Key points are not available for this paper at this time.
This paper examines risk-adjusted performance differences across AI-themed investment vehicles using monthly data from 2005 to 2025. We compare four equal-weighted portfolios: AI-powered ETFs, AI-focused thematic ETFs, AI/machine-learning (AIML) stocks, and a broad technology benchmark. We use factor models, GARCH volatility estimates, Sharpe ratios, quantile regressions, asset-level panels, ETF product-design controls, and non-AI benchmarks. The results show that AIML stocks deliver the strongest cumulative and abnormal returns but exhibit higher volatility, deeper drawdowns, and unstable exposure patterns. AI-powered ETFs show stronger risk-adjusted performance and a more efficient risk-conversion profile, reflected in lower betas, smoother volatility dynamics, milder downside deterioration, and higher return per unit of volatility. They also have a weaker association with lagged attention proxies, consistent with their defensive structure. Strategy-based tests show that the risk-efficiency advantage of AI-powered ETFs is most pronounced among volatility-targeting designs, while their non-volatility-targeting counterparts also compare favorably with AI-focused and non-AI defensive benchmarks.
Sovbetov et al. (Wed,) studied this question.