1# RECORD 52cbefc12720084fe56091430bf37e9fbeba96b92# TITLE: Do 5G cell phone towers decrease house prices? Evidence from Warsaw3# YEAR: 2024.0 VENUE: Journal of Housing and the Built Environment DOI: 10.1007/s10901-024-10144-34# AUTHORS: Bartłomiej Marona; R. Gaca; Michał Głuszak5# PAGES: 22 CANDIDATE TABLE PAGES (1-based): [12, 14, 16]6# ABSTRACT: This study investigates the impact of telecommunication infrastructure on residential property prices in Poland. This study contributes to the discussion of economic externalities related to new public infrastructure within the urban landscape. We use hedonic regression, matching techniques, and a difference-in-differences estimator to assess the impact of base transceiver stations on apartment prices in Warsaw, the capital of Poland, which covered market data on the sales of residential premises located in Miasteczko Wilanów estate in Warsaw from 2016 to 2021. In the analysed period, 1,825 residential sales were recorded and used for econometric modelling. The results do not confirm the influence of the vicinity of the wireless communication technology infrastructure on residential real estate prices in the studied local market. In particular, we did not observe a detrimental effect of cell phone towers on housing prices. Additionally, we investigate whether residential sale prices in proximity to BTS changed significantly after the introduction of the 5G standard. This particular issue has not been addressed in the economic literature. We found that the sale prices of apartments located in Warsaw were not statistically affected by the introduction of a fifth-generation technology standard for broadband cellular networks. Our research contributes to a better understanding of stigmatisation effects related to telecommunication infrastructure, and in particular, the links between the presence of cell phone towers and residential property values in the neighbourhood. The results may be of interest to all potential agents involved in neighbourhood conflicts arising from investments in cell phone towers and the development of new communication infrastructure in urban landscapes. The proximity of the base transceiver stations does not significantly affect the transaction prices. The introduction of 5G technology did not have a significant effect on residential property prices near cell phone towers. These findings contribute to a better understanding of the complex interplay between urban development, community conflicts, and housing prices.789===== PAGE 1 (score 3, numeric-paren 0, table-words 0, var-words 3) =====1011ARTICLE12Received: 24 January 2024 / Accepted: 28 June 2024 / Published online: 1 August 202413© The Author(s) 202414Extended author information available on the last page of the article15Do 5G cell phone towers decrease house prices? Evidence 16from Warsaw17Bartłomiej Marona118 · Radosław Gaca119 · Michał Głuszak120Journal of Housing and the Built Environment (2024) 39:1821–184221https://doi.org/10.1007/s10901-024-10144-322Abstract23This study investigates the impact of telecommunication infrastructure on residential prop24erty prices in Poland. This study contributes to the discussion of economic externalities 25related to new public infrastructure within the urban landscape. We use hedonic regres26sion, matching techniques, and a difference-in-differences estimator to assess the impact 27of base transceiver stations on apartment prices in Warsaw, the capital of Poland, which 28covered market data on the sales of residential premises located in Miasteczko Wilanów 29estate in Warsaw from 2016 to 2021. In the analysed period, 1,825 residential sales were 30recorded and used for econometric modelling. The results do not confirm the influence 31of the vicinity of the wireless communication technology infrastructure on residential real 32estate prices in the studied local market. In particular, we did not observe a detrimental 33effect of cell phone towers on housing prices. Additionally, we investigate whether resi34dential sale prices in proximity to BTS changed significantly after the introduction of the 355G standard. This particular issue has not been addressed in the economic literature. We 36found that the sale prices of apartments located in Warsaw were not statistically affected 37by the introduction of a fifth-generation technology standard for broadband cellular net38works. Our research contributes to a better understanding of stigmatisation effects related 39to telecommunication infrastructure, and in particular, the links between the presence of 40cell phone towers and residential property values in the neighbourhood. The results may 41be of interest to all potential agents involved in neighbourhood conflicts arising from