# RECORD 7670a6d2dc2183fc23719395205238727f178317 # TITLE: Spatial Estimation of the Impact of Airport Noise on Residential Housing Prices # YEAR: 2008.0 VENUE: nan DOI: 10.1007/bf03399267 # AUTHORS: M. Salvi # PAGES: 30 CANDIDATE TABLE PAGES (1-based): [11, 18, 22, 23] # ABSTRACT: SummaryThe paper applies spatial econometric techniques to measure the impact of airport noise on the price of single-family homes in the Zurich Airport area. We exploit a large database of geo-referenced noise measurements to investigate the reaction of prices to different noise metrics. The particular institutional setting of Zurich Airport, with changing patterns of runways configurations allows to distinguish the impact of noise at different times of the day. The use of neighborhood fixed-effects is compared to the results given by a costlier modelling strategy involving a rich set of location descriptors. The paper documents the impact of airport noise on housing prices. In the base model specification the Noise Discount Index is 0.97% with typical discounts in the range of −2% to −8%. Accounting for the spatiality of the data has little effect on the results. ===== PAGE 1 (score 6, numeric-paren 0, table-words 0, var-words 6) ===== © Swiss Journal of Economics and Statistics 2008, Vol. 144 (4) 577–606 a Ecole Polytechnique Fédérale, Lausanne, and Zürcher Kantonalbank, Zürich. Email: marco. salvi@epfl.ch. I am grateful to Patrik Schellenbauer, Philippe Thalmann, Adrian Lüscher, Gion-Reto Hassler, Alice Hollenstein and the editors for their helpful comments and sug- gestions. I thank the ZKB GIS-team for invaluable help with the GIS. The views expressed herein are my own and do not necessarily represent the views of the ZKB. Spatial Estimation of the Impact of Airport Noise on Residential Housing Prices Marco Salvia JEL-Classification: Q53, Q51, R31, L93, C21 Keywords: noise, airports, housing prices, hedonic pricing, spatial econometrics 1. Introduction The impact of airport proximity on neighboring communities is a hotly debated topic. The positive aspects of proximity are related to the provision of communi- cation links and to the direct economic importance of large airports. The down- side of proximity are the adverse environmental effects primarily associated with aircraft noise. The growing interdependence and global connectivity of modern economies is bound to increase the economic importance of air travel. On the other hand, environmental quality tends to be a superior good – as per capita income rises, the social demand for environmental quality increases more than proportionally. It is fair to say that airport noise is likely to fuel ongoing public debate. The recent story of Zurich Airport is in this regard exemplary. Until the year 2000 the use of German airspace for approaching and leaving Zurich airport had been governed by a bilateral agreement between Switzerland and Germany. The agreement was terminated by Germany in 2000, forcing the Swiss authorities to set out new landing procedures. As a result, the number of residents subject to aircraft noise increased significantly. Under the Swiss noise protection law these residents may be entitled to compensation and over 19 000 have filed a claim. This has drastically increased the public awareness for the evaluation of the cost of noise. Specifically, the use of the hedonic method for the scope of evaluation and compensation has been proposed. Recently, the Federal Supreme Court of ===== PAGE 11 (score 34, numeric-paren 0, table-words 4, var-words 22) ===== Spatial Estimation of the Impact of Airport Noise on Residential Housing Prices 587 Table 1: Descriptive Statistics of the House Characteristics Variable Mean Median Min Max Std. Dev House price [CHF 1000] 772.16 727.00 155.00 2 600.00 262.94 Continuous variables) Lot