隨著科技進步，金融科技(Fintech) 成為眾多研究的主要議題，因此藉由科技協助人們做投資組合也是重點研究之一。在金融市場上，高收益的產品都伴隨著高風險，因此如何有效選取不同投資標的，卻能獲得一定收益水準是主要研究主題。在本研究我們使用台灣股市作實例研究，為了從複雜多變的金融市場數據中，萃取出更深層的特徵，我們建議使用基於遞迴神經網路(Recurrent Neural Networks，RNN) 中的長短期記憶(Long Short-Term Memory，LSTM) 作為自編碼器的基本架構，並加入預測解碼器，擷取新特徵供排序學習中的RankSVM 使用，從30 檔股票裡，挑出前10 名下週預測漲跌幅最高的股票，作為每週的投資組合，之後我們回測了在2019 年和2020年之收益成效，從收益結果來說，當整體環境趨勢明顯的情形下，加入預測解碼器的組合式模型，更能提升收益回報率，但在大盤處於盤整時，僅將數據經由LSTM 自編碼器較能保持收益率。由於2020 年全球經歷COVID-19 疫情後，全球股市重創，因此從數值結果發現，在輸入更多類似的事件至RankSVM 後，對於重大事件的應變能力會有所提升，最後我們比較了均值-變異數模型動態尋找每週最佳夏普值之投資組合和台灣50 這兩年的收益成效，結果發現我們的每週的投資組合普遍都比均值-變異數模型和台灣50 更能創造收益。

With the advancement of technology, financial technology (Fintech) has become the main topic of many researches, so using technology to help people make portfolios is also one of the key researches. In the financial market, high-return products are accompanied by high risks. Therefore, how to effectively allocate the weight of different investment targets while maintaining a certain level of return is also a major research topic. In this research,
we propose a method that uses an LSTM-based autoencoder and an autoencoder that adds a predictive decoder to try to extract deeper features from complex and changeable
financial market data, and Borrow new features. RankSVM, which learning to rank, tried to select the top 10 stocks with the biggest gains and declines next week from the 30
stocks in the Taiwan stock market, and used these 10 stocks as a weekly portfolio. After that, we conducted a backtesting on the cumulative return for 2019 and 2020. From the perspective of profitability, when the overall environment trend is obvious, adding a combination model of predictive decoders can increase the return. However, when the market is in a consolidation state, only by passing data through the LSTM autoencoder can the rate of return be maintained. As the global stock market is hit hard by the COVID-19 in 2020, we found that after more similar incidents enter RankSVM, the ability to respond to major incidents will be improved. Finally, we compared the mean-variance model. and Taiwan Top50 Tracker Fund (TTT). It turns out that our weekly portfolio is generally more profitable than the mean-variance model and the TTT.

[1] H. M. Markowitz. Portfolio selection in the journal of finance vol. 7. J Finance, 1952.
[2] M. C. Jensen, F. Black, and M. S. Scholes. The capital asset pricing model: Some empirical tests. 1972.
[3] E. F. Fama and K. R. French. Common Risk Factors in the Returns on Stocks and Bonds. University of Chicago Press, 2021.
[4] E. F. Fama and K. R French. The Cross-Section of Expected Stock Returns. University of Chicago Press, 2021.
[5] D. Duffie and J. Pan. An overview of value at risk. Journal of derivatives, 4:7–49, 1997.
[6] S. Rockafellar, R. T.and Uryasev et al. Optimization of conditional value-at-risk. Journal of risk, 2:21–42, 2000.
[7] Y. Ma, R. Han, and W. Wang. Prediction-based portfolio optimization models using deep neural networks. IEEE Access, 8:115393–115405, 2020.
[8] S. Obeidat, D. Shapiro, M. Lemay, M. K. MacPherson, and M. Bolic. Adaptive portfolio asset allocation optimization with deep learning. International Journal on
Advances in Intelligent Systems, 11:25–34, 2018.
[9] C.-M. Lin, J.-J. Huang, M. Gen, and G.-H. Tzeng. Recurrent neural network for dynamic portfolio selection. Applied Mathematics and Computation, 175:1139–1146,
2006.
[10] X. Y. Fu, J. H. Du, Y. F. Guo, M. W. Liu, T. Dong, and X. W. Duan. A machine learning framework for stock selection. arXiv preprint arXiv:1806.01743, 2018.
[11] V. Kedia, Z. Khalid, S. Goswami, N. Sharma, and K. Suryawanshi. Portfolio generation for indian stock markets using unsupervised machine learning. In 2018 Fourth
ICCUBEA, pages 1–5. IEEE, 2018.
[12] O. Jin and H. El-Saawy. Portfolio management using reinforcement learning.
Preprint, 2016.
[13] Z. Jiang, D. Xu, and Jinjun Liang. A deep reinforcement learning framework for the financial portfolio management problem. arXiv preprint arXiv:1706.10059, 2017.
[14] Z. Jiang and Jinjun Liang. Cryptocurrency portfolio management with deep reinforcement learning. In 2017 Intelligent Systems Conference (IntelliSys), pages
905–913. IEEE, 2017.
[15] J Patel, S Shah, P Thakkar, and K Kotecha. Predicting stock and stock price index movement using trend deterministic data preparation and machine learning
techniques. Expert systems with applications, 42:259–268, 2015.
[16] V-D Ta, C-M Liu, and D Addis. Prediction and portfolio optimization in quantitative trading using machine learning techniques. In Proceedings of the Ninth International Symposium on Information and Communication Technology, pages 98–105, 2018.
[17] Z. Zhang, S. Zohren, and S. Roberts. Deep learning for portfolio optimization. The Journal of Financial Data Science, 2:8–20, 2020.
[18] C. Krauss, X. A. Do, and N. Huck. Deep neural networks, gradient-boosted trees, random forests: Statistical arbitrage on the s&p 500. Eur J Oper Res, 259:689–702,
2017.
[19] H. Li. A short introduction to learning to rank. IEICE TRANSACTIONS on Information and Systems, 94:1854–1862, 2011.
[20] C.-P. Lee and C.-J. Lin. Large-scale linear RankSVM. Neural Computation, 26:781–817, 2014.
[21] T. Joachims. Optimizing search engines using clickthrough data. In Proceedings of the eighth ACM SIGKDD international conference on Knowledge discovery and data
mining, pages 133–142, 2002.
[22] C. Burges, T. Shaked, E. Renshaw, A. Lazier, M. Deeds, N. Hamilton, and G. Hullender. Learning to rank using gradient descent. In Proceedings of the 22nd international
conference on Machine learning, pages 89–96, 2005.
[23] S. Hochreiter and J. Schmidhuber. Long short-term memory. Neural Computation,
9:1735–1780, 1997.
[24] P. Baldi. Autoencoders, unsupervised learning, and deep architectures. In Proceedings of ICML workshop on unsupervised and transfer learning, pages 37–49, 2012.
[25] N. Srivastava, E. Mansimov, and R. Salakhudinov. Unsupervised learning of video representations using LSTMs. In ICML, pages 843–852, 2015.
52