
1 to 2 km distance -0.364∗∗ (0.00649) -0.127∗∗ (0.00529) -0.0615∗∗ (0.00424)
2 to 3 km distance -0.328∗∗ (0.00608) -0.105∗∗ (0.00490) -0.0560∗∗ (0.00390)
3 to 4 km distance -0.301∗∗ (0.00611) -0.0834∗∗ (0.00487) -0.0441∗∗ (0.00381)
4 to 5 km distance -0.265∗∗ (0.00682) -0.0718∗∗ (0.00494) -0.0416∗∗ (0.00384)
5 to 6 km distance -0.221∗∗ (0.00705) -0.0528∗∗ (0.00502) -0.0294∗∗ (0.00394)
6 to 7 km distance -0.191∗∗ (0.00775) -0.0494∗∗ (0.00505) -0.0253∗∗ (0.00393)
7 to 8 km distance -0.164∗∗ (0.00910) -0.0314∗∗ (0.00548) -0.0139∗∗ (0.00413)
8 to 9 km distance -0.128∗∗ (0.00863) -0.0130 (0.00570) -0.000786 (0.00427)
Housing characteristics Yes Yes Yes
Locality characteristics No No Yes
Year dummies Yes Yes Yes
Municipality dummies No Yes Yes
Number of Observations: 2,855,466 2,855,466 2,855,466
R20.435 0.667 0.711
Note: ∗∗ and indicate statistical significance at the 1% and 5% level, respectively; standard errors are clustered at the GEO-Grid level.
Figure A3: Effects of a Wind Turbine on logged House Prices of Rural and Urban
Houses (50km Radius)
Note: Standard errors are clustered at the GEO-Grid level.
Figure A4: Conditional Average Treatment Effect (CATE) of a Wind Turbine Condi-
tional on the Year of Construction
Note: Confidence Intervals are given by dashed lines.
Table A.3: OLS Regression Results of Equation 1 with Rural/Urban Interaction
Coefficients Standard Errors
Wind turbine within
1 km distance -0.0841∗∗ (0.00882)
1 to 2 km distance -0.0697∗∗ (0.00482)
2 to 3 km distance -0.0610∗∗ (0.00445)
3 to 4 km distance -0.0492∗∗ (0.00438)
4 to 5 km distance -0.0486∗∗ (0.00459)
5 to 6 km distance -0.0325∗∗ (0.00457)
6 to 7 km distance -0.0251∗∗ (0.00491)
7 to 8 km distance -0.00971 (0.00521)
8 to 9 km distance 0.000229 (0.00536)
Interaction
1 km distance * urban 0.0681∗∗ (0.0153)
1 to 2 km distance * urban 0.0595∗∗ (0.00997)
2 to 3 km distance * urban 0.0442∗∗ (0.00845)
3 to 4 km distance * urban 0.0363∗∗ (0.00866)
4 to 5 km distance * urban 0.0424∗∗ (0.00817)
5 to 6 km distance * urban 0.0239∗∗ (0.00881)
6 to 7 km distance * urban 0.0199(0.00834)
7 to 8 km distance * urban 0.00563 (0.00905)
8 to 9 km distance * urban 0.00634 (0.00935)
Housing characteristics Yes
Locality characteristics Yes
Year dummies Yes
Municipality dummies Yes
Number of Observations: 2,855,466
R20.687
Note: ∗∗ and indicate statistical significance at the 1% and 5% level,respectively;
standard errors are clustered at the GEO-Grid level.
Table A.4: OLS Regression Results of Equation 1 with Old/New Interaction
Coefficients Standard Errors
Wind turbine within
1 km distance -0.0449(0.00666)
1 to 2 km distance -0.0319∗∗ (0.00437)
2 to 3 km distance -0.0281∗∗ (0.00398)
3 to 4 km distance -0.0225∗∗ (0.00395)
4 to 5 km distance -0.0223∗∗ (0.00389)
5 to 6 km distance -0.0150∗∗ (0.00412)
6 to 7 km distance -0.00735 (0.00400)
7 to 8 km distance -0.00113 (0.00425)
8 to 9 km distance 0.00538 (0.00413)
Interaction
1 km distance * build until 1949 -0.189∗∗ (0.0314)
1 to 2 km distance * build until 1949 -0.191∗∗ (0.0112)
2 to 3 km distance * build until 1949 -0.166∗∗ (0.00949)
3 to 4 km distance * build until 1949 -0.134∗∗ (0.00966)
4 to 5 km distance * build until 1949 -0.116∗∗ (0.0116)
5 to 6 km distance * build until 1949 -0.0861∗∗ (0.00987)
6 to 7 km distance * build until 1949 -0.0947∗∗ (0.0125)
7 to 8 km distance * build until 1949 -0.0794∗∗ (0.0121)
8 to 9 km distance * build until 1949 -0.0380∗∗ (0..0134)
Housing characteristics Yes
Locality characteristics Yes
Year dummies Yes
Municipality dummies Yes
Number of Observations: 2,855,466
R20.689
Note: ∗∗ and indicate statistical significance at the 1% and 5% level,respectively;
standard errors are clustered at the GEO-Grid level.
