Objective
Independently replicate the distinct RecessionAlert inventory row A Recession Fear Indicator from the checked-in canonical snapshot without a live-site request, credentials, private account data, or canonical crawl-database mutation. The source is incomplete, so this run performs the required labeled best-effort proxy exploration rather than stopping at a gap-only disposition.
Approach and provenance
The canonical page is https://recessionalert.com/a-recession-fear-indicator/, selected from research/artifacts/recessionalert/manifest.json, research/findings/recessionalert_inventory.md, and read-only data/recessionalert.sqlite. Its full SHA256 is sha256:2cb8464ee21e8b28af0f8dcbf7bacbd3ddeffc040c53717a198c0b701fa019dd; it is 83,313 bytes and was fetched at 2026-08-22T23:39:13Z. The inventory intentionally exposes only the digest prefix. The crawl log has five same-URL bodies, while the manifest selects the cited final body. Referenced GIFs have no target-page asset rows and are not redistributed.
The campaign reads only tracked public data from data/market.sqlite through SQLite mode=ro: FRED UNRATE monthly observations from 1948-01 through 2026-06, Yahoo SPY from 1993-01 through 2026-07, IEF from 2002-07 through 2026-07, and BIL from 2007-05 through 2026-07, reduced to observed month-end. VIXCLS, USREC, and the native Google Trends series are absent from the cache; no live fetch was authorized.
Disclosed rule and causal boundary
The March 6, 2012 article describes global Google search volume for the term “recession” as a promising recession indicator and fear gauge. It claims a peak within a two-week lag of the 2008 NBER recession start, mentions an August 2011 search-volume spike near a forming S&P 500 bottom, and describes a subsequent SuperIndex fall. These are contextual statements, not a buy/sell/hold algorithm.
The source does not disclose query geography/category/normalization, machine-readable observations, native first/last dates, formula, lookback, smoothing, threshold, cadence, release clock, vintage, state side, target, sizing, rebalance, holding period, costs, slippage, exit, or re-entry. The article-to-archive delay is not a publication lag, and the two-week NBER phrase is an event comparison rather than a recurring release policy.
For the inferred proxy, one available monthly bar after publication is the causal baseline. Lag 0 is retained only as research-only same-close sensitivity; lag 2 is conservative sensitivity. Warm-up and unavailable observations remain unavailable and are not converted to risk-on.
Best-effort proxy results
The inferred best guess is monthly UNRATE.pct_change(6) >= 0.0 shifted one available month. Six nearby labeled variants were executed on the longest local history:
| Variant | Lookback / threshold / lag | Classification | IEF total vs SPY | BIL total vs SPY |
|---|---|---|---|---|
...lb3.lag0.t0p0 | 3 / 0.0 / 0 | research-only lookahead | 9.1922 vs 11.3974 | 5.3567 vs 5.7754 |
...lb6.lag1.t0p0 | 6 / 0.0 / 1 | causal best guess | 6.3458 vs 11.3974 | 3.2554 vs 5.7754 |
...lb12.lag2.t0p5 | 12 / 0.5 / 2 | conservative sensitivity | 6.7471 vs 11.3974 | 3.0411 vs 5.7754 |
...lb6.lag1.t0p5 | 6 / 0.5 / 1 | threshold neighbor | 6.3458 vs 11.3974 | 3.2554 vs 5.7754 |
...lb12.lag1.t1p0 | 12 / 1.0 / 1 | higher-threshold sensitivity | 6.8259 vs 11.3974 | 3.0373 vs 5.7754 |
...lb21.lag1.tm0p5 | 21 / -0.5 / 1 | negative long-lookback | 3.9215 vs 11.3974 | 1.7476 vs 5.7754 |
IEF comparisons span 2002-08-31 through 2026-06-30; BIL comparisons span 2007-06-30 through 2026-06-30. All six variants underperformed the aligned SPY total-return benchmark in both descriptive defensive sleeves. Diagnostics preserve timing sensitivity, GFC/context windows, top UNRATE events, union/intersection combinations, and failure modes. The lag-1 proxy was risk-off throughout the 19-month GFC context but risk-on in the one-month August 2011 and October 2012 windows; these are proxy observations, not validation of the source narratives.
Decision
Exploratory publication under Other / Research, with no source-fidelity promotion. The best-effort requirement is satisfied with selected public proxies, an inferred best guess, six executed nearby variants, longest-history coverage, causal timing handling, diagnostics, and preserved negative results. The native Google Trends series and documented market rule remain unavailable. No official RecessionAlert indicator is registered.
Files and public paths
- Finding:
research/findings/1345.md - Task-specific SignalSpec:
research/signal_spec/fixtures/recessionalert-fear-indicator-proxy.json - Executable campaign runner:
research/research_campaign/recessionalert_1345.py - Campaign artifacts:
research/campaigns/recessionalert-fear-indicator/ - Executed notebook and generated analysis:
research/experiments/1345.ipynb,docs/experiments/1345-analysis.html - Public campaign page and payload:
docs/research/recessionalert-fear-indicator.html,docs/site-data/research/recessionalert-fear-indicator.json - Focused tests:
tests/test_recessionalert_fear_indicator_proxy.py
Public outputs contain only source summaries, hashes, proxy identifiers, derived metrics, causal assumptions, and explicit gaps. Cached HTML/GIF/workbook bytes, credentials, cookies, and private account data are excluded.