Central Bank of Luxembourg publishes research comparing machine learning and traditional apartment price indices

The Central Bank of Luxembourg published a Cahier d’études examining how different methods can be used to construct an apartment price index for Luxembourg, including a machine learning approach alongside more established techniques. The paper finds that the resulting indices show broadly similar trends over time, which can reduce uncertainty that arises when interpreting a single index. The study contrasts the quarterly hedonic index produced by Luxembourg’s National Institute of Statistics and Economic Studies (STATEC) with an alternative “repeat sales” index and a new index built using a simple machine learning model applied to Luxembourg real estate transaction data. It estimates that repeat sales account for 44% of apartment transactions since 2007, and assesses each index using volatility, the size of revisions, coherence across indices, and out-of-sample indications. All three methods confirm rapid price growth in 2018–2021, a sharp slowdown in 2022, and a price decline in 2023; the machine learning-based index closely tracks the traditional indices, is less volatile, but is subject to larger revisions. The central bank notes the paper reflects the authors’ views and not necessarily those of the Central Bank of Luxembourg or the Eurosystem.