1# RECORD 4e94c698b9b6217c36e2e8842148142f5bccfb362# TITLE: A RE-ASSESSMENT OF HOUSE PRICE INDICES: EVIDENCE FROM THE SPANISH MARKET3# YEAR: 2018.0 VENUE: International Journal of Strategic Property Management DOI: 10.3846/ijspm.2019.63664# AUTHORS: Paloma Taltavull de La Paz; Stanley McGreal5# PAGES: 13 CANDIDATE TABLE PAGES (1-based): [9]6# ABSTRACT: This paper using evidence from the Spanish housing market contributes significantly to the debate concerning the different results obtained from house price indices depending on the method used to build the index. Utilising a large database over the period 1994 to 2012, the paper constructs a time dummy hedonic index (HD) and an imputed hedonic index using a Laspeyres approach (HI), and compares the different effect on the price index evolution. The paper discusses control by quality changes and identifies those attributes experiencing structural changes over the analysis period, identified by the HI index but not by the HD index. Results indicate that changes in quality stem from socio-demographic conditions rather than changes to housing quality (other than size). The paper also shows that improvements in neighbourhood quality rather than change in a ‘typical house’ affects house price and argues that these considerations are important in both the method selected to calculate house price indices and the application of the methodology to estimate price changes.789===== PAGE 1 (score 7, numeric-paren 0, table-words 0, var-words 7) =====1011Copyright © 2019 The Author(s). Published by VGTU Press12This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unre-13stricted use, distribution, and reproduction in any medium, provided the original author and source are credited.14*Corresponding author. E-mail: paloma@ua.es15International Journal of Strategic Property Management16ISSN: 1648-715X / eISSN: 1648-9179172019 Volume 23 Issue 1: 23–3518https://doi.org/10.3846/ijspm.2019.6366 19A RE-ASSESSMENT OF HOUSE PRICE INDICES: 20EVIDENCE FROM THE SPANISH MARKET21Paloma TALTAVULL DE LA PAZ1,*, Stanley McGREAL2,3221 Department of Applied Economy Analysis, International Economy Institute, University of Alicante, 23Campus de San Vicente del Raspeig, 03080 Alicante, Spain242 Built Environment Research Institute, School of the Built Environment, University of Ulster, UK253 School of Commerce, Division of Business, University of South Australia, Australia26Received 11 September 2015; accepted 10 January 201827Abstract. This paper using evidence from the Spanish housing market contributes significantly to the debate concerning 28the different results obtained from house price indices depending on the method used to build the index. Utilising a large 29database over the period 1994 to 2012, the paper constructs a time dummy hedonic index (HD) and an imputed hedonic 30index using a Laspeyres approach (HI), and compares the different effect on the price index evolution. The paper discusses 31control by quality changes and identifies those attributes experiencing structural changes over the analysis period, identi-32fied by the HI index but not by the HD index. Results indicate that changes in quality stem from socio-demographic condi-33tions rather than changes to housing quality (other than size). The paper also shows that improvements in neighbourhood 34quality rather than change in a ‘typical house’ affects house price and argues that these considerations are important in both 35the method selected to calculate house price indices and the application of the methodology to estimate price changes.36Keywords: price index, housing market, hedonic models, Spain.37Supplementary material associated with this article can be found, in the online version, at https://doi.org/10.3846/ijspm.2019.636638Introduction39In the post Global Financial Crisis era there has been in-40creasing attention given to the construction of house price 41indices globally and particularly in European countries 42with a growing number of public institutions and govern-43ments publishing house price indices in an open-access 44format. This stems from the perception that house prices 45have macro-economic effects and that continuous moni-46toring in support of the identification of economic imbal-47ances (European Economy [EU], 2012) facilitates trans-48parency in the market. Given this context, a number of 49studies have focussed on the macro effects of house price 50deviation from their long term trend, the role of monetary 51channels of transmission influences, and the effect of the 52‘housing channel’ on consumption prices (Mishkin, 2007; 53Bank for International Settlements [BIS], 2008; Taltavull & 54White, 2016; R. Grover & C. Grover, 2014), demonstrating 55the relevance of housing price indices. Furthermore, at the 56European level, the EU is advocating the building of price 57indices for member countries using the same methodolo-58gy (Eurostat, 2013) and the requirement for central banks 59to capture market data (also for non-residential real es-60tate markets1) in order to observe patterns in house prices. 