Abstract
Photovoltaic solar farms are utility-level ground-mounted arrays of interconnected panels that convert sunlight into electricity. Little is known about the extent of any disamenity impact from these farms, despite numerous communities objecting to their construction. This study uses a property fixed-effects model to examine the disamenity impact of photovoltaic solar farms on households in England and Wales, as revealed by changes in property prices. Properties located ≤ 750 m south of an operational solar farm greater than 5 megawatts in capacity suffer a 5.4% reduction in relative prices. The effect of solar farms ≤ 5 megawatts in capacity or located more than 750 m south of properties is statistically insignificant.
1. Introduction
One of the earliest discussions in the economics literature regarding substituting coal with solar power is in Jevons (1865). Jevons’s concern was primarily about coal as the source of Britain’s industrial supremacy and the finite nature of those reserves. He argued that even if there were some way to harvest sunbeams, Britain’s economic advantage would still slip away because it had fewer sunny days than many competitive countries. Some 150 years later, worldwide coal is now, indeed being substituted with solar power, although not because it is running out. Rather, the spectacular growth of solar power is the result of global concern about climate change.
Globally, there was an increase in photovoltaic (PV) capacity of 108 gigawatts (GW) in 2019 (International Energy Agency 2021a). In the same year, the quantity of electricity generated via this technology rose by 22%. Furthermore, although the amount of electricity produced by PV (720 terawatt-hours) still lags hydro and onshore wind, this increase was greater than for any other form of renewable energy. The 2019 global share of PV solar in electricity production is, however, still only 3% (International Energy Agency 2021b). Of the additional 108 GW capacity added, 30 GW was in China, 42 GW was as distributed solar PV, and the remaining 66 GW was as centralized utility-level PV (International Energy Agency 2021a). We use the term “solar farm” to refer exclusively to centralized, ground-mounted utility-level installations.
Numerous life cycle analyses (LCAs) confirm the enormous potential for greenhouse gas (GHG) emissions reductions through the large-scale deployment of PV: GHG emissions per unit of energy produced are an order of magnitude less than those for oil, coal, or gas and are comparable to those of nuclear power. Goel, Bhat, and Prakash (2009) review seven LCAs in terms of gCO2eq/kWh for PV.1 These estimates range from 53.4 to 250 gCO2eq/kWh. Weisser (2007) identifies those factors giving rise to the sometimes significantly different results that emerge out of LCAs focusing on GHG emissions. For PV, these include differences in panel technology, installation methods, and fuel mix for the manufacturing phase, module efficiency, panel lifetimes, and differences in solar intensity. Although PV compares favorably with fossil in terms of GHG emissions, a comparison between different means of electricity generation in terms of the levelized financial cost of electricity in Aman et al. (2015) confirms that, even ignoring the problem of intermittency, PV compares poorly with fossil fuels.2
Apart from reductions in GHG emissions, there are nonclimate environmental effects associated with solar farms, some of which are less benign. Similar to GHG emissions, these environmental effects occur across the entire life cycle, from the materials acquisition and manufacturing phases through construction, operation, decommissioning, and ultimately disposal of the panels. For a recent review of the diverse environmental implications of various PV technologies, see Rabaia et al. (2021), whose study focuses chiefly on the manufacturing phase and panel disposal. Particular concern relates to the toxicity of materials used in the manufacture of some PV panels (e.g., cadmium, a known carcinogen). Other concerns relate to the use of scarce materials (e.g., indium). Together, these point to the importance of greater resource efficiency during the manufacturing phase and the need for the safe recycling or disposal of PV panels at the end of their working lives.
According to hedonic theory, the value of any in situ environmental effects associated with solar farms should be reflected in the price of nearby property.3 Indeed, although not distinguishing between CSP and PV technologies, a national survey in the United States conducted by Carlisle et al. (2015) found that 70% of people believed that construction of a “large” solar farm within view would impair property prices. In the United Kingdom, Jones, Comfort, and Hillier (2015) also report homeowner fears about the effect of such developments on property prices. Currently, however, few empirical analyses attempt to measure the disamenity effects of solar farms using the hedonic technique. This is surprising, given the number of solar farms and the numerous hedonic analyses of wind farms, a form of renewable energy presenting similar issues.4
We attempt to rectify this omission by analyzing the effect of solar farms on property values using data on planning applications for solar farms in England and Wales. These data contain the location of every planning application to construct a solar farm with an installed capacity ≥ 1 MW submitted to a local planning authority, as well as other important information (e.g., the dates the developers were granted planning permission and when the site became operational).
