On inference and causality in change point regressions

Gavioli-Akilagun, Shakeel (2023) On inference and causality in change point regressions. PhD thesis, London School of Economics and Political Science.

Text - Submitted Version Download (3MB)

Abstract

Change point analysis, broadly defined, concerns the setting in which one observes time indexed data whose distribution is liable to change at a certain number of unknown locations in time. These locations are known as change points, and data indexed between change point locations can be understood to be in some sense homogeneous. This thesis studies two relatively neglected problems in change point analysis. Namely, statistical inference and causal structure discovery. For the first problem, we propose two methods for recovering disjoint intervals each contain a change point location uniformly at some significance which may be tuned by the user. We focus principally on the piecewise polynomial change point model, in which the data are modeled as weakly dependent noise fluctuating around a piecewise trend. For the second problem we consider a multivariate time series and model change points across the series as arrival times of a marked point process. We introduce a procedure for recovering a graph which encodes causal information about the process, in the sense that an edge in the graph can be (under some conditions) understood as indicating that change points in one time series cause change points in another time series.

Item Type:	Thesis (PhD)
Additional Information:	© 2023 Shakeel Gavioli-Akilagun
Library of Congress subject classification:	H Social Sciences > HA Statistics
Sets:	Departments > Statistics
Supervisor:	Fryzlewicz, Piotr
URI:	http://etheses.lse.ac.uk/id/eprint/4764

Actions (login required)

Record administration - authorised staff only