Li Shengnan, Xue Lei
School of Communication and Information Engineering, Shanghai University, Shanghai, China.
PeerJ Comput Sci. 2024 Jan 30;10:e1819. doi: 10.7717/peerj-cs.1819. eCollection 2024.
Stock price prediction is crucial in stock market research, yet existing models often overlook interdependencies among stocks in the same industry, treating them as independent entities. Recognizing and accounting for these interdependencies is essential for precise predictions. Propensity score matching (PSM), a statistical method for balancing individuals between groups and improving causal inferences, has not been extensively applied in stock interdependence investigations. Our study addresses this gap by introducing PSM to examine interdependence among pharmaceutical industry stocks for stock price prediction. Additionally, our research integrates Improved particle swarm optimization (IPSO) with long short-term memory (LSTM) networks to enhance parameter selection, improving overall predictive accuracy. The dataset includes price data for all pharmaceutical industry stocks in 2022, categorized into chemical pharmaceuticals, biopharmaceuticals, and traditional Chinese medicine. Using Stata, we identify significantly correlated stocks within each sub-industry through average treatment effect on the treated (ATT) values. Incorporating PSM, we match five target stocks per sub-industry with all stocks in their respective categories, merging target stock data with weighted data from non-target stocks for validation in the IPSO-LSTM model. Our findings demonstrate that including non-target stock data from the same sub-industry through PSM significantly improves predictive accuracy, highlighting its positive impact on stock price prediction. This study pioneers PSM's use in studying stock interdependence, conducts an in-depth exploration of effects within the pharmaceutical industry, and applies the IPSO optimization algorithm to enhance LSTM network performance, providing a fresh perspective on stock price prediction research.
股价预测在股票市场研究中至关重要，但现有模型往往忽视同一行业内股票之间的相互依存关系，将它们视为独立个体。认识并考虑这些相互依存关系对于精确预测至关重要。倾向得分匹配法（PSM）是一种用于平衡组间个体并改善因果推断的统计方法，尚未在股票相互依存性研究中得到广泛应用。我们的研究通过引入PSM来检验制药行业股票之间的相互依存关系以进行股价预测，填补了这一空白。此外，我们的研究将改进粒子群优化算法（IPSO）与长短期记忆网络（LSTM）相结合，以优化参数选择，提高整体预测准确性。数据集包括2022年所有制药行业股票的价格数据，分为化学制药、生物制药和中药。我们使用Stata软件，通过处理组平均处理效应（ATT）值来识别每个子行业内显著相关的股票。结合PSM，我们为每个子行业的五只目标股票与各自类别中的所有股票进行匹配，将目标股票数据与来自非目标股票的加权数据合并，用于IPSO-LSTM模型的验证。我们的研究结果表明，通过PSM纳入同一子行业的非目标股票数据可显著提高预测准确性，凸显了其对股价预测的积极影响。本研究率先将PSM用于研究股票相互依存关系，深入探讨了制药行业内部的影响，并应用IPSO优化算法提升LSTM网络性能，为股价预测研究提供了新视角。
