The impact of new metro stations on rental prices in Moscow

Control group (apartments from this group do have a metro station nearby at the moment of publishing an offer) consists of 374 observations. The mean value for the rental prices is 34400.80 rubles, the maximum is 310000 rubles, the minimum is 7500 rubles, while standard deviation is estimated around 23492.98 rubles.

All functional forms were tested including linear, semi-logarithmic and logarithmic forms in order to exclude the possibility of model misspecification caused by inappropriate function. As well as for the treatment group, dummy variables for the first floor and the last floor were always insignificant while models' coefficient of determination remained the same and joint significance of the set of exploited variables rose meaning that these variables can be skipped. From that point of view, assumptions that the first floors are undesirable while the last floors seem plausible to consumers are not confirmed. It can be explained by the fact that bars can be put on the first floor to prevent burglary, whereas the last floor has no advantage over the penultimate floor, for example. As for the squared distance, this variable makes sense only for linear and semi-logarithmic forms (both in terms of interpretation and in terms of mathematical accuracy) but for both forms it was found to be insignificant.

Control group's best functional form was found to be the same as for the treatment group:

log(rental prices) =

= a 4 Я » log (distance to the closest metro station) 4

+ io * dummy for parks 4 <p * dummy for lake or river 4 i/> * floor 4

40 ¦ number of storeys 4 у * log (appartment area)

Table 4 demonstrates the resulting coefficients, their standard errors, t-statistics and p-values. T- tests state as a null hypothesis the insignificance of a coefficient. By these tests we reject the null hypotheses on the 5% significance level for the constant, for the logarithm of distance to the closest metro station, for the floor number and for the logarithm of a flat area as their p-values (0, 0.0018, 0.0001 and 0) are lower than 0.05. For a control group, one percentage increase in distance leaded to 0.06 percentage decrease in rental price while 1% increase in flat area leaded to 0.586% increase in rent. Also, one floor upper added approximately 2.2% to rental price. Hypothesis 2 about verification of the sign of change caused by proximity to metro is confirmed by this model meaning that the relationship is actually inverse. In the same way as for the treatment group, for all functional forms F-tests have always shown joint significance of variables. Nevertheless, coefficients were also significant for other functional forms and coefficient of determination is not relatively high for cross-section, and it seems that problem of multicollinearity is absent. Even if this conclusion is wrong, there can be used a fact that estimated equation had a good explanatory power and estimates are unbiased.

Table 4

Control group

When testing possible heteroscedasticity, again, Breush-Pagan-Godfrey and White tests are applied. Both tests' null hypothesis states homoscedasticity. Also, both tests assume computation of residuals and running auxiliary regression, then the coefficients of determination are taken from this regression to calculate F-statistic which is asymptotically distributed with chi-square distribution with the number of degrees of freedom equal to parameter restrictions under the null hypothesis of homoscedasticity. Breush-Pagan-Godfrey F-statistic is 0.0436, White test demonstrates F-statistic equal to 0.0009. At 5% significance level, both tests confirm the presence of heteroscedasticity. In order to get reliable results, heteroscedasticity-consistent standard errors can be used, combined with GLS method that is more efficient than OLS under heteroscedasticity (see Table 5). Heteroscedasticity consistent standard errors and GLS solve the problem of violation of Gauss-Markov conditions and make standard errors homoscedastic, also taking into account certain degree of correlation between residuals. Changes from removing the problem are not large in terms of significant coefficients and their interpretation - on 5% significance level the constant, the logarithm of distance to the closest metro station, the floor number and the logarithm of a flat area are significant as their p-values (0, 0.023, 0.0003 and 0) are lower than 0.05. One percentage increase in distance on average leaded to 0.039 percentage decrease in rental price while 1% increase in flat area leaded to 0.629% increase in rent. Also, one floor upper on average added approximately 1.8% to rental price. Hypothesis 2 about verification of the sign of change caused by proximity to metro is still confirmed. Rn-squared statistic is a robust version of a Wald test of the hypothesis that states that all coefficients are equal to zero. As p-value for it is 0, it confirms statistical significance of this set of variables. Now the whole sample of treatment and control group together can be examined.

Table 5

Control group by GLS

The random sample from different parts of Moscow consists of 500 observations. The mean value for the rental prices is 35600.80 rubles, the maximum is 310000 rubles, the minimum is 7500 rubles, while standard deviation is estimated around 25205.50 rubles.

The procedure was repeated with accordance to previous results.

Best functional form for the sample:

logfrental prices) =

= a + Я * log (distance to the closest metro station) +

+ a) * dummy for parks + cp * dummy for lake or river + ip * floor +

+6 * number of storeys + у * log (appartment area)

* dummy for existing metro station

As functional forms for treatment and control groups coincided, it was not surprising that once again, logarithmic relationship of rents and distance with areas will be the best in terms of significance of every variable separately and of the whole set and in terms of best explanation of dependent variable. Dummy variables for the first floor and the last floor were again insignificant while models' coefficient of determination remained the same and joint significance of the set of exploited variables rose meaning so these variables were dropped. In the same manner, squared distance was dropped losing both in terms of interpretation and in terms of mathematical accuracy. Nevertheless, this model is different due to the inclusion of Dummy_for_metro_station, that helps to distinguish between treatment and control groups.

floor number and the logarithm of a flat area are significant as their p-values (0, 0.0001, 0.0125 and 0) are lower than 0.05 meaning that we reject the H0 that stated zero coefficients.

