US housing starts SAAR, June 2026

Framing and exact resolver: this forecast targets the Census/HUD New Residential Construction first print for privately owned housing units started, United States total, seasonally adjusted annual rate, June 2026. The table variant is Table 3a, seasonally adjusted annual rate, not not-seasonally-adjusted starts and not permits or completions.

Reference class and base rate: over the latest official 13-month current-vintage starts sequence, the level has mostly run between 1.27 million and 1.43 million SAAR before the May 2026 drop to 1.177 million. A persistence base rate from the latest print would be 1.177 million, but the component mix suggests May was unusually weak in multifamily rather than a broad collapse.

Level, momentum, and mechanisms: single-family starts were only 17,000 below April and single-family permits were 886,000, so I keep that component near 0.88 million. Multifamily starts at 0.284 million were far below recent starts and below May 5-plus-unit permits at 0.474 million, so I add a partial rebound of about 0.060 million rather than a full snapback.

Prior/update/interval: persistence prior is May 2026 starts at 1.177 million SAAR using the official Table 3a reference class from May 2025 through May 2026; adjustments are +0.000 million for single-family level, +0.060 million for partial multifamily mean reversion, and +0.000 million for policy/timing effects, giving 1.177 + 0.060 = 1.237. Successive monthly changes in millions are +0.090, +0.053, -0.141, +0.028, -0.046, +0.046, +0.059, +0.007, -0.039, +0.176, -0.130, -0.215, so sigma = 0.113; using an approximately normal 80% half-width, 1.28*sigma = 0.145, giving 1.237 - 0.145 = 1.092 and 1.237 + 0.145 = 1.382.

Counter-considerations: upside risk is a faster multifamily rebound toward the 0.40-0.48 million area implied by recent starts and permits, which would land above the interval if single-family also improves. Downside risk is that high mortgage rates and builder caution keep starts near May's depressed multifamily level, which would land below the interval if single-family starts also break below 0.85 million.

Review disposition: accepted the optional reviewer suggestions to state the normal 80% interval mapping explicitly and to identify the 13-month history as current-vintage Table 3a context; no required fixes were raised.

Framing and exact resolver: this forecast targets Census/HUD New Residential Construction first-print privately owned housing units started, United States total, seasonally adjusted annual rate, for June 2026. The target variant is Table 3a starts SAAR, not not-seasonally-adjusted starts, permits, completions, or later revised historical data.

Base rate/reference class: the comparable Table 3a starts SAAR series over May 2025 through May 2026 is volatile but mostly in the 1.27 to 1.43 million range before the May 2026 drop; the last three target-variant readings, 1.522, 1.392, and 1.177 million, average 1.364 million.

Level and momentum: a naive one-month persistence prior would stay near 1.18 million, but May's 15.4 percent fall is unusually large and permits remained at 1.413 million, so I partially mean-revert June toward the recent starts range rather than extrapolating another decline.

Policy and one-off mechanisms: high financing costs, cautious builders, and multifamily volatility argue against returning all the way to March's 1.522 million; still, the permits-starts gap and Census sampling noise argue for a rebound from the May preliminary print.

Prior/update/interval: persistence prior from May is 1.177 million; reference class is recent Table 3a total starts SAAR from May 2025 through May 2026. Successive changes across the fetched starts sequence are +0.090, +0.053, -0.141, +0.028, -0.046, +0.046, +0.059, +0.007, -0.039, +0.176, -0.130, and -0.215 million, giving sigma = 0.110. The Gaussian 80 percent half-width is roughly 1.28*sigma = 1.28*0.110 = 0.141 million. I widen to about 0.170 million because the May print carried a 7 percent RSE and an unusually large one-month drop, yielding 1.300 - 0.170 = 1.130 and 1.300 + 0.170 = 1.470.

Counter-consideration: upside risk is a rapid normalization of multifamily starts or delayed projects moving into June after May's weak print; downside risk is that the May decline reflects a real builder pullback rather than sampling noise.

Outside-the-interval scenarios: a June first print above 1.47 million would land above the interval and likely require a broad multifamily rebound plus stable single-family starts; a print below 1.13 million would land below the interval and likely require May weakness to persist or worsen despite permits near 1.41 million.

Framing and exact resolver: this forecast is for Census/HUD Table 3a, New Privately-Owned Housing Units Started, United States total, seasonally adjusted annual rate, June 2026 first print, converted from thousands to millions. The release variant, anchors, and history are all SAAR first-print or current-release Table 3a values, not NSA or later historical backfills.

Base rate/reference class: for a one-month-ahead total starts forecast, persistence plus partial mean reversion is the right starting point because the official notes say month-to-month seasonally adjusted construction statistics are irregular and it can take 6 months to establish an underlying trend for total starts.

Prior/update/interval: persistence prior is May 2026 total starts at 1.177 million; historical sample is the 12 successive monthly changes from May 2025 through May 2026 in Table 3a, with sigma = 0.110 million and 80% half-width = 1.28*sigma = 1.28*0.110 = 0.141 million. Adjustment components: +0.020 million for stable single-family starts/permits near 0.88 million, +0.050 million for partial rebound in volatile 5+ starts from 0.284 million, and -0.010 million for soft overall momentum after the May drop. Point = 1.177 + 0.020 + 0.050 - 0.010 = 1.237 million; interval = 1.237 +/- 0.141 = [1.096, 1.378].

