Daily ML-driven volatility research data for 10 exchange-traded minerals across 14 and 30-day horizons. One structured dataset, delivered daily by 08:00 UTC — for research and information, not advice.
Accuracy and AUC figures are derived from hypothetical walk-forward backtesting (2016–present), were not achieved in live trading, and are shown for research and information only. Past and backtested performance is not indicative of future results and is not financial advice. See disclaimer.
Hypothetical backtest on out-of-sample data. The signal turned hot on Feb 16 — two weeks before the 2022 LME squeeze drove nickel to $44.7k.
Source: Volterra 7-day model, out-of-sample backtest · Nickel · LME · Feb–Mar 2022. Single illustrative example, not representative of all outcomes — the model does not detect every event. Backtested and past performance is not indicative of future results. Not financial advice; see the disclaimer.
Illustrative and based on historical backtesting — for informational and research purposes only. Not financial advice and not indicative of future results. See disclaimer.
We predict the probability of abnormal price volatility — not whether prices will rise or fall. For most institutional use cases, the size of a move matters more than its direction.
A HIGH reading for copper describes an elevated probability of a large price swing in either direction within the forecast window — it describes magnitude, not action.
One dataset. Research and risk-monitoring context across 12 minerals and 2 horizons — for informational use only, not advice.
The model's raw output: probability (0–100%) that volatility exceeds the trailing 60-day baseline + 1.5σ within the forecast horizon.
Five tiers from LOW to EXTREME, calibrated to the probability of volatility exceeding the trailing 60-day baseline + 1.5σ.
Boolean field for automated monitoring. True only when HIGH or EXTREME with sufficient confidence. Pipe directly into alerting systems.
Three named signal drivers per prediction — news volume, production risk, country concentration, geographic spread, and more.
Model certainty rating: low, medium, or high. Reflects how strongly the underlying signals supported the prediction at output time.
One day's predictions across all minerals. This is exactly what arrives in your pipeline each morning.
| Mineral | Probability | Risk Level | Alert | Confidence | Factor 1 | Factor 2 | Factor 3 |
|---|
Sample: 2026-03-05 · 0 rows · 7d horizon
The probability is the model's estimate that realised volatility will exceed its trailing 60-day average by more than 1.5 standard deviations within the forecast window — a measure of expected turbulence in either direction, not price direction. Because volatility spikes are rare (historically 13–22% of days, depending on the mineral), most readings sit in single digits; what matters is the level relative to a mineral's own baseline, and how it moves day over day. Risk levels band the probability into five descriptive tiers, and each model applies its own backtested alert threshold, so the same reading can sit in different tiers for different minerals.
Sample data shown for illustration. For informational purposes only — not financial advice. See disclaimer.
Probability of realised volatility exceeding trailing 60-day average + 1.5 standard deviations within forecast horizon
The top 3 risk factors are named in every row. 5 signal categories across 250+ individual features.
The daily prediction signal fuses several non-price signal families: machine-read global news and event coverage spanning 100+ languages, refreshed daily; speculative-positioning data; physical-market inventory levels; futures market activity (open interest and volume); search-interest trends; policy-uncertainty and geopolitical-risk indices; structural supply-concentration benchmarks, so disruptions concentrated in a dominant producing country carry more weight than the same event spread across many producers; and export-restriction indicators tracking government measures that affect mineral supply. The model deliberately uses no price data — not the mineral's own price, nor equities or FX — so it forecasts turbulence from real-world information rather than from price behaviour itself. The specific sources, construction, and weighting inside each signal family are proprietary.
370+ features per mineral per day, constructed across rolling windows of 3, 7, 14, and 30 days from the news stream and non-price fundamentals. Covers news volume and velocity, event severity, source credibility, geographic and supply-chain concentration, disruption type classification, sentiment, cross-mineral contagion, speculative positioning, physical inventory, futures open-interest and volume, search interest, and policy/geopolitical-risk indices. A per-mineral blocklist prevents cross-contamination between unrelated signals.
A proprietary ensemble of machine-learning classifiers trained with walk-forward expanding-window cross-validation — no random splits, no future data leakage. Each mineral has its own independent model per forecast horizon. Auto feature selection picks the highest-signal features per fold, per mineral.