Our analysis uses a variant of the difference-in-differences (DID) methodology in which property fixed effects (FEs) control for all the time-invariant structural and locational attributes of properties. Subsequent analyses investigate whether the disamenity effect depends on site characteristics and whether the same disamenity effect occurs in different regions of the country. To anticipate our main findings, it appears that depending on the installed capacity of the facility and its distance from and geographical orientation vis-à-vis property, there is sometimes a statistically significant disamenity effect. These disamenity effects pass several tests of credibility and seem to imply that glare is an important issue.
2. In Situ Environmental Impacts
In situ environmental impacts associated with solar farms affect human well-being, both directly via emissions and visual or economic impacts and indirectly via ecological impacts. Surveys of these impacts are in Toutsos, Frantzeskaki, and Gekas (2005), Chiabrando, Fabrizio, and Garnero (2009), Turney and Fthenakis (2011), and Hernandez et al. (2014). Some impacts will be common to all sites, whereas others may be highly site-specific.
Ecological impacts may occur because of the on-site use of chemicals (e.g., herbicides and lubricants). The removal of vegetation, any landscape remodeling, and the creation of access roads can result in the loss of topsoil. There may be impacts on hydrological resources, including those caused by soil erosion and water pollution.5 Another concern includes the alteration of the microclimate due to the heating up of the panels. Although surprisingly little is known about the impact of solar farms on flora and fauna impacts are possible from changes to, and fragmentation of, habitat. One particular concern is that bats and birds might mistake the panels for water bodies resulting in collisions.
Turning now to direct impacts, there may be short-term inconvenience during the construction phase (e.g., from delivery vehicles as well as positive impacts on the local economy).6 Airborne particles generated by on-site activities might affect human health. There are emissions in the form of electromagnetic radiation, although it appears that these do not exceed public safety limits beyond the perimeter of the solar farm. The inverters that convert direct current into alternating current generate a humming noise. Nevertheless, beyond the perimeter of the site, any noise rapidly drops to background levels (Tell et al. 2015).
Concerning the visual impact of solar farms, del Carmen Torres-Sibille et al. (2009) investigate the size of the installation, the color contrast against the background, the repeating pattern of panels (something not encountered in nature), and whether the panels are of a uniform design. The authors elicit preferences from a sample of respondents for pictures displaying solar farms with different characteristics. Tellingly, the respondents are not indifferent to what they are shown and have preferences for particularly sorts of solar farms. Furthermore, digitally altered pictures of the same locations with and without the solar farm elicit different emotions (e.g., the affection that respondents have for a landscape).7 Additional impacts from transmission lines required to connect to the grid are possible.
Although intended to absorb sunlight, glint and glare can occur when sunlight is reflected off the flat, shiny surface of panels. Glint is defined as a brief flash of light, whereas glare is of a longer duration.8 Glare can result in a temporary loss of vision caused by the luminance of an object exceeding the capacity of the human eye.
Recently, glare has emerged as a problematic issue for PV solar. Rose and Wollert (2015) discuss court cases concerning glare from panels. Environmental impact assessments conducted for planning an application for constructing a solar farm commonly include an assessment of the potential for glare at various receptor points. There is nonetheless the ability to mitigate glare by modifying the surface of the panels.
3. Valuing the Disamenity Impact of Solar Farms
Turning to economic valuation exercises, Welsch (2016) states that there are no revealed or stated preference valuation studies for solar farms, although since his review was written, two nonhedonic valuation studies have emerged (one of which is coauthored by Welsch). More recently, in their review of the impact on property prices of energy supply infrastructure, Brinkley and Leach (2019) bemoan the absence of any hedonic valuation studies looking at solar farms. We searched for instances of the words “solar” or “photovoltaic” and “hedonic” or “house price” or “property price” appearing in the text using the ECONLIT search engine. On these grounds, we assert that there is currently only one other analysis of the impact of solar farms on the price of nearby property.9 Dröes and Koster (2021) examine the impact of 107 solar farms (and wind farms) on property prices in the Netherlands. Using a DID approach and an arbitrarily chosen cutoff of 1 km, the authors find statistically significant negative impacts, the size of which is determined by the assumptions made concerning the control group: 1–2 km, 2–5 km, or the whole of the Netherlands.10
Von Mollendorff and Welsch (2017) analyze the self-reported subjective well-being of Germans living close to various “undesirable” land uses, including solar farms. Controlling for numerous individual characteristics, they find that residing in the same area as a solar farm is not a statistically significant determinant of subjective well-being, although living in an adjacent area appears to have a small but statistically significant negative impact.
Botelho et al. (2017) present a stated preference study for solar farms in Portugal, which they claim to be the first such study. They interview a sample of local people about the minimum amount they would be willing to accept (WTA) as compensation for the construction of a solar farm at one of three different sites. The payment vehicle used is a credit applied to the respondent’s household electricity bills. They discover that the impact of solar farms is far from negligible and is on average €29.20 per household per month. They also find that WTA is related to the scale of the development such that for the largest site, which was 250 ha, average WTA was €53.24 per household per month.
