机器学习
Search documents
量化专题报告:“机器学习”选股模型系列研究(一):量价指纹模型的构建与应用初探
GOLDEN SUN SECURITIES· 2026-01-16 13:34
Quantitative Models and Construction Methods - **Model Name**: Volume-Price Fingerprint Model **Model Construction Idea**: Inspired by large language models, the volume-price fingerprint model treats market transaction data as a special "language" and uses self-supervised learning to extract semantic features from intraday volume-price behavior [1][8][9] **Model Construction Process**: 1. **Minute-level Feature Preprocessing**: Select 32 dimensions of minute-level features, including price features (e.g., high, low, close, price position) and transaction features (e.g., turnover, order cancellation, fund flow). Standardize these features to eliminate dimensional and historical volatility effects [1][16][17] - Price feature standardization formula: $$\tilde{p}_{t,d}=\frac{p_{t,d}}{p_{\mathrm{open}}}-1$$ [16] - Transaction feature standardization formula: $${\tilde{f}}_{t,d}={\frac{f_{t,d}}{S_{d}}},\quad S_{d}={\frac{1}{N_{\mathrm{hist}}}}\sum_{i=1}^{N_{\mathrm{hist}}}\sum_{t=1}^{T}f_{t,d}^{(i)}$$ [17] 2. **Dual-task Self-supervised Learning Framework**: - **Forward Causal Prediction Task**: Predict price features causally using past transaction and price information. A triangular attention mask ensures strict causality [18][21] - **Backward Feature Reconstruction Task**: Randomly mask transaction features and reconstruct them using global sequence information [18][22] 3. **Anti-collapse Regularization**: Introduce diversity, orthogonality, and uniformity regularization terms to ensure high differentiation, low redundancy, and rich information in fingerprint vectors [1][43][44][46] - Total loss function: $${\mathcal{L}}_{\mathrm{total}}=\lambda_{f}{\mathcal{L}}_{\mathrm{forward}}+\lambda_{b}{\mathcal{L}}_{\mathrm{backward}}+{\mathcal{L}}_{\mathrm{diversity}}+{\mathcal{L}}_{\mathrm{orthogonality}}+{\mathcal{L}}_{\mathrm{uniformity}}$$ [47] **Model Evaluation**: The model provides a structured representation of market dynamics, capturing semantic features beyond traditional numerical predictions [1][9][14] - **Model Name**: GRU Model with Volume-Price Fingerprint **Model Construction Idea**: Use GRU to predict future stock returns based on volume-price fingerprint features [2][50][51] **Model Construction Process**: 1. Input features include 128-dimensional volume-price fingerprint vectors and basic daily features (e.g., open, high, low, close, volume, turnover) [51][52] 2. GRU structure: Two-layer GRU + fully connected layers + LayerNorm + Relu + dropout + fully connected layers [53] 3. Training details: Batch size of 512, learning rate of 1e-4 with warmup, early stopping after 10 rounds without improvement [53][54] **Model Evaluation**: The GRU model effectively utilizes the semantic features of volume-price fingerprints for stock prediction, outperforming traditional factor-based models in certain metrics [2][50][54] - **Model Name**: Dual-stream GRU Model **Model Construction Idea**: Combine volume-price fingerprints and traditional volume-price factors using dual-stream GRU to leverage complementary information [67][68] **Model Construction Process**: 1. Separate GRU streams for volume-price fingerprints and traditional factors, followed by feature fusion through configurable weights [67][69] 2. Training details: Similar to single-stream GRU, with parallel training using different random seeds for robustness [68][69] **Model Evaluation**: The dual-stream GRU model improves prediction accuracy and stability, reducing overfitting risks associated with single data sources [68][69] Model Backtesting Results - **Volume-Price Fingerprint Model**: - Weekly RankIC Mean: 0.106 - Annualized Return (10-group long-short): 83.88% - IR: 5.41 - Weekly Win Rate: 73.87% - Max Drawdown: 11.65% [2][59][65] - **GRU Model with Volume-Price Fingerprint**: - Weekly RankIC Mean: 0.106 - Annualized Return (10-group long-short): 83.88% - IR: 5.41 - Weekly Win Rate: 73.87% - Max Drawdown: 11.65% [2][59][65] - **Dual-stream GRU Model**: - Weekly RankIC Mean: 0.109 - Annualized Return (10-group long-short): 90.89% - IR: 5.95 - Weekly Win Rate: 76.46% - Max Drawdown: 11.54% [68][74] Quantitative Factors and