GICS issue, not point-in-time?

Report a problem or error
1

I recently encountered an issue with my stock recommendation model that I hope you can help me understand. My model, which uses a universe that specifically excludes stocks from the financial sector (rule: 'GICS(40)=false'), recommended purchasing the stock with ticker BTMD on March 20, 2024. Surprisingly, the model later advised selling the stock as its ranking changed to 'N/A'.

Upon further investigation with the screener, I discovered that BTMD changed its sector from 'Health Care' to 'Financials' on August 5, 2023. This classification change occurred well before my model issued the buy signal.

This scenario raises a couple of concerns:

  1. Data Accuracy: Could this be a data issue?
  2. GICS Data not Point-in-Time: Is it possible that GICS classification data is not point-in-time, leading to discrepancies in historical data accuracy?

I would greatly appreciate any insights or similar experiences you might share regarding the handling of GICS data and ensuring accuracy in sector-based filtering in models.

Thank you for your help!

2

BTMD should not have been changed to Financials. It should still be 35151015 HCAREADMIN, effective 2022-08-23. I confirmed that with Yahoo and Fidelity. We are looking into the reason that it was changed but it seems to be a bug where the classification of a security that was previously linked to BTMD changed and our code mistakenly changed the classification of BTMD. So it appears to be a bug and not a data issue. We do not have a change history for the table that stores the classifications, so we don't know the date that FactSet sent the record which changed BTMD to Financials. But once the bug is fixed, that case would not re-occur.

FYI - we recommend that all Factset users switch to using FactSet RBICS codes instead of GICS.

1 Like
3

Thank you, Dan, for investigating the issue with the classification change for BTMD. It's reassuring to know the change to 'Financials' was probably a bug and not an error in data accuracy. Based on your recommendation, I will use FactSet RBICS codes instead of GICS to hopefully avoid similar issues in the future.

As there is no change history, would it be a crazy idea to consider implementing some kind of logging mechanism to track such changes?

Thank you for your support.