Exploring the impact of renovation subsidies on housing markets – evidence from the Swedish property market

Abstract

Purpose

The primary aim of this research is to examine the effects of the Renovation, Conversion, and Extension (ROT) tax deduction for renovations on the scope and quality of renovations and its subsequent impact on house prices across various Swedish municipalities.

Design/methodology/approach

This study utilises a two-way fixed effect instrument variable (IV) spatial Manski approach, analysing balanced panel data from 2004 to 2020 at the municipal level (290 municipalities) in Sweden. The methodology is designed to assess the impact of the ROT subsidy on the housing market.

Findings

The study reveals that the ROT subsidy has significantly influenced house prices, with noticeable variations between municipalities. These differences are attributed to the varying amounts of tax reductions for renovations and the extent to which property owners utilise these subsidies.

Research limitations/implications

The research is limited to the context of Sweden and may not be generalisable to other countries with different housing and subsidy policies. The findings are crucial for understanding the specific impacts of government subsidies on the housing market within this context.

Practical implications

For policymakers and stakeholders in the housing market, this study highlights the tangible effects of renovation subsidies on property values. It provides insights into how such financial incentives can shape the housing market dynamics.

Social implications

The research underscores the role of government policies in potentially influencing equitable access to housing. It suggests that subsidies like ROT can have broader social implications, including the distribution of housing benefits among different income groups and regions.

Originality/value

This study contributes original insights into the field of applied real estate economics by quantitatively analysing the impact of a specific government subsidy on the housing market. It offers a unique perspective on how fiscal policies can affect property values and renovation activities at the municipal level in Sweden.

Citation

Wilhelmsson, M. and Warsame, A. (2024), "Exploring the impact of renovation subsidies on housing markets – evidence from the Swedish property market", Journal of European Real Estate Research, Vol. 17 No. 3, pp. 412-430. https://doi.org/10.1108/JERER-01-2024-0001

1. Introduction

Housing properties are considered long-term investments and durable assets but are not immune to depreciation. As a result, they eventually require maintenance and additional expenditures to maintain their quality standards and expected market values. Many studies have examined the implications of these unique housing characteristics on housing markets and prices, quality standards, depreciation, and maintenance, as well as related government housing policies (Arnott et al., 1983; Chinloy, 1980; Lowry, 1960; Myers, 1984; Sweeney, 1974). In the study of housing economics (DiPasquale and Wheaton, 1994; Kenny, 1999; Wilhelmsson, 2008a), the interplay between housing prices and renovation costs has garnered considerable interest (Gyourko and Saiz, 2004; Mamre and Sommervoll, 2022; Mayer, 1981; Mendelsohn, 1977; Munneke and Womack, 2015; Potepan, 1989; Schuetz, 2020; Wilhelmsson, 2008b).

To explain housing price fluctuations, traditional models typically focus on interest rates, supply, and demand. Still, they often overlook renovation costs, significantly impacting a homeowner's decision to sell or upgrade (Gyourko and Saiz, 2004). Renovation costs are influenced by past housing prices, current interest rates, and market demand (Steiner, 2010). Housing prices are determined by market demand and supply, which are affected by the quality and extent of property renovations (Gyourko and Saiz, 2004).

The relationship between property prices and renovation costs becomes even more complex when government incentives for home improvements are considered (Studer and Rieder, 2019). Such subsidies often promote property improvements, energy efficiency, and historical preservation. According to Studer and Rieder (2019), these subsidies can potentially enhance the financial attractiveness of renovation projects by reducing homeowners' out-of-pocket expenses. This, in turn, could lead to an increase in the general demand for renovations. Furthermore, subsidisation of these improvements could improve the overall quality of the housing stock, exerting a positive influence on the upward trajectory of housing prices.

In Sweden, a wide range of housing policies have been implemented over the years to promote well-functioning housing markets in terms of availability and affordability and maintain tenure neutrality. The government established guidelines for housing policy that emphasise three aspects of tenure naturality: equal standards, costs, and influence over housing management, maintenance, and modernisation (Lundqvist, 1987). Different housing subsidies and allowances were implemented to enhance housing demand and construction (Turner and Whitehead, 2002). A system of subsidising renovations of housing called ROT – an acronym for the Swedish words for Renovation, Refurbishment, and Extension was introduced in 1984 and is still in operation today (Tunefalk and Legnér, 2019)—the subsidies aimed at modernising the building stock in relation to living standards and energy efficiency.

Given this context, two research questions (RQ1 and RQ2) can be examined to offer a more complete view of housing market dynamics and to capture the bidirectional relationship between housing prices and renovation costs.

RQ1.

Do renovation costs get reflected in housing prices? This question examines if increased subsidised renovation spending raises the market value of houses.

RQ2.

Do housing prices influence renovation decisions? This question investigates whether higher housing prices encourage homeowners to invest more in renovations when subsidies are available, aiming to increase their property's value further.

The paper is organised as follows. Section two covers literature dealing with issues related to housing quality standards and housing markets, such as depreciation, maintenance, and house prices. Section 3 provides the case study and descriptive statistics of the available data. The appropriate methods used in the article are explained in Section 4. Section 5 contains the empirical method used in the study, while section 6 covers detailed information on the results of the econometric analysis and discussions. The last two sections address policy implications and conclusions.