Construction Methods - **Factor Name**: Volume-Price Fingerprint Factor **Factor Construction Idea**: Extract semantic features from intraday volume-price behavior using self-supervised learning [1][9][14] **Factor Construction Process**: 1. Generate 128-dimensional daily fingerprint vectors using the volume-price fingerprint model [1][16][18] 2. Ensure high differentiation, low redundancy, and rich information through anti-collapse regularization [43][44][46] **Factor Evaluation**: The fingerprint factor captures hidden market patterns and semantic information, complementing traditional factors [1][9][14] Factor Backtesting Results - **Volume-Price Fingerprint Factor**: - Weekly RankIC Mean: 0.106 - Annualized Return (10-group long-short): 83.88% - IR: 5.41 - Weekly Win Rate: 73.87% - Max Drawdown: 11.65% [2][59][65] - **Fusion Factor (Volume-Price Fingerprint + Traditional Factors)**: - Weekly RankIC Mean: 0.109 - Annualized Return (10-group long-short): 90.89% - IR: 5.95 - Weekly Win Rate: 76.46% - Max Drawdown: 11.54% [68][74] Index Enhancement Results - **CSI 300 Enhanced Portfolio**: - Annualized Return: 11.00% - Excess Annualized Return: 7.12% - Tracking Error: 1.74% - IR: 4.10 - Monthly Win Rate: 86.11% - Max Drawdown: 1.85% [75][77] - **CSI 500 Enhanced Portfolio**: - Annualized Return: 13.32% - Excess Annualized Return: 11.38% - Tracking Error: 3.47% - IR: 3.28 - Monthly Win Rate: 83.33% - Max Drawdown: 4.76% [78][80] - **CSI 1000 Enhanced Portfolio**: - Annualized Return: 13.23% - Excess Annualized Return: 14.84% - Tracking Error: 3.45% - IR: 4.30 - Monthly Win Rate: 83.33% - Max Drawdown: 2.95% [82][83]
全球首座储能电池“灯塔工厂”来了
行家说储能· 2026-01-16 10:19
Core Viewpoint - The article highlights the significance of "Lighthouse Factories" in the context of the digital transformation of global manufacturing, particularly in the energy storage battery sector, with the successful inclusion of the HaiCheng Energy Storage Chongqing Base in the 2026 Lighthouse Factory list by the World Economic Forum [2][3]. Group 1: Lighthouse Factory Concept - "Lighthouse Factories" are selected by the World Economic Forum and McKinsey to represent leading examples of smart manufacturing in the Fourth Industrial Revolution, focusing on productivity and sustainability [3]. - As of now, there are 224 Lighthouse Factories globally, with the HaiCheng Energy Storage Chongqing Base being the first in the energy storage battery sector, marking a significant advancement for China's position in smart manufacturing and digital upgrades [3]. Group 2: Intelligent Manufacturing System - The Chongqing Base aims to address three major challenges in the lithium battery storage industry: high compound annual growth rate in market demand, downward pressure on product prices, and increasing quality consistency requirements, which traditional manufacturing models struggle to meet [4]. - The Chongqing Base implements near-zero defect manufacturing, systematic cost reduction, and full-process intelligent operations to enhance digital and intelligent upgrades, significantly improving product quality, capacity, and overall equipment efficiency [4]. Group 3: Technological Advancements - The Chongqing Base is described as a "thinking" factory, utilizing over 40 digital solutions that integrate generative AI, machine learning, and AIoT technologies, creating a comprehensive intelligent management system covering R&D, material selection, production, and product testing [5]. - At a recent energy summit, HaiCheng showcased its long-duration energy storage solutions, including the ∞Power 6.9MWh system and the ∞Cell 1300Ah battery, marking their international debut [6]. Group 4: Future Outlook - The Chongqing Base has established the world's first dedicated production line for long-duration energy storage batteries, with plans for mass production of the ∞Cell 1175Ah battery by June 2025 and subsequent global delivery of the ∞Power 6.25MWh system [10]. - The company aims to leverage its Lighthouse Factory status to enhance smart manufacturing and digital capabilities, drive digital upgrades across the supply chain, and contribute to the global green and sustainable energy development with replicable manufacturing experiences [10].