Although not a valuation study as such, Al-Hamoodah et al. (2018) contains a survey in which 18 U.S. residential property assessors were asked to estimate the impact of a solar farm of 1.5 MW, 20 MW, and 102 MW on property values at distances ranging from 100 ft. to 3 miles. The average estimate was negative up to 1,000 ft. for a 1.5 MW solar farm, up to 0.5 mile for a 20 MW solar farm and up to 1 mile for a 102 MW solar farm.
There are also quite a few studies investigating the public’s willingness to pay (WTP) for energy from different renewable sources rather than conventional sources (e.g., Borchers, Duke, and Parsons 2007). However, these reflect only the WTP for a different generation mix; consequently, it is unclear the extent to which these estimates reflect the local in situ environmental impacts of the type discussed earlier.
Despite the almost complete absence of any hedonic studies measuring the disamenity impact, there is a literature that might serve as a guide for the successful application of the hedonic technique to solar farms. This literature investigates the disamenity impact of wind farms. The in situ environmental effects associated with wind farms include shadow flicker, visual intrusion, bird kill, and noise and therefore resemble those for solar farms (apart from noise).11 We briefly review applications of the hedonic technique to wind farms while confining ourselves to the literature dealing with onshore wind farms.
Hoen et al. (2011) provide an early example of a hedonic analysis of wind farms. They consider 7,459 property transactions in the United States. These properties are located within 10 miles of wind farms. They estimate a hedonic model including the as-the-crow-flies distance. The model includes property transactions before and after construction. Distance to the wind farm has no statistically significant impact on property prices; such findings are not unusual. Studies involving such large distances are not uncommon: Vyn and McCullough (2014) include property located up to 50 km from wind farms in their study in Ontario, Canada.
Although Sunak and Madlener (2015) analyze the effect of proximity to a single wind farm in Germany (and find statistically significant results), most studies analyze multiple sites, or in some cases, all the wind farms in a country. Dröes and Koster (2016), for example, use a DID approach to examine the disamenity impacts of all the wind farms in the Netherlands. Their data set contains 2.2 million property transactions over 25 years and 1,800 wind turbines. They observe more pronounced effects for larger developments and uncover evidence of considerable regional heterogeneity.
Many hedonic analyses of wind farms adopt a postcode FE/census-block FE approach, whereas others use a property FE/repeat-sales approach. For an example of a hedonic analysis using both, see Heintzelman and Tuttle (2012). Some studies (e.g., Lang, Opaluch, and Sfinarolakis 2014) also investigate the disamenity impacts at different phases of the project cycle (e.g., the time the announcement was made vs. the time the site became operational). Gibbons (2015) presents evidence for England and Wales that a wind farm within 2 km and visible is associated with a statistically significant 5%–6% decrease in property prices.
We adopt many of the modeling strategies as those used by researchers investigating the disamenity impacts of onshore wind farms but with two important differences. The first relates to the much greater distance over which wind farms theoretically remain visible. This is because the height of the turbines is often more than 100 m. Specifically, we concentrate on measuring the change in the price of property within 750 m of a solar farm. This decision is based on the findings of the survey of realtors’ views in Al-Hamoodah et al. (2018) and in conjunction with information on the size of the typical solar farm in England and Wales.
The second important difference is that although several hedonic analyses of wind farms (e.g., Lang, Opaluch, and Sfinarolakis 2014) adopt a DID “doughnut” strategy, we use as our control group those locations where there either is already or would be one day an operational solar farm. This approach is the same as that used by Gibbons (2015), who argues that it avoids the implicit assumption of common trends underlying DID. We exploit the existence of many operational solar farms and the differences in the dates they became operational to measure the impact on property prices.
4. Methodology
The price (PRICE) of property i transacted at time t is a function of j = 1…J (structural attributes) and k = 1…K (locational attributes). Also included is the indicator variable TREATMENT denoting those observations that have been treated in some way. The treatment variable takes the value unity if the transaction occurs after the treatment and is zero otherwise. The coefficient on TREATMENT measures the change in property prices that occurs because of the treatment. We account for more than one sort of treatment simply by adding further indicator variables. Following Kuminoff and Pope (2014), however, we interpret the coefficients on the treatment variables as evidence of the capitalization of disamenity impacts rather than as a measure of welfare change. Last, we include θt, representing a time trend assumed identical for treated and untreated properties. The basic specification of the model is 
, where βj and γk represent coefficients associated with the structural and locational attributes of the properties, respectively. The key assumption for identification is that E[εit |TREATMENTit] = 0. However, we rely on a simple modification of this specification. Specifically, we include property FE, so that the model becomes Ln(PRICEit) = βi + δTREATMENTit + θt + εit. Now both the locational and structural attributes of the property are modeled using FEs represented by βi. By including property FE, we control for all time-invariant characteristics. Any unobserved time-varying locational and property characteristics are, by contrast, consigned to the error term. Kuminoff, Parmeter, and Pope (2010) discuss different strategies to address the problem of omitted variables in hedonic studies.
