Valuing Affordability: Utility and Accuracy Optimisation An Econometric and Data Driven Approach

Abstract

The problem of sustainable and affordable housing stock which is accessible to low income earners is particularly pronounced in developing economies. Housing solutions have the highest potential for social impact in these economies. Nonetheless, uncertainty exists around stock development choices in resource constrained environments. Where the provision of access has failed, issues of land redistribution become topical. The literature review outlines both current and past work on understanding effective responses to housing needs, considering both demand and supply side interventions for increasing housing stock. The contribution of this research is presented in two related themes, the first of which is focussed on combining house value data with other metrics focussed on social outcomes (including wellbeing and health) in order to quantify the extent to which there is correlation. Using factor analysis and instrumental variables, we are able to show that a range of ownership and wellbeing metrics are improved through subsidy programmes, whereas land grants and land restitution lead to neutral or adverse outcomes. Equally, measuring social attitudes as it relates to ownership of properties within estimated value ranges is presented as a novel function for measuring housing access and its impact. We found that government support in general as well as support for government handling of housing and land issues is increased by as much as 11%, when asset scores increase. The second theme is focussed on finding accurate measures of property preference drivers and its occupancy, and by extension property value drivers, across geographies. This is achieved using both conventional hedonic regression modelling and emerging machine learning applications. We use increased accuracies to test the feasibility of building a more robust model for estimating value. Initial analysis is focussed on a subset of a comprehensive dataset in a generic setting, while further iterations expand the scope of datasets subjected to the constructed model to include both new locations as well as a broader sweep of features. Further work still applies industry leading cloud computing infrastructure to a supervised learning environment to attain accuracy gains. Likewise, our stepwise approach to modelling internet usage allowed for up to 12% increases in the per utilised hour value estimation of house sharing units.