1# RECORD 2075796388c43cd5aedc64728dd832af194aa20d2# TITLE: Effects of air quality on house prices: evidence from China’s Huai River Policy3# YEAR: 2020.0 VENUE: nan DOI: 10.1080/00779954.2020.18270144# AUTHORS: Xinghua Liu; Qiang Li; S. Chand; K. Sharpe5# PAGES: 16 CANDIDATE TABLE PAGES (1-based): [7, 10, 12, 13, 14]6# ABSTRACT: We employ a regression discontinuity (RD) design to estimate the impact of air pollution on house prices across a river that demarcates regions with and without coal-fired heating resulting from the Huai River Policy. This policy was decreed by the Chinese government in the 1950s and mandated the burning of coal for indoor heating at subsidised prices north of the Huai River. Employing quasi-experimental variation in particulate matter of 10 micrometres or less in aerodynamic diameter (PM10) generated by this arbitrary policy and a regression discontinuity (RD) design based on distance from Huai River, we estimate the local average treatment effect (LATE) to provide new evidence on the capitalisation of PM10 air pollution into house values. By using panel data covering 30 large cities on either side of the river for the period 2006–2015, we found that 1 µg/m3 (micrograms per cubic metre) reduction in average PM10 is associated with an approximately 1% increase in house prices. The results are robust to using parametric and nonparametric estimation methods, adjustment to a rich set of covariates, and using a subsample excluding first-tier cities.789===== PAGE 1 (score 1, numeric-paren 0, table-words 0, var-words 1) =====1011Effects of air quality on house prices: evidence from China’s12Huai River Policy13Author:14Liu, X; Li, Q; Chand, S; Sharpe, K15Publication details:16New Zealand Economic Papers17v. 5518Chapter No. 119pp. 52 - 65200077-9954 (ISSN); 1943-4863 (ISSN)21Publication Date:222021-01-0123Publisher DOI:24https://doi.org/10.1080/00779954.2020.182701425License:26https://creativecommons.org/licenses/by-nc-nd/4.0/27Link to license to see what you are allowed to do with this resource.28Downloaded from http://hdl.handle.net/1959.4/unsworks_72524 in https://29unsworks.unsw.edu.au on 2026-10-20303132===== PAGE 7 (score 16, numeric-paren 0, table-words 3, var-words 7) =====3334NEW ZEALAND ECONOMIC PAPERS35536Figure 2. The comparison of PM10 and house prices in northern and southern Chinese cities.37Notes: This is a provincial map of provinces of mainland China as well as Taiwan Province. The PM10 levels and house prices in38the capital city of each province are shown in the map. The 30 cities are matched with the provinces in which they are located39except for Tibet, Taiwan, Hong Kong and Macau due to unavailability of data. The unit for PM10 and house prices is μg/m3 and40yuan/m2, respectively. Data on PM10 are from China Environment yearbook and data on house prices are from China Real Estate41Statistical Yearbook.42Table 1. Summary statistics for key variables.43Variable44Observations45Mean46Standard deviation47Min48Max49PM10 (μg/m3)5030051102.995231.1453305430555Real house prices(yuan/m2)56300575916.31584243.115915416049,659.8261Per capita real GDP(10,000 yuan)62300636.537922645.270577651.2214416670.4104367Real wage(yuan)682976945,112.187027,029.217116,9117239,8203.373Fiscal surplus (100 million yuan)7429975–152.1876248.8577–2315.9378601.8379Population density(people/km2)80299811617.66821373.9783223.318411,449.385Per capita floor space (m2)86300878.06884.32890.979024.81911 (Coastal city)92300930.13940.3495096197Students enrolled at higher education9830099433,557.1100225,334.110110,2801021,043,221103No. of hospitals owned per 10,000 people1042991050.531060.421070.231084.701091(North of Huai River)1103001110.51120.5011301141115Latitude distance from Huai River1163001170.571187.06119-13.712015.08121Notes: The entries are calculated across 30 cities during the 2006–2015 period.122north are generally exposed to substantially higher level of PM10 concentrations compared to those123in the south. This is consistent with the claim that air pollution is more severe in northern Chinese124cities. However, the figure fails to reveal some of the systematic patterns of the spatial distribution of125housing price across the river line.126Table 1 reports summary statistics for the key variables in this analysis. China’s mean value for127PM10 concentrations was 102.99 μg/m3 over the period 2006–2015. For the purpose of comparison,128we calculate the US PM10 concentrations for the same period based on data reported by the United129States Environmental Protection Agency (EPA). The figure is 68.58 μg/m3 for the US, making China’s130PM10 concentrations larger by a factor of 1.5 to US levels.131132133===== PAGE 8 (score 