2. Housing price and renovations

2.1 Housing price

The relationship between house prices and economic activity is affected by income levels, housing stock, demographic patterns, credit availability, interest rates, institutional changes such as land supply and land use planning, and the appreciation of housing values over time (Muellbauer and Murphy, 2008). Bourassa and Hendershott (1995) posit that the main determinants impacting residential property values are the pace of expansion in real wage earnings, driven by job expansion, and population growth due to net international migration.

DiPasquale and Wheaton (1994) developed a structural model, known as a stock-flow model, to explain housing prices for single-family homes. Based on macroeconomic fundamentals, this model focuses on predicting future housing prices and serves as an explanatory model. The model assumes that demand equals supply in the housing market, with changes in supply determined by new production, factor prices, housing prices, and the depreciation of the housing stock. Demand is influenced by housing prices, user cost, rent in rental stock, and several exogenously determined variables. Their theoretical model has informed many empirical macroeconomic housing models.

In the realm of housing markets, the establishment of long-term equilibrium is contingent on the interplay between interest payments and household income rather than the association between property values and income (de Vries and Boelhouwer, 2009). Short-term factors, such as income and mortgage payments, influence the OECD's primary determinants of house prices. In the UK, house prices have tracked actual income since the 1940s, with demographic trends and building society financing causing disequilibrium (Holly and Jones, 1997). In contrast, long-term factors, such as acquisition costs, influence house prices (Madsen, 2012).

Regional housing prices consistently respond to national economic factors like mortgage rates. However, local factors influence home prices, including population movements, job patterns, and income trends (Reichert, 1990). The impact of national trends and policies on local housing prices is substantial, although local socioeconomic factors also significantly influence market outcomes (Zhang et al., 2014). The disparities in housing prices between Canadian cities are influenced by various factors, including demand dynamics, household income levels, projected inflation rates, and the proportion of nonfamily households (Kushner and Fortura, 1986).

The impact of fundamental factors, such as the unemployment rate, the real mortgage rate, and the price-to-rent ratio, on property values in Australia has exhibited a consistent and predictable pattern (Peng and Chen, 2016). Furthermore, the valuation of recently constructed residential structures, which serve as an indicator of housing standards, and the overall crime rate, reflecting the quality of the residential environment, both had statistically significant influences on property prices (Peng and Chen, 2016).

2.2 Housing renovations

Renovation activities affect property values and housing market dynamics, particularly those to improve energy efficiency. Housing interventions, such as improvements in thermal characteristics and heating and ventilation systems, have been found to benefit occupant health and contribute to reducing carbon emissions (Howden-Chapman and Chapman, 2012). Therefore, sustainable renovations are crucial to effectively mitigating energy consumption and greenhouse gas emissions, improving interior climate conditions, and ensuring the availability of affordable housing. However, successful renovations require active collaboration among stakeholders and the establishment of comprehensive tools and systems designed for renovation purposes (Jensen et al., 2018). For instance, energy renovation measures in multi-apartment buildings positively influence the economy and climate change, primarily by reducing energy consumption and mitigating dangerous particle emissions (Mikulić et al., 2020).

The property's age plays a major role in physical depreciation and functional obsolescence. Renovation and maintenance expenses are often necessary to counteract the effects of depreciation on the value of housing services provided by residents (Knight and Sirmans, 1996). Lowry (1960) emphasises that quality declines with age, leading to various forms of depreciation. Adequate maintenance is crucial to address issues related to housing depreciation, such as wear and tear due to physical deterioration and technological obsolescence (Lowry, 1960; Wilhelmsson, 2008b). While achieving a zero-depreciation rate is impossible due to external and certain functional obsolescence, owners can reduce physical deterioration through appropriate maintenance (Wilhelmsson, 2008b). Numerous studies (Chinloy, 1979; Knight and Sirmans, 1996; Wilhelmsson, 2008b) demonstrate a strong relationship between depreciation and maintenance, with Manganelli (2013) clarifying that improvement or renovation is necessary to maintain an acceptable quality level during the economic life of the property. Bravo et al. (2019) found in Germany that the profitability of renovation measures, income, and favourable opportunities are critical factors influencing homeowners' decision to renovate energy.

Another driving force behind energy efficiency renovations among Dutch homeowners is the desire to improve their overall quality of life. However, several obstacles hinder the widespread adoption of such renovations, including financial constraints, the intricate nature of the process, and limited access to relevant information (Ebrahimigharehbaghi et al., 2019). For example, achieving effective energy-efficient renovations in housing cooperatives is contingent on various elements, including, but not limited to, the acquisition of information, the structure of ownership, regulatory frameworks, and incentive mechanisms.

Wilhelmsson (2008b) found that depreciation rates were considerably lower for maintained than non-maintained property. Although maintenance delayed the physical deterioration of the property, an appreciation of the price was also observed due to design and preferences. Chinloy (1980) compared maintenance rates of houses built in different years for two cities in Canada (London and St. Catharines) and concluded that maintenance is not independent of the age structure in either city. However, a reported annual maintenance requirement rate of 0.94 and 1.19%, respectively. Harding et al. (2007) found that maintenance increases the home's value by roughly 1% per year while lowering the depreciation rate to approximately 2%.

Manganelli (2013) distinguishes physical depreciation into curable and incurable, where the former is a loss in value retrievable with an intervention whose cost is less than the building's increased value, and the latter's intervention cost is higher than the generated value. In other words, owners' renovation and maintenance costs are not worth the incurable physical depreciation unless other strategies, such as subsidies, are available.

