1# RECORD 3fff47f2ad1427c42ba357bc9b2363011924b07a2# TITLE: Short-term rentals and the housing market: evidence from Portuguese metropolitan areas3# YEAR: 2026.0 VENUE: Portuguese Economic Journal DOI: 10.1007/s10258-026-00295-24# AUTHORS: Francisco Nobre; Diogo Jardim Gonçalves; Ronize Cruz5# PAGES: 35 CANDIDATE TABLE PAGES (1-based): [11, 17, 20, 21, 22, 27, 31, 32]6# ABSTRACT: 7 In this paper, we study the impact of short-term rentals (STR) on housing markets in Portugal, exploiting a 2014 policy reform that eased registration requirements. Using a novel dataset of property transactions from 2010 to 2017 for the Lisbon and Porto metropolitan areas, we estimate the effect of STR expansion using a two-way fixed effects model and a difference-in-differences design that leverages staggered treatment across postcodes. The analysis controls for detailed property characteristics, location and time fixed effects. We find that each additional STR per quarter and postcode raises transaction values by 2.3–6.0%, representing an average increase of €104–€329 per transaction. Effects are larger in higher-end market segments and in parishes with lower unemployment and population density. Event-study estimates show that effects emerge gradually and become significant around ten quarters after the first STR listing. We also find positive spillovers to commercial property values, particularly for retail stores, and in suggesting short-term rentals drive demand for local consumption amenities.8910===== PAGE 1 (score 14, numeric-paren 0, table-words 0, var-words 14) =====1112ORIGINAL ARTICLE13Portuguese Economic Journal14https://doi.org/10.1007/s10258-026-00295-215 Francisco Nobre16f.nobre@kingston.ac.uk17Diogo Jardim Gonçalves18diogo.jardimgoncalves@bristol.ac.uk19Ronize Cruz20ronize.cruz@fe.uc.pt211 22Kingston Business School, Kingston University, London, United Kingdom232 24School of Economics, University of Bristol, 12 Priory Road, Clifton, Bristol 25BS8 1TU, United Kingdom263 27Centre for Business and Economics Research, Faculty of Economics, University of 28Coimbra, Av. Dr. Dias da Silva 165, Coimbra 3004-512, Portugal29Received: 22 January 2024 / Accepted: 26 July 202630© The Author(s) 202631Short-term rentals and the housing market: evidence from 32Portuguese metropolitan areas33Francisco Nobre134 · Diogo Jardim Gonçalves2 · Ronize Cruz335Abstract36In this paper, we study the impact of short-term rentals (STR) on housing markets 37in Portugal, exploiting a 2014 policy reform that eased registration requirements. 38Using a novel dataset of property transactions from 2010 to 2017 for the Lisbon 39and Porto metropolitan areas, we estimate the effect of STR expansion using a 40two-way fixed effects model and a difference-in-differences design that leverages 41staggered treatment across postcodes. The analysis controls for detailed property 42characteristics, location and time fixed effects. We find that each additional STR 43per quarter and postcode raises transaction values by 2.3–6.0%, representing an 44average increase of €104–€329 per transaction. Effects are larger in higher-end 45market segments and in parishes with lower unemployment and population density. 46Event-study estimates show that effects emerge gradually and become significant 47around ten quarters after the first STR listing. We also find positive spillovers to 48commercial property values, particularly for retail stores, and in suggesting short-49term rentals drive demand for local consumption amenities.50Keywords Policy analysis · Housing market · Portugal · Short-term 51accommodation525354===== PAGE 11 (score 18, numeric-paren 0, table-words 4, var-words 8) =====55561 357Short-term rentals and the housing market: evidence from Portuguese…58or more of its attributes. Hence, the dataset contains information not only on property 59attributes but also on the specific reasons prompting each evaluation. Lastly, we link 60each transaction to its most recent evaluation through a unique property identifier 61(Evangelista et al. 2020).62Our sample consists of properties with either a single transaction, more than one 63transaction on the same day or with multiple transactions on different days. In total, 64over the time period considered, we have a sample of 134,079 properties, corre65sponding to 185,654 transactions. Moreover, we have 41,526 properties in the met66ropolitan area of Porto and 92,553 in Lisbon. Table 1 presents descriptive statistics 67of the properties and number of transactions, for those that we were able to match 68Number of 69transactions (per 70property)71Mean72Std. Dev.73Min74Max751.385760.69777178979Number of 80transactions 81(per quarter and 82postcode)832.015843.7868518615187Number of 88transactions (per 89quarter and civil 90parish)9140.7329248.1999319432895Number of trans96actions (per year 97and postcode)983.280998.8811001101407102Number of trans103actions (per year 104and civil parish)105134.046106179.99910711081160109Number of 110transactions (per 111postcode)1129.34811331.65711411151511116Number of trans117actions (per civil 118parish)119851.6241201,011.05312111224610123Number of 124transactions (per 125quarter)1265,801.6881273,737.2541281,54112913,011130Number of 131transactions (per 132year)13323,206.7513415,336.261358,65013647,909137Value of 138transactions139209,655140569,802.41412,002.491427,883,690143Number of 144bedrooms1452.9071461.203147014812149Number of floors1501.0901510.385152115313154Area (square 155meters)156101.25415743.3011583.6159762.73160Age of the build161ing (years)16223.81316321.4221640165298166Table 1 Summary statistics of 167properties transacted168Total number of transactions: 169185,654. This table reports 170the mean, standard deviation, 171minimum, and maximum 172values for key summary 173statistics of the transaction 174data. Transaction counts are 175presented across several time 176and geographical dimensions, 177including per property, quarter, 178year, postcode, and civil parish. 