1# RECORD bc3577b4d2c3cc062519a4199d4f0a8834e89c892# TITLE: Points of Interest and Housing Prices3# YEAR: 2023.0 VENUE: Real Estate Management and Valuation DOI: 10.2478/remav-2023-00074# AUTHORS: R. Cellmer5# PAGES: 9 CANDIDATE TABLE PAGES (1-based): [7]6# ABSTRACT: ABSTRACT Points of Interest (POI) are an inherent element of the urban landscape, and their number and density reflect, among other things, the degree of urbanization and the city’s spatial structure. The very presence of POI in the closest vicinity of a residential property may indirectly or directly affect housing value. This paper aims to demonstrate the usefulness of POI density information of different categories in assessing the quality of a property’s immediate surroundings. While the mere presence of POIs in the nearest neighborhood may affect real estate prices, the influence of specific categories may not necessarily be positive. Therefore, the study attempted to classify POIs and determine their importance in the price formation process using spatial regression models. The results indicate that a high density of POIs in the immediate area is a stimulant for housing prices. The detailed analysis indicated that only some POI categories might be related to transaction prices, while in certain situations, some POI categories may negatively impact prices.789===== PAGE 1 (score 11, numeric-paren 0, table-words 0, var-words 11) =====1011 12Journal homepage: 13www.remv-journal.com 14 15 16REAL ESTATE MANAGEMENT AND VALUATION - vol. 31, no. 1, 2023 17eISSN: 2300-5289 | Received 2022-03-31 | Revised 2022-05-26 | Accepted 2022-07-12 186919POINTS OF INTEREST AND HOUSING PRICES 20 21Radosław Cellmer1* 221Department of Spatial Analysis and Real Estate Market, Institute of Spatial Management and Geography, University 23of 24Warmia 25and 26Mazury 27in 28Olsztyn, 29ul. 30Oczapowskiego 312,10-719 32Olsztyn, 33Poland, 34email: 35radoslaw.cellmer@uwm.edu.pl,ORCID: 0000-0002-1096-8352. 36* Corresponding author 37 38ARTICLE INFO 39ABSTRACT 40Keywords: 41Points of Interest (POI) are an inherent element of the urban landscape, and their number and 42density reflect, among other things, the degree of urbanization and the city's spatial structure. 43The very presence of POI in the closest vicinity of a residential property may indirectly or 44directly affect housing value. This paper aims to demonstrate the usefulness of POI density 45information of different categories in assessing the quality of a property's immediate 46surroundings. While the mere presence of POIs in the nearest neighborhood may affect real 47estate prices, the influence of specific categories may not necessarily be positive. Therefore, 48the study attempted to classify POIs and determine their importance in the price formation 49process using spatial regression models. The results indicate that a high density of POIs in the 50immediate area is a stimulant for housing prices. The detailed analysis indicated that only 51some POI categories might be related to transaction prices, while in certain situations, some 52POI categories may negatively impact prices. 53points of interest, housing 54market, spatial modelling 55JEL Classification: 56C51, R30 57 58Citation: 59Cellmer, R. (2023). Points of interest and housing prices. Real Estate Management and 60Valuation, 31(1), 69-77. https://doi.org/10.2478/remav-2023-0007 61 621. Introduction 63The characteristics of urban space determine the 64quality of location values of residential properties, 65which translates into the value and, consequently, 66prices 67of 68dwellings. 69The 70spatial 71heterogeneity 72of housing prices in urbanized areas is thus the result 73of the diversified characteristics of their surroundings 74(Kiel & Zabel, 2008). These features are related to both 75tangible amenities affecting the quality of everyday 76life (e.g. access to transportation, stores, services) and 77intangible factors affecting, among other things, the 78sense of safety and comfort of habitation (e.g. the 79appearance of the neighborhood, noise level). On the 80one hand, the development of urban infrastructure 81may contribute to the attractiveness of places to live 82in but, on the other hand, it may harm the quality of 83life in some cases. The problem, in this case, is the 84evaluation of the quality of a particular location and its 85closest surroundings. In this assessment, points of 86interest (POI), which are an integral part of the urban 87landscape, with their number and density indicating, 88among other things, the degree of urbanization and 89the spatial structure of the city, are of particular 90importance (Lu et al., 2020; Steiniger et al., 2016; Yue 91et al., 2017). 92This paper aims to demonstrate the usefulness of 93POI density information of different categories 94in assessing the quality of a property's closest 95environment. While the very presence of POI in the 96nearest neighborhood may influence real estate 97prices, the influence of specific categories may not 98necessarily be positive. Therefore, during the research, 99an attempt was made to classify these points and 100determine their importance in price formation using 101spatial regression models. The use of publicly available 102spatial information about the location of significant 103places can improve modeling quality and should be 104an integral part of real estate market analysis, 105especially under conditions of the rapid development 106of spatial data infrastructure. 