Establishing the Price Prediction Model based on the TAIEX Futures Technical Indicators with LSTM Neural Network

• Main Authors: Lin, Chun-You, 林俊佑
• Other Authors: Han, Chien-Shan
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/e4h7y7

碩士 === 輔仁大學 === 金融與國際企業學系金融碩士班 === 107 === This study develops 3 minute K-bar of the stock index prediction model by using Long Short Term Memory (LSTM) with input data including the opening price, the highest price, the lowest price, and the closing price in combination with the technical indicators of different days. The data period is from January 2, 2018 to June 5, 2018. The one-month data is used as the training period, and the next week's forecast is the test period, totally 5 periods. Research the price prediction capabilities of LSTM. In this study, KD, MACD, MA, W%R, VR were selected, and the technical indicators of the combined price and quantity were divided into long-day and short-day models, and based on MAE, trend accuracy and K-bar Hit rate to view forecast performance. The empirical results show that the trend accuracy of both long-day and short-day models is less than 50%, and it is not good to predict the 3 minute index. The performance of the K-bar hit rate is better, and some parameters can be set to hit more than 70% in the next 15 minutes. Regardless of MAE or K-bar hit rate, the short-day model has better predictive performance than long-day model.