Investigating the spillover effects of neighbouring housing rents on housing prices: evidence from Beijing, China

Housing markets are typically described as two interlinked segments – owner-occupied and rental – operating through distinct institutional settings, consumer preferences and regulations yet connected via asset–space market fundamentals (DiPasquale and Wheaton, 1992). In rapidly urbanising contexts such as China, these connections are amplified because housing serves as both consumption and an investment, so shifts in financial conditions, supply and neighbourhood amenities can propagate across tenure sectors (Molloy et al., 2022; Wu et al., 2012). The user-cost framework highlights that when the cost of owning changes – via interest rates, inflation, taxes or expected capital gains – equilibrium requires prices to adjust relative to rents for comparable housing services (Poterba, 1984; Davis et al., 2008). Empirically, high and rising price-to-rent ratios observed in major Chinese cities imply very low user costs and strong expectations of capital gains, underscoring the tight coupling between the two segments (Wu et al., 2012).

The user cost model posits that, in equilibrium, owning and renting an equivalent dwelling should yield the same present value of housing services, so persistent departures of price-to-rent ratios from user cost suggest mispricing or changing expectations (Poterba, 1984; Davis et al., 2008). Asset-pricing views of housing formalise this link by valuing dwellings as the discounted stream of expected future rents, tying ownership prices directly to rental market fundamentals (DiPasquale and Wheaton, 1992; Hill and Syed, 2016). In practice, quality differences between sold and rented units are very important. Hedonic methods show that quality-adjusting price–rent ratios can significantly change conclusions about equilibrium (Hill and Syed, 2016; Chen et al., 2022). A substantial body of work tests whether prices and rents co-integrate in the long run. Evidence for cointegration supports market integration, while persistent gaps are consistent with speculative or regime-switching dynamics (Gallin, 2008; Davis et al., 2008). Episodes of fast price growth with relatively sluggish rents have been documented, often coinciding with low user cost and strong capital-gain expectations (Hill and Syed, 2016; Wu et al., 2012).

Recent studies have changed from city-wide averages to spatial dynamics. Across China’s first-tier cities, price and rent spillovers transmit across space, indicating that local shocks in one market can affect others (Chen et al., 2021; Yang et al., 2018). Within cities, time-varying lead–lag relations between prices and rents have been found, implying that the rent–price linkage is not stable over time (Chen and Chiang, 2021). Neighbourhood-level policies that expand or reconfigure the rental stock can also reshape local consumption and demographic composition – potential channels through which rental conditions feed into owner-occupied prices (Ko et al., 2025). Meanwhile, fine-grained rental-index work shows that segmentation by building type, location and market cluster is meaningful to understand rent formation and its transmission (Song et al., 2022). Together, these findings point to spatially heterogeneous and evolving rent–price connections that standard city-averaged metrics cannot properly reveal.

On the demand side, changes in mortgage costs alter user cost and shift relative demand between owning and renting; local evidence shows that rents and mortgage rates jointly predict neighbourhood house price movements (Shi, 2011). On the supply side, constraints in rental supply or frictions in rental transactions can raise effective rental costs, changing the rent-to-price elasticity and potentially capitalising into owner-occupied prices (Molloy et al., 2022; Yang et al., 2023). In China, price-to-rent ratios have reached levels that require largely expected appreciation to reconcile with user cost, highlighting the importance of short-term rental conditions during transitions (Wu et al., 2012; Chen et al., 2022). In addition, city-to-city spillovers in prices and rents (Chen et al., 2021), time-varying rent–price causality (Chen and Chiang, 2021) and high price-to-rent ratios tied to expectations (Wu et al., 2012; Chen et al., 2022) all suggest that short-term rental conditions can be associated with local price formation. Complementing this, work by scholars frequently cited in this field documents how affordability, tenure choice and market segmentation shape behaviour and price discovery in Chinese housing markets (e.g. Chen, Hu, and Lin, 2019; Dong and Hui, 2022; Song et al., 2022).

