Real estate market

Abstract
Housing market is a market with the heterogeneous sellers trading durable goods. Like in any other market, the state of the housing market is determined mostly by the agents’ behavior. The literature related to housing demand is extensive, while much less attention has been paid to the investigation of the supply side. In this project we expand on the existing literature by focusing on the sellers’ determinants in setting the asking prices. It is important that there are two main types of sellers in the housing market which are private individuals and professional real estate agents. Both types of agents may list the offers on own platforms or aggregators of both types of sellers’ offers like multilisting systems (MLS). When setting the asking price sellers look for internal characteristics of property (area, number of rooms etc.), characteristics of a building (age and material of construction, location etc.), characteristics of surroundings (transport accessibility, closeness and level of infrastructural objects) as well as current market prices of close marketed objects. The traditional approach for real estate price modelling is hedonic price model. In this model price is regressed on internal and external property characteristics. The extension of this model is a model of spatial autoregression where price is also affected by prices of other marketed objects weighted by geographic distance for modeled object (see LeSage, Pace, 2009 as an example). The geographic distance is measured as Euclidian distance on coordinates, other coordinate-based metrics or time-based metrics. It is well known that sellers when setting the asking price look also on the prices of marketed objects which are close by characteristics. Yong and Sasaki (2015) show that when the offer is listed on MLS with online calculator of price based on characteristics and location seller tend to set their price based on the estimate from this calculator along with their own valuation of property. This research project is aimed to explore the new determinants of pricing behavior, such as distance measures between objects in geographic and characteristics spaces that give the best prediction of real estate prices within the spatial autoregression framework, pricing behavior of agents in time and external property characteristics influence on housing prices.

Data
In order to identify the effects on price, we need to observe housing prices in time, geographic and characteristics spaces. We use the Metrosphera.ru website to collect primary data on advertisements about selling of property in Perm, Russia. Perm is 11th largest city in Russia and has approximately 1 million residents. Metrosphera.ru is the most extensive source concerning housing market information in this city with, expertly, near 80% of all selling ads placed on it. In order to extract the data, we developed a tool for daily downloading of all advertisements available.

• The first dataset has starting date of downloading is 26 October 2014. Every day the data on near 6000 of secondary market objects is available. The whole dataset contains over 15000 unique objects observed over 98 days that generated over 500 000 daily observations. Except list prices, an offer contains information about property characteristics (address, area, a number of rooms, floor, a number of floors in the building, material and type of the building, age of construction). We follow each ad and observe changes in price and description on a daily basis until its withdrawal. Every ad is placed only for a week. When the time is up a seller has to prolong it, otherwise it will be withdrawn. This means that we observe the actual week of ad withdrawal and control for possible sample selection.
• The second dataset contains all the ads listed at a certain day, downloaded once at May of 2013, May of 2014, May of 2015, February of 2016 and February of 2017. Each part of the dataset contains around 3000 observations. To add the geographical coordinates we take the GIS data of all buildings in Perm with their addresses and geographic coordinates and match it with prices data by address of a building. This allows constructing the geocoded data on prices and calculating the geographic distances between listed objects. The GIS dataset also contain the full list of firms and its industry belonging. We use this firms data to calculate the distances between listed objects and firms to calculate the density of social and commercial objects within a close distance to the object.

• We study the dependence between housing prices and measures of surrounding area commercial development (proximity of large malls, density of commercial objects).
• We test the hypothesis that sellers when set the price look for the closer objects in the terms of similar characteristics also. We construct various distance metrics in characteristics space and use spatial autoregression (SAR) model of housing price with two weighting matrices: weighting based on distance in geographic space and weighting based on distance in characteristics space.

• Ozhegov E. M., Sidorovikh A. Sellers' Heterogeneity in Housing Market: Difference in Pricing Strategies //  Journal of Housing Economics, 2017
• Sidorovikh A. Estimation of effects of transport accessibility on housing prices //  Applied Econometrics, Vol. 47, 2015 (in Russian)
• Ozhegov, Kosolapov, Pozolotina. Study of school quality effect on the surrounded real estate price value //  Applied Econometrics,  2017 (in Russian, revise & resubmit)