179The table also reports the value 180of transactions and property-181specific characteristics such 182as the number of bedrooms, 183floors, area (in square meters), 184and the age of the building. 185Source: IMT and IMI tax 186registries, Statistics Portugal187 188189190===== PAGE 12 (score 17, numeric-paren 0, table-words 3, var-words 11) =====1911921 3193F. Nobre et al.194between the two tax registries.9 On average we have approximately 1.4 transactions 195per property, with the maximum being 9. We have around 2 transactions per quarter 196and postcode, with a sample mean value of e 209,655. On average, the properties 197contain, 3 bedrooms, 1 floor, and around 100 square meters and are in 23-year-old 198buildings.1993.3 Imovirtual200We employ a novel dataset from Imovirtual, one of the largest online real estate 201platforms in Portugal, which started to operate in 2011. The data provided by the 202website refers to listings of commercial properties, containing information on the 203type of property (store, office and storage unit), asking price, geographical location, 204square footage, operation (rental or sale) and the exact date the advertisement of the 205property became available online. Although we possess data from 2011, we choose 206as period of analysis the years 2015–2017, as we have a more balanced number of 207online advertisements during this period, as depicted in Figure A3.208In total, we possess 39,205 observations, with stores for sale comprising approxi209mately 51% of our sample. Around 60% of our listed properties are located in the 210Metropolitan area of Lisbon. Table 8 reports the summary statistics for the (asking) 211price and number of advertisements (per quarter and postcode) for the different types 212of properties in our sample, with offices for sale having a higher average value and 213stores for sale being the most advertised.2144 Identification strategy215When estimating the effects of the development of local lodging establishments on 216surrounding house prices several factors could confound our estimations. For exam217ple, it could be the case that in a particular neighborhood a considerable number of 218properties are being transacted to then be transformed into local lodging establish219ments. We attempt to shed some light on this issue, by taking advantage of the unique 220features of our dataset and geolocating the properties in transaction dataset to the 221closest local lodging, through ArcGIS.10 As a result, for each property transacted we 222obtain all the short-term rentals that are within 35 m. Moreover, since we also have 223information pertaining to the characteristics of these local lodgings, we drop transac224tion for properties that have the same number of bedrooms of a nearby short-term 225rental. In Table A6 we present some descriptive statistics for properties that were 226dropped due to being near to a short-term rental and having the same number of 227bedrooms. In total, we identify 7,901 properties that are within 35 m of the nearest 228short-term rental and have the same number of bedrooms. These properties have on 2299 For some properties, a match between the two tax registries could not be established, resulting in the 230inability to obtain their characteristics.23110 The geolocation described was provided by Statistics Portugal, as the local Safe center do not possess 232geospatial software.233234235===== PAGE 17 (score 46, numeric-paren 16, table-words 10, var-words 8) =====2362371 3238Short-term rentals and the housing market: evidence from Portuguese…239Table 2 Baseline model240Variables241Prices242Log(prices)243Log(prices)244Log(prices)245Log(prices)246Log(prices)247Log(prices)248Log(prices)249Residualized250Number of Local Lodging2510.0394**2520.0394*2530.0394**2540.0605**2550.0431**2560.0233*2570.0280**2580.0492***259(0.0166)260(0.0201)261(0.0196)262(0.0258)263(0.0196)264(0.0125)265(0.0128)266(0.0159)267Number of bedrooms268✓269✓270✓271✓272✓273✓274✓275✓276Number of floors277✓278✓279✓280✓281✓282✓283✓284✓285Area (square meters)286✓287✓288✓289✓290✓291✓292✓293✓294Age of building295✓296✓297✓298✓299✓300✓301✓302✓303Coefficients of location and quality304✓305✓306✓307✓308✓309✓310✓311✓3127-digit postcode FE313✓314✓315✓316No317No318✓319No320No321Quarter FE322✓323✓324✓325✓326✓327✓328✓329✓3304-digit postcode FE331No332No333No334✓335No336No337No338No339Civil parish FE340No341No342No343No344✓345No346✓347✓348Civil parish-time trends349No350No351No352No353No354✓355✓356✓357Number of clusters35815,833359157360217361158362165363157364165365165366Observations367181,626368181,626369181,626370185,647371185,653372181,626373185,653374185,653375R23760.5103770.5103780.5103790.3233800.3283810.5173820.3423830.178384 Results for the baseline model at the quarterly level, where we regress the logarithm of transactions prices on the number of local lodging establishments opening in a 385quarter and postcode. Columns (1)–(3) present the results including full postcode and quarter fixed effects, while clustering standard errors at the full postcode, postcode 386sector (first 4 digits) and civil parish, respectively. In