5, numeric-paren 0, table-words 1, var-words 3) =====1341356136X. LIU ET AL.1374. Econometric model138We use both the conventional and the RD approach to estimate the association between PM10 and139house prices. The basic idea and structure of this section follows closely that of Chen et al. (2013)140and Ebenstein et al. (2017). The conventional strategy uses the city fixed effects regression model to141estimate the following equation:142HPit = β0 + β1PMit + Xit + γi + εit143(1)144where i denotes a city and t indexes a year, and145n146i=1147γi = 0. PMit is the PM10 concentration in city i148in year t, Xit is a vector of observed covariates used to control for other determinants of house prices149other than air quality, γi are city fixed effects, and εit is an error term. The outcome variable is the150natural logarithm of the average residential house prices in city i in year t. β1 is the coefficient of151interest for Equation (1).152The estimation of Equation (1) faces problems of endogeneity, with the risk that the estimates of β1153are biased. Recall that the validity of the conventional approach depends on the assumption that the154adjustment of covariates rules out all the confounding factors. Although we have controlled for a set155of observed characteristics for which the requisite data is available, unobserved determinants of house156prices may also co-vary with PMit. The use of panel data analysis allows us to eliminate the effects157of any time invariant unobservables by inclusion of city fixed effects, but this is far from sufficient to158address potential problems of endogeneity. Measurement error of pollution concentrations is a serious159risk to potential endogeneity as there is evidence that air quality readings tend to be underreported160by the officials. Consequently, the estimates of the impact of air quality on the price of housing from161Equation (1), our parameter of interest, may be biased.162The second approach leverages the regression discontinuity (RD) design where we exploit the163exogenous variation caused by the Huai River Policy to estimate its impact on both PM10 concentra-164tions and house prices. The development of RD identification is based on the idea that the treatment165status is determined by a continuous variable crossing a cut-off due to the imposition of some arbi-166trary rule. This feature provides a local randomisation for estimating treatment effects. RD design has167emerged almost simultaneously in a wide range of disciplines but has only recently become prevalent168in applied econometrics. We exploit an RD design that has a break in the spatial dimension, whereby169a treatment is applied along a geographical boundary. This is done using two different approaches.170First, we use a sharp regression discontinuity (sharp RD) design to test any discontinuity in PM10 and171house prices across the river. Under this approach, the Huai River Policy is treated as a treatment vari-172able and its impacts on the two endogenous variables are estimated. In this case, the treatment status173is a deterministic and discontinuous function of the running variable (the degree distance north of174the Huai River boundary), making the estimates a sharp RD design. Second, we use a fuzzy regres-175sion discontinuity (fuzzy RD) design to estimate the impact of PM10 on house prices. A fuzzy RD176is where the treatment status is no longer deterministically related to the threshold-crossing rule. In177our context, the PM10 pollutant exists on both sides of the river but the probability of PM10 expo-178sure increases considerably across the river, making our study a fuzzy RD in nature. Consequently,179the discontinuity becomes an instrumental variable for PM10. It is worth noting that the validity of180our RD design rests on the assumption that any unobserved determinants of PM10 and house prices181change smoothly as they cross the river line. If this identification assumption is true, adjustment for182flexible polynomial terms in degrees latitude north of the river (i.e. local linear regressions (LLR))183will eliminate all sources of bias and hence allow us to draw causal inferences.184Of course, one may argue that heating facilities might increase housing construction costs and185therefore raise house prices in Northern China. Since we hypothesise that, holding all else constant,186the housing price will be higher for cities just to the south of the river, our estimations in this paper187are expected to be downwardly biased.188189190===== PAGE 10 (score 42, numeric-paren 17, table-words 8, var-words 5) =====1911928193X. LIU ET AL.1945. Empirical results195Table 2 reports the summary statistics for several key determinants (PM10 and control variables) of196house prices and lends support to the validity of our RD design. The first two columns report the197means and the standard derivations in cities located north and south of the Huai River. Column [3]198presents the mean difference between the south and the north along with the standard errors.199The table shows that PM10 exposure differs significantly among northern and southern cities.200Concentrations of sulphur dioxide (SO2) and nitrogen dioxide (NO2) are also reported in the China201Environmental Yearbook. Indeed, Table 2 demonstrates that the differences among these pollutants202are also statistically significant among southern and northern cities. This difference is reassuring since203sulphur dioxide and nitrogen dioxide are lighter and therefore can travel further than PM10, which is204our measure of environmental dis-amenity on house prices caused by living north of the river.205Table 2 also suggests that the mean value for both students enrolled at higher education institutions206and the number of hospitals owned per 10,000 people are statistically different across northern and207southern cities. To gather evidence on the validity of the RD estimates, we check for discontinuities in208these two variables that may confound the results. Figure A1 in the appendix shows the relationship209between degrees north of the Huai River and these two control variables. The figure indicates that210these control variables do not show significant jumps at the cut-off, suggesting no breaks around the211cut-off.2125.1. Estimates of the effect of the Huai River Policy on PM10 and house prices213First, we graphically show the effect of the Huai River Policy on air pollution in Figure 3. The dots in214Figure 3 represent the mean of PM10 concentrations falling within each bin. The fitted values repre-215sented by the solid line are from a kernel weighted 2nd-order polynomial regression fit of the PM10216on the score. All control and treated observations are included in the control and treated fit, respec-217tively. The plot reveals a positive jump at the cut-off: the PM10 concentration seems to be higher in218cities north of the river than those to the south.219Figure 4 is a scatterplot of house prices of various cities against their degree distance north of the220river. A simple linear regression was also run, and we plotted the fitted values on top of the scatterplot.221The slight monotonic decline of the fitted line indicates a negative correlation between distance in222latitude from the Huai River and the house prices.223In addition to the graphical evidence, Table 3 reports the empirical results from estimating Equa-224tions (2a) and (2b). All the estimates are adjusted for the full set of covariates reported in the data225sources section. Column [1] reports the fixed effect estimates of the coefficient on a dummy variable226Table 2. Differences in key determinants of house prices at the Huai River.227Variable228North229South230Difference in means231PM10 (μg/m3)232119.413323386.56234−32.85333***(3.058324)235SO2 (mg/m3)2360.05086672370.0339867238−0.01688***(0.0020688)239NO2 (mg/m3)2400.043542410.0416533242−.00018867*(0.0013518)243Real House Price (yuan/m2)2445442.612456390.02246−947.41**(487.70)247Real House Price (log)2488.432498.622500.19*** (0.06)251Per capita real GDP(10,000 yuan)2526.4493672536.6264772540.17711(0.6095278)255Real wage(yuan)25645047.325745,176.63258129.3319(3142.108)259Fiscal surplus (100 million yuan)260−148.8867261−155.4967262−6.610044(28.82844)263Population density(people/km2)2641652.9982651582.079266−70.91901(159.132)267Per capita floor space (m2)2687.9378152698.1886312700.2508162(0.4990873)271Students enrolled at higher education272379,014.2273488,100.1274109,085.9*** (25,285.29)275No. of hospitals per 10,000 people2760.5988742770.4537101278−0.1451639*** (0.0477961)279***Significant at the 1% level.280** Significant at the 5% level.281* Significant at the 10% level.282283284===== PAGE 11 (score 3, numeric-paren 0, table-words 1, var-words 1) =====285286NEW ZEALAND ECONOMIC PAPERS2879288Figure 3. Fitted value from RD regression of PM10 exposure on distance from the Huai River.289Figure 4. A linear fit from regression of housing price on distance from the Huai River.290indicating whether it is located to the north of the river after controlling for a third-order polyno-291mial term in distance from the river as well as its interaction with the north river dummy using the292full sample. Columns [2] and [3] apply the parametric and non-parametric RD approach to estimate293these two equations. Column [2] uses the subsample within 7° latitude of the Huai River which was294imposed manually as