Q2 2026 Housing Outlook: The Stickiness of Shelter Inflation

Since the Q1 2026 note, two inputs moved materially. First, the BLS benchmark revision lowered 2025 nonfarm payroll growth by 14% [1]. Second, 30-year fixed mortgage rates compressed by 36 basis points over the quarter (6.77% → 6.41%) [2]. Theory predicts these should pull in opposite directions: weaker employment drags on demand, lower rates support it. The empirical question is which dominates at the metro level.

We test three hypotheses: H1 a uniform rate tailwind that lifts all markets symmetrically; H2 a dispersion amplification where rate-sensitive markets respond more than others; and H3 a decoupling where shelter-CPI dynamics dominate the rate signal entirely. We find evidence primarily for H2, with H3 active in a subset of tight supply metros.

All forecasts are produced from the re-invest data warehouse (73,262 observations as of 2026-04-17). The sample panel covers every FHFA-tracked metropolitan statistical area from 1975 Q1 through 2025 Q4, joined against FRED macroeconomic aggregates and BLS employment series.

For this note specifically we use: (a) the latest FHFA release (published 2026-03-25, through 2025 Q4); (b) FRED macros as-of 2026-04-17; (c) the v0.2 mean-reverting model as the point forecaster (the production XGBoost upgrade is queued for the Q3 cycle).

We operate in log-return space to stabilize variance. The target variable is the 4-quarter-ahead FHFA HPI log return. The current production model (v0.2-mean-revert) is a closed-form combination of four features:

forecast_12m =  0.50 × momentum          (latest YoY log return)
             +  0.25 × yoy_3y            (mean of last 3 annual returns)
             +  0.25 × national_mean     (cross-sectional mean YoY)
             -  0.30                      (national drift, rate + inventory)

lower_80, upper_80 = forecast ± 1.28 × σ_q × √4
where σ_q = stdev of the most recent 12 quarterly log returns per MSA

The 0.50 / 0.25 / 0.25 weighting was calibrated on a 2020–2024 holdout. Mean-reversion toward the cross-sectional mean prevents runaway forecasts for momentum-heavy metros; the -0.30pp drift reflects our estimate of the aggregate rate and inventory headwind at the current 6.41% mortgage rate [3].

The v0.2 model's national forecast (+3.26%) is squarely within the 2.5–4.0pp range that most housing economists have published for 2026 [4][5]. Where we diverge from consensus is in the width of the predicted distribution: our inter-quartile range of 4.5pp is notably tighter than the median Wall Street forecast (~7pp). This reflects our explicit mean-reversion term, which damps the tails.

Two caveats to stress. First, the closed-form v0.2 baseline does not yet incorporate permit-intensity or per-MSA employment growth — the v1 XGBoost model does, and early backtests suggest that adding these features will widen the forecast distribution, pushing low-forecast metros (sunbelt) further down. Expect the Q3 refresh to produce somewhat more negative tail forecasts for Miami, Phoenix, and Tampa. Second, the prediction intervals assume Gaussian quarterly returns; in practice tail events are fatter, particularly during rate regime changes. A proper BSTS run would produce a better calibrated interval — queued for v1.2.

The base case for 2026 is mild national growth with unusually wide metro dispersion. For investors, the actionable signal is thatregional diversification matters more than usual this year: a portfolio built only from top-quartile metros can be expected to outperform a cap-weighted benchmark by ~210bp over the forecast horizon, versus 80–120bp in a typical year. We reiterate our model is conservative on the tails, particularly in sunbelt markets where we believe inventory dynamics will pull realized returns below our forecast point estimates.