in42vestments in cell phone towers and the development of new communication infrastructure 43in urban landscapes.44Highlights45 46●47The proximity of the base transceiver stations does not significantly affect the transac48tion prices.49 50●51The introduction of 5G technology did not have a significant effect on residential prop52erty prices near cell phone towers.53 54●55These findings contribute to a better understanding of the complex interplay between 56urban development, community conflicts, and housing prices.57Keywords Housing prices · BTS · Cell phone towers · 5G · Warsaw · Externalities581 3596061===== PAGE 12 (score 24, numeric-paren 8, table-words 2, var-words 9) =====6263B. Marona et al.64Table 2 Descriptive statistics (N = 1825)65Abbreviation66Category67Measurement68Mean69SD70Min71Max72Price73DV74Apartment’s transaction price 75in PLN (Natural logarithm)76717,799.377(13.39)78341,466.5 79(0.41)80215,000 81(12.28)822,900,000 83(14.88)84BTS_Prox85IV (N)86Proximity to BTS. 1 if the 87apartment being sold was 88located within a 200 m radius 89from BTS and 0 otherwise900.48910.5092093194BTS_Dist95IV (N)96Linear distance to the nearest 97elementary school (in 100 m)982.13990.871000.291014.98102BTS_Num103IV (N)104Number of BTSs in proximity 105(within a 200 m radius from 106the apartment being sold)1070.691080.9210901105111BTS_5G112IV (N)113Proximity to 5G BTS. 1 if 114the apartment being sold was 115located within a 200 m radius 116from 5G BTS and 0 otherwise1170.281180.4511901201121School122IV (N)123Linear distance to the nearest 124elementary school (in 100 m)1255.941262.501270.5912810.78129Bus130IV (N)131Linear distance to the nearest 132public transport stop (in 100 m)1332.251341.181350.221365.01137Noise138IV (E)139Road noise measured in inter140val brackets (less than 55 dB 141− 1, 55–60 dB 2, 60–65 dB – 3, 14265–70 dB – 4)1433.391440.8614511464147Area148IV (S)149Apartment’s usable area in m2150(Natural logarithm)15171.13 152(4.19)15329.43154(0.38)15526.09156(3.26)157287.94158(5.66)159Rooms160IV (S)161Number of rooms in the 162apartment1632.911641.0616511667167Floori168IV (S)169Location on floor. Dummy 170variables that take a value of 1 171if the apartment was located at 172a given floor and 0 otherwise 173(Floor1 -ground floor, base 174category omitted in regression 175models, Floor2 – other floors; 176Floor3 – first and second floor).177Floor1 (318 obs.), Floor2 (874), Floor3 (633)178Trans_C179IV (c)180The transaction involves a 181corporate participant. 1 when 182either buyer or seller is not a 183private person (household). 0 184otherwise1850.071860.2618701881189Timek190IV (t)191Time dummy variables taking 192a value of 1 if a transaction oc193curred in a given quarter within 194the study period (from 2016Q1 195to 2021Q2)1962016q1 (40 obs.); 2016q2 (58); 2016q3 (74); 1972016q4 (42); 2017q1 (50); 2017q2 (74); 1982017q3 (87); 2017q4; (77); 2018q1 (74); 1992018q2 (88); 2018q3 (95); 2018q4 (110); 2002019q1 (91); 2019q2 (114); 2019q3 (124); 2012019q4 (114); 2020q1 (101); 2020q2 (64); 2022020q3 (104); 2020q4 (111); 2021q1 (107); 2032021q2 (26)204Note DV = dependent variable; IV = independent variable; N = neighbourhood attribute; S = structural 205attribute; E = environmental attribute; c = contract-related control variable; t = time-related control variable2061 32071832208209210===== PAGE 13 (score 16, numeric-paren 0, table-words 3, var-words 10) =====211212Do 5G cell phone towers decrease house prices? Evidence from Warsaw213prices of residential properties. However, we acknowledge the risk that we cannot account 214for all salient characteristics that may affect housing prices. The unobserved heterogene215ity in housing quality is a major problem in hedonic price modelling (Bajari et al., 2012; 216Francke & Van de Minne, 2021; Osland, 2013), but to our defence, we were restricted by 217data availability. We did not have access to information on apartment interior design, techni218cal condition, aspect, or view from the window. The housing stock in Miasteczko Wilanow 219is relatively homogenous in terms of building height, architectural design, building con220struction technology, overall quality, and age (construction started in 2002). We believe 221that despite not directly controlling for other neighbourhood quality features, there are no 222substantial unobserved differences within the study area.2234 Results and discussion2244.1 Hedonic regression and matching estimators225To perform a comprehensive inference, calculations were performed for six models in the 226two groups (Table 3). The first three (Mod. 1 - Mod. 3) are baseline hedonic regression 227models based on all transactions within the sample (N = 1825). The remaining models (Mod. 2284 - Mod. 6) use the