Sweeney (1974) studied the rate of maintenance with respect to housing tenure modes (owner-occupants and landlords) and their expected maintenance behaviour. Differences in observed maintenance occur because the owner-occupant is affected by the quality level of the housing unit and the costs and benefits associated with changes in the maintenance expenditure rate. Similarly, Myers (1984) described similar maintenance behaviour but added that maintenance subsidies could incentivise long-term occupants to undertake renovation and maintenance that property owners would not have carried out without subsidies.

A few studies analyse factors influencing renovation decisions, particularly for subsidised housing renovations. Theoretical models indicate that neighbourhood variables, factor prices, and property values influence these decisions. Household income, property value, and building age are also expected to positively impact renovation. Although renovation subsidies may raise the quality standard of properties and increase their prices, it is unclear whether they are intended to improve the property or maximise potential capital gain. Investments in housing renovation are primarily influenced by the difference in market value before and after renovation and the cost of renovation, affecting the investment size (Zavadskas et al., 2004).

The methodology should consider how housing prices are influenced by income levels, housing stock, demographics, and institutional changes, while renovations impact property values by counteracting depreciation and enhancing energy efficiency (Muellbauer and Murphy, 2008; Bourassa and Hendershott, 1995; Howden-Chapman and Chapman, 2012). Additionally, household income, property age, and maintenance practices are critical in understanding the relationship between housing prices and renovation decisions (Knight and Sirmans, 1996; Bravo et al., 2019).

3. The case study and data

3.1 The case of Sweden

In this study, we will analyse the subsidy system in Sweden. For several reasons, Sweden can be a relevant case study regarding tax subsidies for home renovations. First, Sweden is known to have a high level of social welfare, and one of its main areas is housing. Second, in Sweden, there is extensive regulation of the housing sector, including various tax subsidies and grants to support the ownership and renovation of housing (Turner and Whitehead, 2002). Analysing Sweden as a case study can yield crucial insights into the effectiveness and possible disadvantages of tax subsidies for housing renovations and how they affect different groups in society. Another important factor is that Sweden has more home ownership than other countries. This means that owners of houses in Sweden may be more likely to influence political decisions regarding the housing sector. It can also create incentives for the government to support homeownership through tax subsidies.

In the 1980s, the Swedish government introduced a programme called ROT to promote renovations and refurbishments of older buildings. The programme aimed to modernise 425,000 homes and focused on energy efficiency and modernisation. This programme consisted of loans and subsidies to modernise houses older than 30 years (Tunefalk and Legnér, 2019). The programme was later abandoned.

A tax reduction for repairs, alterations, and extensions of residential buildings has existed since the early 1990s. When introduced in 1993, it was a stimulus measure aimed at reducing unemployment in the construction industry (Skatteutskottets betänkande 2001/02:SkU28, 2002). The tax reduction corresponded to 30% of the labour cost; its previous form was applied until 31 December 1994.

The tax reduction was reintroduced in the spring of 1996, but now it is also applied to condominiums. ROT was successively extended but was discontinued in 1999. ROT was reintroduced in 2004–2005 with a tax reduction of 30% of labour costs. The reason for the reintroduction was the weak employment trend in the construction industry. Unemployment was high, and the investment subsidies for the building of apartments were insufficient (Proposition, 2003/04:163, 2004).

In its current form, ROT was introduced in 2008 (Regeringens Proposition, 2008/09:178, 2009). The tax reduction, initially 50% from 2008 to 2016, decreased to 30%. Reintroduction was carried out to increase the labour supply and, in part, to reduce the informal economy. Households would have increased incentives to purchase services legally rather than performing repairs, thereby increasing the labour supply. In 2006, a similar tax reduction system with the same justification was introduced for household services (Ceccato and Benson, 2016). They claim that alterations made to tax policy during this period played a role in increasing tax adherence.

Thus, ROT was introduced as a stimulus measure specifically aimed at the construction industry to stimulate more work and, therefore, reduce unemployment, being a measure to reduce undeclared work (under-the-table work or off-the-books work), that is, to reduce paid work without reporting or paying income tax and social security contributions. Currently, the ROT can amount to 30% of the debited labour cost, including value-added tax. The tax reduction for the ROT may amount to a maximum of SEK 50,000. If more than one person owns the property, each property owner may deduct a maximum of SEK 50,000, i.e. SEK 100,000, together if there are two property owners.

3.2 The data and descriptive statistics

We will test two hypotheses on the relationship between house prices (HP) and tax reduction for renovations, conversions, and extensions (ROT). To do this empirically, we have used aggregated data regarding tax reduction and variables on the housing market (source: Statistics Sweden). In addition, we have information on unemployment rates (unemp), income (Inc), taxes (Tax), and population (Pop), as well as population density (Density) and urbanisation rates (Urban). Table 1 presents the variables included in the empirical analysis.

Data for 290 municipalities in Sweden have been used from 2004 until 2020. The data used are panel data with n = 290 and t = 15 years. The empirical analysis focuses on the tax reduction that can be obtained from renovating single-family homes in Sweden. A variable in the database is the total amount of tax reduction (ROT) in each municipality and year. These variables have been related to the average price of single-family homes (HP), unemployment (Unemp), population density (Density) and urbanisation (Urban). The expectation is that housing prices, density and urbanisation are positively related to tax reduction. The same applies to income (Inc) and taxes (Tax).

