Model-risk warning
Detect high-coherence states where HAR under-forecasts forward realized volatility and tail exposure grows faster than that backward-looking baseline implies.
Coherence Risk Monitor • HAR blind-spot detector
See when volatility models become dangerously optimistic.
The Coherence Risk Monitor does not promise a magic trade. It shows something more useful: the exact regime where a standard HAR realized-volatility model — a backward-looking baseline — underestimates what comes next. This demo turns the verified finding into a working product story for risk teams, allocators, and internal quant research.
Product thesis
The value is not “predict anything.” The value is telling a desk or risk engine when a familiar statistical baseline — HAR — becomes structurally too calm.
Regime language
Convert abstract market-state physics into an interpretable signal that PMs, CROs, and allocators can read without black-box machinery.
Honest boundary
Preserve credibility by showing the IV wall explicitly: useful against backward-looking models, not a free pass through traded market prices.
Live demo dashboard
Every number below is computed in your browser: the coherence order parameter R1, the HAR baseline (fit out-of-sample), and the blind-spot flags. Pick an asset to explore.
Coherence R1 over time
Harmonic-1 order parameter |Z1| from seven trailing indicators. Shaded band = high-coherence regime (≥ 75th pct). Marks = HAR blind-spot flags.
HAR forecast vs. realized volatility
Where realized (solid) rises above the HAR baseline (dashed) in high-coherence states, the model is blind.
Realized vol
HAR forecast
Blind-spot
In-sample vs. out-of-sample validation
HAR is fit on the first chronological half; the coherence–forward-vol relationship is re-measured on the held-out half.
Workflow
The product is a decision-support layer on top of existing vol infrastructure, not a replacement for everything a desk already trusts.
1. Ingest
OHLCV + realized vol
Pull daily market data, compute leak-free phase fields, and standardize all inputs with a trailing window.
2. Diagnose
Compute coherence
Map seven trailing oscillator fields to phases and compress them into the R1 hub magnitude for monitoring and historical comparison.
3. Alert
Prioritize review
Surface only the moments where the backward-looking HAR forecast deserves human attention: high-R1, signed-positive blind spots.
Verified result map
This demo bakes in the current boundary: one strong forecasting-science result, one strong trading null.
| Claim | Status | Meaning | Product implication |
|---|---|---|---|
| HAR blind-spot | Confirmed | High R1 marks states where the backward-looking HAR model under-forecasts forward realized volatility. | Core enterprise signal. |
| Tail asymmetry | Confirmed | The signal concentrates in upside volatility surprises, not symmetric error noise. | Useful for escalation logic. |
| 30d long-gamma edge | Rejected | Traded implied volatility already prices the incremental realized move in SPY and GLD. | Do not sell as standalone fund alpha. |
| Rates / credit IV edge | Open | Needs paid historical options data for TLT and HYG before any claim is made. | Future enterprise research module. |
Partial correlations reported in the Scalar Flower working paper (Table 1): equities/bonds/commodities r=0.062 (13 instruments, 100% positive, 92% chronological sign-agreement), FX r=0.054 (5 pairs, 100%/100%), crypto r=0.049 (27 coins, 96%/48% — active research frontier), belief markets r=0.022 (lawful-but-faint — active research frontier). The deep-market legs (equities/FX/rates/metals) are the settled tier; crypto and belief are where the lawful structure is confirmed but the tradeable edge is time-unstable and under active, resourced investigation. Cross-sectional synchronization predicts forward index vol incremental to the priced correlation premium (r=0.09, p=0.003), with flagged states preceding ~1.6× volatility elevation at 1 day. Direction is null everywhere (sovereignty index S≈0.95–0.99).