our discontinuity sample. Conversely, in Column [3], the estimates are based on295the local linear regression with a triangular kernel and optimal bandwidth (h) selected by the method296prescribed by Imbens and Kalyanaraman (2012).297Overall, Table 3 presents statistical evidence of a considerable rise in PM10 and a decrease in hous-298ing price at the Huai River divide, which is similar to what we have presented in Figures 3 and 4. At the299300301===== PAGE 12 (score 29, numeric-paren 16, table-words 5, var-words 2) =====30230310304X. LIU ET AL.305Table 3. RD estimates of the impact of the Huai River Policy.306Outcome307[1]308[2]309[3]310PM1031119.1157**31241.251***31347.47836*314(8.64)315(7.7153)316(28.10 87)317Housing price (log)318–0.272**319−0.42026***320–0.4306474***321(0.107)322(0.05896)323(0.1575027)324RD type325–326polynomial327Local Linear Regression(LLR)328Polynomial function329Third330Linear331Sample332All3337°334***Significant at the 1% level.335** Significant at the 5% level.336* Significant at the 10% level.337Table 4. Comparing fixed effect and RD estimates of PM10’s impact on house prices.338Outcome339[1]340[2]341[3]342Housing price (log)343–0.001344–0.0101651*345–0.0083*346(0.0008)347(0.0058542)348(0.00467)349Estimation method350Fixed effect351IV352IV353RD Type354–355Local Linear Regression(LLR)356Local Linear Regression(LLR)357Sample358All3597°360***Significant at the 1% level.361** Significant at the 5% level.362* Significant at the 10% level.363threshold, the PM10 increase by 19 μg/m3 as indicated by conventional method, whereas in the RD364design, the increase is substantially larger, by 41–47 μg/m3, respectively. In comparison, the log term365of house prices declines as we move across the border by 27% and 42% for the conventional and para-366metric RD approach. The estimates using nonparametric approach are of slightly larger magnitude:367the estimated decline in log (house prices) of cities north of the river is 43%.3685.2. Estimates of the effect of PM10 on house prices369Table 4 reports the estimated effect of PM10 on house prices based on a variety of approaches. Col-370umn [1] reports the results from application of the conventional fixed effect method, which is used as371a baseline model to compare the estimates with those based on RD and 2SLS approaches. Specifically,372the fixed effect estimator of the coefficient of PM10 exposure on house prices is statistically insignifi-373cant and marginal in size: a 1 unit increase in PM10 is correlated with a 0.1% increase in housing price,374but as noted above the parameter estimate is insignificantly different from zero. Additionally, fuzzy375RD using a restricted sample based on the data-driven approach and bandwidth selected manually376are presented in Columns [2] and [3], respectively. They are statistically significant at 10% significant377level. Overall, the results indicate that the fuzzy RD approach delivers a larger coefficient in magni-378tude compared to conventional estimates. This lends support to the proposition that omitted variables379may have attenuated the estimate of the critical parameter, and the true value of clean air.3806. Robustness381We examine the robustness of our results by (i) using placebo cut-off lines; and (ii) testing whether our382results were affected qualitatively by the decisions made in our paper along several dimensions, such383as adjustment for covariates, sample selection, functional forms of our models, and kernel weighting384methods.385First, we repeat the estimation of fuzzy RD using placebo cut-offs. While we know the Huai386River line is the true cut-off of interest due to the indoor heating policy in the 1950s, we consider387388389===== PAGE 13 (score 28, numeric-paren 14, table-words 6, var-words 1) =====390391NEW ZEALAND ECONOMIC PAPERS39211393Figure 5. RD placebo tests on the impact of PM10 on house prices.394Table 5. Comparing the fixed effect and regression discontinuity estimates of PM10’s impact on house prices.3952SLS Polynomial RD396(within 7 degrees)397Fixed effect398Fixed effect (within3997 degrees)400Local Linear Regressions401[1]402[2]403[3]404[4]405[5]406[6]407[7]408[8]409House prices410–0.001411–0.0011412–0.0197***413–0.0083*414–0.0204***415–0.0087*416–0.0237***417–0.0116*418(0.0008)419(0.0008)420(0.00557)421(0.00467)422(0.00582)423(0.0047)424(0.00757)425(0.0066)426Obs.427294428200429196430200431196432200433196434Controls435Yes436Yes437No438Yes439No440Yes441No442Yes443Kernel444Triangle445Triangle446Epanech.447Epanech.448Uniform449Uniform450***Significant at the 1% level.451** Significant at the 5% level.452* Significant at the 10% level.453demarcations close to the Huai River as the cut-offs for the