EQX-M1 — Model Card
EQX-M1 is an Experimental-series research mark built to test a lightweight, regime-aware equity policy. It blends price action with flow and simple macro proxies, decouples decisions by regime, and keeps risk honest via tight calibration and a “pause-o-meter” that can down-shift when signal quality degrades.
Key Metrics
Performance Viewer
Performance Viewer
Series normalised to start = 1.0; includes costs. Equities: model, SPX, PID-governed book.
Policy Control Trace
Policy Control Trace
- Risk-off days55
- Pause share7.8%
- Threshold0.576 / 0.599 / 0.629
- Max gross0.200 / 0.598 / 0.702
Shared axis; raw PID/control magnitudes. Caps tighten in stress and relax as conditions normalise.
By-Regime Performance
| Regime | Ω | Hit rate | Days |
|---|---|---|---|
| Bull | 3.38 | 48.1% | 1467 |
| Neutral | 1.28 | 44.7% | 790 |
| Bear | 0.27 | 19.6% | 1118 |
Long-only profile is pro-cyclical: strong in Bulls, steady in Neutral, destructive if unconstrained in Bears. The controller down-shifts exposure in stress to preserve capital and shorten recovery arcs.
Risk & Pauses
- Gross caps (min/med/max)0.20 / 0.60 / 0.70
- Thresholds (min/med/max)0.576 / 0.599 / 0.629
Pauses and tighter caps are the “seatbelt.” They reduce tail-risk during volatile or drawdown phases, trading a bit of upside for a cleaner equity path.
Calibration
- ECE0.1607
- Bins10
Bins around 0.60–0.85 are over-confident. Treat thresholds as dials; a dedicated calibration pass (e.g., isotonic) is planned for the next iteration.
Sector Impact
Green chips show sectors where odds (Ω) lean positive; red chips highlight headwinds. The PID router can use this texture to tilt or to route toward safer policies in fragile groups.
Ticker Consistency
- Portfolio max win streak138 d
- Portfolio max loss streak121 d
- Top ticker win streakVTR · 35 d
Streaks contextualise variability. Long win runs cluster in calm regimes; long loss runs often overlap bear stress. Controller caps and abstention dampen the worst streaks without muting resilience.
Operations
| Stage | dev |
| Active | Yes |
| Last update (UTC) | 2025-09-19 |
| Artifacts | equity_25y.json, pid_policy_trace.json |
| Hashes | — |
Ops health looks normal; policy traces and equities render from local artifacts for reproducibility.
Limitations
Research model; parameters subject to change. Backtest assumes fixed costs and a long-only book. Regime labeling uses simple hysteresis rules and may differ from alternative taxonomies.
Methodology (summary)
Selection via probability thresholding; sizes scale with surplus-over-threshold to a regime-aware gross cap. For deep dives, see the Workstation (Sim) pages.