申万金工因子观察第5期：OpenClaw能否实现零代码基础构建量化策略？

1. Report Industry Investment Rating The report does not mention the industry investment rating. 2. Core Viewpoints of the Report - AI development has significantly improved the efficiency of quantitative work, evolving through three stages, with OpenClaw potentially enabling zero - code construction of quantitative strategies [4][5][6]. - OpenClaw can perform tasks from data extraction to code writing and strategy execution, but its execution effect needs improvement [1]. - OpenClaw has achieved the construction of machine - learning strategies, realizing a certain degree of "equal rights" in quantitative implementation [1]. - Currently, OpenClaw still has a long way to go before it becomes "useful" and faces various problems [1]. 3. Summary According to the Directory 3.1 AI's Development Significantly Improves the Efficiency of Quantitative Work - First stage: Large - model data hallucinations made it difficult for AI to assist in quantitative research. In early 2025, the data hallucinations of large models prevented direct data processing in quantitative work [4]. - Second stage: AI Coding rapidly enhanced the efficiency of quantitative research. Although data hallucinations persisted, AI Coding was widely adopted, and quantitative researchers became code supervisors, significantly improving work efficiency [5]. - Third stage: OpenClaw might achieve zero - code construction of quantitative strategies. It can autonomously handle the quantitative work environment and data extraction, theoretically allowing strategy construction and backtesting based on researchers' ideas [6]. 3.2 OpenClaw's Deployment and Preparation Work - Deployment method comparison: Cloud vs. Local: OpenClaw requires core permissions of the deployed machine. Four deployment methods are compared, and for investors unfamiliar with system installation, using a cloud - server mirror + API to access large models is recommended [9][10]. - Data interface: It is both OpenClaw's greatest advantage and current pain point. OpenClaw can directly call data API interfaces, but commercial API interfaces are expensive, while free or low - cost third - party data interfaces have issues with data quality and reading speed [12]. - OpenClaw's environment preparation: After deployment on Tencent Cloud, OpenClaw can autonomously install commonly used Python libraries for quantitative processing and configure data source tokens, but there is a risk of token theft if non - official Skills are installed [13][16]. 3.3 OpenClaw's Attempt to Implement Quantitative Strategy Construction - Simple quantification: Strategy backtesting based on ideas: OpenClaw can quickly complete the backtesting of simple quantitative strategies. For example, it calculated strategies of buying after two consecutive daily limit - up or limit - down, providing clear information on strategy characteristics [21][27]. - Multi - factor quantitative stock selection within the CSI 500 Index: OpenClaw can complete the process of data extraction, factor construction, industry neutralization, and factor measurement in a multi - factor quantitative strategy. However, it may make low - level errors during the process, and the analysis of factors is better done on other platforms [30][57]. - Machine - learning strategy: OpenClaw can autonomously complete the development environment and software installation for a GRU machine - learning strategy, perform feature engineering, and generate a preliminary strategy, achieving a certain degree of "equal rights" in quantitative implementation [59][65]. 3.4 Current Shortcomings - Response mode: OpenClaw can only respond passively and cannot initiate contact actively. It requires constant prompting for progress updates [67]. - File sending: It has problems sending Excel files, often misunderstanding the difference between file and message sending, and the correction may not be long - lasting [70]. - Calculation errors: It makes low - level understanding errors in basic calculations, reducing work efficiency. The performance of DeepSeek in OpenClaw is weaker than that on the web version [74]. - Occasional malfunctions: It may suddenly stop responding, give irrelevant answers, or have serious data hallucinations [74].