Disentangling spatial and structural drivers of housing prices through Bayesian networks: A case study of Madrid, Barcelona, and Valencia

Abstract

Understanding how housing prices respond to spatial accessibility, structural attributes, and typological distinctions is central to contemporary urban research and policy. In cities facing affordability challenges and uneven development, models that combine predictive validity with interpretability are increasingly required. Using over 180,000 geo-referenced listings, this study employs discrete Bayesian networks to model residential price formation across Madrid, Barcelona, and Valencia. The learned probabilistic structures reveal distinct city-specific pricing logics: Madrid is characterized by amenity-driven stratification, Barcelona by typology-based differentiation, and Valencia by a spatial–structural pricing core. By supporting joint inference, scenario-based simulation, and sensitivity analysis, Bayesian networks offer a transparent and auditable alternative to black-box models. This transparency is essential not only for equitable governance but also for designing sustainable housing strategies that balance accessibility, land use, and market resilience.