5. Data
According to the publicly available Renewable Energy Planning Database (REPD) for September 2017, of 1,860 planning applications, 1,060 have resulted in operational PV solar installations, with a further 18 solar installations under construction and permission granted for 231 more.12 The combined capacity of operational sites was 8.1 GW.
Planning applications are unevenly distributed, with 549 for schemes in the South West region. The first solar installation became operational on July 1, 2011; the last became operational on August 24, 2017.
The REPD data include the eastings and northings of all planning applications as well as the installed capacity (CAPACITY) in MW. Only planning applications to construct solar installations ≥ 1 MW are included in the REPD.
Applications to the local planning authority to construct a solar installation are either accepted or rejected and are possibly subject to appeal or even intervention by the Secretary of State, who can overturn local decisions. If permission is granted for building a solar installation, it results in the construction of a solar installation, which then becomes operational, or the site is abandoned or permission is allowed to expire. A notable feature of the REPD is that it includes exact dates when planning permission was sought and granted, when construction began, and when the solar installation became operational.
The REPD usually records whether the solar installation was built on the ground or on a roof or even floating on a reservoir. Most utility-level solar installations, particularly the larger ones, of necessity are placed on the ground, and in the absence of any explicit information to the contrary, this is our assumption.
From the England and Wales Land Registry (EWLR), we obtained every property transaction for those six-digit postcodes whose centroids are within 1,000 m of the location of a planning application for the construction of a solar installation.13 Property transaction data in our analysis range from January 2, 1995, to November 27, 2017. An important limitation of the EWLR database is that although it provides information on all property transactions, it includes few details concerning property characteristics (property type, whether the property is a new build and freehold). It is important to note that this is not a limitation for our analysis, since we use property FEs to control for all property characteristics. We also obtain from the EWLR the monthly house price index (HPI) for each English and Welsh unitary authority or upper-tier authority.14 Note that the HPI is property type–specific, covering detached properties (DETACHED), semidetached properties (SEMI), terraced properties (TERRACED), and flats/maisonettes (FLATS). Similar to Beltrán, Maddison, and Elliott (2018, 2019), the HPI is used to adjust property prices to their January 2015 values.
Because we use data from the EWLR, we exclude from the analysis a small number of solar installations in Northern Ireland and Scotland. We drop any solar farm that is not operational or whose coordinates are missing. This leaves 1,059 solar installations. Of the 309,832 residential property transactions in postcodes whose centroids are within 1,000 m of these solar farms, 271,551 are within 1,000 m of only one operational solar installation; properties within 1,000 m of more than one solar installation are dropped. Some operational solar installations do not have any transactions within 1,000 m. Given our focus on solar farms, we drop 49,201 and 14,819 observations for installations that are described as mounted on a roof or floating on water, respectively. We also drop any transactions described as nonstandard, since these might not represent an arm’s-length transaction. The geographical location of operational solar farms is shown in Figure 1.
Operational Solar Farms in England and Wales
The final data set contains 204,315 transactions within 1,000 m of 898 operational solar farms, of which 96,315 are within 750 m of 772 solar farms. Using information on the date of each property transaction and the history of each solar farm, we create four dummy variables. The variable SUBMITTED takes the value unity for property transactions occurring after submission of a planning application to construct the solar farm and zero otherwise. Analogously defined variables relate to property transactions made after planning permission has been granted, after construction has begun, and after the solar farm has become operational corresponding to the dummy variables PERMITTED, CONSTRUCTION, and OPERATIONAL.
Table 1 shows summary statistics for the data set. Note that the number of observations for OPERATIONAL is greater than the number of observations for any other phase: in the REPD there are many solar farms where either the date the planning application was made or permitted or the date construction began is missing. The mean number of days from submitting a planning application to the application being granted is 148. The corresponding figure for the number of days from a planning application being granted to the start of construction is 250. From the start of construction to when the site became operational, the time elapsed is 98 days. We also include information on the geographical orientation of properties vis-à-vis the solar farm, specifically a dummy variable indicating whether they are situated to the north (NORTH) and the distance (DISTANCE) measured in meters. The rationale for creating the dummy variable NORTH is discussed later.
