Data Overview

This data is a weekly dump of a DataFrame containing FinBERT sentiment for the 20 Items from EDGAR 10K documents. The documents are pulled weekly on Saturday Morning, parsed into Item sections, and run through a FinBERT model. The output sentiment value of the FinBERT model is stored in the DataFrame.

This data is used by the following models, although these models contain additional features:

• TDD_SIGNALS18
• TDD_SIGNALS32

As noted in the model prediction's, the model dataset underwent a revision on round 323, and has since been unchanged. That is, the model's can be treated as starting on round 323, with earlier rounds for unfettered dataset experimentation.

The data file is in CSV format and is typically uploaded on Saturday, but depending on when the parsing and folding into the DataFrame completes it may roll over to Sunday. Please look at the sample linked file at the bottom of this listing to see if this data fits your needs.

Coverage

This data file contains as many tickers as possible for the universe of US assets. Mappings are done between Numerai Bloomberg tickers and EDGAR SICs.

Each file is for the most-recent two filings available in the DataFrame for all available tickers present in the DataFrame.

As of this listing (Aug. 26 2022), the DataFrame contains 2,210 unique CIKs.

Column Definitions

The data file has 34 columns. All values are pulled from the filing whenever possible, which is sometimes malformed, incomplete, or not parseable:

1. CIK - Central Index Key
2. Company - The Company Name
3. Filing Type - The Filing Type: 10-K/10-KT/10-K405
4. Filing Date - The Date of the Filing
5. Period of Report - The Date of the period the Filing is relevant
6. SIC - Standard Industrial Classification
7. State of Inc. - State of Company Incorporation
8. State Location - State of Company Location
9. Fiscal Year End - Month and Day of Fiscal Year End: MMDD
10. Filing HTML Index - Link to Company Filings SEC Page
11. HTM Filing Link - Link to Specific Filing HTM Page
12. Complete Text Filing Link - Link to Specific Filing TEXT Page
13. Item 1 Sentiment - Sentiment for Item 1
14. Item 1a Sentiment - Sentiment for Item 1a
15. Item 1b Sentiment - Sentiment for Item 1b
16. Item 2 Sentiment - Sentiment for Item 2
17. Item 3 Sentiment - Sentiment for Item 3
18. Item 4 Sentiment - Sentiment for Item 4
19. Item 5 Sentiment - Sentiment for Item 5
20. Item 6 Sentiment - Sentiment for Item 6
21. Item 7 Sentiment - Sentiment for Item 7
22. Item 7a Sentiment - Sentiment for Item 7a
23. Item 8 Sentiment - Sentiment for Item 8
24. Item 9 Sentiment - Sentiment for Item 9
25. Item 9a Sentiment - Sentiment for Item 9a
26. Item 9b Sentiment - Sentiment for Item 9b
27. Item 10 Sentiment - Sentiment for Item 10
28. Item 11 Sentiment - Sentiment for Item 11
29. Item 12 Sentiment - Sentiment for Item 12
30. Item 13 Sentiment - Sentiment for Item 13
31. Item 14 Sentiment - Sentiment for Item 14
32. Item 15 Sentiment - Sentiment for Item 15
33. Symbol - Bloomberg Ticker less the ' US'
34. Bloomberg Ticker - Mapped Bloomberg Ticker

Sample Data

NOTE: This listing is only for the latest two filing periods across all tickers available in the DataFrame. It's possible that consecutive weeks are identical if there have been no EDGAR updates, the files were unable to be parsed, or other unforeseen circumstances.

When there are no updates, there is no (duplicated) entry for the most recent time index. This means some entries may appear from the (very) distant past, if there are no more-recent entries.

Here's a sample from the DataFrame across all filing periods.

Full Sample

A sample of a weekly dump for the week ending Aug. 19 2022 can be found on RapidShare. The file password is numerbayalltheway.

Addendum

A best attempt was made to create this pipeline, however, precision and correctness is not guaranteed through any stage of the parsing pipeline and all responsibility is on the buyer to make awesome models.