.png){kind=icon}
.png){kind=icon}
Return
Good portfolios from bad forecasts: The anatomy of LLM volatility estimates
.png){kind=icon}
OA
AI
A
DOI：10.1016/j.frl.2026.110572.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• No LLM among 30+ significantly beats log-HAR at forecasting realized variance. • The verdict survives sub-periods, prompt designs, history lengths, and protocols. • Capability buys calibration and responsiveness, but only matches the benchmark. • Reported LLM portfolio gains are a covariance-geometry tilt, not forecast skill. • The tilt is insignificant, regime-concentrated, and dies at realistic costs.
Keywords:
C53
C58
G11
G17
Large language models
Realized volatility
Volatility forecasting
Portfolio construction
Covariance structure
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
6.9
Papers:
8.9K
Citations:
2.8W
Organization
No organization information available
Cited Papers