Most prior research treats the rent–price nexus as a long-term equilibrium condition or studies city-wide averages. Far fewer studies consider short-term, micro-spatial links whereby neighbouring rents act as a locational attribute capitalised into prices at the parcel or building level – particularly in transitional systems where rental supply is thin, tenure is segmented and speculative demand is uneven. We address this gap and aim to estimate spatial spillovers of neighbouring rents on sale prices using pooled cross-sectional transactions from Beijing 2016–2018.

This paper makes three concise contributions. First, unlike the user-cost model that focuses on long-term equilibrium between prices and rents, we treat neighbouring rent as a short-term locational attribute in a hedonic model and estimate its spillover to sale prices. Second, we show that the spillover is lagged and heterogeneous across submarkets defined by physical and locational features (e.g. unit size and age, school-district status, administrative district). Third, we identify a mechanism: changes in rental housing supply shift the rent-to-price elasticity, so tighter rental supply raises the sale-price premium via this channel, and vice versa. This finding would also contribute to the current governance and design of housing-market policies.

The rest of the paper is structured as follows. Section 2 outlines the data and method, and Section 3 presents the main results and robustness cheques and investigates heterogeneity across submarkets. Finally, Section 4 discusses the policy implications, and Section 5 concludes the paper.

The housing market in China has been growing rapidly since the end of the welfare housing system in 1998 ^[1], when it was characterised by the start of housing marketisation and financialization (Chen and Wu, 2022). Investment in owner-occupied housing is a rooted notion among Chinese citizens, resulting in soaring housing prices in the following two decades. For individual households, purchasing housing for resale and renting it out beyond daily residence is regarded as a guaranteed investment to accumulate household wealth because housing investment returns would be promising and profitable in the short term. Such investment in additional housing is significantly prevalent in first-tier cities – Beijing, Shanghai, Shenzhen and Guangzhou. Due to speculative expectations in home value and rent premium, home buyers carefully consider all possible factors influencing housing market dynamics before purchasing. For example, they need to evaluate the neighbouring housing sales, rent variation and the built environment, such as distances to the public transits, shopping centres, the Central Business District, and if the housing is located in the quality school district (Feng and Lu, 2013; Zhang and Chen, 2018).

Unlike other countries, China’s household registration system, “hukou”, is unique. This system was derived from the beginning of China’s urbanisation strategy in 1958, segregating people from urban and rural areas. Also, the hukou identity determines where individuals can have the social benefits and influences local homeownership (Wu and Logan, 2016; Chan, 2019; Wu and Wallace, 2021; Song and Wu, 2022). Compared with local hukou residents, non-local hukou residents in first-tier cities such as Beijing and Shanghai must always pay social insurance for a couple of years to purchase owned housing. For residents without local homeownership, rental housing is vital in providing residential space. However, the development of the rental housing sector in China is much lagging behind the owner-occupied sector (Chen and Chen, 2018; Kuang et al., 2018; Song et al., 2022). For immigrants in these first-tier cities, rental housing is mainly provided by private owners rather than other institutions in the housing market. Since 2016, central and local governments have launched many regulatory documents to promote synergetic development between the rental and owner-occupied sectors and have encouraged multiple suppliers of rental housing.

Nevertheless, residential inequality between homeowners and renters is still a challenging issue, such as equal access to quality schools in the neighbourhood. Many scholars have regarded it as “tenant discrimination” (Tao et al., 2014; Mao and Chen, 2021; Song and Wu, 2022; Song, 2022). Compared with home buyers, renters are given relatively lower priority to access neighbourhood primary schools, particularly rental households without local hukou and homeownership.

This study uses two pooled cross-sectional data sets obtained from Lianjia (Link to the website of bj.lianjia.), a major real estate agency in China. The owner-occupied housing data set initially includes 175,822 transactions recorded between 2015 and 2018, while the rental data set contains 75,129 lease transactions from 2016 to 2018. Filtering out the outliers and missing variables, we finally retain 22,121 observations of owner-occupied housing transactions varying from Feb. 2016 to Apr. 2018, with characteristics such as floor area, construction year and number of bedrooms, living rooms and bathrooms. To address the research question of the spillover effects of neighbouring rents on housing prices, all these 22,121 transactions are matched with their average neighbouring rents transacted within the same community.

