###Flexible modelling of high-dimensional multivariate time series in macroeconomics
Project ID: 2228bd1134 (You will need this ID for your application)
Research Theme: Mathematical Sciences
UCL Lead department: Statistical Science
Lead Supervisor: Jim Griffin
Project Summary:
Multivariate time series naturally arise in areas of macroeconomics where interest centres on the inter-relationship between a set of macroeconomic indicators. Vector autoregressive (VAR) models and time-varying parameter versions (TVP-VAR) have been commonly used to model these type of data. This allows forecasting of the macroeconomic indicators or the effects of interventions (such as changes to interest rates) on the indicators through impulse response function. VAR models assume a linear model and, usually, normal errors. These assumption are often not supported by data and can lead to poor performance. TVP-VAR can provide better forecasting performance by allowing the parameters of the VAR to evolve over time. However, they are prone to overfitting (particularly, in high-dimensional settings with large numbers of indicators) which can lead to poor predictive performance.
This project will build on recent work in the Bayesian literature which addresses the problems of overfitting and inflexible models for multivariate time series. Overfitting can be addressed by using dimension reduction methods such as shrinkage priors or tensor decomposition to avoid overfitting in high-dimensional linear models. Inflexibility of the VAR models can be addressed by using a Bayesian nonparametric version of VAR models or a model based on Bayesian additive regression tree. In this project, you will develop and combine these two streams of research to allow the estimation of flexible models for high-dimensional multivariate time series.
The project is supervised by Jim Griffin who is Professor of Statistical Science at UCL. He has worked extensively on the development of Bayesian nonparametric methods, shrinkage methods for inference in regression models with many regressors, and their application to problems in economics and finance. Student should have a strong background in a highly numerate subject and have a good knowledge of statistics.