Stacking框架下Boosting算法融合模型量化投资策略设计及应用  

Design and Application of Quantitative Investment Strategies for Boosting Algorithm Fusion Models under Stacking Framework

作　　者：陈创练[1,2] 邹湘妮 CHEN Chuanglian;ZOU Xiangni(School of Economics,Jinan University,Guangzhou 510632,China;Southern China Institute of Finance,Jinan University,Guangzhou 510632,China)

机构地区：[1]暨南大学经济学院,广州510632 [2]暨南大学南方高等金融研究院,广州510632

出　　处：《计量经济学报》2024年第2期425-441,共17页China Journal of Econometrics

基　　金：国家自然科学基金面上项目(72071094)。

摘　　要：金融市场瞬息万变,对于量化投资策略,需要及时地调整和优化.融合模型可以根据市场变化动态调整模型的权重和组合方式,实现自适应的调整和优化.基于此,本文尝试从融合模型的角度来设计量化投资策略.本文基于LightGBM、Adaboost、XGBoost三种不同的Boosting类算法构造了三个不同的融合两层Stacking模型,通过沪深300成分股上进行选股回测实证分析来对比择优来得到最合适的基学习模型和次级模型最佳的融合效果.实证结果表明,三种不同的融合模型在股票市场的预测表现均优于单一算法模型,其中表现最为优异的是基学习器为XGBoost和LightGBM算法,AdaBoost算法作为次级学习器的融合模型.在持仓数量为20只时,平均年化收益为13.57%,夏普比率为1.23,最大回撤为0.48.此外该回测结果表明,融合模型在市场波动性较大时有更好的适应性和有效性.本研究能够为投资者提供一种新的投资思路,也对如何推动融合模型在金融实践中运用具备一定的启示意义.The financial market is changing rapidly,and quantitative investment strategies need to be adjusted and optimized in a timely manner.The fusion model can dynamically adjust the weights of the model and the way of combination according to the market changes to achieve adaptive adjustment and optimization.In this paper,three different fusion two-layer Stacking models are constructed based on three different Boosting class algorithms,namely LightGBM,Adaboost,and XGBoost,and empirical analyses of stock picking backtesting are carried out on CSI 300 constituent stocks to compare and select the most suitable base learning model and the best fusion effect of the secondary model.The empirical results show that the three different fusion models outperform the single-algorithm models in stock market prediction,with the best performance being the fusion model where the base learners are the XGBoost and LightGBM algorithms,and the AdaBoost algorithm is used as the secondary learner.When the number of holdings is 20,the average annualized income is 13.57%,the Shape ratio is 1.23,and the maximum retracement is 0.48.In addition,the results of this backtest show that the fusion model has better adaptability and effectiveness in times of high market volatility.This study can provide investors with a new way of thinking about investment,and also provides some insights into how to promote the use of fusion models in financial practice.

关 键 词：BOOSTING算法 多因子选股模型 Stacking融合方法 量化投资策略 

分 类 号：F832.51[经济管理—金融学] F224