The NBER co-incident Recession Model (confirmation of last resort) | RecessionAlert — exploratory proxy campaign

article date

2012-07-18

fetched at

2026-08-22T23:39:30Z

native span

1959-07 through 2012-07 optimization window (580 months scored); 7 optimization vintages from Dec 1968; 1st vs 6th vintage correlation 0.90.

public boundary

Publish original summaries, source hashes, proxy provenance, inferred rules, aggregate metrics, and gaps only; omit cached HTML/media, credentials, cookies, nonce-bearing URLs, private data, and unverified native series.

publication lag

Monthly co-incident, at least 1-month lagging by construction; NBER proclamations lag up to 12 months; model signals up to 8 months before NBER; subscriber charts updated monthly in a separate report; exact release clock undisclosed.

redacted url

https://recessionalert.com/the-nber-recession-model-project/

sha256

sha256:a0a711709c28e199e85e0a6d79b9cafc44aafeef39d4df6ec183e3bf5dc8a682

title

The NBER co-incident Recession Model – confirmation of last resort | RecessionAlert

• The NBER examines 4 monthly co-incident indicators: Industrial Production; real personal income less transfers deflated by PCE; nonfarm payrolls; real retail sales deflated by CPI. Less weight to IP/retail as sectoral; GDP/GDI are broadest but quarterly via a separate GDPI Model.

• This monthly NBER Model plus the quarterly GDPI Model cover all NBER inputs as a confirmation of last resort (at least 1-month lagging), not real-time forecasting; models signal up to 8 months before NBER proclamations.

• IP uses a 6-month smoothed growth rate (Prof. Moore); income a 3-month smoothed growth; payrolls a 3-month smoothed growth less 0.415; retail sales a 12-month percent change not smoothed.

• Syndrome triggers IP -1.77%, income -0.25%, payroll +0.89%, retail +0.44% (AUC/ROC optimized); each breach counts one vote; index = 2 - votes; below zero calls recession.

• Weighted composite growth index: IP 17%, income 31%, payroll 30%, retail 12%, syndrome diffusion 10% (AUC/ROC optimized); below zero triggers a recession call.

• Six-factor Probit over the 4 growths plus syndrome diffusion plus weighted composite; above 0.10 calls recession in a timely way (2 historical false alarms); above 0.40 never false-alarmed but sacrifices 1-2 months lead.

• Seven optimization vintages from Dec 1968 perform comparably out of sample (1st vs 6th correlation 0.90); the most pessimistic vintage was not flagging recession at the article date.

• The 4 series are taken from FRED St Louis monthly updated charts; no series IDs, vintages, or download hashes are pinned on this page.

• The page supplies no Moore smoothing formula, PCE/CPI deflator pins, Probit coefficients, optimizer detail, release clock, or source-linked observation workbook; 12 chart GIFs are uncrawled with no asset SHA256.

• This campaign tests INDPRO/CPIAUCSL/UNRATE/RRSFS monthly proxies with SMA growth stand-ins, UNRATE-inverse payroll and INDPRO-adjacent income stand-ins, and SPY/IEF overlays; all transforms and action semantics are inferred.

• The NBER examines 4 monthly co-incident indicators: Industrial Production; real personal income less transfers deflated by PCE; nonfarm payrolls; real retail sales deflated by CPI. Less weight to IP/retail as sectoral; GDP/GDI are broadest but quarterly via a separate GDPI Model.

• This monthly NBER Model plus the quarterly GDPI Model cover all NBER inputs as a confirmation of last resort (at least 1-month lagging), not real-time forecasting; models signal up to 8 months before NBER proclamations.

• IP uses a 6-month smoothed growth rate (Prof. Moore); income a 3-month smoothed growth; payrolls a 3-month smoothed growth less 0.415; retail sales a 12-month percent change not smoothed.

• Syndrome triggers IP -1.77%, income -0.25%, payroll +0.89%, retail +0.44% (AUC/ROC optimized); each breach counts one vote; index = 2 - votes; below zero calls recession.

• Weighted composite growth index: IP 17%, income 31%, payroll 30%, retail 12%, syndrome diffusion 10% (AUC/ROC optimized); below zero triggers a recession call.

• Six-factor Probit over the 4 growths plus syndrome diffusion plus weighted composite; above 0.10 calls recession in a timely way (2 historical false alarms); above 0.40 never false-alarmed but sacrifices 1-2 months lead.

• Seven optimization vintages from Dec 1968 perform comparably out of sample (1st vs 6th correlation 0.90); the most pessimistic vintage was not flagging recession at the article date.

• The 4 series are taken from FRED St Louis monthly updated charts; no series IDs, vintages, or download hashes are pinned on this page.