The coordinates of each community or neighbourhood (longitude and latitude) were matched to capture locational attributes using Baidu Map (https://map.baidu.com[Link to the map.baidu.]). Based on these coordinates, Euclidean distances to the city centre, the nearest subway station and the closest business centre were calculated to represent accessibility. Given the substantial price premium associated with primary school districts in Chinese housing markets (Song, 2022), we further identified whether each housing unit was located within a designated school catchment area. Following established methods, a threshold-based criterion was applied to define school district boundaries (Zheng et al., 2016; Song, 2022).

Table 1 shows the descriptive statistics. Housing prices range from 31 to 123 thousand Chinese Yuan (RMB) per square meter, with a mean price of 66 thousand RMB (approximately US$9.9 thousand). The average monthly rent in the neighbourhood is 5,924 RMB (1 RMB ≈ US$0.15), varying from 2,800 to 14,000 RMB (US$420–US$2,100). On average, owner-occupied housing units have a floor area of 80 square metres and were built approximately 20 years before the time of the transaction. A typical unit includes two bedrooms, one living room and one bathroom. The average distance to the city centre (measured from Tiananmen Square, denoted as D_Center) is 13.30 km. The mean distances to the nearest subway station (D_Subway) and business centre (D_Business) are 0.93 and 2.64 km, respectively. Approximately 70.00% of the owner-occupied units are located outside designated school districts. Figure 1 displays our data covers the spatial administrative distribution on the map, including nine administrative districts.

We also conducted a Pearson correlation analysis to examine the linear relationships among the key variables. The results reveal that most correlations are statistically significant (see  Appendix). Notably, the unit housing price is significantly and negatively correlated with the distance to the city centre (D_Center), indicating that housing units closer to the central area tend to have higher prices per square metre. Similarly, significant negative correlations are observed between price and the distance to the business district (D_Business) and subway stations (D_Subway), with coefficients of −0.346 and −0.209, respectively, suggesting a price premium for better accessibility and urban convenience. Interestingly, housing price also shows a significant negative correlation with housing size, implying that, ceteris paribus, larger newly constructed homes tend to have lower unit prices. A positive correlation with access to neighbourhood schools (SCHOOL) further supports the argument that proximity to high-quality education facilities contributes to housing value appreciation.

Regarding structural features, variables such as the number of bedrooms, living rooms and bathrooms are positively and significantly correlated with the size of the home, confirming their collective role in representing the internal structure of the home. The positive correlation between housing age and price is somewhat counterintuitive, possibly because of Beijing’s largely monocentric urban form and the central concentration of older neighbourhoods that primarily reflects location advantages rather than a quality premium. Similar findings could be traced back from previous theories and research in Alonso (1964), Zheng and Kahn (2008) and Gan et al. (2020); consistent with this interpretation, Age is negatively correlated with D_Center, D_Business and D_Subway. The correlation analysis offers preliminary empirical support for selecting explanatory variables and highlights key associations between housing price, neighbouring rent, location and structural attributes.

We empirically test the hypothesis that rental market conditions in neighbouring areas exert a measurable spillover effect on prices in the owner-occupied housing market. This hypothesis is motivated by the user cost theory of housing (Poterba, 1984), which posits that, in equilibrium, the cost of owning a home should approximate the cost of renting an equivalent dwelling. When prices in the ownership market diverge from underlying rental values, arbitrage mechanisms – through tenure choice or investment allocation – are expected to restore balance (DiPasquale and Wheaton, 1992). Given spatial segmentation across housing submarkets, we argue that lagged rents in nearby rental areas influence price formation in the owner-occupied segment via spatial spillovers.

To test this hypothesis, we adopted an extended hedonic price model. The hedonic approach, developed initially by Rosen (1974), has been widely used to assess the contribution of structural and locational characteristics to housing prices (Kain and Quigley, 1970). The baseline specification is expressed as:

where $Xi,t$ indicates a vector of housing attributes, such as size, age and amenity. $α1$is a vector of estimated coefficients of$Xi,t$⁠, indicating the marginal effects on housing prices. $εi,t$ is the error term.

