Spatial Autoregressive Analysis and Modeling of Housing Prices in City of Toronto

Abstract

Previous housing price studies based on hedonic price modeling have mainly focused on applying various factors, including built environment variables in the analysis, without establishing a comprehensive theoretical framework as a basis for the model formulation. To address this gap, this study introduces a more systematic framework for decomposing housing prices into land prices as determined by built form, neighborhood socioeconomic characteristics and individual dwellings' physical conditions. Following this logic, this study experiments with the related variables through regression analysis, including consideration of spatial lags, as well as develops a housing price model using a random forests (RF) algorithm. A comprehensive time-series database of housing transaction data for the City of Toronto is used. Modeling results show that neighborhood socioeconomic factors contribute the most to the explanation of housing prices, while housing characteristics and accessibility measures are also significantly influential. The RF model achieves an overall accuracy of 85%, a relatively good performance in reproducing observed prices. The findings provide insights for planners concerning factors influencing housing prices and, hence, residential location decision-making.