Please use this identifier to cite or link to this item: http://hdl.handle.net/10397/84706
DC FieldValueLanguage
dc.contributorDepartment of Computing-
dc.creatorKwong, Wai-man Raymond-
dc.identifier.urihttps://theses.lib.polyu.edu.hk/handle/200/1172-
dc.language.isoEnglish-
dc.titleIntelligent Web-based agent system (iWAF) for e-finance application-
dc.typeThesis-
dcterms.abstractE-Finance is a complex challenge, requiring complex strategy development and technology implementation. AI techniques applied to stock market prediction include: expert system, fuzzy logic, neural networks, genetic algorithms and some statistical models. But neither one can give confidence for investors to rely on due to the lack of unbiased market movement analysis, trend prediction, human behaviour and psychological implication studies. In this study, we propose a multi-agent framework which combines all the advantages of different forecasting methods, for predicting the best timing of stock market for investment. The system contains five forecasting agents using fundamental, technical and statistical analyses to forecast the market. Each agent is a domain expert in a particular forecasting method and they work together to fill the knowledge gap. As the proposed framework contains different forecasting techniques, we anticipate that the system can provide different measure of the financial market and allow investors to position their investment better and make it more effective. The proposed system consists of five forecasting agents and two non-forecasting agents. Forecasting agents include Fundamental Agent, Technical Agent, Associate Agent, Adaptive Agent and Expert Agent. Different agents give different recommendations in the investment process. Furthermore, we just focus on the construct of 3 agents which are the Coordinate Agent, Technical Agent and Adaptive Agent. The Coordinate Agent is to collect all the recommended strategies from different forecasting agents, leading to a final recommended strategy for users. As different forecasting agents can give rise different prediction results, it is difficult to find a mechanism to handle it. In the literature, the most well known mechanism is majority rule [36] but it can lead to intransitive group preference, which can result in no final recommendation made. We studied how to quantify the recommendation which is originally in terms of 'Buy', 'Hold' and 'Sell' into newly introduced prefer ratio. Using a score table to represent different agents' accuracy in different trading period. Also, two weighting methods had been investigated, they are simple weighting and exponential weighting. Empirical testing shows that the overall system prediction performance can be greatly improved by introducing the Coordinate Agent. The Technical Agent uses technical analysis to predict the market moves. Technical analysis mainly focuses on analyzing the chart patterns, which is a non-trivial task. Because one time scale alone cannot be applied to all analytical chart pattern discovery processes, the identification of typical patterns on a stock price chart requires considerable knowledge and experience. There have been attempts in the last two decades to solve such non-linear financial forecasting problems using AI technologies such as neural networks, fuzzy logic, genetic algorithms and expert systems which may be accurate, but lack of explanatory power or are dependent on domain experts. This agent is a case based reasoning (CBR) agent that can provide an explainable method of financial forecasting that is not dependent on the input of domain experts. Also, in this agent, we have proposed an algorithm, PXtract which identifies and analyses possible chart patterns, making dynamic use of different time windows. We introduce a wavelet multi-resolution analysis incorporated within a radial basis function neural network (RBFNN) matching method that can be used to automate the chart pattern matching process. The automatic process of identifying stock chart pattern is scarce in literature, our proposed algorithm does well and has achieved an identification rate above 80% on average. We believe that it is helpful to investors. The Adaptive Agent uses genetic algorithms and artificial neural networks to predict the market moves. Stock price movements are influenced by many factors and indexes, and they may include changes in gold price, prime rates, deposit call, oil price, exchange rates and other factors. Also, every stock has its own characteristics [38]. Building an artificial neural network with fixed input and topology for all stocks is not feasible to obtain accurate prediction results. This agent provides a genetic approach to address the problem. It includes the input selection, network architecture and the output format. A decision threshold is also introduced to define the best strategy for decision making by investors. Having introduced the decision threshold, we transformed the forecasting problem into a classification problem and got resolved relatively easier with accurate result. Finally, with the integrated multi-agent system for different approaches in solving financial market forecasting problems, we believe that the proposed system can benefit both the experienced and novice investors.-
dcterms.accessRightsopen access-
dcterms.educationLevelM.Phil.-
dcterms.extent142 leaves : ill. ; 30 cm-
dcterms.issued2004-
dcterms.LCSHHong Kong Polytechnic University -- Dissertations-
dcterms.LCSHIntelligent agents (Computer software)-
dcterms.LCSHFinancial services industry -- Computer networks-
dcterms.LCSHElectronic trading of securities-
dcterms.LCSHInternet banking-
Appears in Collections:Thesis
Show simple item record

Page views

49
Last Week
0
Last month
Citations as of Mar 24, 2024

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.