1# RECORD f8260edb97a0dbb7c5a255fca59c10da9cde8ca82# TITLE: Real estate listings and their usefulness for hedonic regressions3# YEAR: 2020.0 VENUE: Empirical Economics DOI: 10.1007/s00181-020-01992-34# AUTHORS: Jens Kolbe; Rainer Schulz; Martin Wersing; A. Werwatz5# PAGES: 31 CANDIDATE TABLE PAGES (1-based): [10, 21]6# ABSTRACT: Real estate platforms provide a new source of data which has already been used as a substitute for transaction data in hedonic regression applications. This paper asks whether it is valid to do so in the established research areas of (1) willingness to pay estimation, (2) automated valuations, and (3) price index construction. It therefore compares listings and transaction data and regression results derived from them. We find that ask prices stochastically dominate sale prices, mainly because the composition of characteristics differs between the two data sets. But estimates of implicit prices also differ. As a result, willingness to pay estimates from listings data can be widely off when compared with estimates from transaction data. Listings data are not very useful to predict market values of individual houses either, as these predictions suffer from upward bias and large error variance. We find, however, that an ask price index complements a sale price index, as it is useful for nowcasting.789===== PAGE 1 (score 4, numeric-paren 0, table-words 0, var-words 4) =====1011Empirical Economics (2021) 61:3239–326912https://doi.org/10.1007/s00181-020-01992-313Real estate listings and their usefulness for hedonic14regressions15Jens Kolbe1 · Rainer Schulz2 · Martin Wersing2 · Axel Werwatz116Received: 14 January 2020 / Accepted: 15 November 2020 / Published online: 13 January 202117© The Author(s) 202018Abstract19Real estate platforms provide a new source of data which has already been used as20a substitute for transaction data in hedonic regression applications. This paper asks21whether it is valid to do so in the established research areas of (1) willingness to pay22estimation, (2) automated valuations, and (3) price index construction. It therefore23compares listings and transaction data and regression results derived from them. We24findthataskpricesstochasticallydominatesaleprices,mainlybecausethecomposition25of characteristics differs between the two data sets. But estimates of implicit prices26also differ. As a result, willingness to pay estimates from listings data can be widely27off when compared with estimates from transaction data. Listings data are not very28useful to predict market values of individual houses either, as these predictions suffer29from upward bias and large error variance. We find, however, that an ask price index30complements a sale price index, as it is useful for nowcasting.31Keywords Hedonic modelling · Nowcasting · Price prediction · Stochastic32dominance33Supplementary Information The online version contains supplementary material available at https://doi.34org/10.1007/s00181-020-01992-3.35B Axel Werwatz36axel.werwatz@tu-berlin.de37Jens Kolbe38j.kolbe@tu-berlin.de39Rainer Schulz40r.schulz@abdn.ac.uk41Martin Wersing42martin.wersing@abdn.ac.uk43144Chair for Econometrics and Business Statistics, Institute for Economics and Business Law,45Technische Universität Berlin, Straße des 17. Juni 135, 10623 Berlin, Germany46247University of Aberdeen Business School, Edward Wright Building, Dunbar Street, Aberdeen AB24483QY, UK49123505152===== PAGE 10 (score 15, numeric-paren 0, table-words 3, var-words 7) =====5354324855J. Kolbe et al.56Table 3 Stochastic dominance tests. The statistic d j,k ( j ̸= k) tests the null hypothesis that distribution j57dominates distribution k weakly. In Panel A, the variable tested for is −AGE, d j,k is the signed two sample58KS test statistic, defined in Eq. (1). The p values for the null are calculated as exp{−2(d j,k)2}, see Barrett59and Donald (2003, p. 78). In Panel B, d j,k is the KS maximal t-statistic as defined in Chernozhukov et al.60(2013, p. 2222). The p-values for the null are calculated as R−1 61r 1(d∗62j,k,r > d j,k), where d∗63j,k,r is the64r′th bootstrap test statistic, see Barrett and Donald (2003, p. 8). The number of bootstrap replications is6520066da,s67p value68ds,a69p-value70Panel A. Marginal distributions71Price720.000731.0007420.577750.00076Age770.286780.8497924.689800.00081Floor area820.000831.0008434.102850.00086Plot area870.027880.999899.007900.00091Panel B. Price decomposition92Price930.000940.9159548.777960.00097Characteristics980.000990.86010053.1321010.000102Implicit prices1031.0761040.70510510.9781060.000107Ask data1080109.02110.04111.06112.08113Density114011520116401176011880119100120Age121Sale data1220123.02124.04125.06126.08127Density128012920130401316013280133100134Age1350136.005137.01138Density1390140200141400142600143Floor area144Ask data145Sale data1460147.0005148.001149.0015150.002151Density152015350015410001551500156Plot area157Ask data158Sale data159Fig. 2 Distributions of house characteristics. Top panel shows histograms of building age for the observa-160tions in the ask and the sale data, respectively. Lower