Modelling and predicting house price indices in Canada via a hybridisation of machine learning methods

House price dynamics are relevant economic indicators since they affect individual wellbeing, the broader financial system, and the real economy, with significant effects on sectors such as construction and retail. Predicting housing prices is thus crucial for households, policymakers, and investors. This study contributes to the existing literature by introducing novel multi-layer hybrid Machine Learning (ML) models—some combining up to four base learners—for the prediction of Canadian house price indices. Using an extensive, longitudinal 34-year data set covering twelve major Canadian cities, our results show that hybrid ML models offer superior predictive accuracy than traditional models, thereby providing new benchmarks and methodological advances for real estate analytics and forecasting. Hybrid models combining Decision Trees with Adaptive Boosting outperform single ML models, and pairing Adaptive Support Vector Regression and Bayesian Ridge Regression consistently reduces prediction errors. Our findings further reveal that the cost of land is the most important factor in predicting Canadian house prices. Other key factors include financial- and credit-related variables, in particular, mortgage lending conditions. Demand pressures coming from population growth and rising rental rates play a relatively influential role. Our findings highlight the importance of careful regulation of urban land-use, mortgage lending, and credit access to prevent supply constraints and speculative bubbles. Individual homebuyers and investors should consider prudent borrowing, diversification, and market timing to manage risks.