NO FULL TEXT AVAILABLE. Access is restricted indefinitely. ----- In stock markets, to assist investors in identifying intrinsic opportunities and to support investors’ decision-making, a host of finance methods (such as fundamental analysis and technical analysis) and information technologies (such as decision-support systems and data mining methods) have been investigated or proposed from multiple disciplines technologically and practically. However, the performance of these traditional technology-based investment methods is limited, because such methods only take into account single-dimensional event factors. For instance, fundamental analysis focuses on fundamental factors, while technical analysis investigates patterns of stock price and volume movement, and psychological finance considers investors’ thinking. Further, these traditional technology-based methods and integration-related studies do not integrate all dimensions and all levels of stock market structures, and neither do they incorporate advantages of different finance methods in a systematic way. Accordingly, such methods have limited value in investment decision-making, and they may not meet the needs of the finance industry. These limitations derive from different and incompatible features of different dimensions of stock market structures and different investment methods.
This thesis shows how the integration of multi-dimensional event factors can be achieved and thus offer improved performance. It proposes a novel three-layer integrated investment decision- support framework composed of Analysis, Synthesis, and Investment Decision Support. At the first layer, multi-dimensions and multi-levels of stock market structures are identified, including intra-dimensions (such as unique trends of stocks, investors’ demand and supply, and fundamental factors), and inter-dimensions (such as a two-way reflexivity model of investors’ decisions and market reactions). The dimensions are modeled using concepts of pattern components, which are basic unit components that derive from each dimension of stock market structures and that are interpretable, understandable, usable, reusable and linkable to other dimensions. At the second layer, working and upcoming multi-dimensional pattern components are synthesized to reveal real and potential market situations, thereby indicating investment opportunities. At the third layer, a prototype of Decision Support System is designed to integrate the functions of the first two layers for investment decision support.
This integrated framework recognizes and covers multi-dimensions and multi-levels of market structures, as well as incorporating concepts and advantages of conventional investment methods. The framework is promising, because experimental results show that it outperforms single-dimensional traditional methods and market baselines.
In this integrated framework, I focus on two key aspects that previous studies have neglected: unique trends of stocks - unique patterns which relate only to individual stocks, and a two-way reflexivity model of investors’ decisions and market reactions. Identification of unique trends of stocks is important, as unique trends indicate characteristics and unique patterns of individual stocks, which are integral to further synthesis and investment decision-making (layers 2 and 3 of the framework). However, related conventional technologies (such as the AutoSplit method) do not integrate finance domain knowledge and so their results (such as hidden trends or variables) are non-interpretable, non-usable and non-reusable. To address this problem, I propose a novel Domain Knowledge-dependent AutoSplit method which can identify unique trends by integrating finance domain knowledge (such as pattern components identified) at its preprocessing, processing and post-processing stages. My performance studies show that the unique trends identified are interesting, interpretable and useful for trading practitioners and finance academics.
Additionally, the novel two-way reflexivity model of investors’ decisions and market reactions reveals an important aspect of stock markets - investors make decisions based on market situations, and the market then reacts to their decisions (especially influential investors’ decisions) in a two-way reflexivity mechanism. Modes and applications of the model were investigated in a novel way using Vector Auto-Regression methods. Experiments show that the model plays an important role in investment decision-making, and is of particular value in portfolio setup and risk controls.