Menu
After hearing about the AI tool Perplexity from friends and about how it helped investing legend Stanley Druckenmiller find five companies in Argentina to invest in, I started trying it out and have found it very useful.
The big advantage I find with Perplexity is that hallucinations (or made up junk) appear to be reduced and data points are linked to their sources.
When I found that LinkedIn Premium members get one year of free access to Perplexity Pro, I jumped at the opportunity and started using the Pro version, which gives me access to ChatGPT-4.o and Claude 3.5 Sonnet from Anthropic.
At a recent GenAI meetup in the San Francisco Bay Area, a friend recommended that I try Claude instead of ChatGPT as my model and he was right.
Back in 2010, coming out of the Great Recession and recovering from the wounds of the deepest bear markets I had experienced, I was searching for answers about ways to improve my investing process. This led me to event-driven investing, which eventually took me down the path of creating InsideArbitrage and writing The Event-Driven Edge in Investing.
The other thing I did back then was create a multi-factor stock rating system that looked at twenty different factors ranging from revenue growth to balance sheet strength to come up with a score for each company. Different weights were assigned to each factor based on my personal investing experience up to that point. I eventually abandoned the model and looking back at it now with an additional 14 years of investing experience, I can see why I abandoned it. The old model suffered from weighting certain valuation metrics too heavily and not assigning enough weight to trends in revenue growth and profitability.
Earlier this week, I once again embarked on building a multi-factor stock rating system (without first peeking at that old model from 2010). Yes, we all know that there is no substitute for qualitative analysis. However I come across many new companies every single week as I look at insider transactions, new M&A activity, stock buyback announcements and spinoffs and wanted a quick filter to identify companies worth exploring.
This new rating system encapsulates my current manual process and in the coming weeks I will play with both the factors and their weighting as I run the model through dozens or hundreds of stocks.
Getting back to Perplexity, I was curious about how it would respond to a question about the kind of factors that should be incorporated into a stock rating system and quite frankly I was blown away. Just the initial question generated most of the factors I had in my new model and factors I intentionally decided not to use or substitute with a better alternative. I’ve included the response below.
An obvious follow-up question would be to ask Perplexity to come up with weights for these factors. Eventually, if we are so inclined, it would make sense to run a full scale machine learning project to test various permutations and combinations of weights against current as well as historical market data and see how the selected companies perform.
Considering I am bootstrapping InsideArbitrage for now and having working on a project like that in a past life, I know that I don’t have the time or financial resources to pull off a machine learning project like that. But this experience with Perplexity was exciting enough that I figured it was worth sharing with fellow investors and genAI explorers.