1072. Literature review 108The drivers of housing prices are highly complicated 109and can be considered on different scales. At the 110micro-level, three main categories of attributes are 111considered important determinants of property price: 112structural, neighborhood, and location characteristics 113(Malpezzi et al., 2002; Zhang et al., 2019). In particular, 114115116===== PAGE 7 (score 36, numeric-paren 0, table-words 7, var-words 6) =====117118 119Journal homepage: 120www.remv-journal.com 121 122 123REAL ESTATE MANAGEMENT AND VALUATION - vol. 31, no. 1, 2023 124eISSN: 2300-5289 | Received 2022-03-31 | Revised 2022-05-26 | Accepted 2022-07-12 12575126Table 4 127Results of spatial regression (error SAR) and multiple regression (OLS), including POIs 128 129Error SAR 130 131 132OLS 133variable 134estimate 135std. error 136p-value 137 138 139estimate 140std. error 141p-value 142VIF 143intercept 1445000.621 145143.399 146< 0.001 147 148 1495016.850 150119.391 151< 0.001 152- 153 154date 15582.713 1567.158 157< 0.001 158 159 160102.824 1617.028 162< 0.001 1631.023 164area 165-21.749 1661.162 167< 0.001 168 169 170-21.072 1711.101 172< 0.001 1731.075 174market 175715.935 17673.616 177< 0.001 178 179 180450.830 18155.709 182< 0.001 1831.532 184floor 18546.240 18623.775 1870.052 188 189 19070.138 19125.655 1920.006 1931.005 194high 195116.940 19667.993 1970.085 198 199 200272.109 20149.650 202< 0.001 2031.179 204year 205214.661 20615.342 207< 0.001 208 209 210207.272 21110.999 212< 0.001 2131.577 214POI 2150.823 2160.178 217< 0.001 218 219 2200.492 2210.109 222< 0.001 2231.063 224 225Lambda: 0.453, AIC: 36254 226 227 228 229AIC: 36457 230Source: own study. 231Table 5 232Results of spatial regression (error SAR) and multiple regression (OLS) concerning individual POI categories 233 234Error SAR 235 236OLS 237variable 238estimate 239std. error 240p-value 241 242estimate 243std. error 244p-value 245VIF 246intercept 2474861.668 248144.832 249< 0.001 250 2514806,7787 252121,843 253< 0.001 254- 255 256date 25785.124 2587.130 259< 0.001 260 261104,491 2627,004 263< 0.001 2641.054 265area 266-21.999 2671.163 268< 0.001 269 270-21,548 2711,126 272< 0.001 2731.167 274market 275794.159 27674.517 277< 0.001 278 279597,354 28058,487 281< 0.001 2821.753 283floor 28447.938 28523.671 2860.043 287 28872,536 28925,238 2900.004 2911.009 292high 293103.195 29467.361 2950.125 296 297252,588 29851,323 299< 0.001 3001.308 301year 302227.864 30315.581 304< 0.001 305 306229,437 30711,984 308< 0.001 3091.943 310motorisation 31118.132 3126.297 3130.003 314 31514,906 3164,288 317< 0.001 3181.102 319culture 3206.568 3214.176 3220.116 323 3242,854 3253,334 3260.392 3271.873 328education 3291.299 3303.202 3310.685 332 333-2,394 3342,423 3350.323 3361.120 337finances 3381.337 3393.395 3400.694 341 3421,991 3432,509 3440.428 3451.997 346gastronomy 3470.585 3481.426 3490.681 350 3512,140 3521,216 3530.079 3541.897 355health 3560.703 3573.022 3580.816 359 3600,388 3612,211 3620.861 3631.924 364offices 36561.787 36618.474 3670.001 368 36969,045 37015,823 371< 0.001 3721.212 373recreation 3741.125 3750.197 376< 0.001 377 3780,642 3790,124 380< 0.001 3811.177 382religion 383-0.287 38414.493 3850.984 386 387-18,441 38810,246 3890.072 3901.101 391security 392-0.644 3931.552 3940.678 395 3960,012 3971,242 3980.993 3991.555 400services 401-4.025 4021.552 4030.009 404 405-4,798 4061,169 407< 0.001 4082.951 409telecom 41010.426 4113.479 4120.003 413 41415,023 4152,523 416< 0.001 4171.519 418tourism 41918.128 42011.529 4210.118 422 42310,477 4248,697 4250.228 4261.222 427other 428-4.726 4292.440 4300.053 431 432-5,022 4331,782 4340.005 4351.310 436 437Lambda: 0.416, AIC: 36233 438 439 440AIC: 36388 441Source: own study. 442 443The Akaike criterion (AIC) indicates that the SAR 444model slightly fits the data better than the OLS model. 445Variance inflation factor showed no multicollinearity. 446The lambda coefficient in the SAR model, indicating 447the autocorrelation of the residuals, was as high as 4480.453. Most of the parameters next to the control 449variables were statistically significant in both the SAR 450and OLS models. The signs (plus and minus) and order 451of magnitude of the parameters are also not 452questionable. However, special attention should be 453paid to the parameter next to the POI variable 454reflecting the density of points in places where 455transactions took place. In the SAR model, this 456parameter was 0.823, which means that each 457additional POI per 1 km2 in a given neighborhood is 458associated with a higher price of an apartment by 0.82 459PLN/m2. It does not seem like much, but considering 460that, in the central part of the city, the calculated 461density is even more than 1700 points/km2 (the 462smaller the bandwidth, the greater the variation in 463density), this impact turns out to be significant. 464Considering the market conditions in Olsztyn (the 465average price of 1m2 of an apartment is about 6000 466PLN), the maximum impact of the proximity of POIs 467may be close to 25% of the average price. The results 468obtained confirm the studies conducted so far (Tang 469et al., 2018). While there is no doubt that in places 470where the density of POIs is higher, prices are also 471higher, the assumption of causality should be treated 472with some distance. The estimated parameters of the 473model testify rather to a statistical dependence and 474not necessarily to a cause-effect relationship. 