Counter-considerations: upside risk is a quick multifamily rebound toward the 0.40-0.48 million 5+ unit range seen earlier in 2026, which would land above the interval if single-family also holds firm. Downside risk is that May's fall reflected a broader financing or demand pullback rather than noise, which would land below the interval if total starts stay near or under 1.10 million.

Framing and exact resolver: this forecast is for Census/HUD Table 3a, New Privately-Owned Housing Units Started, United States total, seasonally adjusted annual rate, June 2026 first print, converted from thousands to millions. The release variant, anchors, and history are all SAAR Table 3a values from the current Census/HUD release, not NSA values, smoothed trend charts, or later historical backfills.

Base rate/reference class: for a one-month-ahead total starts forecast, persistence plus partial mean reversion is the right starting point because the Census/HUD explanatory notes say month-to-month seasonally adjusted construction statistics are irregular and it can take 6 months to establish an underlying trend for total starts.

Prior/update/interval: persistence prior is May 2026 total starts at 1.177 million; historical sample is the 12 successive monthly changes from May 2025 through May 2026 in Table 3a, with sigma = 0.110 million and 80% half-width = 1.28*sigma = 1.28*0.110 = 0.141 million. Adjustment components: +0.020 million for stable single-family starts/permits near 0.88 million, +0.050 million for partial rebound in volatile 5+ starts from 0.284 million, and -0.010 million for soft overall momentum after the May drop. Point = 1.177 + 0.020 + 0.050 - 0.010 = 1.237 million; interval = 1.237 +/- 0.141 = [1.096, 1.378].

Counter-considerations: upside risk is a quick multifamily rebound toward the 0.40-0.48 million 5+ unit range seen earlier in 2026, which would land above the interval if single-family also holds firm. Downside risk is that May's fall reflected a broader financing or demand pullback rather than noise, which would land below the interval if total starts stay near or under 1.10 million.

Framing and exact resolver: this forecast is for Census/HUD Table 3a, New Privately-Owned Housing Units Started, United States total, seasonally adjusted annual rate, June 2026 first print, converted from thousands to millions. The release variant, anchors, and history are all SAAR Table 3a values from the current Census/HUD release, not NSA values, smoothed trend charts, or later historical backfills.

Base rate/reference class: for a one-month-ahead total starts forecast, persistence plus partial mean reversion is the starting point because the Census/HUD explanatory notes say month-to-month seasonally adjusted construction statistics are irregular and it can take 6 months to establish an underlying trend for total starts.

Ladder: P(X <= 1.05) = 0.04; P(X <= 1.08) = 0.075; P(X <= 1.096) = 0.10; P(X <= 1.13) = 0.18; P(X <= 1.17) = 0.31; P(X <= 1.21) = 0.43; P(X <= 1.237) = 0.50; P(X <= 1.27) = 0.60; P(X <= 1.31) = 0.72; P(X <= 1.35) = 0.84; P(X <= 1.378) = 0.90; P(X <= 1.42) = 0.95; P(X <= 1.47) = 0.98. Linear interpolation gives the 0.10 quantile at 1.096, median at 1.237, and 0.90 quantile at 1.378, all in millions.

Prior/update/interval: persistence prior is May 2026 total starts at 1.177 million; historical sample is the 12 successive monthly changes from May 2025 through May 2026 in Table 3a, with sigma = 0.110 million and 80% half-width = 1.28*sigma = 1.28*0.110 = 0.141 million. Adjustment components: +0.020 million for stable single-family starts/permits near 0.88 million, +0.050 million for partial rebound in volatile 5+ starts after the May 0.284 million print versus the higher total-starts environment in March-April 2026, and -0.010 million for soft overall momentum after the May drop. Point = 1.177 + 0.020 + 0.050 - 0.010 = 1.237 million. The ladder-implied 80% interval width is 1.378 - 1.096 = 0.282 million, or 0.141 million half-width, matching the 1.28*sigma rule.

Counter-considerations: upside risk is a quick multifamily rebound toward the 0.40-0.48 million 5+ unit range seen earlier in 2026, which would land above the interval if single-family also holds firm. Downside risk is that May's fall reflected a broader financing or demand pullback rather than noise, which would land below the interval if total starts stay near or under 1.10 million. A shock from permitting-to-start conversion or weather timing could also put the print outside the interval.

Review disposition: accepted the leakage critique by removing the earlier same-target Thesis run as an input; the forecast now rests on official Census/HUD anchors, the release calendar, the persistence prior, explicit adjustment components, the threshold ladder, and the stated realized-dispersion interval method. Also accepted resolver clarity on the mutable current PDF and tied the multifamily rebound adjustment to the weak May 0.284 million 5+ unit print versus the March-April 2026 context.

Derived run: the pointwise median of the CDFs of 3 independent thesis.analyst fast rollouts on this target — median prediction sampling per Turtel et al. 2025 (arXiv:2505.17989). No new model call; this run is a deterministic aggregate of the recorded rollouts at 2026-07-08T02:53:21Z, 2026-07-08T02:53:35Z, 2026-07-08T02:57:35Z. Drivers and resolver fields mirror the rollout closest to the median.

Median CDF quantiles: q10 = 1.096, q50 = 1.237, q90 = 1.38. Constituent points [1.3, 1.237, 1.237] with 80% widths [0.34, 0.282, 0.282]; the median interval inherits the central rollout mass rather than averaging tails.