Applications of Deep Learning to Financial Time Series Forecasting

Abstract

Deep learning has recently risen as a dominant technique in a variety of settings comprising large-scale and high-dimensional data. In the particular case of financial modeling, one of the most important data analysis problems consists of predicting the future volatility of a given asset. In this thesis, we investigate how the Transformer architecture performs at the task of volatility forecasting by comparing its performance against that of previously explored deep learning architectures such as the LSTM.