Table 2 shows that the total tax reduction amounts to 37 million SEK on average, substantially varying between municipalities and over the years (from 0.1 million SEK to 2,227 million SEK). Housing prices are close to an average of 1,600,000 SEK, ranging from SEK 200,000 to 12,000,000. Additionally, the percentage of rental homes varies widely among municipalities, ranging from only 2%–74%. On average, 33,000 people live in municipalities with an average income of SEK 244,000 and an income tax of 32%. However, there is considerable variation between municipalities for all these variables.

In summary, we can assert that the spatial variation in tax reduction between municipalities is significant. Still, we can also note that the housing market varies significantly between municipalities and other background variables. The Moran's I statistics show substantial evidence of spatial dependence in the error terms for all variables [1]. On the other hand, the unit root test shows that all variables are trend stationary in natural logarithmic form.

Figure 1 shows the spatial distribution of the average tax reduction per person in 2020. The bluer a municipality, the higher the tax reduction per person. We can observe that the tax reduction is higher in the three metropolitan regions but also generally in the southern part of Sweden, although there are exceptions where some municipalities show a relatively lower tax reduction than other neighbouring municipalities. There are also exceptions in northern Sweden, with municipalities where the tax reduction per person is high. It is the largest municipality in the mountains, with many holiday homes. Figure 2 illustrates the spatial distribution of average house prices. The brighter the red of the municipality, the higher the prices. The two maps give a relatively consistent picture, and we can also observe a higher tax reduction per person in municipalities with higher housing prices.

4. Stochastic models

We have estimated two models to explore the potential relationship between house prices and tax-deducted maintenance. The first model is intended to explain the spatial variation of tax reduction for renovations, conversions, and extensions, incorporating various underlying variables. The model is similar to the model (Mayer, 1981) presented. The second model aims to explain the spatial variation in house prices. In the following section, we will report on the two empirical models we have estimated: the housing price and renovation models.

4.1 Renovation model

While few studies explicitly analyse factors influencing renovation decisions, some explore the decision-making process and estimate renovation or home improvement models (Potepan, 1989; Mayer, 1981; Mendelsohn, 1977; Munneke and Womack, 2015). Mayer's theoretical model from 1981 explains the renovation decision through neighbourhood variables, factor prices, and property value. In the model (Mendelsohn, 1977), based on (Mayer, 1981), household expenditure on renovation is explained, among other things, household income, the property's value, and the building's age. All of these factors positively impact the frequency of renovations by households. Munneke and Womack (2015) partly address the issue when they analyse what influences the decision to renovate or demolish a building, considering factors like the building's age and variables in the local neighbourhood. Our empirical model builds on these early theoretical models, aiming to explain the level of renovations in different municipalities using income levels, average housing prices, taxes, unemployment, population and population density, as well as urbanisation and fixed area effects. Based on the theoretical models, a simple deterministic renovation model at the municipality level can be expressed as follows:

(1)Renovation=f (House prices,X )

Here, renovation represents maintenance and reinvestment at the municipal level. The average housing prices and other housing market indicators can explain the variation in renovation between municipalities. It is assumed that the price of housing positively impacts renovations. A larger housing market can positively and negatively impact per capita renovation. In addition to the housing market variables, other fundamental variables, such as average income, income tax, and population size, can affect the renovations per capita.

4.2 House price model

The second model focuses on house prices, considering property attributes significantly affecting housing prices. This model, which resembles that of DiPasquale and Wheaton (1994) and later of Wilhelmsson (2008a), includes various variables that are attributed to house price variations, such as income, single-family stocks, the share of different-age houses, as well as population and income taxes of municipalities. However, the focus will be on the tax reduction variable to estimate the spatial variation of house prices attributed to differences in subsidised renovation (ROT) between different municipalities. Renovation subsidies might impact the level of property maintenance in certain municipalities; however, it is not apparent whether the intention behind such maintenance or renovation is to improve the property and not to move or increase the potential capital gain when it is sold, which, in either way, could disrupt the expected residential filtering process (Allon, 2008). The housing price model can be described as the inverse of the housing renovation model.

(2)House prices=f (Renovations,X )

where the average price of houses in the municipality can be described as a function of the level of renovations and other variables of the housing market and income, tax and population size and density, urbanisation and employment rates.

5. Empirical method

The data utilised are unbalanced panel data, consisting of cross-section panel (i) of municipalities over time (t). The panel data model can be formulated in Equation (3) (Baltagi, 2021; Hsiao, 2022).

(3)Yi,t=αi,t+βXi,t+εi,t

where Y in the renovation model represents tax-subsidised renovations, and X includes housing prices and other fundamental factors explaining renovation. In the house price model, Y will equal house prices, and X will consist of tax-subsidised renovations. The subscript i denotes cross-sectional panels, and t represents the year. We allow the constant α to vary randomly or with fixed effects in space and time. The parameter β measures the impact of the independent variable X on the dependent variable Y, and ε is the stochastic error term. Both fixed-effects and random-effects models are considered, with the fixed-effects model preferred based on the Hausman test results.

