AI Analysis: Canada Mortgage and Housing Corporation, housing starts in all centres 10,000 and over, Canada, provinces, and census metropolitan areas, seasonally adjusted at annual rates, monthly

Executive Summary

Canada's housing starts data (Statistics Canada, Table 34100156) spans 36 years of monthly construction activity across 47 regions, revealing a long-run average of 172,628 seasonally adjusted annual units with significant cyclical swings — from a trough of 75,103 units in January 1996 to an all-time peak of 290,954 units in March 2021. Multi-unit housing has grown increasingly dominant over the period, while the COVID-era boom (2020–2022) averaged roughly 230,000 units annually — approximately 33% above the long-run mean. As of February 2026, starts stand at 230,500 units, modestly below the recent 12-month average of 242,194 units, suggesting a gradual cooling from the post-pandemic surge.

Key Findings

• Canada's housing starts peaked at 290,954 annualized units in March 2021, driven by pandemic-era demand and historically low interest rates, and hit their lowest recorded level of 75,103 units in January 1996 following the early 1990s recession.
• The long-run monthly average across the full 36-year period is 172,628 units, with the COVID era (2020–2022) running approximately 33% above this baseline at an average of ~230,000 units annually.
• Multi-unit housing (Multiples) has become the dominant construction category over time, with an average of 17,585 units and a maximum well above single-detached homes, reflecting a structural shift toward higher-density residential development.
• Single-detached units show a median of 4,670 and a mean of 11,148 annualized starts, with a historical maximum of 144,496 units, highlighting substantial variability across the business cycle.
• Larger provinces and CMAs — led by Canada overall, Ontario, and Quebec — dominate regional rankings, while smaller census metropolitan areas consistently record the lowest average housing starts over the 35+ year period.
• Despite extreme highs and lows, no monthly observation technically qualifies as a statistical outlier under the IQR method, as the natural variability of the data produces wide bounds spanning approximately 34,800 to 303,100 units.
• The dataset is high quality, with only 3% missing values across 51,212 rows and 118 unique time-series vectors, supporting reliable long-term trend and regional analysis.