[Feature]: Imputed GICS codes

[theisij]

Feature Description

@kellybry suggested in #55, that we add a backfilled GICS industry code.

Use Case / Motivation

GICS codes are current gold standard for industry codes, but they only start in late 1999.

Academic / Industry Justification (if applicable)

No response

Proposed Implementation (Optional)

Create a new column in the main output called gics_imp (imp for imputes).

Create the column in its own separate function.

Proposes implementation:

• if gics exist, use that
• if gics does not exist:

1. Create sic-to-gics mapping table. Specifically, take all non-missing pairs of sic and gics codes. For each sic code, find the most frequent gics code.
2. Impute the missing gics code of a firm based on its SIC code and the mapping table from 1.

@kellybry: let us know if prefer another approach

Willingness to Contribute

No, but I can provide detailed specifications

Alternatives Considered

No response

Additional Context

No response

Pre-submission Checklist

• I have searched existing issues to ensure this feature has not already been requested
• I have read the documentation to confirm this feature does not already exist

[kellybry]

GICS have multiple levels, with 11,25,74,163 groups depending on the level: https://www.msci.com/indexes/index-resources/gics

I do not recall of the top of my head how hard it is to maps these backwards. If easy to do all, we can do all. If easier to do just, say 11 and 25, that is fine too. @FernandoRDLL I'm happy to advise as you poke around a bit, let me know if you have questions.

[capellini]

We tried to do this as part of a V-Lab research project and did not have much luck, for a few reasons:

• SIC industries haven't been updated since the 1987 final release and thus many industries are outdated or missing
• SIC is a production-based classification system (what does the company make) whereas GICS is a market-based system (who buys the product and how it's used)

That said, perhaps we missed something when we did our V-Lab project. If you were to do this, as @kellybry suggested, I would focus on the higher-level SIC codes -- things may get very murky when you dive down into the more detailed codes. For example, a SIC code like 7370 (Computer Programming Services) could map to GICS Information Technology, Communication Services, or Financials depending on who the customers are.

However, I'm not sure if a SIC-to-GICS mapping would be any less ad-hoc than the FF industries. What if the GICS-based FF industry analysis started in 1999, when GICS started? Then a mapping from another industry classification system wouldn't be necessary. If we wanted to go back to 1997, NAICS may be a cleaner mapping, as it's more conceptually aligned with GICS.

[kellybry]

We also have a lot of data on SIC/GICS correspondence from CRSP/COMPUSTAT. So we could train a mapping. This is more involved, but would be a nice little project for a masters student.