學術產出-Theses

Article View/Open

Publication Export

Google ScholarTM

政大圖書館

Citation Infomation

題名 應用強化學習與卷積神經網路於投資組合配置
Application of Reinforcement Learning and Convolutional Neural Networks to Portfolio Allocation
作者 林冠宇
Lin, Guan-Yu
貢獻者 廖四郎
Liao, Szu-Lang
林冠宇
Lin, Guan-Yu
關鍵詞 卷積神經網路
Black-Litterman 模型
風險趨避參數
強化學習
Convolution Neural Network
Black-Litterman Model
Risk Aversion
Reinforcement Learning
日期 2022
上傳時間 1-Aug-2022 17:27:39 (UTC+8)
摘要 本研究嘗試將強化學習方法應用於投資組合資產配置,且利用卷積神經網路(CNN)以金融資產的價量相關資料及技術指標作為輸入資料,進行資產價格漲跌方向及漲跌幅度的預測,並結合Black-Litterman模型建構風險分散的投資組合。將神經網路模型預測的結果作為Black-Litterman模型的投資人觀點,利用強化學習動態調整Black-Litterman模型中的風險趨避參數進行資產配置。實證發現,卷積神經網路在預測資產價格漲跌方向方面有過度配適的情況,使得測試期間準確度不高;而在預測資產價格漲跌幅度方面則有不錯的表現。在績效表現上面,無論是以iShares Russell 1000 ETF作為狀態資料來進行學習的投資組合一或是以S&P 500作為狀態資料來進行學習的投資組合二,皆大幅超越市值加權投資組合、等值加權投資組合,且投資組合一更是優於iShares Russell 1000 ETF且有更小的最大策略虧損,顯示能在控制風險的同時獲取更好的報酬。
In this thesis, we try to apply reinforcement learning to portfolio allocation. Historical price and volume related data and technical indicators are used as in put data to predict following week’s excess return. We also combine the forecasts with the Black-Litterman model and construct diversified portfolio. We use the forecasts to be investor views in Black-Litterman model and use reinforcement learning to adjust risk aversion.The empirical results show that CNN is overfitting in predicting the sign of asset price, which makes the accuracy of the test period not well; but it has a good performance in predicting the magnitude of excess return. We also find that both portfolio 1 that iShares Russell 1000 ETF is used as the state data and portfolio 2 that S&P 500 is used as the state data significantly outperform the benchmark portfolios. Portfolio 1 is even better than iShares Russell 1000 ETF and has a smaller MDD, showing better returns while controlling risk.
參考文獻 [1] Ballings, M., Van den Poel, D., Hespeels, N., & Gryp, R. (2015). Evaluating multiple classifiers for stock price direction prediction. Expert Systems with Applications, 42(20), 7046-7056.
[2] Black, F., & Litterman, R. (1991). Asset allocation: combining investor views with market equilibrium. The Journal of Fixed Income, 1(2), 7-18.
[3] Black, F., & Litterman, R. (1992). Global portfolio optimization. Financial Analysts Journal, 48(5), 28-43.
[4] Donthireddy, P. (2018, July 19). Black-Litterman portfolio with machine learning derived views. ResearchGate. Retrieved September 22, 2019, from https://www.researchgate.net/publication/326489143_Black-Litterman_Portfolios_with_Machine_Learning_derived_Views.
[5] Deng, Y., Bao, F., Kong, Y., Ren, Z., & Dai, Q. (2017). “Deep direct reinforcement learning for financial signal representation and trading.” IEEE transactions on neural networks and learning systems 28.3: 653-664.
[6] Hoseinzade, E., & Haratizadeh, S. (2018). CNNpred: CNN-based stock market prediction using a diverse set of variables. Expert Systems with Applications, 129, 273-285.
[7] He, G., & Litterman, R. (1999). The intuition behind Black-Litterman model portfolios. New York, NY: Goldman Sachs.
[8] Idzorek, T. (2004). A step-by-step guide to the Black-Litterman model, incorporating user specified confidence levels. Chicago, IL:Ibbotson Associates.
