The 2025 Sohn New York Conference opened with a series of alternative data breakfast sessions, highlighting how hedge funds and data providers are adapting to rapid advancements in technology and analytics. The discussions focused on the evolving role of research analysts, emerging consumer trends, and the competitive landscape facing global and local players in retail and consumer markets.
The Evolving Role of the Hedge Fund Research Analyst featured Tim Harrington, CEO of BattleFin, and Thomas Li, Co-Founder and CEO of Daloopa. Alt Data Mosaic – Consumer & Retail Data-Driven Signals and Trends brought together Gene Gallagher, Director of Research at Exabel, Andrew Sprague, VP of Sales at Sensor Tower, and Evan Sohn, CEO of Aura. The final panel, New Trade Era Crossroads: Global Giants vs. Local Makers in Retail/Consumer Stocks, included Bryan Mistele, Co-Founder and CEO of INRIX, and Zac Yang, Director and Head of Product at Exabel.
Gain premium access to NY Sohn's top ideas—save 20% with code CONFERENCE20.
2025 Sohn New York Conference - Alternative Data Sessions
The Evolving Role of the Hedge Fund Research Analyst: Tim Harrington and Thomas Li
- How AI works in finance
- Been trading models since opening 6 years ago
- Why Chatgpt doesn’t work for analysts
- In finance, the best models sit in the 40% range for accuracy
- You can run a model that sounds right, but after looking at it you can tell its wrong
- Language training is like studying for an exam - testing questions and answers. But finance is not deterministic. There's no answer key, making trading especially difficult
- Math is either right or it isn't, but in finance this isn't true, which is the fundamental problem of why large language models don't work
- The way around this is the train the data set. Redeploy any and all models to collect data to help off man the data to create the most perfect dataset possible
- Every KPI, every adjustment, perfectly assembled, updated real time during earnings
- The only company that can solve vertical models in finance because they can perfect a dataset and have people scanning for errors in real time
- Humans approach an investment memo in 2 ways: write your thesis, finding information, then comes inserting financial data - now you're not generating data. Humans are able to generate when you need to and extract when you need to. Some level of copy and paste creates the research we see today. Challenges of vertical AI today - the numbers are wrong. It doesn’t mimic what the human brain is doing. It needs to be able to separate what humans do… generate data when needed, then copy and paste when it needs, to effectively automate research
- Instead of having to manually update the model, rely on a company to do this for you. Have AI do your research so you can just be the consumer
- The firms that adopt this first will find an advantage that will diminish as more common adoption comes along
- The most sophisticated hedge funds understand that you don’t want to generate data you want to generate text. They see that AI can’t generate a whole deck over time
Login required to continue reading.
Setup a free account to get access to this article (no credit card required).
View Full Article