columns (4)–(5), employ postcode sector, civil parish fixed effects, respectively, while in columns (6)-(7) we use full 387postcode and civil parish fixed effects, respectively, and civil parish time trends. In columns (4)-(7), standard errors are clustered at the postcode sector level. In column 388(8) we use residualized prices as outcome variables. All regressions include a constant and property-specific controls. Standard errors are clustered at postcode sector. *** 3891%, ** 5%, * 10%. Data sources: IMT and IMI tax registries (Statistics Portugal); RNAL (Turismo de Portugal)390391392===== PAGE 18 (score 9, numeric-paren 0, table-words 3, var-words 3) =====3933941 3395F. Nobre et al.396is lower between 2.3 and 2.8%, but still statistically significant. In column (8), we 397show the results for our detrending procedure outlined in the previous section, where 398we first regress our outcome variable on quarter, parish fixed effects and parish-spe399cific time trends, capture the residuals, and then regress the residuals on the number 400of local lodgings, property characteristics and time dummies. The point estimate is 401larger, around 5%, and highly significant. Overall, our main effects range from 2% to 4026%, showing robustness to different specifications.403For the remainder of the section, we use one of our most demanding specifications 404with property attributes, civil parish time trends, while also employing civil parish 405fixed effect, to control for neighborhood unobservables, instead of full postcode fixed 406effect, which could be too strict. In the next exercise, we provide suggestive evidence 407of whether short-term rentals affect the number of transactions and transactions of 408new built properties, which can be seen in Table 9. In the top panel we use as outcome 409the number of transactions in a given postcode and quarter, focusing on those with at 410least one transaction. We find a positive and significant impact for the entire sample 411and both metropolitan areas, with the coefficient being of larger magnitude in Porto. 412In the bottom panel, we use as outcome variable a binary variable that takes the value 413of 1 if in a postcode and quarter there is a transaction occurring after an evaluation 414from the fiscal authority was made due to a property being new.415Although our outcome variable of interest constitutes an imperfect measure of 416housing supply, it is the closest we have to know whether a transaction has happened 417in a new building. We observe a negative, although, non-significant, effect of local 418lodgings on the transactions of newly built properties for the overall sample, with 419this relationship being negative and statistically significant in the metropolitan area 420of Lisbon. However, for the metropolitan area of Porto, the coefficient is positive 421and marginally significant. Overall, although the impact on the transaction of new 422housing is inconclusive, there is some indication that in certain areas, it can lead to a 423reduction in housing supply.424Overall, our main results indicate that a rise in the quantity of local lodging estab425lishments corresponds to an increase in transaction prices. Given that, on average, 4260.026 new venues open per quarter and postcode with transaction, this implies that 427over the duration of our study period, there was a quarterly increment of between 4280.05 and 0.16% in transaction values, which is a significant and noteworthy change.429It is important to place our findings within the context of the broader literature on 430the effects of short-term rentals on house prices. However, this can be a challenging 431endeavor, as other studies have looked at different types of policies, used alternative 432specifications and data and even employed other measures to assess local lodging 433activity. For example, Gonçalves et al. (2022); Barron et al. (2021); Koster et al. 434(2021), have looked at the effect of placing restrictions on short-term rentals in Lis435bon, Chicago, San Francisco and Los Angeles, respectively. Overall, they find that 436it led to a decrease in house prices ranging from 2% to 10%. Even though, we also 437document a positive relationship between new short-term rental listings and housing 438prices, we analyze a period of continuous growth in short-term rental activity, provid439ing it a very different context that those in the studies mentioned.440To the best of our knowledge, the two papers more closely related to our research 441are Franco and Santos (2021) and Garcia-López et al. (2020), who investigate the 442443444===== PAGE 20 (score 43, numeric-paren 12, table-words 11, var-words 7) =====4454461 3447F. Nobre et al.448In column (4), we drop observations in the 95th percentile in terms of number of 449local lodging establishments by civil parish and quarter, with estimates around higher 450than the baseline model and still significant at the 10% level. In column (5), we use 451the years before the policy change as a placebo test, finding no significant results 452and showing that it was the exponential increment in short-term rentals after 2014 453that led to an increase in house prices. In column (6), we use a measure of price per 454square meter as outcome variable, which provides again stronger and highly signifi455cant results than our baseline model. Given an average of 0.026 local lodgings per 456quarter and postcode, our point estimates suggest that it led to a marginal increase 457Table 3 Robustness checks458Variables459Drop be460fore 