placebo tests. We redraw a window of454four displacements at 0.5°-latitude intervals north and south of the Huai River across China as well455as at the actual Huai River (which is equivalent to the 0° displacement) and re-estimate our key coef-456ficient of interest. The results from these falsification exercises are shown in Figure 5. According to457Figure 5, the coefficient is only statistically significant at the actual Huai River. This suggests that the458previous findings reported in this paper are a result of the Huai River Policy rather than an artefact459of the application of the RD approach to the data.460In addition, we examine the robustness of our results based on tests from several dimensions,461including adjustment for covariates, functional forms of our models, and kernel weighting methods.462In Table 5, we expand the results of Table 4 by including a full set of parametric estimates with and463without adjusting for the covariates, and local linear estimates using a variety of kernel weighting464methods. The entries in columns [1] and [4] are identical to those in Table 4, which are adjusted465using the full set of covariates and a linear polynomial in distance to the Huai River. Columns [3],466[5] and [7] report the estimated coefficients excluding the covariates. Using a triangular kernel, the467entries in column [4] match the local linear results in Table 4, while estimates using an Epanechnikov468and uniform kernel are presented in columns [5] to [8], respectively.469470471===== PAGE 14 (score 32, numeric-paren 15, table-words 5, var-words 6) =====47247312474X. LIU ET AL.475Table 6. Robustness checks of choice of functional form for latitude.476R¸estricted sample (within 7 degrees)477Linear478Quadratic479Cubic480Quartic481Quintic482[1]483[2]484[3]485[4]486[5]487Housing price488–0.00833*489–0.0081**490–0.01204***4910.08037***492−0.01249***493(0.00467)494(0.00333)495(0.00284)496(0.01939)497(0.00081)498***Significant at the 1% level.499** Significant at the 5% level.500* Significant at the 10% level.501Table 7. Comparing fixed effect and RD estimates of PM10’s impact on house prices (subsample excluding first-tier cities).502Outcome503[1]504[2]505[3]506Housing price (log)507–0.0010686508–0.0063145*509–0.00743*510(0.0007496)511(0.0036558)512(0.00489)513Estimation method514Fixed effect515IV516IV517RD Type518-519Local Linear Regression(LLR)520Local Linear Regression(LLR)521Sample522All5237°524Moreover, we report estimates from a number of specifications to check if the results are sensi-525tive to the order of the polynomial used. In Table 6, we adopt alternative approaches to implement526the parametric RD design by using various functional forms, varying from a linear polynomial inter-527acted with a North dummy to a quartic polynomial subsequently as one moves from left to right. We528consider the restricted sample of cities within 7 degrees latitude of the Huai River in columns [1] to529[5]. For example, columns [1], [2] and [3] include degree distance to the Huai River, its square, and its530cube, and each term’s interaction with a North dummy in the discontinuity sample. The results show531that except for column [4], the figures are all qualitatively similar to those presented in the baseline532specifications.533Noticeably, two out of three first-tier cities of our sample (Shanghai, Guangzhou and Beijing) are534located south of the Huai river (see Appendix for the location of our sample cities relative to the Huai535river), and they all have extremely high house prices. This may affect the major finding of the paper,536i.e. the negative correlation between air quality and housing price. To account for this big city effect537which may drive our findings, we conduct the same analysis as the baseline using a subsample that538excludes Beijing, Shanghai and Guangzhou. The results are presented in Table 7. As can be seen, the539results remain qualitatively similar to those of Table 4. This may, to some extent, reduce the concern540about the big city effect.5417. Conclusion542Having good estimates of the value of clean air is critical to the cost–benefit analysis of policy inter-543vention regarding environmental protection. In this paper, we draw on a unique intervention with544respect to the use of coal-fired heating amongst households in China in order to estimate the elastic-545ity of house prices with respect to concentrations of PM10. The Huai River Policy generated a stark546difference in air quality in southern and northern China. Our analysis exploits the structure of this547unique Chinese Huai River Policy divide to provide new evidence of the extent to which amenities548from the level of pollution is reflected in