Predicting Taiwan Stock Index Futures Closing Price By Three Institutional Investors With LSTM Artificial Neural Networks

• Main Authors: Li,Jhe-Ruei, 李哲睿
• Other Authors: Han, Chien-Shan
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/y3gqny

碩士 === 輔仁大學 === 金融與國際企業學系金融碩士班 === 107 === The data period is from July 3, 2007 to April 26, 2019. This study uses information provided by the Taiwan Stock Exchange and the Taiwan Futures Exchange to provide information of three institutional investors, including changes in futures open positions and changes in options open positions, and net purchases or net sales, and then predict closings price with LSTM Model. From the network structure of foreign investors and investment trust and the minimum RMSE network structure of the three institutional investors and the best trend hit rate, and the network structure with the best target accuracy, it seems that it is impossible to evaluate which network in one way. The road structure is the best network structure belonging to the chip. This study finds that, based on a one-month contract analysis, the foreign investors open interest and the net buy of the foreign investors on the settlement date are better than the non-settled date, and are better than the dealer and the investment trust. In addition, for the average five-day average accuracy and five days before the settlement date, it is found that the open interest of the dealer and the excess amount of the purchase and sale are used to predict the price and are more accurate than foreign investors and investment trust.