凌晨点外卖次数过多,银行卡被风控冻结,银行反诈系统引争议
Mei Ri Jing Ji Xin Wen· 2026-01-16 03:36
凌晨点外卖次数过多,银行卡竟被风控冻结。一套基于机器学习的风控系统,正让寻常消费行为与电诈特征在深夜的支付路口狭路相逢。 日前,一名网友在社交平台展示了一张带有反诈中心公章的解封证明,其"可疑交易说明"一栏写着"主要是晚上凌晨点外卖次数较多被风控"。这位网友调 侃道:"没玩梗,是真的,盖章之后解开了。" | | | C-C-C | | | --- | --- | --- | --- | | 联系地址 | | S | | | 人联系方式 | | | | | 史发说说明 | 十两具唯一济量占外工作领导 安散人宜。 | | | | 中心审核意见 | | | | | 是否建议解除官 | | | | | 控) | | | | | 心审核人(签章) | 联系方式 | | 官 | | | 注:此表一式两份,银行(金融机构)开户网点、公安机关反 | | | 金融消费者如何降低被"误伤"的概率?万一被管控,如何高效解决?银行业研究人士表示,银行的风控系统主要是通过机器学习模型,识别与已知电信诈 骗等非法活动相似的可疑交易模式。金融消费者要注意避开敏感交易特征、保持信息真实有效、远离风险账户等。 消费者被"误伤": 风控模型难以 ...
深圳“技客”故事|黄佳杰:从“00后”专科生到云计算“世界冠军”
Sou Hu Cai Jing· 2026-01-15 11:16
Core Viewpoint - Huang Jiajie, a representative from Shenzhen City Vocational College, won the gold medal in the cloud computing category at the 47th World Skills Competition, marking a significant achievement for the Chinese team in this event [1][4]. Group 1: Competition Overview - The cloud computing project at the World Skills Competition is unique as it provides real-time scoring, adding pressure to competitors [4]. - Huang faced high-level competition from engineers at major companies like Samsung and Google, as well as students from prestigious universities [4]. - Despite challenges, including unexpected changes in the competition format, Huang demonstrated strong adaptability and strategic planning, leading to his success [4][6]. Group 2: Training and Preparation - Huang trained for over five years, dedicating more than 11 hours daily to practice, focusing on various aspects of cloud computing [9]. - His training included participation in multiple competitions, which helped him build the necessary skills to compete at the national level [9][10]. - The collaboration between his school and leading tech companies provided him with a solid foundation in cloud computing, including obtaining top-level certifications [10]. Group 3: Personal Growth and Future Aspirations - After winning the championship, Huang chose to return to his alma mater as a teacher, aiming to pass on his knowledge and experience to future generations [13][14]. - He emphasizes the importance of continuous learning in the rapidly evolving field of cloud computing and encourages young people to stay updated with new technologies [15]. - Huang's commitment to education is reflected in his students' success, as they achieved gold in a provincial competition, showcasing the impact of his mentorship [14].
