近年來，因為 GPU 的發展，網路上龐大公開資料的可取得，資料分析成了
一個熱門的問題，深度學習也因此走入大眾眼中，因為深度學習可以從龐大的、
無序的資料中發現一些我們人類不容易找到的關聯性。隨著深度學習理論的完善，
他也被應用在各個領域中。其中一個蓬勃發展的應用就是金融領域。金融價格預
測與分析是一個持續被研究了數十年，包括技術分析、基因演算法、時間序列預
測、機器學習、深度學習等等。因為金融市場的不確定性，所以到目前為止沒有
一個保證可以獲利的模型或是方法被提出。
本論文建立了一個簡單容易使用的系統，包含了兩個深度學習模型以及一個
交易模擬程式。深度學習模型用來預測價格的趨勢，他們分別是長短期記憶神
經網路 ( Long Short-Term Memory Neural Networks) 和遞歸神經網路 (Recurrent
Neural Networks) 。期貨市場的交易模擬程式用以驗證我們的模型產生的預測資
料在不同交易策略下的獲利情況。

In recent years, because of the development of GPU and the availability of large
and open data on the Internet, data analysis has become a hot issue. Therefore,
deep learning comes into public attention, because deep learning can find some
correlations that we humans can’t easily find from the vast and disorderly data.
With the improvement of deep learning theories, they have also been applied in
various fields. One of the thriving applications is in the financial area. Financial
price forecasting and analysis is a continuous study for decades, including technical
analysis, gene algorithms, time series prediction, machine learning, deep learning,
and so on. Because of the uncertainty of the financial market, so far, no model or
method has been proposed to guarantee profitability.
In this paper, we build a simple and easy-to-use system, which includes two
deep learning models and a trading simulation program. The deep learning is for
predicting price trends. They are Long Short-Term Memory Neural Networks and
Recurrent Neural Networks. The future market trading simulation program is used
for evaluating the forecast data generated by our models under different trading
strategies.

1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Contribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Background 4
2.1 Neural Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Recurrence Neural Network . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Long Short-term Memory Network . . . . . . . . . . . . . . . . . . . 7
3 Related works 10
3.1 Price-based Classification Models . . . . . . . . . . . . . . . . . . . . 10
3.2 Text-based Prediction Models . . . . . . . . . . . . . . . . . . . . . . 11
4 Implementation 13
4.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.1.1 Trend Prediction . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.1.2 Trading Simulation . . . . . . . . . . . . . . . . . . . . . . . . 14
4.2 Data Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.3 Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.3.1 RNN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.3.2 LSTM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.4 Back-trading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.5 Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.5.1 Baseline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.5.2 Martingale Strategy . . . . . . . . . . . . . . . . . . . . . . . 20
4.5.3 Anti-Martingale Strategy . . . . . . . . . . . . . . . . . . . . 21
4.5.4 Great Martingale Strategy . . . . . . . . . . . . . . . . . . . . 21
4.6 Profit Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.7 Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
5 Experiment 24
5.1 Accuracy and Loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
5.2 Profit Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5.2.1 Stop loss = 3.5% . . . . . . . . . . . . . . . . . . . . . . . . . 26
5.2.2 Stop Loss = 4% . . . . . . . . . . . . . . . . . . . . . . . . . . 31
6 Conclusion 36
6.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
6.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Bibliography 38

[1] Johan Bollen, Huina Mao, and Xiao-Jun Zeng. Twitter mood predicts the
stock market, 10 2010.
[2] Xiao Ding, Yue Zhang, Ting Liu, and Junwen Duan. Deep learning for
event-driven stock prediction. pages 2327–2333, 2015.
[3] Alex Graves. Generating sequences with recurrent neural networks.
https://arxiv.org/pdf/1308.0850.pdf.
[4] Victor Lavrenko, Matt Schmill, Dawn Lawrie, Paul Ogilvie, David Jensen, and
James Allan. Mining of concurrent text and time series. pages 37–44, 2000.
[5] Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg Corrado, and Jeffrey Dean.
Distributed representations of words and phrases and their compositionality.
pages 3111–3119, 2013.
[6] Xiongwen Pang, Yanqiang Zhou, Pan Wang, Weiwei Lin, and Victor Chang.
Stock market prediction based on deep long short term memory neural
network.
[7] Yangtuo Peng and Hui Jiang. Leverage financial news to predict stock price
movements using word embeddings and deep neural networks, 2016.
[8] Luca Di Persio and Oleksandr Honchar. Artificial neural networks
architectures for stock price prediction: comparisons and applications.
International Journal of Circuits, Systems and Signal Processing, 2016.
[9] Lei Shi, Zhiyang Teng, Le Wang, Yue Zhang, and Alexander Binder. Deepclue
- visual interpretation of text-based deep stock prediction. IEEE Transactions
on Knowledge and Data Engineering, 2019.
[10] Lawrence Takeuchi and Yu-Ying (Albert) Lee. Applying deep learning to
enhance momentum trading strategies in stocks. Technical report, Stanford
university, 2013.
[11] Avraam Tsantekidis, Nikolaos Passalis, Anastasios Tefas, Juho Kanniainen,
Moncef Gabbouj, and Alexandros Iosifidis. Forecasting stock prices from the
limit order book using convolutional neural networks.