1# RECORD b51ae7fa8ad6ec948d8787f76cdf0f315d38c6f62# TITLE: Spatial Heterogeneity in Hedonic House Price Models: The Case of Austria3# YEAR: 2014.0 VENUE: nan DOI: 10.1177/00420980134922344# AUTHORS: M. Helbich; Wolfgang A. Brunauer; E. Vaz; P. Nijkamp5# PAGES: 24 CANDIDATE TABLE PAGES (1-based): [14, 16]6# ABSTRACT: nan789===== PAGE 1 (score 1, numeric-paren 0, table-words 0, var-words 1) =====1011TI 2013-171/VIII 12Tinbergen Institute Discussion Paper 13 14Spatial Heterogeneity in Hedonic House Price 15Models: 16The Case of Austria 17 18 19 20Marco Helbich1 21Wolfgang Brunauer2 22Eric Vaz3 23Peter Nijkamp4 24 25 26 271 University of Heidelberg, Germany; 282 UniCredit Bank Austria AG, Austria; 293 Ryerson University, Canada; 304 Faculty of Economics and Business Administration, VU University Amsterdam, and Tinbergen 31Institute. 32 33 34 353637===== PAGE 14 (score 76, numeric-paren 70, table-words 1, var-words 4) =====383912 40 41Burgenland have a significantly lower price compared with houses situated in Lower Austria, serving 42as reference category. 43Tab. 2: Estimation results of the global models. 44OLS 45SAR 46S2SLS 47Coeff. 48HC Std. 49err. 50t-val. 51Coeff. 52Std. 53err. 54z-val. 55Coeff. 56SHAC Std. 57err. 58t-val. 59Intercept 609.391 610.195 6248.278*** 6.872 630.328 6420.969*** 6.542 650.548 6611.939**67* 68lnarea_total 690.410 700.017 7123.988*** 0.409 720.018 7322.825*** 0.411 740.017 7523.864**76* 77lnarea_plot 780.111 790.012 808.945*** 810.111 820.013 838.855*** 840.111 850.012 868.891*** 87cond_house_1 88-890.052 900.014 91-3.742*** 92-930.045 940.014 95-3.330** 96-970.044 980.014 99-3.229** 100heat_1 101-1020.102 1030.020 104-5.018*** 105-1060.107 1070.023 108-4.602*** 109-1100.105 1110.021 112-5.005*** 113bath_1 114-1150.076 1160.019 117-3.910*** 118-1190.074 1200.022 121-3.339*** 122-1230.075 1240.019 125-3.985*** 126attic_1 127-1280.024 1290.012 130-2.018* 131-1320.024 1330.011 134-2.077* 135-1360.024 1370.012 138-2.020* 139basement_1 1400.132 1410.013 1429.936*** 1430.126 1440.013 1459.597*** 1460.124 1470.014 1489.069*** 149garage_1 150-1510.082 1520.012 153-6.899*** 154-1550.084 1560.012 157-7.162*** 158-1590.084 1600.012 161-7.064*** 162terr_1 1630.070 1640.012 1655.893*** 1660.066 1670.012 1685.673*** 1690.066 1700.012 1715.574*** 172age 173-1740.006 1750.000 176-17716.976*** 178-1790.006 1800.000 181-18217.617*** 183-1840.006 1850.000 186-16.434 187time 1880.009 1890.003 1903.550*** 1910.010 1920.002 1934.100*** 1940.010 1950.003 1963.925*** 197muni_pp_ind 1980.005 1990.001 2005.800*** 2010.005 2020.001 2035.890*** 2040.005 2050.001 2065.709*** 207muni_acad 2080.011 2090.002 2105.695*** 2110.009 2120.002 2134.384*** 2140.008 2150.002 2163.692*** 217muni_unempl 218-2191.454 2200.649 221-2.241* 222muni_age_ind 223-2240.020 2250.004 226-5.538*** 227-2280.017 2290.004 230-4.788*** 231-2320.016 2330.004 234-4.255*** 235ln_muni_popd 2360.033 2370.007 2384.518*** 2390.020 2400.006 2413.153** 2420.020 2430.006 2443.245 245enumd_acad_dmuni 2460.008 2470.001 2486.072*** 2490.006 2500.001 2514.997*** 2520.005 2530.001 2544.232** 255enumd_unempl_dmun256i 2571.758 2580.607 2592.898** 260enumd_age_ind_dmu261ni 2620.004 2630.003 2641.568*** 265Burgenland 266-2670.129 2680.023 269-5.644*** 270-2710.096 2720.025 273-3.905*** 274-2750.087 2760.024 277-3.671*** 278Carinthia 279-2800.082 2810.024 282-3.397*** 283-2840.093 2850.024 286-3.912*** 287-2880.092 2890.024 290-3.801*** 291Upper Austria 2920.018 2930.021 2940.870 2950.016 2960.019 2970.832 2980.012 2990.019 3000.648 301Vorarlberg & Tyrol 3020.241 3030.042 3045.791*** 3050.182 3060.036 3075.023*** 3080.179 3090.041 3104.340*** 311Salzburg 3120.231 3130.030 3147.748*** 3150.190 3160.028 3176.796*** 3180.185 3190.031 3206.054*** 321Styria 322-3230.055 3240.020 325-2.760** 326-3270.056 3280.020 329-2.805** 330-3310.055 3320.020 333-2.759** 334Vienna 3350.266 3360.034 3377.759*** 3380.220 3390.031 3407.011*** 3410.215 3420.034 3436.401*** 344𝜌 3450.205 346p < 0.001 3470.229 348p < 0.001 349Adj. R² 3500.473 3510.486 352AIC 3532783 3542707 355AIC without regions 3562988 3572850 358RMSE 3590.344 3600.339 3610.339 362RMSE without regions 3630.353 3640.345 3650.345 366Signif.: '***' 0.001, '**' 0.01, '*' 0.05, '.' 0.10 367 368 369370371===== PAGE 15 (score 8, numeric-paren 1, table-words 1, var-words 5) =====37237313 374 375 3765.3 Local and semi-local model 377 378Next, GWR and MGWR are estimated. A data-driven approach to explore functional instability 379replaces the federal states dummies. For both models the analysis is restricted exclusively to the 380main effects. Otherwise, compositional effects would result in complex and difficult interpretational 381local effects. Because of a significant compositional effect of the unemployment rate and an 382insignificant corresponding main effect in the S2SLS model, the unemployment rate is additionally 383considered at the enumeration district level. To achieve a trade-off between computational burden 384and estimation accuracy, (a) a random sample of 50% of the overall data set is used and (b) only 385those variables are chosen that are significant in the global S2SLS model (Table 1). 