size [m2] 521.62 447.00 150.00 2 430.00 315.61 Size, volume of building [m3] 723.12 691.00 350.00 1 995.00 220.67 Number of rooms 5.49 5.00 2.00 11.00 1.15 Number of bathrooms 2.27 2.00 1.00 6.00 0.73 Age of building [years] 30.76 21.00 0.00 156.00 32.67 Categorical variables Condition of the dwelling New 0.36 Renovated 0.13 Well maintained 0.41 Needs renovation 0.10 Brick building 0.95 Detached 0.46 Cellar 0.95 Double glazing 0.66 Floor heating 0.45 Garage 0.49 Underground garage 0.25 Modern kitchen 0.33 Swimming pool 0.02 Sauna 0.04 The table reports the descriptive statistics of the 3 737 observations included in the final sample. The original data set contained 3 967 transactions single-family homes sold at arm’s length in the Canton of Zurich between 1995 and 2005. 230 observations (5.8%) were excluded because of likely data entry mistakes, of which 178 transactions with very small lot sizes under 150 m2. We suspect that for these transactions, the house living area was erroneously entered. ===== PAGE 12 (score 24, numeric-paren 0, table-words 1, var-words 15) ===== 588 Marco Salvi Table 2: Descriptive Statistics of the Location Characteristics Variable Mean Median Min Max Std. Dev Aircraft noise (in excess of 50 dB) Leq 16h [dB] 1.11 0.00 0.00 20.40 2.84 Leq Evening [dB] 0.24 0.00 0.00 16.10 1.25 Leq 1h Peak [dB] 0.67 0.00 0.00 8.20 1.50 Location variables Road traffic noise, Lr 16h [dB] 0.78 0.00 0.00 18.50 2.55 View on lakes [ha] 512.00 0.00 0.00 6 971.00 1 236.00 View on Alps [km2] 187.83 168.83 0.00 986.48 132.56 Travel distance to CBD [min] 32.9 33.00 12.00 56.00 8.01 Near power line ( < 200m) 0.03 0.00 0.00 1.00 0.16 Near railway ( < 100m) 0.09 0.00 0.00 1.00 0.28 Slope terrain [%] 4.59 3.83 0.13 21.33 3.49 Aspect: East 0.28 0.00 0.00 1.00 0.45 Aspect: West 0.44 0.00 0.00 1.00 0.50 Share of Swiss residents 0.84 1.00 0.00 1.00 0.36 Maximum built density [%] 43.48 40.00 15.00 333.00 18.89 Community tax revenue [Index] 106.78 90.93 42.34 511.42 51.95 The location variables were matched at the hectare level with each of the 3 737 single-family homes, sold in the Canton of Zurich between 1995 and 2005. The aircraft noise exposure and, where pos- sible, the other location variables are those of the transaction year. All noise variables are reported as the values in excess of 50 dB(A). Daytime car travel noise is measured with the Lr metric (rating level). The distance to CBD is measured as the car travel time to Zurich Main station. Aspect (orientation) dummies are computed only for terrains with a slope greater than 1%. The share of Swiss residents is computed in a neighborhood of 300m around the transaction. View variables are computed at 4 meters above ground. The density variables report the maximum allowed built density on the lot expressed as the ratio of the living space to the lot size. The community tax is the communal per capita tax revenue. See the main text for further explanations. ===== PAGE 18 (score 27, numeric-paren 0, table-words 2, var-words 13) ===== 594 Marco Salvi 11 ML estimation requires the repeated evaluation of the information matrix, which is computa- tionally demanding in this context (Pace and Gilley, 1998). Therefore, we use a sparse matrix implementation. of the spatial model in Table 5 (Model 1 and 2) for different specifications of the aircraft noise, alongside the results of the OLS estimation of a non-spatial specification (Model 3).11 Table 5: Estimation Results: SAR Models and Non-Spatial Hedonic Regression Model 1 (ML SAR) Model 2 (ML SAR) Model 3 (OLS) Variable Estimate SE Estimate SE Estimate SE Intercept 9.2801 0.1337 9.3899 0.1083 9.2325 0.1253 Leq 16h −0.0098 0.0010 −0.0070 0.0013 −0.0098 0.0009 Leq Evening −0.0064 0.0019 Leq 1h Peak −0.0063 0.0023 Age of building −0.0028 0.000 −0.0029 0.0003 −0.0026 0.0003 