Does Proximity to a Power Plant Affect Housing Property Values of a City in South Korea? An Empirical Investigation

Abstract

The South Korean government plans to switch from a centralized power generation method to a distributed one. However, due to opposition from local residents, construction of distributed power plants is frequently delayed or suspended. This study attempts to investigate whether proximity to a power plant negatively affects housing property values, using the hedonic pricing technique and quantitatively analyzing the level of impact. To this end, 2291 apartment sales data from a specific city in the South Korean Seoul Metropolitan area with a power plant were used. As a result of the analysis, it was found that proximity to a power plant had a negative effect on apartment prices, which was statistically significant at the significance level of 5%. The difference in apartment prices per 1 km direct distance from the power plant to the apartment was derived as KRW 8 million (USD 7.1 thousand). This value is about 0.7% of the average price of apartments in the area (KRW 1102 million = USD 0.98 thousand). The results of this study can be used as a useful reference when the government determines the size of subsidies for local residents near power plants.

1. Introduction

South Korea is a peninsular country, but it has a military confrontation with North Korea to the north and is not connected to any other countries’ power grids. Although South Korea is not an island in terms of geography, it is a representative electricity island country from a power system perspective. Since it is impossible to buy electricity from or sell electricity to foreign countries, it is necessary to have enough flexible, large-capacity generators such as natural gas-fired power plants. In fact, in September 2011, power demand exceeded power supply, resulting in a nationwide rolling blackout, which caused great economic damage. In addition, South Korea has a relatively small amount of land and many mountainous areas. The degree of urbanization is high, with the population concentrated mainly in large cities. As a distributed power source, a natural gas-fired power plant must be located in a large city to provide stable power supply. Japan, located in East Asia along with South Korea, is also a representative electricity island country. It is geographically an island. Israel is neither an island like Japan nor a peninsula like South Korea, but it is a representative electricity island. These electricity island countries are striving to secure power supply stability by constructing a sufficient number of power plants for economic development and survival.

About 70% of South Korea’s total power generation is produced by coal-fired and nuclear power plants, and most of them are located on the seashore to secure cooling water and emit hot water [1]. Large-scale power transmission and substation facilities are needed in the process of supplying electricity produced by coal-fired and nuclear power plants located on the seashore to urban areas and industrial complexes, which are areas of high demand, but they are typically hated facilities and it is difficult to construct additional ones [2,3,4,5,6]. In addition, such a centralized energy system causes conflicts between regions [7,8]. Most of the coal-fired and nuclear power plants are located in Chungnam and Gyeongbuk, respectively, in South Korea. As of 2021, the amount of power generation in Seoul, Gyeonggi, Chungnam, and Gyeongbuk were 5343 GWh, 82,232 GWh, 111,228 GWh, and 81,381 GWh, respectively [9]. The values of annual electricity consumption of those regions as of 2021 were 47,295 GWh, 133,445 GWh, 48,801 GWh, and 44,258 GWh, respectively. Among them, Seoul and Gyeonggi are included in the Seoul Metropolitan area. The power independence levels in Seoul, Gyeonggi, Chungnam, and Gyeongbuk, calculated by dividing the power generation in each region by electricity consumption, are 11%, 62%, 228%, and 184%, respectively.

In other words, about twice as much electricity is produced in Chungnam and Gyeongbuk. On the other hand, power generation in Seoul, which consumes the most electricity, accounts for only 11% of the electricity consumed in the region. Therefore, residents around the coal-fired and nuclear power plants oppose the expansion of power plants, arguing that it is a one-sided sacrifice to supply electricity to the urban area. In addition, protests against the construction of power transmission towers and power lines are taking place throughout the country. As such, in South Korea, local residents’ opposition to the construction of power plants and transmission lines to deliver local electricity to cities is growing, which can lead to conflicts between regions. Therefore, securing the acceptance of residents has become an important factor in constructing power plants and transmission lines, and as a result, it is pointed out that this is a national loss because much more time and costs must be spent on construction than in the past.

Therefore, the government plans to switch from a centralized power generation method to a distributed one. Specifically, it is intended to expand the proportion of distributed power generation from 12% in 2017 to 30% by 2040 [10]. In South Korea, small power generation facilities of 40 MW or less or power generation facilities of 500 MW or less that can access 154 kV transmission lines near the demand site are recognized as distributed power sources. However, despite the government’s plans and the need for distributed power sources, the construction of power plants has been hampered or delayed due to residents’ opposition. For example, in Incheon, the construction of a 40 MW fuel cell power plant was suspended for about a year due to opposition from local residents. A solid refuse fuel cogeneration plant completed in Naju in September 2017 has not yet operated normally due to opposition from residents. In other words, power plants as distributed power sources are recognized as hated facilities in South Korea. Public acceptance is an important determinant of business performance in the energy sector [11].

