Multi-agent and artificial neural networks prediction framework development for stock investment strategy

• Main Author: Phang, Wai San
• Format: Thesis
• Language: English
• Published: 2016
• Subjects: HC Economic history and conditions
• Online Access: https://eprints.ums.edu.my/id/eprint/26595/1/Multi-agent%20and%20artificial%20neural%20networks%20prediction%20framework%20development%20for%20stock%20investment%20strategy.pdf; https://eprints.ums.edu.my/id/eprint/26595/
• Institution: Universiti Malaysia Sabah

In personal wealth management, it is necessary to have a plan before making investment in order to ensure a profitable return for the investors. The process of generating an investment portfolio with good investment options is complex as it needs to consider a lot of parameters such as the track record of the companies, the company’s revenue projection, the risk assessment, the political conditions and the nature of business. In this case, a multi-agent framework can be applied to solve the problem. This thesis focuses on the development of a multi-agent framework for wealth management particularly on stock market investment. The core objective is to develop an Intelligent Investment Planner which utilizes multiple agents that work together to plan, predict, assemble and generate a profitable investment portfolio for its investor. Kuala Lumpur Stock Exchange (KLSE) was selected as the targeted stock market. Four types of agents were developed, including the Web Mining Agent (WMA), the Wealth Forecasting Agent (WFA), the Strategy Agent (SA), and the Wealth Planning Agent (WPA). WMA comprises of an algorithm for web mining which enables it to mine and extract semi-structured information and create new structured information using ontology. The ontology developed is not limited to just a knowledge base to store data in structured format but it plays an important role as an inference sources in the decision making of the buying and selling of stock by performing fundamental analysis. WFA consists of a forecasting model to predict the stock price. This work involves the investigation of the performance of different classifiers (established/combinations/new prediction methods) that are used in stock market prediction. Artificial Neural Network (ANN) was chosen as the target candidates for the forecasting model in this work because of its ability to solve complex problems such as the stock price prediction. Feed Forward Neural Network (FFNN), Elman Recurrent Neural Network (ERNN), Jordan Recurrent Neural Network (JRNN) and Ensemble Neural Network (ENN) were tested in the experiments. Based on the results, ENN outperformed the other ANNs and so it was used in the stock market prediction. SA is responsible to generate the buy-sell signal based on the predicted stock prices. WPA generates the investment portfolio based on the buy-sell signal and the fundamental analysis of stock. It selects potential stocks based on investor’s preferences and passes these potential stock candidates to WFA for stock price prediction. In turn, WPA decides on a suitable trading strategy that gives the most profitable investment returns and presents the investment portfolio to the investor. Several experiments were conducted to investigate the performance of the Intelligent Investment Planner in different environments using two trading strategies and the results obtained showed that the proposed planner was able to generate a profitable investment portfolio.