386 387First, it is necessary to distinguish between global and local effects using non-parametric GWR with a 388Gaussian kernel. CV determines an optimal adaptive bandwidth that includes 227 neighboring 389houses. The test statistic after Leung et al. (2000a) does not reject the H0 of stationary parameters 390for six out of 16 variables. In a MGWR specification, those six variables are thus held spatially 391constant, while the remaining 10 variables are allowed to vary. Again, applying a Gaussian kernel 392function, CV results in an optimal bandwidth of 177 neighboring houses for the MGWR. Compared to 393the GWR the AIC score is noticeably reduced from 1,396 (GWR) to 1,381 (MGWR), while the RMSE 394for both models is quasi similar (0.338). Statistically the MGWR is favored. Sensitivity analysis with 395alternative kernels (i.e. bi-square kernel) shows no significant differences. 396 397The spatial variation of the model fit ranges from 0.24 to 0.49. The lowest values are located in 398Vienna plus its surroundings and in the south of Austria. The best fit is achieved in the northern parts 399of Austria. The reason for the low fit in Vienna might be that flats dominate the market in the capital. 400Figure 2 visualizes the parameter surfaces of the local R² and the predicted dwelling prices in Euro, 401achieved through ordinary kriging (Pebesma, 2004). The F(3)-test (Leung et al., 2000a) confirms that 402now all spatially varying covariates are statistically significant. No significant residual dependence on 403a 0.01 level is detected by the Moran's 𝐼 (Leung et al., 2000b). 404 405 406Fig.2: Local model performance and the predicted house prices. 407 408409410===== PAGE 16 (score 35, numeric-paren 24, table-words 2, var-words 7) =====41141214 413 414Table 3 lists the MGWR results and its global counterparts. Additionally, Figures 3 through 5 visualize 415the parameter surfaces. Based on pseudo-t-values, only significant coefficients on 𝛼= 0.1 have gray 416tones. 417 418Tab. 3: Estimation results of the semi-local model. 419 420Global coeff. 421 422Local coeff. 423 424 425Estim. Std.err. t-val. 4261. quantile Median 3. quantile F(3)-test 427lnarea_total 4280.422 0.024 42917.453*** 0.367 4300.398 4310.458 432*** 433lnarea_plot 4340.077 0.018 4354.268*** 4360.061 4370.087 4380.111 439*** 440heat_1 441-0.111 0.031 442-3.544*** 443-0.163 444-0.125 445-0.072 446* 447Age 448-0.005 0.000 449-11.441*** -0.006 450-0.005 451-0.004 452*** 453Time 4540.006 0.003 4551.990* 4560.007 4570.011 4580.013 459*** 460enumd_unempl -2.103 0.587 4613.581*** 462-2.335 463-1.520 464-2.236 465*** 466muni_pp_ind 4670.004 0.001 4684.442*** 469-0.001 4700.006 4710.007 472*** 473muni_acad 4740.012 0.003 4754.567*** 4760.007 4770.015 4780.022 479*** 480muni_age_ind 481-0.038 0.005 482-8.260*** 483-0.037 484-0.028 485-0.022 486*** 487ln_muni_popd 4880.067 0.008 4898.623*** 4900.036 4910.065 4920.079 493*** 494cond_house_1 495-0.038 0.020 496-1.953. 497bath_1 498-0.064 0.031 499-2.029* 500attic_1 501-0.028 0.016 502-1.674. 503basement_1 5040.144 0.019 5057.665*** 506garage_1 507-0.068 0.017 508-4.060*** 509terr_1 5100.076 0.017 5114.587*** 512Signif.: '***' 0.001, '**' 0.01, '*' 0.05, '.' 0.10 513 514 515 516 517Fig. 3: Physical covariates and temporal effect. 518 519 520521522===== PAGE 17 (score 4, numeric-paren 0, table-words 0, var-words 4) =====52352415 525 526 527Fig. 4: Temporal effect and neighborhood covariates. 528 529 530 531Fig. 5: Neighborhood covariates. 532 533The negative MGWR intercept is not, by itself, meaningful. The spatially fixed coefficients consist 534solely of physical covariates, meaning buyers value these housing characteristics equal across space. 535Furthermore, these stationary coefficients illustrate that real estate markets are economically 536connected. Compared to the S2SLS estimates, the MGWR coefficients do not considerably deviate, 537which is an indication of model suitability. 538 539The total floor area is statistically significant across the entire study site and shows strong spatial 540variation (Figure 3). Its positive effect on house prices is less pronounced closer to Vienna and 541reaches its largest positive effects in parts of the federal states of Salzburg and Tyrol. These areas 542include well-established winter sports sites (e.g., Kitzbühel) where tourism and vacation homes, 543secondary residences etc. increase house prices. Particularly, the scarcity of space leads to higher 544prices for additional floor area in these regions. In contrast, a different pattern can be found for 545