過去有許多學者致力於探討影響股價的重要因素及研究以機器學習及深度學習模型準確預測股價，多獲得比統計模型更佳的準確率，但仍有許多可改良之處。為取得比基礎深度學習模型更好的模型績效與投資報酬率，本研究參閱過去文獻整合了多種技術面特徵（動能、價量特徵、威廉指標、KD指標等）和基本面特徵（本益比、股價淨值比、市值、股利殖利率等）作為模型的輸入，並採用了添加兩層注意力機制和圖神經網路的HAD-GNN模型進行台股市值前50大公司之報酬預測與漲跌預測，再與其他機器學習和深度學習模型進行比較。研究結果顯示，HAD-GNN模型在報酬預測方面表現較佳，具有較低的均方誤差。在投資組合選擇方面，HAD-GNN模型的投資組合累積報酬率在訓練過程中逐漸穩定並呈現正報酬，尤其在多頭行情時表現優異，長期而言有超越買進持有0050ETF策略的績效。最後本研究進一步運用SHAP模型進行特徵解釋，理解股價預測中重要特徵及其對預測結果的影響程度，使模型的預測過程更透明。綜觀30年樣本，研究顯示短期動量、交易量標準差的自然對數以及中期動量等流動性和動量相關特徵在股價預測中具有重要的預測能力。此外，針對預測結果影響較小的特徵或含有較多雜訊的情況，本研究並將其排除後重新訓練模型，導致均方誤差降至0.139。

In the past, many scholars have dedicated their efforts to exploring the key factors influencing stock prices and researching the accurate prediction of stock prices using machine learning and deep learning models. These efforts have often yielded better accuracy than statistical models, yet there remain areas for improvement. To achieve better model performance and investment returns than basic deep learning models,this study referred to past literature and integrated various technical indicators (such as momentum, price-volume features, Williams %R, KD indicators) and fundamental features (such as P/E ratio, P/B ratio, market capitalization, dividend yield) as inputs for the model. The study employed HAD-GNN model that incorporates two layers of attention mechanisms and graph neural networks to predict returns and price movements for the top 50 companies in the Taiwan stock market. This model's performance was then compared to other machine learning and deep learning models. The research findings indicate that the HAD-GNN model outperforms other models in terms of return prediction, with lower mean squared error. In terms of portfolio selection, the HAD-GNN model's accumulated portfolio return gradually stabilizes during training, showing positive returns, particularly in bullish market conditions,and exhibiting better performance over the long term than a simple "buy and hold"strategy using the 0050 ETF. Finally, the study further utilizes the SHAP model for feature interpretation, enhancing the understanding of important features and their impact on price prediction results, thus making the prediction process more transparent. Over a span of 30 years, the study shows that short-term momentum, the natural logarithm oftrading volume standard deviation, and medium-term momentum are crucial features with predictive power for stock price prediction. Moreover, the study addresses less influential or noisy features in prediction results by excluding them during model training, resulting in a reduced mean squared error of 0.139.

中文文獻
林怡汝(2022)。「以機器學習預測元大臺灣50的股價」，國立臺灣大學國際企業學系碩士學位論文。

鍾毅(2020)。「以深度學習 LSTM 方法進行台灣加權股價指數預測」，國立交通大學科技管理研究所碩士學位論文。

外文文獻
Adadi, A., & Berrada, M. (2018). Peeking inside the black-box: a survey on explainable artificial intelligence (XAI). IEEE access, 6, 52138-52160.

Banz, R. W. (1981). The relationship between return and market value of common stocks. Journal of financial economics, 9(1), 3-18

Berkson, J. (1944). Application of the logistic function to bio-assay. Journal of the American statistical association, 39(227), 357-365.

Breiman, L. (2001). Random forests. Machine learning, 45(1), 5-32.

Campbell, J. L. (1997). Mechanisms of evolutionary change in economic governance: interaction, interpretation and bricolage. Evolutionary economics and path dependence (pp. 10-32).

Carhart, M. M. (1997). On persistence in mutual fund performance. The Journal of finance, 52(1), 57-82.

Chan, L. K., Lakonishok, J., & Swaminathan, B. (2007). Industry classifications and return comovement. Financial Analysts Journal, 63(6), 56-70.

Chang, C. C., & Lin, C. J. (2011). LIBSVM: a library for support vector machines. ACM transactions on intelligent systems and technology (TIST), 2(3), 1-27.

Chen, S., & Ge, L. (2019). Exploring the attention mechanism in LSTM-based Hong Kong stock price movement prediction. Quantitative Finance, 19(9), 1507-1515.

Chen, T., & Guestrin, C. (2016, August). Xgboost: A scalable tree boosting system. In Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining (pp. 785-794).

Chung, J., Gulcehre, C., Cho, K., & Bengio, Y. (2014). Empirical evaluation of gated recurrent neural networks on sequence modeling. arXiv preprint arXiv:1412.3555.

Fama, E. F., & French, K. R. (1996). Multifactor explanations of asset pricing anomalies. The journal of finance, 51(1), 55-84.

Feng, F., He, X., Wang, X., Luo, C., Liu, Y., & Chua, T. S. (2019). Temporal relational ranking for stock prediction. ACM Transactions on Information Systems (TOIS), 37(2), 1-30.

Gu, S., Kelly, B., & Xiu, D. (2020). Empirical asset pricing via machine learning. The Review of Financial Studies, 33(5), 2223-2273.

Hochreiter, S., and Schmidhuber, J. (1997). Long short-term memory, Neural Computation 9(8):1735-1780

Hopfield, J. J.(1982). Neural networks and physical systems with emergent collective computational abilities. Authors Info & Affiliations. 79 (8), 2554-2558

Ibáñez, V. R., Gómez-Plana, A. G., & Arzoz, P. P. (2021). Raising the Accuracy of Shadow Economy Measurements. Hacienda Publica Espanola, (239), 71-125.

Lundberg, S. M., and Lee, S. I. (2017). A unified approach to interpreting model predictions. Advances in neural information processing systems 30.

Qiu, J., Wang, B., & Zhou, C. (2020). Forecasting stock prices with long-short term memory neural network based on attention mechanism. PloS one, 15(1), e0227222.

Rios, J., Zhao, K., Street, W. N., & Blackhurst, J. (2020). Stock Price Movement Cross-Predictability in Supply Chain Networks. Available at SSRN 3737658.

Sharpe, W. F. (1964). Capital asset prices: A theory of market equilibrium under conditions of risk. The journal of finance, 19(3), 425-442.

Tian, H., Zheng, X., Zhao, K., Liu, M. W., & Zeng, D. D. (2022). Inductive Representation Learning on Dynamic Stock Co-Movement Graphs for Stock Predictions. INFORMS Journal on Computing. 34(4):1940-1957

Tian, Z., Shen, C., Chen, H., & He, T. (2019). Fcos: Fully convolutional one-stage object detection. In Proceedings of the IEEE/CVF international conference on computer vision (pp. 9627-9636).

Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., ... & Polosukhin, I. (2017). Attention is all you need. Advances in neural information processing systems 30.

Zolfaghari, M., and Gholami, S.(2021) A hybrid approach of adaptive wavelet transform, long short-term memory and ARIMA-GARCH family models for the stock
index prediction. Expert Systems with Applications. Volume 182, 115149

Zou, Z., & Qu, Z. (2020). Using LSTM in Stock prediction and Quantitative Trading. CS230: Deep Learning, Winter.