A Study on Mining Group Stock Portfolio by Using Grouping Genetic Algorithms

• Main Authors: Cheng-Bon Lin, 林政邦
• Other Authors: Chun-Hao Chen
• Format: Others
• Language: en_US
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/2n2ctc

碩士 === 淡江大學 === 資訊工程學系碩士在職專班 === 104 === Due to variance of financial market, the financial data mining is always an attractive research issue and a real challenge to researches. For example, stock price prediction and stock portfolio mining are topics of financial data mining. It is easily to understand that the stock portfolio mining problem is an optimization problem. Hence, in the past decades, based on genetic algorithms, many approaches were proposed to deal with it. However, the problem of them is that only one stock portfolio is suggested. When only one stock portfolio is provided, some problems may happen in real application. For example, investors may think the price of the suggested stock is too high to buy, or the stock price reach the daily limit such that investor cannot buy it. Besides, stock portfolio mining should not only consider the objective criteria but also investors'' subjective criteria. The objective criteria are return on investment (ROI) and value at risk (VaR). This thesis two approaches for solving the mentioned problems. Firstly, we propose a grouping genetic algorithm based approach for mining group stock portfolio (GSP) and its goal is to divide n stocks into K groups. Stocks in the same group means that they have similar properties. To achieve this goal, a chromosome consists of three parts. They are grouping part, stock part and stock portfolio parts. The grouping and stock parts are used to represent how n stock are divided into K groups. For each group in the stock portfolio part uses two real number to indicate whether it is purchased group and it’s purchased units. Each chromosome is then evaluated by the group balance and portfolio satisfaction. The group balance is used to make the number of stocks in groups can as similar as possible. The portfolio satisfaction is utilized to measure the satisfaction degree of objective and subjective criteria of a chromosome. As a result, the derived GSP can provide various stock portfolios to investors. Then, to improve the similarity of groups and profit stability, the second algorithm has been proposed. In second approach, the price balance and unit balance are designed to make the stock prices of stocks and purchased units in groups can as similar as possible. At last, the two proposed algorithms are verified on 31 stocks which are selected from Taiwan 50 ETF, including the derived GSP analysis, impact of the fitness functions to the derived GSPs and the ROI of the derived GSPs. The experimental results show that the two proposed algorithms can mine GSPs that provide higher ROI than benchmark.