πŸŸͺ The new math of new news

The BreakdownΒ·Β·4 min read
FinanceAI/MLTechnology
Share𝕏in

AI Summary

A new study titled 'The Inefficient Pricing of News' finds that markets are surprisingly slow to incorporate news into stock prices, with an LLM-based trading strategy using 'news shocks' (residual embeddings) generating ~30% annual returns by trading only once a month. The research builds on Alfred Cowles' 1932 finding that professional forecasters cannot beat the market, but reframes the inefficiency: instead of human skill being the problem, it's the market's failure to process publicly available information using modern NLP tools. The author predicts quantitative hedge funds will exploit this anomaly within nine months, closing the gap.

Key Facts

βœ“A new study finds an LLM-based trading strategy using residual news embeddings (4,096-dimension vectors) generates ~30% annual returns by trading only once per month.
βœ“Alfred Cowles' 1932 paper established markets were efficient because forecasters couldn't beat random chance β€” the new research reframes this: markets are inefficient because they fail to process publicly available news.
βœ“The author predicts quantitative hedge funds will close this news-pricing anomaly within nine months by adopting LLM-based residual embedding strategies.

Author Takes

BearishThe Breakdown

Market efficiency and LLM-based news trading

The persistence of news-pricing inefficiency feels as shocking as Cowles' 1932 result, and the author predicts quant funds will close the anomaly within nine months of adopting LLM embedding strategies.

NeutralThe Breakdown

Future informational efficiency of markets

Markets will eventually become informationally efficient through LLM adoption, but this won't prove the wisdom of crowds β€” only that technology has advanced beyond pencil and paper.

Contrarian Angle

Trading Stocks Monthly Using LLM News Embeddings

Researchers built a long-short portfolio trading exclusively on 'news shocks' β€” residual embeddings from LLM-processed news articles β€” that returns ~30% annually while only rebalancing once per month.

In an era of high-frequency trading and real-time data, simply reading the news once a month with an LLM outperforms nearly all professional strategies β€” contradicting the assumption that modern markets are informationally efficient.

Related topics

More from The Breakdown

πŸ“°Today⚑FeedπŸ“‘SignalsπŸ’°Capital