准确率达97%,普林斯顿大学等提出MOFSeq-LMM,高效预测MOFs能否被合成
3 6 Ke· 2026-01-15 11:10
Core Insights - A joint research team from Princeton University and the Colorado School of Mines has developed a machine learning-based method for efficiently predicting the free energy of Metal-Organic Frameworks (MOFs), significantly reducing computational costs and enabling high-throughput thermodynamic assessments [2][12]. Group 1: Research Methodology - The proposed method utilizes a large language model (LLM) to predict free energy directly from the structural sequences of MOFs, achieving an F1 score of 97% in determining whether the free energy exceeds a threshold for synthetic feasibility [2][29]. - The research team constructed a large dataset named MOFMinE, which includes approximately 1 million MOF prototypes, providing comprehensive information from component selection to functional modifications [7][10]. - The MOFSeq-LMM model framework was developed to facilitate efficient free energy predictions, transforming MOF structural information into a computer-readable sequence representation [12][13]. Group 2: Data Characteristics - MOFMinE encompasses 1,393 topological templates, 27 inorganic building blocks, 14 organic building blocks, and 19 basic edge building blocks, ensuring diversity in chemical and topological structures [10]. - A subset of 65,574 structures within MOFMinE contains free energy data, which is utilized for fine-tuning and testing the LLM [11]. Group 3: Model Performance - The LLM-Prop model, designed for material property predictions, achieved an average absolute error of 0.789 kJ/mol per MOF atom in free energy predictions, with a high correlation coefficient (R² = 0.990) [21]. - The model demonstrated a successful rate of approximately 78% in identifying the most stable polymorphs among 7,490 polymorphic families, indicating its potential for high-throughput screening [30][32]. Group 4: Implications for the Industry - The integration of AI in MOFs research is reshaping the methodologies and innovation pace within materials science, moving from traditional experimental approaches to data-driven predictions [34][36]. - The development of structured knowledge graphs like MOF-ChemUnity aims to standardize naming conventions and enhance the accessibility of MOF-related data, further facilitating research and development in this field [35].
量化、宏观、CTA,到底选谁?
雪球· 2026-01-15 08:06
Core Viewpoint - The article discusses the increasing trend of quantitative macro strategies in the investment landscape, emphasizing their growth and effectiveness in asset allocation and risk management [9][10]. Group 1: Growth of Quantitative Macro Strategies - Over the past seven years, the global management scale of quantitative macro strategies has experienced explosive growth, surpassing 60% of the global macro strategy proportion in 2023, with this percentage continuing to rise [9]. - Quantitative macro strategies shift investment decision-making from narrative-driven approaches to rule-based execution through quantitative models, integrating both quantitative trading and macroeconomic logic [10]. Group 2: Addressing Criticisms of Quantitative Macro Strategies - Criticism regarding the rationality of macro strategies is addressed, highlighting that while traditional macro strategies rely on low-frequency economic data, quantitative macro can utilize a broader range of high-frequency data through advancements in machine learning and AI [12][13]. - The article counters the skepticism about the performance of quantitative macro strategies, asserting that many strategies have demonstrated significant excess returns, particularly in volatile market conditions, where they can quickly respond to market signals [16][18]. Group 3: Strategy Implementation - An example of a quantitative macro strategy is provided, which divides its approach into Beta and Alpha components, with the Beta portion using a risk parity model to allocate equal volatility weights to equity indices, government bonds, and commodity futures [15]. - The Alpha component employs machine learning models to identify short-term signals for timing and trading, enhancing returns without increasing overall portfolio risk [15][18]. Group 4: Risk Management and Leverage - Quantitative macro strategies are noted for their cautious approach to leverage, utilizing a more diversified trading portfolio and a programmatic risk control mechanism to monitor leverage usage in real-time [20]. - The article emphasizes that the flexibility in using leverage is a significant advantage of macro strategies, particularly when employing CTA methods to amplify returns during certain market conditions [18][20].