Since power plant facilities are generally considered ‘not in my backyard’ facilities, securing public acceptance for them is a major challenge for governments and power generation operators. In this respect, a number of studies related to the public’s perception of power plant facilities have been conducted, and they are largely classified into three categories. First, there are studies analyzing public acceptance of power plant facilities. Specifically, public acceptance studies were conducted on nuclear power plants [12,13,14], renewable energy [15,16,17], and natural gas-fired power generation [18,19]. Second, there are research cases in which the external effects of power plant facilities are evaluated in monetary units. For example, Sun et al., [20] and Sun et al., [21] examined public WTPs to avoid the construction of nuclear power plants and waste-to-energy power plants near their residences, respectively. In addition, there are studies that investigated WTP to reduce greenhouse gases emitted during power generation [22,23].

Third, there are studies that analyzed the effect of power plants on housing prices. It was found that energy facilities generally had a negative effect on housing prices. For example, Boes et al. [24], using a hedonic approach, reported a 2.3% drop in rent for apartments near nuclear power plants in Switzerland after the Fukushima accident. According to Jensen [25], as a result of applying the hedonic method, adding one onshore wind turbine within 3 km in Denmark reduces residential house prices by 0.2% to 1.1%. Joly and De Jaeger [26] applied a hedonic technique to real estate transaction data in Belgium and found that proximity to wind turbines reduced real estate prices by an average of 2 to 3%. Using a geographically weighted regression model, Tsai [27] reported that the price of a house with a coal-fired power plant within 15 km was about 25% lower than that of a house without one within this radius. On the other hand, an analysis by Dastrup et al. [28] that applied the hedonic technique showed that household solar panels have a premium of about 3.5% of the house price.

In South Korea, there are two main reasons for the local residents’ opposition to the construction of distributed power plants. The first is that harm is caused by air pollutants emitted from power plants. Second, there are concerns that power plants will negatively affect housing property values. The former can be substantially improved by power plant operators installing remote monitoring systems for the air environment to prevent air pollutants from being discharged into the atmosphere. In addition, the government limits the total amount of air pollutants emitted from power plants through the Air Environment Conservation Act and conducts real-time monitoring. Besides, technologies for reducing pollutant emissions from power plants and utilizing renewable energy are carried out [29,30,31]. On the other hand, since a fall in housing property values is difficult to quantitatively evaluate, it is hard to resolve the conflict between the power plant operator, the government, and local residents. The government has enacted a law on supporting areas around power plants and has been implementing various support projects for local residents within a 5 km radius of a power plant. However, despite the government’s support policy, it is not easy to mediate the conflict over the installation of distributed power sources.

The basis for the possibility that proximity to a power plant may affect housing prices in South Korea is that power plants are regarded as a kind of ‘not in my backyard’ facility. The biggest reason for this is the aesthetic discomfort caused by smoke emitted from chimneys in power plants. Whether the smoke is harmless water vapor or air pollutants, it is a nuisance to nearby residents. People are anxious that the closer they are to the power plant, the worse it will be for their health, even if this is not true. Thus, residents and real estate agents believe that proximity to power plants has a negative impact on housing prices.

In 2021, the share of non-financial assets, including real estate, in household assets in South Korea was 64.4%, which is significantly higher than that of the US (28.5%), Japan (37.0%), and the UK (46.2%) [32]. Therefore, the South Korean public has a strong sense of resistance to falling real estate prices. This study attempts to determine whether the proximity to a power plant negatively affects housing property values using the hedonic pricing technique. In addition, if the proximity has a significant impact on the value of housing assets, the level of the impact will be quantitatively analyzed. To this end, 2291 apartment sales data from a specific city in the South Korean Seoul Metropolitan area with a distributed power plant were used. An apartment is a typical housing type in South Korea. Variables that affect apartment prices were selected through previous studies and the direct distance from the power plant to the apartment was added as a variable. The rest of this paper is largely composed of three sections. Section 2 deals with the method and analysis model. The data, analysis results, and a discussion of the results are presented in Section 3. The last section is devoted to conclusions.

2. Materials and Methods

2.1. Method: Hedonic Price Technique

The value of a good consists of the value of its detailed attributes [33]. Applying this to the housing market means that the buyer chooses from among several residences the one with the greatest sum of values of its detailed attributes [34]. These two concepts are the basis for the hedonic price technique. The basic hypothesis of the hedonic price model is that consumers select a number of housing characteristics when choosing a residence, and the technique is mainly used to measure the value of each of the residence’s intrinsic attributes [24,25,26,27,28,35,36,37,38,39,40]. The hedonic price technique is widely used to estimate the effect of each attribute on the housing price or the value of each attribute of a house. In other words, the relationship between housing price and attributes of housing can be explained through the hedonic price technique. Therefore, the basic model of the hedonic price technique can be expressed as $Y_i=\mathrm{f}\left(X_i\right)$, where $Y_i$ is the price of the house and $X_i$ is the attribute vector of the house. In this study, the hedonic price technique is applied to analyze the effect of proximity to a power plant on housing prices in South Korea.

