Understanding Housing Bubble in Vancouver, BC
Δlog HPI (2005–2021): ADF; SARIMA (2,1,0)×(2,0,0)_12 via AIC/BIC; Ljung–Box; 12-month forecasts. VAR(1) on first differences of HPI, GDP, 10-yr yield, CAD/USD, employment, earnings, CPI, crime + Bill-28 dummy; Granger, Cholesky IRFs, FEVD. Results: HPI variance mostly self-driven; GDP predictive (95%); employment marginal (90%); CPI/yield short-run; policy coefficient negative/significant.
Overview
- Univariate modeling (HPI growth, 2005–2021):
- ADF stationarity tests on $\Delta\log(\mathrm{HPI})$.
- ACF/PACF-guided SARIMA $(2,1,0)\times(2,0,0)_{12}$ chosen by AIC/BIC.
- Ljung–Box diagnostics; 12-month out-of-sample forecasts.
- Multivariate dynamics (VAR(1) in first differences):
- Variables: HPI, GDP, 10-year yield, CAD/USD, employment, earnings, CPI, crime; exogenous Bill 28 policy dummy.
- Granger causality, Cholesky IRFs, and FEVD.
- Findings: HPI variance mostly self-driven; GDP predictive (95%); employment marginal (90%); CPI/yield shocks are short-run; policy coefficient negative and significant.