瑞士百达资管雷德玮： AI驱动量化投资进入2.0时代

Core Insights - AI-driven quantitative investment strategies are evolving, moving from traditional models that rely on a limited number of factors to more advanced models that can identify hundreds of high-frequency signals and non-linear relationships in data [1][5]. Group 1: Company Overview - Swiss Bank Asset Management, part of the Swiss Bank Group with a 220-year history, has an asset management scale of 711 billion Swiss Francs as of June 30, 2025 [2]. - The quantitative investment team led by David Wright manages $25 billion, with plans to expand AI quantitative strategies into emerging markets, including A-shares in China [2][3]. Group 2: Market Interest and Strategy - Global capital interest in China is recovering, with plans to include A-shares in AI quantitative strategies as the team develops a version for emerging markets [2][3]. - The current AI quantitative strategy products are primarily focused on developed markets, tracking the MSCI World Index, but there is a push to include A-shares in the future [2][3]. Group 3: AI Model Adaptability - AI models can adapt to the Chinese market, with backtesting showing that identified signal relationships are transferable to emerging markets, including China [3]. - The potential for excess returns in emerging markets is higher than in developed markets, although trading costs are also higher [3]. Group 4: AI Application in Stock Ratings - AI models can enhance the predictive power of stock ratings by incorporating various signals, such as the timing of earnings reports, to improve decision-making [4][5]. - Traditional quantitative methods typically use around 20 company-level signals, while Swiss Bank's AI strategy utilizes approximately 400 high-frequency signals [5]. Group 5: Differentiation in AI Strategies - Swiss Bank's AI quantitative strategy focuses on a holding period of about 20 days, contrasting with many competitors that prefer shorter holding periods [5][6]. - The strategy emphasizes factor neutrality, maintaining balanced exposure across various investment styles without overexposing to any single factor [5][6]. Group 6: Mitigating Overfitting Risks - The company employs several methods to mitigate overfitting risks in AI models, including using economically sound signals, integrating numerous simple models, and applying cross-validation techniques [6]. - The role of fund managers is evolving, shifting from model building to training AI models and conducting factor research, while still maintaining oversight of model outputs and portfolio construction [6][7].