A scoping study of Geographically Weighted Regression (GWR) analysis of house price estimation: with applications to impacts of crime, ethnic/religious segregation and landlord portfolio optimisation
Dr Ellie Bates, University of Edinburgh and Professor Gwilym Pryce, University of Sheffield, Joe Frey, Northern Ireland Housing Executive and Dr John Boyle, Rettie and Co.
This project explored new inter-disciplinary uses of Geographically Weighted Regression (GWR) in criminology, sociology, housing economics and real estate finance, developed in collaboration with the Northern Ireland Housing Executive and Rettie and Co. Ltd. While there are theoretical and statistical problems in using GWR to estimate housing market segregations directly, GWR offered a powerful way to estimate the first step in a new generation of housing market estimation methods based on house price dynamics rather than attribute coefficients. The potential power of GWR, therefore, is not that it revealed directly where sub-market boundaries lay, but that it offered a highly flexible way of estimating the first stage in a Cross Price Elasticities of Prices (CPEP) analysis.
This project sought to establish whether GWR house price models would offer sufficient improvements on existing methods applied to CPEP, and particularly the Location Value Signature (LVS) approach. We then explored a number of potential applications including the extent to which the housing market is blind to race and religion, the house price impact on crime, and computing optimal landlord portfolios. We also reflected on the wider range of future potential applications, such as adding nuance and detail to local authority housing market assessments, and assisting with Council Tax revaluation.
The main outputs from the project were: a a working paper highlighting the value of using GWR compared to other techniques for modelling house prices; an academic journal article based on the working paper; short research reports on the application of GWR to the impacts of crime and religious/ethnic divisions and the application to landlord portfolios with non-technical summaries to be circulated among relevant stakeholders. In addition, this project enabled sharing of expertise with AQMeN and Urban Studies research staff on how to develop and produce Geographic Weighted Regression models. A knowledge exchange seminar and a two day training event were offered to AQMeN members and training materials for use in UG and PG quantitative methods teaching to social scientists was developed.