Paper studies how processing old versus new news affects portfolio allocation
Get up to speed
Bolin Song and Surui Xiang have published quantitative finance research comparing two portfolio allocation approaches for handling stale news signals: retaining the signal and providing timeliness information to the portfolio strategy, versus filtering out stale news before portfolio selection. This study uses a fixed large language model to extract signals from SEC public disclosure filings, conducting backtests based on 2023-2025 historical data to compare net-of-fee returns of reinforcement learning strategies and large language model strategies that directly select stocks. Backtest results show that for large language model direct stock selection strategies, filtering out stale news achieves the highest average annual net return of 18.43% in scenarios where news is delivered with a four-week delay.
AI-generated from reports · 1 hr agoUpdated
Coverage timeline
Follow the coverage to understand different sides of the event.
- arXiv q-finWhen News Is Stale: Portfolio Allocation with Reinforcement Learning and Large Language Models
A quantitative finance study compares two portfolio allocation approaches for handling stale news signals: retaining signals and providing timeliness information to the portfolio strategy, versus filtering out stale news before portfolio selection. The study conducts backtests based on historical data from 2023 to 2025, using a fixed large language model to extract signals from SEC public disclosure filings, and compares the after-fee returns of reinforcement learning strategies and direct stock-picking LLM strategies.
Trend of this event
Not enough consecutive observations yet; trend is not drawn.