Exploring sub national housing markets with the Kohonen map - evidence from three European country contexts

To explore and monitor the housing market fundamentals (prices, dwelling features, area density, residents etc.) on a macro-locational level is motivated because of various policy or business related goals. Housing market segmentation, a concept with increasing relevance, means differentiation of housing due to income and preferences of the residents and administrative circumstances (i.e. the emergence of housing submarkets). In order to capture the segmentation empirically, the study applies a fairly new and emerging technique known as the Îself-organizingÌ map (SOM, ÎKohonen mapÌ). The SOM is a type of (artificial) neural network, a non-linear and flexible (i.e. non- or semi-parametric) regression and Îmachine learningÌ technique. By utilising the ability of the SOM to visualise patterns, it is possible to analyse various dimensions within the variation of the data set. Segmentation may then be detected depending on the resulting patterns across the map layers, each of which represents the data variation for one input variable. Following an inductive modelling strategy, cross-sectional and nationwide data of the owner-occupied housing markets of Finland, the Netherlands and Hungary, respectively, are run with the SOM. On the basis of the resulting configurations certain regularities (similarities and differences) across the three contexts may be identified. In Finland, housing market segmentation depends on two factors: relative location within the region, and house type. In the Netherlands, the picture is more heterogeneous within urban areas of all sizes, but more homogeneous across municipalities. In Hungary the structure shares features from both previous contexts: features of relative macro-location can be identified, as can a sharp market segmentation within counties and regions. In all three cases the segments are determined by physical and institutional differences between the housing bundles and localities. The exercise demonstrates how the inductive SOM-based approach is well suited for illustrating the contextual factors that determine housing market structure. Additionally, some empirical results are generated that may have some further significance in respective contexts.