Do People Value Environmental Goods? Evidence From the Dutch Housing Market

Abstract

This article studies the relationship between environmental goods and housing prices in the Netherlands. Applying a hedonic pricing model, we estimate simultaneous correlations between housing prices and air quality, noise pollution, green scenery, and water scenery, demonstrating the importance of including related sets of environmental goods to avoid omitted variable bias. We find that households particularly value noise pollution and green and water amenities in close proximity to their house. Air quality does not appear to substantially impact housing prices, neither do green spaces located more than 200 meters away. These results suggests that households mainly value those environmental goods that are directly experienced. A comparison of our hedonic price results to existing monetary estimates of the health effects of environmental goods indicates that people only partially internalize these health effects.

Keywords:

• Hedonic pricing
• environmental goods
• functional form

JEL CLASSIFICATION:

• Q51
• Q53

Acknowledgements

We are grateful to Henri de Groot, Hans Koster, Rob Euwals, and Guus Velders for comments and suggestions on this article. We also thank seminar participants at the Global Conference on Economic Geography 2022, CPB bureau for economic Policy Analysis, CE Delft, and the Dutch Institute for Public Health and the Environment (RIVM). We also thank two anonymous reviewers for their comments and suggestions. Clémence Sautai provided excellent research assistance. Views expressed in this article are our own and do not necessarily reflect those of the CPB.

Disclosure Statement

The authors report there are no competing interests to declare.

Notes

1 As alternative measures, we separately estimate the effects of PM₁₀ and nitrogen dioxide (NO₂).

2 These people live at locations at which either the recommend standards by the World Health Organization (WHO) of noise pollution (54 dB road noise) or air pollution (PM_2.5 over 5 mg/m3) are exceeded. The recommended standards by the World Health Organization can be found here (link).

3 Particles at this level can, when being inhaled, be transmitted from the lungs to the blood stream and brain. There is an associated increased health risk, such as respiratory complaints, cardiovascular disease, and lung cancer (see Anderson et al., 2012; Raaschou-Nielsen et al., 2013). As a consequence, the exposure to PM_2.5 is associated with an increased risk of earlier mortality (Chen et al., 2013).

4 The Lden measure (level day, evening, night) reflects the average noise transmitted over a day, although noise transmitted during the evening and night are “penalized” with a 5db and 10db increment factor, respectively.

5 A detailed description of the air pollution model is listed in Sautar et al. (2020).

6 Further details of the noise pollution model are listed in Schreurs et al. (2010).

7 Since the measure of trees is based on aerial photos, the method cannot determine the land use below the trees (e.g., the presence of grass and shrubs). For this reason, our coefficients that exhibit the willingness to pay for trees may also partly comprise the willingness to pay for green scenery below the trees.

8 The Netherlands is also known as the “Lowlands,” because over 25% of the Netherlands is located below sea level To illustrate this: our data shows that over three-fourths of all housing transactions have water within a distance of 200 meters.

9 The steps conducted to clean the data set on residential property transactions is listed in the Data appendix Table A1.

10 A conceptual description of the hedonic price method in the context of environmental (dis)amenities is listed in Appendix section A2.

11 Although we have a large set of potential control variables, we may not want to include them all in the regression equation due to risks of overfitting (Bateman et al., 2001). To infer whether our model includes too many controls, we perform a lasso methodology as a robustness check (see Appendix section A3 for econometric details). We also use the method developed by Oster (2019) to assess the importance of omitted variables in our regression model.

12 On average, there are about three neighborhoods located within each postal-code area.

13 The national model that determines the air pollution levels in the Netherlands uses the same 1 × 1km spatial grids. The national model is complemented by local inputs on a spatially detailed level (e.g., to compute the transmission of air pollution from highways; further details listed in the data section). As a robustness check, we therefore also use clustered errors on the lowest spatial grids possible (the lowest neighborhood level).