matching transaction data. We applied the Coarsened Exact Matching 229(CEM) algorithm(Iacus et al., 2011) to match residential units in proximity to BTS (treated 230group) to relatively similar apartments that were not directly affected by the presence of 231BTS (control group). The sample size in these models was small (N = 834). Within each 232group, we used different measures of exposure to BTS, as described in detail in the previous 233section. In particular, we control for the presence of BTS using a binary proximity measure 234(BTS_Prox), linear distance (BTS_Dist), and number of BTS in proximity to the property 235(BTS_Num).236We use a log-linear specification; thus, the natural logarithm of the sale price is a depen237dent variable (Price). Table 3 presents the estimation results.238The R2 coefficients obtained from the hedonic regression models range from 0.883 for 239the baseline models (Mod. 1- Mod. 3) to 0.875 for the CEM matching. The models exhibited 240a relatively good fit with the empirical data. All models show price growth in the subsequent 241years of the analysis, which we controlled for using quarterly fixed effects time dummies.242Generally, the coefficients estimated for the statistically significant variables have intui243tive (logical) signs. Along with expectations, the area is positively linked with sale prices; 244bigger apartments are more expensive, and others remain equal. The estimated price elas245ticities (0.88–0.92) have an intuitive economic interpretation that is also consistent with the 246law of diminishing marginal utility derived from an increase in apartment size. The number 247of rooms did not have a statistically significant effect on apartment sale prices after control248ling for the usable area.249A similar interpretation applies to the Floori variables (the story where the apartment was 250located). There was a significant discount related to the location on the ground floor (base251line category) compared with the other floors. Other floors remained equal, whereas the sale 252prices were significantly higher in the case of apartments located on the first or second floor 253(Floor3). Street noise (Noise) did not have a significant impact on property prices within 254our sample, potentially because it was not seen as a crucial factor differentiating properties. 2551 32561833257258259===== PAGE 14 (score 129, numeric-paren 97, table-words 6, var-words 6) =====260261B. Marona et al.262Distance from the closest school has negative impacts on apartment prices. The result is 263also intuitive; in general, the farther from the school, the lower the apartment price, and the 264other stay equal. However, we report that the positive coefficient for the bus variable – the 265increase in distance from the bus stop–is associated with an increase in apartment prices. 266This finding is somewhat surprising; however, the marginal increase in accessibility may be 267due to disturbances caused by the traffic associated with the operation of buses in this par268ticular district. Additionally, in all models, we controlled for parties involved in apartment 269sales (Trans_C). We observed that transaction prices, where either a buyer or seller was a 270business organisation, were significantly lower than in the case of scale between private 271market participants, while others remained equal. One potential explanation is that, on the 272one hand, companies are less likely to be anchored by unrealistic offer prices compared to 273private market participants when selling real estate. However, when buying apartments, 274Table 3 Hedonic regression models. The dependent variable is the natural logarithm of the sales price275Variables276Mod. 1277Mod. 2278Mod. 3279Mod. 4280Mod. 5281Mod. 6282BTS_Prox2830.014*284-0.001285(0.007)286(0.008)287BTS_Dist2880.0062890.008290(0.004)291(0.005)292BTS_Num2930.0062940.003295(0.004)296(0.005)297Ln Area2980.913***2990.919***3000.912***3010.882***3020.886***3030.880***304(0.017)305(0.017)306(0.017)307(0.023)308(0.023)309(0.023)310Rooms3110.0083120.0063130.0083140.0113150.0103160.011317(0.006)318(0.006)319(0.006)320(0.008)321(0.008)322(0.008)323Floor23240.0143250.0153260.014327-0.007328-0.007329-0.007330(0.009)331(0.009)332(0.009)333(0.013)334(0.013)335(0.013)336Floor33370.076***3380.077***3390.076***3400.037**3410.037**3420.037**343(0.010)344(0.010)345(0.010)346(0.013)347(0.013)348(0.013)349Noise3500.000351-0.001352-0.000353-0.006354-0.006355-0.006356(0.004)357(0.004)358(0.004)359(0.005)360(0.005)361(0.005)362School363-0.006***364-0.005***365-0.006***366-0.006***367-0.006**368-0.006**369(0.001)370(0.001)371(0.001)372(0.002)373(0.002)374(0.002)375Bus3760.010***3770.009**3780.011***3790.013***3800.011**3810.013***382(0.003)383(0.003)384(0.003)385(0.004)386(0.004)387(0.004)388Trans_C389-0.097***390-0.097***391-0.097***392-0.039*393-0.039*394-0.040*395(0.013)396(0.013)397(0.013)398(0.018)399(0.018)400(0.018)401Constant4029.302***4039.278***4049.307***4059.525***4069.496***4079.525***408(0.062)409(0.064)410(0.062)411(0.084)412(0.086)413(0.084)414Time