Studies in Econometric Modelling and High Dimensional Inference

Student thesis: Doctoral ThesisDoctor of Philosophy

Abstract

This thesis examines a number of econometric approaches to advance the state of economic and financial forecasting for the UK economy. Novel techniques are established to accommodate alternative indicators with a particular focus on text information. A vast part of this work is devoted to the development of high frequency text indicators and to the evaluation of their predictive power to key economic variables. We show that textual time series can act as a complement to traditional survey-based data to enhance forecasts of GDP growth, unemployment rate, business investment and consumption mostly over longer horizons as well as during times of distress. The effectiveness of news is further improved when exploiting this information on a more granular level, coupled with the aid of machine learning, regression-like models. Moreover, we extend a standard dynamic factor model to accommodate the arbitrary flow of text information which is a main characteristic of this type of information. To contribute to the debate between large and medium sized models, our study demonstrates that maintaining the unstructured nature of these data is preferred than taking a simple average of these series. A study on structural change is also introduced which refines two existing machine learning algorithms (i.e. neural network and support vector regression) and compares the time-varying version of these models to their fixed co-efficient counterpart. We include theoretical evidence of the time-varying estimators while a variety of interesting macroeconomic and financial applications are provided.
Date of Award1 Jan 2022
Original languageEnglish
Awarding Institution
  • King's College London
SupervisorGeorge Kapetanios (Supervisor) & Martin Weale (Supervisor)

Cite this

'