| --- |
| pretty_name: SEC-RPT-Panel |
| license: other |
| license_name: sec-public-domain |
| license_link: https://www.sec.gov/privacy#dissemination |
| language: |
| - en |
| task_categories: |
| - tabular-classification |
| - text-classification |
| tags: |
| - finance |
| - accounting |
| - sec-edgar |
| - related-party-transactions |
| - fraud-detection |
| size_categories: |
| - n<1K |
| configs: |
| - config_name: company_universe |
| data_files: company_universe.csv |
| - config_name: filing_manifest |
| data_files: filing_manifest.csv |
| - config_name: proxy_manifest |
| data_files: proxy_manifest.csv |
| - config_name: extraction_manifest |
| data_files: extraction_manifest.csv |
| - config_name: nlp_scores |
| data_files: nlp_scores.csv |
| - config_name: xbrl_financials |
| data_files: xbrl_financials.csv |
| - config_name: aaer_manifest |
| data_files: aaer_manifest.csv |
| - config_name: aaer_company_matches |
| data_files: aaer_company_matches.csv |
| - config_name: fraud_labels |
| data_files: fraud_labels.csv |
| - config_name: master_panel |
| data_files: master_panel.csv |
| --- |
| # SEC-RPT-Panel v2.0.0 |
|
|
| A firm-year research panel for studying related-party-transaction (RPT) fraud |
| risk, built only from official SEC sources: EDGAR 10-K filings, DEF 14A proxy |
| statements, XBRL company facts and Accounting and Auditing Enforcement Releases |
| (AAERs). Published as `codewithdark/SEC-RPT-Panel`. |
|
|
| Built with `sec_rpt_dataset` v2.0.0 (git `bdc3712`) on |
| 2026-10-07. |
|
|
| ## Tables |
|
|
| | File | Content | |
| |---|---| |
| | `company_universe.csv` | One row per company (CIK): SEC metadata, SIC, case role. | |
| | `filing_manifest.csv` | One row per annual report (10-K family) with fiscal year and `is_primary_filing`. | |
| | `proxy_manifest.csv` | DEF 14A proxy statements used for related-party text when Item 13 is incorporated by reference. | |
| | `extraction_manifest.csv` | Per-filing extraction status, word counts, `rpt_text_source`, Item 9A internal-control flags. | |
| | `nlp_scores.csv` | Readability and dictionary scores for Item 1, Item 13, the related-party disclosure and the RP note. | |
| | `xbrl_financials.csv` | Annual XBRL values as first reported (FY2009+). | |
| | `aaer_manifest.csv` | SEC Accounting and Auditing Enforcement Releases crawled from the official listing. | |
| | `aaer_company_matches.csv` | Automatic AAER-to-company name matches: candidates for manual review, not labels. | |
| | `fraud_labels.csv` | Company x fiscal-year labels (`fraud_aaer_label`, `event_time`, `misconduct_period`). | |
| | `master_panel.csv` | Everything joined: one row per annual report. | |
|
|
| All tables are keyed on the 10-digit `cik`. Firm-year tables use `fiscal_year` |
| (Compustat convention: fiscal years ending January-May belong to the previous |
| calendar year). `master_panel.csv` has one row per annual report; use |
| `is_primary_filing == 1` for one row per company-year (first original 10-K, |
| as-first-reported). |
|
|
| ## Labels (`fraud_labels.csv`) |
| |
| - `fraud_aaer_label`: 1 for a **case** firm's fiscal years inside the |
| assessment window (the five fiscal years before the first public revelation |
| of the misconduct), 0 for matched peers and for years outside the window. |
| Case firms, misconduct periods and revelation dates come from a hand-verified |
| cases file, not from automatic matching. |
| - `event_time`: fiscal year minus revelation year (peers inherit their matched |
| case's revelation year). Negative values are pre-revelation. |
| - `misconduct_period`: 1 inside the misconduct period named in the SEC order. |
| - `case_role`: `case` or `peer`; `case_id` / `matched_case_id` link them. |
| - `aaer_release_in_year`, `aaer_numbers`: **unverified** automatic name |
| matches of AAERs *released* in that calendar year. These are an |
| enforcement-date flag, not a misconduct label; review |
| `aaer_company_matches.csv` before using them. |
|
|
| Enforcement data capture detected and pursued misconduct only. For most |
| cases the window overlaps the misconduct period, so signals are |
| pre-*detection*, not pre-*misconduct*. |
|
|
| ## Related-party text (`rpt_text_source`) |
|
|
| | value | meaning | |
| |---|---| |
| | `10k_item13` | Item 13 body of the 10-K was substantive | |
| | `def14a` | Item 13 was incorporated by reference; the related-party section of the matching proxy statement was used | |
| | `none` | no related-party text of at least the minimum length was found | |
|
|
| Texts shorter than the minimum word count are treated as missing: their NLP |
| scores are blank and `*_text_available` is 0. |
|
|
| ## Raw files |
|
|
| ```text |
| raw/filings/<CIK>/<FY>/<ACCESSION>/{filing,item1,item9a,item13,rp_note,rpt_disclosure}.txt |
| raw/proxies/<CIK>/<ACCESSION>/{proxy,proxy_rpt}.txt |
| raw/xbrl/CIK<CIK>.json |
| raw/aaer/<YEAR>/<AAER_NO>.{pdf,txt} |
| ``` |
|
|
| ## Coverage (dataset_qa.json) |
| |
| | Metric | Value | |
| |---|---| |
| | `companies` | 24 | |
| | `dataset_version` | 2.0.0 | |
| | `extraction_errors` | 0 | |
| | `filings` | 384 | |
| | `fiscal_years` | 2004: 13, 2005: 14, 2006: 14, 2007: 14, 2008: 16, 2009: 16, 2010: 16, 2011: 16, 2012: 16, 2013: 16, 2014: 16, 2015: 17, 2016: 18, 2017: 23, 2018: 23, 2019: 23, 2020: 24, 2021: 24, 2022: 24, 2023: 21, 2024: 20 | |
| | `forms` | 10-K: 384 | |
| | `fraud_label_positive_rows` | 17 | |
| | `label_source` | verified_cases_file: 384 | |
| | `primary_filings` | 384 | |
| | `rpt_text_available` | 339 | |
| | `rpt_text_source` | 10k_item13: 32, def14a: 307, none: 45 | |
| | `share_item13_incorporated_by_reference` | 0.8307 | |
| | `share_item9a_available` | 0.9505 | |
| | `share_rp_note_found` | 0.2031 | |
| | `share_xbrl_total_assets` | 0.7891 | |
|
|
| ## Citation |
|
|
| If you use this dataset, cite the accompanying thesis / working paper and this |
| repository. SEC filings are public domain; this compilation is released for |
| research use. |
|
|
