The Study of Taiwan Weighted Index Forecasting Model Construction Base on Machine Learning

• Main Authors: WANG, JING-YAO, 王靖瑤
• Other Authors: SHEU, JIA-SHING
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/762k77

碩士 === 國立臺北教育大學 === 資訊科學系碩士班 === 107 === Investing in the stock market is a high-risk activity. As the country's economic development reflects on index stocks and the excess return of the investment will be very attractive. Therefore, as the average age of the investors continues to drop, discussions on the stock market index has become an extremely hot topic. Being able to predict the stock market means one will expect to earn excess return in the short term. This paper attempts to predict the stock price of Taiwan Weighted Stock Index after a certain period of time using Back Propagation Neural network (BPN). This will allow us to increase the reward and control the level of risk in advance. We also compared the forecasting result of BPN and Random Forest. In this paper, the BPN model significantly outperformed the Random Forest model. The accuracy achieves 60%, and the Mean Square Error achieves 0.15. Simulated trading on the Exchange Traded Funds of Taiwan Weighted Stock Index using the model proposed by this paper obtained 2.91% of return on investment in the simulation interval. Hence the prediction model proposed by this paper can help conduct the transaction smarter and generate more stable returns on investment.