To capture the influence of neighbouring rental conditions, we include spatially lagged average rents as a locational attribute. The full empirical model is specified as follows:

where $lnPricei,t$ is the logarithm form of housing price, $i$ denotes observations in the data sets; $t$equals period, $k$ equals the number of temporal lags (e.g. one-quarter, half-year and one-year lags). The variable $Xi$ represents other independent variables. In addition, $m$ denotes the neighbourhood of estimated owner-occupied housing, e.g. housing community; $Rent¯m,t−k$ is the average rent in the neighbourhood. $WilnPricei,t$ indicates the housing spatial dependency (Anselin, 2001; Wilhelmsson, 2002). $Wi$ is the spatial weight matrix, built on the longitude and latitude coordinates of the community. In this study, we use the k-nearest neighbours approach to construct the spatial weight matrix, where k = 10.

The hypothesis aims to estimate $β2$⁠, indicating the extent to which spillover effects of neighbouring rents affect housing price volatility. $β1$ is the coefficient of spillover effects of neighbouring housing prices. $β3$ is a vector of coefficients of control variables, including observed physical and locational attributes of owner-occupied housing. $β0$ is the constant.

This section outlines the empirical strategy and presents the results. First, we construct average housing prices and community monthly rents at three horizons – one quarter, half a year and one year – to estimate equation (4). Second, we estimate the baseline specification in equation (2). To assess robustness, we use migration-based instruments and estimate IV–2SLS (instrumental variable-two-stage least squares) models; apply the Oster (2019) selection-on-observables approach to assess sensitivity to selection on unobservables; and implement randomization-based placebos as falsification tests. We then examine heterogeneity in rent spillovers across housing age, housing size, school-district status and administrative districts. Finally, we explore the mechanism by leveraging a rental-market intervention to test whether tighter rental conditions amplify the pass-through from lagged rents to sales prices.

The baseline regression uses a log-log specification, allowing all coefficients to be interpreted as elasticities. That is, they represent the percentage change in housing prices associated with a 1% change in each explanatory variable.

As shown in Table 2, the model demonstrates strong explanatory power, with R-squared values between 0.79 and 0.80, indicating that the included variables explain approximately 80% of the variation in housing prices. Variance inflation factor (VIF) values remain between 2.99 and 3.00, well below the threshold of 10, suggesting no multicollinearity issues. Individual VIFs are reported in  AppendixTable A2.

Both the spatial dependence of housing prices and rent spillovers show significant effects. The coefficient on the spatial lag of housing prices is consistently 0.11, indicating that a 1% increase in surrounding housing prices leads to a 0.11% increase in the price of a given unit. This confirms the presence of strong spatial spillover effects. We select one-quarter, half-year and one-year lag rent into the baseline model to address the temporal spillovers, which is similar to previous studies from Feng (2020), Baltagi and Li (2015) and Chen et al. (2021). Rent spillovers are also significant across all time lags. A 1% increase in neighbouring rents, lagged by one quarter, is associated with a 0.35% rise in housing prices. This effect grows to 0.38% with a half-year lag and 0.39% with a one-year lag. The increasing trend suggests that past rental conditions exert a persistent and slightly strengthening influence on housing prices, possibly reinforcing long-term price expectations in the market.

Among structural attributes, a 1% increase in the number of bedrooms significantly raises housing prices by approximately 0.14–0.15%, while additional living rooms and bathrooms are associated with increases of roughly 0.09–0.10% and 0.09%, respectively. We also observe diminishing returns to scale. A 1% increase in total housing space slightly reduces the price per unit by 0.38–0.41%, likely reflecting bulk discounts or lower marginal value of space. Similarly, a 1% increase in housing age lowers prices by 0.08% on average, reflecting depreciation, rising maintenance costs, or reduced housing quality.

Locational characteristics also play an important role. Being located within a school district increases housing prices by approximately 1.8%–1.9%, indicating that access to quality education is capitalised in the housing market. Distance-based factors show negative impacts. A 1% increase in distance to the city centre, subway stations and business districts reduces housing prices by about 0.15%, 0.02% and 0.01%, respectively. These results underline the stronger influence of city centre proximity and transit accessibility compared to proximity to business districts.

