Efficient Bayesian Markov chain Monte-Carlo for Ultra-High-Dimensional Time Series

Macroeconomic dynamics are vast and complex, making Bayesian Vector Autoregressions (BVAR) and their diverse extensions indispensable models for understanding the dynamic interactions within multidimensional time series data. These models are routinely implemented in various statistical software packages such as Matlab, R, and Python. However, current algorithms encounter substantial computational bottlenecks when dealing with ultra-high-dimensional data, i.e., well beyond 100 variables. This limitation stems from a somewhat naive construction within the multivariate linear regression framework, often failing to harness specific features of the models. This project will develop a suite of efficient algorithms tailored for conditionally Gaussian space VARs, with the aim of addressing these computational challenges and unlocking the model's full potential. In collaboration with Dr Aubrey Poon and Dr Alfred Duncan at the University of Kent, I will apply this innovative framework to widely used macroeconomic and time-series models, providing policymakers and practitioners
with powerful new tools for analysis.