利用 TCN 及 Residual LSTM 建立股票投資深度學習回測模型

以作者查詢圖書館館藏

、以作者查詢臺灣博碩士

、以作者查詢全國書目

、勘誤回報

、線上人數：38

、訪客IP：18.225.156.91

姓名

張博鈞(ZHANG, BOJUN) 查詢紙本館藏

畢業系所

資訊管理學系在職專班

論文名稱

利用 TCN 及 Residual LSTM 建立股票投資深度學習回測模型

相關論文

★ 針對提昇資料倉儲資料庫執行效能之知識管理與相關系統設計	★ 以關聯規則探勘為基礎，探討詐騙車手提領型態互動之研究
★ 部落格之網路口碑評比機制平台管理與應用	★ 虛擬貨幣交易平台之實現
★ 適用於多種設備的可否認鑑別協定之設計	★ 交易程式最佳化的多維度分析平台之設計與建置
★ 多商品多策略程式交易績效曲線比較和分群機制之研究	★ 整合分位數回歸分析與單因子選股的股票選股策略回測與績效評估之研究
★ 以工作流程與Portlet為基礎整合學習管理系統以支援課程編組	★ 使用服務導向技術建構具支援二線廠客製化能力的電子中樞系統之研究
★ 以流程為中心的Portlet重用性分析	★ 應用資料倉儲技術建構平衡計分卡資訊系統之研究-以某消費性電子製造公司人力資源計分卡為例
★ 自動化的產品平台管理與應用	★ 以代理人為基礎的資訊系統協助新產品開發流程的自動化
★ 以整合式的教練引導開發以框架為基礎的專案	★ 支援新產品研發的整合性知識管理系統

檔案

[Endnote RIS 格式]

[Bibtex 格式]

[相關文章]

[文章引用]

[完整記錄]

[館藏目錄]

[檢視]

[下載]

本電子論文使用權限為同意立即開放。
已達開放權限電子全文僅授權使用者為學術研究之目的，進行個人非營利性質之檢索、閱讀、列印。
請遵守中華民國著作權法之相關規定，切勿任意重製、散佈、改作、轉貼、播送，以免觸法。

摘要(中)

過去深度學習應用於股票之研究已有很多，但大多應用僅止於預測股價，該類研究衡量方式大多為統計量，如均方差(MSE;Mean square error)、Accuracy 衡量、衡量股價變動方向的正確率，以上之研究多停留提升預測之精準度，較少人將研究拓展至將預測股價應用於實際交易，並檢驗何種模型產生之投資績效較好，本研究的貢獻在於從深度學習模型得到預測股價後，利用程式交易的方法進一步將預測股價應用於實際交易，並驗證投資績效好壞。過去研究曾對TCN及LSTM、GRU(Gated recurrent units network)優劣做廣泛性的比較，發現 TCN 優於傳統 LSTM、GRU，但尚未有TCN 與加上殘差連接(residual connection)的 LSTM網路的比較。
　　本研究將TCN及RES LSTM(Residual LSTM)比較，應用在四個常被交易的標的上(兩個指數、微軟股票、JPMorgan Chase 股票)，檢視兩者何者創造較高好的投資績效，同時與買進持有(buy and hold)的投資績效比較。
　　研究結果發現，發現不論是預測次營業日(? =1)或次週(? =5)後收盤價、或是以TCN網路還是RES LSTM網路預測，這兩種參數變動，都得到神經元數越少，年化報酬率越高這個結論，但是神經元數越少，風險報酬比不一定越高。
　　不考慮神經元數的 8 種參數組合中，以年化報酬率而言有 5 個組合為 RES LSTM網路較好，但是考量到風險的獲利指標，風險報酬比，8 種組合中有 5 個組合為TCN網路較好。TCN 參數、訓練時間均較RES LSTM 少，但是平均年化報酬率較高、平均風險報酬比較高，也能產生較多的獲利高原，且在各參數組合中以風險報酬比比較，獲勝組數較多，大抵而言，以動量交易策略應用，TCN 網路結果優於RES LSTM 網路。

摘要(英)

