This repository contains the codes and data for the paper “Machine Learning-driven Quantamental Investing" published in a Chinese journal.
Bin Li, Xinyue Shao, and Yueyang Li. 2019. "Machine Learning-driven Quantamental Investing". China Industrial Economics, (08): 61-79.
Abstract: Quantamental investing is an emerging hot topic in financial technology and quantitative investments. As a representative technique in Artificial Intelligence (AI), machine learning can significantly improve the prediction task in economics and management. This paper investigates the application of machine learning in quantamental investing. Based on 96 anomaly factors in Chinese markets ranging from January 1997 to October 2018, we adopt LASSO regression, Ridge regression, Elastic Net, Partial Least Square, Forecast Combination, Support Vector Machine, Gradient Boost Tree, Extreme Gradient Boost Tree, Ensemble Neural Network, Deep Feedforward Network, Recurrent Neural Network, and Long-Short Term Memory to build stock return prediction model and construct portfolios. Empirical evidence shows that machine learning algorithms can efficiently identify complex patterns among anomaly factor and excess return, the investment strategy can deliver better performance than the traditional linear model and all factors. Long-short portfolios based on the forecast of Deep Feedforward Network can obtain a monthly return of 2.78%. We further evaluate factors’ importance in the prediction model, and find that trading friction factors demonstrate better predictive ability in the Chinese stock markets. Deep Feedforward Network driven quantamental investing models running on the selected feature set provide the best performance of 3.41% per month. This study introduces the machine learning techniques to the research on quantamental investing, which will further facilitate the joint research on artificial intelligence, machine learning and economics and management and finally will boost the national strategy of artificial intelligence.
Key Words: quantamental investing; machine learning; market anomaly factors; deep learning
JEL Classification: C8, G0, 011
本站点包含了李斌等（2019）的附件材料，包含完整的数据集和程序，以及论文附录等。
李斌, 邵新月, and 李玥阳. 2019. “机器学习驱动的基本面量化投资研究.” 中国工业经济 (08): 61–79.
摘要：基本面量化投资是近年来金融科技和量化投资研究的新热点。作为人工智能的代表性技术，机器学习能够大幅度提高经济学和管理学中预测类研究的效果。本文系统性地运用机器学习，来提升基本面量化投资中的股票收益预测模块。基于1997年1月至2018年10月A股市场的96项异象因子，本文采用预测组合算法、Lasso回归、岭回归、弹性网络回归、偏最小二乘回归、支持向量机、梯度提升树、极端梯度提升树、集成神经网络、深度前馈网络、循环时序网络和长短期记忆网络等12种机器学习算法，构建股票收益预测模型及投资组合。实证结果显示，机器学习算法能够有效地识别异象因子-超额收益间的复杂模式，其投资策略能够获得比传统线性算法和所有单因子更好的投资绩效，基于深度前馈网络预测的多空组合最高能够获得2.78%的月度收益。本文进一步检验了因子在预测模型中的重要性，发现交易摩擦类因子在A股市场具有较强的预测能力，深度前馈网络在筛选因子数据上的多空组合月度收益达到了3.41%。本文尝试将机器学习引入基本面量化投资领域，有助于促进人工智能、机器学习与经济学和管理学的交叉融合研究，为推进国家人工智能战略的有效实施提供参考。
关键字：基本面量化投资； 市场异象因子； 机器学习； 深度学习
中图分类号：F832