標題: 建構以事件評估為基礎之分類元系統於股市交易事件之實證研究
Construction of Event-Evaluation-Based Classifier System in the Stock Market Transaction Event of the Empirical Study
作者: 游政霖
Yo, Janq-Lin
陳安斌
Chen, An-Pin
管理學院資訊管理學程
關鍵字: 分類元系統;事件評價;事件評估;知識挖掘;法人籌碼;除權息;股市交易;classifier system;event evaluation;knowledge mining;ex-rights;ex-dividend;stock market trading
公開日期: 2009
摘要: 金融市場為一變化迅速且知識不斷演化的動態環境,充滿各種影響價格走勢的事件。因此,如何迅速掌握金融事件的脈動,並迅速評估事件資訊的價值,篩選出具有交易價值的交易知識,掌握最佳時機投入市場進行交易以獲取利潤,成為本研究之目標。 本研究之目的為運用人工智慧方法學中的分類元系統,建構一套以事件評估為基礎,具有知識探勘、評估及演化功能的智慧型金融投資決策系統,稱之為「事件分類元系統(ECS)」。 本研究首先發展一套事件評估模式作為ECS運作之核心,其可提供具有預測性質之出場目標價位(或報酬率)的功能。在該模式為基礎下,採用人工智慧方法學中兼具基因演算法擇優演化能力及專家系統功能之分類元系統的架構,建構具有提供事件評估之知識演化功能及提供知識事件交易資訊之專家系統功能的事件分類元系統,以迅速提供精準且智慧化的金融投資交易資訊。為驗證所發展之方法與系統是否具有可行性與有效性,本研究透過建立法人籌碼面事件策略模組及除權息事件策略模組,設計及開發出事件分類元交易系統,並採用投入真實資金(新台幣3,300萬)依據ECS的交易訊號在台灣股票市場進行期間約8個月的實際交易之實證研究方法。 實證結果顯示,以報酬率、波動率及夏普指數衡量,本研究之事件分類元系統績效均優於同時期之台灣加權股價指數。
This study used classifier system of artificial intelligence methodology to construct an intelligent financial investment decision system, which is based on Event-Evaluation model, and has functions of knowledge mining and knowledge evolution. We named it as "Event Classifier System (ECS)". First, we have developed an Event-Evaluation model as the operational core of ECS. Second, we developed the ECS transaction system with user-friendly interface and intelligent operational functions for traders. Third, for the reason of application, two trading modules were designed and developed: institutional investors’ chips event strategy trading module and ex-dividend event strategy trading module. Final, to verify the effectiveness of the system, we invested real money (about NTD 33 million) into the Taiwan’s stock market, and according to the ECS's signal to do the actual trading of empirical experiments about 8 months. By comparison of the three performance indicators: the rate of return, volatility and Sharpe ratio, the empirical results show that the performance of ECS is better than Taiwan weighted stock index during the same period.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT079264512
http://hdl.handle.net/11536/40462
Appears in Collections:Thesis