Table 6

Sample

Hypothesis 1 that there is a difference between the rental offers for apartments that already have an access to a metro station and those who will get it in 2018/2019 is not confirmed by this model but it means that future benefits can already be incorporated into prices. Hypothesis 2 about an inverse relationship of distance and rents is confirmed in the same way as for both treatment and control groups. These observations concerning significance and Hypotheses 1 and 2 are not the final ones as it is highly probable that heteroscedasticity problem from control group will be caught here, thus, t-test and F-test can be invalid and t-statistics can be overestimated giving rise to the misleading impression of the precision of regression coefficients.

In order to verify it, Breush-Pagan-Godfrey and White tests are applied. Both tests' null hypothesis states homoscedasticity. After computation of residuals and running auxiliary regression, F-statistic is calculated which is asymptotically distributed with chi-square distribution with the number of degrees of freedom equal to parameter restrictions under the null hypothesis of homoscedasticity. Breush-Pagan-Godfrey F-statistic is 0.0017, White test demonstrates F-statistic equal to 0.0004. At 5% significance level, both tests confirm the presence of heteroscedasticity. In order to get reliable results, heteroscedasticity-consistent standard errors are applied combined with GLS method (see Table 7).

Table 7

Sample by GLS

Changes in regression from removing the problem of heteroscedasticity matter (in terms of significant coefficients and their interpretation) as variable for the floor number became insignificant on 5% significance level while the variable controlling for capitalization of future benefits into rents became significant. This result “flips” the situation completely but it can be considered as more convincing. The constant, the logarithm of distance to the closest metro station, the logarithm of a flat area and dummy variable for existing metro station are significant as their p-values (0, 0.024, 0 and 0.0189) are lower than 0.05. One percentage increase in distance on average leaded to 0.042 percentage decrease in rental price while 1% increase in flat area leaded to 0.635% increase in rent. The most important result is that having already functioning metro station nearby increases rental prices on average by 8.1%. Hypothesis 1 that there is a difference between the rental offers for apartments that already have an access to a metro station and those who will get it in 2018/2019 is confirmed by this model. Even with awareness of renters about future benefits and exploitation of this fact in offers, rental prices differ in accordance with the metro stage of working/still in construction. Thus, an appearance of metro can increase rent at least by 8%. Taking into account that the rise in prices is gradual, overall effect of new metro station can be even higher as the market have already included new opportunities, at least partially, that is seen from offers. An inverse relationship of distance and rents (Hypothesis 2) is proved but expected quadratic form for distance, that could capture negative effects like pollution or noise caused by new metro station, is not confirmed.

It should be mentioned that surprisingly, in all the regressions the presence of natural amenities was never significant despite the fact that usually households take them into account when choosing a flat. It raises a discussion on the interesting phenomenon that could be seen looking at the data. Good and furnished apartments with park and lake nearby are not priced higher than apartments that are much worse and have no access to natural amenities. It is possible that in such a way we observe irrational pricing as rents are not always provided by specialists or agencies but by real people who seem to overestimate or underestimate the benefits provided by a flat.

These results have a large list of limitations starting from the data itself and finishing with doubtful hedonic price models. First of all, there might be dynamic adjustments that the crosssection data fail to pick up. Secondly, hedonic models fail to account for spatial autocorrelation and omitted variable. This last problem may alter some conclusions. Number of housing attributes is much larger and it was easy to miss an important variable that could lead to model misspecification and biased estimated coefficients with standard errors, t tests, and F test that are invalid. For example, this variable could be a type of house: brick, panel and monolithic houses can have a direct effect on price, but this information is unavailable for given dataset. Endogeneity could also be the case as there may be omitted variables that are correlated with at least one of the included explanatory variables (for example, if type of house is correlated with number of storeys). However, it is not possible to test it as a valid instrument was not obtainable. The problem with data applies also to the effects of primary and high schools, that parents take into account when they make a decision where to live. In fact, the presence of school nearby was constant and could not explain the fact that parents choose school according to its rating as well as to leading subjects. For that reason, even with the access to such technology as ArcGIS platform, there can be doubts about the inclusion of all the important variables.

