DataScience@BI seminar with Valeria Gargiulo

Mean and Variance Regimes in Threshold VARs: Theory and Regime-dependent Fiscal Multipliers

Abstract

This paper provides the theory for threshold estimators implemented in two different threshold VAR models, both characterized by regime-dependence in the conditional variance. The first model is one where the conditional mean and variances of the TVAR switch following the same regimes, allowing the researcher to incorporate information from breaks in the variance in the identification of the thresholds. The second, more general model, allows the conditional means and conditional variances to evolve according to different regimes driven by possibly different state variables, enabling the researcher to separately study mean- and variance-regimes in the data. I establish consistency for these estimators and show their faster-than-parametric convergence rates. These are then used to analyze regime-dependent fiscal multipliers, contributing to a vast and growing literature. The application follows Ramey and Zubairy (2018). Differently from their work, which imposes theory-motivated thresholds, I estimate them from the data using the developed methodology. Following this approach, I uncover significant differences in the fiscal multipliers across regimes.

• Read more about the speaker
• Visit our website to read more about the Department of Data Science and Analytics