证券研究报告 宏观 关税、财政不确定性扰动经济与市场 华泰研究 2025 年 6 月 01 日│中国内地 图说全球月报 研究员 易峘 evayi@htsc.com SAC No. S0570520100005 SFC No. AMH263 研究员 胡李鹏，PhD +(852) 3658 6000 hulipeng@htsc.com +(86) 10 6321 1166 研究员 陈玮 SAC No. S0570524030003 SFC No. BVH374 chenwei023580@htsc.com +(86) 10 6321 1166 联系人 赵文瑄 SAC No. S0570124030017 zhaowenxuan@htsc.com +(86) 10 6321 1166 然偏弱；黄金高位震荡。 增长：5 月关税缓和提振增长动能，但关税冲击或仍待显性化。美国经济动 能低位反弹。美国 5 月 Markit 制造业与服务业 PMI 均超预期扩张，推动综 合 PMI 上行至 52.5。不过，制造业价格与库存分项仍明显上行，显示关税对 价格影响仍在积存，关税短期缓和背景下企业存在抢进口行为。地方联储 PMI 也 ...

证券研究报告 宏观 中美关税降级，美经济动能低位反弹 华泰研究 图说美国宏观月报 | 2025 年 5 月 第二十一期 本篇为图说美国宏观月报第二十一期。5 月中美关税降级，部分美国调查指 标低位回升，就业市场仍有韧性；虽然美国经济衰退风险下降，但关税对增 长和价格的影响或逐步显现。关税缓和，美股情绪修复，但美国政策不确定 性、对美国财政风险的担忧等因素仍压制美元回升，推高美债期限溢价。全 月来看，美股显著回升，美元小幅走弱，美债收益率整体上行。往前看，关 注 5 月非农（6 月 6 日）、5 月 CPI（6 月 11 日）、6 月 FOMC（6 月 19 日）。 增长：关税缓和提振部分调查指标，企业投资意愿偏弱，地产数据改善 关税缓和提振美国 5 月部分调查指标，居民消费有所回落，企业投资意愿 延续下滑，地产需求有所改善。增长方面，5 月 Markit 制造业和服务业 PMI 均超预期扩张，推动综合 PMI 上行 1.7 至 52.5，制造业、服务业 PMI 均为 52.3，均高于预期和前值；企业预期、新订单、产出等分项均有明显改善， 但价格与企业库存大幅上行。一季度 GDP 增速上修 0.1pp 至-0. ...

证券研究报告 金工 AI 行业轮动模型看好石油石化、家电等 华泰研究 AI 主题指数轮动模型：今年以来超额 10.43%，下周推荐中证中药等 AI 主题指数轮动模型使用全频段量价融合因子对 133 个概念指数打分并构 建周频调仓策略，每周选取 10 个主题指数等权配置。模型 2018 年初回测 以来年化收益率为 16.03%，相对等权基准年化超额收益率为 13.10%，今 年以来超额 10.43%。截至 2025 年 5 月 30 日，模型未来一周将推荐持有中 证中药、国证油气、浙江国资、央企股东回报等指数。AI 概念指数轮动未 来一周将推荐持有银行精选、中药精选、万得微盘股、保险精选等指数。 今年以来 AI 量价因子多头超额 15.04%，中证 1000 增强超额 8.35% 截至 2025 年 5 月 30 日，全频段融合因子今年以来 TOP 层相对全 A 等权基 准的超额收益为 15.04%，自 2017 年初回测以来 TOP 层年化超额收益率 31.47%，5 日 RankIC 均值 0.116。根据全频段融合因子构建的 AI 中证 1000 增强组合本周超额收益为 0.05%，今年以来超额收益为 8 ...

证券研究报告 出口回补带动 PMI 边际改善 华泰研究 宏观 2025 年 5 月 31 日│中国内地 动态点评 研究员 易峘 evayi@htsc.com SFC No. AMH263 +(852) 3658 6000 研究员 吴宛忆 wuwanyi@htsc.com SAC No. S0570524090005 SFC No. BVN199 +(86) 10 6321 1166 联系人 王洺硕，CFA，PhD SAC No. S0570123070085 SFC No. BUP051 wangmingshuo@htsc.com +(86) 10 6321 1166 造业企业可能处于去库进程中。 免责声明和披露以及分析师声明是报告的一部分，请务必一起阅读。 2. 非制造业商务活动扩张速度边际放缓 期外需不确定性扰动犹存，由此，财政政策有望进一步加力、以夯实名义 增长回升的基础。外需方面，5 月 1-30 日，华泰出口需求日度指数（HDET） 同比录得 0%左右，显示关税降级后、出口景气度较 5 月前两周的低点持续 回升，港口集装箱吞吐量及美线集装箱运价指数亦持续上行，整体二季度出 口仍有望对经济形成一定支撑 ...

