Efficient group stock portfolio optimization algorithms with natural group detection

機構典藏 > College of Engineering > Graduate Institute & Department of Computer Science and Information Engineering > Thesis > Item 987654321/114683

Please use this identifier to cite or link to this item: https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/114683

Title:	Efficient group stock portfolio optimization algorithms with natural group detection
Other Titles:	具群組偵測之有效率的群組股票組合最佳化演算法
Authors:	喬納森;Coupe, Jonathan
Contributors:	淡江大學資訊工程學系全英語碩士班陳俊豪
Keywords:	分群效度指標;多樣群組股票投資組合;群組遺傳演算法;分組問題;組合最佳化;Cluster validation indices;diverse group stock portfolio;grouping genetic algorithm;grouping problem;portfolio optimization
Date:	2017
Issue Date:	2018-08-03 15:00:39 (UTC+8)
Abstract:	由於金融市場的多變性，投資組合最佳化至今仍是相當吸引人的研究主題。過去幾十年間，許多不同的演化式演算法針對不同的投資組合亦不斷的被提出，其中一種就是多樣群組投資組合。然而，本研究發現現存的多樣群組投資組合最佳化技術仍有三個問題待解決，分別是：如何設定合適的群組數、演化過程耗時、投資組合風險差異過高。故為解決這些問題，本論文使用群組遺傳演算法提出兩個多樣群組股票投資組合最佳化方法。第一個方法主要用於解決前兩個問題。針對設定合適的群組數問題，所提的方法首先透過股東權益報酬率(ROE)與本益比(P/E)兩屬性將股票分群後，之後設計分群效度因子(cluster validation factor)並將之當成適合度函數的一部份，使演算法可以自動搜尋較佳之分群結果。為解決演化過程耗時問題，在方法一則設計暫存染色體(Temporary chromosome)透過降低需要評估的組合數使演化過程得以加速。第二個方法則著手解決投資組合風險差異過高問題。首先，方法中設計風險比例因子(Risk ratio factor)計算多樣群組股票投資組合可產生之最大組合風險。接著，所提的演算法結合自我調適之交配(Adaptive crossover)、突變(Adaptive mutation)運算與排序為基礎之輪盤選擇法(Rank-based roulette wheel selection)以達更好的搜尋能力。最後，實驗透過31與50家公司之真實股市資料驗證所提方法的效率與效能確實優於現存的最佳化技術。與現存方法比較顯示，方法一不但在獲利上可達到相似結果，在執行時間上亦可減少約原來的百分之八十五。而方法二所找出之多樣群組投資組合其風險上也有明顯降低。 Due to the variety of financial markets, stock portfolio optimization is an attractive research topic. In the past decades, many evolutionary-based algorithms have been proposed to optimize different types of stock portfolios, and one of them is named diverse group stock portfolio (DGSP). However, this study found that three problems remain to be solving in the existing DGSP approaches. They are how to set an appropriate group size, evolution process is time-consuming, and difference between risks of portfolios is too high. To solve these problems, two approaches by grouping genetic algorithms (GGA) are proposed for optimizing DGSPs in this thesis. The first approach is used to deal with the first two problems. For setting an appropriate group size, the two attributes, Return on Equity (RoE) and Price Earnings Ratio (P/E), are utilized to group stocks. Then, cluster validation factor, which is used as a part of fitness function, is designed to derive better stock groups. To solve time-consuming problem, a temporary chromosome is designed to reduce number of stock portfolios should be evaluated to speed up the evolution process. The second approach is then proposed to handle the third problem. It first designs risk ratio factor to calculate the maximum risk of a given DGSP. Then, by combining adaptive crossover, adaptive mutation, and rank-based roulette wheel selection, the second approach has higher searching ability to find better solution. At last, experimental results on the two real datasets that contain 31 and 50 stocks were made to verify the two proposed approaches are effective and efficient. Comparing with the existing approach, the results show that the first approach can not only reach similar return but also reduce execution time up to 85%. The risk of optimized DGSP by second approach is significantly lower than that by the existing approaches.
Appears in Collections:	[Graduate Institute & Department of Computer Science and Information Engineering] Thesis

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	147	View/Open

Loading...