Who owns Lausanne?

Abstract

Why are rents in Lausanne so expensive? And who profits from it?

The real estate market is usually quite opaque to the public. Being ourselves residents of Lausanne, we know how hard finding affordable housing can be. Therefore we would like to know more about the market situation that causes these difficulties. Our goal is to analyse cadastral and rental data in order to find and visualise the ownership proportions of real estate in the city of Lausanne and to understand how prices are composed.

More precisely, by leveraging public-domain data from the administration of both the city of Lausanne and the canton de Vaud we attempt to relate real estate owners and high cost of rents. Furthermore, we try to understand the data from a political point of view. Thereby we hope to improve the transparency of the real estate situation and its effects – for the good of our society.

Our results are narrated as a data story on the website who-owns-lausanne.github.io

Repo contents

This repository contains all the code and documentation about the project "who owns Lausanne". The following directory structure helps you while exploring:

• /cleaning: This folder contains scripts that are used to clean some datasets after they have been acquired.

• /data: Load all the data in this folder. All the scripts expect the data to be contained there. This folder has the following subfolders:

  • maps contains the geographical data
  • raw contains scraped data before it is cleaned,
  • owners contains the data about the ownership of parcels
  • rents contains the data about current rent announcements
  • xlsx_commune features some spreadsheets from the statistical office of the city of Lausanne.

• /scraping: Contains all scripts needed to get data from the different web services.

• /Who owns lausanne.ipynb: The main notebook featuring data analysis and explanations. This is your next chapter after this document.

Research questions

• How does the position influence rent prices? Property and rent prices vary greatly with position within Switzerland. Is the same true also within the city of Lausanne? Are any neighbourhoods significantly cheaper or more expensive depending on their distance from the centre?

• How does the type of owners influence prices Several categories of owners are invested in the real estate market: private citizens, companies, pension funds, and public institutions. We ask ourselves whether the proportions of these categories influence rent prices. For example, is a certain part of the city more expensive because most real estate there is possessed by companies?

• Does the position influence the composition of property owners? As position might influence prices, it could also influence the profile of real estate investors. We look for the way in which position influences the composition of the categories of owners. As mentioned before, this is highly related to the dependence of prices and location.

Analysis by quartier

Our analysis will use the quartiers (districts) of the city of Lausanne as a natural spatial unit. Lausanne is divided, for geographical and historical reasons, in 18 quartiers. From a data science position, one might say that this unit is far too large and one might be inclined to use a finer grid in order to perform the analysis.

However, if we remind ourselves of the goal of this project, it is to help society understand a problem: how are rent prices composed? As human beings we are used to think of a place by its political divisions like the districts of a city. If someone is asked where in the town they live the usual answer will be the name of the quartier. The same should be true if someone asks, where in the city the rents are affordable. Therefore, we will use this conventional division to perform our analysis and to showcase our story.

Story outline

We want to turn this project into a story that will be told on a nicely designed webpage. The story will take the perspective of someone trying to find accommodation in Lausanne, think of a newly arrived student.

Driven by the difficulties of finding an affordable home in Lausanne, we start asking the question: Who owns all this real estate? And why are the rents in some quartiers cheaper than in others? Following this, there will be an analysis of the ownership patterns overall and for different quartiers. Where are the most big companies etc. (see research questions).

The main part of the story will explain our findings about the composition of prices in different quartiers. It will combine the knowledge from the previous part and it will bring all the calculations together. We hope to be able to give an explanation whether the distance from the centre or the ownership pattern influences the price and where a newly arrived student should search for affordable accommodation.

The entire story will feature various maps showing for example the ownership patterns or the differences in rent prices.

Datasets

Even though our datasets are not listed on the site opendata.swiss, we still consider them to be "open swiss data". They come from official swiss administrations or swiss websites/webservices and they are open to the public.

Cadastral data

The basis for our analysis is the cadastral data which is published by the city of Lausanne on map.lausanne.ch. It features information for each parcel including the owner, the area, and the position. The dataset is described on asitvd.ch. Here is an example screenshot and data for the Bel-Air building in the centre of Lausanne:

The cadastral dataset is available for free for research institutions. We contacted the service du secrétariat général et cadastre of the city of Lausanne on Friday, 2 November. They gave us the access to their ftp server.