To sum up, hypotheses about a difference between the rental offers for apartments that already have an access to a metro station and those who will get it in 2018/2019 (Hypothesis 1) and an inverse relationship of distance to the closest metro station and rental prices (Hypothesis 2) were confirmed. Having already functioning metro station nearby increases rental prices on average by 8.1% in comparison with those places where new stations were expected over the next two years. Hedonic price regressions of linear, semi-logarithmic and logarithmic forms were used, basing on finding values that are formed by people's willingness to pay for a commodity when different characteristics change. Cross-section of ten thousand strings of rental offers all over Moscow was exploited to conduct a research. There were randomly selected five hundred observations that were divided into two groups (treatment and control group) that were examined separately and together. The preferred estimates, based on GLS method with robust squared errors uncovered the relationship of rental prices depending on distances to the closest metro station, areas in square meters, numbers of storeys, floor numbers and the presence of a park or forest of any size near the house, as well as the presence of a river or lake of any origin, where distance and areas played a key role. Surprisingly, it was found that natural amenities do not bring much value to rental prices. Negative effects like pollution or noise caused by new metro station failed to be captured. Overall effect of new metro station can be higher than eight percent as many renters were aware of future benefits and exploited this information in offers, thus, prices could rise after the announcements about new metro.

Conclusion

The aim of this paper was to primarily measure the impact of a construction of new metro stations on rents as it will exclude opportunities of developers, and focus on capturing the trends created by market presented by renters who will react to the news provided by mayor. In other words, the main idea was to measure the effect of transportation infrastructure and land-use characteristics and to investigate the impact of the announcement of a future metro stations on the area. Research was based on rental offers published in 2018 that account for announcements on 2018 and 2019 openings of stations. It investigated modern methods including difference-indifference models, multilevel models, spatial quantile hedonic price models, traditional hedonic price models, repeat sales models and spatial econometric models such as spatial error model etc. The worldwide results confirmed the capitalization of new benefits and proved the existence of positive externality. The most interesting problem for our research was verifying the causes for Thessaloniki and its negative result which was, as we see, in indirect costs that can alter prices in any directions. At the time for which the research was conducted, property prices had an increasing trend and wise versa, rental market demonstrated diminishing prices that were predicted to fall further. At the same time renters were moving to apartments of lower class and with smaller number of rooms. The predictions for the influence on rents stated inverse relationship of distance on prices that was confirmed empirically. Of course, cross-sectional data has more limitations as it does not allow to consider the changes before/after metro development. Hypothesis about a difference between the rental offers for apartments that already have an access to a metro station and those who will get it in 2018/2019 was also confirmed. Having already functioning metro station nearby increases rental prices on average by 8.1% in comparison with those places where new stations were expected over the next two years.

Hedonic price regressions of linear, semi-logarithmic and logarithmic forms were used, basing on finding values that are formed by people's willingness to pay for a commodity when different characteristics change. Surprisingly, it was found that natural amenities do not bring much value to rental prices. Negative effects like pollution or noise caused by new metro station failed to be captured. Overall effect of new metro station can be higher than eight percent as many renters were aware of future benefits and exploited this information in offers, thus, prices could rise after the announcements about new metro. Most important limitations were the absence of panel data and the probable misspecification of the model due to the physical impossibility to include more variables.

List of References

1. “Real estate development in anticipation of the Green Line light rail transit in St.Paul” Xinyu(Jason) Cao, Dean Porter-Nelson 2016

2. “When and how much does new transport infrastructure add to property values? Evidence from the bus rapid transit system in Sydney, Australia” Corinne Mulley, Chi-Hong(Patrick)Tsai 2016

3. “Transit premium and rent segmentation: A spatial quantile hedonic analysis of Shanghai Metro” Yiming Wang, Suwei Feng, Zhongwei Deng, Shuangyu Cheng 2016

4. “The effect of new metro stations on local land use and housing prices: The case of Wuhan, China” Ronghui Tan, Qingsong He, Kehao Zhou, Peng Xie 2019

5. “The impact on neighbourhood residential property valuations of a newly proposed public transport project: The Sydney Northwest Metro case study” Yuer Chen, Maziar Yazdani, Mohammad Mojtahedi, Sidney Newton 2019

6. “The Anticipated Capitalisation Effect of a New Metro Line on Housing Prices” Claudio A. Agostini and Gaston A. Palmucci 2008

7. “Measuring the effects of transportation infrastructure location on real estate prices and rents: investigating the current impact of a planned metro line” Dimitrios Efthymiou, Constantinos Antoniou 2013

8. “The impact of metro services on housing prices: a case study from Beijing” Shengxiao Li, Luoye Chen, Pengjun Zhao 2017

9. “The dynamic effects of subway network expansion on housing rental prices using a repeat sales model” Chang-Moo Lee, Kang-Min Ryu, Keechoo Choi, Jin Yoo Kim 2018

10. “Location and land use” William Alonso, 1964

11. https://www.irn.ru/IRN.RU real estate agency

12. https://baza-winner.ru/Winner informational real estate agency

13. https://www.ceicdata.com/CEIC data Informational economic base

14. https://www.azbuka.ru/Azbuka Zhylya real estate agency

15. https://stroi.mos.ru/Complex of urban planning policy and construction of the city of Moscow

Размещено на Allbest.ru