2014461Only 1 462transaction 463per day464No new 465built466Dropping 46795pct LL468Drop after 4692013470SUTVA471Log(prices) Log(prices) Log(prices) Log(prices) Log(prices) Log(prices/472m2)473Log(prices)474Number 475of LL4760.0233*4770.0314**4780.0342***4790.0306*480−0.003164810.0342***482(0.0128)483(0.0128)484(0.0128)485(0.0178)486(0.0519)487(0.0120)488LL 489neigh490bourhood 491postcodes4920.00775493(0.0544)494Property 495attributes496✓497✓498✓499✓500✓501✓502✓503Civil par504ish FE505✓506✓507✓508✓509✓510✓511✓512Quarter 513FE514✓515✓516✓517✓518✓519✓520✓521Civil 522parish-523time 524trends525✓526✓527✓528✓529✓530✓531✓532Num533ber of 534clusters535164536165537165538163539139540165541165542Observa543tions544144,579545177,182546167,032547178,32754841,067549185,653550185,653551R25520.3725530.3415540.3045550.3405560.2935570.2265580.341559This table reports the estimates for a set of robustness checks. In column (1) we consider only transactions 560from 2014 onwards, in column (2) only properties with one transaction per day, in column (3) we drop 561the first transaction after an evaluation by fiscal authorities occurred due to the building being new, in 562column (4) we drop areas in the 95th percentile in terms of number of local lodging establishments per 563quarter and civil parish, in column (5) we only consider the period before 2014, as a placebo test. In 564columns (1)–(5), the outcome variable is the log value of transactions in the present period for a given 565postcode. In column (6), the outcome variable is a measure of price per square meter. In column (7), 566we test for the SUTVA condition, with the outcome variable being the log value of transactions for 567a given postcode and quarter and the main explanatory variable being a quarterly average of Local 568Lodgings of all postcodes with the same 6-digits. All regressions include a constant, property-specific 569controls, quarter and civil parish fixed effects, civil parish specific time trends. We exclude properties 570that are within 35 m from a Local Lodging and have the same number of bedrooms. Standard errors are 571clustered at postcode sector. *** 1%, ** 5%, * 10%. Data sources: IMT and IMI tax registries (Statistics 572Portugal); RNAL (Turismo de Portugal)573574575===== PAGE 21 (score 29, numeric-paren 5, table-words 8, var-words 8) =====5765771 3578Short-term rentals and the housing market: evidence from Portuguese…579of around €186 of transactions prices per square meter, still lower than the results 580reported in Franco and Santos (2021).581Lastly, in column (7) we test for whether the SUTVA condition holds in our set582ting. Formally, this assumption states that treatment in one unit should not be affected 583by treatment in other units. To present some suggestive evidence that this condition 584holds, we run a regression where the outcome refers to the log of transactions value 585for a given postcode and quarter and as main explanatory variable, we employ an 586average per quarter of local lodgings opening in the postcodes with the same first 587six digits, which represent nearby areas. If the assumption holds, presumably the 588coefficient would be non-significant. Indeed, this is the case, as seen in column (7), 589supporting the validity of the SUTVA condition in our context.590Then, we analyze the timing and persistence of our effects and perform a falsifica591tion exercise, displayed in Table 4. In column (1), we use as variable of interest the 592number of local lodging establishments opening one quarter before, respectively. The 593effects exhibit a considerable level of persistence, with the number of new venues 594opening 1 quarter before having a positive and significant impact on the value of 595transactions in the current period, being larger in magnitude than the baseline model.596The persistence in these effects could be ascribed to the renovation and revamping 597necessary to turn the establishments into temporary accommodation services, or local 598lodging. This transformation process might increase the attractiveness of a particular 599area, potentially leading to a higher level of tourism, raising the value of the prop600erties located there. These effects can persist over subsequent periods, influencing 601property values in the long run.602Table 4 Falsification exercise and timing of the effects603Variables604Log(prices) 605Log(prices) 606Log(prices)607Number of Local Lodgingt−16080.0734***609(0.0237)610Number of Local Lodgingt6110.0280**612(0.0128)613Number of Local Lodgingt+16140.0280615(0.0172)616Property attributes617✓618✓619✓620Civil parish FE621✓622✓623✓624Quarter FE625✓626✓627✓628Civil parish time trends629✓630✓631✓632Number of clusters633165634165635165636Observations637183,316638185,653639172,640640R-squared6410.3436420.3426430.337644Results for the falsification and timing of the effects exercises, at the quarterly level. Column (1) presents 645the results when using the number of local lodging establishments opening the quarter before. In column 646(2) is the baseline model. Column (3) displays the estimates when using the number of establishments 647opening the quarter after. In all specifications, the outcome variable is the log value of transactions in 648the present period and include a constant, the full set of property-specific controls, quarter and civil 649parish fixed effects, civil parish specific time trends. We exclude properties that are within 35 m from a 650Local Lodging and have the same number of bedrooms. Standard errors are clustered at postcode sector. 