Interpretable Machine Learning for Real Estate Market Analysis

Authors
Lorenz, Felix 
Willwersch, Jonas 
Cajias, Marcelo 

No Thumbnail Available
Type
Article
Change log
Abstract

While Machine Learning (ML) excels at predictive tasks, its inferential capacity is limited due to its complex non-parametric structure. This paper aims to elucidate the analytical behavior of ML through Interpretable Machine Learning (IML) in a real estate context. Using a hedonic ML approach to predict unit-level residential rents for Frankfurt, Germany, we apply a set of model-agnostic interpretation methods to decompose the rental value drivers and plot their trajectories over time. Living area and building age are the strongest predictors of rent, followed by proximity to CBD and neighborhood amenities. Our approach is able to detect the critical distances to these centers beyond which rents tend to decline more rapidly. Conversely, close proximity to hospitality facilities as well as public transport is associated with rental discounts. Overall, our results suggest that IML methods provide insights into algorithmic decision-making by illustrating the relative importance of hedonic variables and their relationship with rental prices in a dynamic perspective.

Publication Date
2022-05-31
Online Publication Date
Acceptance Date
2022-05-06
Keywords
Interpretable Machine Learning, Microeconomic Hedonic Pricing, Housing Markets, Rental Markets
Journal Title
Real Estate Economics
Journal ISSN
1080-8620
1556-5068
Volume Title
Publisher
Wiley