Housing Starts Forecast for Japan and the United States | YHS

Yamaken Housing Starts Forecast

Housing Starts Forecast for Japan and the United States | YHS

YHS-FE / YHS-Wood

日本語
Data originsJapan Jul. 2026U.S. Jul. 2026Public research edition / updated after official releases

Independent research based on official statistics. Forecast values are not guarantees.

Japan Housing Market

Japan Jul. 2026 · monthly units

Latest Housing-Starts Forecast

SeriesTarget month / yearValue80% range / status
Official actual2026-0766,439 homesPublished official value
YHS-FE 1-month2026-0864,210 homes56,809–69,361 homes
YHS-FE 3-month2026-1069,258 homes59,881–75,108 homes
YHS-FE 6-month2027-0159,991 homes49,224–64,855 homes
YHS-FE 12-month2027-0768,611 homesEmphasize the range over the point estimate
YHS-FE FY2026FY2026747,928 homesYear over year +5.2%
YHS-FE FY2027FY2027758,930 homesYear over year +1.5%

This summary is provided for search and comparison. Sources, ranges, and limitations appear in the sections below.

Current Market Position

Housing-starts marketRecovering

Includes rebound from a low base

FY2026747,928

+5.2% YoY

Next Month+6.5%

forecast YoY

Six Months59,991

-9.7% vs. latest actual

Japan’s housing-starts market is recovering from a low base. YHS-FE places FY2026 at 747,928 homes, +5.2% year over year. The next-month forecast is +6.5% year over year, while the six-month point is 59,991 homes, -9.7% versus the latest actual. This looks more like stabilization and rebound than a broad economic boom.

This classifies the housing-starts market, not the broader economy. Fixed four-signal count: 3 positive, 1 negative, 0 neutral.

Official actuals: MLIT Building Starts Statistics via e-Stat

Current Reading

Japan recorded 66,439 housing starts in July 2026, including 39,547 wooden units. YHS-FE places the national total near 64,000 one month ahead, 69,000 at three months, and 60,000 at six months.

The Japanese model draws more information from the recent mix of owner-occupied, rental, and for-sale starts than from one leading indicator. The twelve-month result remains close to a simple persistence forecast.

Japan also has a separate wooden-housing branch and a 47-prefecture panel. Medium-term prefectural values remain research candidates with only one complete temporal holdout.

Market Momentum

Next-Month Forecast64,210

homes

Reference MoM-2.1%

YHS reference seasonal adjustment

Forecast YoY+6.5%

2026-08

Latest 3-Month Average+18.7%

year over year

FY2026 Landing747,928

+5.2% year over year

Forecast RevisionNo prior match

Forecast ledger starts with this release

Reference month-over-month growth uses YHS multiplicative month factors estimated only from the trailing 120 observations. It is not an official seasonal adjustment. Same-target forecast revisions and realized surprises start accumulating with this release.

Latest Contributions and Forecast Risk

Contributions by Use

Owner occupied
+0.1pt
Rental
+4.5pt
For sale
+3.8pt
Employer housing
-0.2pt

Contribution to the latest actual year-over-year change

Year-over-Year Trend by Three Main Uses

Owner occupied +0.4%Rental +10.0%For sale +14.6%
-40%-20%+0%+20%+40%Owner occupied 2026-07: +0.4%Rental 2026-07: +10.0%For sale 2026-07: +14.6%2025-082026-022026-07

Latest 12 months of official actuals. Employer housing is excluded because its small count makes percentage changes unusually volatile.

Downside Frequency and Regional Breadth

Empirical frequency below one year earlier26.3%n=319
80％ forecast range56,809–69,361homes

Prefectures above one year earlier

  • 1 months35/4774%
  • 3 months30/4764%
  • 6 months28/4760%
  • 12 months21/4745%

This is an empirical backtest frequency, not a calibrated or guaranteed probability.