fixed-effects (Quarterly)415yes416yes417yes418yes419yes420yes421Matching (CEM)422no423no424no425yes426yes427yes428R-squared4290.8834300.8834310.8834320.8754330.8754340.875435Adj R-squared4360.8814370.8814380.8814390.8704400.8704410.870442AIC443-1935.767444-1933.516445-1933.813446-1153.1447-1155.791448-1153.508449BIC450-1764.977451-1762.727452-1763.024453-1006.586454-1009.277455-1006.995456N457182545818254591825460834461834462834463Note Standard errors are in parentheses. Statistical significance * p < 0.05, ** p < 0.01, *** p < 0.0014641 34651834466467468===== PAGE 15 (score 12, numeric-paren 1, table-words 4, var-words 3) =====469470Do 5G cell phone towers decrease house prices? Evidence from Warsaw471corporate buyers may be more rational and not guided by emotions. The results were statis472tically significant for all specifications.473In summary, taking into account the economic interpretation of the results obtained for 474the control variables, both the coefficients’ signs and values are fairly consistent with the 475knowledge of the particular factors’ impact on real estate prices. Generally, it should also 476be noted that the results obtained for the control variables are consistent and stable for all 477specifications used, as they are not always statistically significant.478The most relevant estimates were related to the BTS (BTS_Prox, BTS_Dist, and BTS_479Num). The signs of the coefficient do not necessarily align with a priori expectations based 480on results from previous studies and widespread public beliefs. Nonetheless, both of these 481variables proved to be statistically significant. In the case of proximity to BTS (BTS_Prox, 482which was used in Mods 1 and 4), the coefficient was positive (0.004), which is counterin483tuitive. If statistically significant, the results would suggest that apartments located within 484a 200 m radius from the BTS were generally more expensive. Paradoxically, and more 485in line with expectations, the coefficient is also positive in the case of distance from BTS 486(BTS_Dist) (Mod 2 and Mod 5), and the number of BTS within the 200 m radius from the 487apartment (BTS_Num) variables that were used in alternative specifications. The results are 488mixed, but due to the lack of statistical significance, they are not conclusive.4894.2 Difference in differences estimator490We used differences-in-differences regression (DID) to assess the causal effect of the intro491duction of 5G technology in May 2020 and compared the transaction prices of residential 492properties in proximity to BTS with 5G antennas (treatment group) to the sales prices of 493comparable housing units not affected by the event (control group). In line with the empiri494cal specifications used in hedonic regression models, we control for the presence of BTS 495using a binary proximity measure (BTS_Prox), linear distance (BTS_Dist), and the number 496of BTS in proximity to the property (BTS_Num). The estimation results of the three DiD 497models (Mod. 7 - Mod. 9) are listed in Table 4.498DiD is reasonably well suited to empirical data. Additionally, we tested the parallel-499trend assumption. Based on the test results (Prob > F = 0.073), we could not reject the null 500hypothesis that the linear trends were parallel in the pretreatment period. Additionally, 501based on the Granger causality test, we note that anticipation of treatment had no effect 502(Prob > F = 0.1425).503We used the Average Treatment Effect on the Treated (ATET) to assess the causal effect. 504ATET coefficients (ranging from − 0.022 to -0.021) were not statistically significant in any 505specification. We conclude that the introduction of 5G technology did not have a significant 506effect on residential property prices in the study area. We note that there is no solid support 507for the claim that sale prices dropped significantly in the vicinity of those BTS that offered 5085G signals. This effect is presented graphically in Fig. 2.509Telecommunication infrastructure cannot create in my back yard (NIMBY) opposition 510of residents living in proximity. Wireless communication infrastructure, similar to other 511similar developments related to conflicting land uses, may generate neighbourhood conflicts 512due to the potential stigmatisation of the area or subjective assessment of environmental and 513health risks by the local community (Bennett & Davies, 2015; Llurdés et al., 2003; Mei et 514al., 