Table A.5: OLS Estimation Results of Equation 1 with Placebo Treatment and Various Sets of Control Variables
No Regional Controls Local Fixed Effects Only Full Controls
Coefficients Standard Errors Coefficients Standard Errors Coefficients Standard Errors
No Wind turbine within
1 km distance -0.0729 (-1.66) -0.0714∗∗ (-3.68) -0.0227 (-1.36)
1 to 2 km distance -0.0285 (-1.03) -0.0619∗∗ (-2.96) -0.0264 (-1.80)
2 to 3 km distance -0.0701(-2.52) -0.0318(-1.97) 0.0108 (0.74)
3 to 4 km distance -0.0846∗∗ (-3.92) -0.0273 (-1.79) 0.00274 (0.22)
4 to 5 km distance -0.0555∗∗ (-3.05) -0.0194 (-1.28) 0.00399 (0.32)
5 to 6 km distance -0.0371(-2.47) 0.0000343 (0.00) 0.0135 (1.52)
6 to 7 km distance -0.0853∗∗ (-4.96) -0.0143 (-1.15) -0.00425 (-0.45)
7 to 8 km distance -0.102∗∗ (-7.11) -0.0514∗∗ (-5.02) -0.0188(-2.18)
8 to 9 km distance -0.0819∗∗ (-6.68) -0.0482∗∗ (-4.27) -0.0201(-2.30)
Housing characteristics Yes Yes Yes
Locality characteristics No No Yes
Year dummies Yes Yes Yes
Municipality dummies No Yes Yes
Number of Observations: 986,862 986,862 986,862
R20.400 0.686 0.733
Note: ∗∗ and indicate statistical significance at the 1% and 5% level, respectively; standard errors clustered on GEO-Grid level.
Figure A5: Conditional Average Treatment Effect (CATE) of a Wind Turbine Condi-
tional on the Distance to the Next City
Note: Confidence Intervals are given by dashed lines.
... Sunak & Madlener [15] found that those with a view that from "medium to extreme" was dominated by wind turbines had fallen in price by 9-14%. Still, another German study [16] found an average treatment effect of up to −7.1% for houses within 1 km of a wind turbine, an effect that reached 0 at a distance of 8 to 9 km. A study in the Netherlands [17] found a 1.4% price decrease for houses within 2 km of a turbine. ...
... Exact longitude and latitude coordinates are also included in the model. It has been shown that the inclusion of coordinates can remedy the problem of potential spatial dependency [13,16,42]. Table 2 shows the results of the logistical regression. ...
Article
Full-text available
The expansion of wind turbines plays a significant role in developing the ability of a country like Sweden to achieve climate-neutral energy production without relying on nuclear power plants. Wind-turbine energy production is expected to grow in the coming decades. Conflicts may arise between, on the one hand, the government and the energy authority, and, on the other hand, municipalities and property owners, especially if this expansion affects other economic activities, such as tourism and reindeer husbandry, or property values. This report aims to analyse the negative capitalization of wind turbines on property values in Sweden over the last ten years. Our conclusions clearly show a relatively significant capitalization and that this capitalization is relatively local, within eight kilometers of the wind power plant. Large wind turbines, or larger clusters of wind turbines in wind farms, impose a greater socio-economic cost on lower value properties.
... Existing studies that estimate the disamenity costs of wind energy based on housing prices yield ambiguous results. While many US studies find no significant effects (Lang et al. 2014;Hoen et al. 2015;Hoen and Atkinson-Palombo 2016), several European studies find a significant reduction in prices of nearby houses of − 1.4% to − 14% (Jensen et al. 2014;Gibbons 2015;Dröes and Koster 2016;Sunak and Madlener 2016;Frondel et al. 2019). Because property value data for all potentially affected houses across Europe is unavailable, we reject hedonic pricing for evaluating disamenity costs of wind turbines within the European scope of this paper. ...
