使用GBDT-SVM多层次模型优化多因子选股系统

Optimizing Multi-Factor Stock Selection System Using GBDT-SVM Multi-Level Model

摘要: 在量化投资领域中，多因子选股模型凭借稳定性高、资金容纳量大等优势被A股市场的广大专业投资者接受和使用。但近年来，模型的同质化愈发严重，基于多因子模型的投资难以获取可观的收益率。本文提出了基于大量因子的GBDT-SVM多层次选股模型，希望使用机器学习技术对因子选取和因子权重动态调整方面进行优化，提高多因子模型对股票超额收益的获取能力。之后，使用2013年至2017年中国A股市场数据进行实证研究，并与经典多因子模型和其改进模型进行比较。研究结果表明，GBDT-SVM多层次选股模型具有更高的预测准确性，历史回溯测试中获得了更高的收益率和夏普比。

Abstract: In the field of quantitative investment, the multi-factor model is widely accepted and used by investors in A-share market because of its high stability and high capital capacity. But in recent years, model homogeneity has been getting worse and investment based on multi-factor model can hardly obtain a considerable rate of return. In this work, we presents GBDT-SVM multi-level model based on big factor database, hoping to promote the multi-factor model’s ability to acquire excess return in stock investment by optimizing factor selection and factor weight dynamic adjustment using machine learning techniques. Then, we conduct empirical research using China’s A-share market data from 2013 to 2017 and compare the model with the classical multi-factor model and its improved version. The research results show that the GBDT-SVM multi-level stock selection model has higher prediction accuracy and gains higher yields and Sharp ratio in historical backtesting.