Abstract:Nonstationary time series data exist in various scientific disciplines, including environmental science, biology, signal processing, econometrics, among others. Many Bayesian models have been developed to handle nonstationary time series. The time-varying vector autoregressive (TV-VAR) model is a well-established model for multivariate nonstationary time series. Nevertheless, in most cases, the large number of parameters presented by the model results in a high computational burden, ultimately limiting its usage. This paper proposes a computationally efficient multivariate Bayesian Circular Lattice Filter to extend the usage of the TV-VAR model to a broader class of high-dimensional problems. Our fully Bayesian framework allows both the autoregressive (AR) coefficients and innovation covariance to vary over time. Our estimation method is based on the Bayesian lattice filter (BLF), which is extremely computationally efficient and stable in univariate cases. To illustrate the effectiveness of our approach, we conduct a comprehensive comparison with other competing methods through simulation studies and find that, in most cases, our approach performs superior in terms of average squared error between the estimated and true time-varying spectral density. Finally, we demonstrate our methodology through applications to quarterly Gross Domestic Product (GDP) data and Northern California wind data.

Modeling of Stationary Periodic Time Series by ARMA Representations

On maxima of periodograms of stationary

Modeling for a Class of Nonstationary ARMA Series 1

Periodically stationary multivariate autoregressive models

Unified Analysis of Periodization-Based Sampling Methods for Matérn Covariances

Multidimensional Rational Covariance Extension with Applications to Spectral Estimation and Image Compression

Multidimensional Rational Covariance Extension with Approximate Covariance Matching

Positive time series regression models: theoretical and computational aspects

Spatio-Temporal Stationary Covariance Models

Autoregressive Times Series Methods for Time Domain Astronomy

Bayesian Circular Lattice Filters for Computationally Efficient Estimation of Multivariate Time-Varying Autoregressive Models

Parameter-Covariance Maximum Likelihood Estimation

Time-Varying Multi-Seasonal AR Models

On several properties of a novel class of generalized Humbert autoregressive moving average process

An irregularly spaced ARMA(1,1) model and an application to contamination data

Cyclical Long Memory: Decoupling, Modulation, and Modeling

Multivariate Matérn Models -- A Spectral Approach

The fundamental properties of time varying AR models with non stochastic coefficients

General estimation results for tdVARMA array models

Binomial ARMA count series from renewal processes

On the Approximability of Stationary Processes using the ARMA Model