Financial-Conditions Monitor

10-Year JGB2.77%

+12bp month over month

Flat 35 Minimum Rate3.46%

+17bp month over month

Short-Run Model StatusMonitor / not adopted

No stable significant gain

Sources：Japan MOF JGB yields／Japan Housing Finance Agency Flat 35 rate history

National Housing Starts

YHS-FE

monthly units
ActualYHS theoretical / forecast80% forecast range

RMSE Reduction vs. Same Month One Year Earlier

1 months
38.4%Error-weighted ensemble
3 months
20.0%Error-weighted ensemble
6 months
11.7%Error-weighted ensemble
12 months
0.0%Latest observed value

A larger reduction means a smaller historical error than the seasonal-naive benchmark.

Forecast Values

1-month horizon · 2026-0864,210 homes

80% forecast range: 56,809–69,361

3-month horizon · 2026-1069,258 homes

80% forecast range: 59,881–75,108

6-month horizon · 2027-0159,991 homes

80% forecast range: 49,224–64,855

12-month horizon · 2027-0768,611 homes

80% forecast range: 55,950–74,561

Forecast-versus-Actual Scorecard

2026-07 ・ 1-month horizon

Latest Forecast Check

0.40%absolute error rate
YHS-FE forecast
66,175
First observed actual
66,439
Difference264 homes

264 homes below the actual

1 published forecast(s) have a realized target so far. Both bars share a zero-based scale, and the forecast remains fixed at its published value.

Forecast dateHorizonTargetForecastFirst observed actualForecast minus actualAbsolute error rateStatus
2026-09-0212 months2027-0768,611———Pending
2026-09-026 months2027-0159,991———Pending
2026-09-023 months2026-1069,258———Pending
2026-09-021 months2026-0864,210———Pending
2026-08-2112 months2027-0668,822———Pending
2026-08-216 months2026-1264,428———Pending
2026-08-213 months2026-0965,959———Pending
2026-08-211 months2026-0766,17566,439-2640.40%Realized

Published forecasts remain fixed. The first official value seen by the monthly pipeline is stored once; later revisions are tracked separately. homes

Fiscal-Year Housing Starts

ActualYHS-FE
FY2026 YHS-FE747,928 homes

year over year +5.2% · May 2026 origin

FY2027 YHS-FE758,930 homes

year over year +1.5% · May 2026 origin

FY2021–FY2025 are actual totals. FY2026 and FY2027 are YHS-FE forecasts with a May 2026 origin. Year-over-year rates use the preceding value within the same series.

Data Foundation

Official monthly series738 months

1965-01–2026-06

YHS-FE1 / 3 / 6 / 12

Direct horizon-specific forecasts

Forecast ledgerNo overwrite

Origin, forecast, actual, and error archive

YHS Methodology

YHS Model Construction and Validation

Equations, backtests, publication-time comparisons, and rejected extensions.

Methodology Article →

National YHS-Wood

Japan Jul. 2026

Forecast Values

National YHS-Wood · 1-month horizon38,731 homes

2026-08

National YHS-Wood · 3-month horizon41,045 homes

2026-10

National YHS-Wood · 6-month horizon32,986 homes

2027-01

National YHS-Wood · 12-month horizon39,547 homes

2027-07

National YHS-Wood

monthly units
ActualYHS theoretical / forecast

Validation Accuracy

HorizonRMSEBest Model
1 months3,256AR(12)
3 months3,427AR(12)
6 months3,369Seasonal naive
12 months3,532Latest value
Lumber-Statistics Rejection

Short-horizon wooden-housing forecasts for Japan. Lumber production, shipments, and inventories are not part of the adopted model.