Article
Full-text available
Numerical optimization models are used to develop scenarios of the future energy system. Usually, they optimize the energy mix subject to engineering costs such as equipment and fuel. For onshore wind energy, some of these models use cost-potential curves that indicate how much electricity can be generated at what cost. These curves are upward sloping mainly because windy sites are occupied first and further expanding wind energy means deploying less favorable resources. Meanwhile, real-world wind energy expansion is curbed by local resistance, regulatory constraints, and legal challenges. This presumably reflects the perceived adverse effect that onshore wind energy has on the local human population, as well as other negative external effects. These disamenity costs are at the core of this paper. We provide a comprehensive and consistent set of cost-potential curves of wind energy for all European countries that include disamenity costs, and which can be used in energy system modeling. We combine existing valuation of disamenity costs from the literature that describe the costs as a function of the distance between turbine and households with gridded population data, granular geospatial data of wind speeds, and additional land-use constraints to calculate such curves. We find that disamenity costs are not a game changer: for most countries and assumptions, the marginal levelized cost of onshore wind energy increase by 0.2–12.5 €/MWh.
... The latter is found to be the dominant, negative external effect of wind turbines (Mattmann et al., 2016). Accordingly, several strands of literature seek to quantify the negative impact of wind turbines, for example, through their effect on property prices within a revealed preference framework (Dröes and Koster, 2016;Gibbons, 2015;Heintzelmann et al., 2017;Jensen et al., 2018;Kussel et al., 2019;Lang et al., 2014;Sims et al., 2008;Sunak and Madlener, 2016;Vyn and McCullough, 2014) or through choice experiments and surveys (Drechsler et al., 2011;Mattmann et al., 2016;Meyerhoff et al., 2010) in a stated-preferences context. Zerrahn (2017) provides a comprehensive survey of the literature on wind power externalities. ...
Article
Full-text available
Austria aims to meet 100% of its electricity demand from domestic renewable sources, such as wind energy. Apart from the benefit of reducing CO2 emissions and, potentially, system cost, wind power is also associated with negative impacts at the local level, particularly its interference with landscape aesthetics. Some of these negative impacts might be avoided by using alternative renewable energy technologies. To better understand the trade-offs involved, we quantify the opportunity cost of wind power versus its best feasible alternative, which is solar photovoltaics in Austria. Methodologically, our analysis relies on the power system model medea, a technology-rich, partial-equilibrium model of the Austro-German electricity and district heating markets. Our findings suggest that the cost of undisturbed landscapes is considerable, even more so when solar PV is mainly realised on rooftops. Under a wide range of assumptions, the opportunity cost of wind power is high enough to allow for significant compensation of the ones affected by local, negative wind turbine externalities.
... Accordingly, several strands of literature seek to quantify the negative impact of wind turbines, for example through their impact on property prices within a revealed preferences framework [12,20,23,28,33,35,48,50,55], or through choice experiments and surveys [11,38,39] in a stated preferences-context. Zerrahn [59] provides a comprehensive survey of the literature on wind power externalities. ...
Preprint
Full-text available
By 2030 Austria aims to meet 100 % of its electricity demand from domestic renewable sources, predominantly from wind and solar energy. While wind power reduces CO 2 emissions, it is also connected to negative impacts at the local level. such as interference with landscape aesthetics. Nevertheless, wind power comes at lower system integration cost than solar power, so that it effectively reduces system cost. We quantify the opportunity cost of replacing wind turbines with solar power, using the power system model medea. Our findings suggest that these cost of undisturbed landscapes are considerable, particularly when PV is not entirely rolled out as utility scale, open space installations. The opportunity cost is likely high enough to allow for significant compensation of the ones affected by local wind turbine externalities.
Chapter
Die Energiewende verändert nicht nur das Energiesystem. Ihre physische Manifestation durch den Aus- und Neubau von Stromtrassen, Wind- oder Biogasanlagen schafft zudem ‚Energielandschaften‘. Im Zuge dieser Veränderungen entstehen sich dynamisch entwickelnde öffentliche Konflikte. Landschaft als Konstrukt gesellschaftlicher Aushandlungs- und Verständigungsprozesse nimmt bei diesen Kontroversen eine zentrale Rolle ein, etwa im Hinblick auf eine veränderte Flächennutzung durch Energiepflanzen, sichtbare Überlandtrassen oder großflächige Windparks. Der Beitrag greift entsprechend Konflikte in den drei Transformationsfeldern Netzausbau, Biogas und Windkraft auf und stellt Gemeinsamkeiten und Unterschiede der Konfliktanlässe auf Grundlage aktueller Forschungsliteratur heraus. Anhand dieser Beispiele legen die Autor*innen dar, inwiefern ein verändertes Konfliktverständnis zur konstruktiven Bearbeitung von Protesten und Mobilisierungen rund um ‚Landschaft‘ in der Energiewende beitragen kann und diskutieren die Potenziale dialogorientierter Bürgerbeteiligung für eine Konfliktbearbeitung.
