Impact of ground source heat pumps on house sales prices in Finland

Buildings contribute to approximately 28% of global energy-related emissions. Heat pumps are a key technology for decarbonising the heating emissions of buildings. This study focuses on ground source heat pumps (GSHP), which are increasingly used in colder regions. Since, for an average home, the capital expenditure of GSHP can be an order of magnitude higher than that of traditional heating, it is important to understand whether GSHP has an impact on house transaction prices. A hedonic price model was constructed to estimate the sales prices of detached houses, where heating type is the main variable of interest. The hedonic analysis revealed that for detached houses, GSHP had a statistically significant positive impact of 5.33% on house sales prices. Further analysis puts the premium in the context of housing prices in different locations in Finland. An average house in the Helsinki Metropolitan Area (HMA) could cover the required capital expenditure of a GSHP system with the sales price premium, whereas in other areas in Finland, 5 years of energy savings are required on top of the premium. Hence, in locations with lower housing prices, the house must be owned for a longer period to recoup the investment costs. This is important to understand when national energy aid policies are planned to accelerate investments in heat pumps.


Introduction
Buildings contribute to approximately 28% of global energy-related emissions (IEA, 2020a). In the USA, 20% of the nation's greenhouse gas emissions come from heating, cooling and powering households (Goldstein et al., 2020). In Germany, private households' energy consumption causes approximately 31% of Germany's total CO2 emissions (Destatis, 2021;Ritchie & Roser, 2020). In the UK, heating in the residential sector accounts for approximately 21% of CO2 emissions (National Statistics, 2020a, b). In Finland, building heating and electricity consumption caused 36% of all emissions, of which roughly twothirds were from heating (SYKE, 2020).
Heat pumps are a key technology for decarbonising the heating sector. Currently, only 5% of global heat is delivered with heat pumps, but this may triple by 2030, and eventually, heat pumps could cover 90% of the global space and water heating needs for low emissions (IEA, 2020b). Ram et al. (2019) calculated that by 2050, over €7 trillion will be invested into individual heat pumps that heat or cool buildings. Currently, the European Commission (2022) and its member countries are planning major subsidy policies to encourage the adoption of heat pumps to displace natural gas for heating.
Heat pumps use electricity to draw energy from the surrounding air, water or ground (Staffell et al., 2012). They are highly energy efficient, as one unit of electricity can deliver many units of heating or cooling. Heat pumps can reduce emissions by over 90% compared to fossil fuel heaters if the electricity system powering them has low emissions (Vimpari, 2021). Although they have higher initial investment costs than traditional fossil-fuel-based heaters, heat pumps have lower operating costs due to their high efficiency in energy conversion. This high capital expenditure can hinder investment, even if high economic and environmental benefits are proven.
One important practical question for households is whether the upfront investment is reflected in house sales prices. The lifecycles of these investments can be up to 30 years, with payback periods ranging from 5 to 15 years, whereas homes are switched in shorter periods. A homeowner might wonder whether the annual energy savings are enough to cover the initial investment cost. If the heating system has a positive impact on sales prices, the homeowner does not have to rely only on energy savings to recoup upfront costs. This topic is related to a growing body of literature that has measured the effects of different energy efficiency measures' impacts on housing prices using statistical methods. Previous research suggests that decreased energy costs are among the key reasons for price premiums for energy-efficient homes (e.g. Dastrup et al., 2012;Shen et al., 2021).
This paper adds to the literature by examining the effect of GSHP, a particular type of heat pumps, 1 on house transaction prices in Finland. A dataset of 19,008 transactions in eight large cities in Finland between 1999 and 2018 was used. For detached houses, four heating types dominate in these cities: direct electricity, district heating, oil and ground source heat pump (GSHP), with an approximate market share of 52%, 21%, 20% and 7%, respectively. Traditionally, district heating has been very costeffective, but tightening environmental regulations and the cost-effectiveness of heat pumps have increased the competition for district heating companies. Figure 1 presents the average historical heating prices for these four heating types for the period 1998-2019, as well as housing price development in the Helsinki Metropolitan Area (HMA) and the rest of Finland (Statistics Finland, 2021a, b). For GSHP, an average coefficient of performance (COP) of three was used, i.e. one unit of electricity is required to produce three units of heating (Vimpari, 2021). Oil heaters are assumed to have an average efficiency of 80%, as older boilers may have an efficiency between 60 and 70% and modern boilers up to 95%.
According to Fig. 1, the oil price increased by 340%, electricity by 131% and district heating by 142%, whereas house prices in the Helsinki Metropolitan Area increased by 119% and in the rest of Finland by 60%. District heating was cheaper than oil and electricity, but GSHP was the most costefficient heating type. However, GSHP requires a high upfront investment compared to the other heating methods. For example, in the UK, for a detached house requiring 10 kW of heating capacity, approximately €1 600 must be invested for a direct electricity-based system, €2300 for a gas-or oil-based system, and €13,700 for a GSHP-based system (Scottish Government, 2021).
Given that GSHP requires a high upfront investment and significantly reduces heating costs, the following main hypothesis was set: There is a sales price premium for GSHP-heated houses compared to other heating systems. This was tested by constructing a hedonic price model.

