HDM — Methodology & Experiment

The Hormuz Divergence Monitor. A daily research indicator measuring the gap between how ships behave and how markets price risk at the Strait of Hormuz. Published daily at 17:00 UTC. Not for settlement.**

This document is the public methodology of record. It is comprehensive by design — it states mechanisms, not operational trading values. The construction constants were frozen and cryptographically timestamped before any data was ingested (see Data integrity), so the empirical results this experiment produces cannot be accused of specification search.

1. The experiment

Two communities price Strait-of-Hormuz risk every single day, and they do not talk to each other.

• Physical actors — tankers, gas carriers, and cargo ships — reveal risk through behaviour: how many transit, how fast, whether they loiter or go dark, where they choose to load.
• Financial markets reveal risk through instruments: oil and equity volatility, option skew, defense stocks, sovereign credit, gold, tanker equities.

When something happens in the strait, these two populations do not necessarily move together. HDM measures both on one common, standardized scale and publishes the difference between them.

The hypothesis under test: when ships and markets disagree, does the disagreement carry information about what comes next? The experiment is designed to answer that honestly — including the possibility that the answer is no. A well-identified null result is itself a publishable finding about how this risk is intermediated.

Why now: through recent Middle-East stress cycles, war-risk insurance premiums for a Hormuz transit have moved from a fraction of a percent of hull value to multiples of that, and physical-vs-financial dislocations have been visible in the tape. The raw ingredients — daily transit data, AIS behaviour, market volatility — are now available at daily frequency, most of them free and public.

2. The two indices and the headline

HDM computes two composite indices and publishes their difference.

• PSI — Physical Stress Index. How stressed the strait looks, from ship behaviour. Higher = more physical stress.
• MRI — Market-priced Risk Index. How much risk markets are charging, from financial instruments. Higher = more market-priced risk.
• D = PSI − MRI — the headline divergence.
• D > 0 — physical-hot: ships are behaving as if a crisis is underway that markets have not yet repriced.
• D < 0 — market-hot: markets are pricing more risk than the strait is physically showing.
• D ≈ 0 — ships and markets broadly agree.

Regime bands (on |D|): neutral (|D| < 1) · watch (1–2) · action (> 2).

The raw D is the research/trading reference; a 3-day trailing mean is published for display.

3. Components

Each index is an equal-weighted composite of standardized components.

Physical Stress Index (PSI) — P1–P6

• P1 — Transit count. The deseasonalized log-deviation of daily transits from their trailing 90-day, day-of-week-matched median. Fewer transits = stress (P1 enters with a negative sign).
• P2 — Loitering. The intensity of vessels stopping or slow-steaming in the Gulf approaches — waiting behaviour that rises under threat.
• P3 — Dark events. AIS “gaps” — vessels that go dark while underway, a signal of evasion or disabling.
• P4 — Loading migration. The share of Gulf-origin oil loading that shifts to terminals seaward of the strait (Fujairah cluster) versus inside the Gulf. Hormuz has no sea alternative, so stress reveals itself as loading-point substitution.
• P5 — Differential speed. How much slower laden tankers move in the Hormuz approach lane than normal — minus the same deviation on a control lane (Malacca) — isolating Hormuz-specific hesitation from global fuel-price effects.
• P6 — Insurance conditions. The war-risk premium level for a Hormuz transit, plus changes to the insurers’ listed-areas designations.

Market-priced Risk Index (MRI) — M1–M7

• M1 oil volatility (OVX) · M2 equity volatility (VIX) · M3 crude option skew (escalation fear shows as call skew) · M4 defense-basket relative strength versus the broad market · M5 sovereign-credit stress (a regional credit spread) · M6 gold · M7 tanker-equity volatility.

4. Data sources — and why

HDM is built in tiers: a free, public, fully-reproducible tier-0 that anyone can replicate, with commercial upgrades added only where a component genuinely needs them. Point-in-time discipline governs every source: each day’s value uses only data available by publication time, and every input’s vintage is archived.

Pending components (by design, not oversight): P3 dark events and P5 differential speed require raw AIS vessel tracks; P4 loading migration needs a curated map of Gulf oil-loading terminals (those seaward of the strait versus inside it); P6 war-risk premium is relationship-distributed and partly hand-collected; M3 option skew needs options data. These enter as the index earns the commercial data (own-AIS detectors, a broker premium panel) — a documented tier-1 upgrade, not a change to the frozen method.

Design principle: the free tier-0 must produce a defensible index on its own, and a replication package that runs entirely on free sources is a first-class deliverable. Commercial data improves precision; it is never a precondition to publish.

5. How the scores are computed

Every raw component is standardized the same way, so physically- and financially-derived numbers live on one comparable scale.

1. Rolling z-score. Each component value is converted to z = (x − μ) / σ, where μ and σ are computed over the trailing 250 trading days, excluding the current day, and the result is winsorized at ±4 (clipped, so no single extreme input dominates).
2. Availability, not imputation. A component that hasn’t updated for five business days is flagged stale and excluded from that day’s composite — never guessed. The composites are means over whatever components are available.
3. Composites. PSI = mean of stress-signed physical z-scores, MRI = mean of market z-scores, and the headline D = PSI − MRI. A 3-day trailing mean is published for display.
4. Regime bands are applied to |D| (neutral / watch / action).