There were many studies that applied deep learning to stocks, but most of them only stopped at the prediction of stock prices. These studies focused on statistical measurements, e.g. MSE; mean square error, accuracy or the directions of price movements and stopped at the level of increasing the accuracy of the predictions. Few people extended the research from the accuracy of the predictions to the application of real-life stock trading and evaluated investment performance derived from deep learning models.
This study represents a novel deep learning stock backtesting framework, which incorporated into two parts. Firstly, using a deep learning model to retrieve the prediction of stock prices. Secondly, applying the prediction of stock prices to real-life trading. We then evaluate investment performance created by the deep learning model. A prior study compared the performance of TCN and LSTM, but there haven’t been studies to compare the performance of TCN and LSTM with residual connections.
Our study compares the performance of TCN and LSTM with residual connections and applies them to four common targets (two of them are benchmarks, and the remaining two are Microsoft stock and JPMorgan Chase stock.) and then evaluates which model produces better investment performance and compares their performance with a buy-and-hold strategy.
We empirically find that no matter which business day after today we choose (i.e., one or five) and no matter which neural network we choose (i.e., TCN or RES LSTM), the fewer the neurons, the better the rate of return will be. But the reward/ risk ratio doesn’t follow the same rule.
Regarding the rate of return, RES LSTM model outperforms TCN model with five out of eight parameter sets. On the other hand, considering risk rewards, TCN model outperforms RES LSTM model with five out of eight parameter sets. Although TCN model takes less time and parameters to train, the average rate of return TCN model produces is higher and the average risk rewards TCN model produces are higher. When using TCN model, more profitable plateaus can be generated. All in all, with the application of the momentum trading strategy, the TCN network is better than the RES LSTM network.

關鍵字(中)

★ 深度學習
★ 股票投資回測模型
★ 程式交易

關鍵字(英)

★ Temporal Convolutional Network
★ Residual LSTM Network

論文目次

摘要 i
Abstract ii
致謝 iii
目錄 iv
表目錄 vii
壹. 緒論 1
1.1 研究背景 1
1.2 研究動機 1
1.3 研究目的 1
1.4 論文架構及流程 2
貳. 文獻探討 3
2.1 相關文獻簡介 3
2.1.1 遞迴網路(RNN)簡介 3
2.1.2 TCN簡介 3
2.2 長短期記憶模型 4
2.2.1 細胞狀態及三道門 4
2.2.2 遺忘門 5
2.2.3 輸入門 5
2.2.4 更新記憶區塊 5
2.2.5 輸出門 5
2.2.6 殘差遞迴網路(Residual RNNs) 6
2.3 時序卷積神經網路(TCN) 8
2.3.1 因果卷積(Causal convolutions) 8
2.3.2 空洞卷積(Dilated convolutions) 8
2.3.3 殘差連接(Resudual connections) 9
2.4 啟動函數(Activation function) 10
2.5 損失函數(Loss function ; Cost function) 11
2.6 程式交易 11
2.7 預測標的 11
參. 研究方法 13
3.1 研究設計及流程 13
3.2 Python deep learning framework 13
3.2.1 TCN模型建立 14
3.2.2 RES LSTM模型建立 14
3.2.3 模型共通性參數 14
3.2.4 資料來源 15
3.2.5 資料分割 16
3.3 MultiCharts framework 16
3.3.1 利用準確預測之收盤價的動量交易策略 17
3.4 效能評估方法 17
肆. 研究結果分析 19
4.1 環境設定 19
4.1.1 Keras api 19
4.1.2 程式交易平台MultiCharts 19
4.1.3 電腦環境設定 19
4.2 系統流程 19
4.3 系統驗證：程式交易回測之獲利比較 20
4.3.1 各參數與持有天數的交互影響 20
4.3.1.1. 不同標的 20
4.3.1.2. 預測次日或次週收盤價、不同神經網路: 23
4.3.1.3. 不同神經元數 24
4.3.1.4. 預測標的、預測次營業日(x=1)或次週(x=5)之收盤價、神經網路3個參數對報酬率勝過買進持有天數的影響明細: 26
4.3.2 持有天數外的參數對投資績效比較彙總 28
4.3.2.1. 不同神經元數投資績效: 28
4.3.2.1.1. 不同神經元數、預測次日或次週收盤價: 28
4.3.2.1.2. 不同神經元數、不同神經網路 29
4.3.2.2. 不同神經網路投資績效: 29
4.3.2.2.1. 不同神經網路、不同神經元數: 29
4.3.2.2.2. 不同神經網路、預測次日或次週收盤價: 29
4.3.2.3. 不同標的、預測次日或次週收盤價、不同神經網路投資績效明細 30
伍. 結論 32
5.1 研究限制 32
5.2 研究結果 32
5.3 後續研究方向 32
參考文獻 33
附錄 37
附錄1. 投資績效明細 37
附錄2. Multicharts程式碼 74
附錄3. 技術指標 75
附錄4. 經濟指標 77

參考文獻

一、書籍
1. Chollet, F. (2018). Deep learning with Python.
2. Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep learning: MIT press.
3. 張林忠. (2014). 分析師關鍵報告2：張林忠教你程式交易: 寰宇出版社.

