LLM-Augmented Financial Networks Lift Quant Sharpe Ratio to 0.82

iblanco_finance · x · 2026-08-05

An arXiv paper proposes a two-stage framework: constructing a firm-level network from 10-K filings and then using an LLM to filter out spurious edges that lack genuine economic connections.

In a backtest on S&P 500 constituents (2011–2019), LLM-based edge filtering improved the long-short Sharpe ratio from 0.74 to 0.82 and reduced the max drawdown from -10.5% to -7.9%. The author notes this modest improvement is exactly the right size for a sane factor enhancement, suggesting LLMs should be used to clean inputs rather than invent trading signals.

Original post →

More from Research

Research channel →