透過您的圖書館登入
IP:54.163.14.144
  • 學位論文

整合主成分分析與多目標粒子群演算法結合蜜蜂交配演算法運用在投資組合選擇

Integrating Principle Component Analysis with Multi-objective Particle Swarm Optimization and Honey-Bee Mating Optimization for Portfolio Selection

指導教授 : 邱垂昱

摘要


近幾年來,股票市場變動不定,尤其在金融海嘯襲捲之下,空頭氣氛凝重,市場走向相當不明確,因此本研究希望透過資訊技術與啟發式演算法的協助,建立一資訊系統,希望透過資訊系統從台灣五十指數與台灣中型一百指數裡來篩選出投資組合來分散風險,並調整適合的資產配置,讓投資人以少許的金融知識,獲取低風險且獲利穩定的投資方法。 此資訊系統分為兩階段,第一階段主成分分析從近百項財務指標中篩選出具解釋力的少數主成分,並藉由簡化出來的主成分來篩選每季的投資組合,並藉由多目標粒子最佳化演算法結合蜜蜂交配最佳化演算法調整適合的資產配置及技術指標權重,並藉此配置比例進行投資,預估下季報酬率及預測買賣後的股票漲跌,讓投資人不單能以少許的金融知識且貼近實務面,獲取低風險,穩定的報酬。 由實證結果可證明利用主成分分析挑選出投資標的,再由多目標粒子群演算法結合蜜蜂交配演算法最佳化技術指標即資產權種配置,做為下季的買賣決策模型,此方法三年間共獲利23.86%, 平均預測率高達七成以上,不單單報酬率較優於台灣加權指數、台灣五十指數、中型一百指數,且獲得穩定的收益。

並列摘要


In the recently years, the stock market is very unstable in Taiwan; especially after attacked by the worldwide financial crisis, the market trend becomes quite unclear. Under today’s unpredictable stock market, how to use financial instruments to make asset allocation and gain the best profit return is the most important topic to address. Therefore, through the help of information technology and heuristic algorithms, this research aimed to establish an information system to identify the best portfolio out of stocks in Taiwan 50 Index and Taiwan Med-Cap 100 Index so that financial risks could be diversified; furthermore, through investment method of this research, investors with little financial knowledge could obtain an investment portfolio with low risks and stable profits. This research was divided into two stages. First, we adopted principle component analysis to create main components with high explanatory power from hundreds of indicators in financial statements. Through these main components, the best portfolio in each quarter could be identified from Taiwan 50 Index and Med-Cap 100. Second, Multi-objective Particle Swarm Optimization combined (MOPSO) with Honey-Bee Mating Optimization (HBMO) was used to develop appropriate weights of asset allocation and weights of technology indicators. Then, investors with little financial knowledge could apply the derived information from our model to make investment, estimate the return rate of the next quarter, and predict the rise and fall of the purchased stocks The component stocks of Taiwan 50 Index and Mid-Cap 100 Index were our study target, and the time span of this study was from the fourth quarter of 2006 to the third quarter of 2009. Empirical results showed the investment portfolio built in this study would generate a return rate of 23.86% and a prediction rate of 70% over this three-year period, which were superior to those of other indexes such as Taiwan 50 Index and Med-Cap 100. Therefore, we can conclude that this method could not only predict the future trend of the stock market but also obtain stable returns in each quarter while reducing investment risks.

參考文獻


[45] 張景閔,技術分析的切入時點,碩士論文,國立成功大學會計學研究所,台南,2005。
[42] 李舒禾,台灣股市選股策略獲力能力分析,碩士論文,國立成功大學,台南,2004。
[44] 邱怡瑛,質群演算法於多組解方程最佳化問題之研究,碩士論文,私立元智大學工業工程與管理研究所,中壢,2004。
[28] R. C. Eberhart and Y. Shi, “Tracking and optimizing dynamic systems with particle swarms,” Proceedings of the 2001 Congress on Evolutionary Computation, Seoul, Korea, 2001, pp.29-36.
[1] H.M. Markowitz, (1952) ”Portfolio Selection” Journal of Finance 7 77-91

延伸閱讀