Data Set
Finally, we tag any transactions that share the same primary addressable object name, secondary addressable object name (normally used to identify flats), street name, town, and unitary authority or upper-tier authority and postcode as well as property type and ownership type. The only property characteristic permitted to change between sales without signaling a different property is the variable NEW, which takes the value unity if the property is a new build and zero if the property has been preowned. The data set is summarized in Table 1.
6. Results
We are now able to determine whether events associated with the development of a solar farm resulted in a statistically significant change in the price of property. The following regression equations include, along with the variable NEW, property FEs and a common time trend modeled using a restricted spline function, whose 22 knots are evenly spaced at approximately yearly intervals. We also include the following treatment variables: SUBMITTED− PERMITTED, PERMITTED−CONSTRUCTION, CONSTRUCTION−OPERATIONAL, and OPERATIONAL. The purpose of differencing these dummy variables is so that their coefficients represent the effect on prices during each phase of the development cycle. For example, the coefficient on the differenced variable SUBMITTED−PERMITTED represents the effect on prices during the period after a planning application was submitted but before the application was permitted. By contrast, the variable OPERATIONAL is not differenced because there are no examples of a solar farm being decommissioned and the site it occupied being restored to its former state.
Each developmental phase provides fresh information to households and a potentially different set of disamenity impacts, some of brief duration. For example, the effect of submitting a planning application on property prices might be limited because even if plans are submitted, planning permission might be withheld. Furthermore, although it might seem to be the moment when any disamenity impacts are crystallized, the effect on property prices of permission being granted is still somewhat uncertain; as previously mentioned, there are sites where the development status report in the REPD indicates that the site has been abandoned or that planning permission has been allowed to expire. The construction phase includes building activities that cease when the site enters operation and perhaps only then can households gauge any disamenity effects, particularly whether a property is affected by glare. The results of the regression analysis are shown in Table 2.
Regression Results
Model 1 includes all properties ≤ 750 m from a solar farm following Al-Hamoodah et al. (2018) and in consideration of the average size of the solar farms contained in the data set. None of the developmental phases is statistically significant, even at the 10% level of significance. Note that, following Cameron and Miller (2015), standard errors throughout are clustered at the level of the solar farm. The only statistically significant variable is NEW, which as expected, points to a premium for new as opposed to preowned properties.15 In model 2, we continue to include all properties ≤ 750 m from a solar farm but exclude the common time trend. This assumes that in the absence of a solar farm, prices would have followed the HPI for the upper tier or unitary local authority. Once again, the results point to no statistically significant impact from an operational solar farm. These results do not change much, even if the common time trend varies over NUTS1 regions.16 These results are largely unaffected by using year FEs instead of a restricted spline function. We also changed the distance to ≤ 1,000 m and ≤ 500 m (models 3 and 4, respectively), but the results continue to show no statistically significant impact from an operational solar farm. Last, the results do not change if we use a standard DID doughnut model in which the treatment group is ≤ 750 m and the control group is between 750 m and 1,000 m: there is no statistically significant impact associated with the operational phase of a solar farm.
7. Discussion
The regression results shown in Table 2 refer to all of England and Wales. But in this specification, the disamenity impact of each developmental phase, along with all the other coefficients, is constrained to be the same across all regions. To examine the consequence of relaxing this assumption, thereby allowing for different implicit prices caused by differences in supply and demand conditions, we analyze the disamenity impact of solar farms by NUTS1 region, following Michaels and Smith (1990).
Models 5–9 in Table 3 present results for each NUTS1 region of England, while model 10 presents results for Wales, which is both a NUTS1 region and a country. Because of the very small number of solar farms in certain NUTS1 regions, we are compelled to combine observations from the geographically neighboring regions of Yorkshire and Humberside, North East, and North West. We also combine observations from the West Midlands and East Midlands and from London and the South East. The coefficient on the variable OPERATIONAL is not statistically significant for any of the regions/combined regions, even at the 10% level of significance.
Regression Results by Region
We turn to consider the impact of solar farms according to site characteristics. We measure the scale of the development by installed capacity. Note immediately that this is hardly the ideal measure of the scale of the disamenity impact because it is implausible to suggest that households care about this attribute. Unfortunately, other information (e.g., on the exact area covered by each solar farm) is unavailable. Although there is likely to be a correlation between the two, this is affected by a range of other factors (e.g., the desire of the site owner to obtain agricultural co-benefits in the form of grazing, implying greater spacing between the panels, and the efficiency of the panels).
The geographical orientation of properties vis-à-vis the solar farm is another potentially important characteristic; it determines whether a property is capable of being subject to glare. The panels used in solar farms may have either a fixed mounting or one with a single or dual axis. Panels with a fixed mounting should, in the Northern Hemisphere, point south, at least if the goal is to maximize electricity production. Because sunlight reflected by panels cannot affect locations that are more than 90° away from the normal to the panels, it follows that it is impossible for any property situated to the north of a solar farm with a fixed mounting to be affected by glare.
