Paper Type
Complete
Abstract
Can Generative AI democratize investment analysis? We collect 2,845 directional stock forecasts on Nasdaq 100 constituents generated by Gemini Deep Research across weekly, monthly, and quarterly horizons from May to October 2025. This prospective design eliminates the look-ahead bias that has confounded prior backtesting-based evaluations of large language models (LLMs). The model’s directional accuracy is statistically indistinguishable from chance at every horizon, and signal-based portfolios underperform the index. We find that LLM acts as a sentiment parrot: its outlook tracks source sentiment and visibility, neither of which predicts returns. We identify a democratization paradox: Generative AI democratizes the process of investment research but not its outcome, because the binding constraint on quality is the architecture of publicly available internet content rather than model capability.
Paper Number
1640
Recommended Citation
Pan, Qian; Cui, Zhenyu; and Yang, Zonghao, "Can Generative AI Truly Democratize Investment Analysis? An Evaluation of Gemini Deep Research" (2026). AMCIS 2026 Proceedings. 18.
https://aisel.aisnet.org/amcis2026/sig_dsa/sig_dsa/18
Can Generative AI Truly Democratize Investment Analysis? An Evaluation of Gemini Deep Research
Can Generative AI democratize investment analysis? We collect 2,845 directional stock forecasts on Nasdaq 100 constituents generated by Gemini Deep Research across weekly, monthly, and quarterly horizons from May to October 2025. This prospective design eliminates the look-ahead bias that has confounded prior backtesting-based evaluations of large language models (LLMs). The model’s directional accuracy is statistically indistinguishable from chance at every horizon, and signal-based portfolios underperform the index. We find that LLM acts as a sentiment parrot: its outlook tracks source sentiment and visibility, neither of which predicts returns. We identify a democratization paradox: Generative AI democratizes the process of investment research but not its outcome, because the binding constraint on quality is the architecture of publicly available internet content rather than model capability.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
SIG DSA