To assess robustness, we performed three exercises that addressed endogeneity, omitted-variable bias and falsification. First, we estimate IV–2SLS models that instrument community rents with district-level long-term migrant inflows. Second, we apply the Oster (2019) selection-on-observables procedure to bound the rent elasticity under plausible degrees of selection on unobservables. Third, we implement randomization-based placebo tests as falsification tests. In what follows, the one-quarter lag of community rents serves as the baseline rent measure.

To address endogeneity in the rent–price specification, we estimate 2SLS models that instrument community rents with district-level long-term migrant inflows from the Beijing Statistical Yearbook (2017–2019). Using migration as an instrument follows established housing-market practice that treats demographic inflows as exogenous demand shifters in rental markets which – conditional on time and area fixed effects – affect sale prices only through rents (Saiz, 2007; González and Ortega, 2013; Sanchis-Guarner, 2023).

Because migration data are annual, we reconcile them with monthly frequency using Denton–Cholette proportional benchmarking, which preserves high-frequency dynamics while honouring annual control totals (Denton, 1971; Dagum and Cholette, 2006). To relax functional-form restrictions – because the migration–rent linkage may be nonlinear – the instrument set includes level, log and squared terms. In the first specification, the instruments are the t − 3 migrant inflows; the second replaces these with the cumulative t − 4 to t − 6 inflows. The second stage retains the same physical and locational controls and month fixed effects as in the baseline.

Instrumental variable (IV) estimates indicate a large and precisely estimated rent–price elasticity. In column (1) the coefficient on lagged community rent equals 0.229; in column (2) it equals 0.236. Both are significant at the 1% level. Relative to the OLS elasticity reported in Table 2 (0.349***), the IV magnitudes are smaller but economically meaningful, consistent with attenuation once endogeneity is addressed and with a causal rent-to-price spillover. First-stage diagnostics confirm instrument strength: the Kleibergen–Paap LM statistics reject under-identification (3139.18; 3161.09), and the Cragg–Donald F (1300.48; 1313.60) and Kleibergen–Paap rk Wald F (1302.27; 1314.41) values far exceed the Stock–Yogo 10% maximal IV size critical value of 22.3. Model fit is also high with R-squares of 0.6454 and 0.6471. The stability of the estimates across alternative lag windows and nonlinear instrument sets reinforces this interpretation.

We use two approaches to address concerns about omitted variables. First, we estimate an instrumental variable model (IV-2SLS) in which community rents are standardized with long-term migrant inflows at the district level. In spite of the potential endogeneity of the rents, the IV estimates reported in Table 3 remain positive and precise (0.2290–0.2362), consistent with the baseline pattern in Table 2 and indicating a causal spillover of rent-to-price.

Second, following Hu et al. (2023) and Lin et al. (2024), we apply the selection-on-observables method developed by Oster (2019), which extends Altonji et al. (2005) to quantify how much unobserved selection would be required to explain the baseline estimate. We set the relative selection parameter to $ρ=1$ (equal selection on observables and unobservables). Let $Rmax$ denote the attainable $R2$ for a model including both observed and unobserved determinants, and $R˜$ is $R2$ from our fully controlled specification (the baseline model). Following Oster (2019), we consider$Rmax∈{1.1, 1.2, 1.3}×R˜$⁠. The bias-adjusted elasticities for $lnRentt−3$ decline from the OLS coefficients of 0.3490–0.2823, 0.2156 and 0.1489, respectively (Table 4). These values remain positive and economically meaningful, suggesting that omitted-variable bias is insufficient to overturn the rent–price relationship. Moreover, the Oster-bounded range nests the IV–2SLS estimates, reinforcing the interpretation that the observed spillover is not driven by unobserved confounding.

To check whether the observed rent–price pass‐through could be an artefact of confounding or selection, we implement a placebo (falsification) design based on a Freedman–Lane residual permutation (Freedman and Lane, 1983). Specifically, we first partial out covariates, district and month-fixed effects from both the outcome and the lagged neighbourhood‐rent regressor, obtaining residualised variables (price and rent). We then permute the rent residual within districts and re-estimate the model by regressing the outcome residual on the permuted residual without an intercept; repeating this 500 times yields an empirical null distribution for the rent elasticity. This block choice preserves local market composition while breaking the original price-rent alignment; given that month effects are partialled out, cross‐month permutations within a district do not reintroduce common time shocks. This procedure follows established theory on residual permutation for linear models with nuisance structure and is widely used in fixed‐effects specifications (Freedman and Lane, 1983; Zhao and Ding, 2021).

