Bayesian forecasting
Awen.
Forecasting that knows what it knows — and shows its working.
A research project from the world of state-space forecasting
Awen is a platform for dynamic linear models and sequential Bayesian learning — built so every forecast arrives with its assumptions, its diagnostics, and its uncertainty in the open. At its core is DLMAX, a JAX-based library for dynamic linear modelling with dynamic model averaging.
What it does
Dynamic linear models. A flexible component-based builder — local level, trend, seasonal, and regression components — composed into state-space models and fitted with a fast, numerically stable square-root filter.
Dynamic model averaging. Rather than committing to one model, DLMAX maintains a universe of candidates and weights them by their out-of-sample performance as data arrives — adapting which structure is in favour over time.
Honest uncertainty. Forecasts come with calibrated predictive intervals and the diagnostics to judge them, not just point estimates.
Tools and writing to follow. For enquiries, hello@awenforecasting.com.