Zhao Cheng, Liu Xiaohui, Zhou Jie, Cen Yuefeng, Yao Xiaomin
School of Economics, Zhejiang University of Technology, Hangzhou, Zhejiang, China.
College of Computer Science and Technology, Zhejiang University of Technology, Hangzhou, Zhejiang, China.
PeerJ Comput Sci. 2022 Aug 11;8:e1057. doi: 10.7717/peerj-cs.1057. eCollection 2022.
Most stock price predictive models merely rely on the target stock's historical information to forecast future prices, where the linkage effects between stocks are neglected. However, a group of prior studies has shown that the leverage of correlations between stocks could significantly improve the predictions. This article proposes a unified time-series relational multi-factor model (TRMF), which composes a self-generating relations (SGR) algorithm that can extract relational features automatically. In addition, the TRMF model integrates stock relations with other multiple dimensional features for the price prediction compared to extant works. Experimental validations are performed on the NYSE and NASDAQ data, where the model is compared with the popular methods such as attention Long Short-Term Memory network (Attn-LSTM), Support Vector Regression (SVR), and multi-factor framework (MF). Results show that compared with these extant methods, our model has a higher expected cumulative return rate and a lower risk of return volatility.
大多数股价预测模型仅仅依靠目标股票的历史信息来预测未来价格，而忽略了股票之间的联动效应。然而，先前的一系列研究表明，股票之间相关性的杠杆作用可以显著提高预测效果。本文提出了一种统一的时间序列关系多因素模型（TRMF），该模型包含一种能够自动提取关系特征的自生成关系（SGR）算法。此外，与现有研究相比，TRMF模型将股票关系与其他多维度特征相结合用于价格预测。我们在纽约证券交易所和纳斯达克数据上进行了实验验证，将该模型与注意力长短期记忆网络（Attn-LSTM）、支持向量回归（SVR）和多因素框架（MF）等流行方法进行了比较。结果表明，与这些现有方法相比，我们的模型具有更高的预期累积回报率和更低的回报波动风险。