Some solar farms in the Northern Hemisphere nevertheless use panels facing an east–west direction.17 East–west facing panels generate more electricity in the morning and afternoon compared with south-facing panels and their noontime peaks. We suppose that such a design might subject properties situated immediately to the north to glare. However, there are at present no solar farms with an east–west orientation in the United Kingdom.18 The ability to track the motion of the sun could have implications for which properties are subject to glare. Once again, however, there are no examples of solar farms using trackers in our data set.19 We conclude that only those properties to the south of solar farms are (perhaps) subject to glare.
In Table 4, we stratify the data into different groups according to installed capacity ≤ 5 MW or greater than 5 MW, which is approximately the average, and properties to the north or south of the solar farm. Stratifying the data in this way yields coefficients identical to one in which all possible group interaction terms are included in a single regression while allowing the variance of the error term potentially to differ across groups. For most models, the coefficient on the variable OPERATIONAL is still not statistically significant. This is important information in that it suggests that in many cases, solar farms do not impose significant disamenity impacts. It is nonetheless extremely interesting to see that, while most disamenity impacts are statistically insignificant, in model 14, the impact is statistically significant, even at the 1% level of significance. One would undoubtedly have described such a situation as a worst-case scenario: located ≤ 750 m south of a large (> 5 MW) solar farm thereby potentially subject to glare. Using the formula of Halvorsen and Palmquist (1980), this equates to a 5.4% reduction in property prices.20 Such findings also point to the potential importance of technologies, such as antireflective coatings, capable of overcoming the problem of glare from PV panels.
Regression Results by Site Characteristics
Given the statistical significance of the results of model 14, we subject that model to additional tests of robustness. Beginning with model 15 in Table 5, we include only those observations where the installed capacity is greater than 10 MW while continuing to include only those properties to the south whose distance is ≤ 750 m. This causes the coefficient in the model to almost double, and the statistical significance to increase further with the t-statistic now pointing to statistical significance at the 0.1% level of significance. Thus, the disamenity effects of solar farms appear to be sensitive to the scale of the development. Model 16 reverts to installed capacity greater than 5 MW with properties to the south but includes only those properties whose distance is between 750 m and 1,000 m. The coefficient on OPERATIONAL is now no longer statistically significant, even at the 10% level of significance. It is interesting to note, however, that the variable SUBMITTED−PERMITTED is negative and significant at the 1% level of significance. This might suggest that households located at a distance of more than 750 m initially fear that they might be impacted but realize that they are not after more information becomes available. This provides support for the views of the surveyed realtors in Al-Hamoodah et al. (2018), namely, that significant disamenity impacts beyond 0.5 mile (750 m) are unlikely. In model 17, we examine how the disamenity impacts change when the distance to the solar farm is reduced from ≤ 750 m to ≤ 500 m. Being closer should increase (in absolute terms) the disamenity impact, and this is precisely what happens: the coefficient on OPERATIONAL increases while retaining its significance at the 1% level of statistical significance.
Further Regression Results
Continuing our scrutiny of model 14, we present a placebo test with “false” treatments. Specifically, rather than using the actual dates when plans for solar farms were submitted, we randomize these dates. The time elapsed before planning permission is granted, construction begins, and the site becomes operational stays the same. This procedure results in a set of false treatment dummies. Clearly, the statistical significance of these dummies calls into question our identification strategy. However, as Table 6 shows, none of the false treatment dummies are statistically significant, even at the 10% level. Also shown in Table 6 is model 19, in which we investigate whether the results are robust to eliminating outlier observations. Specifically, we reestimate model 14 but drop any observations where the real price of property is in either the 1st or 99th percentile. The results are virtually identical to those for model 14.
Regression Results for Tests of Robustness
We consider a further test of model 14. A shortcoming of the property FE model is that it is necessary to wait for a before-and-after sale of the same property to identify the impact of an operational solar farm on property prices. Therefore, many property sales do not contribute to identification. There are also concerns about the property FE model in that it assumes that properties are indeed the same and have not, for example, undergone significant alterations in between sales (Shiller 1993).
In model 20, therefore, following Gibbons (2015), we present a postcode FE model. In this model, we include the (very) limited range of property characteristics available through the EWLR and include FEs for different postcodes. Because the average postcode contains only 15 properties, these FEs do a good job of controlling for neighborhood characteristics, but it is nonetheless quite clear such a model does not fully control for property characteristics.