651*** 1%, ** 5%, * 10%. Data source: IMT and IMI tax registries (Statistics Portugal); RNAL (Turismo 652de Portugal)653654655===== PAGE 22 (score 44, numeric-paren 10, table-words 13, var-words 7) =====6566571 3658F. Nobre et al.659In column (3) we present a falsification exercise. As previously stated, if our speci660fication is correct, the number of establishments opening periods ahead should not 661affect house prices at the current period. This is confirmed by the non-significant, and 662even negative, coefficient obtained when using the number of new local lodgings in 663a civil parish in the quarter ahead. In sum, the robustness checks performed reinforce 664the conclusions of the main analysis in Table 2, while validating the appropriateness 665of the identification strategy.6665.1.3 Heterogeneous effects667We proceed by leveraging the extensive amount of information in our dataset to 668perform a series of heterogeneous exercises, using our specification at the quarterly 669level and including quarter and civil parish fixed effects and civil parish specific time 670trends. Our aim is to understand how the effects differ by geographical locations and 671by types of property, which addresses a gap in the literature.672We start by dividing our sample by metropolitan areas, with specific emphasis on 673the municipalities of Lisbon and Porto. For the Portuguese case, we believe this focus 674is particularly relevant to understand how the effects differ in the two biggest cities, 675addressing a crucial gap in the existing literature. The results for this comparative 676analysis are displayed in Table 5. In columns (1) and (4), we look at the metropolitan 677areas of Lisbon and Porto, respectively. The effects are stronger for the metropolitan 678area of Porto, both in terms of magnitude and significance. In fact, for the metropoli679tan area of Porto the coefficient estimated is even slightly higher than the baseline 680model. In column (2) and (5), we narrow down the geographical scope of our analy681sis and focus on the municipalities of Lisbon and Porto, respectively. We now find 682Table 5 Heterogeneous effects: by metropolitan area683Variables684Lisbon685Porto686Metropolitan area687Log(prices)688Municipality689Log(prices)690Historical center691Log(prices)692Metropolitan area693Log(prices)694Municipality695Log(prices)696Historical 697center698Log(prices)699Number of LL7000.02017010.0303*7020.0576*7030.0394*7040.01347050.0185*706(0.0149)707(0.0165)708(0.0298)709(0.0230)710(0.0122)711(0.00966)712Property attributes713✓714✓715✓716✓717✓718✓719Quarter FE720✓721✓722✓723✓724✓725✓726Civil parish FE727✓728✓729✓730✓731✓732✓733Civil parish time trends734✓735✓736✓737✓738✓739✓740Number of clusters7411187423574313744477451274611747Observations748127,34774933,9997506,97375158,30675214,1767535,087754R27550.3567560.3337570.3707580.3097590.3727600.410761Results for heterogeneous exercises, by metropolitan area, at the quarterly level. Columns (1) and (3) present the results when 762considering separately the metropolitan areas of Lisbon and Porto, respectively. Columns (2) and (4) display the results when 763considering separately the municipalities of Lisbon and Porto, respectively. Columns (3) and (6) show the estimates when 764considering separately the historical centers of Lisbon and Porto, respectively. In all specifications, the outcome variable is 765the log value of transactions in the present period and include a constant, property-specific controls, quarter and civil parish 766fixed effects and civil parish time trends. We exclude properties that are within 35 m from a Local Lodging and have the same 767number of bedrooms. Standard errors are clustered by postcode sector. *** 1%, ** 5%, * 10%. Data sources: IMT and IMI tax 768registries (Statistics Portugal); RNAL (Turismo de Portugal)769770771===== PAGE 23 (score 12, numeric-paren 0, table-words 3, var-words 6) =====7727731 3774Short-term rentals and the housing market: evidence from Portuguese…775stronger effects for the municipality of Lisbon, with the point estimate being again 776slightly larger than the corresponding specification in the baseline model. Lastly, in 777columns (3) and (6), we consider solely the Historical centers of Lisbon and Porto, 778respectively.11 Our findings suggest that the impact on house prices is stronger in 779the Historical center of Lisbon, with the coefficient being 3 times larger than for the 780Historical center of Porto, while being equally significant.781The results in Table 5 reveal some interesting features. Firstly, when controlling 782for the time, civil parish fixed effects and civil parish time trends, our findings sug783gest that the effects in Lisbon are quite constricted to its municipality and historical 784center, while in Porto the impact of local lodgings is broader and felt across the entire 785metropolitan area. Secondly, and recalling that civil parish-specific linear time trends 786allow us to explore variations from the parishes’ own trend, our results indicate that 787the impact of local lodgings on these deviations from the trend are very similar across 788the municipality of Porto. In this regard, the point estimates are nearly identical when 789considering the entire municipality or just the historical center of Porto. On the other 790hand, for Lisbon, the coefficients estimated increase in magnitude and significance as 791we narrow down our scope of analysis. For the entire metropolitan area, the impact 792of local lodgings on deviations from the individual