Figure 2 reports the placebo results. The permutation coefficients are tightly concentrated around zero and are well approximated by a normal density. In contrast, the baseline estimate (β = 0.349, see the vertical dashed line) lies in the extreme right tail of the empirical null and exceeds all permutation estimates in absolute value. Taken together, the falsification evidence is consistent with a true neighbourhood rent spillover rather than spurious co-movement induced by confounding or selection.

As confirmed by robustness tests, the spillover effects of neighbouring rents remain stable in explaining housing price volatility. However, given the housing market segmentation, examining how these effects vary across different (non-geographical) submarkets is important, as discussed in existing studies. Housing as a special good exhibits significant heterogeneity segmented based on geography, property characteristics and household demographics (Schnare and Struyk, 1976; Watkins, 2001; Goodman and Thibodeau, 1998; Bourassa et al., 2007; Wilhelmsson, 2004). Housing spillover effects would also vary across different submarkets (Malpezzi, 1999; Holly et al., 2011). As an important indicator, school district is always capitalised in housing price and influences housing price dynamics (Gibbons and Machin, 2003; Wang and Li, 2022).

Based on this, we conduct a heterogeneity analysis to assess the variation in rent spillover effects across different housing segments. The results are presented in Tables 5–,8. As shown in Table 2, both physical attributes (e.g. size and age) and locational characteristics (e.g. school district status and proximity to subway stations) significantly influence price volatility.

Following Zhang and Chen (2018), we partition the sample into five size bands (≤50 m², 50–70 m², 70–90 m², 90–110 m² and ≥110 m²) to examine heterogeneity. Starting with the rent spillover (⁠$lnRentt−3$⁠), all coefficients are positive and significant at the 1% level, ranging from 0.274 to 0.388. The effect peaks for mid-sized homes (70–90 m²: 0.388). It remains sizable for 50–70 m² (0.357) and 90–110 m² (0.338) but is relatively weaker at the two ends of the size distribution (≤50 m²: 0.274; ≥110 m²: 0.307). Economically, a 1% increase in neighbouring rents raises sale prices by about 0.27%–0.39%, depending on the size segment. The particularly strong response in the 70–90 m² band likely reflects its broad appeal to both owner-occupiers and renters, which tightens the rent–price linkage in this market tier.

Turning to spatial price dependence (⁠$WlnPrice$⁠), the spatial-lag coefficients are again significant at the 1% level and lie between 0.057 and 0.118. The strongest price co-movement is observed for the smallest units (≤50 m²: 0.118), followed by the largest (≥110 m²: 0.107). The weakest effect appears in the 50–70 m² group (0.057), with intermediate magnitudes for 70–90 m² (0.085) and 90–110 m² (0.089). Overall, rent spillovers are the central result – non-monotonic across size with a clear peak in mid-sized homes – while spatial price spillovers corroborate stronger cross-neighbourhood price spillover at the small-unit end, where substitutability and price competition are more intense.

Following prior studies, we use housing age as a proxy for quality (Zheng et al., 2016; Zhang and Chen, 2018). We define 2000 as the threshold year, as a sharp rise in private construction followed the end of the welfare-based housing allocation system in 1998.

Table 6 shows that rental spillovers are present in both groups. A 1% increase in neighbouring rents results in a 0.35% increase in prices of newly built homes and 0.32% for housing constructed before 2000. The slightly larger rental elasticity for newer properties may reflect greater investor activity and rental yield considerations. Newly constructed housing units are often in growth areas with higher turnover and more volatile rental dynamics. Older homes, by contrast, may serve more stable owner-occupiers, reducing the transmission of rental trends to sale prices.

