Market Performance - The recent 5 trading days saw the Sci-Tech 50 Index increase by 0.24%, the ChiNext Index rise by 4.13%, large-cap value stocks up by 1.55%, large-cap growth stocks up by 2.05%, the SSE 50 Index up by 1.46%, and the small-cap represented by the CSI 2000 up by 3.77% [1] - The defense and military industry, as well as the communication sector, performed well, while steel and retail sectors lagged behind [1] Risk Premium Analysis - The static PE of the CSI All Index minus the yield of 10-year government bonds indicates a risk premium, which has historically reached extreme levels at two standard deviations above the mean during significant market bottoms, such as in 2012, 2018, and 2020 [1] - As of April 26, 2022, the risk premium reached 4.17%, and on October 28, 2022, it was 4.08%, with a recent reading of 4.11% on January 19, 2024, marking the fifth occurrence since 2016 of exceeding 4% [1] Valuation Levels - As of May 9, 2025, the CSI All Index's PETTM is at the 50th percentile, with the SSE 50 and CSI 300 at 61% and 47% respectively, while the ChiNext Index is close to 11% [2] - The ChiNext Index's valuation is relatively low compared to historical averages [2] Long-term Market Trends - The technical analysis of the Deep 100 Index indicates a pattern of bear markets every three years followed by bull markets, with previous declines ranging from 40% to 45% [2] - The current adjustment cycle began in Q1 2021, suggesting a potential for upward movement from the bottom [2] Fund Flow and Trading Activity - In the last 5 trading days, ETF funds saw an outflow of 17.9 billion yuan, while margin trading increased by approximately 4.4 billion yuan [2] - The average daily trading volume across both markets was 1.2918 trillion yuan [2] AI and Machine Learning Insights - A convolutional neural network (CNN) was utilized to model price and volume data, mapping learned features to industry themes, with a current focus on banking [2][7] Market Sentiment - The proportion of stocks above the 200-day moving average is being tracked to gauge market sentiment [9] Equity and Bond Risk Preference - Ongoing monitoring of risk preferences between equity and bond assets is being conducted [11]

Core Viewpoint - The article presents a comprehensive analysis of macroeconomic and technical perspectives on major asset classes, indicating a bearish outlook for equities and industrial products, while being bullish on bonds and gold [1][3][21]. Group 1: Macroeconomic Perspective - The macroeconomic indicators suggest a negative outlook for equity assets, a positive outlook for bond assets, and a negative outlook for industrial products, while gold assets are viewed positively [3][5][21]. - Specific macro indicators such as PMI, CPI, and social financing stock growth rates are analyzed to determine their impact on asset performance [6][21]. Group 2: Technical Perspective - The technical analysis indicates a downward trend for equities, bonds, and industrial products, while gold shows an upward trend [10][11][21]. - The article employs various methods to assess asset trends, including historical price averages and specific trend indicators [7][11]. Group 3: Asset Valuation and Fund Flow - The equity risk premium (ERP) for the CSI 800 index is reported at 86.07%, indicating a low valuation level for equity assets [14][15]. - As of April 30, 2025, the net inflow for equity assets is recorded at 557 billion, suggesting a state of capital inflow [17][18]. Group 4: Performance Tracking of Asset Allocation Combinations - Historical performance data shows that the fixed ratio combined with macro and technical indicators yielded a return of 0.05% in April 2025, with an annualized return of 11.87% since March 2006 [2][26]. - Other combinations, such as volatility control and risk parity, also demonstrated positive returns, with annualized returns of 9.33% and 9.64% respectively [26][27].

Core Viewpoint - The article emphasizes the increasing importance of factor development and iteration in multi-factor models due to the declining returns from traditional factors and the challenges posed by factor crowding [1][3][62]. Factor Construction - The "Index Enhanced ETF Factor" is constructed using daily subscription and redemption data from index-enhanced ETFs, comparing the actual allocation weights of fund managers to the benchmark index weights to derive relative allocation (also known as "underweight") ratios [1][8]. - This process allows for the creation of signals based on fund managers' actual stock preferences, enhancing active management strategies [1][8]. Empirical Analysis - The constructed "Index Enhanced ETF Factor" shows a significant monotonic increase in returns across various indices (CSI 300, CSI 500, CSI 1000, and CSI 2000) during weekly backtesting, with notable excess returns for the top groups compared to the bottom groups [2][22]. - The factor's Information Coefficient (IC) performance is robust, with IC win rates of 62.42% for CSI 300, 64.33% for CSI 500, 72.32% for CSI 1000, and 60.00% for CSI 2000, indicating strong predictive power [2][40][43]. High-Frequency vs. Low-Frequency Data - High-frequency data offers advantages in factor development due to its larger volume and the ability to create diverse features through advanced techniques like machine learning, despite the challenges of noise and complexity [4][5][6]. - Low-frequency data, while more traditional, has limited incremental information, making it harder to extract significant alpha, thus necessitating innovative approaches to factor construction [6][62]. Strategy Explanation - The strategy involves tracking fund managers' preferences through the ETF's daily disclosure of holdings, allowing for the identification of stocks with higher expected returns based on their relative underweight status [8][62]. - The performance of index-enhanced ETFs has shown consistent outperformance against their benchmarks, validating the strategy's rationale [9][62]. Backtesting Results - The backtesting results indicate that the "Index Enhanced ETF Factor" has demonstrated significant cumulative returns across the four major indices, with a clear upward trend in group returns from low (G1) to high (G5) [22][62]. - The factor's IC values have shown a steady increase over time, particularly in the CSI 500 and CSI 1000 indices, highlighting its effectiveness in capturing excess returns [62][63]. Conclusion - The "Index Enhanced ETF Factor" effectively tracks fund managers' actual stock preferences, showing significant empirical validity in its ability to generate excess returns across various indices [62][63]. - The strategy is particularly well-suited for capturing structural opportunities in a rapidly changing market environment, outperforming traditional passive strategies [63].

