International Migration, Housing Demand Measurement, and Vancouver New-House Price Forecasting
Better housing-demand accounting does not automatically produce better forecasts. This study separates a coherent migration measure from the evidence needed to show incremental predictive value.
Overview
International migration is often discussed as housing demand, but raw population flows can double-count temporary-to-permanent transitions and ignore differences in household formation. This paper combines population components for 41 Canadian metropolitan areas with status- and tenure-specific housing-use rates to estimate migration-associated occupied-dwelling use per 1,000 dwellings.
The study tests whether that measure improves forecasts of Vancouver’s New Housing Price Index using release-delay-safe pseudo-real-time and historical-release evaluations. Some short-horizon models improve point estimates, but the gains are imprecise or do not persist across horizons and benchmarks. The central result is therefore disciplined: accounting coherence and incremental forecast value are separate questions.
Core contributions
- 01
Reproducible demand measure
Constructs a comparable migration-associated occupied-dwelling-use flow while reconciling temporary-to-permanent transitions instead of counting them as new external arrivals.
- 02
Release-aware evaluation
Uses current-vintage pseudo-real-time tests and a separate reconstruction from exact Statistics Canada historical releases to reduce look-ahead bias.
- 03
Multi-horizon comparison
Compares benchmarks, autoregressions, pooled models, forecast combinations, and a neural specification across quarterly and annual horizons.
- 04
Transparent negative evidence
Reports where apparent gains are statistically imprecise, fail at longer horizons, or do not clear incremental-information criteria.
Evidence and scope
This SSRN preprint studies measurement and predictive association. It does not identify a causal policy effect or establish prospective forecast superiority. Several exercises are exploratory, historical data can be revised, and results for Vancouver new-house prices should not be generalized to all housing markets or outcomes.
Abstract
This paper constructs a reproducible measure of occupied-dwelling use associated with international migration and asks whether the measure improves forecasts of Statistics Canada’s Vancouver New Housing Price Index. We combine annual population components for 41 census metropolitan areas over 2002–2024 with national status- and tenure-specific housing-use rates. The accounting applies a nationally calibrated reconciliation of temporary-to-permanent transitions instead of treating them as new external arrivals, and expresses the resulting flow per 1,000 occupied dwellings. We conduct a release-delay-safe, current-vintage pseudo-real-time quarterly evaluation and a retrospective annual expanding-origin comparison. An exploratory extension evaluates realized-future information diagnostics, a 19-feature monthly leading layer, horizon-specific cross-CMA partial pooling, strictly historical-error forecast combinations, and a two-hidden-layer tanh neural network. On 102 common Vancouver origins, the exponential-weight combination has four-quarter MAE 2.826 versus 3.256 for the random walk, a 13.2% reduction, but the paired Newey–West comparison is imprecise (p = .224). At eight quarters its MAE is 6.146, worse than the random walk’s 5.736. Neural MAEs are 0.907, 3.392, and 6.560 at one, four, and eight quarters; none improves on both benchmark criteria. A separate reconstruction from exact Statistics Canada The Daily releases produces 585 historical-first-release retrospective forecasts. At one quarter, AR(4)-direct MAE is 0.781 percentage points versus 0.887 for the random walk on 51 origins; the candidate-minus-random-walk descriptive interval [-0.283, 0.071] crosses zero. In the annual metropolitan-area exercise, the composition-adjusted central sensitivity likewise fails its incremental-value criterion against both the pooled base and raw total population flow in every evaluation set. A secondary purpose-built-rental exercise also fails its incremental-information criterion for both rental outcomes. The evidence therefore separates accounting coherence from incremental predictive value. The results concern new-house prices and predictive association; they do not identify a causal policy effect or establish prospective forecast superiority.
Cite this work
Use the DOI record for stable bibliographic information and citation formats.