本研究旨在開發執行自動化投資管理、時間序列分析以及投資組合最佳化的理財機器人系統。本研究開發的核心演算法，利用人工智慧執行金融資產價格預測，並作為理財機器人的子系統。演算法交易以及投資組合最佳化的研究中已有許多實務上的應用，然而只有少部分文獻專注於理財機器人的應用開發。本研究利用深度學習中的長短期記憶 (Long Short-Term Memory) 方法，來解決時間序列分析中序列相依的複雜問題，並進一步開發自動化系統來執行時間序列預測於當沖交易策略以及投資組合最佳化的應用。本研究開發的可擴充系統可做為自動化投資管理中理財機器人系統的開發基礎。

This research aims to develop the system for Robo-Advisor to perform time-series forecasting and portfolio optimization in automated investment management. We designed the core algorithm, artificial intelligence to perform time-series forecasting in the financial market, as part of the system for Robo-Advisor in automated investment. There are many studies in algorithmic trading and portfolio management in various forms of applications; however, there is a paucity of literature focusing on the applications which are designed for Robo-Advisors. In this research, we used a deep learning approach, Long Short-Term Memory (LSTM), to develop our algorithm to solve the complexity of sequence dependence in time-series forecasting. We developed an automated system to perform time-series forecasting and used the result to construct a day trading strategy and to perform portfolio optimization to show that LSTM based algorithm added value. The system we built is expandable and can be used as a framework when developing Robo-Advisors for automated investment management.

Table of Contents
1.	Introduction	1
1.1.	Objective	1
1.2.	Background and Motivation	2
1.3.	Importance of the Study	2
1.4.	Robo-Advisors	3
1.5.	The Black-Litterman Model	4
1.6.	Deep Learning	5
1.7.	Long Short-Term Memories for Time Series Forecasting	6
2.	Research Methods and System Framework	7
2.1.	Research Design	7
2.1.1.	Construct a Conceptual Framework	7
2.1.2.	Develop a System Architecture	7
2.1.3.	Analyze & Design the System	7
2.1.4.	Build the System	7
2.1.5.	Observe & Evaluate the System	7
2.2.	System Architecture	8
2.2.1.	Data Collection Module	9
2.2.2.	Data Preprocessing Module	9
2.2.3.	LSTM Module	9
2.2.4.	Forecasting Module	9
2.2.5.	Data Visualization Module	10
2.2.6.	Portfolio Optimization Module	10
2.3.	Subjects and Data Collection	11
2.4.	Time-Series to Supervised Learning	11
2.5.	Training and Validation	13
2.6.	LSTM Model Configuration	13
2.7.	Forecasting and Labeling	16
2.8.	The Black-Litterman Formula	16
3.	Application and Results	18
3.1.	An LSTM Based Day Trading Strategy	18
3.1.1.	Result and Analysis	19
3.1.2.	Comparison with Longer Time Horizon	22
3.2.	Portfolio Optimization Using the LSTM Based Investor Views	27
3.2.1.	The LSTM Investor Views	27
3.2.2.	The Black-Litterman Model	31
3.2.3.	Portfolio Performance in 2016	31
4.	Conclusion and Recommendations	38
References	40
Appendix A.2 Comparison of Cumulative Return	41

List of Figures
Figure 1.  How LSTM Works	5
Figure 2. System Development Research Method Flow Chart	8
Figure 3. System Architecture	10
Figure 4. Model Training Losses	14
Figure 5. Distribution of the Cumulative Returns	20
Figure 6. 275_OIL Cumulative Return Under LSTM Investing Strategy in 2015	23
Figure 7. 275_OIL Cumulative Return Under LSTM Investing Strategy in 2016	23
Figure 8. Price Movement of 275_OIL in 2015	24
Figure 9. Price Movement of 275_OIL in 2016	24
Figure 10. Fluctuation in 2015	25
Figure 11. Fluctuation in 2016	25
Figure 12. Confusion Matrix for 275_OIL in 2015	26
Figure 13. Confusion Matrix for 275_OIL in 2016	26
Figure 14. Convert Forecasted Daily Returns into Quarterly Return	28
Figure 15. Pick Matrix and View Matrix for each Quarter	30
Figure 16. Portfolio Cumulative Returns	37
Figure 17. Assets Cumulative Returns	37
 
List of Tables
Table 1. Selected ETFs Categories	12
Table 2. Model Configurations	15
Table 3. Model Summary	15
Table 4. Summary of Cumulative Return Comparison	19
Table 5. Performance of the LSTM Based Strategy in Each Class	21
Table 6. Six Selected Assets Portfolio	27
Table 7. Date of Portfolio Weight Rebalancing for Each Quarter	27
Table 8. Forecasted Quarterly Cumulative Returns and Prices	29
Table 9. Predicted Quarterly Returns	30
Table 10. Quarterly Performance of each Portfolio	31
Table 11. Historical Covariance Matrix for each Quarter	32
Table 12. Posterior Covariance Matrix for each Quarter	33
Table 13. Return Vectors and Resulting Portfolio Weights for each Quarter	34
Table 14. Annual Portfolio Statistics	36

Bicksler, J. L. (1984). Modern Portfolio Theory and the Capital Asset Pricing Model: A User's Guide (Book Review) (Vol. 39, pp. 568-569).
Black, F., & Litterman, R. (1992). Global portfolio optimization. Financial analysts journal, 48(5), 28-43. 
Chishti, S., & Barberis, J. (2016). The FINTECH Book The Financial Technology Handbook for Investors, Entrepreneurs and Visionaries. New York: New York : Wiley.
Grauer, R. R., & Janmaat, J. A. (2010). Cross-sectional tests of the CAPM and Fama– French three-factor model. Journal of Banking and Finance, 34(2), 457-470. doi:10.1016/j.jbankfin.2009.08.011
Gruber, M. J., & Ross, S. A. (1978). THE CURRENT STATUS OF THE CAPITAL ASSET PRICING MODEL (CAPM. Journal of Finance, 33(3), 885-901. doi:10.1111/j.1540-6261.1978.tb02029.x
Idzorek, T. M. (2002). A step-by-step guide to the Black-Litterman model. Forecasting expected returns in the financial markets, 17. 
Längkvist, M., Karlsson, L., & Loutfi, A. (2014). A review of unsupervised feature learning and deep learning for time-series modeling. Pattern Recognition Letters, 42, 11-24. 
LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521(7553), 436-444. doi:10.1038/nature14539
Markowitz, H. (1952). Portfolio selection. The Journal of Finance, 7(1), 77-91. 
Nunamaker, J. F., Chen, M., & Purdin, T. D. M. (1990). Systems Development in Information Systems Research. Journal of Management Information Systems, 7(3), 89-106. doi:10.1080/07421222.1990.11517898
Rättyä, J. (2016). March of the Robo-advisors: The potential for global expansion of digital asset management platforms. 
Sironi, P. (2016). FinTech Innovation From Robo-Advisors to Goal Based Investing and Gamification. Somerset: Somerset : Wiley.
Walters, C. (2014). The Black-Litterman model in detail. The Black-Litterman Model in Detail (June 20, 2014). 
Zhou, G. (2009). Beyond Black-Litterman: letting the data speak. Journal of Portfolio Management, 36(1), 36.