2.2. Model

Theoretically, the functional form of the hedonic price equation is not limited. In the literature, linear-linear [41], linear-log [42], log-linear [43], log-log [44], and Box-Cox transformation [35,45,46] models have been used as functional forms of the hedonic price equation. Among them, the Box-Cox transformation model is sometimes preferred because it has more flexibility and robustness than other models [47]. It has also been reported that the log-log model has higher explanatory power than other models [48,49]. However, in general, no particular functional form is superior and the researcher should choose one appropriately [50,51,52]. In this study, a total of six price equations are estimated without arbitrarily adopting the price equation, and the one with the smallest prediction error is selected to interpret the analysis results.

First, the authors apply four types of functions: linear-linear, linear-log, log-linear, and log-log. This is because these functional forms are the most widely used in literature and are quite easy to be dealt with by researchers. The functional forms of linear-linear, linear-log, log-linear, and log-log may be expressed as follows.

$$P_i=\alpha +{\gamma }^{\prime }{C}_i+{\beta }^{\prime }{D}_i+e_i$$

(1)

$$P_i=\alpha +{\gamma }^{\prime }\ln C_i+{\beta }^{\prime }{D}_i+e_i$$

(2)

$$\ln P_i=\alpha +{\gamma }^{\prime }{C}_i+{\beta }^{\prime }{D}_i+e_i$$

(3)

$$\ln P_i=\alpha +{\gamma }^{\prime }\ln C_i+{\beta }^{\prime }{D}_i+e_i$$

(4)

where $i$ is the $i$th observation,$P_i$ is the price of $i$th apartment, ln stands for natural logarithm, $\mathsf{\alpha }$ is a constant term, $C$ is a vector of continuous variables affecting apartment price, $\gamma$ is a parameter vector related to $C$, $D$ is a vector of dummy variables, $\beta$ is a parameter related to $D$, and $e$ is a disturbance term.

However, limiting the form of a function to linear or natural logarithm can be too restrictive. For this reason, it is necessary to use a more comprehensive and flexible form of function that includes linear or natural logarithms as a special case. Under this awareness of the problem, the main application in the literature is the Box-Cox transformation model. Therefore, this study also intends to additionally apply a Box-Cox transformation model. This model can be expressed in the form of converting the dependent variable to $F^{\left(\mu \right)}=\frac{F^{\mu }-1}{\mu }$. If $\mu \to 0$, $F^{\left(\mu \right)}$ becomes $\ln F$, and if $\mu \to 1$, $F^{\left(\mu \right)}$ becomes $F$, so it becomes a more flexible function form while covering Equations (1) to (4). Therefore, Equations (5) and (6) are also estimated.

$$F^{\left(\mu \right)}=\alpha +{\gamma }^{\prime }{C}_i+{\beta }^{\prime }{D}_i+e_i$$

(5)

$$F^{\left(\mu \right)}=\alpha +{\gamma }^{\prime }\ln C_i+{\beta }^{\prime }{D}_i+e_i$$

(6)

That is, in this study, a total of six apartment price equations, (1) to (6), are estimated. The one with the smallest prediction error is selected to interpret the analysis results.

2.3. Collection of Data

The area to be analyzed and the data collection method of this study were determined according to four principles. First, the area selected should be where the power plant recognized as a distributed power source is located. Second, the operating period of the distributed power plant in the region must have been more than five years. This is because unless an attribute is sufficiently recognized by a home buyer, it cannot be included in the asset value of a house [51]. Considering that the proportion of power generation of distributed power in South Korea is low at 12% as of 2017, a region where a distributed generation power plant has been in operation for at least five years will be selected.

Third, it is necessary to be able to secure enough observations and variable data representing the apartment characteristics and neighborhood. Theoretically, actual transaction data for apartments should be collected, but sufficient transaction data cannot be obtained due to a significant reduction in demand for apartments caused by the global interest rate hike. Therefore, in this study, it is intended to use apartment sales data to allow for sufficient data to be secured and because this is convenient for collecting characteristic variable data. In addition, statistics from public institutions are used for apartment neighborhood variables. Fourth, credible data should be used. Since there are various databases in which apartment sales information is provided in South Korea, the choice of dataset can be an issue. Therefore, data from the real estate information system provided by Kookmin Bank, a national bank established by the government, are used.

As a result, a city in Gyeonggi-do, a Seoul Metropolitan area with a distributed power plant that has been in operation for more than five years and for which sufficient apartment sales data are available, was selected. As of 2021, the city had 320,000 residents, representing about 0.6% of South Korea’s total population. For a month in July 2022, apartment sales information and characteristic variables were collected from Kookmin Bank’s real estate information system, and neighborhood variables were obtained from 2021 statistical data for the city. The direct distance from the distributed power plant to each apartment was measured directly using a satellite map. The object to be investigated in this research is a combined heat and power plant that produces both heat and electricity, with a capacity of 399 MW and 534 Gcal per hour. The power plant started operations in October 2015, with the goal of supplying heat to a total of 121 thousand households. The fuel used is natural gas. To avoid unnecessary controversy, the name and specific location of the power plant are not disclosed here.

