Data pipeline, causal discovery and DAG
Everything the model learns from is public, trusted, and lands on a weekly panel. The causal graph is built before any forecast is fitted, so only drivers with a defensible mechanism are considered.
This is what we’re forecasting and explaining the movements of.
Role in the framework: Target variable.
How hard refineries are running.
Role in the framework. Supply-side driver.
Crude oil going into refineries, before it becomes a specific product
Role in the framework. Upstream shared cause.
Product-specific volumes once refining is done
Role in the framework. Supply-side driver.
Raw domestic extraction, upstream of everything else
Role in the framework. Upstream shared cause.
Barrels sitting in storage, by fuel
Role in the framework. Supply-side driver.
Cross-border flows, by fuel
Role in the framework. Supply-side driver.
How much the economy is producing: output, factory activity, and sentiment
How much people are actually driving: the strongest observed driver of gasoline and jet demand
Inflation and the cost of borrowing
What refiners pay for crude oil, before refining
What drivers pay at the pump: downstream of everything else
Bulk benchmark prices traders and refiners deal in, upstream of the pump
The dollar's broad strength
National population-weighted heating/cooling demand proxy
