Farming Lobsters Heat Up Fund Circle; Public Funds Undergo Deep Workflow Revolution

Securities Times Reporter Zhao Mengqiao and Pei Lirui

“We won’t be replaced by AI, but we will definitely be replaced by those who are skilled at using AI, especially in the field of public fund research and investment, where science and art intertwine, and rationality and emotion coexist,” said a public fund manager to Securities Times.

Recently, a craze for “raising lobsters” has swept from the tech industry into the financial sector. AI Agents, represented by OpenClaw, are gradually attracting the attention of public fund managers. Securities Times has learned that many fund companies are cautiously evaluating the application of this tool in fund research and investment. Some fund managers, especially quantitative fund managers, have already begun experimenting with OpenClaw for strategy development. AI is evolving from a super tool to an autonomous collaborator.

However, on the other side of the coin, the fund industry is also re-examining the impact of AI on traditional research and investment models. Whether it’s processing vast amounts of financial data, signal recognition in quantitative investing, or previously high-threshold research models, the public fund industry is experiencing a gentle yet profound workflow revolution. At the same time, new challenges such as human-machine replacement and data leaks are emerging.

“Lobster-Raising” Fever Spreads in the Fund Circle

“I initially thought of it as an intern that could help us backtest scripts and handle data, but over the past half month, I found it has strong autonomy. It can independently extract good factors from raw data around the clock, broadening our sources of Alpha, and its accuracy is very high—like having a senior fund assistant by your side 24/7,” described a quantitative fund manager in Shanghai to Securities Times.

Recently, open-source AI Agent projects like OpenClaw have become very popular, sparking a nationwide “lobster-raising” craze, especially in the complex and information-dense field of fund research and investment.

Cao Hongyuan, Chief Digital Officer of Bosera Fund, revealed that Bosera already has teams using OpenClaw on public clouds under compliant conditions, and is also exploring domestic software for internal security and compliance scenarios.

Additionally, E Fund has formed a dedicated team to conduct functional verification and technical exploration of OpenClaw in isolated network environments, though it has not yet entered production deployment. According to insiders from E Fund’s fintech division, application scenarios mainly focus on automated market information collection and analysis, and corporate data governance.

“OpenClaw, as an open-source, highly customizable AI agent, ignites a new wave of enthusiasm for AI applications in the public fund industry. Its significance goes far beyond the tool itself,” Cao Hongyuan said. “OpenClaw is mainly aimed at individual users. It is expected to greatly unleash each person’s innovative capacity. In application, individual initiative is crucial. Currently, the first movers are research personnel, as it provides them with a super digital assistant that helps release individual creativity and productivity.”

Yimin Fund believes that OpenClaw is not just an enhancement of existing tools but is gradually triggering a gentle yet profound workflow revolution in fund research and investment.

“The core of traditional research tools is passive response—people input commands, tools output results. The breakthrough with AI Agents like OpenClaw is proactive execution, capable of autonomously completing the closed loop of ‘information gathering—data organization—preliminary analysis—result feedback’ based on preset goals. For example, previously, researchers spent 1-2 days organizing public opinion and financial data for a certain industry. Now, with OpenClaw configured with relevant skill modules, it can automatically fetch, classify, and archive data 24/7. Researchers can then focus solely on data interpretation and logical validation. This is fundamentally a reconstruction of the research workflow, not just a simple efficiency boost,” Yimin Fund stated.

Wang Ying, Deputy Director of Quantitative Investment at CITIC Prudential, said that their quantitative team has already integrated AI technology into their daily research system. Currently, about 30% of their strategies rely on quant factors trained via machine learning, mainly applied in price-volume trading strategies.

“We found that trading signals identified by AI models yield better returns when executed on the same day compared to next day,” she explained. “The logic behind this is that AI is good at capturing short-term pulses driven by liquidity amplification. Intervening at these moments not only allows us to seize fleeting opportunities but also reduces trading costs due to ample liquidity. The entire process automates signal generation.”

Human-Machine Replacement or Coexistence

From large language models capable of acting as super brains to autonomous AI Agents that plan and execute independently, AI’s rapid evolution is directly impacting fundamental, repetitive tasks in traditional research—such as information collection, data organization, and report writing. For public funds and research professionals, does this mean AI will lead to a new wave of industrial revolution, turning researchers into “weavers” of the past?

A fund manager from South China said, “I see AI as a ‘mature intern’ or ‘research and investment newcomer.’ Tasks like data collection, cross-validation, and simple analysis are quite mature now. Once AI handles these basic tasks, research staff can devote more time and energy to things AI still can’t do.”