Group 1 - The overall holding value of northbound funds reached 2.24 trillion RMB as of March 31, 2025, an increase of approximately 25.7 billion RMB compared to the end of Q4 2024, accounting for about 5.5% of the free float market value of A-shares [1][8][11] - Long-term allocation funds from foreign banks held 1.71 trillion RMB, increasing by about 10.8 billion RMB, representing 4.2% of the free float market value, while short-term trading funds from foreign brokerages held 0.38 trillion RMB, increasing by approximately 11.2 billion RMB, accounting for 0.93% [1][8][11] Group 2 - Northbound funds showed a significant increase in allocation to momentum, liquidity, and growth styles in Q1, reversing the previous quarter's reduction in these areas [2][17][22] - The overall style preferences of northbound funds included overweight positions in market capitalization, momentum, volatility, profitability, growth, and leverage, while underweight positions were noted in beta, BP, and liquidity [2][20][25] Group 3 - The highest holding value proportion of northbound funds was in the consumer sector at 6.9%, followed by financials at 6.0%, with a slight increase in the cyclical sector [3][28][32] - Northbound funds were overweight in consumer and financial sectors compared to the overall A-share market, while they were underweight in stability, technology, and cyclical sectors [3][38][42] Group 4 - The top five industries for northbound funds in terms of holding proportion changes were automotive, retail, consumer services, machinery, and electronics, while the bottom five included utilities, financials, telecommunications, real estate, and construction [3][42][45] - Northbound funds were overweight in industries such as power equipment and new energy, food and beverage, home appliances, banking, and automotive, while underweight in computer, basic chemicals, machinery, defense, and electronics [3][51][52] Group 5 - In terms of index allocation, northbound funds showed a decrease in holding proportions for the Shanghai 50 (-0.5%), CSI 300 (-0.3%), and CSI 500 (-0.2%), while there was a slight increase for the CSI 1000 (+0.1%) [4][58][62] - Northbound funds were overweight in the Shanghai 50 and CSI 300 compared to the overall A-share market, while underweight in the CSI 500 and CSI 1000 [4][67]

广发证券首席金工分析师 安宁宁 SAC: S0260512020003 anningning@gf.com.cn 广发证券资深金工分析师 陈原文 SAC: S0260517080003 chenyuanwen@gf.com.cn 广发证券资深金工分析师 王小康 SAC: S0260525020002 wangxiaokang@gf.com.cn 广发金工安宁宁陈原文团队 摘要 公式化因子挖掘与AlphaForge框架介绍 ： 各类神经网络模型作为一种编码方案，能较好预测未来一段时间股票截面收益率的差异。而构造更多的公式化 特征作为模型输入也是比较重要的环节，从理论上具有丰富模型输入，代替一部分编码器职能的效果。传统的框架包括遗传规划、OpenFE等，均无法实 现具有方向性的优化迭代，而AlphaGen虽然通过将因子表现作为奖励不断优化生成动作，但依然存在超参数敏感、容易过拟合的情况。AlphaForge通过创 新性的框架设计，一定程度上解决了上述问题。 基于AlphaForge的因子挖掘 ： 该框架首先通过设计若干算子、回看天数和基础特征得到潜在的因子库。因子会依次经过生成器和预测器完成训 练，其中预测器的主 ...

广发证券首席金工分析师 安宁宁 SAC: S0260512020003 anningning@gf.com.cn 广发证券资深金工分析师 张钰东 SAC: S0260522070006 zhangyudong@gf.com.cn 广发金工安宁宁陈原文团队 摘要 最近5个交易日，科创50指数跌0.40%，创业板指涨1.74%，大盘价值跌0.30%，大盘成长涨0.89%，上证50跌0.33%，国证2000代表的小盘涨2.38%，汽车、 美容护理市场表现靠前，食品饮料、房地产表现靠后。 风险溢价，中证全指静态PE的倒数EP减去十年期国债收益率，权益与债券资产隐含收益率对比，历史数次极端底部该数据均处在均值上两倍标准差区 域，比如2012/2018/2020年（疫情突发），2022/04/26达到4.17%，2022/10/28风险溢价再次上升到4.08%，市场迅速反弹，2024/01/19指标4.11%，自2016年 以来第五次超过4%。截至2025/04/25指标3.99%，两倍标准差边界为4.75%。 估值水平，截至2025/04/25，中证全指PETTM分位数48%，上证50与沪深300分别为60%、46%， ...