61Furthermore, financial and investment institutions need 62reliable indices in updating their balance sheets in accord-63ance with the mandatory regulations (Nikolov, 2014).64Most studies are based on the current price index 65published in each country but discussion has intensified 66concerning to what extent indices correctly reflect the true 67housing values. Consequently, the debate on house price 68index methodology has had a renewed interest in different 69forums. However, complexity arises from the heterogene-70ous nature of housing in terms of location, structure, com-71position and behaviour with price considered to represent 72how a household evaluates attributes according to income, 73education, quality and other characteristics. Such multi-74plicity of attributes makes it difficult to determine the 75best approach to the construction of house price indices. 76The consensus is that to capture the contribution of such 771 See the Greece Central Bank or the British Treasury as exam-78ples of different indices.798081===== PAGE 9 (score 69, numeric-paren 35, table-words 9, var-words 6) =====8283International Journal of Strategic Property Management, 2019, 23(1): 23–35843185Table 1. Urban area house prices growth rate in different periods (accumulate %)86Base 199887Urban level (Imputed hedonic − HI)88‘Imputed-89hedonic (HI)᾽90(a)91‘Dummy 92hedonic (HD)᾽93(b)94Differences 95(%)96[(a/b) – 1] · 1009719829931004101Alicante1021998−201210389.5104117.710584.8106114.3107108.6108146.1109−25.71101998−200411193.311295.411388.311475.311588.9116167.8117−47.01182004−200711938.512016.512121.51221.212316.212429.9125−45.81262007−2012127−20.3128−6.0129−16.113020.8131−5.3132−28.6133−81.4134Madrid1351998−201213654.0137157.513896.9139179.7140−46.11411998−2004142−2.114330.614416.114539.414631.1147152.2148−79.61492004−2007150−39.215155.815224.315366.815456.915525.7156121.41572007−2012158−8.3159−8.516050.716119.0162−12.2163−255.7164Murcia1651998−201216679.716769.516834.416983.117070.9171130.0172−45.41731998−200417480.417567.717618.017758.517867.2179134.6180−50.11812004−200718226.218315.518421.718526.318619.718732.0188−38.61892007−2012190−18.9191−11.91928.2193−11.9194−12.1195−23.8196−49.4197Valencia1981998−2012199118.1200111.020122.5202109.0203100.4204103.5205−3.12061998−2004207101.4208104.620937.021089.421193.4212130.3213−28.32142004−200721515.221621.321722.121818.121919.222047.1221−59.32222007−2012223−11.8224−12.2225−26.3226−9.4227−11.5228−38.6229−70.3230Table 2. Hedonic HI model for Alicante province231Alicante232var dep: Log of asking prices233b234Standardized b235Stand. error236constant2376.47238−2390.017240***241City (7)242PD2430.0442440.0412450.001246***247Pdev248−0.004249−0.0062500.001251***252PGR2530.0392540.0262550.001256***257P (logs)2580.0352590.1152600.000261***262Eco2630.0352640.0572650.000266***267UENV2680.1582690.0252700.004271***272Neighborhood (12)273Inc2740.0542750.0572760.001277***278Resarea2790.0922800.0722810.001282***283Con2840.0012850.0172860.000287***288URR289−0.001290−0.0302910.000292***293Qroad2940.0202950.0052960.003297***298Qw-pipe299−0.634300−0.5623010.001302***303Qlig304−0.009305−0.0033060.002307***308Qsh3090.0063100.0083110.001312***313Qschool3140.0123150.0113160.002317***318Qchurch3190.0493200.0443210.002322***323Qleis3240.0803250.0883260.001327***328Qhealth329−0.113330−0.1153310.001332***333Accessibility (3)334Bus3350.0073360.0113370.000338***339Train340−0.002341−0.0023420.001343***344Und-Tram3450.0293460.0693470.000348***349House (12)350TB3510.1013520.0983530.001354***355dwel356−0.001357−0.0273580.000359***360lift3610.0583620.0753630.001364***365age366−0.008367−0.1943680.000369***370Qshop_n3710.1183720.2113730.000374***375Inc_t3760.0963770.0983780.001379***380PD_t381−0.020382−0.0183830.001384***385Ori3860.0033870.0133880.000389***390Vi3910.0193920.0213930.001394***395ConQ3960.0333970.0553980.001399***400m2 (log)4010.8114020.4004030.001404***405m2_O (log)4060.0344070.0554080.000409***410Adj R24110.768412Se24134357.188414F41558081.547***416N417596170418Tolerance and VIF test indicates the absence of colineality419+ Categorical variable420*** p-value < 0.001, ** p-value < 0.01, * p-value < 0.05421Note: This Table 2 contains the average value of four urban level betas estimated for Alicante in 2010 (4 quarters). Total number of usable observations 422for Alicante (for the whole period) are 478418, and 11390 on average for every quarter of 2010.423424425===== PAGE 10 (score 11, numeric-paren 0, table-words 0, var-words 11) =====42642732428P. Taltavull de la Paz, S. McGreal. A re-assessment of house price indices: evidence from the Spanish market429housing characteristics have smaller coefficients and in 430essence a lower ranking concerning the effect on prices 431(quality of construction, burb_env = 0.158, orientation, bo-432rient = 0.003, view, bview = 0.019, additional surface area 433bm2_other = 0.034, number of lifts blift = 0.058, and qual-434ity of construction bqconstr = 0.033). This suggests that the 435main determinant for house prices are neighbourhood 436attributes. Hence, indices failing to control by location/437characteristics of the surrounding area could be biased. 