[Back to Results | New Search]

URN	etd-0621118-142449
Author	Jia-Hong Liu
Author's Email Address	No Public.
Statistics	This thesis had been viewed 5614 times. Download 73 times.
Department	Computer Science and Engineering
Year	2017
Semester	2
Degree	Master
Type of Document
Language	English
Title	Portfolio Investment Based on Neural Networks
Date of Defense	2018-07-30
Page Count	56
Keyword	neural network stock investment gene expression programming convolutional neural network portfolio
Abstract	In this thesis, we combine the trading signals generated by the gen expression programming (GEP) method of Lee et al. and the portfolio generated by convolutional neural network (CNN) structure of Jiang et al. to form a stock investment method with portfolio management. The method of Jiang et al. focuses on the investment of the cryptocurrency. We change the invested target of Jiang et al. from cryptocurrency to stocks. We recompute the weights of the portfolio when the method of Lee et al. generates a trading signal (buy or sell). To test our method, we choose 213 stocks which always exist during 1995/1/5 to 2017/12/29 on stock market in Taiwan. Our training period starts from 1995/1/5. We perform the trading from 2002/1/2 until 2017/12/29. There are three cases in our experiments: Trading 100 stocks with the 100-stock features, trading 100 stocks with the 213-stock features, and trading 213 stocks with the 213-stock features. The annualized returns for the three cases are 25.00%, 26.52% and 27.32%, respectively. Our method is better than the buy-and-hold 12.36% for 100 stocks, and 12.21% for 213 stocks. Our method is also better than the method of Lee et al. without portfolio management 12.94% for 100 stocks, and 12.67% for 213 stocks.
Advisory Committee	Shih-Chung Chen - chair Chiou-Yi Hor - co-chair Chien-Feng Huang - co-chair Chang-Biau Yang - advisor
Files	etd-0621118-142449.pdf Indicate in-campus at 0 year and off-campus access at 1 year.
Date of Submission	2018-07-21

[Back to Results | New Search]