广发证券资深金工分析师 张钰东 SAC: S0260522070006 SAC: S0260517080003 chenyuanwen@gf.com.cn 广发金工安宁宁陈原文团队 摘要 ETF市场概况： 指数化投资理念愈发受到投资者认可，ETF产品凭借透明、低费率、交易便捷等优势，成为居民资产配置的重要工具，ETF 规模持续创新高，ETF资金流变动逐渐成为市场中的关注重点。 ETF交易机制特点： ETF具有独特的双层交易机制，即一级市场的申购赎回和二级市场的买卖交易，一级申赎指用一篮子股票换取ETF份 额，申赎会直接增减ETF的总份额。本报告旨在针对申赎导致的ETF资金流数据，探索用于ETF轮动的配置效果。 zhangyudong@gf.com.cn 广发证券首席金工分析师 安宁宁 SAC: S0260512020003 anningning@gf.com.cn 广发证券 资深金工分析师 陈原文 风险提示： 本专题报告所述模型用量化方法通过历史数据统计、建模和测算完成，所得结论与规律在市场政策、环境变化时可能存在失效风 险；策略在市场结构及交易行为的改变时有可能存在策略失效风险；因量化模型不同，本报告提出的 ...

广发证券首席金工分析师 安宁宁 SAC: S0260512020003 anningning@gf.com.cn 广发证券资深金工分析师 张钰东 SAC: S0260522070006 zhangyudong@gf.com.cn 广发金工安宁宁陈原文团队 摘要 最近5个交易日，科创50指数跌0.31%，创业板指跌0.64%，大盘价值涨2.62%，大盘成长跌0.24%，上证50涨1.45%，国证2000代表的小盘涨0.05%，银行、 房地产市场表现靠前，国防军工、农林牧渔表现靠后。 风险溢价，中证全指静态PE的倒数EP减去十年期国债收益率，权益与债券资产隐含收益率对比，历史数次极端底部该数据均处在均值上两倍标准差区 域，比如2012/2018/2020年（疫情突发），2022/04/26达到4.17%，2022/10/28风险溢价再次上升到4.08%，市场迅速反弹，2024/01/19指标4.11%，自2016年 以来第五次超过4%。截至2025/04/18指标4.05%，两倍标准差边界为4.74%。 估值水平，截至2025/04/18，中证全指PETTM分位数47%，上证50与沪深300分别为60%、45%， ...

摘要 自由现金流（Free Cash Flow, FCF）：是衡量企业财务健康状况和投资价值的重要指标之一，反映了企业在满足必要的资本支出后，能够自由支配的现金 金额，反映真实的现金流生成能力。 自由现金流是基于现金流的指标，反映企业实际可用的现金 ，可以用于企业现金留存、分红回购、还债付息等，是衡量企业财务健康和盈利能力的重要指 标之一。 自由现金流充裕的企业相对更能规避价值陷阱，同时分享长期增长红利， 相对更易穿越经济周期，在政策支持与市场风格切换的背景下， 自由现 金流组合有望成为资产配置的 "压舱石"。 中证全指自由现金流指数： 选取100只自由现金流率较高的上市公司证券作为指数样本，以反映现金流创造能力较强的上市公司证券的整体表现。 成分股覆盖周期和消费权重相对较高。 根据Wind，截至2025年4月14日，中证自由现金流指数的前5大行业合计权重为65%，重仓行业为煤炭、石油石化、 交通运输、食品饮料和家用电器。覆盖大中小市值股票，其中 小市值股票数量相对较多。持仓相对集中， 前十大重仓股合计权重为65.53%. 历史业绩长期优于沪深300等宽基指数。 以2014年至今作为观察周期，考虑分红再投资， ...

实习时间： 每周至少实习3天以上，实习时间不少于3个月，不满足的请勿投递，实习考核优秀者有留用机会。 岗位职责： 1、负责数据处理、分析、统计等工作，协助研究员完成量化投资相关课题的研究； 实习生招聘 工作地点： 深圳、广州、上海、北京 ，要求线下实习 简历投递截止日期： 2025年4月30日 2、协助进行金融工程策略模型的开发与跟踪等工作； 3、完成小组安排的其他工作。 基本要求： 1、数学、统计、物理、计算机、信息工程等理工科专业，或金融工程相关专业，硕士或博士在读，特别优秀的大四 保研亦可，非应届（2026年及之后毕业）； 2、熟练掌握Python等编程语言，熟悉SQL数据库，有优秀编程能力与编程规范； 3、有责任心，自我驱动能力强， 具有良好的信息搜集能力、逻辑思维能力、分析判断能力、言语和书面表达能力、 人际沟通能力。 加分项： 4、 具备扎实的金融市场基础知识，熟悉股票、债券、期货、指数及基金等核心概念； 5、数学基础好，有科研项目经历、有学术论文被SCI或EI收录； 6、熟悉Wind、 Bloomberg、天软等金融终端； 7、熟悉机器学习、深度学习，熟悉PyTorch、Linux，有GPU服务 ...