Tag Archives: pricing

Quantifying the effect of location on Swiss multi-family house prices

It doesn’t take long before one discovers that in real-estate location is key:

Location, location, location. You may have heard this mantra when talking to an agent about the home values. In a nutshell, it means homes can vary widely in value due to their location. For example, the median cost of a single-family home in Decatur, Ill., is $107,900. The median cost of a single-family home in the Honolulu, Hawaii, area is $813,500. Location is essential when it comes to the value of a property.

thebalance.com/what-location-means-in-real-estate-1798766

… only homebuyers who choose the best locales will be holding the most valuable property that also depreciates at a much slower rate. This difference in value is largely a result of a home’s location.

investopedia.com

Recently I worked on real-estate price prediction at CrowdHouse. We’d developed an in-house method to quickly qualify incoming properties (exclusively multi-family ones). It’s mostly based on construction/renovation year and location. Our analysts make heavy use of a thing called “micro/macro location“: https://www.wuestpartner.com/data/ratings. In short, it’s a single number representing location’s quality (1-6, the higher the better), based on a number of factors such as public transport infrastructure, schools, noise, shops, leisure possibilities, etc. Here’s our blog post explaining the concept in more detail: https://crowdhouse.ch/de/blog/lagebeurteilung-von-immobilien-mikro-und-makrobetrachtungen/ , but you can imagine it as a heat-map:

hypothetical macro-location heat-map, source: immomapper.ch

While benchmarking our price estimator on some 160+ properties, I looked into micro/macro location <-> price correlation, expecting to see a strong one, because the better the location the higher the price, right? Surprise, surprise:

0.13 and -0.05, in other words nothing, nix, nada, zip. Property price has nothing to do with its (macro/micro) location.

To put this into perspective, consider these: property’s construction year and price correlate with ~0.3; rental income and price ~0.98.

How’s that possible? Couple of reasons:

  • inaccurate data (this is for sure the case, but not sure to which extent)
  • our data sample is not representative
  • the differences are so small that they don’t show up, i.e. all locations are pretty good
  • these models behind micro/macro location are incorrect
  • bug in my python script

It’s possible that these micro/macro location models do not even try to capture pricing information, but some other qualities, such as how fast does it sell, how well does it maintain its value, etc.

And yes, correlation does not mean causation:

spurious correlations, source tylervigen.com