Minerals Volatility Explained: How Price Dispersion Drives Supply Chain Risk

Volatility in Critical Minerals Is Structurally Different

Equity vol desks operate in deep, liquid markets with continuous price discovery and well-understood mean-reversion dynamics. Minerals markets share almost none of these properties. Tin, cobalt, lithium, and nickel trade across fragmented venues with varying contract specifications, thin order books outside front-month, and structural information asymmetries between producers, intermediaries, and end-consumers. The result is a volatility surface that behaves differently from anything in rates or equities.

Realized volatility in LME nickel, for example, clustered above 50% annualized for extended periods in 2022, driven not by macro repricing but by a single counterparty's short squeeze. Cobalt prices on the LME have exhibited multi-month regimes of suppressed vol followed by rapid dislocation when DRC export policy shifts or artisanal supply disruptions propagate. These are not fat tails in a stationary distribution. They are regime changes driven by identifiable, trackable factors: geographic concentration, policy intervention, and inventory drawdowns.

This is the core problem Volterra addresses. The model ingests 96 daily GDELT GKG news files alongside supply chain concentration indices and market context features to classify forward-looking volatility into five risk levels (LOW through EXTREME) across 7-day, 14-day, and 30-day horizons. The signal is not a point forecast of realized vol; it is a probability distribution over volatility regimes, which maps more directly onto how risk managers and options desks actually think about exposure.

Geographic Concentration as a Volatility Amplifier

The Herfindahl-Hirschman Index for mine-stage production of many critical minerals exceeds 2,500, the threshold the DOJ uses to define highly concentrated markets. China's dominance in rare earth processing, the DRC's share of cobalt extraction, and Indonesia's control of nickel ore exports each represent single points of failure that convert local disruptions into global price shocks.

This concentration matters for volatility modeling because it introduces event-driven asymmetry. A policy announcement from Jakarta on nickel ore export levies does not produce symmetric price movement. It creates a skewed, leptokurtic return distribution that standard GARCH models systematically underestimate. Volterra captures geographic concentration through HHI-derived features at both the extraction and processing stages, allowing the model to weight event sensitivity proportionally to structural vulnerability. When HHI is elevated and GDELT event intensity spikes for a given producing region, the model's probability mass shifts toward ELEVATED or HIGH regimes well before realized vol confirms the move.

From Vol Signal to Actionable Risk Framework

For options desks trading LME metals, the practical question is whether implied vol is correctly priced relative to the forward risk environment. A Volterra signal at HIGH for 14-day nickel volatility, when the LME nickel 1M implied vol surface is pricing at 30% annualized, represents a direct input to vol-selling or vol-buying decisions. The signal does not replace a desk's proprietary view, but it provides an independent, systematically generated prior calibrated at a walk-forward cross-validated AUC of 0.815.

For procurement teams and physical traders, the use case differs but the underlying logic is the same. A 30-day ELEVATED signal on cobalt informs hedge ratio adjustments, triggers for fixed-price contract negotiation windows, and inventory buffer decisions. The signal converts an abstract concern about "supply chain risk" into a quantified probability that the next 30 days will see above-normal price dispersion.

Systematic funds can integrate the signal as a feature in multi-asset vol timing strategies or as a filter for minerals-exposed equity baskets. The five-level classification maps naturally onto position sizing rules without requiring bespoke threshold calibration.

Measuring What Matters: News Sentiment, Supply Data, and Market Context

Traditional commodity vol models rely on price history and perhaps term structure slope. Volterra extends the feature space to include real-time news event classification from GDELT, supply-side metrics (production concentration, inventory levels, shipping and logistics indicators), and market context variables that capture cross-commodity correlation regimes and macro backdrop.

The GDELT GKG layer is particularly relevant for minerals because price-moving information in these markets often originates outside the financial press: government gazettes, mining ministry announcements, environmental agency rulings, and labor action reports in producing regions. Processing 96 files daily allows the model to detect shifts in event tone and volume before they are fully reflected in exchange-traded prices.

Figures from the Volterra daily pipeline. Full historical backfill available on AWS Data Exchange. The dataset covers 12 exchange-traded critical minerals across LME, COMEX, NYMEX, and SGX, providing a structured, machine-readable foundation for integrating minerals volatility into institutional risk workflows.