“AI and human capabilities in research are actually non-overlapping and even complementary,” said Wang Yue, a fund manager at Minsheng JiaYin. “A good researcher should constantly ask good questions. Their goal isn’t to get a certain answer but to continually question the current situation, asking ‘why’ to truly understand the core variables of an industry or company. A good AI is a tool that can provide good answers. While AI lacks strong reasoning and thinking abilities, it can give sharp, accurate responses to researchers’ questions, improving research efficiency.”

Veyu, a fund manager at HSBC Jintrust, also believes that AI cannot replace fund managers and researchers at this stage. AI can assist in processing vast amounts of historical data, identifying key points, and summarizing patterns—more like a research assistant. Based on this information, researchers or fund managers can rely on their long-term experience to make more accurate industry judgments and investment decisions.

“Many tasks, however, remain beyond AI’s reach. For example, on-site due diligence involves direct communication with company executives or management teams. An important purpose is to perceive their work state—this may sound subjective, but it is reflected in company performance indicators, often as early signals. Also, uncovering non-public information—AI can only organize and analyze existing materials, but under compliance, non-public information can have high analytical value,” the South China fund manager added.

From Breadth to Depth in Alpha Sources

While AI has become a powerful tool for research, the professional barriers of fund managers and research teams remain clear and even more pronounced.

“AI replaces low-value-added tasks, not the entire position,” said Yimin Fund. “It threatens those unwilling to adapt or with limited capabilities, not core research personnel.”

Yimin Fund states that in the AI era, information gaps in the market will gradually narrow because AI can quickly gather and analyze vast amounts of information. Almost all research institutions can access the same basic data and information through AI. Therefore, future excess returns will no longer come from who can access information faster but from who can interpret it more deeply, judge trends more accurately, and better control risks. Simply put, it’s not just about computing power but more about algorithms—this is the core source of individual alpha for fund managers and the research barriers of fund companies.

Wang Yue emphasized, “We value the depth of research thinking more than the breadth of information collection. Research personnel should ask the key questions, not just gather as much information as possible. Information is endless; finding the most critical variable and capturing it sharply is the real source of excess returns.”

Cao Hongyuan also said that AI can help improve information processing efficiency by automating multi-modal data analysis, prompting research personnel to shift toward higher-level skills like logical reasoning, industry insights, and cross-validation. This could evolve the research system into a human-AI collaborative network, changing the linear structure of ‘researcher pushes tickets—fund manager makes decisions.’ Researchers will work alongside AI in clue discovery, strategy development, and risk management.

“We won’t be replaced by AI, but we will definitely be replaced by those who are proficient in using AI,” said the South China fund manager. “At the company level, building research barriers still depends on creating an ecosystem suited to the company. To do that, you need good mechanisms, culture, talent, and tools. In the future, these tools will definitely be related to AI. Ultimately, in this environment, everyone plays to their strengths, drawing nourishment from shared information on a unified platform, forming a coordinated ecosystem.”

Embracing Efficiency While Vigilant of Risks

AI undoubtedly boosts research efficiency, but many public funds have deeply realized that AI is a double-edged sword—it can greatly improve productivity but also harbors risks that, if ignored, could lead to investment losses.

Yimin Fund’s Quantitative Finance Lab points out the need to be cautious of the “black box” risk of AI models, which is the most critical and concerning. Most AI models, especially deep learning ones, operate as “unexplainable boxes”—we only know the inputs and outputs, not how the model arrived at its results. This is the “black box problem.”

The lab believes this risk manifests mainly in two ways: first, in factor discovery being “pseudo-effective”—AI might identify factors that seem significant but are actually results of historical fitting, which may not generate returns in future markets and could even cause losses; second, in misleading decision suggestions—AI might base recommendations on flawed logic or biased data, leading to poor investment decisions if blindly followed. For example, AI might recommend buying a stock because its historical data performed well, ignoring deteriorating fundamentals, which could cause significant losses. Additionally, the “unexplainability” of AI models makes risk traceability difficult, complicating root cause analysis.

Wang Ying also remains cautious about the broad application of AI. She notes that markets are “adaptive,” and trading behaviors in A-shares are a constantly evolving process. “The historical data used to train models already contains the behaviors of all market participants. Once your model starts trading, its actions become new market data, influencing the market itself. It’s like a model influencing itself in a feedback loop,” she explained.

This feedback loop causes the factors derived from AI models, especially price-volume factors, to show volatile excess returns. She cited that some machine learning factors that performed well in 2023 have shown huge performance swings recently, due to this “profit and loss being from the same source.” “The biggest challenge is knowing when to stop,” Wang Ying admitted.

Wang Yue added that AI applications in fund research should also be cautious about sensitive information leaks, which are major hidden risks. “Therefore, we mainly use AI for reasoning dialogues and collecting public information, with strong privacy protection mechanisms and measures to prevent AI from accessing corporate confidential information,” she said.