Applications of Deep Learning Models in Financial Forecasting

In financial markets, deep learning techniques sparked a revolution, reshaping conventional approaches and amplifying predictive capabilities. This thesis explored the applications of deep learning models to unravel insights and methodologies aimed at advancing financial forecasting. The crux of the research problem lies in the applications of predictive models within financial domains, characterised by high volatility and uncertainty. This thesis investigated the application of advanced deep-learning methodologies in the context of financial forecasting, addressing the challenges posed by the dynamic nature of financial markets. These challenges were tackled by exploring a range of techniques, including convolutional neural networks (CNNs), long short-term memory networks (LSTMs), autoencoders (AEs), and variational autoencoders (VAEs), along with approaches such as encoding financial time series into images. Through analysis, methodologies such as transfer learning, convolutional neural networks, long short-term memory networks, generative modelling, and image encoding of time series data were examined. These methodologies collectively offered a comprehensive toolkit for extracting meaningful insights from financial data. The present work investigated the practicality of a deep learning CNN-LSTM model within the Directional Change framework to predict significant DC events—a task crucial for timely decisionmaking in financial markets. Furthermore, the potential of autoencoders and variational autoencoders to enhance financial forecasting accuracy and remove noise from financial time series data was explored. Leveraging their capacity within financial time series, these models offered promising avenues for improved data representation and subsequent forecasting. To further contribute to financial prediction capabilities, a deep multi-model was developed that harnessed the power of pre-trained computer vision models. This innovative approach aimed to predict the VVIX, utilising the cross-disciplinary synergy between computer vision and financial forecasting. By integrating knowledge from these domains, novel insights into the prediction of market volatility were provided.