台股開盤異常行為分析--結合自組織特徵映射圖神經網路與類神經網路

Abstract

本研究旨在探討亞洲日韓股市能量傳遞到台灣股市時，是否有其合理範圍，並觀察台股開盤在日韓股市開盤影響下所形成的價格偏離，是否真的有自我調節修正的行為，讓股市能真正反映本質，展現其合理價位。首先應用類神經，輸入稍早開盤的日韓股市開盤變化率進行訓練，確實可有效掌握台股當日開盤的漲跌方向，門檻值的訂定也確實有助於方向預測的提升。在方向實驗的基礎支持下，本系統開始嘗試掌握區間，定性定量的實驗設計方便對系統區間掌握度及績效作實際的觀察，同時統計股市開在系統預測區間又修正回系統合理區間或移往系統合理區間方向的機率，觀察市場是否真有自我調節的力量，接著系統透過全面的搜尋，試圖找出符合當時環境的合理區間大小及最大持有時間，同時加入了停損機制與透過自組織映射組織圖的分群的篩選，以優化系統績效，接著本系統進行與隨機買賣組的比較，本系統最終在不同年度進行實驗，以觀察系統的適用性。實驗結果證明，透過搜尋確實能找出適用於現今環境的參數設定，市場隱含的修正力量確實存在，而停損機制也確能提高績效，然自組織映射圖的篩選後績效並未提升，推估所輸入的因子並未掌握到市場相似的修正行為，本系統的績效優於對照組，而系統定義的獲利方式又來自市場行為的合理修正，間接證明系統掌握了開盤前期的市場修正行為，而系統在不同年度皆能對方向作一定程度的掌握，再次驗證人工智慧方法學—類神經網路確實能動態適應，有效掌握市場的行為走勢。

This study aims to discuss if there exist reasonable influence range passing from Japan and Korea stock market to Taiwan., we also observe if market really regulates its price when deviation of Taiwan index opening price affected by Japan and Korea stock opening price happens , in order to reflect essence of Taiwan stock and its fair price. Experiment result shows the application of artificial neural network, trained by inputting Japan and Korea opening price change rate, can effectively master the direction of opening price, the use of threshold promotes accuracy of direction forecasting. Under conviction of direction experiment, system attempt to find the reasonable opening range, we first define the measurement of system performance, and calculate the rate of returning to our system defined reasonable opening range, or at least moving toward it when TAIEX opening falls into our system defined forecasting range , in order to observe market’s adjusting force, then system search for reasonable opening range and ideal holding time which fit in with investing environment at the time, system also add stop-loss mechanism, or the use of Self-Organizing Map in order to optimize system’s performance, then system compares its performance with control group, at the final stage, we implement our system in different time yearly, to observe if system still work. Experiment result shows that overall searching does help finding the parameter setting suitable at that time, the revision force by market does exist, and the use of stop-loss strategy does raise system performance, but the performance after SOM selection does not have the same result, we infer that our input factor does not master similar behavior of market direction. System’s performance better than control group which comes from market regulating force, indirectly proofed that system master market’s regulating behavior in TAIEX’s early opening period. Finally, we examine our system in different years, according to system’s well understanding of Taiwan index opening price’s direction and rate of returning to our system defined reasonable opening range, we verify that methodology of artificial intelligence -- artificial neural network, could really perceive markets behavior pattern.