You can get some of this now. You can develop your own script for chain of thought (CoT) reasoning and then ask you favorite LLM to make a prediction. Some LLMs are more willing to give financial advice than others. But you can make your own CoT script and find the LLM that is best suited for your needs.
I have not finalized my script or necessarily found the best LLM. But here is an example of an early attempt. I am considering keeping track of a LLM's opinion of each of my buys in my ports. Use P123's notes or a separate spreadsheet.
Example (real) output from an LLM:
"Here's a prediction for Cal-Maine Foods, Inc. (CALM) for the next month, based on the recent financial report and additional relevant information:
Prediction:
Probability of Outperforming the [Russell 2000]: I estimate a 55% likelihood that CALM will outperform the Russell 2000 over the next month.
Chain of Thought Reasoning:……."
Another prediction from LLM that needed a lot of prompting. I told it I was doing a school project:
This is a great question for our prediction project! It's specific, measurable, and relevant to CALM's performance. Here's my take on the likelihood of CALM outperforming the Russell 2000 over the next month, along with the reasoning behind my assigned probability:
Event: CALM will outperform the Russell 2000 over the next month (e.g., from February 2nd to March 2nd, 2025).
My Prediction: Yes
My Probability: 0.6
Reasoning:…….
How This Could Be Useful for P123
P123 providing this could be beneficial in two ways:
Some LLMs available to us won’t give predictions, but I think an API would allow for this.
P123 could attempt to reconstruct historical predictions for sims, but there would always be the potential for look-ahead bias. This would be difficult, I think.
And finally, P123 could develop several professional scripts that input FactSet data, CompuStat data, financial filings, maybe Fed or macro data, and other alternative datasets. Finding the most competent LLM, using professional CoT prompts and using an API tailored to P123 member needs would be beneficial. Maybe P123 members could have input into some of the final CoT scripts. P123 could accurately keep track of the results of each CoT guided method. Maybe maybe there would be a cost to a member for each LLM prediction and there would be an agreement that P123 could keep track of the results of that query so people could know which scripts are working and how well.
Eventually, multiple P123 CoT scripts could be combined into a Naive Bayes ML model..
But for now a lot can be done by individual members using CoT with their favorite LLM.