3. Results and Discussion

3.1. Data

A total of 2291 apartment sales data were collected in this study. To investigate the effect of proximity to a distributed power plant on apartment prices, the distance from each apartment to the power plant was investigated and included as a variable. A total of 15 variables affecting apartment prices were selected through a literature survey, consisting of 11 characteristic variables of apartments and four neighborhood variables. The characteristic variables of apartments are Area, Household, Room, Bathroom, Age, Parking, Subway, South, Type, DHS, and Height. The neighborhood variables are Foreigner, Public, Enterprises, and Store. The definition and sample statistics of each variable are presented in

Table 1. Description of variables.

Variables	Description	Mean	Standard Deviation	Expected Sign
Sale	Housing price for sale (unit: hundred million Korean won a)	11.02	3.14
Area	Area of the house (unit: m2)	108.63	19.41	(+)
Household	Total number of households in the apartment complex	898.28	395.25	(+)
Room	Number of rooms in the house	3.13	0.52	(+)
Bathroom	Number of bathrooms in the house	1.90	0.30	(+)
Age	Age of the house	10.00	7.18	(−)
Parking	Number of parking spaces available per household	1.29	0.33	(+)
Foreigner	Number of foreigners in the neighborhood per 10,000 population	72.51	73.47	(−)
Public	Number of public officials in the neighborhood per 10,000 population	4.85	2.98	(+)
Enterprise	Number of enterprises in the neighborhood per 10,000 population	529.04	485.42	(+)
Subway	Walking distance from house to the nearest subway station (unit: km)	1.11	0.56	(−)
Store	Number of distribution stores in the neighborhood per 10,000 population	0.18	0.78	(+)
South	Whether the house faces south (0 = no; 1 = yes)	0.93	0.25	(+)
Type	Structure in front of the house’s entrance (0 = corridor-type; 1 = stairway-type)	0.95	0.22	(+)
DHS	Heating method of the house (0 = individual or central heating system; 1 = district heating system)	0.73	0.44	(+)
Height	Whether the house is on the middle floor or higher (0 = no; 1 = yes)	0.63	0.48	(+)
Plant	The direct distance from house to a power plant (unit: km)	2.58	2.18	(+)

^a At the time of obtaining the data, KRW 1123 was approximately equal to USD 1.

.

Among the apartment characteristic variables adopted in this study, Area, Household, Room, Bathroom, Parking, South, Type, DHS, and Height are expected to have a positive effect on apartment prices. On the other hand, the Age and Subway variables are expected to have an inverse relationship with apartment prices. In the case of neighborhood variables, Public, Enterprises, and Store are expected to have a positive correlation with apartment prices because they contribute to improving the quality of life of residents. Meanwhile, since there is still a negative view of foreigners living in South Korea, the proportion of foreign residents is expected to have a negative correlation with apartment prices. Lastly, the direct distance from the power plant to the apartment, which is the main interest of this study, is expected to have a positive effect on apartment prices.

3.2. Estimation Method

The most widely applied method in estimating Equations (1)–(4) is the least squares (LS) method. In place of the LS method, the maximum likelihood (ML) estimation method can be applied. However, as is widely mentioned in econometric textbooks, estimates of the coefficients to which the two methods are applied are the same, and only the standard error for the estimated coefficients varies. Therefore, in situations where both methods can be applied, it is customary in the literature to apply only the LS that is easier to apply. On the other hand, the least absolute deviations (LAD) method can be adopted as an alternative to the LS method. Therefore, in order to estimate Equations (1)–(4), both the LS and LAD methods are utilized.

The former is a mean regression analysis method and involves assumptions about linearity and distribution [52]. The latter is a median regression analysis method and does not assume the distribution but only assumes linearity. Therefore, since the latter is less affected by outliers than the former, more robust estimation results can be derived [53]. Equations (5) and (6), to which the Box-Cox transformation model is applied, are estimated using the ML estimation method. Finally, LS and LAD are applied to Equations (1)–(4), respectively, and Equations (5) and (6) are analyzed through the ML estimation method, giving a total of 10 estimation results.

This is because Equations (5) and (6) are representative nonlinear models. Both the LS estimation method and the LAD estimation method cannot be applied to the estimation of such a nonlinear model. Instead, the ML estimation method can be easily applied. Since the ML estimation method obtains the value of the parameter that maximizes the likelihood function, the nonlinear function form does not cause any difficulty in obtaining the coefficient estimates.

One of the 10 estimation results must be selected to draw a meaningful conclusion through the estimation results. In general, $Adjusted-R^2$ is widely used for model comparison because it represents a value that quantifies the goodness of fit of a model. That is, the model with the largest $Adjusted-R^2$ value can be selected. However, since the dependent variables of the 10 equations estimated in this study are not all the same, $Adjusted-R^2$ cannot be used as a criterion. Therefore, it is intended to use the root mean square percent error (RMS%E) as a standard, which can be used for comparison between models with different dependent variables. The model with the smallest RMS%E value can be selected as the optimal model. The form of RMS%E is expressed as follows.

$$\mathrm{RMS}\%\mathrm{E}=100\sqrt{\frac{1}{T}{\displaystyle \sum }_{i=1}^T{\left(\frac{\widehat{S_i}-S_i}{S_i}\right)}^2}$$

(7)

where $T$ is the sample size, $S_i$ is the observed value of the dependent variable, and $\widehat{S_i}$ is the fitted value of the dependent variable.

