自注意力機制模型於股價指數預測之應用

Application of Self-Attention Models to Stock Index Prediction

呂佳晏(Jia-Yan Lu)

指導教授 ： 呂育道

國立臺灣大學/電機資訊學院/資訊網路與多媒體研究所/碩士(2024年)

https://doi.org/10.6342/NTU202401343

摘要

如何在非線性、動態的金融市場中預測價格一直是一個極具挑戰性的任務，以往的研究中，與時間序列相關的模型大多數都是選擇以長短期記憶模型或是閘門循環單元為主。
本論文中使用自注意力機制模型預測股票指數，並結合一些能夠提高模型性能的策略如位置編碼、滾動式窗口驗證、早停法等，最後將自注意力機制模型產出的預測結果與長短期記憶模型比較，結果表明，自注意力機制模型在處理不同長度的時間上表現均較佳。而本模型也能夠良好的預測股價指數，在不同時間窗口的平均絕對誤差百分比均介於 0.05%~0.11%。

關鍵字

自注意力機制 ； 股價指數預測 ； 時間序列

並列摘要

Predicting price in nonlinear and dynamic financial markets has been a highly challenging task. In past studies, models dealing with time series are mostly Long Short-Term Memory (LSTM) models or Gated Recurrent Units (GRU).
This thesis employs a self-attention mechanism for stock index prediction, integrating several strategies to enhance model performance such as positional encoding, rolling window validation, and early stopping. We compare the predictive outcomes using self-attention mechanism and those of the LSTM model. The results indicate superior performance of the self-attention mechanism model across varying time spans. Additionally, the proposed model exhibits strong predictive capabilities for stock indices, with mean absolute errors percentage ranging between 0.05% and 0.11% across different time windows.

並列關鍵字

Self-Attention Mechanism ； Stock Index Prediction ； Time-Series Analysis

參考文獻

Bachelier, L. Théorie de la spéculation. (1900). Annales scientifiques de l'École Normale Supérieure, Serie 3, vol. 17, pp. 21–86.

Bengio, Y., Frasconi, P., & Simard, P. (1993). The problem of learning long-term dependencies in recurrent networks. In IEEE International Conference on Neural Networks, vol. 3, pp. 1183–1188.

Bengio, Y., Simard, P., & Frasconi, P. (1994). Learning long-term dependencies with gradient descent is difficult. IEEE Transactions on Neural Networks, vol. 5, no. 2, pp. 157–166.

Bergmeir, C., & Benítez, J. M. (2012). On the use of cross-validation for time series predictor evaluation. Information Sciences, vol. 191, pp. 192–213.

Bollerslev, T. (1986). Generalized autoregressive conditional heteroskedasticity. Journal of Econometrics, vol. 31, no. 3, pp. 307–327.