AGENDA

Lutnick Theater, Intrepid

DATA SCIENCE TRACK CONTENT

1:00PM 

Registration

Please join us at the Data Science Track at BattleFin N.Y. on Tuesday May 21st for a deep-dive into:

2:00PM - 3:00PM

The Integration of Structured and Unstructured Text/Data, and The Use of LLM's to Find Signal

Short term signals: Using news and social media during nontrading hours to explain subsequent price behaviour of stocks at the market opening. Building on the 2022 academic research paper from Gan, Alexeev and Yeung, the POC dives into the approaches traders can take to inform their decisions on selecting algos for specific trades on their blotter.

Long term signals: Using the sentiment from news and social media to develop a multi-factor model for alpha strategies with monthly investment horizons.  Based on the research jointly done by the StarMine and MarketPsych teams, the model considers scores spanning the company’s equity, business and management.

The session will be split into three 20 minute sections:

3:00PM - 3:15PM

Intros

3:15 - 3:45PM

Introducing LLM Efficiency Gain with S&P Data/Research on LLMs, Utilizing Snowflake AI for Vectorization and a Risk Chat App

How to safely introduce S&P data into your models
Create vector embedding of SEC filings natively in Snowflake
Prompt Engineering: Strategies for refining prompts to elicit desired outputs
Deploy Arctic and Mistral Large LLM's to enable natural language question and answer
Deliver results via a Streamlit chat user interface

COFFEE BREAK

LLM for Alpha Generation – Sentiment Extraction with Snowflake AI and S&P Research on Alpha Generation with LLM Driven Features

Merging traditional NLP techniques with LLMs to enhance efficiency 
Transforming textual data into numerical features through sentiment extractio

Question and Answer