Article
Many scientific and engineering challenges---ranging from personalized medicine to customized marketing recommendations---require an understanding of treatment effect heterogeneity. In this paper, we develop a non-parametric causal forest for estimating heterogeneous treatment effects that extends Breiman's widely used random forest algorithm. Given a potential outcomes framework with unconfoundedness, we show that causal forests are pointwise consistent for the true treatment effect, and have an asymptotically Gaussian and centered sampling distribution. We also discuss a practical method for constructing asymptotic confidence intervals for the true treatment effect that are centered at the causal forest estimates. Our theoretical results rely on a generic Gaussian theory for a large family of random forest algorithms that, to our knowledge, is the first set of results that allows any type of random forest, including classification and regression forests, to be used for provably valid statistical inference. In experiments, we find causal forests to be substantially more powerful than classical methods based on nearest-neighbor matching, especially as the number of covariates increases.
Article
The Fukushima Daiichi accident in Japan in March 2011 caused a fundamental change in Germany's energy policy which led to the immediate shut down of nearly half of its nuclear power plants. Using data from Germany's largest internet platform for real estate and employing a difference-in-differences approach, we find that Fukushima reduced housing prices near nuclear power plants that were in operation before Fukushima by 4.9%. Housing prices near sites that were shut down right after the accident even fell by 9.8%. Our results suggest that on the German housing market, the negative economic effects of the closure of nuclear power plants dominate potential positive changes in local amenities.
Article
In this study, we investigate the effect of planned or operating wind turbines on urban home values. Previous studies, which largely produced non-significant findings, focused on rural settings. We analyzed more than 122,000 home sales, between 1998 and 2012, that occurred near 41 turbines in densely populated Massachusetts communities. Although we found the effects from various negative features (such as electricity transmission lines) and positive features (such as open space) generally accorded with previous studies, we found no net effects due to turbines in these communities. We also found no unique impact on the rate of home sales near wind turbines.
Article
Throughout the world, governments foster the deployment of wind power to mitigate negative externalities of conventional electricity generation, notably CO2 emissions. Wind turbines, however, are not free of externalities themselves, particularly interference with landscape aesthetics. We quantify these negative externalities using the life satisfaction approach. To this end, we combine household data from the German Socio-Economic Panel Study (SOEP) with a novel panel dataset on over 20,000 installations. Based on geographical coordinates and construction dates, we establish causality in a difference-in-differences design. Matching techniques drawing on exogenous weather data and geographical locations of residence ensure common trend behaviour. We show that the construction of wind turbines close to households exerts significant negative external effects on residential well-being, although they seem both spatially and temporally limited, being restricted to about 4000 m around households and decaying after five years at the latest. Robustness checks, including view shed analyses based on digital terrain models and placebo regressions, confirm our results.
Article
In many countries, wind turbines are constructed as part of a strategy to reduce dependence on fossil fuels. In this paper, we measure the external effect of wind turbines on the transaction prices of nearby houses. A unique Dutch house price dataset covering the period 1985–2011 is used, as well as the exact location of all wind turbines that were built in the Netherlands. Using a difference-in-differences methodology we find a 1.4% price decrease for houses within 2 km of a turbine. There is also evidence for anticipation effects a few years before placement of a turbine. The effect is larger for taller turbines and in urban areas. Especially the first turbine built close to a house has a negative effect on its price.
Article
Wind power in Germany has triggered much debate regarding the impacts on landscape and vista. This paper investigates the impact of wind farms on surrounding property values using a hedonic pricing model in a spatial fixed effects and a locally weighted specification. We find that proximity generally causes negative impacts on the surrounding property values. Thereby, local statistics reveal varying spatial patterns across the study area and provide evidence for visibility effects as being a key driver of negative impacts captured by distance measures. The analysis was performed for a study area in Germany.
Article
In this paper we propose methods for estimating heterogeneity in causal effects in experimental and observational studies and for conducting hypothesis tests about the magnitude of differences in treatment effects across subsets of the population. We provide a data-driven approach to partition the data into subpopulations that differ in the magnitude of their treatment effects. The approach enables the construction of valid confidence intervals for treatment effects, even with many covariates relative to the sample size, and without "sparsity" assumptions. We propose an "honest" approach to estimation, whereby one sample is used to construct the partition and another to estimate treatment effects for each subpopulation. Our approach builds on regression tree methods, modified to optimize for goodness of fit in treatment effects and to account for honest estimation. Our model selection criterion anticipates that bias will be eliminated by honest estimation and also accounts for the effect of making additional splits on the variance of treatment effect estimates within each subpopulation. We address the challenge that the "ground truth" for a causal effect is not observed for any individual unit, so that standard approaches to cross-validation must be modified. Through a simulation study, we show that for our preferred method honest estimation results in nominal coverage for 90% confidence intervals, whereas coverage ranges between 74% and 84% for nonhonest approaches. Honest estimation requires estimating the model with a smaller sample size; the cost in terms of mean squared error of treatment effects for our preferred method ranges between 7-22%.