3.3. Estimation Results

The estimation results of the 10 equations are presented in

Table 2. Various housing price equations estimated in this study.

	Least Squares Estimation b				Least Absolute Deviations Estimation b				Box-Cox Transformation b
Variables a	Linear-Linear	Linear-Log	Log-linear	Log-Log	Linear-Linear	Linear-Log	Log-Linear	Log-Log	Linear	Log
Constant	–0.6563 (–1.73)	–23.8521 (–18.54) *	1.1532 (36.67) *	–1.2400 (–12.23) *	–1.8420 (–7.15) *	–17.5856 (–19.37) *	1.1269 (52.42) *	–0.7827 (–10.67) *	−331.1298 (−4.26) *	–35.3609 (–10.63) *
Area	0.0807 (31.87) *	8.6305 (31.59) *	0.0073 (35.07) *	0.7938 (36.88) *	0.0684 (39.87) *	7.6592 (39.73) *	0.0067 (46.65) *	0.7354 (47.22) *	6.4713 (12.92) *	14.8631 (14.02) *
Household	0.0008 (7.80) *	0.5725 (8.17) *	0.0000 (11.81) *	0.0681 (12.32) *	0.0006 (9.28) *	0.4907 (9.92) *	0.0000 (14.97) *	0.0689 (17.24) *	–0.0487 (–2.73) *	0.7233 (5.46) *
Room	0.8114 (10.00) *	1.9044 (7.05) *	0.0665 (9.92) *	0.1727 (8.11) *	0.5228 (9.52) *	1.4008 (7.35) *	0.0396 (8.64) *	0.1286 (8.35) *	76.0997 (7.90) *	2.8155 (5.94) *
Bathroom	–0.6764 (–4.55) *	–2.0808 (–8.97) *	–0.0067 (–0.54)	–0.1072 (–5.86) *	–0.1267 (–1.26)	–0.7504 (–4.58) *	0.0206 (2.46) *	–0.0325 (–2.46) *	–78.7614 (–2.89) *	–3.6805 (–7.14) *
Age	–0.0897 (–8.88) *	–1.1346 (–10.23) *	–0.0090 (–10.75) *	–0.0945 (–10.80) *	–0.0715 (–10.45) *	–0.9159 (–11.70) *	–0.0080 (–14.02) *	–0.0891 (–14.08) *	–10.6551 (–5.01) *	–1.9757 (–6.61) *
Parking	0.8062 (6.36) *	2.1859 (10.99) *	0.0599 (5.71) *	0.1820 (11.61) *	2.7182 (31.69) *	3.7325 (26.60) *	0.1994 (27.84) *	0.2812 (24.80) *	29.6876 (3.40) *	3.9482 (12.19) *
Foreigner	–0.0149 (–6.07) *	–3.6730 (–14.19) *	–0.0020 (–9.68) *	–0.3476 (–17.05) *	–0.0101 (–6.10) *	–3.9191 (–21.46) *	–0.0017 (–11.98) *	–0.3740 (–25.35) *	1.6815 (4.20) *	–8.2091 (–10.41) *
Public	0.1324 (2.84) *	5.0199 (14.76) *	0.0145 (3.75) *	0.4200 (15.68) *	0.0135 (0.43)	5.7041 (23.78) *	0.0099 (3.76) *	0.5015 (25.88) *	–52.0093 (–6.51) *	12.1128 (11.59) *
Enterprises	0.0009 (4.58) *	0.0071 (0.07)	0.0002 (9.93) *	0.0388 (4.88) *	0.0009 (6.49) *	–0.1924 (–2.70) *	0.0001 (12.64) *	0.0160 (2.78) *	–0.0216 (–0.56)	–0.5150 (–2.38) *
Subway	–0.4014 (–5.38) *	–0.9143 (–10.92) *	–0.0385 (–6.24) *	–0.0860 (–13.04) *	–0.5420 (–10.73) *	–1.0504 (–17.78) *	–0.0456 (–10.82) *	–0.0913 (–19.14) *	–55.0533 (–4.40) *	–3.4499 (–12.73) *
Store	–0.3552 (–7.58) *	–0.3750 (–8.20) *	–0.0649 (–16.76) *	–0.0596 (–16.54) *	–0.3150 (–9.94) *	–0.4512 (–13.98) *	–0.0603 (–22.77) *	–0.0721 (–27.66) *	15.2882 (1.82)	–0.7796 (–8.71) *
South	–0.2742 (–2.24) *	–0.2715 (–2.12) *	0.0035 (0.34)	0.0046 (0.46)	–0.1166 (–1.41)	–0.1962 (–2.17) *	–0.0015 (–0.21)	0.0064 (0.88)	–40.0007 (–2.00) *	–0.6015 (–2.79) *
Type	–0.9452 (–5.57) *	–0.7062 (–3.89) *	–0.0181 (–1.29)	0.0075 (0.53)	–0.6807 (–5.92) *	–1.4939 (–11.66) *	–0.0125 (–1.31)	–0.0446 (–4.30) *	–129.2213 (–2.78) *	–1.4013 (–3.12) *
DHS	1.5478 (9.98) *	1.0532 (6.15) *	0.1289 (10.05) *	0.1086 (8.05) *	1.3059 (12.44) *	0.8894 (7.36) *	0.0964 (10.99) *	0.0738 (7.56) *	130.0102 (3.59) *	2.2672 (5.40) *
Height	0.3378 (5.82) *	0.3574 (5.88) *	0.0259 (5.40) *	0.0273 (5.69) *	0.3000 (7.64) *	0.3195 (7.45) *	0.0264 (8.05) *	0.0280 (8.09) *	30.0676 (2.90) *	0.5869 (4.50) *
Plant	0.4200 (19.29) *	0.2583 (3.52) *	0.0307 (17.03) *	0.0194 (3.36) *	0.4365 (29.60) *	–0.0483 (–0.93)	0.0314 (25.47) *	–0.0067 (–1.62)	89.1558 (15.03) *	0.3353 (1.96) *
$\mu$									2.7668 (143.33) *	1.2144 (51.61) *
Adjusted R2 c	0.8199	0.8024	0.8640	0.8645	0.7962	0.7884	0.8482	0.8570	0.1888	0.3893
Wald statistics (p-values)	371.91 (0.00)	12.37 (0.00)	290.13 (0.00)	11.28 (0.00)	876.34 (0.00)	0.87 (0.35)	648.94 (0.00)	2.63 (0.10)	575.61 (0.00)	3.98 (0.05)
RMS%E d	12.71	15.40	9.82	7.77	13.33	16.25	10.68	8.25	1608.36	56.23
Sample size	2291	2291	2291	2291	2291	2291	2291	2291	2291	2291