Previous literature
There is a vast body of literature that utilised hedonic regression to estimate how different variables explain housing prices. Within the field of this study, i.e. how energy-related characteristics impact housing prices, this methodology has been used within the context of energy efficiency ratings, rooftop photovoltaics, solar heaters and, most recently, air source heat pumps, Deng et al. (2012) examined whether high energy efficiency, as measured with the Green Mark label, impacts the sales prices of apartments in Singapore. The label was found to command a 4 to 6% premium. Kahn and Kok (2014) investigated how green labels 1 3 Vol.: (0123456789) such as Energy Star or LEED affect housing transaction prices. Their results suggested a 2 to 4% premium for energy-efficient homes, depending on location and building characteristics. Walls et al. (2017) examined house transactions in the USA. Spatial matching, propensity score matching and regression analysis found a 2% premium for Energy Star in Portland, Oregon, but no premium in Austin, Texas. Local certificates had larger premiums (4% and 9%, respectively), but these certificates often represended more qualities than energy-related improvements. Cerin et al. (2014) investigated the impact of the European Energy Performance Certificate (EPC) on housing prices in Sweden. They found that energy performance was not always rewarded in terms of price, depending on building age and price class. Fuerst et al. (2015) investigated whether a highenergy performance rating, as measured by the European energy performance rating (EPC), has an impact on housing prices in the UK. The findings suggest that an A/B rating commands a 5.0% premium and a C rating commands a 1.8% premium compared to the holdout rating of D. Similarly, Fuerst et al. (2016) examined whether high-energy performance ratings commanded a price premium for apartment transactions in Finland. A premium of 3.3% was found for the top three energy performance categories, which dropped to 1.5% when a set of neighbourhood characteristics was added to the model. In contrast, Yoshida and Sugiura (2015) analysed the transaction prices of green buildings in Tokyo. Their model suggests that a green building with renewable energy and recycled materials can result in a price discount due to higher lifecycle costs for the user. Dastrup et al. (2012) examined the impact of rooftop photovoltaics (PV) on house transaction prices in California. They noted that the value was generated through energy savings and communicating that a home is green. A hedonic pricing model, as well as a repeat sales approach, found a 3.6% premium for homes with PVs. Similarly, Hoen et al. (2013) examined the impact of PV on house prices in California. A hedonic pricing model revealed a 3.6% price premium with a 1% significance. The price premium was found to be slightly higher than the system's upfront costs. Qiu et al. (2017) examined the impact of PV and solar heaters on residential home values in Arizona. They employed semi-parametric, non-parametric and hedonic regression to test whether a treatment group with solar systems has a price premium compared to a control group. The study found a 17% premium on sales prices for properties with installed PV; no premium was found for solar heaters. The percentage premium is rather large due to low housing and land prices in Arizona. Shen et al. (2021) estimated how the installation of an air source heat pump impacted house prices in the USA. They used different methods and found a premium of between 4.3 and 7.1% for heat pump transactions. The results also showed that the premium was higher in regions with more environmentally conscious and middle-class people, as well as in regions with a milder climate.

