Non-Gaussian Nonlinear Forecast Reconciliation
Background: Nonlinear constraints in forecasting often involve variables with non-Gaussian distributions, yet existing reconciliation techniques frequently rely on Gaussian assumptions.
Question / Future Work: The extension of nonlinear forecast reconciliation methods to accommodate intermittent, discrete, or mixed-type time series remains an open problem, as current state-of-the-art methods like UKF-based approaches are restricted by Gaussian assumptions.
Why It Matters: Many real-world forecasting applications, such as demand or count-based monitoring, involve non-Gaussian variables that render current nonlinear reconciliation techniques inapplicable.
Evidence: However, our specific algorithm based on UKF is unsuitable to reconcile intermittent or discrete time series with nonlinear constraints due to its underlying Gaussian assumption.
Metadata & Links
- created_at
- 2026-05-02T05:49:29Z