^a They are defined in Table 1. ^b The t-values are presented in the parentheses below the estimates and * denotes that the estimate secures statistical significance at the 5% level. ^c The values for Box-Cox transformation indicate McFadden’s peusdo-R². ^d RMS%E denotes the root mean square percent error.

. In the course of the estimation, an econometric software of TSP 5.1 was used. Usually, which econometric software is used has no effect on the estimation results. The estimation result of Equation (1) shows little difference between the LS estimation method and the LAD estimation method in terms of the sign and size of the estimated coefficients. In other words, for the data used in this study, the application of the two estimation methods with different background philosophies for obtaining estimates does not result in significant differences. Similar reasoning is possible even by examining the estimation results of Equations (2)–(4). Of course, differences definitely exist in the information on RMS%E. Since the dependent variables in Equations (5) and (6) are different from each other, it is impossible to directly compare the estimation results of the two equations. In addition, for the same reason, the estimation results of Equations (5) and (6) cannot be compared with the estimation results of Equations (1)–(4).

Looking at the Wald statistics, 7 out of 10 estimation results have p-values less than 0.01. This means that the null hypothesis that the model is meaningless can be rejected at the significance level of 1%. On the other hand, Wald statistics of the estimation results of the linear-log model and log-log model to which LAD is applied and the Box-Cox transformation log model show that the null hypothesis cannot be rejected at the significance level of 1%. Therefore, these three estimation results are excluded from the comparison to select the optimal model. That is, the RMS%E values of the other seven models are compared with each other to select the model with the smallest value. As a result, the log-log model estimated with the LS method, which had the smallest value of RMS%E of 7.77, was selected as the optimal model. Thus, the subsequent interpretations are based on the log-log model estimated with the LS method.

The estimated coefficients for all variables except South and Type were statistically significant. The signs of the statistically significant estimated coefficients appeared to be generally the same, as expected, but the estimated coefficients of the Bathroom and Store variables were unexpectedly derived as negative signs. The reason why the estimated coefficient for the Bathroom variable was negative is that since the number of family members in South Korean households has been recently decreasing and the number of single-person households has been increasing, it seems that a larger living room area or higher number of rooms is preferred over a larger number of bathrooms. Regarding the estimated coefficient of the Store variable, online shopping has become popular in South Korea and the delivery period for purchased items is usually about three days, so the lack of a distribution store near a person’s residence may not cause much inconvenience. Rather, large retailers can cause traffic congestion in nearby areas. For this reason, it is judged that the estimated coefficient for the Store variable was negative, contrary to expectations.

