利用動態式準則評估於共演化策略解決股票投資決策問題

USING DYNAMIC MULTI-CRITERIA EVALUATION ON CO-EVOLUTION STRATEGIES FOR SOLVING THE STOCK TRADING DECISION PROBLEMS

吳姿婷(Tz-Ting Wu)

指導教授 ： 張應華 ； 楊承亮

大同大學/經營設計學院/資訊經營學系所/碩士(2009年)

摘要

投資組合中風險和報酬是互相依存，獲取期望報酬的同時必須面臨相對風險，然而一般投資者不容易預測各投資標的風險和報酬，常因為不當的資金配置造成無法獲取較佳利潤。股票投資為多準則決策問題，但傳統解決多準則的理論有兩大缺點：一是無法隨著環境的改變而做適當的決策變動，二是各準則間權重的指派過於簡化而不切實際，不符合人類的思維模式。1965年Rechenberg提出演化策略以解決實數參數最佳化的問題，改善傳統演算法只做單點搜尋且存在高可能性落入區域最佳解的缺點。1992年Hillis提出共演化概念，即生物藉由與環境(多準則)的相互演化，不斷改良基因以利生存，並加速演化的計算。
因此本研究整合共演化準則評估模式與演化策略於股票投資的多準則決策問題上，由於共演化式演化策略的進化過程有著自我修正的特性，其準則評估可以隨著時間、環境的改變而做適應性調整，符合人類的決策模式（評估準則的動態變化），解決傳統在求多準則決策問題上的缺點（各準則間權重的指派簡化）。由共演化式演化策略找出最佳資金比例組合，協助投資者在有限的資金下依據最佳分配比例做適當的配置，以獲取報酬。實驗中與一般演化策略和類神經預測模型做比較的結果得知，共演化式演化策略優於一般演化策略，更優於類神經預測模型。共演化評估方式改善一般演算法所採用的簡化適應函數和多準則決策問題的準則偏好權重不能隨著環境改變做適應性調整，使各準則依據各資金配置染色體變動而有所調整，而資金配置染色體也會隨著適應各準則作演化以達成更符合人類決策思考的模式，亦於2004年到2008年各年中找到更好夏普值之資金配置。

關鍵字

最佳化 ； 類神經網路 ； 動態股票投資決策 ； 演化策略 ； 共演化模式

並列摘要

Risk and return in the risk portfolio are interdependent. In acquiring anticipated return, we must face comparative risk at the same time. However, complexity of investment environment and dynamic change of decision-maker’s criteria,makes investors not easy to forecast risk and return of various investment objects, and the investors often fail to acquire better profits because of improper capital allocation. Although stock investment concerns Multi-Criteria Decision-Making (MCDM), such traditional MCDM theory bears two shortfalls: first, it is unable to conduct properly evolving decision-making with changing environments; second, the weight assignment on various criteria becomes too simplified to be realistic, which does not correspond with human thinking pattern.
In 1965, Rechenberg proposed Evolution Strategies to solve optimal problem on real number parameters, and improve the flaws of only engaging point search and existing high probability of falling into optimal solution area in traditional algorithms. In 1992, Hillis introduced Co-Evolutionary Concept, under which living creatures engage interactive evolution with environments (multi-criteria) to constantly improve their genes for survival and thus expedite evolutionary computation. Therefore, this research is aimed to integrate Co-Evolutionary Criteria Evaluation Model with Evolution Strategies so as to solve the Multi-Criteria Decision-Making problems for stock-trading investment. Since the progress course of Co-Evolutionary Evolution Strategies bear a self-calibration nature, its criteria evaluation can be made adaptively adjustment to the changes of time and environment, which not only corresponds with human decision-making pattern (i.e., evaluation on dynamic change of criteria), but also solves the shortfall of Multi-Criteria Decision-Making problems (i.e., simplified assignment on weights among various criteria).
From the Co-Evolutionary Evolution Strategies, we can identify the optimal capital portfolio, and help investors to acquire maximum return according to optimal capital preoperational allocation under limited capital. During the experiment of this research, we compare the General Evolution Strategies with Artificial Neural Forecast Model, and then come up with the result that the Co-Evolutionary Evolution Strategies are better than the General Evolution Strategies, and much better than Artificial Neural Forecast Model. The Co-Evolutionary Criteria Evaluation Model has improved the problem of simplified adaptive functions adopted by general algorithms and the problem of favoring weights but failing to engage adaptively adjustment to the change of environment in the traditional Multi-Criteria Decision-Making. So doing allows various criteria to act adaptively in accord with the change in various capital allocation chromosomes. Also, the capital allocation chromosomes shall adapt themselves to various criteria and evolve to the model more accordant to human thinking pattern. Accordingly, a much better Sharpe Ratio capital allocation was spotted repeatedly in 2004 and 2008.

並列關鍵字

Optimization ； Evolution Strategies ； Co-Evolutionary Model ； Artificial Neural Network ； Dynamic Sock-Trading Decision-Making

參考文獻

[10]姚凱齡，“應用限制滿足式遺傳演算法於股票投資策略制定”，大同大學資訊經營研究所碩士論文，2009。

[11]張宏瑋，“動態式準則評估於旅遊行程安排之最優化”，大同大學資訊經營研究所碩士論文，2007。

[14]余芳佩，“以共演化限制式遺傳演算法解決供應鏈產銷整合問題-以台灣織布業為例”，大同大學資訊經營研究所碩士論文，2008。

[21鄭克文，“應用高效率螞蟻演算法於多其動態資產配置最佳化的研究”，國立暨南國際大學資訊管理研究所碩士論文，2007。

[22]林佳禾，“應用高效率隨機規化演算法於多其動態資產配置最佳化研究”，國立暨南國際大學資訊管理研究所碩士論文，2007。