An over time comparison of linear and spatial regression model parameters to predict average housing values: Evidence from two Canadian cities

Residential real estate models are a key component of planning support systems, which aid in shaping future urban land use and transportation. These models typically rely on parameters estimated from historical data corresponding to a specific base year, assuming that such parameters remain stable over time. This paper examines the temporal stability of parameters linked to variables influencing average housing values at the dissemination‐area level. It does so by comparing parameter estimates across multiple time periods in two Canadian cities: Ottawa, Ontario and Calgary, Alberta. Using Canadian census data, pooled regression analyses were conducted for the years 2006, 2011, and 2016 to model average housing values. The findings indicate general consistency in the factors affecting housing values across both cities. Interestingly, the parameters of the Ottawa pooled models suggest strong stability over time compared to Calgary. Additionally, the Spatial Autoregressive model outperformed both linear and spatial error models in terms of accuracy and predictive performance .