Datawetenschap in financiële investeringen

In recent years, the remarkable progress of data science, fueled especially by the significant power of neural networks and deep learning in feature extraction and pattern recognition, has increasingly attracted attention from the field of quantitative financial investment. However, current deep learning based financial research focuses on stock prediction, while exploration in other domains, such as index tracking and green AI-based stock forecasting, remains limited. This research gap serves as the starting point of this dissertation. In Chapter 1, we introduce fundamental concepts of financial investment and data science and present the motivation and research objectives, providing a foundation for subsequent chapters. Chapter 2 performs a systematic literature review of index tracking, which reviews past relevant research works and points out potential future research directions. Chapter 3 proposes a one-dimensional pointwise autoencoder to capture latent characteristics of the market and employs Shapley additive explanations to select component stocks to create tracking portfolios. Chapters 4 and 5 present two sparse Softmax functions to solve the index tracking problem in an end-to-end way and financial characteristics of portfolios based on different approaches are explored in Chapter 5. In Chapter 6, a lightweight model is designed using only convolutional neural networks achieving better prediction and investment performance compared to other state-of-the-art models. In conclusion, this dissertation provides valuable insights into the use of data science-based solutions for financial investment. On the one hand, index tracking research with deep learning is extended significantly, enabling financial investors to obtain better and more robust investment performance. On the other hand, a lightweight stock prediction model is proposed while considering green AI, inspiring researchers to take into account the influence of AI techniques when they design new models.