The question
From hourly noise to daily signal.
Commodity prices move through regimes. A single forecasting model that is good in calm markets tends to break exactly when a shock makes the forecast matter most.
Approach
How it works
Five neural forecasters. Five neural time-series models are trained as ensemble members.
Regime-aware, online-adaptive blending. The ensemble adapts its weighting online, conditioned on the detected market regime.
Statistical validation. The ensemble is statistically validated, including through a live +103% geopolitical price shock in Brent crude.
Results & insight
What came out of it
Full results will be published with the preprint: “From Hourly Noise to Daily Signal: A Regime-Aware, Statistically Validated Ensemble for Brent Crude Oil Forecasting.”
Stack & sources
Under the hood
Pythonneural time-series models