Besides, spatial spillovers are significant for both segments but are more pronounced for old housing constructed before 2000. A 1% increase in neighbouring house prices raises the price of an older home by 0.11%, compared to 0.09% for newly built homes. The stronger spatial dependency in older properties may reflect their location in mature neighbourhoods with more stable and transparent pricing structures. In contrast, newer developments likely exhibit more heterogeneity in quality, design and location, reducing their sensitivity to nearby prices.

We further examine how school district boundaries influence spatial and rental spillovers. In China, access to public school resources varies by tenure status and can affect property values (Zhang and Chen, 2018; Zheng et al., 2016; Yan and Han, 2020; Song, 2022).

Table 7 shows that rental spillovers are significant in both segments. A 1% increase in nearby rents raises housing prices by 0.36% outside school districts and 0.32% within school districts. The marginally higher elasticity outside school districts implies that rental market conditions exert a more decisive influence when school access does not add value directly. One explanation is that affordability pressures are more acute in these areas, making homebuyers more responsive to rent changes when forming price expectations. Spatial dependence is weaker within school districts. A 1% increase in nearby housing prices leads to a 0.13% increase in local prices outside of school districts, compared to 0.05% within them. This suggests that the embedded value of school access dampens the influence of neighbouring prices.

Table 8 shows that the one-quarter–lagged neighbouring rent effect is positive and highly significant in all nine districts, with elasticities between 0.151 and 0.356. All specifications control for housing structural attributes, locational amenities (including school-district status and accessibility), district-fixed effects and month-fixed effects. The largest responses appear in Chaoyang district (CY), Haidian district (HD) and Shijingshan district (SJS). A second tier of impacts is observed in Xicheng district (XC), Fengtai district (FT) and Daxing district (DX). The weakest rent pass-through occurs in Dongcheng district (DC), Changping district (CP) and Tongzhou district (TZ). Economically, a 1% increase in neighbouring rents raises local sale prices by 0.15%–0.36% depending on the district. Stronger rent spillovers in CY/HD/SJS are consistent with dense employment, education clusters and vibrant rental markets that tighten the rent–price linkage; the muted effects in DC and TZ likely reflect older-core idiosyncrasies (heritage, regulation and limited turnover) and new-town supply dynamics, respectively.

As to spatial price dependence, WlnPrice is significant in several districts but not universal. Spatial co-movement is strongest in XC (0.155) and remains significant in CY (0.071), HD (0.057) and DX (0.10), with CP showing a marginal effect (0.126). In contrast, spatial-lag terms are statistically weak in DC, FT, TZ and SJS. This pattern suggests that while rent spillovers are robust across all locations, price spillovers are selective, emerging more clearly along corridors and peripheral axes (e.g. XC and DX) but decreasing in tightly regulated cores or segmented submarkets.

Beyond investigating heterogeneity, we examine the mechanism through which rents transmit their impacts to prices. We focus on a supply-tightening channel and treat Beijing’s November 2017 enforcement against informal or illegal leasing as an exogenous contraction of rental supply. One possible consequence of the November 2017 enforcement is reducing effective rental housing, tightening the rental market and placing upward pressure on rents. Because the baseline model shows a significant spillover at the one-quarter horizon, we take $lnRentt−3$ as a reference measure and introduce a dummy variable$Shock$⁠. Because the enforcement occurred in November 2017, sales in December 2017 and January 2018 therefore use rents that occurred in September and October 2017, which is pre-shock in the rental market. The first month which $lnRentt−3$ reflects the post-shock environment is February 2018. Therefore, we set $Shock=1$ for owner-occupied housing transactions from Feb. 2018; 0 otherwise. The interaction term $lnRentt−3 × Shock$ could capture the rent spillover effect under post-shock rental conditions.