We aim to test the hypothesis that renovations impact housing prices and vice versa. In the research question itself, there is an endogeneity that we must address. We have done that by using the instrument variables approach (Angrist and Pischke, 2009). This means that we try to identify exogenously given instrument variables (IV) correlated with the endogenously given variable(s) (X) but not correlated with the error term. We have estimated this through a two-stage procedure. First, we have estimated a model where the exogenous variable X is explained by all other exogenously given variables and the instrument variable Z. In the second step, the predicted variable of X is used in the model where Y is explained according to Equation (4) below.

(4)Xˆi,t=πi,t+ηZi,tYi,t=αi,t+βXˆi,t+εi,t

X is the endogenous independent variable, and Z is the instrument variable. π and η are parameters to be estimated. To find a strong IV is difficult. We will need IV for both house prices and renovations. We are considering variables measuring municipal public services such as culture, infrastructure and school for house prices. To test whether the IV is strong or weak, we have used different tests such as the F-test of excluded instruments, Andersson Canonical Correlation LM Statistics, and Cragg-Donald Wald F statistics (all available in ivreg2 in Stata). We have used municipal spending on culture per capita for house prices, and for renovation, we have used construction costs as IV.

We know that spatial dependency is present in the data. We have, therefore, estimated several spatial models from Manski (1993) to Durbin models (Durbin, 1960). For model comparison and weight matrix selection, a specific-to-general test procedure is followed, as proposed by Elhorst (2010). A two-stage fixed effect IV model is initially estimated, and Moran's I test is conducted to test for spatial dependency. If the tests on residuals are significant, a general Manski model is estimated. The log-likelihood ratio test (LR-test) is used to compare the suitability of Manski compared to the Kelejia-Prucha model (Kelejian and Prucha, 1998), Spatial Durbin model and Spatial Durbin Error model according to Elhorst (2010) terminology. The general Manski model is equal to:

(5)Y=ρWY+αi+Xβ+WXθ+uu=λWu+ϵ

where W equals the spatial weight matrix, WY, WX, and Wu are the spatial interaction effects between dependent, independent, and error terms. The parameters to be estimated are ρ (spatial autoregressive coefficient), λ (spatial autocorrelation coefficient) and θ. We will test these by one to determine whether a Kelejian-Prucha (θ = 0) model, Spatial Durbin model (λ = 0) or Spatial Durbin error model (ρ = 0) is preferred. As with the panela data fixed effect model, one of our X will be instrumented by the variable Z.

The spatial weight matrix is then selected using two types of matrices: distance-based and contiguity-based matrices (Chasco, 2013). These types have been widely used in spatial econometric literature (Elhorst, 2010). The selection of the weight matrix is based on the mobility pattern between municipalities, with contiguity and inverse distance matrices used in studies like Mendiola et al. (2015).

Although the choice is somewhat arbitrary, Elhorst (2010) notes that the wrong choice of the spatial weight matrix can distort the estimates, but the likelihood of this occurring is small if spatial dependence is strong. LeSage and Pace (2014) argue that the sensitivity of estimates to the choice of spatial weight matrix is a myth in spatial econometrics. It has been recommended to use goodness-of-fit measures to differentiate between spatial matrix specifications, as there are no clear theoretical preferences (LeSage and Pace, 2009; Stakhovych and Bijmolt, 2009; Halleck Vega and Elhorst, 2013). We use the log-likelihood value to differentiate between spatial weight matrices (Elhorst, 2010).

6. The econometric analysis

This section presents the findings of the empirical analysis, which has taken place in two different stages. In Step 1, we estimate the panel data IV models; in Step 2, we have estimated the spatial panel data IV model. Table 3 presents the results of two separate 2SLS IV models, with the dependent variables being house prices (HP) in model 1 and renovation per capita (ROT) in model 2. Both models include fixed effects for municipalities and years.

Model 1 has an AIC of −8038.7, and Model 2 has an AIC of −5972.1, indicating that the house price model fits the data better. The p-value of the Hausman test is 0.00 for both models, suggesting that the fixed-effects model is preferred over the random-effects model. The F values of the excluded instruments are 30.73 and 19.42 for Models 1 and 2, respectively, indicating that the instruments used are not weak. Additionally, the Anderson Canonical Correlation LM Statistics values are 33.01 and 20.91 for Models 1 and 2, respectively, indicating that the models are not underidentified. Lastly, the Cragg-Donald Wald F statistics are 30.73 and 19.42 for Models 1 and 2, respectively, confirming the strength of the instruments against weak identification concerns. Both models exceed the critical value of 16.38, indicating that the instruments are strong.

In Model 1, the coefficient for ROT is 0.404, significant at the 0.01 level, indicating that an increase in renovation per capita correlates with higher house prices. Since the variables are in natural logarithm form, a 1% increase in renovation per capita corresponds to a 0.4% rise in house prices. As expected, higher unemployment rates are linked to lower house prices. However, the size of the rental housing market does not affect the owner-occupied housing market. Population density is positively correlated with house prices, but higher levels of urbanisation are associated with lower house prices, suggesting potential multicollinearity issues. The population coefficient is 0.263, significant at the 0.05 level, indicating that a larger population is associated with higher house prices. Unexpectedly, the coefficient for an income variable is 0.137, but it is not statistically significant. Lastly, the income tax coefficient is 0.523, which is significant at the 0.05 level, indicating that higher income tax is associated with higher house prices.