Address data

We also need to convert addresses to coordinates at some point. There is luckily another cadastral layer of building addresses, provided by the Cadastral offic of Lausanne. It can again be found on the ftp server.

Maps of the quartiers

To be able to capture the space-dependent behaviour of rent prices and ownership structure, we will need to aggregate our data by position. As said before, we will use the official quartiers delimitation of the city of Lausanne. The reference map is hosted on Google Maps. It is possible to download it as a KML file.

Rent prices

To collect data points on the cost of rent in Lausanne, we retrieved the current listings for rents from three websites: anibis, homegate and tutti. We were able to download about 900 offers from Anibis, 400 from Homegate and 600 from Tutti. The data needed extensive cleaning to select only offers for which the address is known and valid, and the surface area of the offer's object is available. Entries present in all datasets were detected and deduplicated.

Implementation

Sourcing the data

Getting our hands on the required data was already a challenging aspect of this project. Several scrapers were developed for this purpose. The Jupyter notebook describes the scraping and cleaning phase.

Data pipeline

The total datasets size is under 1 GB. We can therefore run all of our analysis on a single local machine.

At the source, the cadastral data of Lausanne is available in ShapeFile format. The quartiers boundaries are in KML format. The rent offers are in the raw formats used by the respective website's UI. To process the spatial data in a coherent way in Python, we converted the ShapeFile and KML files to GeoJson by using the QGIS software application. Ad-Hoc parsers were needed for the rent offers data.

Model

The main analysis goal of this project is to determine how prices vary as a function of the ownership pattern of a quartier and as a function of the quartier's distance from the centre. In order to do this, we will use a linear regression model that is detailed in the notebook. We hope that this will reveal information about price composition of rents.

Further ideas and discussion

Having now a clear line (the calculation and estimation of the prices in the quartiers of Lausanne) guiding us through our project, we can assess what benefit the project can make of certain datasets. Clearly, we needed some additional rental data in order to estimate the mean rent of a quartier. This data was obtained by scraping some of the most used websites for real estate announcements in Switzerland (anibis.ch, homegate.ch and tutti.ch). Together with the ownership data we already had this completes our needs for the regression model.

There were additional ideas that came up during our discussions of the model and the story we wanted to tell. Some of them are listed here. For most of them we won't have the time and data. But if (for some miracle) there is still time we might want to use one of them:

• During our research we stumbled upon some statistics done by the statistical office of the city of Lausanne. They contain similar data as the results of our analysis (rental prices by quartiers, etc.) However they date back to the year 2000. Nonetheless, it could be interesting to compare our results in the end to those from 2000.

• Rating the quality of life in each quartier by counting the number of shops, restaurants, bars, bus stops and the like. This could be done using the Google Maps API. The life quality factor would then be another covariate in the regression model. However, coming up with the scoring function is very complicated.

• Estimating the real estate surface of the buildings using the data for building heights extracted from a LIDAR scan of the canton. We already mentioned this in the last milestone and the dataset is also described on asitvd.ch.

• Another idea was to cluster the parcels without a notion of quartiers.
  However, as described before quartiers are a very meaningful and human delimiters. Also, we weren't sure how such a clustering result would have to be interpreted for our analysis and what we could deduce from it for our story.

Individual Contributions

• Jonathan Besomi:

  • responsible for the Jupyter Notebook
  • scraped tutti.ch for rent offers
  • price extrapolation with K-nearest neighbours on parcelles
  • will prepare the final presentation

• Yann Bolliger:

  • wrote text/concept for project proposal
  • scraped Homegate for rent offers
  • linear regression on distance and surface
  • designed website
  • wrote data-story
  • will prepare the final presentation

• Pietro Carta:

  • sourced the cadastral dataset from the Lausanne office of cadastre
  • scraped the missing details of the cadastral dataset
  • explored the cadastral dataset with QGIS
  • scraped anibis for rent offers
  • matched rent offers to geographical positions and owners
  • linear regression on price depending on owner type
  • denoised the owner types map