二、英文會議論文(Conference Proceedings)
4. Chen, K., Zhou, Y., & Dai, F. (2015). A LSTM-based method for stock returns prediction: A case study of China stock market. Paper presented at the 2015 IEEE International Conference on Big Data (Big Data).
5. Collobert, R., & Weston, J. (2008). A unified architecture for natural language processing: Deep neural networks with multitask learning. Paper presented at the Proceedings of the 25th international conference on Machine learning.
6. Dauphin, Y. N., Fan, A., Auli, M., & Grangier, D. (2017). Language modeling with gated convolutional networks. Paper presented at the Proceedings of the 34th International Conference on Machine Learning-Volume 70.
7. El Hihi, S., & Bengio, Y. (1996). Hierarchical recurrent neural networks for long-term dependencies. Paper presented at the Advances in neural information processing systems.
8. He, K., Zhang, X., Ren, S., & Sun, J. (2015). Delving deep into rectifiers: Surpassing human-level performance on imagenet classification. Paper presented at the Proceedings of the IEEE international conference on computer vision.
9. He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. Paper presented at the Proceedings of the IEEE conference on computer vision and pattern recognition.
10. Hermans, M., & Schrauwen, B. (2013). Training and analysing deep recurrent neural networks. Paper presented at the Advances in neural information processing systems.
11. Jozefowicz, R., Zaremba, W., & Sutskever, I. (2015). An empirical exploration of recurrent network architectures. Paper presented at the International Conference on Machine Learning.
12. Klambauer, G., Unterthiner, T., Mayr, A., & Hochreiter, S. (2017). Self-normalizing neural networks. Paper presented at the Advances in neural information processing systems.
13. Long, J., Shelhamer, E., & Darrell, T. (2015). Fully convolutional networks for semantic segmentation. Paper presented at the Proceedings of the IEEE conference on computer vision and pattern recognition.
14. Martens, J., & Sutskever, I. (2011). Learning recurrent neural networks with hessian-free optimization. Paper presented at the Proceedings of the 28th International Conference on Machine Learning (ICML-11).
15. Pascanu, R., Mikolov, T., & Bengio, Y. (2013). On the difficulty of training recurrent neural networks. Paper presented at the International conference on machine learning.
16. Santos, C. D., & Zadrozny, B. (2014). Learning character-level representations for part-of-speech tagging. Paper presented at the Proceedings of the 31st International Conference on Machine Learning (ICML-14).
17. Sutskever, I., Vinyals, O., & Le, Q. V. (2014). Sequence to sequence learning with neural networks. Paper presented at the Advances in neural information processing systems.
18. Tsantekidis, A., Passalis, N., Tefas, A., Kanniainen, J., Gabbouj, M., & Iosifidis, A. (2017). Using deep learning to detect price change indications in financial markets. Paper presented at the 2017 25th European Signal Processing Conference (EUSIPCO).

