Canada is forming households faster than it is building homes

What we observe. As of May 2026, Canada was starting homes at a seasonally adjusted annual rate of about 247,000 units, or roughly 22,600 actual starts in the month. Toronto, the country's largest construction market, completed just 2,250 units in February. Set against that, household formation pressure was still negative in January (about -4.0). That indicator nets the pace of new household formation against the pace of completed homes, so a negative reading means demand is outrunning delivery: families are forming faster than the housing stock is growing.

Why it matters. Every month this gap persists, the problem compounds. A household that forms without a matching completion joins a backlog that eventually has to be built, and that backlog is what underwrites developer demand for financing, land and labour over the next several years. It also puts a floor under residential values even with affordability as stretched as it is. The risk for lenders, developers and policymakers is misreading the gap as a slow patch. Treat it as cyclical and you under-allocate capital to the one sector where the structural case for new supply is clearest.

The chain runs from population and migration flows through to starts, with financing as the binding constraint at each stage. Net inflows of non-permanent residents add to formation pressure with about a 90-day lag, though the effect per person is weaker than for permanent residents because new arrivals tend to share accommodation at first. The flow of people understates the eventual flow of households. Building permits in Toronto lead national housing starts by roughly six months, so today's permit pipeline is the best forward read on starts in late 2026; softer permit values now mean fewer starts then. The most direct link is credit. Construction lending velocity drives starts with only a 30-day lag and carries the strongest measured relationship in the model: when credit to builders tightens, starts fall almost immediately, whatever demand is doing. Purpose-built rental starts feed the national total with no meaningful lag, so a change in the economics of rental development, whether cap rates, construction costs or rent expectations, shows up in the aggregate right away. Put together, the supply side is constrained by financing, led by a permit pipeline that takes six months to materialise, and exposed to any credit tightening, while the demand side keeps accumulating on population momentum and migration.

None of the five macro conditions the model tracks (an inflationary backdrop, liquidity stress, deflation, asset-bubble dynamics, a credit contraction) is currently active, and that matters. In a credit contraction, the link between construction lending and starts would be amplified by roughly 1.8 times and the supply shortfall would accelerate. Nothing like that is in force. Lending conditions are close to normal, which makes the current starts shortfall harder to explain away as a credit-market distortion. The gap between formation and starts is opening in a benign financing environment, which points to a structural constraint: zoning, construction capacity, project economics. A temporary financing freeze would self-correct when credit normalises. This will not.

Two relationships in this analysis carry a confirmed record against real outcomes. The construction-lending-to-starts link has been confirmed twice and contradicted zero times, most recently in June 2026, which backs the read that financing availability is the proximate driver of starts. The Toronto-permits-to-national-starts link is also two for two, checked at the same date, which backs its use as a six-month leading indicator. Both point the same way as the thesis. The rest of the relationships here, covering migration flows, population momentum, purpose-built rental starts and provincial contributions, have not yet been validated against outcomes in this model. The economic logic behind them is sound and some are close to accounting identities, but their empirical record is unestablished. Weight the lending and permit channels more heavily, and treat the migration-to-formation pathway as plausible but unproven.

The most serious challenge comes from a statistical anomaly in the construction data. Construction employment and housing starts are diverging at 3.18 standard deviations beyond their normal 18-month relationship. In a healthy cycle workers follow projects and the two move together. A divergence this large means one of two things: starts are about to recover sharply toward employment levels, which would close the gap and weaken the thesis, or employment is lagging a genuine collapse in the project pipeline, which would confirm it. The model itself flags that this pair should be trusted less than usual right now. There is a second problem. The formation-pressure indicator may overstate true demand if non-permanent residents are doubling up at higher rates than history suggests; the migration-to-formation link already carries a weaker per-person effect, and if that discount should be bigger still, the effective shortfall is smaller than it looks.

What would falsify this. If seasonally adjusted national starts hold above 280,000 units over the next two quarters while formation pressure moves back toward zero or positive, the thesis fails. That combination would mean supply is catching up with formation and the deficit is closing, not deepening.

What to watch. Watch monthly construction lending velocity against the seasonally adjusted starts count. If lending stays available but starts keep undershooting what the permit pipeline implies on the normal six-month lag, the constraint has moved from financing to something harder to fix (approvals, construction capacity, project viability), and the deficit will widen faster than the lending channel alone would predict.