Roeterseilandcampus - Gebouw G, Straat: Nieuwe Achtergracht 129-B , Ruimte: S.04
This study compared human and GPT-4's ability to classify companies using Hamilton Helmer's 7 Powers framework. Six humans rated 15 well known companies , showing good inter-rater reliability, while GPT-4 analyzed 10-K filings for 53 companies. Human ratings strongly correlated with the performance metric Return on Invested Capital (ROIC) and also enhanced its prediction, validating the framework. However, GPT-4's classifications showed low temporal stability and no correlation with ROIC. This suggests either current AI limitations in complex analysis or insufficient information in 10-K filings.