New methods for time series
File(s)
Author(s)
Palasciano, Henry Antonio
Type
Thesis
Abstract
This thesis explores a range of topics within the vast realm of time series analysis and forecasting, focussing first on locally stationary wavelet processes and then on generalised network autoregressive models. We begin by extending the test for the absence of aliasing and/or white noise confounding developed for locally stationary wavelet processes to the two-dimensional spatial setting. Next, we introduce a continuous-time counterpart to the locally stationary wavelet processes, developing both associated theory and spectral estimation techniques for this new class of models. We then move to multivariate time series, returning to the stationary setting, with a focus on the generalised network autoregressive processes. In this setting, we first use generalised network autoregressive processes to forecast consumer price inflation in the United Kingdom, comparing our results to various benchmarks including the Bank of England. In the final project, we undertake a theoretical comparison of standard vector autoregressive and generalised network autoregressive models, from a bias-variance trade-off perspective.
Version
Open Access
Date Issued
2025-08-06
Date Awarded
01/12/2025
License URL
Advisor
Nason, Guy
Sponsor
Engineering and Physical Sciences Research Council
Publisher Department
Department of Mathematics
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)