英国伦敦大学学院副校长Geraint Rees院士加入欧洲经济研究院
Sou Hu Cai Jing· 2026-01-14 15:03
欢迎英国医学科学院院士、英国皇家医师学会会士、英国伦敦大学学院副校长、伦敦大学学院认知神经 学教授、伦敦大学学院生命科学学院前院长、认知神经科学研究所前所长、谷歌DeepMind公司前高级 科学顾问Geraint Rees院士加入欧洲经济研究院! 项目和荣誉 Geraint Rees院士的研究兴趣是人类认知的本质和神经基础,特别是潜在的意识和相关现象;以及将高维 多变量推理("机器学习"和类似工具)应用于医疗保健提供、创新理解和其他领域的重大挑战。在伦敦 大学学院担任高级职务时,他与同事合作,作为亨廷顿舞蹈症治疗干预神经基础的联合研究员(与 Sarah Tabrizi 教授担任 PI),以及作为神经科学和医学高维推理的联合研究员(与 Parashkev Nachev 教 授)。 EUROPEAN ECONOMIC RESEARCH INSTITUTE 职业生涯 Geraint Rees 是伦敦大学学院的副教务长(研究、创新和全球参与),Geraint 负责为伦敦大学学院世界 领先的研究、知识交流和全球参与以及支持它的职能、服务和资源提供愿景和学术领导。 2014 年至 2022 年,Geraint Rees院 ...
Cast AI获10亿美元估值融资 推出统一GPU市场平台
Sou Hu Cai Jing· 2026-01-13 13:07
云原生环境优化专业公司Cast AI Group Inc.今日宣布,获得Pacific Alliance Ventures的新一轮融资,公司 估值突破10亿美元,并将推出统一的云GPU市场平台。 继去年由G2 Venture Partners和软银愿景基金2期领投的1.08亿美元C轮融资后,Cast AI的融资总额已超 过1.8亿美元。该公司未透露此次战略融资的具体金额。 Cast AI成立于2019年,专注于利用机器学习技术为Kubernetes云环境提供自动化性能优化,通过"合理 调配"资源、自动扩缩容和管理竞价实例来实现成本节约和安全保障。这使企业能够更高效地使用云基 础设施进行FinOps和云工程等操作。 Cast AI联合创始人兼首席执行官Yuri Frayman表示:"企业需要的不仅仅是更便宜的基础设施,而是能 够随着工作负载和约束条件变化而自动适应的基础设施。这正是我们的自动化智能体的设计目标,这笔 投资将帮助我们在全球范围内扩展这一能力。" 甲骨文公司作为首个主要云服务提供商加入Omni Compute,通过Cast AI为客户提供多余的GPU容量。 甲骨文云基础设施高级副总裁Karan Ba ...
OneSpaWorld(OSW) - 2026 FY - Earnings Call Transcript
2026-01-12 20:02
Financial Data and Key Metrics Changes - The company pre-announced preliminary fourth quarter results with a slight downtick in revenue guidance, attributed to weaker performance in November, but December rebounded strongly, leading to a positive outlook for the fourth quarter and 2026 [3][4] - Guest spend reached the highest level ever, with metrics indicating strong performance during the holiday season, particularly Christmas and New Year cruises [12] Business Line Data and Key Metrics Changes - The company is seeing significant growth in the acupuncture and med spa segments, which currently account for about 8% of total revenue and are growing at 10% annually [9] - Changes in revenue recognition in Europe will not impact EBITDA, as the company will now recognize management fees instead of direct revenue from certain cruise lines [6][7] Market Data and Key Metrics Changes - The company noted that pre-booking rates are around 22%, with a goal to increase this to 30%, as pre-booked guests tend to spend 35% more than those who book on board [18][22] - The company is piloting revenue enhancement features on 80 vessels, with plans to expand to 185 vessels by the end of the second quarter [24] Company Strategy and Development Direction - The company is focusing on enhancing its wellness offerings, particularly in the med spa segment, and is exploring options to integrate longevity services into its offerings [10][33] - The company aims to maintain a collaborative relationship with cruise line partners, focusing on growing the overall business rather than competing for smaller slices of revenue [41][42] Management's Comments on Operating Environment and Future Outlook - Management expressed confidence in consumer spending trends and the ability to maintain pricing power across different macro environments, despite some concerns about consumer nervousness [14][15] - The company is optimistic about the future, with a focus on improving operational efficiency through AI and machine learning initiatives [25][26] Other Important Information - The company returned $92.9 million to shareholders in 2025 through share repurchases and dividends while also investing in debt reduction [38] - Staff retention has improved significantly, with a retention rate of 76%, which is expected to enhance productivity and reduce training costs [31][32] Q&A Session Summary Question: Can you provide insights on the preliminary fourth quarter results? - Management noted a slight revenue guide downtick due to November's performance but highlighted a strong December, leading to a positive outlook for the fourth quarter and 2026 [3][4] Question: What are the implications of closing the Asia land-based operation? - The exit from the Asia land-based operation will