证券研究报告 研究员 易峘 evayi@htsc.com SAC No. S0570520100005 SFC No. AMH263 研究员 胡李鹏，PhD +(852) 3658 6000 hulipeng@htsc.com +(86) 10 6321 1166 宏观 关税、财政不确定性扰动经济与市场 华泰研究 2025 年 6 月 01 日│中国内地 图说全球月报 研究员 陈玮 SAC No. S0570524030003 SFC No. BVH374 chenwei023580@htsc.com +(86) 10 6321 1166 联系人 赵文瑄 SAC No. S0570124030017 zhaowenxuan@htsc.com +(86) 10 6321 1166 然偏弱；黄金高位震荡。 增长：5 月关税缓和提振增长动能，但关税冲击或仍待显性化。美国经济动 能低位反弹。美国 5 月 Markit 制造业与服务业 PMI 均超预期扩张，推动综 合 PMI 上行至 52.5。不过，制造业价格与库存分项仍明显上行，显示关税对 价格影响仍在积存，关税短期缓和背景下企业存在抢进口行为。地方联储 PMI 也 ...

证券研究报告 金工 GPT-Kline：MCoT 与技术分析 华泰研究 2025 年 5 月 30 日│中国内地 深度研究 人工智能 91：GPT-Kline：MCoT 与技术分析 本研究对多模态大模型的推理能力及其在投研中的应用进行了深入探索。多 模态思维链——MCoT 是近期多模态领域炙手可热的研究方向，其将大模型 的多模态与推理能力相结合，大大提升了大模型应对复杂多模态任务时的表 现。本研究尝试将 MCoT 应用于投研中 K 线技术分析场景，构建了智能化、 自动化的技术分析平台 GPT-Kline，实现了从 K 线绘图、走势分析、指标标 注到报告输出的技术分析全流程自动化，并构建了网页端方便直观的用户交 互入口。测试结果显示，GPT-Kline 能基于 K 线图像进行准确识别和深入推 理，技术分析全流程中能实现可靠的走势分析、精准的技术指标及形态标注、 逻辑连贯的技术分析报告输出。 多模态协同是大模型通往通用人工智能的必经之路，这就要求大模型不能仅 仅是大语言模型，而需成为多模态全能的通才。早期计算机视觉领域的蓬勃 发展为大模型带来了感知图片的能力，而近期思维链技巧的迅速崛起则为大 模型提供了缜密的逻辑 ...

Investment Rating - The report does not explicitly provide an investment rating for the A-share market but emphasizes the construction of a multi-dimensional timing framework to enhance investment decisions. Core Insights - The multi-dimensional timing framework developed from various indicators has shown a significant improvement in performance, with a Sharpe ratio of approximately 1 and an annualized return of about 22% compared to the 3.77% annualized return of the underlying assets during the same period [1][3][42]. Summary by Sections Single Indicator Testing: Valuation, Sentiment, Capital, and Technical - A series of indicators with timing capabilities and good complementarity were selected, including financing buy amounts, Bollinger Bands, individual stock price change ratios, options put-call ratios, implied volatility, and futures positions [2][11][16]. - These indicators provide insights into the current market state, but single indicators face issues such as excessive noise and inconsistent performance across different samples [2][42]. Multi-Dimensional Framework Construction: Layered Synthesis of Timing Signals - The report proposes a layered synthesis approach to combine multiple original signals, achieving a synergistic effect where 1+1>2 [3]. - The framework integrates valuation and sentiment indicators as left-side indicators, while capital and technical indicators form right-side indicators, balancing win rates and payoffs [3][39]. Evaluation and Analysis: Overfitting Risk and Generalization Ability - The multi-dimensional framework exhibits lower overfitting risk and higher applicability across major A-share indices, particularly benefiting small-cap stocks [4][39]. - The framework's adaptability allows investors to tailor it according to their specific needs, with valuation indicators being more suitable for large-cap stocks and capital and technical indicators being more effective for small-cap stocks [4][39].