As shown in Table 9, the pre-shock spillover elasticity is 0.35 with a t-statistic of 55.16, closely matching the baseline. The interaction between lagged rent and the post-shock indicator is 0.03, implying a post-shock spillover elasticity of about 0.38. In economic terms, a 10% increase in neighbourhood rent maps into roughly 3.5% higher sale prices prior to the policy and 3.8% afterward – an amplification of about 0.3 percentage points (approximately nine percent relative to the baseline elasticity). The coefficient on the spatial lag of price remains stable at 0.11 with a t-statistic of 12.47, and overall fit is unchanged with an $R$-square near 0.794 for 20,213 observations. In economic terms, a 10% increase in neighbourhood rent can increase roughly 3.5% higher sale prices prior to the policy and 3.8% afterward – an amplification of about 0.3 percentage points (approximately 9% relative to the baseline elasticity). The coefficient on the spatial lag of price remains stable at 0.11 with a t-statistic of 12.47, and overall fit is unchanged with an R-square near 0.794 for 20,213 observations.

This research mainly examines the short-term spillover effects of neighbourhood rents on housing prices. The spillover is heterogeneous across submarkets. For example, it is stronger for mid-sized units and somewhat larger in newer stock, but relatively weaker within school-district neighbourhoods, where access to education is already capitalised in prices. Statistically, the spillover effect of the rental is greater in the Chaoyang, Haidian and Shijingshan districts, where there are more employment sources and educational sources, and smaller in the districts where the neighbourhoods are characterised by cultural places (e.g. Dongcheng district) and newly built dwellings (e.g. Changping and Tongzhou districts).

Our findings are in line with previous evidence that rental dynamics shape local price levels and that rent changes tend to precede price changes in dense urban spatial patterns (Chen et al., 2021; Ji and Bhandari, 2021). Furthermore, the findings are in accordance with the documented capitalisation of school-district access in Beijing, which helps explain the weaker incremental role of rents in these neighbourhoods (Zheng, Hu and Wang, 2016). Our mechanism analysis also shows that the spillover increases after the November 2017 enforcement against informal and illegal leasing, with the spillover elasticity increasing from roughly 0.35 to approximately 0.38. That is to say, under constrained rental supply, rent variation could result in higher sales prices, leading to a larger spillover effect rents to transaction prices.

Against these findings, empirical implications could be drawn as follows. First, considering the effects of the spillover of rental income on sales prices, expansion of suppliers and a multifaceted supply of rental housing are needed in the housing market to mitigate soaring housing prices. Priorities include expanding affordable rental housing and encouraging institutional suppliers to provide affordable and decent rental housing, which can not only mitigate the variation in rental housing but also the housing price premium in the short run. In addition, the governance of the housing market should be spatially differentiated. In school-district areas, where large amenities premia are already embedded, rent-side measures will have a smaller effect on prices than in districts with high rental turnover. More broadly, in terms of rental spillover effects, converting for-sale stock into professionally managed rentals helps reduce inventories and deepen rental supply, reinforcing an owner-occupied market orientated to residence rather than speculation.

Using a pooled cross-sectional spatial hedonic price–rent model, we examine how the spillover and spillover effects of neighbouring rents influence housing prices while controlling for spatial dependence in prices. The results show a significant causal relationship between unit prices and neighbouring rents: a 1% increase in one-quarter-lagged neighbouring rents is associated with a 0.35% rise in unit prices on average. A more conservative estimate using a migration-based IV and 2SLS is about 0.23%. We also analyse heterogeneity across submarkets segmented by physical and locational characteristics – housing size, housing age, school-district status and administrative districts. Finally, we investigate the spillover mechanism from neighbouring rents to prices. Using the November 2017 rental-market shock as a case, we find that the decrease in rental housing supply raised the rent premium and thereby amplified the spillover by about 0.3 percentage points in the post-shock period (see Table 9).

Although these findings inform policy, the study has several limitations that warrant further research. First, the analysis relies on pooled cross-sectional data over a short horizon, which limits our ability to identify long-term causal relationships between housing prices and neighbouring rents. Second, although we control for housing quality, district fixed effects and accessibility, other factors, such as local economic conditions, interest rates and credit availability, may also affect the rent–price relationship and deserve further analysis. Finally, rental-market depth and housing-supply dynamics differ between cities and countries, so comparative studies would help assess how these spillover effects operate in other contexts.

The authors wish to acknowledge the funding provided by the China Scholarship Council (CSC201700260251) and the National Natural Science Foundation of China (No. 72011530136). The authors would also like to thank the editor and reviewers for their helpful suggestions on improving the quality of this paper.