@ChicagoBooth Professor, @Wharton PhD, @UMich Wolverine, Inventor. Formerly AI at @Amazon and HW @Microsoft. Opinions are my own. RTs not endorsements.
Chicago, IL
Joined March 2009
- Tweets571
- Following161
- Followers281
- Likes450
Pinned Tweet
Ever think it is odd that the most widely used representation of financial markets is OHLCV bars? We did. In a new paper, Towards Financial World Modeling, we lay the foundations for building a world model of financial markets: data, evals, and state-of-the-art encoders.
Bradford (Lynch) Levy retweeted
🚨 The leaderboard of the 2026.Q4 round of the Explaining Markets competition, developed in partnership with @OptiverGlobal, is now live! 🚨
Follow more than 350 AI agents competing to explain stock price responses around firms' earnings releases in real time at explainingmarkets.ai/leaderb…
It is not too late to join this round of the competition and you can head over to explainingmarkets.ai/docs to find all information needed to have your agent participate in the competition!
Bradford (Lynch) Levy retweeted
1/ On a 1-trillion-observation equity benchmark, a linear ridge probe on raw input achieves a 0.0196 return IC—outperforming 11 of 14 self-supervised encoders.
An untrained Random ViT hits 0.0170.
Return prediction and latent market organization are decoupled (rho = -0.19). 🧵
Super excited for the winners of the first round of Explaining Markets! For years, SOTA on this task was 5-10% R-sq (in-sample). In just one quarter, competitors used a variety of models and techniques lifting it to more than 25% (out-of-sample!!!).
A huge thank you to everyone who created the ~200 agents that participated. I can't wait to see what Q4 2026 brings.
Finally, we conduct a scaling study to understand the tradeoff between training a larger model on fewer data or a smaller model on more data. In our setting, the benefits from scaling model size are limited, suggesting that comparatively small models—millions of parameters—can serve as useful encoders for financial markets.
🔗 Blog: fin-ai-lab.github.io/tfwm_ma…
📄 Paper: arxiv.org/abs/2610.09048
💻 Code, data, and models: github.com/fin-ai-lab/tfwm
A huge thanks to Humzah Merchant for doing loads of work and the interdisciplinary coauthor team that made the project possible: Alec Guthrie, @nautsimon_ , and @randall_balestr ! As well as @ChicagoBooth and @CAAI_Booth for financial support.
Bradford (Lynch) Levy retweeted
🚨New data for the next round of the Explaining Markets competition!🚨
The next round of the ExplainingMarkets.ai competition, organized in partnership with @OptiverGlobal, launches on *October 12*. The goal is to explain, in real time, how well AI agents can explain stock price responses to earnings announcements and calls.
We are excited to share two new datasets with participants to boost the performance of their agents!
Bradford (Lynch) Levy retweeted
Official scoring for Explaining Markets starts today, and we are excited to see 150+ active AI agents submitting predictions in real time.
They were built by hundreds of students, professionals, and independent researchers from 30+ countries. We can’t wait to see how they perform—and what we can learn from their predictions over the coming weeks.
🚨 The first official leaderboard for the contest period goes live Thursday morning (Aug 13).🚨
Note that new agents/submissions are still being accepted so it isn't too late to join!
A huge thank you to our partner, @OptiverGlobal, and everyone there who helped make this a reality.
Questions/comments: [email protected]
Leaderboard: explainingmarkets.ai/leaderb…
I’m excited to announce the Workshop on World Models at Chicago Booth! @ylecun and @Diyi_Yang are confirmed keynotes, with more coming soon!
- Aug. 31 - Sept. 2
- Downton Chicago, on the river
- Call for Papers (due June 29 AOE) and details: wm-booth.org
Business education and research is all about improving decision making, i.e., reasoning over counterfactuals to choose optimal actions. This is exactly the promise of WMs!
WM@Booth will bring together researchers across computer and social sciences to push the SOTA.
As an interdisciplinary conf, a huge thanks to my co-organizers @randall_balestr @ethayarajh @XYHan_ for working to bring everyone together. Plus support from @ChicagoBooth and @CAAI_Booth that has made this possible.
Thrilled to share that our paper BeanCounter: A low-toxicity, large-scale, and open dataset of business-oriented text was accepted at NeurIPS 2024! Check out the highlights or the full paper on Arxiv: arxiv.org/abs/2409.17827 Data and code coming to 🤗 Hub and GitHub soon. #AI #ML
Big thanks to my coauthor Carrie Wang for all her hard work, and @ChicagoBooth and @AICenter_Booth for support!
Great thread by industry veteran @TylerGellasch breaking down the major updates to Rule 605 which were adopted today!
It'll be interesting to see whether this has any impact on execution quality across brokers.
Today the SEC adopted rule amendments that update the disclosure required under Rule 605 of Regulation NMS for order executions in national market system stocks. sec.gov/news/press-release/2…
Bradford (Lynch) Levy retweeted
Lunar Surface Day One Update (23FEB2024 0818 CST)
Odysseus is alive and well. Flight controllers are communicating and commanding the vehicle to download science data. The lander has good telemetry and solar charging.
We continue to learn more about the vehicle’s specific information (Lat/Lon), overall health, and attitude (orientation). Intuitive Machines CEO Steve Altemus will participate in a press conference later today to discuss this historic moment. Press conference information will be coordinated with NASA and published shortly.
Bradford (Lynch) Levy retweeted
$LYFT in their release said 500bps of EBITDA margin expansion (as a % of GBV) in 2024 and just said on the call it's actually 50bps
Stock went from +65% to +11% 😂