Investment Rating - The report maintains a "Buy" rating for the company with a target price of 15.30 HKD [8][9]. Core Insights - The company reported a revenue of 16.98 billion USD for 4QFY25, representing a year-over-year increase of 22.8%, exceeding Bloomberg consensus expectations by 8.4% [1]. - Non-HKFRS net profit attributable to shareholders was 280 million USD, up 24.7% year-over-year, indicating business resilience despite tariff uncertainties [1]. - Revenue forecasts for FY26 and FY27 have been cautiously adjusted downwards by 5.1% and 3.1% to 75.44 billion USD and 81.69 billion USD, respectively, due to tariff impacts [1]. - The company is expected to maintain strong growth in high-margin businesses driven by the rollout of AI agent products [1]. Summary by Sections IDG (Intelligent Devices Group) - IDG revenue increased by 12.9% year-over-year, with a pre-tax profit margin of 6.8% [2]. - PC and smartphone shipments grew by 10.9% and 5.6% year-over-year, respectively, outpacing global shipment growth rates [2]. - Revenue from PCs rose by 13.2% year-over-year, driven by robust sales of gaming laptops and high-end workstations [2]. ISG (Infrastructure Solutions Group) - ISG revenue surged by 62.7% year-over-year, continuing its strong growth trajectory [3]. - The growth was primarily fueled by AI server demand, with "Neptune liquid cooling" revenue increasing by 68% year-over-year [3]. - FY26 revenue is projected to grow by 22.2% year-over-year to 17.74 billion USD [3]. SSG (Solutions and Services Group) - SSG achieved a pre-tax profit margin of 22.7%, the highest since FY22, indicating robust growth in AI solutions and services [4]. - AI solutions and "as-a-service" offerings accounted for 58% of SSG revenue, reflecting a 4 percentage point increase year-over-year [4]. - FY26 revenue is expected to grow by 10.0% year-over-year, driven by digital solutions and cloud services [4]. Financial Projections - The company anticipates FY26/27/28 net profits of 1.47 billion USD, 1.50 billion USD, and 1.53 billion USD, respectively, with year-over-year growth rates of 2.0%, 2.1%, and 2.1% [5]. - The projected EPS for FY26, FY27, and FY28 are 0.13 USD, 0.14 USD, and 0.15 USD, respectively [5]. - The report applies a 15x FY26 PE valuation, reflecting a premium over the average PE of comparable companies [5].

Investment Rating - The industry investment rating is "Overweight" [5] Core Views - The report highlights a positive outlook for the lithium battery supply chain, with a month-on-month increase in production driven by policies promoting electric vehicle (EV) demand and a favorable export environment for energy storage [1] - The domestic market for new energy vehicles (NEVs) is expected to grow significantly, with a year-on-year increase in retail sales and a notable rise in commercial vehicle electrification [2] - The energy storage market is supported by strong bidding volumes in China and continued growth in Europe and emerging markets, indicating robust future demand [3] Summary by Sections Lithium Battery Production - In June, battery production reached 107.7 GWh, a month-on-month increase of 2.9% - Positive trends in production for cathodes (133,000 tons, +9.1%), anodes (117,000 tons, +0.9%), separators (1.49 billion square meters, +4.3%), and electrolytes (76,000 tons, +6.6%) were noted [1] New Energy Vehicles - Domestic NEV retail sales reached 980,000 units in May, a year-on-year increase of 21.9% - Cumulative sales from January to May reached 4.306 million units, up 32.8% year-on-year - The penetration rate of new energy heavy trucks reached 23.0%, significantly exceeding previous expectations [2] Energy Storage Market - The bidding scale for energy storage in China reached 132.89 GWh from January to April, a year-on-year increase of 219.1% - The U.S. market saw a 105% year-on-year increase in large-scale storage installations, driven by favorable tariff expectations [3] Supply and Demand Dynamics - The report indicates tightening supply and demand dynamics, with expectations for improved capacity utilization and profitability for leading battery manufacturers [4] - Recommended companies include CATL, EVE Energy, and Sunwoda, which are expected to benefit from these trends [4]

Investment Rating - The report maintains a "Buy" rating for the companies mentioned, including Kingsoft Office, Foxit Software, Wangsu Science & Technology, and Dingjie Smart [10][13][15][17][18]. Core Insights - The R1-0528 model from DeepSeek has achieved a performance improvement of 15%-26% compared to its predecessor, with significant enhancements in mathematical and coding capabilities [1][3][4]. - The introduction of tool invocation in R1-0528 is a key development, enabling its use in complex agent processes, which could disrupt the current reliance on overseas models [1][4]. - The report highlights the potential for R1-0528 to drive advancements in the agent and MCP (Model Context Protocol) sectors, suggesting a positive outlook for related companies [5]. Summary by Sections Model Performance - R1-0528 shows a notable increase in reasoning depth and token usage, with an average of 23K tokens per question compared to 12K in the previous version [2]. - The model's hallucination rate has decreased by 45%-50%, addressing a significant issue present in the earlier version [3]. Tool Invocation - The support for tool invocation in R1-0528 is emphasized as a major highlight, allowing for broader applications in agent development [4]. - The model's performance in the Tau-Bench evaluation is comparable to OpenAI's models, although it still lags behind in some areas [4]. Recommended Companies - The report recommends investing in Kingsoft Office, Foxit Software, Wangsu Science & Technology, and Dingjie Smart due to their potential benefits from advancements in AI and agent technologies [5]. - Other companies in the industry chain are also highlighted, including Hehe Information and HanDe Information, which are positioned to benefit from the growing demand for agent solutions [5].