Abstract
This thesis contributes to the problem of equity portfolio management using computational intelligence methodologies. The focus is on generating automated …nancial reasoning, with a basis in computational …nance research, through searching a space of semantically meaningful propositions. In comparison with classical …nancial modelling, our proposed algorithms allow continual adaptation to changing market conditions and a non- linear solution representations in most cases. When compared with other computational intelligence approaches, the focus is on a holistic design that integrates …nancial research with machine learning. The major aim of the thesis is to develop portfolio allocation techniques for learning investment-decision making that can easily adapt to changes in market processes together with speed and accuracy. We evaluate the algorithms developed in out-of-sample trading framework using historical data sets. The testing is designed to be realistic; for instance, considering factors such as transaction costs, stock splits and data snooping. To demonstrate the robustness of our approach we perform extensive historical simulations using previously untested real market datasets. On all data sets considered, our proposed algorithms signi…cantly outperform existing portfolio allocation techniques, sometimes in a spectacular way, without any additional computational demand or modeling complexity. Before proceeding any further, we stress that setting up abstract and complex mathe- matical models is neither the intention nor the scope of this thesis. Our aim rather is to investigate empirically and possibly capture any existing nonlinearities or non-stochasticities that are apparent in the dynamics of cross sectional returns of stock prices. In doing so we iii utilise some novel techniques, which are mostly based on such methodologies that have been used successfully in the physical sciences were the deterministic dynamics of the phenomena are more easily detected. Our intention is to provide an additional empirical analysis frame- work that could shed new light in the investigation of the nature of financial time-series data generating processes.
Ph.D. (Economics and Financial Sciences)