三、英文期刊論文(Journal Article)
19. Bahdanau, D., Cho, K., & Bengio, Y. J. a. p. a. (2014). Neural machine translation by jointly learning to align and translate.
20. Bai, S., Kolter, J. Z., & Koltun, V. (2018). An empirical evaluation of generic convolutional and recurrent networks for sequence modeling. arXiv:.01271.
21. Bao, W., Yue, J., & Rao, Y. (2017). A deep learning framework for financial time series using stacked autoencoders and long-short term memory. PloS one, 12(7), e0180944.
22. Bengio, Y., Simard, P., & Frasconi, P. (1994). Learning long-term dependencies with gradient descent is difficult. IEEE transactions on neural networks, 5(2), 157-166.
23. Berge, T. (2014). Predicting Recessions with Leading Indicators: Model Averaging and Selection Over the Business Cycle. Journal of Forecasting.
24. Bottou, L., Soulie, F. F., Blanchet, P., & Liénard, J.-S. J. N. N. (1990). Speaker-independent isolated digit recognition: Multilayer perceptrons vs. dynamic time warping. 3(4), 453-465.
25. Cho, K., Van Merriënboer, B., Bahdanau, D., & Bengio, Y. J. a. p. a. (2014). On the properties of neural machine translation: Encoder-decoder approaches.
26. Di Persio, L., & Honchar, O. (2016). Artificial neural networks architectures for stock price prediction: Comparisons and applications. International Journal of Circuits, Systems and Signal Processing, 10, 403-413.
27. Graves, A. (2013). Generating sequences with recurrent neural networks. arXiv:.01271.
28. Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural computation, 9(8), 1735-1780.
29. Kalchbrenner, N., Espeholt, L., Simonyan, K., Oord, A. v. d., Graves, A., & Kavukcuoglu, K. J. a. p. a. (2016). Neural machine translation in linear time.
30. Kim, J., El-Khamy, M., & Lee, J. (2017). Residual LSTM: Design of a deep recurrent architecture for distant speech recognition. arXiv:.03360.
31. Kingma, D. P., & Ba, J. J. a. p. a. (2014). Adam: A method for stochastic optimization.
32. Koutnik, J., Greff, K., Gomez, F., & Schmidhuber, J. (2014). A clockwork rnn. arXiv:.01271.
33. LeCun, Y., Boser, B., Denker, J. S., Henderson, D., Howard, R. E., Hubbard, W., & Jackel, L. D. J. N. c. (1989). Backpropagation applied to handwritten zip code recognition. 1(4), 541-551.
34. Lei, T., Zhang, Y., & Artzi, Y. (2017). Training rnns as fast as cnns. arXiv:.02755.
35. Melis, G., Dyer, C., & Blunsom, P. (2017). On the state of the art of evaluation in neural language models. arXiv:.05589.
36. Merity, S., Keskar, N. S., & Socher, R. (2017). Regularizing and optimizing LSTM language models. arXiv:.02182.
37. Pradhan, S., & Longpre, S. (2016). Exploring the depths of recurrent neural networks with stochastic residual learning.
38. Prakash, A., Hasan, S. A., Lee, K., Datla, V., Qadir, A., Liu, J., & Farri, O. (2016). Neural paraphrase generation with stacked residual LSTM networks. arXiv:.03098.
39. Sejnowski, T. J., & Rosenberg, C. R. (1987). Parallel networks that learn to pronounce English text. Complex systems, 1(1), 145-168.
40. Serrà, J., Pascual, S., & Karatzoglou, A. (2018). Towards a universal neural network encoder for time series. Artif Intell Res Dev Curr Chall New Trends Appl, 308, 120.
41. Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., & Salakhutdinov, R. J. T. J. o. M. L. R. (2014). Dropout: a simple way to prevent neural networks from overfitting. 15(1), 1929-1958.
42. Tu, S. (2017). Passive Market Share to Overtake Active in the US No Later than 2024,. Moody’s Investors Service.
43. Van Den Oord, A., Dieleman, S., Zen, H., Simonyan, K., Vinyals, O., Graves, A., . . . Kavukcuoglu, K. (2016). WaveNet: A generative model for raw audio. SSW, 125.
44. Waibel, A., Hanazawa, T., Hinton, G., Shikano, K., Lang, K. J. J. B. T., Architectures, & Applications. (1995). Phoneme recognition using time-delay neural networks. 35-61.
45. Wu, Y., Schuster, M., Chen, Z., Le, Q. V., Norouzi, M., Macherey, W., . . . Macherey, K. (2016). Google′s neural machine translation system: Bridging the gap between human and machine translation. arXiv:.08144.

四、中文文獻
46. 李映瑾. (2018). 指數股票型基金(ETF) 全球發展概況與可能影響. 台北外匯市場發展基金會委託計畫.
47. 董寶蘭. (2010). 程式交易策略實證研究-以投資 ETF0050 為例. 淡江大學管理科學研究所企業經營碩士在職專班學位論文, 1-65.

五、網路資料
48. Karpathy, A. (2019, 2019). A Recipe for Training Neural Networks. Retrieved from http://karpathy.github.io/2019/04/25/recipe/
49. LLC, G. Keras. Retrieved from https://www.tensorflow.org/guide/keras/
50. Ltd, S. R. (2019, 2019). S&P 500 Sector Weightings 1979 – 2019. Retrieved from http://siblisresearch.com/data/sp-500-sector-weightings/
51. Olah, C. (2015). Understanding LSTM Networks. Retrieved from http://colah.github.io/posts/2015-08-Understanding-LSTMs/
52. 凱衛資訊股份有限公司. 交易平台介紹. Retrieved from http://www.multicharts.com.tw/characteristic.aspx

指導教授

許智誠

審核日期

2019-7-2

推文