Using machine learning to identify spatial market segments. A reproducible study of major Spanish markets

Identifying market segments can improve the fit and performance of hedonic price models. In this paper, we present a novel approach to market segmentation based on the use of machine learning techniques. Concretely, we propose a two-stage process. In the first stage, classification trees with interactive basis functions are used to identify non-orthogonal and non-linear submarket boundaries. The market segments that result are then introduced in a spatial econometric model to obtain hedonic estimates of the implicit prices of interest. The proposed approach is illustrated with a reproducible example of three major Spanish real estate markets. We conclude that identifying market sub-segments using the approach proposed is a relatively simple and demonstrate the potential of the proposed modelling strategy to produce better models and more accurate predictions.

Automated summary

This paper presents a novel approach to market segmentation using machine learning techniques. It demonstrates how classification trees and spatial econometric modeling can improve the fit and performance of hedonic price models, resulting in better models and more accurate predictions.