438The effects of missing neighbourhood quality change thus 439limit the ability of the index to control by quality.4405. Discussion441As the HD estimations do not allow the attribute param-442eters to vary with time, the results suggest that the model 443structure reflects the weights given to each attribute dur-444ing the first part of the time series period but differ from 445the pattern in the second period. The test for breaking 446points in attributes25 supports this argument and suggests 447that a change in quality accounts for part of the observed 448house price growth and that remaining differences reflect 449other price impulses. HI seems better placed at capturing 450the changes in quality while the HD model reflects ex-451ogenous influences26 such as the market influx on house 452prices from the demand side. One of these components 453could also be due to autocorrelation (spatial and time) in 454the data27.455Models from the HI Hedonic Imputed suggest that 456there is a pattern of initial growth characterised by rising 457house prices until 2005 followed by stable house prices 458suggesting that the share of price explained by property 459quality was increasing until 2005 but then stabilized. These 460results contrast with those from the HD model, notably 461over the period 2005–2010, illustrating that differences in 462house price growth and the extent of the level of growth is 463dependent on the methodology used. One interpretation28 464is that the hedonic dummy method does not take into ac-465count variability in the price of attributes (assumed to be 466constant over the period by the method) hence the index 467loses its capacity to capture quality changes over time. By 468contrast HI allows the estimated betas to vary. It seems that 469the type of dwellings sold over the boom and bust periods 47025 Performed in the typical house’s section 5.1.47126 When the Ph_hat is estimate in a ‘HD model, the values 472are equal to the observed asking prices. This is why it is not 473necessary to calculate an index using the traditional method 474(Laspeyres or Paasche), in a model controlled by attributes 475plus time dummies. The time dummies capture any external 476influence on prices.47727 Note that papers explaining imputed price methods based on 478traditional indices (Laspeyres, Paasche or Fisher) using the es-479timated prices obtained directly from a hedonic model do not 480allow intercept and attribute parameters to change over time 481and then capture the time effect.48228 We want to thanks to an anonymous referee for provide this 483highlight.484have varied to the point that either the relative prices of 485attributes changed or that comparing the reference house 486at the end of the period to the same house at the beginning 487does not make sense. Such reaction could be capturing 488changes in the type of house sold over time and may also 489reflect changes in market liquidity. For example, during the 490boom practically any type of unit was easily sold, even if of 491low quality, hence the rising quality adjusted prices while 492during the bust, the number of transactions declined and 493only certain types of dwellings were sold. One would have 494expected that only the very good houses got sold, so that 495the quality increase was such that the price would decline 496even more when adjusting for quality.497The house price index evolution compared in Figure 2 498suggests that HD and HI are measuring different concepts. 499The HI construction estimated price of the total attributes, 500letting the errors capture the increase in price due to other 501factors (as defined by Rosen). In essence, the HI models 502essentially provide an index of quality rather than an index 503controlled by quality, whereas the HD models isolate house 504prices from their components (time and attributes) obtain-505ing an index adjusted by quality. The different methods of 506construction mean that the interpretation of HI indices is 507not the same as that for HD indices; the former are indi-508ces of quality prices and the latter indices of house price 509controlled by quality. Thus both are complementary rather 510than substitutive, and both partially capture true quality 511controlled house price. As a consequence, in the case of 512HI the price growth rate could be calculated based on 513the equation residuals rather than on the imputed prices 514whereas in the case of a HD index, the equation to estimate 515dummy parameters includes a sufficient number of char-516acteristics to control by quality though the effect on prices 517of the unobserved or omitted key attributes may produce 518an index not fully controlled by quality.5195.1. The ‘typical house’520A further point of discussion relates to estimation of the 521‘typical house’ attributes structure to guarantee that the 522index is controlling by changes in quality and the assump-523tion that it can change over time, explaining the results 524found in the two estimated indices. In this respect, some 525evidence about changes in quality of the whole stock is 526needed to identify whether this type of change occurs 527and when. In the case of no statistically significant chang-528es in housing attributes (which determines the ‘type of 529house’ in the long term), a house price index measuring 530the same initial attribute structure should be precise in 531capturing price changes. However, a change in attributes 532could introduce a bias in the estimated house price index 533capturing ‘inexistent shifts’ in prices and in fact actually 534measuring changes in the type of house on the market. To 535illustrate this point further, the current database is used 536to identify shifts in quality (attributes) over the time. Ta-537ble S3 (supplementary material) shows the significance of 538Zivot-Andrews test of breaking points and its application 539to the density distribution for categorical variables and ex-540