panel shows kernel density estimates of floor area161(left) and plot area (right) for the observations in the data. Bandwidths are chosen with Silverman’s (1986)162rule-of-thumb163123164165166===== PAGE 11 (score 20, numeric-paren 0, table-words 6, var-words 8) =====167168Real estate listings and their usefulness...16932491705800000171581000017258200001735830000174370000175380000176390000177400000178410000179Longitude180Latitude18101825183101841518520186Relative frequencies187 of listings [%]188Ask data1895800000190581000019158200001925830000193370000194380000195390000196400000197410000198Longitude199Latitude20002015202102031520420205Relative frequencies206 of transactions [%]207Sale data208Fig. 3 Spatial distribution of ask and sale observations. Shows the relative frequency of observations in the209ask and sales data across Berlin’s 12 administrative districts. Crosses give locations of the 59,502 (12,218)210individual observations in the ask (sale) data for which we have coordinates211Table 2 shows that the proportion of detached houses is higher in the ask than in the212sale data and the proportion of terraced houses is lower.12 This could explain why the213floor and plot area distributions of the ask data dominate those of the sale data. Figure 3214shows that the observations of the two data sets follow a similar spatial cluster and215are, at the level of Berlin’s administrative districts, nearly identical with a correlation216of ρ = 0.97. The observed characteristics can therefore not explain why the houses217in the ask data are dominantly younger.218In summary, the analysis of the core characteristics age, floor and plot area shows219strong dominance of observations in the ask over their counterparts in the sale data.220We explore next whether this is the sole reason for the strong dominance of ask over221sale prices.2222.4 Decomposition analysis223We use the inference procedure of Chernozhukov et al. (2013) to assess the importance224of house characteristics for the differential of ask and sale price distributions. The225procedure is similar in spirit to the decomposition at the means of Blinder (1973) and226Oaxaca (1973). The starting point is the following formula which—akin the law of227iterated expectations—relates unconditional and conditional distribution functions228Fj|k(p) ≡229230Xk231FPj|X j (p|x)dFXk(x)232(2)233FPi|Xi (p|x) is the price distribution conditional on the vector x of characteristics and234FXi (x) is the distribution of these characteristics. If k = j, both the conditional price23512 t-tests (not reported) show that the proportions of the three house types are different between the data at236the usual significance levels.237123238239240===== PAGE 21 (score 28, numeric-paren 0, table-words 2, var-words 17) =====241242Real estate listings and their usefulness...2433259244Table 5 Willingness to pay for house characteristics. Reports WTP estimates and regression diagnostics for penalised least squares estimates of Eq. (6). WTPs for continuous245(discrete) house characteristics are computed with Eq. 9 (Eq. 10). Standard errors are computed using the pairs bootstrap with 200 replications. ¯R2 is the adjusted coefficient246of determination. DCV is the double cross-validation score. Significant at ∗∗∗0.001 level, ∗∗0.01 level, ∗0.05 level247Ask data248Sale data249(1)250(2)251(3)252(4)253(5)254WTP255SE256WTP257SE258WTP259SE260WTP261SE262WTP263SE264Age265243.99∗∗∗26647.25267−78.9626843.88269−1173.73∗∗∗270103.73271−1461.39∗∗∗272106.05273−1562.89 ∗∗∗274110.14275Floor area2761419.07∗∗∗27710.392781293.92∗∗∗2797.412801047.99∗∗∗28117.46282906.73∗∗∗28314.82284830.24∗∗∗28516.87286Plot area2871465.18∗∗∗2887.952891380.25∗∗∗2907.462911152.82∗∗∗29218.142931011.70∗∗∗29414.60295940.84∗∗∗29616.79297Detached2986881.76∗∗∗2991424.7430017771.60∗∗∗3011263.12302−7918.00∗∗3032468.323044130.493052198.433066546.97∗∗3072999.16308Semi-detached309−1042.733101205.123115074.97∗∗∗3121011.21313−8160.82∗∗3142630.253153741.823162315.453173705.943182158.45319Listed32012599.23∗∗3213702.38322Prefabricated323−6544.52∗∗∗3241686.19325Converted attic3263426.77∗∗3271141.96328Swimming pool32914744.26∗∗3305509.92331Flat roof332−5158.22∗∗∗3331426.43334No basement335−21888.12∗∗∗3361724.46337Backland develop.338396.673391363.94340Waterfront34159947.35 ∗∗∗3426417.56343Poor condition344−53313.91∗∗∗3452341.98346Good condition34729031.33∗∗∗3481657.96349Poor amenities350−6263.71∗∗∗3511720.23352Good amenities35316298.74∗∗∗3542243.77355Excel. amenities35637654.98∗∗∗3579613.84358f4(L AT , LON)359No360Yes361No362Yes363Yes364123365366367===== PAGE 22 (score 5, numeric-paren 0, table-words 2, var-words 0) =====3683693260370J. Kolbe et al.371Table 5 continued372Ask data373Sale data374(1)375(2)376(3)377(4)378(5)379WTP380SE381WTP382SE383WTP384SE385WTP386SE387WTP388SE389District dummies390Yes391No392Yes393No394No395Year dummies396Yes397Yes398Yes399Yes400Yes401Buyer/Seller dummies402No403No404No405No406Yes407DCV4080.0484090.0394100.0684110.0564120.050413¯R24140.7754150.8164160.6654170.7314180.761419N42068,07042159,50242212,52442312,21842412,218425123426