Comparative analysis of AI-driven versus human-managed equity funds across market trends

Abstract

Abstract The emergence of AI-driven funds has prompted inquiries regarding their efficacy in comparison with human-managed funds; however, empirical evaluations remain scarce. This study investigates their performance across distinct market conditions, a downtrend in 2022 and a subsequent recovery to an uptrend in 2023 and 2024, with the aim of determining which investment skills are more effective under varying economic cycles. Fund performance is evaluated using risk-adjusted return metrics such as Sharpe, Treynor, and Jensen’s alpha to assess returns relative to risk exposure. Additionally, t-tests on raw returns were conducted to statistically validate whether AI-driven or human-managed funds achieve significantly different returns under each market condition. Findings reveal that AI-driven funds outperform in downtrend markets, effectively mitigating downside risk, whereas human-managed funds achieve higher returns in recovery to uptrend periods, leveraging qualitative judgment to capture market momentum. The t-test results confirm significant performance differences, supporting the risk–return trade-off theory and behavioral finance principles. This study is limited to mutual funds that align with the investment strategies and asset allocation focus of AI-driven funds, specifically global equity funds with ESG and technology sector exposure. These findings may not be generalizable to other fund types, including those managed by human fund managers with different investment mandates. This study provides new empirical insights by comparing AI-driven and human-managed funds across different market conditions, using risk-adjusted performance metrics and statistical validations. It fills a gap in AI investment research by offering preliminary yet valuable evidence on how AI-based strategies perform relative to human fund managers’ discretionary decision-making in investment.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 15 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.