Stagflation stress study — 1968–1983

Monthly reconstruction, generated 2026-09-18T16:13:59+00:00 · Live dashboard · Regime analog · v1 · v2 · v3

PROXY — NOT THE PRODUCTION SYSTEM. The production monitor's series do not exist in this era. Everything below is a monthly proxy reconstruction for stress testing the portfolio, not a validation of the live signal.

Episodes ≥15% (daily closes)

1973–74: credit leg confirmed

20 mo after the peak

1973–74 sim vs equity

+24.2% vs -36.3%

Worst Phase-1 sim max drawdown

-7.8%

Phase 1 (proxy) Armed (proxy) Phase 2 (proxy) ≥15% drawdown episode (numbered) S&P 500 monthly close (log scale)

Drawdown episodes (≥15% on daily closes)

#	Peak	Trough	Depth	Months	Proxy state at peak month	Credit leg first fired
1	1968-11-29	1970-05-26	-36.1%	18	PHASE_1	1968-11 (+0 mo)
2	1973-01-11	1974-10-03	-48.2%	21	PHASE_1	1974-09 (+20 mo)
3	1976-09-21	1978-03-06	-19.4%	18	PHASE_1	never in episode
4	1980-02-13	1980-03-27	-17.1%	1	PHASE_2_PROXY	1980-03 (+1 mo)
5	1980-11-28	1982-08-12	-27.1%	21	ARMED	1980-11 (+0 mo)

Episode detection runs on daily ^GSPC closes (a monthly-close rule would miss the five-week 1980 break entirely); the state machine and simulation run monthly. The brief named four episodes; the 1968–70 bear (−36%) also falls inside the window and is reported.

Portfolio survival simulation (monthly rebalanced)

Episode	Phase 1 sim: total	Phase 1 sim: max DD	60/40: total	60/40: max DD	100% equity: total	100% equity: max DD
1. 1968-11 → 1970-05 (-36%)	-3.2%	-5.6%	-22.6%	-22.6%	-29.4%	-29.4%
2. 1973-01 → 1974-10 (-48%)	+24.2%	-7.8%	-23.7%	-31.0%	-36.3%	-45.2%
3. 1976-09 → 1978-03 (-19%)	+8.4%	-1.6%	-7.0%	-12.8%	-15.2%	-19.0%
4. 1980-02 → 1980-03 (-17%)	-4.7%	-4.7%	-7.0%	-7.0%	-10.2%	-10.2%
5. 1980-11 → 1982-08 (-27%)	+1.0%	-6.2%	-2.2%	-16.1%	-15.0%	-23.8%

1. 1968-11 → 1970-05 (-36%)

Phase 1 sim60/40100% equity

2. 1973-01 → 1974-10 (-48%)

Phase 1 sim60/40100% equity

3. 1976-09 → 1978-03 (-19%)

Phase 1 sim60/40100% equity

4. 1980-02 → 1980-03 (-17%)

Phase 1 sim60/40100% equity

5. 1980-11 → 1982-08 (-27%)

Phase 1 sim60/40100% equity

Proxy state machine (monthly, v1-analogous)

Gates: credit — Baa–GS10 spread ≥ its trailing 36-month 80% percentile OR a 6-month widening ≥ +75 bps; yield — GS10 ≥ its trailing 36-month 90% percentile. Reference windows exclude the current month.
Confirmations (quorum ≥ 2 of 4): UNRATE 3-month change ≥ +0.4pp; ICSA (weekly, averaged to monthly) ≥ trailing 12-month min + 25,000; oil 6-month return ≤ -20%; gold/copper ratio ≥ 1.15× its trailing 24-month mean.
States: red = any gate + quorum; amber = yield gate without quorum; green otherwise. No latches — this proxy is deliberately simpler than v3.

Simulation approximations (every one of them)

Cash 36% = FEDFUNDS/12 per month (no intra-month compounding).
Duration 8% = constant-10-year bond: GS10/12 carry minus 10× the monthly yield change. No convexity, no roll-down, no coupon curve — a deliberately rough long-bond fund stand-in.
TIPS 4% = bills + realized CPI pass-through (TIPS did not exist until 1997; this assumes a perfect inflation hedge with zero real-yield risk — generous).
Equities 27% = 0.9 × S&P 500 price return (COWZ/XLV/GRID did not exist).
Gold 15% = monthly London fix returns. Real assets 10% = 50/50 oil/copper returns (no roll yield, no futures curve — spot proxies).
Monthly rebalancing, no transaction costs or taxes. Benchmarks: 60/40 (S&P + the same duration proxy) and 100% equity, both price-only.
Publication lags ignored: CPI, UNRATE and claims are treated as known in their reference month; in real time they arrive weeks later.
FRED history as it stands today (no vintage adjustment).

Data sources & substitutions

Leg	Series used	Substitution notes
Credit spread	FRED BAA − GS10	No high-yield OAS exists before 1986/1996; Baa is investment grade and runs tighter, so the credit proxy is conservative by construction. AAA is fetched for reference (BAA−AAA inspected; highly correlated).
Oil	FRED WTISPLC (WTI spot)	The requested Brent series (POILBREUSDM) starts 1992 and DCOILBRENTEU starts 1987; WTI spot (monthly, from 1946) stands in. WTI traded at a small premium to Brent in this era; returns are close.
Copper	FRED WPU102502 (PPI copper)	The IMF copper series (PCOPPUSDM) starts 1992; the PPI copper index (from 1953) stands in. It is an index, not $/ton — only its returns are used.
Gold	Bundesbank monthly London fix via datahub CSV	FRED removed the London gold fixing series (GOLDAMGBD228NLBM) in 2020 and Stooq/Nasdaq Data Link are bot-walled. Note: gold was still administered (≈$35–42/oz) until the 1971–73 float, so the gold sleeve is nearly flat before that — historically accurate, not a bug.
Equities	yfinance ^GSPC	Price-only, no dividends. This understates equity and 60/40 returns (1970s dividend yields ran 3–5%), i.e. it flatters the Phase-1 comparison.

Honest conclusion. In 1973–74 the credit proxy confirmed ~10 months after the equity peak; long Treasuries lost money while equities halved; only gold and carry preserved capital. This is the failure mode the 8% duration cap, 5.40% invalidation, and 40% carry/TIPS/gold block exist to price. The monitor is not expected to time stagflation peaks — nothing does — the portfolio is built to survive them.

Reconstruction note (computed, this page): on the proxy parameters above, the credit leg's first fire in 1973–74 prints 1974-09 — 20 mo after the Jan-1973 peak; the “~10 months” figure is parameterization-dependent. The sequencing point is unchanged: credit confirmed long after the peak, and only after much of the decline had already happened.