High-dimensional functional time series modelling of environmental datasets

In a dynamic time-series model, selecting the appropriate lag order is crucial for understanding the predictive relationships among variables and generating accurate forecasts. The standard approach involves minimizing an information criterion (IC) over a range of possible lag orders to estimate the optimal lag-length. However, relying on ICs most often imposes unnecessary constraints on temporal interactions by assuming a uniform lag order across all covariates. This limitation becomes particularly evident when exploring or forecasting complex dynamics, such as those encountered in climate change studies where the length of impulse-response interactions is inherently heterogeneous.

In this project, we intend to develop an innovative methodology for exploring high-dimensional time series dynamics by integrating time series analysis with functional data analysis methods, all within the Bayesian framework. We envision to apply these models to large datasets of climate and economic data with the goal of identifying the predictive effects of climate trends on global economic outcomes and vice versa.

The ideal candidate will be interested in time series analysis, high-dimensional statistics, functional data analysis and Bayesian statistics. Background in at least one of the above subjects will be beneficial, but candidates with other backgrounds will be considered.

The project is in collaboration with Dr Luca Margaritella, Lund University