quantmodel

← Results index

The improved, de-trended Composite Market Health Index (CMHI) | RecessionAlert — exploratory proxy campaign

2026-09-01T23:00:00Z Research completed The SHA-pinned offline CMHI article discloses a daily -1..+1 composite with zero and +0.3 pivot concepts but no native observations or complete BUY/SELL rule. Thirty-three labeled public-proxy trials were executed with causal lag defaults; publication is exploratory Other/Research only.
1375 recessionalert research_campaign proxy cmhi detrended fidelity_mismatch data_limited interesting_negative_result needs_more_research

Source report: research/campaigns/recessionalert-detrended-cmhi-1375/synthesis.json

Objective

Independently replicate the distinct RecessionAlert inventory row for the de-trended CMHI article from the checked-in data/recessionalert.sqlite snapshot without live-site requests, credentials, private account data, or canonical crawl-database mutation. Because the source omits constituent series and a complete executable state, the campaign tests labeled public proxies rather than claiming native parity.

Source and provenance

The selected page https://recessionalert.com/de-trended-cmhi/ is bound to SHA256 sha256:3d0121b55586960ae5624ac588023ecce2416503226c0d5f0ed94d73a7073427, fetched at 2026-08-22T23:46:27Z. The canonical row has no linked numeric asset. Public proxies were loaded from data/market.sqlite in read-only mode.

Disclosed rule, native span, and publication lag

Documented fragments include a daily CMHI bounded from -1 to +1, qualitative regression de-trending for earlier bear warnings, post-1983 horizontal reference near +0.3, and monitoring pivots near zero and +0.3. Claimed history spans 1968 (all models) and 2014 (interactive download) without captured numeric series. Daily cadence is documented; timezone, cutoff, release time, and revision policy are unknown. Campaign causal policy uses month-end observations, one-month macro availability where applicable, and one trading-month lag by default; lag-zero is control-only.

Approach and variants

Primitives live in research/research_campaign/recessionalert_1375.py. Families: expanding_linear_detrend, rolling_linear_detrend, horizontal_level_proxy, zero_pivot_overlay, cmhi_diffusion_bridge. Trial ledger: research/campaigns/recessionalert-detrended-cmhi-1375/state-space.json (33 variants across three panels).

Results (exploratory)

The causal expanding-60 residual best guess (expanding-lb60-resid0-p2-lag1-ief) returned 6.52% CAGR and -23.15% maximum drawdown over 271 live-market months versus 10.75% CAGR and -50.78% drawdown for same-span risk-on. None of the 30 causal frozen trials beat same-span risk-on CAGR. Defensive overlays reduced drawdown in GFC, 2015-16, and pandemic windows but incurred severe opportunity cost in the 2023-2026 expansion.

Files and validation

Validation: focused pytest 10 passed; read-only DBs; no live RA fetch.

Limitations

Public proxies are not official RecessionAlert CMHI observations. Exact constituent formulas, regression estimator vintages, pivot semantics, and publication clock remain undisclosed. Do not treat exploratory equity paths as a registered Other/Indicators signal. Official CMHI indicator promotion is not supported.

Source artifact: research/results/1375.md