Prefectural YHS-Wood

Prefectural Forecast Distribution

100%
OkinawaHokkai DoAomori KenIwate KenMiyagi KenAkita KenYamagata KenFukushima KenIbaraki KenTochigi KenGunma KenSaitama KenChiba KenTokyo ToKanagawa KenNiigata KenToyama KenIshikawa KenFukui KenYamanashi KenNagano KenGifu KenShizuoka KenAichi KenMie KenShiga KenKyoto FuOsaka FuHyogo KenNara KenWakayama KenTottori KenShimane KenOkayama KenHiroshima KenYamaguchi KenTokushima KenKagawa KenEhime KenKochi KenFukuoka KenSaga KenNagasaki KenKumamoto KenOita KenMiyazaki KenKagoshima KenOkinawa Ken

Short-horizon colors are relative categories across prefectures. Select a prefecture to show its exact value above and in the chart below. Use the controls, mouse wheel, or pinch to zoom, and drag to pan. Boundaries: simplified JapanPrefGeoJson (Public Domain).

Short-Horizon Forecast

monthly units

YoY compares each forecast with official wooden starts in the same prefecture twelve months earlier.

2023 Actual and 2024–2028 Average

Demand-and-stock specification / one temporal holdout

Forecasts for All 47 Prefectures

Prefecture1 month3 months6 months12 months2024–2028 averagevs. 2023
1,1351,2378401,11314,810+5.1%
3003132032954,178-1.5%
3323683143244,873-11.5%
81792865188510,302-11.2%
2142351772152,982+4.0%
3433352383303,749-1.6%
4915934244956,409-3.7%
97496581699011,358-6.2%
6256115486217,288-2.6%
6326925526437,580-6.3%
3,2333,4162,7463,39630,779-12.1%
2,5692,6312,2372,68424,693-9.3%
3,9104,1343,4464,04544,104+3.8%
3,0983,3702,7373,06433,733-3.3%
6146784686047,249-2.6%
3373462613454,012-3.9%
4294703704514,218-0.3%
2242612092383,037-4.5%
2663152232763,081+1.4%
6947285717067,513-9.2%
5926375326046,884-3.2%
1,1041,1189061,10211,838-3.7%
2,8462,9472,4752,93427,270-9.9%
4865264335016,062-0.2%
4755203874895,832+0.7%
6587195436767,604+17.2%
2,4202,6402,1392,55928,051+2.9%
1,4381,5291,1751,46716,627+3.5%
3203652813334,386+9.0%
2872422102663,218-0.4%
1401661331491,992+9.6%
1991881691902,477+0.9%
5505754965646,592+0.2%
7087746087059,838-1.9%
3603603383364,593-1.2%
1691791541702,403+3.9%
2712782102783,729-2.8%
3623442903384,726-0.2%
1641651321592,095+14.1%
1,6661,6761,3801,71818,607-1.2%
2762872382723,669-8.1%
2953402602994,466-1.7%
5716144935907,182-8.9%
2733002362664,034-1.0%
2823222552924,022+2.8%
4144413523965,826-4.8%
1651681331743,088+69.7%

Short-horizon figures are monthly wooden starts; medium-term figures are annual averages. Select a prefecture name to update the chart above.

Research candidate

The prefectural panel beat the seasonal benchmark at every tested horizon. At twelve months, RMSE improved by 5.3%, but sMAPE was 21.0% over only 20 target months. Accuracy is lower than at one month, and medium-term values have only one independent temporal holdout.

Japan Medium- and Long-Term Scenarios

Japan only

National Conditional Wooden-Housing Scenarios

BaselineHigher ratesFaster household decline
2024–2028Anchor443,054 homes / year
2030homes / year
  • Baseline446,460 (+0.8%)
  • Higher rates424,138 (-4.3%)
  • Faster household decline440,217 (-0.6%)
2035homes / year
  • Baseline442,413 (-0.1%)
  • Higher rates420,292 (-5.1%)
  • Faster household decline430,788 (-2.8%)
2040homes / year
  • Baseline432,733 (-2.3%)
  • Higher rates411,097 (-7.2%)
  • Faster household decline413,400 (-6.7%)
Conditional Scenarios

These wooden-housing scenarios anchor to the 2024–2028 annual average and apply the 2024 IPSS prefectural household projection. They are conditional sensitivities, not validated long-term point forecasts. The higher-rate case is a fixed 5% downside stress. The faster-decline case doubles household losses versus 2025. Neither is an estimated causal effect.

