應用倒傳遞類神經網路探討國際股市於 台指期貨開盤之行為發現

Abstract

台灣股市是一個淺碟型市場，股市漲跌很容易受國際股市及外來資金的影響，因此，要分析台灣股市的漲跌並提出策略，一定得將國際股市納入重要因子加以分析。開盤價往往透露出主力對市場多空的看法，是研究股市或期貨等金融衍生性商品當日趨勢非常重要的資訊。

本論文應用倒傳遞類神經網路，整合亞洲台灣加權指數、台灣加權指數期貨、上海A股、香港恆生指數、日本日經指數、韓國綜合指數、美國道瓊指數、美國S&P500指數、美國那斯達克指數、歐洲德國DAX指數、法國CAC指數、英國FTSE100指數的物理量變化預測台灣加權指數期貨的開盤漲跌幅，並驗證日股、韓股開盤的已經反映國際金融的事件總和能有效提升台指期貨開盤價預測的準確性。

本研究証實，利用倒傳遞類神經網路學習亞洲、美洲、及歐洲各主要股市收盤價之波動，可以有效的預測日內台指期貨的走勢，代表台灣股市與全球股市連動密切。並且日股及韓股開盤早於台指期貨45分鐘，該股市已率先反應前一夜的國際情勢，而其動能也會傳遞到台股，可當作台股開盤研究的重要參考資訊，以有效提升台股開的方向預測及減少預測之誤差。

Taiwan’s stock is a shallow-plate market, it is susceptible to international stock market and foreign funds. Therefore, to forecast the ups or downs of Taiwan stock market and propose strategies, we must study the international stock market. Opening price of the market often reveals the trend of daily stock market. It is an important information for researching the stock index and futures.

This study uses the opening price of Taiwan Stock Index (TAIEX), Taiwan Stock Index Future (TAIFEX), Shanghai A-share Index (SHASHR), Hang Seng Index (HSI), Dow Jones Industrial Average Index (DJIA), S&P500 (SPX), Nasdaq Composite Index (NASDAQ),The Deutscher Aktien index (DAX), Cotation Assistée en Continu 40 (CAC40), and Financial Times Stock Exchange 100 Index (FTSE100), and opening price of Nikkei 225 Stock Index (Nikkei 225) and Korea Composite Index (Kospi) as input of Back Propagation Neural Network (BPNN) to forecast the opening price of TAIFEX to verify if the opening of Nikkei 225 and Kospi can improve the forecast accuracy of TAIFEX's opening price.

The result reveals the welled trained BPNN model by the closing and opening price of above stock index from Asia, America, and Europe are effective information to forecast TAIFEX’s opening direction and price. It means Taiwan stock market is highly co-movement with internationals. Besides, the opening of Nikkei 225 and Kospi which had been reflected the change of international economics can effectively improve the forecast accuracy of opening direction and price of TAIFEX.