Canadian housing prices : a case study using macroeconomic variables and machine learning

Abstract

This thesis uses cointegration methods to investigate the extent to which segmentation manifests in the Canadian housing markets. By applying the Johansen, Engel-Granger, and FMOLS models, inference on market segmentation and long-run relationships of the associated macroeconomic variables are evaluated. The initial result from the analysis brings to light the issue of using the national housing price index to predict local housing prices and the necessity to study local macroeconomics variables that can influence housing prices. Furthermore, XGBoost, LASSO, and Random Forest will be utilised to predict housing prices using the aforementioned cointegrated variables. Using machine learning methods showed that the variables chosen can reasonably predict local housing prices but ultimately also displayed the limitations of the data set, highlighting the need for more data and features to reduce overfitting.