The meanings of the signs of the estimated coefficients of other variables are as follows: the larger the area of the apartment, the greater the number of households in each complex; the greater the number of rooms, the more recent the year of occupancy; the greater the number of parking spaces per household, the closer the apartment to a subway station on foot; the lower the ratio of foreigner residents per 10,000 people, the higher the ratio of public servant residents per 10,000 people; the more companies nearby, the more use of the district heating system as the heating method; and the higher the floor of the apartment, the higher the apartment price. The direct distance from the apartment to a power plant, which is the main focus of this study, was found to have a positive correlation with the apartment price, and this was statistically significant at the significance level of 5%. This means that a power plant is a non-preferred facility from the perspective of local residents.

3.4. Discussion of the Results

Using the 10 housing price equations estimated above, it is possible to quantitatively calculate the effect of the direct distance from the power plant to the apartment on the apartment price. In other words, the marginal implicit price, which is the difference in apartment price per 1 km of direct distance, is derived. When Equation (1) is used, the difference in apartment prices per kilometer of direct distance can be easily obtained by differentiating the dependent variable from the Plant variable and is calculated as follows:

$$\mathrm{Marginal}\mathrm{implicit}\mathrm{price}={\widehat{\gamma }}_{Plant}=\frac{\partial P}{\partial C_{Plant}}$$

(8)

However, if the natural logarithm is taken for the dependent variable or the natural logarithm is taken for the independent variable, the calculation becomes somewhat complicated. In the case of using Equations (2)–(4), the difference in apartment prices is derived from Equations (9)–(11), respectively. $\overline{Plant}$ and $\overline{P}$ are the sample mean of the value of the direct distance from the power plant to the apartment and the apartment price, respectively.

$$\mathrm{Marginal}\mathrm{implicit}\mathrm{price}=\frac{\partial P}{\partial \left(\ln C_{Plant}\right)}\times \frac{1}{\overline{Plant}}$$

(9)

$$\mathrm{Marginal}\mathrm{implicit}\mathrm{price}=\frac{\partial \left(\ln P\right)}{\partial C_{Plant}}\times \overline{P}$$

(10)

$$\mathrm{Marginal}\mathrm{implicit}\mathrm{price}=\frac{\partial \left(\ln P\right)}{\partial \left(\ln C_{Plant}\right)}\times \frac{\overline{P}}{\overline{Plant}}$$

(11)

Equations for calculating the marginal implicit price using the estimation results of Equations (5) and (6) to which the Box-Cox transformation model is applied are derived as Equations (12) and (13), respectively.

$$\mathrm{Marginal}\mathrm{implicit}\mathrm{price}=\frac{\partial F^{\left(\mu \right)}}{\partial C_{Plant}}\times {\overline{P}}^{\left(1-\mu \right)}$$

(12)

$$\mathrm{Marginal}\mathrm{implicit}\mathrm{price}=\frac{\partial F^{\left(\mu \right)}}{\partial \left(\ln C_{Plant}\right)}\times \frac{{\overline{P}}^{\left(1-\mu \right)}}{\overline{Plant}}$$

(13)

The results of analyzing the difference in apartment prices per 1 km of direct distance from the power plant to the apartment using Equations (8)–(13) are presented in

Table 3. Changes in housing prices caused by the direct distance from a power plant to the house.

	The Marginal Implicit Price Difference a	t-Values
Least squares estimation:
Linear-linear	KRW 42 million (USD 37.4 thousand)	19.29 *
Linear-log	KRW 10 million (USD 8.9 thousand)	3.52 *
Log-linear	KRW 34 million (USD 30.3 thousand)	17.03 *
Log-Log	KRW 8 million (USD 7.1 thousand)	3.36 *
Least absolute deviations estimation:
Linear-linear	KRW 44 million (USD 39.2 thousand)	29.60 *
Linear-log	KRW −2 million (USD −1.8 thousand)	−0.93
Log-linear	KRW 35 million (USD 31.2 thousand)	25.47 *
Log-Log	KRW −3 million (USD −2.7 thousand)	−1.62
Box-Cox transformation:
Linear	KRW 129 million (USD 115.0 thousand)	23.99 *
Log	KRW 8 million (USD 7.1 thousand)	1.99 *

^a At the time of obtaining the data, KRW 1123 was approximately equal to USD 1. * denotes that the estimate secures statistical significance at the 5% level.

. Since the estimation results are different for each model, among the eight statistically significant marginal price difference estimation results at the significance level of 5%, the value resulting from the log-log model estimated with the LS method with the smallest RMS%E is selected. The difference in apartment price per kilometer of direct distance derived from the model is KRW 8 million (USD 7.1 thousand). Since the average price of apartments in the area selected in this study is KRW 1102 million (USD 0.98 million), the price difference is about 0.7% of the average price.

Previous studies that applied the hedonic approach to analyze the effect of proximity to power plants on housing prices and this study are different in four aspects: evaluation target, sample size, study time point, and model. Therefore, it is difficult to directly compare the respective results. However, in two respects, the results of this study can be judged similarly to those of previous studies. First, power plant facilities have a negative impact on housing prices. Second, the level of decline in housing prices reported in previous studies is about 0.2% to 3% [24,25,26]. As a result of this study, the level of decline in the average apartment price was about 0.7%, which is not much different from the results of previous studies.

