Title:	Locally stationary seasonal ARMA models
DNr:	NAISS 2026/4-20
Project Type:	NAISS Small
Principal Investigator:	Ganna Fagerberg <ganna.fagerberg@stat.su.se>
Affiliation:	Stockholms universitet
Duration:	2026-01-12 – 2027-02-01
Classification:	10106
Homepage:	https://www.su.se/forskning/forskningsprojekt/bayesiansk-inferens-för-storskaliga-data-genom-frekvensbaserad-sampling-teori-och-tillämpningar
Keywords:

Abstract

We introduce a locally stationary seasonal ARMA framework with dynamic global–local shrinkage priors capable of capturing multiple seasonal periodicities. The regular and seasonal parameters are parameterized to ensure stationarity and invertibility at every time point. Posterior inference is conducted via a Gibbs sampler employing forward–filtering backward–sampling with nonlinear Kalman filter updates for the time-varying parameters. We analyze the model’s behavior at the boundaries of the parameter space and adjust the specification accordingly to ensure robust posterior inference. Finally, we plan to evaluate the forecasting performance of the proposed model relative to established alternatives.