Age squared 0.0007 0.000 0.0007 0.0002 0.0006 0.0002 New 0.1377 0.012 0.1373 0.0116 0.1358 0.0119 Renovated 0.1451 0.009 0.1452 0.0093 0.1419 0.0096 Well maintained 0.1126 0.008 0.1125 0.0080 0.1096 0.0082 Brick building 0.0284 0.009 0.0290 0.0092 0.0298 0.0094 Detached 0.0256 0.006 0.0264 0.0056 0.0253 0.0057 Cellar 0.0476 0.010 0.0478 0.0103 0.0487 0.0106 Double glazing 0.0312 0.005 0.0317 0.0055 0.0322 0.0056 Floor heating 0.0356 0.006 0.0359 0.0064 0.0357 0.0066 Garage 0.0253 0.005 0.0257 0.0055 0.0253 0.0056 Underground garage 0.0290 0.007 0.0293 0.0071 0.0264 0.0072 Modern kitchen 0.0589 0.005 0.0586 0.0052 0.0592 0.0054 Swimming pool 0.0637 0.015 0.0641 0.0147 0.0653 0.0152 Sauna 0.0396 0.011 0.0396 0.0109 0.0379 0.0113 ===== PAGE 19 (score 32, numeric-paren 0, table-words 2, var-words 14) ===== Spatial Estimation of the Impact of Airport Noise on Residential Housing Prices 595 Model 1 (ML SAR) Model 2 (ML SAR) Model 3 (OLS) Variable Estimate SE Estimate SE Estimate SE Log lot size 0.1691 0.006 0.1692 0.0059 0.1646 0.0060 Log size 0.4163 0.011 0.4155 0.0111 0.4206 0.0113 Log rooms 0.1243 0.013 0.1246 0.0130 0.1223 0.0134 Log bathrooms 0.1005 0.008 0.0998 0.0078 0.1018 0.0080 Share Swiss residents 0.0149 0.006 0.0157 0.0064 0.0159 0.0063 Road traffic noise −0.0050 0.001 −0.0051 0.0009 −0.0053 0.0009 Log travel dist. to CBD −0.2310 0.015 −0.2354 0.0151 −0.2290 0.0133 Low density (< 40%) −0.0034 0.006 −0.0040 0.0059 −0.0044 0.0056 Middle density (40–50%) −0.0024 0.007 −0.0029 0.0068 −0.0034 0.0065 Higher density (> 50%) −0.0117 0.007 −0.0126 0.0072 −0.0132 0.0070 Lake view > 0–20 km2 0.0552 0.007 0.0514 0.0066 0.0563 0.0059 Lake view 20–40 km2 0.1390 0.012 0.1346 0.0118 0.1411 0.0107 Lake view > 40 km2 0.1121 0.014 0.1074 0.0138 0.1103 0.0120 View 50–100 km2 0.0359 0.009 0.0357 0.0091 0.0406 0.0084 View 100–250 km2 0.0236 0.008 0.0244 0.0078 0.0270 0.0071 View > 250 km2 0.0163 0.009 0.0183 0.0090 0.0159 0.0081 Slope 0–9% 0.0194 0.006 0.0191 0.0061 0.0191 0.0060 Slope > 9% 0.0357 0.008 0.0360 0.0080 0.0353 0.0076 Near power line −0.0103 0.015 −0.0133 0.0150 −0.0118 0.0140 Near railway −0.0243 0.008 −0.0237 0.0082 −0.0247 0.0078 Aspect: East −0.0032 0.007 −0.0034 0.0068 −0.0002 0.0064 Aspect: West 0.0102 0.006 0.0093 0.0063 0.0139 0.0060 Log tax revenue 0.1559 0.010 0.1568 0.0099 0.1549 0.0089 ρ 0.1852 0.1853 – LR test (p-value) < 0.001 < 0.001 – Residual standard error 0.1279 0.1275 0.1309 N 3737 3737 3737 Log likelihood 2364 2374 AIC −4625 −4640 For reasons of space, yearly time dummies are not reported here. They are available from the author upon request. ===== PAGE 22 (score 27, numeric-paren 0, table-words 2, var-words 12) ===== 598 Marco Salvi In spatial econometric applications, the choice of spatial matrix may cause greater differences in parameter estimates than the choice of estimation tech- nique. This is unsatisfactory as this choice remains largely untested and many possible matrices are compatible with the underlying economic story. We thus conclude the spatial econometric section by reporting the results of the estima- tion of Model 1 given alternative choices for the spatial matrices. The results are reported in Table 6, in the columns labelled Model 4 (for the weighting matrix with cut-off at 600m) and Model 5 (for the weighting matrix based on nearest- neighbor contiguity). Table 6: ML SAR Estimation Results with Alternative Weighting Schemes Model 4 (c = 600m) Model 5 (NN) Model 6 (No GIS) Variable Estimate SE Estimate SE Estimate SE Intercept 9.3800 0.1137 9.3701 0.1124 9.3226 0.0840 Leq 16h −0.0099 0.0011 −0.0099 0.0010 −0.0086 0.0023 Age of building −0.0030 0.0003 −0.0030 0.0003 −0.0034 0.0003 Age squared 0.0008 0.0002 0.0008 0.0002 0.0010 0.0002 New 0.1359 0.0116 0.1367 0.0116 0.1271 0.0117 Renovated 0.1443 0.0093 0.1448 0.0092 0.1491 