[9] LeCun, Y., Boser, B., Denker, J. S., Henderson, D., Howard, R. E., Hubbard, W., & Jackel, L. D. (1989). Backpropagation applied to handwritten zip code recognition. Neural Computation, 1, 541-551.
[10] Markowitz, H. (1952). Portfolio selection. The Journal of Finance, 7(1), 77-91.
[11] Moody, J., Wu, L., Liao, Y. & Saffell, M. (1998). Performance functions and reinforcement learning for trading systems and portfolios. Journal of Forecasting, 17(5-6), 441-470.
[12] Neuneier, R. (1998). Enhancing Q-learning for optimal asset allocation. In Advances in neural information processing systems (pp. 936-942).
[13] Schulman, J., Wolski, F., Dhariwal, P., Radford, A., & Klimov, O. (2017). Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347.
[14] Thawornwong, S., Enke, D., & Dagli, C. (2003). Neural networks as a decision maker for stock trading: a technical analysis approach. International Journal of Smart Engineering System Design, 5(4), 313-325.
描述 碩士
國立政治大學
金融學系
108352028
資料來源 http://thesis.lib.nccu.edu.tw/record/#G0108352028
資料類型 thesis
dc.contributor.advisor 廖四郎zh_TW
dc.contributor.advisor Liao, Szu-Langen_US
dc.contributor.author (Authors) 林冠宇zh_TW
dc.contributor.author (Authors) Lin, Guan-Yuen_US
dc.creator (作者) 林冠宇zh_TW
dc.creator (作者) Lin, Guan-Yuen_US
dc.date (日期) 2022en_US
dc.date.accessioned 1-Aug-2022 17:27:39 (UTC+8)-
dc.date.available 1-Aug-2022 17:27:39 (UTC+8)-
dc.date.issued (上傳時間) 1-Aug-2022 17:27:39 (UTC+8)-
dc.identifier (Other Identifiers) G0108352028en_US
dc.identifier.uri (URI) http://nccur.lib.nccu.edu.tw/handle/140.119/141055-
dc.description (描述) 碩士zh_TW
dc.description (描述) 國立政治大學zh_TW
dc.description (描述) 金融學系zh_TW
dc.description (描述) 108352028zh_TW
dc.description.abstract (摘要) 本研究嘗試將強化學習方法應用於投資組合資產配置,且利用卷積神經網路(CNN)以金融資產的價量相關資料及技術指標作為輸入資料,進行資產價格漲跌方向及漲跌幅度的預測,並結合Black-Litterman模型建構風險分散的投資組合。將神經網路模型預測的結果作為Black-Litterman模型的投資人觀點,利用強化學習動態調整Black-Litterman模型中的風險趨避參數進行資產配置。實證發現,卷積神經網路在預測資產價格漲跌方向方面有過度配適的情況,使得測試期間準確度不高;而在預測資產價格漲跌幅度方面則有不錯的表現。在績效表現上面,無論是以iShares Russell 1000 ETF作為狀態資料來進行學習的投資組合一或是以S&P 500作為狀態資料來進行學習的投資組合二,皆大幅超越市值加權投資組合、等值加權投資組合,且投資組合一更是優於iShares Russell 1000 ETF且有更小的最大策略虧損,顯示能在控制風險的同時獲取更好的報酬。zh_TW
dc.description.abstract (摘要) In this thesis, we try to apply reinforcement learning to portfolio allocation. Historical price and volume related data and technical indicators are used as in put data to predict following week’s excess return. We also combine the forecasts with the Black-Litterman model and construct diversified portfolio. We use the forecasts to be investor views in Black-Litterman model and use reinforcement learning to adjust risk aversion.The empirical results show that CNN is overfitting in predicting the sign of asset price, which makes the accuracy of the test period not well; but it has a good performance in predicting the magnitude of excess return. We also find that both portfolio 1 that iShares Russell 1000 ETF is used as the state data and portfolio 2 that S&P 500 is used as the state data significantly outperform the benchmark portfolios. Portfolio 1 is even better than iShares Russell 1000 ETF and has a smaller MDD, showing better returns while controlling risk.en_US
dc.description.tableofcontents 摘要 I
Abstract II
目次 III
表次 IV
圖次 V
第一章 緒論 1
第一節 研究背景與動機 1
第二節 研究目的 1
第二章 文獻回顧 3
第一節 機器學習方法相關文獻 3