In Model 2, house prices do not explain renovation per capita. Similarly, unemployment rates are not statistically significant. However, the coefficient for the share of rental housing is significant at the 0.05 level, indicating that a higher share is associated with a lower renovation per capita. Population density negatively impacts renovation, while urbanisation has no effect. The income coefficient is 0.758, which is significant at the 0.001 level, indicating that higher income is associated with higher renovation per capita. Finally, the coefficient for income tax is −1.410, significant at the 0.001 level, suggesting that a higher income tax is associated with lower renovation per capita.

The results in Table 4 present the estimates of the Manski model with fixed effects and instrumented endogenous independent variables, showing the impacts of various independent variables on house prices (HP) and renovation per capita (ROT). The results are divided into two columns, (1) and (2), representing the two dependent variables.

The estimation of the spatial autoregressive model with fixed effects is done by a quasimaximum likelihood estimator (Lee and Yu, 2010). Model 1's results suggest a significant positive impact on instrumented renovation per capita house prices. The coefficient indicates that a 1% increase in ROT leads to a 0.665% increase in HP, significant at the 0.1% level. Unemployment has a significant negative impact on house prices, where a 1% increase in unemployment leads to a 2.28% decrease in house prices, also significant at the 0.1% level. Conversely, the share of rental dwellings significantly positively affects house prices. A 1% increase in the share of rental housing leads to a 6.16% increase in house prices, significant at the 1% level. Population density positively impacts house prices, while urbanisation has a negative effect. Population and income tax positively affect house prices, while income does not significantly impact.

In Model 2, instrumented house prices do not significantly impact renovation per capita. The same applies to unemployment, the share of rental dwellings, and urbanisation. However, income has a positive effect, and income tax has a negative impact.

The likelihood ratio (LR) tests comparing the Manski vs Kelejian-Prucha, Durbin, and Durbin error models indicate significant differences, suggesting that the Manski model is preferred. The Hausman test results are (prob>0.00) for both models, suggesting the appropriate fixed-effects model. The empirical results show differences between the fixed effect model and the spatial fixed effect model.

In the fixed effect model (Table 3), the impact of renovation per capita (ROT) on house prices (HP) is positive and significant, with a coefficient of 0.404. This suggests that a 1% increase in ROT is associated with a 0.4% increase in HP. However, the spatial fixed effect model (Table 4) shows a stronger relationship between ROT and HP, with a coefficient of 0.665, indicating that a 1% increase in ROT corresponds to a 0.665% increase in HP, significant at the 0.1% level.

The Manski model we use decomposes the effects of independent variables into direct, indirect, and total impacts. Decomposing the effects allows for a nuanced understanding of how changes in independent variables influence the dependent variable directly (on the same unit) and indirectly (on neighbouring units). The direct, indirect, and total impacts are shown in Table 5.

The table provides the direct, indirect, and total effects of two independent variables (IV ROT and IV HP) on the dependent variables (HP and ROT), estimated using the Manski model. The Delta method is used for estimation.

House price model: The direct effect suggests that a one-unit increase in IV ROT leads to an average increase of 0.6701 units in the dependent variable HP, holding other factors constant. The high z-value (3.98) indicates this effect is statistically significant. On the other hand, the indirect effect suggests that a one-unit increase in IV ROT results in an average decrease of 0.1181 units in the dependent variable ROT in neighbouring units. The negative sign and significant z-value (−2.71) indicate a significant and inverse spillover effect. The total effect combines the direct and indirect effects. Here, the total impact of a one-unit increase in IV HP is an average increase of 0.5520 units in ROT, which is significant (z = 3.91).

The Renovation model: Direct effect indicates that a one-unit increase in IV HP leads to an average increase of 0.2087 units in the dependent variable ROT. The z-value (0.73) indicates this effect is not statistically significant. The indirect effect suggests that a one-unit increase in IV HP results in an average decrease of 0.0659 units in HP in neighbouring units. The low z-value (0.61) indicates this effect is not statistically significant. Hence, the total impact of a one-unit increase in IV HP is an average increase of 0.1427 units in HP, which is not significant (z = 0.47).

7. Policy implications

Our findings suggest that incentives for renovations increase home values, especially in areas where they are used more frequently. This implies the necessity of reassessing the allocation of subsidies to ensure they do not excessively favour wealthier municipalities or contribute to exacerbating regional price differences.

Additionally, governments should strive to achieve a balance between promoting property enhancements and preserving home affordability. Although subsidies for renovations enhance housing quality, they can also exacerbate the unaffordability of housing for lower-income demographics, mainly when subsidies are predominantly allocated to expensive regions. Therefore, to ensure fair and equal access to housing, the government may consider customising subsidies to assist households with lower incomes or areas where property costs are lower. This could facilitate a more equitable improvement in housing standards among various economic brackets and geographical regions.

Regular and systematic monitoring and evaluation of the subsidy program are essential. This would include evaluating the direct consequences of renovation activities and house quality and the persistent implications of housing market dynamics, encompassing prices and accessibility for various income brackets. Most importantly, the renovation subsidy should be regarded as a more comprehensive component of the housing policy framework. This would require ensuring its compatibility with other housing efforts, such as those focused on increasing housing availability or offering direct assistance to low-income tenants. To mitigate the risk of gentrification, particularly in metropolitan regions, it is necessary to implement laws that protect current low-income residents from being displaced due to rising costs in their localities.

