---
title: "Modelling and Forecasting Housing Investment"
authors: "(Bank of Canada Working Paper 2005-41)"
source: "Bank of Canada"
source_url: "https://www.oar-rao.bank-banque-canada.ca/record/1346/files/wp05-41.pdf"
paper_id: "Bank of Canada WP 2005-41"
harvested: 2026-08-22
year: 2004
keywords: [housing investment, structural model, starts, permits, forecast]
pdf_url: "https://www.oar-rao.bank-banque-canada.ca/record/1346/files/wp05-41.pdf"
---
## Abstract

This paper (by Frédérick Demers, Bank of Canada, Working Paper 2005-41) proposes and evaluates econometric models to explain and forecast real quarterly housing expenditures in Canada over the sample period 1961–2004. Residential investment is split into two categories: construction (new construction plus renovation expenditures) and resale (transaction costs from property transfers), which are modelled separately. The housing sector accounts for roughly six percent of Canadian real aggregate economic activity. The paper finds that the long-run relationship between housing expenditure and its determinants shifted during the late 1970s, and that housing investment became more sensitive to interest rates over time.

The analysis identifies several key fundamental drivers of housing investment, including the relative price of housing, demographics, the labour-force participation rate, wealth, interest rates, and stock-adjustment effects. Long-run specifications are estimated as non-linear cointegration models in one step using constrained maximum-likelihood techniques, with a structural-break test indicating a shift in the cointegrating relationship. A Gregory-Hansen test is used to locate the structural break, and Granger-Lee tests reject non-symmetric equilibrium correction, supporting symmetric adjustment.

On forecasting performance, simple leading-indicator (LI) models deliver relatively accurate near-term forecasts. Among the findings reported in the paper and discussed in the wider literature: the preferred structural model, which allows for a shift in the cointegrating vector, provides good forecast accuracy for the construction category but not for resale, which is considerably more difficult to predict given its much greater variance and abrupt corrections. In addition, disaggregating starts into single-family versus multi-family components does not improve aggregate starts forecasting, indicating that a single aggregate housing-starts series captures the information relevant for forecasting the aggregate. Overall, the paper offers a structural, fundamentals-based framework for forecasting Canadian housing investment.