time trends of each parish trends 793appears to be non-significant, while for the historical center, the effect of short-term 794rentals is around 2 times higher than the baseline model.795Up to this point, our results seem to suggest that the impact of local lodging 796establishments is not homogeneous across regions. The absence of effects in the 797municipality of Porto might be considered surprising as it is at odds with the findings 798documented in Franco and Santos (2021), who evidence a positive and significant 799effect in both municipalities. We further scrutinize the impact of short-term rentals 800on house prices in the municipalities through quantile regressions, which allows us 801to estimate the difference in effects across our distribution. We follow the estimator 802proposed in Machado and Santos Silva (2019), which makes it possible to perform 803quantile regressions with two fixed effects. In order to allow to control for time trends 804as well, we adopt the price residualized approach, by first regressing the log of trans805action values on civil parish, quarter fixed effect and parish time trends, capturing the 806residuals and using those as the outcome in our quantile analysis, while controlling 807for property characteristics and time dummies. We focus on the 0.10,0.25,0.5,0.75 808and 0.90 quantiles and our results are displayed graphically in Fig. 4.809The results reveal some interesting patterns. For the municipality of Lisbon, the 810coefficient estimated is insignificant at 0.10 percentile, while in the case of Porto, the 811point estimate is significant in all quantiles. Moreover, while in both municipalities 812the impact of short-term rentals is stronger at the upper quantiles, which represent the 813higher-end of the housing market, the difference in effects across the distribution is 814much smaller for Porto. In fact, while in Porto, the coefficients range from 0.0550 to 8150.0559, in Lisbon they range from 0.0363 to 0.0535. Nevertheless, the larger effects 81611 We consider part of the Historical center of Lisbon the civil parishes of Campo de Ourique, Estrela, 817Misericórdia, Penha de França, Santa Maria Maior and São Vicente, while the Historical center of Porto 818is composed by the parishes of Bonfim, União das freguesias de Cedofeita, Santo Ildefonso, Sé Miragaia, 819São Nicolau e Vitória and União das freguesias de Lordelo do Ouro e Massarelos.820821822===== PAGE 27 (score 39, numeric-paren 16, table-words 7, var-words 6) =====8238241 3825Short-term rentals and the housing market: evidence from Portuguese…826periods after treatment we see an upward trend, which becomes more evident after 827period 9. The point estimates become significant at period 10 after treatment, which 828corresponds to around two and a half years after the first short-term rental appears. 829Moreover, in periods 10 and above the ATT can reach up to 0.79, with the aggregate 830post-treatment ATT being around 0.20 and statistically significant at 1% level.831Table 6 Spillover effects: Residualized prices of commercial properties832 Panel A: Overall sample833Variables834Offices835Offices836Stores837Stores838for rent839for sale840for rent841for sale842Number of Local Lodging843-0.077844-0.0028450.052*8460.102***847(0.061)848(0.041)849(0.030)850(0.038)851Quarter FE852✓853✓854✓855✓856Civil parish FE857✓858✓859✓860✓861Civil parish time trends862✓863✓864✓865✓866Number of clusters867122868125869147870158871Observations8725,1598735,4128748,61987520,000876R28770.1128780.0428790.1718800.076881Panel B: Metropolitan area of Lisbon882Offices883Offices884Stores885Stores886Variables887for rent888for sale889for rent890for sale891Number of Local Lodging8920.0668930.062894−0.0138950.120896(0.118)897(0.114)898(0.047)899(0.095)900Quarter FE901✓902✓903✓904✓905Civil parish FE906✓907✓908✓909✓910Civil parish time trends911✓912✓913✓914✓915Number of clusters9168791788918105919116920Observations9213,2719222,7309235,32692412,173925R29260.1799270.0599280.2359290.114930Panel C: Metropolitan area of Porto931Variables932for rent933for sale934for rent935for sale936Number of Local Lodging937−0.008938−0.0169390.140***9400.101***941(0.039)942(0.014)943(0.032)944(0.033)945Quarter FE946✓947✓948✓949✓950Civil parish FE951✓952✓953✓954✓955Civil parish time trends956✓957✓958✓959✓960Number of clusters96137962399634696450965Observations9661,8889672,6829683,2939697,827970R29710.3529720.0539730.1329740.115975Results for spillover effects to commercial properties, at the quarterly level. In columns (1) and (2), we 976look at offices for rent and sale, respectively, while in columns (3) and (4) we analyze stores for rent and 977sale. Panel A uses the composite sample, containing both metropolitan areas. Panel B and C analyze 978separately the effects on the metropolitan area of Lisbon and Porto, respectively. We use as outcome 979variable the residualized asking prices, which are obtained by regressing the logarithm of prices on 980quarter FE, civil parish FE and civil parish time trends. The results displayed are of a regression of these 981residuals on the count of local lodging by quarter and civil parish, on the square footage of properties 982and time dummies. Standard errors are clustered at the postcode sector. *** 1 %, ** 5 %, * 10 %. Data 983sources: Imovirtual; RNAL (Turismo de Portugal).984985986===== PAGE 28 (score 3, numeric-paren 0, table-words 0, var-words 3) =====9879881 3989F. Nobre et al.990The lag in the reported effects is in line with other studies on housing markets. 