Analyse de marché
The dominant source remains macro-narrative, but quantitative trading requires microstructure layers missing from the “petrodollar → gold” narrative. Since May 2026, vessels, including VLCCs and LNG carriers, have switched off AIS transponders to cross the Strait of Hormuz in small groups, sometimes under |…| coordination; by August 2026, 62% of tanker crossings occurred with AIS disabled. This degrades data quality and biases physical-flow estimates downward, requiring a latent “dark transits” variable calibrated through data fusion combining Planet Labs satellite imagery, customs data and maritime agency reports.
Steelldy Geopolitical Risk Model can support link analysis and trust graphs to reconstruct routes and relationships from incomplete, noisy, adversarial data. Separately, dark-pool and ATS order-flow data, inferred through models such as Alphascope/Polybro, suggest institutional gold accumulation through blocks exceeding $10 million in Q2 2026, despite declining reported COT net-long positions. This divergence between declared positioning and executed flows constitutes a mosaic signal, quantified as D_t = ||Δw_t^reported − Δw_t^executed||_ℓ1; a high value indicates hidden accumulation or distribution and signals a regime shift rather than direction.
Finally, prediction-market contracts on ceasefires and gold-price targets from Polymarket, Kalshi and CME Event Contracts provide useful implied probabilities but are distorted by liquidity. A Bayesian shrinkage adjustment combines market probabilities with a structural-model prior: p̂_i^oracle = λ_i p_i^market + (1 − λ_i) p_i^model, where λ_i depends on order-book depth and bid-ask spread. Illiquid contracts, especially extreme strikes, should be shrunk strongly toward the TVP-VAR model prior; otherwise, market probabilities systematically overstate tail risks and generate oversized hedging signals.