The frozen-baseline companion (HDM-Level)

The rolling standardization makes D a surprise measure — a disruption sustained past the window is gradually absorbed into μ and σ, and the score reverts even as absolute conditions stay severe. That is correct for a divergence signal, wrong for measuring chronic deprivation. So HDM also publishes a Level series (PSI-Level, MRI-Level, D-Level) computed with μ and σ frozen once over a pre-crisis window (Jan 2019 – Sep 2023) and never updated. D answers “how abnormal is today versus the recent past”; D-Level answers “how far is today from the pre-crisis world.” Any settlement or contracting application references the Level series or raw observables — never the rolling scores.

Registered robustness variants

Alternative constructions run daily in shadow from launch and are disclosed in advance so they cannot be used to shop for a result: a logarithmic tail transform (instead of the hard ±4 clip), inverse-variance and principal-component weightings (as standing challengers to the equal-weight baseline, promotable only on forward out-of-sample outperformance), and a regional-only market composite. Equal weights are the pre-registered baseline — stability over optimization, correct for a benchmark.

6. Data integrity

Credibility here rests on cryptography and discipline, not on trust.

• Pre-registration + cryptographic seal. The complete construction constants — components, windows, weights, the standardization — were written down, hashed (SHA-256 010af39e…), and externally timestamped via OpenTimestamps (Bitcoin-anchored) before any behavioural data was ingested. This is the act that lets later results stand: the method provably predates the data.
• The daily hash chain. Each day’s published record hashes its own canonical content together with the prior day’s hash, and binds to the sealed methodology hash. The result is a tamper-evident chain: any retroactive edit to any past day breaks every day after it. This doubles as the oracle mechanism for any future settlement application.
• Point-in-time discipline. Each day uses only data available by publication; every input carries its dataset vintage; methodology changes are versioned, dated, and never applied retroactively.
• The information wall (for any future licensed use): the index consumes only public and commercially-licensed data; the trading loop consumes only the published, hash-verified record, with no technical path to pre-publication values; provenance for every input is hashed and third-party-auditable.
• Honest labelling. Every record carries NOT_FOR_SETTLEMENT. HDM is today a research indicator; any move toward a settlement-grade benchmark would happen under a structurally separate administrator with independent oversight, before any instrument references the number.

7. The trading experiment

Publishing a number is not the same as showing it was tradeable. So a validation trading program trades HDM’s published signals under frozen, hash-locked rules — an auditable demonstration, explicitly not a fund track record and not marketing.

• Frozen rules. Entries reference the index exactly as published at 17:00 UTC; every ticket stores the value, publication timestamp, and input hash. Rules only ever append, never edit. Pre-publication values are structurally unavailable to the trading loop — the same barrier that enforces no look-ahead.
• Defined-risk only, in every phase. Long options, debit spreads, and binaries; no naked short optionality, ever. Maximum loss is capped by construction.
• Phased. Paper first, with unlimited breadth (the evidentiary book). Then a small, segregated, live autonomous book — sized so its maximum loss is the cash it holds — trading the same signals across brokerage, event-market, and crypto rails with hard breakers (daily-loss and drawdown halts) enforced in code, not memory. The live book carries skin-in-the-game and proves the full end-to-end loop; the paper book carries the evidence.
• Kill criteria are pre-registered. A rule with negative expectancy after a set number of closed tickets is halved, then suspended. Suspension is a result.

The point is simple: the index’s owner eats its own cooking, with real money, under rules that were frozen before the trades.

8. What gets published

This is a pre-registered empirical study, and the commitment is to publish what it finds — whatever it finds.

• A pre-registered protocol (frozen before data): event studies around insurer listed-area changes; predictive regressions of forward tanker-equity, oil-vol, and freight outcomes on D, with controls and multiple-testing corrections; and placebos — the identical index built at chokepoints with no geopolitical-premium channel (Malacca, Gibraltar), which should show no predictability.
• Identification honesty. Divergence means disagreement, not mispricing. The study distinguishes rational sources (private information, insurance-capacity frictions, market-attention frictions) and commits to publishing whichever pattern the data shows — including a clean null.
• A working paper and a replication package that runs on free sources alone, so the result is independently reproducible.
• The live index and its daily records, hash-chained and archived, as the public evidence base.

9. Current status & limitations

• Live today: 8 of 13 components — P1 (transits), P2 (loitering); M1 (OVX), M2 (VIX), M4 (defense RS), M5 (sovereign credit), M6 (gold), M7 (tanker vol). Published daily, sealed, and hash-chained.
• Pending (5): P3 (dark events) and P5 (differential speed) — need raw AIS tracks; P4 (loading migration) — needs a curated Gulf-terminal classification; P6 (war-risk premium) — relationship-sourced; M3 (option skew) — needs options data. The composites use available components until these land.
• Known caveats, stated plainly. AIS is adversarial in this region (spoofing, dark-running) — mitigated, not eliminated. The war-risk-premium input is the weakest link (over-the-counter, partly reconstructed) and is hard-capped in its weight. PortWatch transit data publishes with a short processing lag. Every one of these is a documented limitation, not a hidden assumption.

This methodology is public and editable at its source. The index, its daily records, and this document together are the experiment’s evidence base.