Automated Housing Price Valuation and Spatial Data

Abstract

AbstractThe demand for automated, reliable and understandable housing price valuation mechanisms is increasing. Most efforts have been made to improve model accuracy and prediction power through the well-established standard econometric models based on regression techniques. However, the modelling of the spatial attributes of housing through mass appraisal tools has been given less attention. Incorporating spatial modelling approaches through econometrics frameworks opens new opportunities for improving automated valuation tools.This work presents an exploratory analysis of different approaches to incorporating spatial data into AVM tools, taking advantage of the potential of spatial (big) data, stored on different sources – census data, open street maps and public administration data.Improvement of the standard housing price models embedded in a Portuguese housing appraisal decision system (held by PrimeYield SA) will be presented. Different strategies to incorporate spatial data from public sources are analysed, taking the Sintra municipality and PrimeYield data on this territory as a case study. The focus is the mitigation of the well-known pitfalls of spatial models, such as spatial heterogeneity and spatial dependence.The results show the potential added-value of collecting and (pre)processing a different set of territorial variables – socioeconomic, accessibility, and land use – to improve the explanation power, parsimony and understanding of housing price models. Geographic weight regression models can be a balanced compromise to achieve those objectives which will be investigated.