991Gonçalves et al. (2022) show that the ban on short-term rentals (which is arguably 992a more salient policy than one that promoted a gradual growth of local lodgings), in 993Lisbon, took 3 quarters to have a negative impact on house prices. Cellini and Nobre 994(2026), Blanco and Neri (2025), Tang (2021) show that impact of urban regeneration 995phenomena on housing markets, in London, only started to be felt 3 to 5 years after 996treatment adoption.997In Panels (b) and (c) of Fig. 5, we perform the analysis for the Metropolitan areas 998of Lisbon and Porto, respectively. The figures show similar patterns to the overall 999sample. For Lisbon, the periods before treatment show a non-significant aggregate 1000ATT, with only lead 9 being statistically significant. The post-treatment period exhib1001its a significant aggregate of around 0.19, with the effects starting to be statistically 1002significant only in period 10. In Porto, we also see a non-significant aggregate ATT 1003and a post-treatment ATT of the same size as Lisbon. The effects become statistically 1004significant starting in period 7 after treatment but gain robustness and consistency 1005only from period 10 onwards.1006Overall, our results using the Callaway and Sant’Anna (2021) estimator show the 1007absence of house prices trends in the periods before the arrival of local lodgings, 1008Fig. 5 Event study: Callaway and Sant'Anna (2021). Data sources: IMT and IMI tax registries (Statis1009tics Portugal); RNAL (Turismo de Portugal)1010 101110121013===== PAGE 31 (score 18, numeric-paren 0, table-words 5, var-words 6) =====101410151 31016Short-term rentals and the housing market: evidence from Portuguese…1017Appendix1018Mean1019Std. Dev.1020Min1021Max1022(a) Asking Price1023Offices for rent10241,483.07910252,262.307102670102721,4001028Offices for sale1029229,598.91030328,404.8103175010322,580,0001033Stores for rent10341,287.83510352,598.4681036150103735,0001038Stores for sale1039178,013.21040210,20410414,20010421,500,0001043(b) Advertisements 1044(per quarter and 1045postcode)1046Offices for rent10471.94410483.028104911050531051Offices for sale10522.00710532.96105411055741056Stores for rent10571.75110582.467105911060591061Stores for sale10622.23710633.3561064110651001066Table 8 Summary statistics 1067of commercial properties 1068advertised1069This table reports the mean, 1070standard deviation, as well 1071as minimum and maximum 1072values for asking1073price of the commercial 1074properties advertised on 1075the platform Imovirtual and 1076number of advertisements per1077civil parish and postcode. 1078Source: Imovirtual1079 1080Mean1081Std. Dev.1082Min1083Max1084Total1085Per quarter and 1086postcodes10870.09110880.52710890109070109116,6221092Per quarter and post1093codes (just those with 1094residential houses 1095transactions)10960.02610970.23110980109914110016,6221101Per quarter and civil 1102parish11032.498110415.452110501106457110716,6221108Per quarter and civil 1109(just those with 1110residential houses 1111transactions)11129.841111336.038111401115457111616,6221117Per quarter1118519.4381119664.74711201911212642112216,6221123Per year11242077.7511252604.526112611611277193112816,6221129Number of bedrooms 1130per local lodging11312.17511322.41811330113459113516,6221136Table 7 Summary statistics: 1137Number of new local lodging 1138establishments1139This table reports the number 1140of observations, mean, standard 1141deviation, as well as minimum 1142and maximum values for new 1143local lodging establishments 1144in our sample. We present 1145statistics based on various time 1146dimensions, quarter and year 1147and geographical determinants 1148such as the civil parish and 1149postcode. Source: RNAL 1150(Turismo de Portugal)1151 115211531154===== PAGE 32 (score 62, numeric-paren 20, table-words 14, var-words 11) =====115511561 31157F. Nobre et al.1158Table 10 Heterogeneous effects: by characteristics of the civil parish1159Variables1160Unemployment rate1161Density1162Coastal vs non-coastal1163Below median1164Log(prices)1165Above median1166Log(prices)1167Below median1168Log(prices)1169Above median1170Log(prices)1171Non-coastal1172Log(prices)1173Coastal1174Log(prices)1175Number of Local 1176Lodging11770.0344**11780.030911790.0721*11800.018511810.010611820.0357*1183(0.0155)1184(0.0203)1185(0.0423)1186(0.0114)1187(0.0123)1188(0.0190)1189Property attributes1190✓1191✓1192✓1193✓1194✓1195✓1196Quarter FE1197✓1198✓1199✓1200✓1201✓1202✓1203Civil parish FE1204✓1205✓1206✓1207✓1208✓1209✓1210Civil parish time trends1211✓1212✓1213✓1214✓1215✓1216✓1217Number of clusters1218111121992122010712218912221081223531224Observations122593,116122692,537122794,201122891,4521229114,199123054,3181231R212320.35912330.27112340.32812350.36012360.33612370.3881238Results for heterogeneous exercises, by characteristics of the civil parish, at the quarterly level. In column (1)-(2) we consider 1239civil parishes that are below and above the median in terms of unemployment rate, respectively, while in column (3)-(4) we 1240look at civil parishes that are below and above the median in terms of population density. In column (5) we look at non-coastal 1241civil parishes, while analyzing coastal civil parishes in column (6). In all specifications, the outcome variable is the log value 1242of transactions in the present period and include a constant, property-specific controls, quarter and civil parish fixed effects 1243and civil parish time trends. We exclude properties that are within 35 meters from a Local Lodging and have the same number 1244of bedrooms. Standard