With the current emphasis on energy efficiency and modernisation, there is a better chance to align renovation subsidies with sustainability objectives. This may imply offering supplementary incentives for repairs that substantially decrease a property's carbon emissions or enhance its environmental efficacy. Furthermore, increasing public awareness of these financial aids, especially among low-income homeowners, could improve their accessibility and utilisation. This involves streamlining the application process and offering support or guidance for renovations.

Lastly, promoting and supporting the advancement of affordable and environmentally friendly refurbishment technologies and practices through research and development could potentially reduce the overall expenses associated with improvements, enhancing affordability for a wider demographic.

8. Conclusions

We began our study with the premise that differences in house prices between Swedish municipalities may be partially explained by variations in the renovation tax deduction provided by the Swedish government. However, it is also possible that high-priced houses or affluent neighbourhoods are renovated more frequently to maintain their perceived property values. As we are uncertain whether the tax deduction influences the desire to renovate or whether high-value houses require higher rates of renovation and maintenance, we have estimated several models to address both scenarios.

We suggest that renovation and maintenance tax deductions (ROT) may encourage property maintenance, which could subsequently increase house prices. Our estimates indicate a positive impact on house prices from tax subsidies for renovations. Given that most of the subsidies have been directed to areas with the highest prices, the subsidy has increased the gap in housing prices between regions.

Although the Swedish tax deduction program may have achieved its intended objectives of reducing housing costs, it could also promote other outcomes that affect housing markets. From a well-functioning housing market perspective, one can conclude that the maintenance tax deduction disrupts the perceived residential filtering process. These deductions enable more renovation than warranted, causing renovated properties to command higher prices than non-renovated or infrequently renovated ones. Consequently, property prices become unaffordable for low-income groups. ROT can also interfere with providing equal standards, costs, and influence on housing management, maintenance, and modernisation issues.

Renovating housing to maintain its condition or enhance energy efficiency raises property prices, impacting affordability for different income groups and affecting the housing market's filtering process. The unintended consequences of subsidised renovations on property prices and the potential effects of excessive renovations on this process have not been thoroughly investigated. Further research on how owners decide on renovation and maintenance strategies, particularly in relation to tax deductions (ROT), may reveal this policy's success or unintended consequences.

Figures

Figure 1

Spatial distribution of tax reduction renovations (ROT)

Figure 2

Spatial distribution of average house prices (HP)

Table 1

Variable definitions

Variable	Definition	Period
ROT	Tax reduction for renovations, conversions, and extensions, Total sum, SEK million	2004–2020
HP	Average housing price, single-family houses, SEK thousands	2004–2020
Rentals	Rental apartments as a share of total housing stock, percentage	2004–2020
Unemp	Unemployment rates in per cent in the municipality	2004–2020
Density	Population density in the municipality	2004–2020
Urban	Urbanisation rates in the municipality	2004–2020
Pop	Population, number of persons in all age groups	2004–2020
Inc	Average incomes, population older than 16 years, SEK thousands	2004–2020
Tax	Income tax in municipality and county, per cent	2004–2020

Note(s): The table defines the variables we access in our analysis and the period in which we have the data. It is panel data with 290 municipalities and 15 years (n = 290 and t = 15). The variables refer to three different definitions of tax subsidies; ROT is the total tax reduction. During three years, the ROT was equal to zero. The variable HP refers to the average price of municipal single-family homes per year. Rentals measures the share of rental apartments in the total housing stock in the municipality per year. The variable Unemp is measured as the percentage of the unemployed in the workforce; Density is measured as the population divided by the geographical size of the municipality, and the variable Urban is calculated as the share of the population living in urbanisation. The control variables also refer to Pop, the population size, Inc, which refers to the average income of the population over 16 years of age, and tax (the total income tax in the county and municipality) in the years in question. Period refers to years (t) to which we have access to data

Source(s): Secondary data from Statistics Sweden and calculations by the authors

Table 2

Descriptive statistics

Variable	Obs	Mean	Std. Dev	Min	Max	LLC	Morans’I
ROT	4,060	37.055	92.062	0.1	2227.1	0.000	0.000
HP	4,930	1641.326	1359.26	216	12,214	0.000	0.000
Rentals	4,930	26.699	9.058	2.22	74.19	0.000	0.000
Unemp	4,930	1.318	0.577	0.1	5.1	0.000	0.000
Density	4,930	142.029	501.901	0.2	6076.1	0.000	0.000
Urban	4,930	74.966	14.409	30	100	0.000	0.000
Pop	4,930	33205.35	67085.318	2387	975551	0.000	0.000
Inc	4,930	244.41	45.193	165.3	578.4	0.000	0.000
Tax	4,930	32.422	1.155	28.89	35.15	0.000	0.000

Note(s): The table shows the descriptive statistics for inclusive variables. Statistics regarding the number of observations (Obs), average values (Mean), standard deviation (Std.Dev.), minimum (Min) and maximum (Max) are presented. Included is also a unit root test Levin-Lin-Chu test - LLC (Levin et al., 2002). The test has been performed on the natural logarithmic of the variables with a trend. The test statistics p-value is included in the table. We also perform Moran’s I test for spatial correlation among the residuals using a contiguity spatial weight matrix. The test statistics p-value is included in the table. Stata commands: sum, reg, estat moran, spmatrix, xtunitroot

