Causal Systems builds causal, mechanism-first models of real-world systems that convert upstream signals into forecasts, pricing layers, or risk feeds for markets, energy, agriculture, and policy. The company validates every model on out-of-sample data and publishes its methodology and results, including failures, to demonstrate rigor. Its research demonstrates, for example, a five-month causal lead from fertilizer prices to US consumer food inflation, with Granger p-value of 0.004.
- Artificial Intelligence
- Software Only
Funding
Founders
Product
Problem
Conventional forecasting relies on historical correlations that break down when market regimes shift, leaving traders, insurers, and policymakers blind to impending shocks. These models treat every variable as a single, static signal, failing to capture the distinct causal chains and time lags that actually drive prices, harvests, and supply chains.
Solution
Causal Systems provides causal, mechanism-first models that start from the structure of what drives what, rather than from recent correlations. The company identifies the causal chain, validates it on real out-of-sample data, and publishes the methodology and results—including negative findings—to prove the mechanism. It then delivers a licensed, documented, and monitored model, which may take the form of a signal, a forecast feed, or a full pricing layer that plugs directly into a client's existing stack. Models are built for specific real-world systems, such as energy and power markets, soft commodities, and global food security, with lead times that are empirically measured across the full chain.
Target Audience
Primary customers are power and commodity traders, asset and fund managers, insurers and reinsurers, and supply-chain and logistics firms, alongside policy bodies, humanitarian organizations, and global health and food-security agencies that need early warning on systemic risks.
Features
- Models built from causal structure identification, then validated with out-of-sample data and published (including failures) for transparency
- Empirical lead-lag analysis, as demonstrated by a fertilizer-to-food chain showing a five-month lead to US CPI with Granger p=0.004
- Cross-asset news-sensitivity analytics across 33 instruments and seven asset classes, revealing horizon-dependent response patterns (e.g., equities peak same-day, commodities at one week, crypto has negligible response)
- Tail-risk alerting focused on significant upstream moves rather than continuous regression, ensuring models are useful in volatile regimes
- Research repository (CS/RES series) covering topics from rare-earth supply chains to EU carbon prices, illustrating model applications across sectors