Methodology
In hedonic price regression, a specific set of characteristics is used to form an equilibrium that defines the price of goods (Rosen, 1974). The common denominator in the previous literature is that first, a model containing a set of temporal, locational and building characteristics is used to form an equilibrium model, which is then enriched by a set of energy-related characteristics. Depending on the type of characteristics, these are set as continuous variables or categorical variables. Continuous variables, such as the dependent variable, price, are often defined in levels or in its natural logarithm transformation, which often may increase the explanatory power of the equation. Categorical variables are used to analyse the impact of a specific categorical (dummy) characteristic on price.
In this study, the following equation was used to estimate the price of a dwelling: where ln(P) is the logarithmic for the price of transaction i at year t and month m, 0 is the intercept, B j is a vector of j building variables, including heating type, D t is a vector of t temporal variables, S l is a vector (1) of l locational variables and ϵ is the error term of the model.
This methodology was used to analyse whether having GSHP as a heating type increases transaction price of a detached house. This was studied with two different models, i.e. including one of the following variables in the building variables: 1) GSHP (true/false) 2) Heating type (categorical): direct electricity, district heating, GSHP or oil (hold-out: direct electricity) Given that GSHP has the highest capital expenditure and is the cheapest form of heating, it should have a price premium against other heating types. The second model was used to test and estimate how different heating types perform individually against direct electricity, which is the most expensive heating type, as well as the most used for detached houses in Finland. Python Statsmodels (2022) was used to conduct the analysis, with ordinary least squares (OLS) as the method.

Data and descriptive statistics
The main dataset was a housing transaction database collected by the Central Federation of Finnish Real Estate Agencies (KVKL). It includes nearly two million transactions between 1999 and 2018 in Finland (KVKL, 2019), also including other building types, such as apartments and semi-detached houses, which are almost always run by housing cooperations in Finland. This data often does not include the heating type of the building. However, cities' building departments maintain a technical building database that includes this information. Eight large Finnish cities provided this data, which was merged with the transaction data by using the exact street addresses of buildings within both databases (Building data, 2020). To ensure exact matching, the street addresses in the housing transaction database were cleaned using the Levenshtein distance method, which was used for comparing (and correcting) addresses with the official street addresses used in the cities' building databases. Levenshtein (1966) is a method that calculates distances between words and can be used to clean data. Furthermore, Statistics Finland collects socioeconomic data in a raster database (raster size 250 m × 250 m) in Finland (Statistics Finland, 2021c). This data were added to the database based on the nearest publication year ( Table 1).
The remaining data were then cleaned for outliers (above and below three standard deviations) based on floor area (sqm) and unencumbered transaction price (€). Cleaning was done separately for each city, as there are major differences in housing prices between the cities. Additional cleaning was done by removing transactions with the 'unknown' condition. Finally, new developments were also removed, as this study wanted to focus on the transactions of existing buildings. The final dataset included 19,008 transactions of detached houses in eight cities (Helsinki, Espoo, Vantaa, Turku, Tampere, Lahti, Kuopio and Oulu), home to over two million inhabitants. Tables 2 and 3 provide descriptive statistics of the data used.
In the tables, the transaction prices were adjusted for 2020 using housing price index data available for the Helsinki Metropolitan Area and the rest of Finland (Statistics Finland, 2021b). However, for the hedonic regression model, this adjustment was not made because the temporal variables should capture the market conditions over time. Direct electricity dominates with a 52% share as a heating source, while district heating has a 20% share, oil 21% and GSHP 7%. Houses with GSHP are clearly higher valued than the average, and oil is the opposite. However, this difference fades away when looking at the prices at the city level, where GSHP prices are quite close to the mean prices of all the buildings. For other characteristics, some differences can be identified. Oil-heated houses are older and their condition is not as good as others. Figure 2 shows how different types of heating have evolved over time. Direct electricity and district heating have had a rather stable share of total house transactions, with shares of 50% and 20%, respectively. Oil's market share in the transaction seems to be decreasing, while the share of GSHP has increased over the last few years.

