摘要(英) |
FinTech is a popular application in artificial intelligence applications. Although many studies have been conducted to predict the movement of stock price based on historical data and technical indicators, how to com- bine the prediction with portfolio allocation remains a problem. In this re- search, we consider financial news and technical indexes in the stock price movement prediction to achieve better performance. To support adaptive portfolio optimization, this study further proposes a new task to predict the ranking of stocks return. With a simple strategy of investing top ranked stocks, the prediction model achieves a high return on investment (ROI) than Taiwan 50 ETF and Taiwan mid-cap 100 ETF. The thesis is divided into three parts: binary classification of stock movement on the next trad- ing day, the ranking prediction of stock returns in the next week/month and an application of portfolios trading strategy based on backtesting.
For stock trend prediction, we use stock technical indicators as the model’s input features, and compare the performance with or without news headlines or news sentiment on the next trading day stock movement pre- diction. Experiments show that adding news sentiment has better per- formance than adding news headline about 4%. Comparing it with the baseline (Random Forest), performance has been improved around 5%. Secondly, we formulate a new problem to predict stock rank for the next week/month based on return of investment to reduce frequent trading. Base on technical indicators, we compare the performance of stock rank prediction with or without news headlines and news sentiment. While the model with news headlines achieves lower loss, the model with news sen- timent has better balance in stock rank prediction. Finally, we use the next week/month stock rank prediction to select the top ranked k stocks for adaptive portfolio allocation. Experiments show that the portfolio al- location strategy achieve higher investment return than Taiwan 50ETF, especially when using the model with news sentiment, the average ROI performance can be improved around 23%. |
參考文獻 |
[1] https://pscwm.wordpress.com/2017/10/23/中、美砸錢全球搶 ai 頂
尖人才;ai 操盤 etf 績效跑贏標普/.
[2] Y. Xu and S. B. Cohen(2018, July). Stock Movement Prediction from Tweets and Historical Prices. Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics, Melbourne, Aus- tralia, pages 1970–1979..
[3] L. Zhang et al. 2017. Stock Price Prediction via Discovering Multi- Frequency Trading Patterns. KDD Applied Data Science Paper. Hal- ifax, NS, Canada, pages 2141–2149.
[4] http://www.asimovinstitute.org/neural-network-zoo/
[5] https://www.finlab.tw/
[6] https://hahow.in/courses/5b9d3a6dca498a001e917383/main
[7] https://github.com/mrjbq7/ta-lib
[8] J. Devlin, M. W. Chang, K. Lee, K. Toutanova(2019, June), BERT: Pre-training of Deep Bidirectional Transformers for Language Under- standing, Association for Computational Linguistics, pp. 4171-4186, Minneapolis, USA
[9] https:// www.coderbridge.com/ @weifanhaha/ e2e46480d185435680609a40997644be
[10] https://github.com/google-research/bert
[11] T. HELLSTROM, 2000. Predicting a Rank Measure for Stock Re- turns. Theory of Stochastic Processes, Vol.6 (22), no.3-4, pp. 64-83
[12] S. Obeidat et al. 2018. Adaptive Portfolio Asset Allocation Optimiza- tion with Deep Learning. International Journal on Advances in Intel- ligent Systems. page 25-34.
[13] M. Y. Day and J. T. Lin. 2019. Artificial Intelligence for ETF Mar- ket Prediction and Portfolio Optimization. IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining. Van- couver, BC, Canada, pages 1026–1033.
[14] K. Khare et al. 2017. Short Term Stock Price Prediction Using Deep Learning. IEEE International Conference On Recent Trends in Elec- tronics Information Communication Technology (RTEICT). Banga- lore, India, pages 482–486.
[15] S. McNally et al. 2018. Predicting the Price of Bitcoin Using Ma- chine Learning. 26th Euromicro International Conference on Parallel, Distributed, and Network-Based Processing. Cambridge, UK, pages 339–343.
[16] M. Khairalla et al. 2017. Hybrid Forecasting Scheme for Financial Time-Series Data using Neural Network and Statistical Methods. In- ternational Journal of Advanced Computer Science and Applications. pages 319–327.
[17] J. Patel et al. 2015, Predicting stock and stock price index movement using Trend Deterministic Data Preparation and machine learning techniques, Expert Systems with Applications. pages 259-268
[18] V. S. Pagolu et al. 2017, Sentiment Analysis of Twitter Data for Predicting Stock Market Movements, International Conference on Signal Processing, Communication, Power and Embedded System (SCOPES), Paralakhemundi, India. page 1-6
[19] G. Jacobs et al. 2018, July, Economic Event Detection in Company- Specific News Text, Proceedings of the First Workshop on Economics and Natural Language Processing, Melbourne, Australia. pages 1–10
[20] W. Y. Ma and K. J. Chen. 2003. Introduction to CKIP Chinese Word Segmentation System for the First International Chinese Word Seg- mentation Bakeoff. Proceedings of ACL, Second SIGHAN Workshop on Chinese Language Processing, page 168-171.
[21] C. L. Chou, C. H. Chang, Y. Y. Huang. 2016. Boosted Web Named Entities Recognition via Tri-Training. Transactions on Asian and Low- Resource Language Information Processing, New York, USA, Volume 16 Issue 2.
[22] V. D. Ta, C. M. Liu, D. Addis. 2018. Prediction and Portfolio Op- timization in Quantitative Trading Using Machine Learning Tech- niques. Proceedings of the Ninth International Symposium on Infor- mation and Communication Technology(SoICT), Da Nang, Viet Nam, Pages 98–105.
[23] https://github.com/fxsjy/jieba
[24] http:// www.yuanta.com.tw/ file-repository/ content/ 0919news/ no- tice107_0919.htm
[25] https://ec.ltn.com.tw/article/breakingnews/3073254
[26] https://medium.com/bincentive/量化交易夯什麼-7ee1c25a18ce |