quantmodel — Replicate RecessionAlert indicator: The Average Liquidity Index (ALIX)

Objective

Independently replicate the distinct RecessionAlert row The Average Liquidity Index (ALIX) | RecessionAlert from the checked-in canonical snapshot without a live-site request, credentials, private account data, or mutation of data/recessionalert.sqlite. Because the source does not provide source-linked observations or complete action semantics, this reopened campaign tests labeled public analogues rather than claiming native fidelity.

Disclosed rule and provenance

The cached article describes ALIX as the average of three named components: EMA(21) of VMCOS; EMA(10) less EMA(21) of VMCOS; and EMA(10) less EMA(20) of Net 13-week highs less Net 52-week highs. It describes short-term component values below zero as early warnings and ALIX below zero as a likely volatility or sideways-movement warning, not a certainty of further decline. It claims daily subscriber prints, a downloadable 20-year history, and a relationship extending back to 1990.

The selected page body is the read-only manifest/inventory/database record at https://recessionalert.com/alix/, SHA256 sha256:9aae9dcf82b77c6195e2e586fc79ba20fe792093867ba4f30cec84f0120aca77, fetched 2026-08-22T23:46:53Z, with 82,622 bytes, two same-URL crawl revisions, and zero target-page asset rows. The source map is research/findings/recessionalert_inventory.md; the manifest is research/artifacts/recessionalert/manifest.json; the native source-fidelity contract is research/findings/specs/recessionalert-alix.json.

Best-effort proxy campaign

The campaign uses public series already present in the checked-in data/market.sqlite, read-only: NFCILEVERAGE, BAA10Y, SPY, and IEF. The common monthly panel spans 2002-07-01 through 2026-07-01, limited by IEF availability; macro history is retained as warm-up where applicable. Six labeled variants were executed:

alix_proxy_level_lb12_t0_lag1: rolling 12-month NFCILEVERAGE z-score, threshold 0, one available-bar lag;
alix_proxy_spread_lb12_t0_lag1: EMA(10) minus EMA(21) NFCILEVERAGE spread, threshold 0, one-bar lag;
alix_proxy_trend_lb12_t0_lag1: SPY below rolling 12-month mean stress gate, threshold 0, one-bar lag;
alix_proxy_ensemble_lb21_t05_lag2: average of standardized NFCILEVERAGE, BAA10Y, and SPY trend stress, threshold 0.5, two-bar lag;
alix_proxy_level_lb6_tm05_lag2: six-month level neighbor at threshold -0.5 and two-bar lag;
alix_proxy_level_lb12_t0_lag0_sensitivity: same-close timing sensitivity only, never a causal production interpretation.

These are inferred/proxy reconstructions. None reproduces native VMCOS/HILO inputs, native ALIX values, or a source-defined trade rule. The executable implementation, six SignalSpecs, lab series, state-space ledger, diagnostics, synthesis, and notebook are under research/campaigns/recessionalert-alix-1351/ and research/experiments/1351.ipynb.

Timing and causal assumptions

Native numeric span is undisclosed: the article is one dated narrative snapshot (2019-11-01T12:44:44-0400) that claims a history but supplies no observation payload. Native calculation cutoff, release timestamp, timezone, vintage, late-input handling, and revision schedule are unknown. The approximately 2486-day article-to-crawl interval is archival delay, not publication lag. For proxy exploration, monthly observations are evaluated with one or two available-bar lags; lag zero is retained only as a research-only sensitivity. Any future overlay must use point-in-time inputs and execute no earlier than the next available trading bar. Missing warm-up or input values remain unavailable rather than false or zero.

Validation and publication

The focused offline suite python3 -m pytest -q tests/test_recessionalert_1351.py covers source metadata, read-only database behavior, rolling/EMA warm-up, lagged warning state, SignalSpec validation and DSL parity, non-negative target weights, and deterministic monthly metrics. The canonical publication commands are:

python3 research/campaigns/recessionalert-alix-1351/run_lab_series.py;
python3 -m research.reports research-library --out docs;
python3 -m research.reports results --out docs --source research/results;
python3 research/scripts/check_static_reports.py docs.

The public Research Library entry is docs/research/recessionalert-alix-1351.html with payload docs/site-data/research/recessionalert-alix-1351.json; the Results changelog detail is generated at docs/results/1351.html. Public output contains derived proxy summaries, formulas, hashes, timing metadata, and limitations only. It does not contain cached HTML/media bytes, workbooks, credentials, cookies, nonce-bearing URLs, private account data, or unverified native ALIX series.

Explicit gaps and disposition

No source-linked ALIX/VMCOS/HILO series, exact formula precedence, EMA initialization, formal zero-crossing persistence, executable buy/sell/hold/re-entry mapping, target/sizing/cash semantics, native release lag, revision policy, or independent market outcome window is disclosed. The public proxies therefore remain exploratory and descriptive; costs and slippage are omitted, and no market validation or official indicator registration follows.