Forecasting Canadian Housing Prices: Assessing the Forecasting Performance of ARIMA and GARCH Models.

Abstract

This paper contributes to the existing literature on the Canadian real estate market by analyzing the forecastability of Canadian real estate prices with a particular focus on the regional markets of Vancouver, BC and Toronto, ON. Using a multitude of Canadian real estate data sets, a number of time-series forecasting models are estimated employing a rolling estimation methodology to generate Single-Step-Ahead and Multi-Step-Ahead forecasts. Mean Squared Prediction Errors and Mean Absolute Prediction Errors are then calculated and used to qualitatively compare the relative forecasting performances of the various models for each particular data set. The results suggest that no single model performs best across all data sets, but rather optimal forecasts are obtained by selecting models which are most suitable to the particular data series under consideration.