Models are trained on data from 2016 to present. The walk-forward validation framework means every prediction in backtesting was made using only information available at that point in time.
Mean backtested AUC of 0.78 across the 20 deployed models (10 minerals × 14- and 30-day horizons), rising to about 0.80 at the 30-day horizon. The feed intentionally covers only the 14- and 30-day windows: short-term (7-day) volatility is driven largely by price microstructure, which this news-and-fundamentals model deliberately does not use, so signal strength increases with horizon and the shorter window was excluded rather than shipped weak.
Probability that realised volatility will exceed the trailing 60-day average by more than 1.5 standard deviations within the forecast window. Captures large moves in either direction — not directional price prediction.
The feed covers 10 exchange-traded minerals with liquid, transparent daily prices: copper, aluminium, zinc, nickel, lead, tin (LME base metals), gold, silver, palladium (precious), and iron ore. Cobalt and lithium are deliberately excluded: they trade largely by private contract, and their reference prices are published only as subscription assessments by specialist price-reporting agencies rather than a liquid public exchange. We do not publish a forecast we cannot ground in reliable daily price data. Both are tracked in our research and may be added if a suitable price source is secured.
The methodology and performance figures described above are based on historical backtesting and statistical modelling. All results are hypothetical, were not achieved in actual trading, and are provided strictly for research and informational purposes; they must not be relied upon as a basis for any investment, trading, hedging, or procurement decision. Hypothetical and backtested performance has inherent limitations — it is prepared with the benefit of hindsight, does not reflect transaction costs or market impact, and model behaviour may differ materially during market stress or conditions not represented in the training data. Past performance, whether actual or backtested, is not indicative of future results, and no representation or warranty is made that any model will achieve results similar to those shown. Nothing in this section constitutes financial product advice, investment advice, investment research, or a recommendation, opinion, or endorsement of any kind in any jurisdiction — including under the Corporations Act 2001 (Cth) in Australia, the Investment Advisers Act of 1940 and Commodity Exchange Act in the United States, the Financial Services and Markets Act 2000 and FCA rules in the United Kingdom, Canadian securities legislation, or MiFID II in the European Union. Volterra is not licensed, registered, or authorised as a financial adviser by any regulator in any of those jurisdictions or elsewhere; no adviser-client, fiduciary, or other special relationship is created by your use of this information; and to the maximum extent permitted by law Volterra accepts no liability for any loss arising from reliance on it. You should obtain advice from an independent, appropriately licensed professional before making any financial decision. See disclaimer.
The fields described above, and the values delivered in the dataset, are general statistical information describing volatility risk and its drivers, derived from publicly available news and event data, non-price fundamentals (positioning, physical inventory, futures activity, search interest, policy and geopolitical-risk indices, export restrictions), and production benchmarks processed through statistical models. This information is factual in nature, is provided strictly for research and informational purposes, and does not constitute — and must not be relied upon as — financial product advice, investment advice, trading advice, or a recommendation, opinion, offer, or solicitation to buy, sell, hold, or otherwise deal in any financial instrument or commodity. Volterra is not a licensed or registered financial adviser in any jurisdiction: it does not hold an Australian Financial Services (AFS) licence under the Corporations Act 2001 (Cth); is not registered as an investment adviser under the US Investment Advisers Act of 1940 or as a commodity trading adviser under the US Commodity Exchange Act; is not authorised or regulated by the UK Financial Conduct Authority; is not registered with any Canadian securities regulatory authority; and is not authorised under MiFID II or any equivalent regime in the European Union or elsewhere. No adviser-client, fiduciary, or other special relationship arises from your access to or use of this information in any jurisdiction. Field values are probabilistic model estimates that may be incomplete or incorrect, are provided without warranty of any kind, and do not take into account your objectives, financial situation, or needs. Any decision made in reliance on this information is made solely at your own risk, and to the maximum extent permitted by law Volterra accepts no liability for any resulting loss. You should obtain advice from an independent, appropriately licensed professional before making any financial decision. See disclaimer.
Every post is generated from live pipeline data — real scores, real movements, real disruption events. Not generic commentary.