impact revenue but not EBITDA, as it was not profitable [6] Question: How is the company addressing the growth in guest spend? - The company has reworked service offerings to encourage longer and higher-priced treatments, which has successfully driven guest spend [11] Question: What is the company's strategy regarding pre-booking? - The company aims to enhance pre-booking capabilities, as pre-booked guests tend to spend significantly more [22] Question: How does the company view its relationships with cruise line partners today? - The company emphasized a collaborative approach with cruise line partners to grow the overall business, contrasting with past competitive dynamics [41][42]
OneSpaWorld(OSW) - 2026 FY - Earnings Call Transcript
2026-01-12 20:02
Financial Data and Key Metrics Changes - The company pre-announced preliminary fourth quarter results with a slight downtick in revenue guidance, attributed to weaker performance in November, but December rebounded strongly, leading to a positive outlook for the fourth quarter and 2026 [4][3] - Retail spend increased significantly during the holiday season, with penetration around 11%, marking the best New Year cruises on record [4] Business Line Data and Key Metrics Changes - The company is reorganizing its operations, exiting the Asia land-based operation, which will impact revenue but not EBITDA, as it was not profitable [6][8] - In Europe, the company will now recognize management fees instead of revenue from certain cruise lines, maintaining EBITDA levels while simplifying operations [7] Market Data and Key Metrics Changes - The company noted a strong consumer appetite for wellness services, particularly in acupuncture and med spa offerings, which currently represent about 8% of total revenue and are growing at 10% annually [9][10] - Guest spend is at an all-time high, with December and New Year cruises performing exceptionally well [12] Company Strategy and Development Direction - The company is focusing on enhancing its wellness offerings and exploring new services related to longevity, which is seen as a significant growth area [10] - There is an emphasis on improving pre-booking capabilities, which currently stands at around 22%, with a goal to increase this to 30% [18][29] Management's Comments on Operating Environment and Future Outlook - Management expressed confidence in maintaining pricing power across various macro environments, noting that consumer spending has remained strong despite economic uncertainties [14][15] - The company is optimistic about the impact of tax breaks on consumer spending and overall business performance [15] Other Important Information - The company returned $92.9 million to shareholders in 2025 through share repurchases and dividends while also investing in debt reduction [38] - Staff retention has improved significantly, with a focus on bringing back experienced staff, which enhances productivity and reduces training costs [31][32] Q&A Session Summary Question: Can you provide insights on the preliminary fourth quarter results? - Management noted a slight revenue guide downtick due to November's performance but highlighted a strong December recovery [4] Question: What are the implications of exiting the Asia land-based operation? - The exit will not impact EBITDA as the operation was not profitable, but it will affect revenue numbers [6] Question: How is the company addressing the growth in guest spend? - The company has revamped service offerings to encourage longer and higher-priced treatments, which has successfully driven guest spend [11] Question: What is the company's strategy regarding pre-booking? - Management emphasized the importance of improving pre-booking capabilities, which significantly enhance guest spending [18][22] Question: How does the company view its relationships with cruise line partners today? - The focus has shifted to collaborative growth with cruise line partners, moving away from past practices of aggressive cost-cutting [40][42]