Source(s): Secondary data from Statistics Sweden and calculations by the authors

Table 3

Two-stage least squares instrument variable (IV) model, fixed effects

	(1)	(2)
	HP	ROT
IV ROT	0.404**
	(2.99)
IV HP		−0.00786
		(−0.03)
Unemp	−0.0283***	0.00455
	(−5.24)	(0.44)
Rentals	0.0301	−0.0530*
	(1.41)	(−2.04)
Density	0.340**	−0.425**
	(2.59)	(−3.28)
Urban	−0.171**	0.0351
	(−3.10)	(0.42)
Pop	0.263*
	(2.28)
Inc	0.137	0.758***
	(1.03)	(4.38)
Tax	0.523*	−1.410***
	(2.25)	(−7.81)
Constant	4.648***	−5.776***
	(3.47)	(−3.87)
Fixed municipalities	Yes	Yes
Fixed-year effects	Yes	Yes
AIC	−8038.7	−5972.1
Hausman test	0.00	0.00
F-test of excluded instruments	30.73	19.42
Andersson Canonical correlation LM statistics	33.01	20.91
Cragg-Donald Wald F statistics	30.73	19.42
Stock-Yogo critical values	16.38	16.38
Observations	4,060	4,060

Note(s): p < 0.05, **p < 0.01, ***p < 0.001. The dependent variable is house prices (HP) in the first model and renovation per capita in the second model (ROT). The independent variables included are unemployment (unemp), the share of rental dwelling of total housing stock (rental), population density (density), urbanisation (urban), population (pop), income (inc), and income tax (tax). Municipality spending on culture per capita is the instrument variable for the endogenous variable house price, and the cost of housing construction (excluding the cost of land) is the instrument variable for renovation. All variables are transformed into a natural logarithm. The statistics for t are in parentheses. STATA 17 has been used. Commands: ivreg2, xtivreg, and hausman

Source(s): Secondary data from Statistics Sweden and calculations by the authors

Table 4

Manski model, fixed effects

	(1)	(2)
	HP	ROT
Main
IV ROT	0.665*** (3.99)
IV HP		0.209 (0.73)
Unemployment	−0.0228*** (−5.05)	0.00511 (0.50)
Rentals	0.0616** (3.15)	−0.0338 (−1.30)
Density	0.510*** (3.82)	−0.642*** (−5.15)
Urban	−0.167*** (−3.39)	−0.00621 (−0.07)
Population	0.292** (2.71)
Income	0.113 (0.78)	0.735*** (4.12)
Tax	1.131*** (4.34)	−0.837*** (−4.08)
W
IV ROT	−0.00323 (−0.60)
Unemployment	−0.00704 (−0.64)	0.00306 (0.18)
Rentals	−0.117** (−2.75)	0.0319 (0.52)
Density	0.171 (0.58)	0.157 (0.40)
Urbanisation	−0.230 (−1.51)	0.235 (1.07)
Population	−0.256 (−0.81)	0.271 (0.64)
Income	0.452** (2.92)	0.0307 (0.11)
Tax	−0.142 (−0.45)	−0.853 (−1.85)
HP	−0.249** (−3.19)
e.HP	0.640*** (13.02)
IV HP		−0.0793 (−0.60)
ROT		−0.00443 (−0.48)
e.ROT		0.539*** (24.11)
sigma_e
Constant	0.0729*** (73.48)	0.101*** (85.26)
Fixed Municipalities	Yes	Yes
Fixed-Year effects	Yes	Yes
AIC	−10701.2	−6317.1
Hausman	0.0000	0.0000
LR test Manski vs Kelejian-Prucha	0.0001	0.0018
LR test Manski vs Durbin	0.0000	0.0000
LR test Manski vs Durbin Error	0.0091	0.0001
Observations	4,930	4,060

Note(s): p < 0.05, **p < 0.01, ***p < 0.001. The dependent variable is house prices (HP) in the first model and renovation per capita in the second model (ROT). The independent variables included are unemployment (unemp), the share of rental dwelling of total housing stock (rental), population density (density), urbanisation (urban), population (pop), income (inc), and income tax (tax). Municipality spending on culture per capita is the instrument variable for the endogenous variable house price, and the cost of housing construction (excluding the cost of land) is the instrument variable for renovation. All variables are transformed into a natural logarithm. Spatial weight matrix (W): contiguity. The statistics for t are in parentheses. STATA 17 has been used: commands: spxtregress, spmatrix, lrtest, and hausman

Source(s): Secondary data from Statistics Sweden and calculations by the authors

Table 5

Direct, indirect and total effect from the Manski model

	IV ROT		IV HP
	dy/dx	z	dy/dx	z
Direct	0.6701	3.98	0.2087	0.73
Indirect	−0.1181	−2.71	−0.0659	0.61
Total	0.5520	3.91	0.1427	0.47

Note(s): Estimates the average impact of the direct, indirect, and total impacts of independent variables IV ROT and IV HP on the reduced-form mean of the dependent variable (ROT and HP), Delta method is used. Command: estat impact

Source(s): Secondary data from Statistics Sweden and calculations by the authors

Notes

1. Here we are using the contiguity spatial weight matrix, but Moran’s I statistics using inverse distance as spatial weight matrix shows no spatial dependence in the error terms.