Should I stay or should I go? A latent threshold approach to large-scale mixture innovation models

Abstract

This paper proposes a straightforward algorithm to carry out inference in large time-varying parameter vector autoregressions (TVP-VARs) with mixture innovation components for each coefficient in the system. We significantly decrease the computational burden by approximating the latent indicators that drive the time-variation in the coefficients with a latent threshold process that depends on the absolute size of the shocks. The merits of our approach are illustrated with two applications. First, we forecast the US term structure of interest rates and demonstrate forecast gains of the proposed mixture innovation model relative to other benchmark models. Second, we apply our approach to US macroeconomic data and find significant evidence for time-varying effects of a monetary policy tightening.

References in corpus (3)

• Ancillarity-Sufficiency Interweaving Strategy (ASIS) for Boosting MCMC Estimation of Stochastic Volatility Models
• Dealing with Stochastic Volatility in Time Series Using the R Package stochvol
• Dynamic sparsity on dynamic regression models

Cited by in corpus (7)

• Sparse Bayesian time-varying covariance estimation in many dimensions
• Sparse Bayesian vector autoregressions in huge dimensions
• Dynamic Shrinkage Processes
• Forecasting macroeconomic data with Bayesian VARs: Sparse or dense? It depends!
• A loss discounting framework for model averaging and selection in time series models
• Selective linear segmentation for detecting relevant parameter changes
• Introducing shrinkage in heavy-tailed state space models to predict equity excess returns