2021). These concerns often manifest as decreases in house values in neighbourhoods.5151 35161835517518519===== PAGE 16 (score 63, numeric-paren 45, table-words 2, var-words 7) =====520521B. Marona et al.522The results indicate that this was not the case for BTS in Miasteczko Wilanów, as we did 523not find significant evidence that prices were lower in proximity. The lack of an economic 524impact of BTS on apartment prices requires further comment, as has been observed in many 525prior studies, especially when neighbourhood conflicts were intense and extensively cov526ered by the media (Bond, 2007a). First, the turbulent introduction of BTS has become a 527more acceptable part of urban and rural environments. In recent years, more people have 528used wireless technology and are reluctant to express negative emotional arousal when 529exposed to the presence of BTS. Many recent studies suggest that the negative impact of 530BTS is currently smaller than that during the early 2000s and is often not statistically signifi531cant (Witkowski & McLaughlin, 2021). Second, the technologies also changed significantly 532and blended more aesthetically into the city landscape. Currently, BTS are less visible and 533are often concealed. Other studies have suggested that the potential impact of BTS on house 534prices in the vicinity is significantly reduced when the disamenity is not directly visible 535(Affuso et al., 2018).536Table 4 Difference-in-differences regression models. The dependent variable is a natural logarithm of the 537sale price538Variables539Mod. 7540Mod. 8541Mod. 9542ATET543Treatment (BTS_5G*after)544-0.021545-0.022546-0.022547-0.021548(0.005)549(0.004)550(0.005)551(0.005)552BTS_Prox5530.026554(0.005)555BTS_Dist5560.010557(0.011)558BTS_Num5590.014560(0.012)561Ln Area5620.919*5630.916*5640.915*565(0.019)566(0.015)567(0.020)568Rooms5690.0075700.0075710.007572(0.009)573(0.008)574(0.009)575Floor25760.0155770.0155780.014579(0.004)580(0.005)581(0.005)582Floor35830.0775840.077*5850.077586(0.007)587(0.005)588(0.006)589Noise590-0.001591-0.001592-0.002593(0.003)594(0.003)595(0.003)596School597-0.006*598-0.005599-0.006600(0.000)601(0.001)602(0.001)603Bus6040.0126050.0076060.013607(0.006)608(0.006)609(0.005)610Trans_C611-0.098*612-0.095*613-0.098*614(0.006)615(0.003)616(0.005)617Constant6189.275**6199.278**6209.288**621(0.064)622(0.019)623(0.071)624Time fixed-effects (Quarterly)625yes626yes627yes628N629182563018256311825632Note ATET estimates are adjusted for covariates, group effects, and time effects. Note: Standard errors are 633in parenthesis. Statistical significance * p < 0.05, ** p < 0.01, *** p < 0.0016341 36351836636637638===== PAGE 17 (score 3, numeric-paren 0, table-words 0, var-words 3) =====639640Do 5G cell phone towers decrease house prices? Evidence from Warsaw641Our study had some limitations. First, it is unclear whether the results can be generalised 642to other rural and urban areas in Poland. Most likely, the results would be similar in major 643cities, where telecommunication infrastructure blends within urban landscapes and the 644majority of market participants are relatively unaware of their exact location, but different 645in rural and suburban areas where cell phone towers are less concealed. Extending empirical 646investigations beyond major cities to less urbanised areas seems to be a promising direc647tion for future research. Second, despite our best efforts, we were unable to control for all 648relevant housing attributes. Unobserved heterogeneity may lead to biased estimation results, 649as suggested by econometric literature (Francke & Van de Minne, 2021). We attempted to 650reduce this risk by employing matching techniques and quasi-experimental settings. Finally, 651the fact that the proximity of BTS did not affect residential property prices, may be due to 652asymmetric information between buyers and sellers. Telecommunication infrastructure has 653blended into the urban landscape thus buyers are often not aware of the presence of BTS 654(Klaps et al., 2016). As a result, the information about the proximity of BTS may not be 655revealed in hedonic prices (Pope, 2008). It is worth extending the research to include stated 656preferences data, particularly in terms of buyers’ attitudes towards 5G networks but also 657awareness that this infrastructure exists in the location where they live.6585 Conclusion659This article engages in a discussion on the perception of telecommunication infrastructure 660within an urban landscape. This research provides new evidence on how base transceiver 661stations (BTS) influence housing prices. This study aimed to determine the impact of the 662Fig. 2 Causal effect of the introduction of 5G in Miasteczko Wilanów663 6641 36651837666