Source: IPSS Household Projections by Prefecture, 2024 projection

Prefectural Conditional Scenarios

2023 Actual and 2024–2028 Average

Prefectural Conditional Scenarios

YearBaselineHigher ratesFaster household decline

Unit: annual wooden housing starts. Parentheses show change versus the 2024–2028 average.

United States Housing Market

U.S. Jul. 2026 · thousand SAAR

Latest Housing-Starts Forecast

SeriesTarget month / yearValue80% range / status
Official actual2026-071,239 thousand SAARPublished official value
YHS-FE 1-month2026-081,274 thousand SAAR1,153–1,370 thousand SAAR
YHS-FE 3-month2026-101,272 thousand SAAR1,114–1,384 thousand SAAR
YHS-FE 6-month2027-011,283 thousand SAAR1,031–1,439 thousand SAAR
YHS-FE 12-month2027-071,265 thousand SAAR829–1,474 thousand SAAR

U.S. values are seasonally adjusted annual rates.

Current Market Position

Housing-starts marketSoft / Stagnant

Leading data provide partial support

2026-07 actual-12.4%

month over month

2026-07 actual-13.5%

year over year

2026-07 permits+3.1%

year over year

The U.S. housing market remains soft under high financing costs. The latest starts reading was -12.4% month over month and -13.5% year over year, while permits were +3.1% year over year. YHS-FE places the next month at 1,275 thousand SAAR, but -1.3% year over year does not yet signal a clear upswing.

This classifies the housing-starts market, not the broader economy. Fixed four-signal count: 1 positive, 3 negative, 0 neutral.

Official actuals: U.S. Census Bureau, New Residential Construction

Current Reading

From the July 2026 release, YHS-FE places U.S. housing starts near 1.274 million SAAR one month ahead, 1.272 million at three months, and 1.283 million at six months.

Permits were the clearest short-run signal in the historical test. The wider supply-pipeline model improved the one-month RMSE again, but the advantage did not survive at three, six, or twelve months.

The strict first-release-vintage backtest ends in August 2013. I update the current model each month, but I do not describe the post-2013 period as an equivalent real-time accuracy record.

Market Momentum

Next-Month Forecast1,275

thousand SAAR

Forecast MoM+2.9%

seasonally adjusted annual rate

Forecast YoY-1.3%

2026-08

Latest 3-Month Average-10.4%

vs. previous three months

Six-Month Pace1,283

thousand SAAR

Forecast Revision+0.0%

vs. prior forecast for the same target month

U.S. indicators use the official seasonally adjusted annual rate. Same-target forecast revisions and realized surprises start accumulating with this release.

Latest Contributions and Forecast Risk

Contributions by Housing Type

Single family
-10.5pt
Multifamily / other
-2.9pt

Contribution to the latest actual year-over-year change

Downside Frequency and Signal Breadth

Empirical frequency below one year earlier60.8%n=255
80％ forecast range1,153–1,370thousand SAAR

Starts-and-permits breadth

  • Four core series2/450%

Mean absolute official revision 30.6 thousand SAAR

This is an empirical backtest frequency, not a calibrated or guaranteed probability.

Financial-Conditions Monitor

30-Year Fixed Mortgage6.65%

2026-08-20

Weekly Change-2bp

Freddie Mac PMMS

YHS-FE StatusMonitor / not adopted

Not tested in the current model

Source：Freddie Mac PMMS via FRED MORTGAGE30US

National Housing Starts

YHS-FE

SAAR
ActualYHS theoretical / forecast80% forecast range

RMSE Reduction vs. Same Month One Year Earlier

1 months
61.9%Permit-based ridge
3 months
49.5%Permit-based ridge
6 months
30.3%Latest observed value
12 months
0.2%Error-weighted ensemble

A larger reduction means a smaller historical error than the seasonal-naive benchmark.

