NuclearPhynance Archive

Has anyone done any work on systematic / data driven methods to determine sector classifications, as opposed to relying on ICB/GICS/Bloomberg?



Sure i read a paper on this on SSRN, but having trouble finding it or anything like it

Do you mean clustering?

I've seen the WQ/Kakushadze paper on it, using clustering, and that's a great start (and would love more / similar ideas) , but was also wondering about assessment at a systematic look at fundamentals too - if anyone has even done?

I think the GICS methodology would be a good place to start. Pick earnings apart, and systematize it? From a fundamentals perspective, any company could have a curve-like GICS profile = {10: 50%, 15: 20%, 20: 1%, etc, etc, ..., 60: 0%} rather than one single number.

We had a graduate student rotating with us doing some work on this using unsupervised learning to determine clustering using return data. Some interesting results but ultimately a bit flawed (may have been implementation rather than ideaology).

I find it interesting that a lot of pretty prominent places seem to be deploying students and interns on pretty important topics and then scratch their head when nothing comes out of it.



In this case, this is a critical ingredient that goes into the holy mix. Putting an expert on the case can make all the difference in my experience.

Experts? We live in an age when "expert" is a swear word!

@mtsm: We are not in control of the grad program and they just rotate through if there is a good fit for their skill, they want to join the team and we accept mentoring them. The reason some of these kind of projects wind up with students or juniors is because the benefit is not clear up front so they get employed to do the initial exploratory checks for greenfield stuff with support from a senior researcher. The senior researcher will then generally have to do a lot of cleaning up and review to get concrete results and that's where the project is at currently but that same researcher also had multiple other projects with a more clear benefit that get prioritized over things like this.

also attempted to do systematic/data-driven methods that help to determine sector classification... in my case good results occured when the clustering methods were constructed in accordance with the methods to discern "good/bad" companies within the sectors - something very similar to things mentioned in this topic

[url=/Show%20Post.aspx?PostIDKey=187134]https://nuclearphynance.com/Show%20Post.aspx?PostIDKey=187134[/url]

deleted