• This campaign tests INDPRO/CPIAUCSL/UNRATE/RRSFS monthly proxies with SMA growth stand-ins, UNRATE-inverse payroll and INDPRO-adjacent income stand-ins, and SPY/IEF overlays; all transforms and action semantics are inferred.

• Can the disclosed NBER co-incident syndrome-vote / weighted-composite / Probit-threshold rules be turned into labeled public-proxy confirmation overlays when the Moore formula, deflator pins, and Probit coefficients are undisclosed?

canonical snapshot

data/recessionalert.sqlite

common complete rows

415

proxy database

data/market.sqlite

proxy inputs

• INDPRO

• CPIAUCSL

• UNRATE

• RRSFS

• SPY

• IEF

• BIL

proxy spans

BIL

2007-05 through 2026-07

CPIAUCSL

1947-01 through 2026-06

IEF

2002-07 through 2026-07

INDPRO

1919-01 through 2026-06

RRSFS

1992-01 through 2026-06

SPY

1993-01 through 2026-07

UNRATE

1948-01 through 2026-06

redistribution

Publish derived arrays, hashes, metadata, inferred specs, and aggregate metrics only; do not copy target HTML/media or private data.

revision policy

Use checked-in proxy revisions as stored; no real-time vintage archive is available. Latest-known month-end value is carried after inception; unknown source values remain unavailable.

source

https://recessionalert.com/the-nber-recession-model-project/

source fetched at

2026-08-22T23:39:30Z

source inventory

research/findings/recessionalert_inventory.md

source manifest

research/artifacts/recessionalert/manifest.json

source sha256

sha256:a0a711709c28e199e85e0a6d79b9cafc44aafeef39d4df6ec183e3bf5dc8a682

claim kind

inferred

execution

latest available monthly proxy value, then one or more available-bar lags; lag 0 sensitivity only

exploratory

True

family

syndrome-vote-proxy

id

syndrome_doc_lag1

lag bars

1

metrics

cagr

0.09105411174938238

max drawdown

-0.4277536198650529

observations

402

risk off events

14

risk on share

0.8768115942028986

sharpe

0.7432491542934887

volatility

0.12887687706515097

native fidelity

False

signal side

risk_on_when_index_at_or_above_zero

source provenance

fetched at

2026-08-22T23:39:30Z

sha256

sha256:a0a711709c28e199e85e0a6d79b9cafc44aafeef39d4df6ec183e3bf5dc8a682

url

https://recessionalert.com/the-nber-recession-model-project/

threshold

0.0

Compare syndrome_doc_lag1 against composite_doc_lag1 and probitlike_rare_lag1; vary lag 1/2 and strict/relaxed thresholds.

• priority

  high

  question

  Would ALFRED-vintage PAYEMS/W875RX1 pins rescue the payroll/income stand-ins and change the NO-GO native disposition?

• priority

  medium

  question

  Does the disclosed Moore smoothing formula shift 6m/3m timing enough to beat the SMA stand-in frontier?

• priority

  medium

  question

  Should co-incident confirmation be paired with a leading trigger given its late-by-design nature?

• iteration number

  1

  objective

  Execute labeled NBER co-incident proxy families and publish causal exploratory evidence while preserving source gaps.

  status

  completed

Useful as a documented confirmation overlay for research; not a forecasting edge and not promotable to Other/Indicators without native pins.

Syndrome AND composite is more defensive (risk-on 0.75) than either alone; OR resembles syndrome alone; neither claims Probit parity.

Risk-off months cluster around NBER recessions with at least one month of lag; the payroll stand-in barely fires and is reported as a negative result.

Lag 2 slightly beats lag 1; strict thresholds cut events without dominating Sharpe; lag-0 controls outperform causal lags as expected.

Syndrome/composite Sharpe decay from 1992-2007 (0.85/1.00) to 2008-2026 (0.66/0.75); the best-effort neighbor holds up better (0.93/0.87).

18 labeled variants across five families on 1992-01..2026-06 (402 overlay months); see 1589_trial_ledger.json.

Causal lag>=1 only; no costs modeled; confirmation arrives 1-2 months after ideal timing, mirroring the disclosed Probit tradeoff.

The UNRATE-inverse payroll stand-in almost never exceeds +0.89; the over-defensive composite bar sits out too much upside.

Monthly syndrome votes on proxy growths fire sparse risk-off months (14 events for the documented baseline); composite and Probit-stand-in variants modulate persistence. Labeled inferred, not native NBER.

Composite weighting and the relaxed best-effort threshold cut drawdown versus buy-hold while keeping most upside.

Scale mismatch between unemployment-change units and payroll-growth thresholds; co-incident confirmation lags turning points by construction.