- 學位論文
整合深度學習預測模型於股票投資策略
Integrating Deep Learning Predictive Models into Stock Investment Strategies
摘要
股票投資是一個歷史悠久的問題，人們總是期盼提高投資獲利並且降低風險，分析投資標的之多元資訊遂成為主要挑戰，依照資訊內涵可以粗略分為公司財務價值和股票交易狀況二類的股票指標。本研究將兩類指標都納入考量，以不同架構的深度學習類神經網路建立股價預測模型，進一步預估月報酬率後，以驗證誤差選出最適模型，最後將預測模型整合於股票投資策略。我們採用台灣不同類股的20年資料進行實驗，在不同參數設定和不同策略運用下，分別觀察預測模型的誤差及投資策略的獲利。結果顯示加入深度學習模型的投資獲利超過傳統方法獲利的7倍以上，另一項有趣的發現是預測模型的輸出值在更改預測指標後投資獲利可提高47%。
並列摘要
Stock investment is a long-standing problem. People always expect to increase investment profits but reduce risk. The analysis of various information about investment targets becomes the main challenge. Stock data, according to their semantic, can be roughly classified into two categories of stock indicators; company financial value and stock transaction status. Our study takes both categories into consideration and adopts different deep learning neuro networks to build stock price prediction models. The monthly rates of return are further estimated to find the best model with respect to the errors on validation data. Finally, the best model is integrated into the strategies for stock investment. In experiments, we use the data of 20 years from Taiwan stock market. With various settings of parameter and methods, we observe the errors of predictive models and profits of investment strategies, respectively. The results show that our approach integrating a deep learning model achieves a profit more than 7 times of that obtained by a conventional method. Another interesting finding is that the profit of our approach raises 47% after the output of predictive index is slightly adjusted.