第二節 投資組合理論相關文獻 4
第三節 強化學習相關文獻 4
第三章 研究方法 6
第一節 卷積神經網路 6
第二節 Black-Litterman 模型 16
第三節 強化學習 17
第四章 資料描述及實證分析 26
第一節 機器學習模型預測投資人觀點 26
第二節 強化學習訓練風險趨避參數 32
第三節 實證結果 34
第五章 結論與建議 40
第一節 結論 40
第二節 未來展望 40
參考文獻 41
zh_TW
dc.format.extent 2288731 bytes-
dc.format.mimetype application/pdf-
dc.source.uri (資料來源) http://thesis.lib.nccu.edu.tw/record/#G0108352028en_US
dc.subject (關鍵詞) 卷積神經網路zh_TW
dc.subject (關鍵詞) Black-Litterman 模型zh_TW
dc.subject (關鍵詞) 風險趨避參數zh_TW
dc.subject (關鍵詞) 強化學習zh_TW
dc.subject (關鍵詞) Convolution Neural Networken_US
dc.subject (關鍵詞) Black-Litterman Modelen_US
dc.subject (關鍵詞) Risk Aversionen_US
dc.subject (關鍵詞) Reinforcement Learningen_US
dc.title (題名) 應用強化學習與卷積神經網路於投資組合配置zh_TW
dc.title (題名) Application of Reinforcement Learning and Convolutional Neural Networks to Portfolio Allocationen_US
dc.type (資料類型) thesisen_US
dc.relation.reference (參考文獻) [1] Ballings, M., Van den Poel, D., Hespeels, N., & Gryp, R. (2015). Evaluating multiple classifiers for stock price direction prediction. Expert Systems with Applications, 42(20), 7046-7056.
[2] Black, F., & Litterman, R. (1991). Asset allocation: combining investor views with market equilibrium. The Journal of Fixed Income, 1(2), 7-18.
[3] Black, F., & Litterman, R. (1992). Global portfolio optimization. Financial Analysts Journal, 48(5), 28-43.
[4] Donthireddy, P. (2018, July 19). Black-Litterman portfolio with machine learning derived views. ResearchGate. Retrieved September 22, 2019, from https://www.researchgate.net/publication/326489143_Black-Litterman_Portfolios_with_Machine_Learning_derived_Views.
[5] Deng, Y., Bao, F., Kong, Y., Ren, Z., & Dai, Q. (2017). “Deep direct reinforcement learning for financial signal representation and trading.” IEEE transactions on neural networks and learning systems 28.3: 653-664.
[6] Hoseinzade, E., & Haratizadeh, S. (2018). CNNpred: CNN-based stock market prediction using a diverse set of variables. Expert Systems with Applications, 129, 273-285.
[7] He, G., & Litterman, R. (1999). The intuition behind Black-Litterman model portfolios. New York, NY: Goldman Sachs.
[8] Idzorek, T. (2004). A step-by-step guide to the Black-Litterman model, incorporating user specified confidence levels. Chicago, IL:Ibbotson Associates.
[9] LeCun, Y., Boser, B., Denker, J. S., Henderson, D., Howard, R. E., Hubbard, W., & Jackel, L. D. (1989). Backpropagation applied to handwritten zip code recognition. Neural Computation, 1, 541-551.
[10] Markowitz, H. (1952). Portfolio selection. The Journal of Finance, 7(1), 77-91.
[11] Moody, J., Wu, L., Liao, Y. & Saffell, M. (1998). Performance functions and reinforcement learning for trading systems and portfolios. Journal of Forecasting, 17(5-6), 441-470.
[12] Neuneier, R. (1998). Enhancing Q-learning for optimal asset allocation. In Advances in neural information processing systems (pp. 936-942).
[13] Schulman, J., Wolski, F., Dhariwal, P., Radford, A., & Klimov, O. (2017). Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347.
[14] Thawornwong, S., Enke, D., & Dagli, C. (2003). Neural networks as a decision maker for stock trading: a technical analysis approach. International Journal of Smart Engineering System Design, 5(4), 313-325.
zh_TW
dc.identifier.doi (DOI) 10.6814/NCCU202201024en_US