0.0093 Well maintained 0.1119 0.0080 0.1119 0.0080 0.1111 0.0080 Brick building 0.0290 0.0093 0.0278 0.0093 0.0280 0.0093 Detached 0.0277 0.0056 0.0271 0.0056 0.0267 0.0057 Cellar 0.0468 0.0103 0.0458 0.0103 0.0441 0.0104 Double glazing 0.0340 0.0055 0.0326 0.0055 0.0291 0.0055 Floor heating 0.0333 0.0064 0.0349 0.0064 0.0326 0.0065 Garage 0.0254 0.0055 0.0259 0.0055 0.0278 0.0056 Underground garage 0.0270 0.0071 0.0289 0.0072 0.0287 0.0073 Modern kitchen 0.0587 0.0053 0.0585 0.0052 0.0634 0.0053 Swimming pool 0.0598 0.0147 0.0624 0.0147 0.0587 0.0149 Sauna 0.0391 0.0109 0.0378 0.0109 0.0526 0.0110 Log lot size 0.1700 0.0059 0.1693 0.0060 0.1786 0.0059 Log size 0.4165 0.0111 0.4158 0.0112 0.4223 0.0113 ===== PAGE 23 (score 29, numeric-paren 0, table-words 3, var-words 14) ===== Spatial Estimation of the Impact of Airport Noise on Residential Housing Prices 599 Model 4 (c = 600m) Model 5 (NN) Model 6 (No GIS) Variable Estimate SE Estimate SE Estimate SE Log rooms 0.1199 0.0130 0.1207 0.0130 0.1246 0.0131 Log bathrooms 0.0960 0.0077 0.0969 0.0077 0.0969 0.0078 Share Swiss residents 0.0160 0.0064 0.0169 0.0066 Road traffic noise −0.0053 0.0009 −0.0054 0.0009 Log travel dist. to CBD −0.2335 0.0162 −0.2307 0.0159 Low density (< 40%) −0.0077 0.0059 −0.0045 0.0060 Middle density (40–50%) −0.0065 0.0068 −0.0027 0.0069 Higher density (> 50%) −0.0145 0.0072 −0.0120 0.0073 Lake view > 0–20 km2 0.0545 0.0067 0.0557 0.0067 Lake view 20–40 km2 0.1318 0.0123 0.1357 0.0122 Lake view > 40 km2 0.1080 0.0146 0.1119 0.0144 View 50–100 km2 0.0319 0.0092 0.0344 0.0093 View 100–250 km2 0.0213 0.0081 0.0216 0.0080 View > 250 km2 0.0135 0.0093 0.0137 0.0092 Slope 0–9% 0.0171 0.0060 0.0177 0.0061 Slope > 9% 0.0357 0.0080 0.0345 0.0081 Near power line −0.0090 0.0151 −0.0064 0.0154 Near railway −0.0244 0.0080 −0.0228 0.0083 Aspect: East −0.0032 0.0068 −0.0031 0.0069 Aspect: West 0.0093 0.0063 0.0106 0.0065 Log tax revenue 0.1588 0.0107 0.1599 0.0105 ρ 0.2579 0.2233 0.1453 LR test (p-value) < 0.001 < 0.001 < 0.001 Residual standard error 0.1276 0.1274 0.1263 N 3737 3737 3737 AIC −4640 −4645 −4439 Model 4 uses a contiguity matrix with a cut-off value set at 600m, while in Model 5 it is con- strained to the four nearest neighbors. For Model 6 the same weighting matrix as in Model 1 of Table 5 is used, with GIS-matched location variables replaced by community indicator variables. For reasons of space, both the time and the community dummies are not reported. ===== PAGE 24 (score 4, numeric-paren 0, table-words 1, var-words 2) ===== 600 Marco Salvi Both alternative weighting matrices are associated with marginally higher, posi- tive spatial autocorrelation parameters ρ. Accordingly, the standard error of the estimated coefficients are somewhat higher. Nevertheless, as far as the added value of the spatial econometric specification is concerned, this analysis confirms the previous negative results. 4.2.3 Robustness of Results Higher noise levels may exert a disproportionate impact on house prices – an effect not captured by a crude linear specification. In order to examine this issue we reformulate Model 1 as a generalized additive model. This nonpara- metric estimation method relaxes the assumption of log-linearity of the mean noise exposure, allowing for a more flexible functional form (Hastie and Tib- shirani, 1990). The nonparametric component is modelled as a third-order spline. Figure 3 illustrates the result of the estimation where we report the non- parametric fit with the respective 95%-confidence interval. Figure 3. Non-Parametric Estimation of the Impact of Mean Noise Exposure. Leq 16h over 50 dB Effect on log house price 0.00 −0.05 −0.10 −0.15 −0.20 −0.25 0 5 10 15 20 The nonparametric component is modelled as a third-order spline. The dotted lines represent the 95% confidence interval.