Hedonic regression results
The estimation results are presented in Table 4. In the first column, GSHP was tested against other heating types and in the second column separately against each heating type. The adjusted R 2 numbers of 0.851 indicate high performance for both models.
The building characteristic variables worked as expected. A larger floor area and better conditions increase the transaction price, whereas older buildings, longer distance from the CBD and leasehold decrease the transaction price. Our variable of interest, GSHP as a heating type, increases the transaction  price of detached houses by 5.33% 2 (4.08 to 6.61%, with a 95% confidence interval) compared to other heating types. This finding is statistically significant at the 1% level. Thus, the main hypothesis was supported. This was further tested in the second column by inspecting each heating system individually against direct electricity. Again, GSHP commands a price premium of 4.85% (p < 0.01), while district heating commands a lower price premium of 1.27% (p < 0.05). Oil decreases the transaction price by 2.31% (p < 0.01). These secondary model's findings were somewhat aligned with the heating costs presented in Fig. 1: GSHP is clearly less expensive than direct electricity, and district heating is also less expensive. On the other hand, oil is approximately on the same level as direct electricity, but its price has high volatility, requires effort from the owner to refill the oil boilers and produces local pollutants. These could be the reasons behind the negative price premium. Table 5 provides robustness testing by first stepwise increasing the characteristics and then analysing how removing both age and/or condition changed the models. It is known that both can have a major effect on heating costs, as older buildings that are in a bad condition have lower energy efficiency. When a location is controlled, the GSHP premium drops dramatically. This is expected because in the dataset, GSHP houses had a larger market share in cities with higher prices (see Tables 2 and 3). Adding the lower-level, time-varying neighbourhood characteristics did not have a significant effect on either the performance or the estimates. When building characteristics were added, the performance increased by approximately 0.22 in both regression models, but the price premiums also decreased. This change was then further analysed by excluding first age, then condition and then both from the final model. The performance remained high, but interestingly, these exclusions decreased the price premium. This suggests that GSHP houses did not have some unseen conditional effect that accounted for the estimated GSHP price premium.
Further analysis was conducted by estimating how the premiums had developed over time and in different cities. Two regression models are created with interaction variables, see Table 6.
There positive premiums have been quite consistent since 2010, when the heating costs also started to increase more rapidly than housing prices, as presented in Fig. 1. However, the positive housing price developments were fuelled by higher loans and very low interest rates. Since heating costs are paid by available income rather than the debt that is used for buying the house itself, the increasing trend in premiums could be linked to the ratio between available income and energy costs. Figure 3 presents the indexed development of electricity prices and district heating (oil is excluded given the high volatility), as well as wages and salaries in Finland (Statistics Finland, 2021b, d). Up until the financial crisis in 2007-2009, heating costs and wages followed each other. However, since the financial crisis, there has been a clear and growing gap between these indexes. Regarding the municipality interaction variable, 6 municipalities have a statistically significant positive price premium with quite a bit of volatility between the size of the premium.

Economic return of GSHP in detached houses
The total economic and environmental performance of GSHP was estimated for an average detached house in Finland. At the end of 2020, the average selling price of an old, detached house in HMA was 3 369 €/sqm and in the rest of Finland, it was 1 497 €/sqm (Statistics Finland, 2021b). Based on  Vimpari (2021), the actual energy consumption of these buildings was estimated based on the floor area, construction year and location. For the rest of Finland, the city of Kuopio (located in the centre of Finland) was used as a reference point, as the climate is colder and heating requirements are higher compared to HMA, which is in the south. Oil was used as the current heating system. Table 7 presents details regarding current heating costs and emissions, as well as the numbers of whether GSHP is the heating system. Additionally, some details of GSHP are provided, together with key financial parameters, such as the payback period and the internal rate of return (IRR). The heating costs and emissions are higher for the building in the rest of Finland, even though they are 21% smaller. The locations in a colder region, as well as an older construction year with worse insulation, are the reasons for this. However, savings from GSHP are larger for the building in HMA, a key reason being that the coefficient of performance for GSHP is higher in southern Finland and electricity (distribution) prices are lower.
Heating costs and emissions in HMA are reduced by approximately 71% and 94%, respectively, compared to the buildings in the rest of Finland, with 58% and 93%, respectively. In both cases, the payback period for the investment was approximately 10 years without considering the potential sales price premium. The lifecycle investment performance (IRR) was slightly better for the building in HMA. Finally, the number of years of energy savings on top of the price premium required to cover the capital expenditure of GSHP was calculated. In HMA, 0 years of energy savings plus the price premium cover the investment costs, whereas 5 years are needed in the rest of Finland. This highlights the relationship between the investment costs of a GSHP system and housing prices. The relative investment