475The study of the impact of the proximity of POIs, 476considered without dividing them into categories, may 477478479===== PAGE 8 (score 20, numeric-paren 0, table-words 5, var-words 10) =====480481 482 483REAL ESTATE MANAGEMENT AND VALUATION - vol. 31, no. 1, 2023 484eISSN: 2300-5289 | © 2023 The Author(s) | Article under the CC BY license 48576 486lead to an incorrect conclusion that the proximity of 487each such point is a price stimulant. It should be taken 488into account that POIs may represent, for example, 489both unfriendly sewage treatment plants and friendly 490local amenities, e.g. in the form of parcel machines or 491neighborhood stores. Hence, Table 5 presents the 492estimation results of the SAR model and the OLS 493model in which each POI category is included 494(according to Table 2) 495The Akaike criterion (AIC), also in this case, 496indicates that the SAR model is a slightly better fit to 497the data than the OLS model. As in the previous 498model, the variance inflation factor did not show 499multicollinearity. The lambda coefficient in the SAR 500model, denoting the autocorrelation of the residuals, 501was 5020.416, 503which 504indicates 505a 506relatively 507high 508autocorrelation. Again, most of the parameters next to 509the control variables were statistically significant. 510Among the parameters next to the variables denoting 511the individual POI categories, the parameters next to 512motorization, offices, recreation, services, telecom, and 513others (SAR model) were statistically significant at a 514significance level of less than 0.05. Unfortunately, the 515absolute values of the coefficients do not allow for a 516direct comparison of the importance of individual POI 517categories due to the different levels of the average 518density of points (see Table 2). This would be possible 519if there were the same number of POI in each 520category. 521Overall, the research shows that the impact of POI 522density on prices can be: 523– positive 524(motorization, 525offices, 526recreation, 527telecom), 528– neutral 529(culture, 530education, 531finances, 532gastronomy, health, religion, tourism), 533– negative (services, other). 534In the case of the other category, the negative 535impact is perfectly understandable (see Table 2). In the 536case of retail and services, however, the results are 537somewhat confusing. It is unlikely that the proximity 538of retail and services amenities harms location quality. 539This result is particularly intriguing, and an explanation 540of this problem may contribute to further in-depth 541research. In any case, several causes of this 542phenomenon can be considered as hypotheses, e.g.: 543– apparent correlations - the density of POIs 544indicating retail and service may be strongly 545correlated 546with 547some 548features 549negatively 550influencing prices that were not included in the 551research 552(e.g. 553heavy 554traffic, 555unsafe 556neighborhood, noise, pollution), 557– synergy - perhaps retail and service should be 558considered together with other characteristics 559(e.g. recreation) which it can form synergistic 560systems with, 561– spatial structure of the development - most 562POIs related to retail and service are located in 563locations with predominantly prefabricated 564housing, which may be a factor that negatively 565affects prices, 566– new housing estates - as a rule, better locations 567are in new housing estates where there is still 568little retail and services. 5696. Conclusions 570POI information is widely available and can be very 571useful during real estate price 572analyses. This 573information can be used to assess the spatial structure 574of cities and model housing prices. The study showed 575a significant statistical relationship between the 576density of POI and the level of housing transaction 577prices. However, it should be taken into account that 578this relationship is not always causal. While a large 579number of POI in the vicinity is undoubtedly 580correlated with prices, it should be taken into account 581that there are many categories of POI whose 582relationship with prices is not so obvious. The study 583showed that motorization, offices, recreation, and 584telecom positively influence prices. The negative 585impact on prices, which seems quite obvious, is the 586presence of points of interest related to recycling and 587water treatment plants. Indeed, the catalogue of 588investigated POIs can be extended to include, among 589others, points related to urban transport (e.g. subway 590stations), which can also determine further research 591directions. A key challenge may be to develop a new 592POI system that takes into account the specificities of 593the real estate market. The constructed models also 594indicate that the category retail and services should be 595treated as a destimulant, which means that the 596hypothesis of causality should be treated with 597considerable distance. The obtained results allowed 598the formulation of additional hypotheses determining 599further research directions. 600The study justified using spatial models to 601investigate the relationship and dependence between 602POI density and housing prices. Housing prices show 603not only spatial heterogeneity but also spatial 604autocorrelation. Models using this fact are better 605suited to the realities of the market than classical 606multiple regression models. 607Although the conducted research was aimed at 608investigating 609the 610relationship 611between 612POIs 613