Validation & Methods

Systemic Stress

Hit if any of: VIX ≥ 35; SPX 5‑day ≤ −3.0%; SPX 10‑day ≤ −6.0%; SPX 20‑day ≤ −10.0%.

Volatility Spike

Hit if any of: VIX ≥ 30; SPX 5‑day ≤ −3.0%; SPX 10‑day ≤ −5.0%.

Stress (Advisory)

Hit if any of: VIX ≥ 30; SPX 5‑day ≤ −3.0%; SPX 10‑day ≤ −5.0%.

Turning

Hit if any of: SPX 10‑day ≤ −2.0%; SPX drawdown ≤ −4.0%; VIX ≥ 25 for 2 consecutive trading days.

Calm

Active when all other alarms are inactive (informational regime classification).

What we backtest

• MSD evaluates five production market-state alarms: systemic stress, volatility spike, turning, Stress (Advisory), and a meta calm state.
• Performance summaries are maintained in a versioned internal validation archive (single source of truth). A full validation packet with current figures and provenance is available on request via the validation packet form.

Data and period

• Historical period: multi‑year history (e.g., 2012–2024) as reflected in the validation files.
• Inputs: the canonical PDI time series and public market data (SPX and VIX).
• Time base: business‑day (BD) calendar; timestamps validated and stored in ISO 8601.

What we do / What we don’t do

How an alarm is evaluated (at a glance)

1. Fixed rules: Each alarm is specified by an immutable definition (no in‑sample tuning in the backtest).
2. Daily evaluation: Conditions are checked per business day against historical inputs.
3. Episodes: Consecutive alert days are collapsed into alert “episodes” using business‑day cooldowns and windows.
4. Market events: Independent market events (e.g., VIX and SPX patterns) are identified and deduplicated into event episodes.
5. Scoring: An alert episode is a “hit” if a qualifying event occurs within a forward‑looking scoring horizon (in business days). Otherwise it is a miss; episodes with no event are false positives.
6. Metrics: Precision, recall, and F1 are computed from episodes; figures reported on the website come from these validation files.

Design principles that reduce overfitting risk

• No in‑sample optimization: Backtests run the published rule logic as‑is.
• Forward‑only scoring: Hits require future events within a predefined BD horizon; no look‑ahead.
• Deterministic time handling: All windows/horizons/cooldowns are business‑day based.
• Separation of concerns: Historical performance validation files (manifests) are presentation SoT and are not used by evaluation logic.
• Snapshot parity: Internal regression checks ensure outputs remain stable across versions.

Data sources

We use public market data (SPX, VIX) and public environmental datasets (e.g., NOAA, NASA, USGS). References to third‑party data providers are for sourcing only and do not imply endorsement.

Validation protocols (walk‑forward and robustness)

• Held‑out era evaluation on excluded historical eras; no tuning on the evaluation period.
• Walk‑forward validation: expanding-window approach with 2-year training, 1-year test, 90-day roll. Era splits (SC24/SC25) validated separately.
• Leave‑one‑era‑out to assess robustness by omitting eras and testing excluded windows.
• Era stratification to surface instability and confirm consistency across market conditions.
• Cross‑domain sanity checks where applicable.
• Optional inactive‑day baselines using identical episode windows are available for context in the internal validation materials.

For complete overfitting controls and offline reproduction procedures, we provide an auditor‑friendly validation packet on request via the validation packet form.

Reproducibility

• Public figures are sourced from versioned internal validation records; website statistics are derived directly from this single source of truth.
• Validation Packet — an auditor‑friendly package to reproduce metrics without environment access. Request via the form and we will deliver the packet by email; certain sensitive artifacts may require an NDA.

Compliance framing

All analytics are informational research or risk‑classification only and must not be construed as investment advice, solicitation, or performance guarantees.

FAQ