Adamopoulos, Konstantinos (2024) Deep learning techniques in time series econometrics: forecasting, data generation, and model selection. PhD thesis, Birkbeck, University of London.
![[img]](https://eprints.bbk.ac.uk/style/images/fileicons/text.png) |
Text
Adamopoulos K, thesis for library.pdf Download (3MB) |
Abstract
This thesis examines deep learning techniques in time series econometrics, considering both theoretical frameworks and empirical applications. The design, computational implementation, and estimation of the three main neural network architectures are discussed, and their performance is evaluated across a range of equity market indices. The second core chapter introduces a conditional Generative Adversarial Network model for the generation of univariate time series. The term conditional indicates that the time series is generated based on an additional predetermined characteristic, allowing for the customization of the process to generate data within predefined regimes of interest. The efficacy of the proposed approach is demonstrated through a Monte Carlo simulation to generate AR(1) processes and in an empirical setting for the generation of stock market index data. The last chapter of this thesis leverages the properties of a fully estimated Generative Adversarial Network, and introduces the Adversarial Discriminator, which is a novel goodness of fit measure and model selection criterion. The primary advantage of the proposed approach is that it relies solely on the fit of a candidate series to the target time series of interest, without requiring any assumptions about the potentially high-dimensional independent variables. A Monte Carlo experiment evaluates its performance in estimating the true number of variables (factors) in models, and the results of an empirical application for evaluating linear asset pricing models align with conventional model selection criteria.
Metadata
| Item Type: | Thesis |
|---|---|
| Copyright Holders: | The copyright of this thesis rests with the author, who asserts his/her right to be known as such according to the Copyright Designs and Patents Act 1988. No dealing with the thesis contrary to the copyright or moral rights of the author is permitted. |
| Depositing User: | Acquisitions And Metadata |
| Date Deposited: | 14 Feb 2025 14:55 |
| Last Modified: | 01 Apr 2025 02:38 |
| URI: | https://eprints.bbk.ac.uk/id/eprint/54988 |
| DOI: | https://doi.org/10.18743/PUB.00054988 |
Statistics
Additional statistics are available via IRStats2.
Archive Staff Only (login required)