Despite this, all the property type dummies in model 20 have the expected signs, and these are highly significant. The omitted category of property is FLATS. Likewise, properties that are freehold (as opposed to leasehold) are much more expensive than are properties that are new rather than preowned. More important, the coefficient on OPERATIONAL is still negative and little different than its counterpart in model 14. The statistical significance of the coefficient, however, reduces such that it is now significant only at the 5% level of significance. This is probably due to the lack of control for property characteristics more than outweighing the increase in the number of observations, potentially helping identify the effect of an operational solar farm.
Finally, we investigate the consequences of changing the way we control for generalized property price inflation. Rather than deflating property prices by upper-tier or unitary local authority property type–specific time trends, we include region-specific restricted spline functions, again with 22 knots evenly spaced at approximately yearly intervals. In addition, we take as the dependent variable the log of nominal prices (NOMINAL). The consequences are shown in model 21 in Table 6. It appears that this alternative approach leaves the results largely unaffected (when compared with model 14), both in terms of the size of the treatment effect of interest and in its statistical significance.
8. Conclusions
We find that solar farms impose disamenity impacts, at least on properties located ≤ 750 m south of a solar farm with a capacity greater than 5 MW. Specifically, we estimate a disamenity impact of −5.4%. Such findings could inform discussions about the scale of any compensation paid to affected households and, more importantly, help identify the sorts of solar farm projects that avoid disamenity impacts altogether. The directional heterogeneity observed is readily explained by the ability of panels to subject properties to the south to glare. We find that this disamenity impact increases further for solar farms with capacities greater than 10 MW, but in line with prior expectations, it disappears at distances of more than 750 m. These results are also robust to a placebo test using false treatment dates.
Our results differ from the only other study that examines the impact of solar farms on property prices because we find that only in certain circumstances do solar farms incur disamenity impacts. It would be interesting to compare our findings with those from studies conducted elsewhere, when they emerge. We also see scope for further research aiming to distinguish between the disamenity impacts caused by (1) proximity to the solar farm; (2) a view of the solar farm unobscured by undulations of the land, vegetation, or buildings; and (3) glare from the solar farm. Such studies will require highly detailed site- and property-specific information. Further research is also needed to establish whether disamenity impacts depend on existing land use. For example, are solar farms erected on greenfields deemed worse than those on brownfield sites? Future studies might attempt to measure the size of the disamenity impact by the amount of land covered by the panels rather than by installed capacity.
It will also be interesting to see whether any more evidence emerges from stated preference surveys in which people are asked about their WTP to avoid or WTA compensation for putting up with a hypothetical solar farm. Such studies must inquire about developments in the neighborhood of the respondent rather than in some other, undisclosed location. Perhaps these studies will also help identify what characteristics of the solar farm most concern people. Finally, these figures could be used to estimate the “externality adder” for electricity generated by solar farms in England and Wales and compare them with the external costs from other forms of renewable and nonrenewable electricity generation.
Acknowledgments
We acknowledge comments received at the 2020 online conferences by the Association of Environmental and Resource Economists and the European Association of Environmental and Resource Economists. We express our gratitude to two anonymous referees for helpful comments on an earlier draft.
Footnotes
↵1 Although not the focus of this article, concentrated solar power technologies use mirrors to reflect and concentrate sunlight onto receivers that collect solar energy and convert it to heat. The thermal energy is used to produce electricity via a turbine or heat engine driving a generator.
↵2 McCombie and Jefferson (2016) compare different forms of renewable energy with nuclear power in terms of power density (i.e., watts per square meter of land). The power density of PV depends on solar intensity, which is location specific, but the authors nonetheless find that PV is generally superior to strip-mined coal, biomass, and wind (although greatly inferior to nuclear).
↵3 By contrast, the environmental impacts associated with material acquisition, manufacturing, and disposal do not affect local property markets because these activities take place elsewhere (e.g., where the panels are constructed, which is now mainly in China).
↵4 Official planning guidance in the United Kingdom states that many issues affecting solar farms will be the same as those affecting wind farms, but that the visual effects associated with the former should be geographically more restricted.
↵5 Water is consumed for washing the panels, without which their efficiency begins to decline.
↵6 A 25 MW scheme covering 76 ha at Rampisham Down in Dorset was claimed by the developers to support at least two fulltime operational and maintenance staff. In addition, the construction phase would support up to 30 construction jobs and 200 supply chain jobs.
↵7 In perusing the BBC local news website, one is soon tempted to conclude that the main objection to solar farms is indeed the visual impact. An application at Charborough Estate in Dorset, rejected after no fewer than 650 objections were received, was described as the “industrialization” of the landscape. One resident objecting to the proposed development in South Hams in Devon likewise remarked: “It seems as if the whole of this area is going to be smothered in glass panels, and we really don’t want that.” An application to construct a solar farm in Duns Tew in Oxfordshire was approved despite being described by planning officers as offering an “expanse of dark-colored panels,” which would appear as a “wholly alien” feature of the landscape. One campaigner put it more bluntly: “It’s a power station at the end of the day, and we don’t want them on our green fields.”