Different volatility forecast horizons serve fundamentally different risk functions. This post breaks down how 7-day, 14-day, and 30-day windows map to distinct trading, hedging, and procurement workflows in critical minerals markets.
Standard k-fold cross-validation leaks future information into commodity volatility models. Walk-forward validation enforces temporal ordering, producing reliable out-of-sample performance estimates for production deployment.
How the GDELT Global Knowledge Graph transforms unstructured global news into quantitative volatility signals for critical minerals, and why traditional sentiment analysis falls short for commodity risk.
Posts are generated from live pipeline data using the same signals and ML features that power the predictions. Published weekly across five rotating content categories.
One research dataset, delivered daily by 08:00 UTC.
Important notice — not financial advice
Not financial advice
The information, risk scores, and volatility forecasts provided by Volterra are for research and information only. They are provided for general informational and research purposes and do not constitute financial product advice, investment advice, or a recommendation, offer, or solicitation to buy, sell, or hold any financial instrument, security, derivative, or commodity. Volterra is not a licensed or registered financial adviser, investment adviser, broker, or commodity trading adviser in any jurisdiction.
No warranty — provided “as is”
The data is provided on an “as is” and “as available” basis without warranties of any kind, whether express or implied, including any warranty of accuracy, completeness, timeliness, reliability, or fitness for a particular purpose. Risk scores are probabilistic statistical estimates and may be incorrect. Volterra does not warrant that the data will be uninterrupted, error-free, or that any forecast will prove accurate.
Limitation of liability
To the maximum extent permitted by law, Volterra and its operators shall not be liable for any direct, indirect, incidental, special, consequential, or exemplary loss or damage — including without limitation trading or investment losses, lost profits, or lost opportunity — arising out of or in connection with your access to, use of, or reliance on the data, even if advised of the possibility of such loss.
No advisory or fiduciary relationship
Your access to or use of Volterra does not create any adviser–client, fiduciary, agency, or other special relationship between you and Volterra. The data is not tailored to your objectives, financial situation, or needs, and Volterra does not consider your individual circumstances.
Past performance & model limitations
All risk scores and volatility forecasts are based on historical backtesting and statistical modelling. Past performance and backtested results are not indicative of future results. Any examples shown (including case studies of individual events) are illustrative, are not representative of all outcomes, and should not be taken to imply that the model detects every event — it does not. Model accuracy may differ materially during periods of market stress, geopolitical disruption, or conditions not represented in the training data. Coverage of Chinese domestic policy signals is limited. The product is intended to serve as one input among many in a broader analytical and risk-management process — not as a standalone trading signal.
United States
Volterra is not registered as an investment adviser under the Investment Advisers Act of 1940, nor as a commodity trading adviser under the Commodity Exchange Act. The output is not personalised investment advice and should not be relied upon as a basis for trading decisions in regulated commodity or securities markets.
Australia
In Australia, this service provides factual information only and does not constitute financial product advice within the meaning of the Corporations Act 2001 (Cth). It does not take into account your individual objectives, financial situation, or needs. Volterra does not hold an Australian Financial Services (AFS) licence. You should obtain independent professional advice before making any financial decision.
United Kingdom
Volterra is not authorised or regulated by the Financial Conduct Authority (FCA). Nothing on this site constitutes investment advice under FCA rules, nor a financial promotion or inducement to engage in investment activity within the meaning of section 21 of the Financial Services and Markets Act 2000 (FSMA). The data is provided to recipients on the basis that they do not rely on it as advice.
European Union, Canada & other jurisdictions
The data does not constitute investment advice, investment research, or a personal recommendation under the EU Markets in Financial Instruments Directive (MiFID II) or any equivalent regime, and is not prepared in accordance with legal requirements designed to promote the independence of investment research. The service may not be appropriate or available in all jurisdictions. You are solely responsible for ensuring that your access to and use of the data complies with the laws applicable to you.
Your responsibility
You are solely responsible for your own investment, trading, hedging, and procurement decisions and for any resulting outcomes. You should obtain independent financial, legal, and tax advice appropriate to your circumstances before acting on any information provided by Volterra.
A permanent copy of this notice is available at mineralvolatility.com/disclaimer.