AI in Financial Research

Tim Dong

University of Tennessee

Types of AI uses

• Processing data
• Generating data
• Simulation

Processing data

• Earliest and most common use of AI
• Use AI to read and process (textual) data
• Classifying and quantifying sentiment
• Classifying and evaluate data
• I use ChatGPT to read proposals and:
• Classify whether the proposal is about loans
• Evaluate who is the primary beneficiary if the proposal passes

Generating (synthetic) data

• Use AI as respondents for survey
• Common criticism includes: reducing diversity and not realistic
• Professional services use actual human response to validate against AI responses:
• Crowdwave
• Fairgen

Simulation

• Use AI to simulate and replicate real market
• Can be anything from very basic AI to complex AIs with reinforcement learning
• Only seen in confirmatory research, e.g., AI indeed develop collusion strategy to game market price (AI-Powered Trading, Algorithmic Collusion, and Price Efficiency by Winston Wei Dou, Itay Goldstein, Yan Ji)
• I use agent-based simulation in my work to simulate different scenarios

Conclusion

• Using AI for data processing is getting more popular
• This will become standard soon
• Alpha is in the second and third type
• Using AI to generate hard-to-get data
• Simulate real market to test out theories