Core Insights - The article discusses the advancements made by Galaxy Fund's quantitative team in deep quantitative stock selection research, emphasizing the shift from traditional linear analysis to nonlinear analysis for better market insights and investment opportunities [1][2] Group 1: Quantitative Research Strategies - The quantitative research approach combines linear and nonlinear strategies, utilizing multi-factor models alongside nonlinear machine learning models to achieve stable excess returns and reduce tracking errors [1][2] - The team has developed strategies that include both linear methods, primarily multi-factor models, and nonlinear methods such as XGBoost and LightGBM, which enhance the model's adaptability to market changes [2][3] Group 2: Neural Network Development - The development of complex neural network learning is highlighted, where the approach integrates long-term rules with short-term information to improve the training of supervised learning models [3] - The focus is on extracting features from raw data while addressing the noise present in the data, which aids in the model's ability to adapt quickly to market fluctuations [3] Group 3: Satellite Strategies - To further enhance market adaptability, satellite strategies are employed, including dividend selection and large-cap growth selection, which target specific market characteristics [4] - The dividend selection strategy focuses on high dividend yield stocks, while the large-cap growth strategy emphasizes stocks with large market capitalization and high growth potential [4] Group 4: Risk Management and Product Development - A financial risk management strategy has been developed to mitigate unexpected impacts from risk events, forming a comprehensive quantitative strategy system [5] - The Galaxy Fund has launched two index enhancement products: the Galaxy CSI 300 Index Enhanced Fund and the Galaxy CSI A500 Index Enhanced Fund, with plans to issue the Galaxy CSI 800 Index Enhanced Fund, which offers a balanced representation of both large-cap and mid-cap growth styles [5]

"比如，XGBoost可以对因子的重要性进行展示，通过排序帮助我们识别哪些因子更加重要，增强模型 对市场变化的适应能力。尤其是在今年的结构化行情下，非线性策略能够抓住一些弹性品种的机 会。"罗博表示。 □本报记者 王鹤静 为适应复杂的市场环境，银河基金量化团队近年来在深度量化选股研究领域持续深耕，突破传统线性分 析对历史回测分析的局限，通过非线性分析方式，更加精准地分析市场，挖掘其中的投资机遇。 日前，银河基金量化与FOF投资部总监助理、基金经理罗博在接受中国证券报记者采访时介绍了量化研 究的新思路。在指数样本增强方面，罗博主要采取线性和非线性相结合的方式，由多因子模型与非线性 的机器学习模型互相协作，力争获得相对稳健的超额收益；同时，由于模型间相关性较低，力争有效降 低整体组合的跟踪误差。 开发深度神经网络学习 罗博具有21年证券从业经验、15年公募基金管理经验，长期扎根于指数与量化投资领域。随着市场环境 持续发生变化，罗博意识到，量化投资仅仅依靠线性分析把握市场长期规律愈发难以支撑，在行业趋势 从线性向非线性过渡的过程中，需要不断学习非线性分析技术，紧跟市场变化。 经过近年来的逐步积累和完善，目前罗博针对 ...

□本报记者 王鹤静 为适应复杂的市场环境，银河基金量化团队近年来在深度量化选股研究领域持续深耕，突破传统线性分 析对历史回测分析的局限，通过非线性分析方式，更加精准地分析市场，挖掘其中的投资机遇。 日前，银河基金量化与FOF投资部总监助理、基金经理罗博在接受中国证券报记者采访时介绍了量化研 究的新思路。在指数样本增强方面，罗博主要采取线性和非线性相结合的方式，由多因子模型与非线性 的机器学习模型互相协作，力争获得相对稳健的超额收益；同时，由于模型间相关性较低，力争有效降 低整体组合的跟踪误差。 开发深度神经网络学习 罗博具有21年证券从业经验、15年公募基金管理经验，长期扎根于指数与量化投资领域。随着市场环境 持续发生变化，罗博意识到，量化投资仅仅依靠线性分析把握市场长期规律愈发难以支撑，在行业趋势 从线性向非线性过渡的过程中，需要不断学习非线性分析技术，紧跟市场变化。 在简单的神经网络学习基础上，罗博做了进一步的开发和挖掘，看好复杂神经网络学习。"简单的神经 网络学习主要是根据原始数据来提取个股的特征，用未来一段时间的预期收益率作为标签，进行有监督 的学习。但原始数据可能存在较大的'噪音'，很难训练出一个收 ...