摘要
本文充分结合资本市场中的有效信息,从惯性、估值与成长、交易摩擦、公司基本面、投资、盈利六个维度构造上市公司股票特征集,再利用随机森林、GBDT、XGBoost和LightGBM算法建立四种股价崩盘风险预测模型,比较考察了不同模型的预测效果以及关键影响因素。研究发现,相比单一算法模型以及其他集成算法模型,LightGBM算法模型的预测精度和效率更高。在LightGBM算法模型中,影响股价崩盘风险的关键特征包括惯性、交易摩擦、估值以及公司基本面特征。该研究结果对于上市公司及其他利益相关者动态监控股价具有实际的应用价值。
In order to prevent stock price crash risk and ensure the economic stability of enterprises,this paper fully combines the effective information in the capital market to construct a stock feature set with six dimensions and 42 variables including inertia,valuation and growth,transaction friction,firm fundamentals,invest and earning.We establishes a stock price collapse risk prediction model based on machine learning by using four ensemble learning algorithms,namely Radom Forest,GBDT,XGBoost and LightGBM.This paper found that the prediction accuracy and efficiency of LightGBM algorithm is better than that of other integrated algorithms and single learner algorithm.Market factors based on liquidity volatility and short-term momentum,fundamental factors and valuation indicators are the key indicators to predict the stock price crash risk.The research results have practical application value for listing and other stakeholders to dynamically monitor stock prices.
作者
黄鹤
汤瑛琦
刘延冰
张明媚
Huang He;Tang Yingqi;Liu Yanbing;Zhang Mingmei
出处
《管理会计研究》
2024年第1期21-38,共18页
MANAGEMENT ACCOUNTING STUDIES
