Title:	Smooth Time Varying Parameter VARs
DNr:	NAISS 2024/22-1607
Project Type:	NAISS Small Compute
Principal Investigator:	Alessandro Celani <alessandro.celani@oru.se>
Affiliation:	Örebro universitet
Duration:	2024-12-06 – 2026-01-01
Classification:	50201
Keywords:

Abstract

This paper introduces a novel framework designed to mitigate the overparameterization problem frequently encountered in time-varying parameter (TVP) VARs. We present a methodology grounded on the concept of "moderate time variation", where TVP are allowed to vary just over a subset of the time domain. The dimensionality reduction is performed using two ingredients: a fixed weight matrix, and a low-dimensional parameter that evolves over time. Notably, the model includes both the time-invariant VAR and the standard TVP-VAR as special cases. By using more flexible weighting schemes we allow the parameters to evolve smoothly over time. In this paper, we construct the weight matrix via B-Splines. B-Splines have both the characteristics of smoothness and "forgetting" of far observations, which complies with the stylized fact regarding the evolution of macroeconomic dynamics. We illustrate our technique via a comprehensive simulation exercise and three real applications in growing dimensions.