Triple-model GDP nowcast ensemble: (A) 6-var simplified bridge (baseline), (B) 13-var extended bridge (ISM + credit spreads + real rates + sentiment + permits + SPX 6M), (C) Atlanta Fed GDPNow. Weighted by historical RMSE (20/45/35). Model B coefficients OLS-calibrated on 1985Q1 — 2025Q4, N=48, R²=0.751, OOS RMSE 2.72pp.
2026Q1T-6d
Second
2026-05-29
2026Q1T-34d
Third
2026-06-26
2026Q2T-68d
Advance
2026-07-30
BEA publishes 3 estimates per quarter: Advance (~30d after qtr-end) → Second (+30d) → Third (+30d). Our ensemble nowcast tracks each release.
A. Simple bridge (6 vars)
+4.50%
Baseline
B. Extended bridge (12 vars)
+2.86%
High-fidelity
C. GDPNow (Atlanta)
+4.26%
External baseline
★ Ensemble
+3.68%
low agreement
95% confidence interval (historical out-of-sample RMSE)2.01% ↔ 5.34%
σ ≈ 0.85pp · model dispersion 1.64pp
Ensemble nowcast for 2026Q2: +3.68% SAAR (95% CI 2.01% — 5.34%). Of 12 leading indicators, 7 support growth, 5 oppose, 1 neutral. Top positive contributions: High-Yield OAS (gap from 4.5% median) (+0.85pp), Industrial Production YoY (+0.80pp). Top drags: Building permits YoY (leads housing ~12mo) (-0.00pp), Initial jobless claims (gap from 250k neutral) (-0.02pp). Significant model disagreement (dispersion 1.64pp) — monitor ISM/employment/credit early signals.
12-variable current readings & contribution
High-Yield OAS (gap from 4.5% median)pro
2.78· Pct14· z -0.99+0.85pp
HY OAS 2.78% very tight
Industrial Production YoYpro
1.35· Pct66· z +0.29+0.80pp
IndPro +1.4% mild expansion
Retail Sales YoY (consumption ~70% of GDP)pro
4.87· Pct67· z -0.02+0.68pp
Retail +4.9% moderate
S&P 500 6-month return (leads GDP 6-9 mo)pro
14.29· Pct71· z +0.90+0.41pp
SPX +14.3% positive
ISM Manufacturing PMIanti
48.50· Pct2· z -1.56+0.30pp
PMI 48.5 mild contraction
Nonfarm payrolls 3-mo changeneutral
144.00+0.04pp
+144k slowing
10Y-3M curve (inverted = recession warning)pro
0.88+0.00pp
Curve 0.88% normal
Building permits YoY (leads housing ~12mo)anti
-0.21· Pct32· z -0.34-0.00pp
Permits -0.2% contracting
Initial jobless claims (gap from 250k neutral)pro
209,000· Pct18· z -0.55-0.02pp
Claims 209k very low
ISM New Orders (most leading sub-index)anti
47.20· Pct2· z -1.69-0.17pp
New orders 47.2 contracting
10Y Real Rate (TIPS, gap from 0.5%)anti
2.18· Pct99· z +2.14-0.74pp
Real rate 2.18% very restrictive
UMich Consumer Sentimentanti
49.80· Pct1· z -2.05-1.39pp
Sentiment 49.8 weak
Chicago Fed NFCIpro
-0.52· Pct20· z -0.81-1.50pp
NFCI -0.52 loose
📊 Model methodology & calibrated coefficients
Method: OLS via numpy.linalg.lstsq · Target: GDPC1 YoY % (year-over-year real GDP growth) · Period: 1985Q1 — 2025Q4
OLS regression coefficients (* p<0.05, ** p<0.01, *** p<0.001):
| Variable | β | SE | t | p | |
|---|
| 趋势 (intercept) | +3.5930 | 0.887 | 4.05 | 0.0001 | *** |
| ISM PMI gap from 50 | -0.2005 | 0.116 | -1.73 | 0.0834 | |
| ISM New Orders gap from 50 | +0.0605 | 0.069 | 0.88 | 0.3784 | |
| Industrial Production YoY | +0.5878 | 0.098 | 6.00 | 0.0000 | *** |
| Retail Sales YoY | +0.1385 | 0.066 | 2.10 | 0.0354 | * |
| Payrolls 3-mo change (M) | +0.3036 | 0.122 | 2.49 | 0.0129 | * |
| 10Y-3M curve inverted (1=yes) | +0.0000 | — | — | — | |
| Claims gap from 250k (units of 50k) | -0.0211 | 0.023 | -0.93 | 0.3518 | |
| NFCI level | +3.4473 | 1.819 | 1.89 | 0.0581 | |
| HY OAS gap from 4.5% | -0.4962 | 0.479 | -1.04 | 0.3000 | |
| Real rate gap from 0.5% | -0.4386 | 1.584 | -0.28 | 0.7819 | |
| Sentiment gap from 85 (/20) | +0.0395 | 0.023 | 1.73 | 0.0838 | |
| Building Permits YoY | +0.0111 | 0.024 | 0.45 | 0.6494 | |
| SP500 6-month return % | +0.0289 | 0.054 | 0.53 | 0.5951 | |
Model B Extended bridge equation (coefficients approximated via OLS on 1985-2024 quarterly GDP YoY):
- β₀ = 1.95 (trend potential growth)
- β₁ × (ISM_PMI − 50) — manufacturing momentum
- β₂ × (ISM New Orders − 50) — most leading sub-index
- β₃ × IndPro YoY
- β₄ × Retail Sales YoY (consumption ~70% of GDP)
- β₅ × Payrolls 3-mo Δ (millions)
- β₆ × 1{10Y-3M inverted} → −0.85pp
- β₇ × (250k − ICSA) / 50k
- β₈ × NFCI (tight financial conditions: −0.40pp)
- β₉ × (HY OAS − 4.5%) — credit spread tilt
- β₁₀ × (Real rate − 0.5%) — restriction level
- β₁₁ × (UMich sentiment − 85) / 20
- β₁₂ × Building permits YoY (leads housing ~12 mo)
Each component clipped to [±2.5pp]; total clipped to [−6, +8]. 95% CI based on historical out-of-sample RMSE ≈ 0.85pp.