house prices.549We employ a regression discontinuity (RD) design to estimate the impact of air pollution on house550prices across a river that demarcates regions with and without coal-fired heating resulting from the551Huai River Policy. This policy – which was decreed by the Chinese government in the 1950s – enabled552the burning of coal at subsidised prices for indoor heating only to the north of the Huai River. Using a553quasi-experimental variation in particulate matter of 10 micrometres or less in aerodynamic diameter554(PM10) and a regression discontinuity (RD) design based on distance from Huai River, we estimate555556557===== PAGE 15 (score 14, numeric-paren 0, table-words 0, var-words 14) =====558559NEW ZEALAND ECONOMIC PAPERS56013561the local average treatment effect (LATE) of PM10 air pollution on house prices. By using panel data562of 30 large cities on either side of the river for the period 2006–2015, we found that 1 μg/m3 reduction563in average PM10 results in a 1% increase in the price of housing.564The findings also confirm the importance of instrumenting for local air pollution to reduce prob-565lems of endogeneity. Our econometric analysis reveals that conventional fixed effect regressions tend566to underestimate the value of clean air. Accounting for problems of endogeneity improves the esti-567mates, providing statistically significant parameter estimates of the effect of PM10 on house prices.568These estimates of the economic value of clean air are robust to adjustment in covariates and are less569sensitive to model specification.570Extrapolation from US-based studies suggests that the costs of poor air are substantial in terms of571health effects and welfare in general. This paper shows that the value of clean air in China exceeded572that for the US. A 1 μg/m3 reduction in TSPs results in a 0.2–0.4% increase in mean housing values573in the US (Chay & Greenstone, 2005), compared to an approximately 1% increase in housing price in574China based on our estimates. Our findings therefore suggest that the value of clean air may increase575with the level of pollution.576Notes5771. Total suspended particles (TSP) measures the mass concentration of particulate matter in the air. Within TSP, PM10578stands for particles with a diameter of 10 μm or less, and PM2.5 stands for those with a diameter of 2.5 μm or less.579Particulates that are 10 μm or greater are filtered and generally do not enter the lungs. Particulates smaller than58010 μm are likely to enter the lungs. Particulate matter that is smaller than 2.5 μm (PM2.5) can enter into the Alveoli581where gas exchange occurs. Throughout the world, ambient monitoring now focuses on PM10 and PM2.5.5822. We exclude Lhasa (capital city of Tibet) in our sample and restrict the sample to the year 2006–2015 due to583availability of certain covariates.584Disclosure statement585No potential conflict of interest was reported by the author(s).586ORCID587Satish Chand588http://orcid.org/0000-0003-3524-1878589References590Almond, D., Chen, Y., Greenstone, M., & Li, H. (2009). Winter heating or clean air? Unintended impacts of China’s591Huai River Policy. American Economic Review, 99, 184–190.592Ambrey, C. L., Fleming, C. M., & Chan, A. Y.-C. (2014). Estimating the cost of air pollution in South East Queensland:593An application of the life satisfaction non-market valuation approach. Ecological Economics, 97, 172–181.594Angrist, J. D., & Lavy, V. (1999). Using Maimonides’ rule to estimate the effect of class size on scholastic achievement.595The Quarterly Journal of Economics, 114, 533–575.596Angrist, J. D., & Pischke, J.-S. (2008). Mostly harmless econometrics: An empiricist’s companion. Princeton, NJ: Princeton597University Press.598Bayer, P., Keohane, N., & Timmins, C. (2009). Migration and hedonic valuation: The case of air quality. Journal of599Environmental Economics and Management, 58, 1–14.600Besley, T., & Mueller, H. (2012). Estimating the peace dividend: The impact of violence on house prices in Northern601Ireland. American Economic Review, 102, 810–833.602Black, S. E. (1999). Do better schools matter? Parental valuation of elementary education. The Quarterly Journal of603Economics, 114, 577–599.604Brunekreef, B., & Holgate, S. T. (2002). Air pollution and health. The Lancet, 360, 1233–1242.605Chay, K. Y., & Greenstone, M. (2005). Does air quality matter? Evidence from the housing market. Journal of Political606Economy, 113, 376–424.607Chen, Y., Ebenstein, A., Greenstone, M., & H, L. I. (2013). Evidence on the impact of sustained exposure to air pol-608lution on life expectancy from China’s Huai River Policy. Proceedings of the National Academy of Sciences, 110,60912936–12941.610