Industrial Production Nowcast

Forward 3-month and 6-month INDPRO YoY nowcast from an 8-input composite of leading production indicators (manufacturing payrolls, capacity utilization, manufacturing hours, new orders ex-aircraft, truck tonnage, WEI, initial claims, inventory-to-sales) calibrated by 60-month rolling-origin OLS. Same architecture as the Inflation and Wage Nowcasts. Updated .

Current reading: composite at σ (mildly above long-run mean). INDPRO YoY currently % as of ; rolling-origin OLS calibrated on the most recent 60 months projects forward production growth at % (3m ahead) and % (6m ahead).

Composite vs realized INDPRO YoY

The composite z (cyan) is the equal-weighted z-mean of 8 production-leading indicators. The INDPRO YoY series (gold) is the realized target. Composite turning points lead INDPRO turning points, particularly visible in the 2008-09 collapse, the 2014-16 manufacturing recession, the 2020 COVID shock, and the 2022-23 mid-cycle slowdown. Note the post-2020 era: composite recovered strongly but INDPRO has stayed below trend — the post-pandemic supply-chain and goods-vs-services rotation has decoupled lead indicators from realized output more than any prior cycle on this sample.

Rolling-origin nowcast vs realized (h=3m)

Rolling-origin OOS at h=3m. Aggregate metrics: corr , R² , MAE pp, bias

pp. Pre-2020 corr ; post-2020 corr . MAE caveat: INDPRO YoY is a much higher-variance target than CPI or wage growth — it swings ±10% across cycles, so a ~1.9 pp MAE at h=3m is reasonable in absolute terms but means the nowcast confidence interval is wider in dollars-of-output than it is in pp.

Components at the latest date

Each lens contributes one z-scored input. Manufacturing payrolls YoY and capacity utilization are the two highest-magnitude signals individually (h=3m corr +0.43 and +0.58). New orders ex-aircraft is the cleanest forward-demand read — among the few signals whose post-2020 correlation is HIGHER than its pre-2020 correlation. Truck tonnage and WEI add real-time freight and broader-economy pulse. Initial claims (inverted) picks up labor-side demand contractions early. Inventory-to-sales (inverted) reflects the production-pull from depleted inventories.

Methodology

Composite construction. z_full = mean of 8 expanding-window z-scores. Each input is z-scored using only data available at each month (no full-sample peeking). Sign convention is set so HIGH = production-acceleration signal (initial claims and inventory-to-sales are flipped where appropriate).

Calibration. 60-month rolling-origin OLS — same architecture as the Inflation Nowcast, Wage Nowcast, and CCMI v2. At each historical month, the model is fit on the prior 60 months of (z, forward_indpro_yoy) pairs and used to predict production growth at t+h. The "current" nowcast at the page top uses OLS fit on the most recent 60 months for the latest composite reading.

Why z_full and not z_orders. The Phase 0 probe (scripts/indpro_nowcast_phase0.py) tested four composites — z_full (8 inputs), z_real (3-input hard-data subset), z_orders (3-input forward-demand subset), z_labor (2-input labor subset). z_orders had the highest post-2020 correlation (+0.52) but lower pre-2020 (+0.51) and a much smaller pre-2020 input window. z_full's full-sample correlation (+0.67 at h=3m) is materially higher and the 60-month rolling refit handles the regime drift z_full surfaces. Same Occam principle as inflation/wage: simplest model that maximizes OOS skill across the whole sample, not just one regime.

Why h=3m and h=6m, not h=12m. The probe showed h=12m FAILS — the post-2020 correlation flips negative for almost every signal at the 12-month horizon. Production growth a year out depends more on the credit cycle than on real-economy lead indicators, so we ship the two horizons where the signal is robust and stop there.

Validation summary (from scripts/indpro_nowcast_phase0.py):

Rolling-origin OOS corr (h=3m): , n=
Rolling-origin OOS corr (h=6m): , n=
Pre/post 2020 stability: real but weakened (corr_pre +0.83 → corr_post +0.49 at h=3m); supply-chain dislocation post-pandemic decoupled lead indicators from output more than any prior cycle in the sample
Bias near zero at both horizons; MAE 1.9 pp (h=3m) and 2.4 pp (h=6m) — large in absolute terms because INDPRO YoY itself is high-variance

Sources.

Live track record. Every prediction this page emits is recorded in the Calibration Ledger at the moment it's made and graded against the realized INDPRO YoY when it lands.