DTA

Digital Theses Archive

 

Tesi etd-11042021-215232

Type of thesis
Corsi integrativi di II livello
Author
BOUCHS, ALESSANDRO
URN
etd-11042021-215232
Title
Forecasting Financial Volatility with High Dimensional Data: A Deep Learning Approach
Structure
Cl. Sc. Sociali - Scienze Economiche
Course
SCIENZE ECONOMICHE E MANAGERIALI - SCIENZE ECONOMICHE E MANAGERIALI
Committee
relatore Prof. FAGIOLO, GIORGIO
Relatore Prof. Giuseppe Ragusa
Presidente Prof. BARONTINI, ROBERTO
Membro Prof. MINA, ANDREA
Membro Prof. CINQUINI, LINO
Membro Dott. Andrea Vandin
Keywords
  • deep learning
  • machine learning
  • neural networks
  • realized volatility
Exam session start date
01/12/2021;
Availability
parziale
Abstract
This project tackles the issue of dimension reduction and forecasting with high dimensional financial data. The most common dimension reduction technique is Principal Component Analysis, which forms the backbone of Dynamic Factor Models and is based on linear transformations of the data. Yet, scholars have argued that non-linear structures in the data could be exploited to obtain an improved representation of the original data and more accurate forecasts. In this project, I experiment the use of neural networks, part of the arsenal of techniques coming from deep learning, as a dimension reduction method that exploits non-linear interactions. Using short-term signals from the order book data of 107 stocks observed in 3380 10-minute windows, I derive a measure of realized volatility for each window. This forms the basis to estimate factors by Principal Component Analysis and also by training several autoencoders with differing depth. Subsequently, I generate factor-augmented forecasts of the volatility of a number of stocks in the following 10-minute windows and compare the results in order to assess both methodologies against one another and against traditional models used in the literature.
Files