This is one scary paper for anyone in academic finance, AI-Powered (Finance) Scholarship. It is scary not because of a specific conclusion but because an AI program can not only generate and test new investment signals but because it can also wrap it up in a nice working paper suitable for publication. These AI papers may be better written and presented than papers done by humans. Ouch, this is a game changer.
This paper destroys a lot of academic finance and is the ultimate data mining paper delivered in nice publishable form. Think about it. We have gone from the search for factors that form a factor zoo to now mining all indicators. The authors mine over 30,000 potential stock return predictors form accounting data and apply the "assaying anomalies" protocol to generate 96 signals that pass the protocol test and benchmarks these results against a large group of known anomalies.
We have moved from a world of p-hacking and atheoretical research which does not need careful thought and formation of hypotheses to a world where we can have AI run the research and write the paper. This will put academics out of work - sort of. They will have to reinvent who they are and what they do.
The LLM generates papers that given signals creative names, custom introductions to provide a justification for the signals, and incorporate citations to support the work. The authors provide a cautionary note that this is the ultimate form of HARKing - Hypothesizing After Results are Known.
No comments:
Post a Comment