資訊管理學報
蕭國倫;劉柏辰;蔡泊均;
頁: 177-207
日期: 2024/04
摘要: 股價預測一直是一個很棘手的問題，由於許多因素都會影響股價，因此簡單的模型無法準確預測。但添加太多的特徵將增加模型的複雜度，若能找到關鍵的特徵，模型的準確性將會更好。在股票預測領域中，許多使用長短期記憶(Long Short-Term Memory, LSTM)的研究顯示了良好的結果。時間卷積網絡(Temporal Convolution Network, TCN)在時間序列研究中也取得了不錯的成果。因此，本研究採用了時間卷積網絡(TCN)與長短期記憶(LSTM)相結合的方法，並與RNN-LSTM、CNN-LSTM和長短期記憶(LSTM)等三種深度學習模型進行比較。為了比較不同模型的效能，本研究使用了不同的損失函數進行比較。結果表明，本研究提出的TCN-LSTM模型相比其他三種模型的結果更好。除了使用過去的數據集回測，本研究還基於所提出的方法開發了一種實際操作方法，並使用對沖交易進行驗證。根據股價預測趨勢圖發現，當遇到大的價格趨勢波動時，本研究提出的TCN-LSTM模型的預測結果與真實值的差距比LSTM模型更小。根據結果，我們可以得出結論，即TCN-LSTM更適用於相對活躍且對沖量較高的股票。本研究的發現，結果與討論可為該領域有所貢獻，並啟發相關領域的從業人員。
關鍵字: 股票預測;深度學習;時間卷積網路(TCN);長短期記憶(LSTM);
Abstract: Predicting stock prices has always been a challenging task due to the multitude of factors that can influence them. Adding too many features can make the model overly complex, so identifying key features is crucial for accuracy. In the field of stock price prediction, many studies have shown that Long Short-Term Memory (LSTM) models perform well. Similarly, the Temporal Convolution Network (TCN) has achieved good results in time series research. Therefore, this study combines LSTM and TCN models and compares them with RNN-LSTM, CNN-LSTM, and LSTM models for stock price prediction using various loss functions. The results indicate that the proposed TCN-LSTM model performs better than the other models. This study not only tested the proposed method with historical data sets but also validated it through hedge trading. The TCN-LSTM model proposed in this study outperforms the LSTM model in predicting stock prices during large price trend fluctuations, making it more suitable for active stocks with high hedging volume. These findings can contribute to the field and inspire practitioners in related fields.
Keywords: Stock prediction, Deep learning, Temporal Convolutional Network (TCN), Long Short-Term Memory(LSTM);
瀏覽次數: 1440 下載次數: 123
引用 導入Endnote
頁: 177-207
日期: 2024/04
摘要: 股價預測一直是一個很棘手的問題，由於許多因素都會影響股價，因此簡單的模型無法準確預測。但添加太多的特徵將增加模型的複雜度，若能找到關鍵的特徵，模型的準確性將會更好。在股票預測領域中，許多使用長短期記憶(Long Short-Term Memory, LSTM)的研究顯示了良好的結果。時間卷積網絡(Temporal Convolution Network, TCN)在時間序列研究中也取得了不錯的成果。因此，本研究採用了時間卷積網絡(TCN)與長短期記憶(LSTM)相結合的方法，並與RNN-LSTM、CNN-LSTM和長短期記憶(LSTM)等三種深度學習模型進行比較。為了比較不同模型的效能，本研究使用了不同的損失函數進行比較。結果表明，本研究提出的TCN-LSTM模型相比其他三種模型的結果更好。除了使用過去的數據集回測，本研究還基於所提出的方法開發了一種實際操作方法，並使用對沖交易進行驗證。根據股價預測趨勢圖發現，當遇到大的價格趨勢波動時，本研究提出的TCN-LSTM模型的預測結果與真實值的差距比LSTM模型更小。根據結果，我們可以得出結論，即TCN-LSTM更適用於相對活躍且對沖量較高的股票。本研究的發現，結果與討論可為該領域有所貢獻，並啟發相關領域的從業人員。
關鍵字: 股票預測;深度學習;時間卷積網路(TCN);長短期記憶(LSTM);
Predicting Stock Prices by Combining Long- and Short-Term Memory with Time Convolutional Networks: An Empirical Study on Electronic Stocks in Taiwan
Abstract: Predicting stock prices has always been a challenging task due to the multitude of factors that can influence them. Adding too many features can make the model overly complex, so identifying key features is crucial for accuracy. In the field of stock price prediction, many studies have shown that Long Short-Term Memory (LSTM) models perform well. Similarly, the Temporal Convolution Network (TCN) has achieved good results in time series research. Therefore, this study combines LSTM and TCN models and compares them with RNN-LSTM, CNN-LSTM, and LSTM models for stock price prediction using various loss functions. The results indicate that the proposed TCN-LSTM model performs better than the other models. This study not only tested the proposed method with historical data sets but also validated it through hedge trading. The TCN-LSTM model proposed in this study outperforms the LSTM model in predicting stock prices during large price trend fluctuations, making it more suitable for active stocks with high hedging volume. These findings can contribute to the field and inspire practitioners in related fields.
Keywords: Stock prediction, Deep learning, Temporal Convolutional Network (TCN), Long Short-Term Memory(LSTM);
瀏覽次數: 1440 下載次數: 123
引用 導入Endnote