Data Study on House Selling Price Estimation
The skills the author demoed here can be learned through taking Data Science with Machine Learning bootcamp with NYC Data Science Academy.
Introduction
Selling your house requires finding a buyer willing to pay the price you are ready to accept. This can be a difficult task, especially in the Netherlands, data shows as the final price is determined through negotiation, where buyers try to push down the asking price for the house. In most cases, homeowners ask real estate brokers to assist them in the selling process and in setting an asking price that will maximize the selling price.
In the Netherlands, this process is not different; however, since a few years, homeowners have begun selling their houses without the intervention of an established real estate firm. Home owners do this as this can easily save them a few thousand euros, as the real estate brokerโs fee is a percentage of the selling price (https://www.rtlnieuws.nl/editienl/klaar-met-de-makelaar-verkoop-je-huis-zelf). Consequently, the services originally provided by the real estate broker, like providing a tour through the house to potential buyers or advertising the house, now have to be organized by the homeowners themselves. The homeowners also have the challenge of setting the right asking price for the house.
Data Background
Real estate brokers usually have some experience in assessing the market value of a home because they are familiar with the area and have access to private databases that allows them to easily compare houses on house characteristics and price. Your average homeowners doesnโt have this knowledge or access. Consequently, there is asymmetric information between the real estate firm and the house owner, and it is on that basis that real estate agents can offer a value-added service.
From a commercial perspective, such a practice can be considered absolutely normal; however, in this particular case the information on houses, which is used by real estate firms, is no longer completely private. Nowadays, information on houses, which are for sale on the Dutch housing market, can be found on a large number of websites and are therefore accessible to the public. However, website design still does not allow homeowners to perform easy comparison and therefore prevents them from valuating their house effectively and precisely.
This problem raises the following research question: Can the elimination of asymmetric information improve the seller's position in the Dutch real-estate market?
Methodology and Data
In order to access and store housing information, Python web scraping techniques, such as Beautifulsoup, were deployed. Beautifulsoup creates the possibility for the efficient extraction of information from web pages, which can be stored in comma separated files or Microsoft Excel files. To reduce the load on the server that contains the information, only text fields were extracted.
Additionally, to maximize the scraping speed, TOR network I.P. rotation was deployed. This technique prevents that the server might block the I.P address from which scaping is performed as the server observed load per I.P address is reduced to normal single user load (a maximum of 200 pages are scraped before the I.P address is changed). In the case that the I.P address is blocked, the I.P. address is automatically changed, and the scraping process continues. The TOR network provides access to 7000 proxies worldwide, which makes the scraper very robust against anti-bot technologies. In total 72000 pages were scraped for 35 different house characteristics.
Prediction Model
In addition to the scraping of information, the information was used to construct a prediction model for house pricing. This model uses characteristics of the house, like the size of the house and the type of the house to estimate the selling price. In order to estimate this model, a multiple linear regression model with Whiteโs Standard Errors was used. This particular model was chosen to correct for the violation of the Gauss Markov assumption for orderly least square regression and to aim for the best linear unbiased estimators. The final model obtained an adjusted R2 of 0.87 and therefore provides an overall good estimation.
With the aim of using the estimated model to improve the seller's position in the real-estate market, the model was embedded in a Shiny application. This allows house owners to determine the selling price based on their houseโs characteristics.
Data Results of house selling price estimation
In order to estimate the selling price of a house within the Netherlands and to cross validate the valuation model, multiple houses are selected from the best known housing website in the Netherlands, www.funda.nl; which represents 80% of the available houses on the housing market (https://www.emerce.nl/nieuws/concurrenten-funda-winnen-marktaandeel). From this website 5 houses were randomly selected for estimation.
The table below presents the results from the price estimation and indicates that the model predicts in line with expectations, considering the R2 of 0.87. In most cases, the model overestimates the sales price of the house, except for houses in the higher price range, as indicated by house nr. 5. Overall it can be concluded that having access to market information allows for a close approximation of the sales price of houses. This implies that for determining the sales price of houses that are considered average for the Dutch housing market, a real estate broker is no longer required, assuming that market information is available.
Table 1: Sales price estimation for 5 randomly selected houses in the Dutch real estate market
House 1 | House 2 | House 3 | House 4 | House 5 | |
Type | One family between house | Farmhouse | Two under one roof | Free standing | Villa |
Zipcode | 1966 | 9842 | 3315 | 2181 | 1921 |
Surrounding | Quiet road | Quiet road | Residential | Quiet road | Quiet road |
Living space | 90 | 280 | 179 | 196 | 290 |
Ground area | 144 | 1020 | 229 | 590 | 1165 |
Type of garden |
Front and back yard | Side garden | Front and back yard | Surrounding garden | Surrounding garden |
Number of floors | 1 | 2 | 3 | 2 | 2 |
Number of bathrooms | 1 | 1 | 1 | 2 | 3 |
Insulated | No | No | Yes | Yes | Yes |
Parking | Public parking | Public parking | Public parking | Public parking | Garage |
Estimated sales price | โฌ 266,204 | โฌ 354,203 | โฌ 395,271 | โฌ 673,367 | โฌ 1,015,375 |
Actual sales price | โฌ 249,000 | โฌ 345,000 | โฌ 389,000 | โฌ 650,000 | โฌ 1,149,000 |
Difference | 7% | 3% | 2% | 4% | -12% |
Conclusion
Nowadays house owners are selling their house without the assistance of a real estate broker but are still dependent for the valuation of their house. This dependency exists as market information is not easily accessible for the public, which makes it difficult to make comparisons among houses.. Research has indicated that, with access to market information, it is possible to closely estimate the selling price of the house without the intervention of real estate brokers.
Consequently, removing the dependency on real estate brokers and improving the position of sellers on the market, as a valuation fee is no longer required. Therefore, it is possible to conclude that the elimination of asymmetric information does improve the seller's position in the Dutch real estate market; however, only applies to houses that are considered average in contrast to the existing housing market.
Disclaimer
This publication has been prepared solely for illustration, educational and or discussion purposes. It does not constitute independent research and under no circumstances should this publication or the information contained in them be used or considered for the valuation of real estate within or outside the Netherlands.