↵8 Henceforth, we refer only to glare.
↵9 Note that there is a significant body of literature that considers whether placing solar panels on the roofs of private properties affects the price of the property, and if so, whether the present value net benefit of the solar panels is more than or less than fully capitalized into the price. For a recent example of such a study, see Hoen et al. (2017). Interestingly, Hoen et al. (2017) call for research into the effect of third-party solar developments on property values of precisely the type presented in our article.
↵10 We learned of the results of this study after submitting the present article. Leaving aside methodological differences, there are several other differences between our work and that of Dröes and Koster (2021). We observe 204,315 transactions within 1 km of 898 solar farms compared with 12,650 within 1 km of 107 solar farms for the study undertaken in the Netherlands. In addition, we investigate price effects across the whole planning cycle, from submitting the planning application to its acceptance as well as its construction and the point at which it becomes operational. We also investigate the importance of the scale of the development and the geographical orientation of properties vis-à-vis the solar farm. In addition to Dröes and Koster (2021), a working paper has recently emerged analyzing the impact of solar farms on property values in the United States.
↵11 As a directional externality, we view the issue of shadow flicker as being like the problem of glare for solar farms.
↵12 In the United Kingdom, deciding on applications to construct solar farms is the responsibility of local planning authorities. In the guidance with which they are provided (available at https://www.gov.uk/guidance/renewable-and-low-carbon-energy), there is a presumption against granting applications for construction of solar farms in the greenbelt surrounding cities. Developers are encouraged to focus attention on brownfield sites and land of low environmental importance without heritage value. Permission is also more easily obtained by developments using land that is either nonagricultural or of a poorer quality or where the development is such that some form of agricultural use continues to be possible. Caution is to be exercised where glare from panels might affect transportation, and consideration must always be given to mitigating adverse visual effects (e.g., by planting hedges).
↵13 In practice, the full postcode of a property in the United Kingdom can range between six and eight alphanumeric characters. We use the term “six-digit postcode” to refer to the full postcode of the property. The mean number of properties in a six-digit postcode is 15.
↵14 Outside London, local government in England operates under a one-tier or a two-tier system. The house price indices we use are based on information on the geographical areas covered by single-tier or upper-tier local government (i.e., unitary authorities or metropolitan districts and nonmetropolitan counties). Local government in Wales operates under a one-tier system. There are 141 such areas in total.
↵15 A limitation of our analysis, which we do not attempt to address, is the fact that the technique places more weight on those properties that are transacted more frequently; see Steele and Goy (1997).
↵16 The Nomenclature of Territorial Units for Statistics (NUTS) is a hierarchical classification of administrative areas used across the European Union for statistical purposes. There are 12 NUTS1 regions in the United Kingdom, although because solar farms are unevenly distributed, we combine some regions.
↵17 For an example of a solar farm already using such a design, see the Cestas solar farm in France.
↵18 On May 28, 2020, the Secretary of State gave permission for a 350 MW solar farm at Graveney in Kent. When built, this will be the United Kingdom’s largest solar farm, covering 360 ha and consisting of 800,000 panels, reportedly saving 68,000 tonnes of carbon a year. This will be the first solar farm in the United Kingdom to use the east–west design.
↵19 While industry sources claim that, internationally, between 2009 and 2012, 85% of installations greater than 1 MW in capacity used trackers; in the United Kingdom, the first solar farms to use trackers only became operational in late 2019. The first solar farm to do so was also the first to use bifacial panels, generating energy from both sides of the panel. The desirability of a tracking system depends on many factors, including the relative cost of panels compared with the additional costs of the tracking system. Because the relative cost of panels has fallen and the attractiveness of tracking systems has waned, it has become more economical simply to purchase additional panels; see http://www.reuk.co.uk/wordpress/solar/solar-tracker/.
↵20 Given that in the United Kingdom, as elsewhere, solar is supplanting fossil fuel, it is interesting to compare the disamenity impact of fossil fuel plants notwithstanding that localized effects do not include the damage done by regional air pollution or GHG emissions. Davis (2011) considers 92 such plants that opened 1993–2000 in the United States, all of which were greater than 100 MW. Results point to a reduction in the price of property of 4%–7% within a 2-mile radius (an area of 32.5 km2 compared with the 0.9 km2 that lies 750 m south of a solar farm greater than 5 MW). At the same time, the fossil fuel power stations considered by Davis have a capacity greater than 100 MW. Furthermore, the capacity factor of coal in the United Kingdom is 58.4%, compared with 10.2% for solar.
