應用自組織映射圖神經網路及倒傳遞類神經網路於機構投資人融資券行為分析研究-以台灣50 期貨指數為例

Abstract

近年來機構投資人占股市成交金額的比重逐年上升，至民國96年機構投資人交易已佔總交易三分之一；從許多文獻得知，一般非機構投資人常參考機構投資人的交易動作，來作為投資的參考，因此本研究針對機構投資人交易行為與股票價格間的關係，對股票價格之價格進行預測動作，嘗試從籌碼面的角度切進，探討是否可由籌碼面因子來判斷未來股價之趨勢。 本研究應用人工智慧領域中的自組織映射圖神經網路對台指50指數期貨籌碼面歷史日資料做分群，並透過逐步迴歸，過濾掉各群中和台灣50指數期貨收盤價相關性較低的籌碼面因子，最後將各群中的資料輸入至倒傳遞神經網路，預測3日、5日、7日後相對於今日台灣50指數期貨收盤價的漲幅程度。實驗結果證實，實驗模型自組織映射圖神經網路結合倒傳遞神經網路，其預測準確率顯著優於對照組只有使用倒傳遞神經網路的模型，藉由分群技術輔助更能掌握環境的變化去作動態學習，進而提供投資人明確的買賣交易資訊，輔助決策者做正確抉擇的依據。

In recent years, institutional investors accounted for the proportion of stock market turnover increased year by year, the institutional investor trades up to 2007 has accounted for one-third of the total, from many literature reviews, the general non-institutional investors often refer to institutional investors trading action , as a reference for investment, therefore this study is for institutional investors’ trading transactions behavior and stock price relations , to forecast the price of the stock price movements, and trying to from the traded volume angle to discusses whether can judge in the future tendency of the stock price by the traded volume factor.

This study applied the self-organising map neural network in the field of artificial intelligence to cluster the historical data of the Taiwan 50 index future traded volume, and filter out each group’s of Taiwan 50 index future closed at lower correlation traded volume factor through the stepwise regression, finally, input each group’s data to the back-propagation neural network to forecast the degree of increase of 50 Taiwan index future closing price in 3, 5, and 7 days compared with the today's. The experimental results pointed out that the accuracy rate with the experimental model of self-organizing map neural network combined with Back-propagation neural network was much better than the control group which only adopted back-propagation neural network models. The thesis attempts to apply clustering technique,which to grasp the environmental changes, to make a dynamic learning and to provide investors more explicit information of sale transaction to support decision-makers to do the correct choice.

This study applied the self-organising map neural network in the field of artificial intelligence to cluster the historical data of the Taiwan 50 index traded volume, and filter out each group’s of Taiwan 50 index closed at lower correlation traded volume factor through the stepwise regression, finally, input each group’s data to the back-propagation neural network to forecast the degree of increase of 50 Taiwan index fuclosing price in 3, 5, and 7 days compared with the today's. The experimental results pointed out that the accuracy rate with the experimental model of self-organizing map neural network combined with Back-propagation neural network was much better than the control group which only adopted back-propagation neural network models. The thesis attempts to apply clustering technique,which to grasp the environmental changes, to make a dynamic learning and to provide investors more explicit information of sale transaction to support decision-makers to do the correct choice.