14 Due to the problem of multicollinearity, we did not include PM_2.5 and NO₂ simultaneously in our regression analysis. Therefore, we determined the willingness to pay for NO₂ by excluding PM_2,5 in equation (2)(2) $$\mathrm{\text{log}}\hspace{0.17em}{p}_{i,n,r,t}=\alpha {\sum }_{j=1}^J{{\left(\mathit{\text{EV}}\right)}^k}_i+\beta X_{i,t}+\gamma A_{i,r}+\delta D_{i,r}+\zeta S_{i,r}+{\mu }_n+{\lambda }_t+{\epsilon }_{i,t\mathrm{ }}$$(2) . These results indicate that each $\mu$ g/m³ increase in NO₂ is associated with a decrease in residential property prices by 0.31%.

15 These results are available upon request.

16 We further investigated whether the willingness to pay for noise pollution depends on the number of insulation layers and/or the maintenance quality of a house. We therefore extended our previous strategy by interacting the hedonic price schedule for noise pollution with both characteristics. The results exhibited in Appendix Figure A3 confirm the idea that insulation and the maintenance quality matter in terms of the housing price relationship with noise pollution. In particular, especially within the range up to 50 decibels, the implied marginal willingness to pay for (less) noise pollution becomes stronger as the number of insulation layers (maintenance quality) increases.

17 We did consider the hypothesis that households primarily value a direct view on green scenery instead of the percentage of green scenery around their house. We therefore included dummy variables indicating whether houses have a direct view on (a) a forest, (b) a park, and (c) on water. The results, which are available upon request, indicated this addition primarily affects the coefficients of the willingness-to-pay measures for environmental amenities within 50 meters. Unsurprisingly, these coefficients are adjusted downward, albeit slightly (less than 5%). Taking into account that households also value the percentage of green amenities, the willingness to pay for a direct view on a forest (water) is respectively 2% (4.8%). The willingness to pay for a view on a park is statistically insignificant.

18 Another explanation is that our variables that are proxies for density (number of addresses, degree of urbanity) do not perfectly capture the full extent of density. In that case, the coefficients of the percentage of grass (beyond 200 meters) may partially pick up a lack of density.

19 There are some articles that estimate the willingness to pay for the distance of a house to water: Cho et al. (2006) and Rouwendal et al. (2017). The results of these articles indicate that people are willing to pay for a closer distance of water to their house. The estimated relationship tends to be weak, however, and in the case of Rouwendal et al. (2017) already disappears after 30 to 40 meters. This article primarily shows that households are willing to pay for a direct view on water from their house. In addition, our article suggests that next to this direct view, people also value water at larger distances up to 1,000 meters from their home, as this may enable them to recreate on water (e.g., using a boat).

20 We use a demeaned (within postal-code areas) estimator because of the constraints imposed by the errors-in-variables regression model. The method prohibits correcting for a measurement error that is larger than the $R^2$1. <TQ > Please change hyphen to minus sign.</TQ > Using a demeaned within-estimator leads to the within $R^2,$ which is lower than the overall $R^2.$

21 Because we consider the statistical inferences of a set of variables simultaneously, there is an increased risk of incorrectly rejecting a null hypothesis (type 1 errors). The Bonferroni correction proposes to control for this increased risk by reevaluating the desired significance levels using the formula $\alpha /m,$ where $m$ refers to the number of hypotheses. Since we consider 20 environmental variables simultaneously, we obtain a desired significance level of at least 0.0025 (0.05/20). In Appendix Figure A5, we show that the variables that must be interpreted with caution do not obtain this threshold of significance.

22 Appendix Table A3 provides the descriptive statistics of these variables.

23 We have tested whether the price segments and dwelling types have significant differences in the association between environmental (dis)amenities and housing values by using interaction terms.

24 We use the double selection in combination with cross-validation to select the optimal value of the lasso penalty parameter. Details on the lasso methodology are provided in the Appendix section A3 Methods.

25 Specifically, the method of Oster is not suited to perform the test simultaneously for the linear term and its higher order polynomials.

26 One overlooked point is that the bid curves of consumers and offer curves of producers for environmental (dis)amenities may be related to the bid and offer curves of housing characteristics. That is, the bid function of consumers for noise pollution may be (partly) determined by the bid function for the number of insulation layers of a house or the maintenance quality of the building. We consider the importance of the relatedness of bid functions in our empirical section.