errors are clustered by postcode sector. *** 1 %, ** 5 %, * 10 %. Source: IMT and IMI tax registries 1245(Statistics Portugal); RNAL (Turismo de Portugal).1246Overall sample1247Metropolitan area of Lisbon1248Metropolitan area of Porto1249Variables1250Number of Transactions1251Number of Transactions1252Number of Transactions1253Number of Local Lodging12540.672***12550.620***12560.765***1257(0.149)1258(0.205)1259(0.192)1260Civil parish FE1261✓1262✓1263✓1264Quarter FE1265✓1266✓1267✓1268Civil parish-time trends1269✓1270✓1271✓1272Number of clusters127316512741181275471276Observations127778,662127853,532127925,1301280R212810.02512820.02212830.0341284Transactions1285New built1286Transactions1287New built1288Transactions1289New built1290Number of Local Lodging1291−0.001931292−0.0102*12930.0124*1294(0.00463)1295(0.00605)1296(0.00632)1297Civil parish FE1298✓1299✓1300✓1301Quarter FE1302✓1303✓1304✓1305Civil parish-time trends1306✓1307✓1308✓1309Number of clusters131016513111181312471313Observations131478,662131553,532131625,1301317R213180.10413190.12713200.0591321 Effects on the number of transactions and transaction of new buildings, at the quarterly and postcode level. 1322Column (1) presents the results for the entire sample, column (2) for the metropolitan area of Lisbon and column 1323(3) for the metropolitan area of Porto. In the top panel the outcome variable is the number of transactions in the 1324present period, while in the bottom panel the outcome variable is an indicator variable that takes the value of 1 1325if in a given postcode quarter there was a transaction of a new building. All specification include quarter, civil 1326parish and civil parish time trends. Standard errors are clustered at postcode sector. *** 1%, ** 5%, * 10%. 1327Source: IMT and IMI tax registries (Statistics Portugal); RNAL (Turismo de Portugal)1328Table 9 Outcome variable: number of total transactions and of new built properties1329 133013311332===== PAGE 33 (score 2, numeric-paren 0, table-words 0, var-words 2) =====133313341 31335Short-term rentals and the housing market: evidence from Portuguese…1336Supplementary Information The online version contains supplementary material available at https://doi.1337org/10.1007/s10258-026-00295-2.1338Acknowledgments We would like to thank Yanos Zylberberg, Arnaud Philippe, Martin Foureaux Koppen1339steiner, João Santos Silva, Stefano Cellini, Matthias Parey, João Pereira dos Santos, Sebastian Ellingsen, 1340Rodrigo Martinez-Mazza for their useful comments. We would also like to thank the participants of the 1341Surrey internal PhD workshop, 11th PhD-student WIPE, EEA-ESEM 2023 congress and 2nd Diversity 1342and Human Capital Workshop for their helpful feedback. We are most grateful to the two anonymous 1343reviewers for their insightful comments and suggestions. We are also appreciative to the Portuguese Sta1344tistical Office, Statistics Portugal, for help with accessing the data through their Safe Centre, especially to 1345Idílio Freire, Álvaro Silva, Marcelo Pereira, Bartholomeus Schoenmakers and Alexandra Pessoa and to 1346Diogo Lopes, from OLX group, for providing access to Imovirtual data. 1347Author contributions Francisco Nobre: Conceptualization, Data curation, Formal analysis, Methodology, 1348Software, Writing – original draft, Writing – review & editing.1349Ronize Cruz: Conceptualization, Data curation, Formal analysis, Methodology, Software, Writing – origi1350nal draft, Writing – review & editing.1351Diogo Jardim Gonçalves: Conceptualization, Data curation, Formal analysis, Methodology, Software, 1352Writing – original draft, Writing – review & editing.1353Funding Diogo Jardim Gonçalves acknowledges funding from the Fundação para a Ciência e a Tecnolo1354gia (FCT) under the grant 2021.07555.BD.1355Declarations1356Ethical approval The research complies with standard ethical guidelines for social science research.1357Conflict of interest The authors declare that they have no known competing financial interests or personal 1358relationships that could have appeared to influence the work reported in this paper.1359Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, 1360which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long 1361as you give appropriate credit to the original author(s) and the source, provide a link to the Creative 1362Commons licence, and indicate if changes were made. The images or other third party material in this 1363article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line 1364to the material. If material is not included in the article’s Creative Commons licence and your intended use 1365is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission 1366directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licen1367ses/by/4.0/.1368References1369Acemoglu D, Koster HR, Ozgen C (2023) Robots and workers: Evidence from the Netherlands. Working 1370Paper 31009, National Bureau of Economic Research1371Alexander D, Karger E (2023) Do stay-at-home orders cause people to stay at home? Effects of stay-at-1372home orders on consumer behavior. Rev Econ Stat 105(4):1017–10271373Ang D (2021) The effects of police violence on inner-city students. Q J Econ 09(1):115–1681361374Barata Salgueiro T, Mendes L, Guimarães P (2017) Tourism and Urban Changes: Lessons from Lisbon. 1375Tourism in the City: Towards an Integrative Agenda on Urban Tourism. Springer, pp 255–2751376Barron K, Edward Kung, and, Proserpio D (2021) The effect of home-sharing on house prices and rents: 1377Evidence from Airbnb. Mark Sci 40(1):23–471378