Forecast Values

1-month horizon · 2026-081,274 thousand SAAR

80% forecast range: 1,153–1,370

3-month horizon · 2026-101,272 thousand SAAR

80% forecast range: 1,114–1,384

6-month horizon · 2027-011,283 thousand SAAR

80% forecast range: 1,031–1,439

12-month horizon · 2027-071,265 thousand SAAR

80% forecast range: 829–1,474

Forecast-versus-Actual Scorecard

Forecast dateHorizonTargetForecastFirst observed actualForecast minus actualAbsolute error rateStatus
2026-09-0212 months2027-071,265———Pending
2026-09-026 months2027-011,283———Pending
2026-09-023 months2026-101,272———Pending
2026-09-021 months2026-081,275———Pending
2026-08-2112 months2027-071,265———Pending
2026-08-216 months2027-011,283———Pending
2026-08-213 months2026-101,272———Pending
2026-08-211 months2026-081,275———Pending

Published forecasts remain fixed. The first official value seen by the monthly pipeline is stored once; later revisions are tracked separately. thousand SAAR

YHS Methodology

U.S. Model Construction and Validation

Equations, first-release backtests, permit signals, and rejected extensions.

Methodology Article →

Permits and Supply Pipeline

U.S. Jul. 2026

Forecast Values

1-month horizon · 2026-081,400 thousand SAAR

80% forecast range: 1,304–1,507

3-month horizon · 2026-101,319 thousand SAAR

80% forecast range: 1,191–1,486

6-month horizon · 2027-011,317 thousand SAAR

80% forecast range: 1,092–1,534

12-month horizon · 2027-071,356 thousand SAAR

80% forecast range: 999–1,694

Incremental Pipeline Value

HorizonPipeline RMSEComparatorComparator RMSEResult
1 months83.9Permits86.83.3% lower
3 months117.1Permits115.41.5% higher
6 months182.2Latest value160.313.7% higher
12 months267.9Latest value232.815.1% higher

Unit: thousands, seasonally adjusted annual rate. Lower RMSE indicates smaller forecast error.

One-Month Research Candidate

The pipeline model beat the permit model by 3.3% at one month, then deteriorated at three, six, and twelve months. It is not adopted across all horizons.

U.S. Validation Boundary

First-release and current-series separation

First-release backtest387 months

1981-06–2013-08

Current revised series811 months

1959-01–2026-07

Nonlinear residual layer7.0% worse

Rejected even at one month

The 2013 Validation Boundary

The strict U.S. accuracy record ends in August 2013. Current series through July 2026 support operation, but they are not presented as an equivalent vintage-aware accuracy record.

YHS Methodology

U.S. Validation Method

Publication-time boundaries, equations, backtests, and rejected extensions.

Methodology Article →
Terms and Reading Guide
RMSE
Forecast error measure that penalizes large misses more heavily. Lower is better.
sMAPE
A symmetric percentage error scaled by the forecast and actual values.
SAAR
Seasonally Adjusted Annual Rate, the monthly pace expressed at an annual rate.
Reference MoM
A YHS display transformation using month factors estimated from Japan’s trailing 120 observations. It is not an official seasonal adjustment.
Forecast revision
The change from a previously stored forecast for the same target month. Different target months are not compared.
Empirical frequency
The share of historical backtest residuals meeting the current condition. It is not a guaranteed or calibrated probability.
Breadth indicator
Japan uses the share of prefectures above one year earlier; the U.S. uses the share of four starts-and-permits signals above one year earlier.
80% forecast range
An empirical range based on the 10th and 90th percentiles of historical backtest residuals.
Research edition v1.5 / values, links, and display revalidated after official updates

カテゴリー

アーカイブ