Of course, it is difficult to generalize the results because this study only investigated a specific area in Gyeonggi, but it is significant that the proximity to a power plant has a negative effect on apartment prices. As mentioned in the introduction, the government intends to expand distributed power generation. To this end, the government has enacted and implemented support for areas around power plants by law to improve public acceptance of power plant construction. The results of this study can be used as a useful reference for comparing with the size of government subsidies. In the current government support policy, a radius of 5 km from a power plant is designated as a neighboring area, and KRW 100 (USD 0.09) per MWh based on the amount of power generation and 1.5% of the construction cost are paid as subsidies during the power plant construction period and operation period, respectively.

4. Conclusions

The South Korean government plans to increase the share of distributed power generation from 12% in 2017 to 30% by 2040. In order to expand distributed power generation, it is important to secure the acceptance of local residents because power plants must be built near to residences. Therefore, the government and power plant construction companies are trying to secure the acceptance of local residents in various ways. For example, they build facilities such as gymnasiums and senior citizen centers for local residents or provide subsidies. In addition, recently, methods of designing the exterior of a power plant building like a work of art or drawing murals on the walls of the building are also being promoted. Nevertheless, due to opposition from local residents, construction of power plants is frequently delayed or stopped. Therefore, the government and business operators need objective information to persuade and negotiate with local residents.

In response, this study quantitatively estimated the effect of proximity to power plants on apartment prices, which is one of the main reasons why local residents oppose power plant construction. As a result, it was found that proximity to the power plant had a negative effect on apartment prices, and the size of the effect was analyzed as KRW 8 million (USD 7.1 thousand) per kilometer.

This value can be used as useful information when determining the size of the subsidy to be paid to residents near the plant. As mentioned earlier, the government provides subsidies to areas within a 5 km radius of the power plant. For example, the number of subsidies based on the 500 MW power plant, which is a distributed power source, is as follows. Assuming the total construction cost of a 500 MW power plant is KRW 600 billion (USD 534.3 million), KRW 9 billion (USD 8.0 million), 1.5% of the total construction cost will be paid as subsidies during the construction period. In addition, assuming that the utilization rate of a 500 MW power plant is 60%, the annual power generation is 2628 GWh, and the subsidy is KRW 100 per MWh, so the annual subsidy during the operation period is KRW 263 million (USD 234 thousand). If this value is paid annually for 30 years, the lifetime of the plant, the total subsidy during the operating period will reach KRW 7884 million (USD 7020 thousand).

A total of 2021 apartments are located within a radius of 5 km from the power plant, which is the subject of evaluation in this study. Applying the KRW 8 million (USD 7.1 thousand) per kilometer reduction in apartment prices derived from this study, the total asset value reduction is calculated as KRW 30.2 billion (USD 26.9 million). As a result, the government’s subsidy was found to be lower than the decrease in apartment prices that residents had to endure due to the construction of a power plant near their residence. Therefore, residents may believe that the government does not reasonably compensate for the loss of property caused by construction of the power plant. Thus, in order to expand distributed generation, the government should realize the scale of compensation for the area around power plants.

This study contributes to both policy and research aspects. From a policy point of view, the results of this study can be used as useful reference data when the government pays subsidies to residents near power plants. The results can also be used as data to mediate conflicts between power generation operators and local residents. From a research perspective, this study contributes to the literature as the first case of quantitative analysis of the effect of proximity to a power plant on apartment prices in South Korea. However, this study has three limitations and needs to be further supplemented through future studies. The first limitation is that the study used housing sales data rather than actual housing transaction data. This is because sufficient transaction data cannot be obtained in South Korea as strong policies to curb fluctuations in real estate prices, such as raising housing ownership taxes and strengthening mortgage loan regulations, have been recently implemented. If housing transactions are further activated in the future, it will be possible to secure meaningful results by conducting additional analyses using actual transaction data. Second, among various distributed power sources, only power plants were selected for analysis. There are several types of distributed power sources such as fuel cells, solar power, wind power, and combined heat and power generation. In addition, since the amounts and risks of pollutants emitted by power sources differ, local residents may show different levels of rejection depending on the type of distributed power source. Therefore, if the effect on apartment prices for each distributed power plant is analyzed in the future, more significant implications can be obtained.

Third, the effect of changes in the appearance of power plants, which have been actively considered by operators, on the acceptance of local residents has not been evaluated. Recently, a plan to improve the exterior of the power plant is being promoted to secure resident acceptance. For example, the wall of the power plant site is used as an exhibition space for cultural and artistic activities, and even the lotus stones, an essential facility of the power plant, are decorated like medieval castles to reduce residents’ reluctance to the power plant. In addition, the boundary of the power plant site may be created as a landscaping fence rather than a general iron fence to create an eco-friendly and open appearance. Therefore, analyzing whether such improvements in the appearance of a power plant have the effect of preventing a drop in prices of nearby apartments can be considered as a follow-up study.