README.md

---
license: other
language:
- en
task_categories:
- text-generation
- feature-extraction
pretty_name: Raw SEC Item 1A Risk Factors Text
size_categories:
- "1K<n<10K"
tags:
- finance
- sec-filings
- 10-k
- risk-factors
- item-1a
- edgar
- raw-text
- document-corpus
configs:
- config_name: default
  data_files:
  - split: train
    path: data/item1a_raw.parquet
---

# Raw SEC Item 1A Risk Factors Text

## Dataset Summary

This dataset is an English-language corpus of Item 1A Risk Factors sections
from SEC Form 10-K annual filings. Each row represents one curated
company-year filing. The corpus was created for research on temporal changes
in corporate risk disclosure, with particular interest in environmental risk
language.

The dataset contains raw disclosure text only. It does not include
environmental enforcement records, annotations, model outputs, labels or
scores.

## Supported Tasks

This is a document corpus, not a supervised benchmark. It can support document
representation learning, retrieval over risk-factor disclosures, topic
modeling, lexical analysis and temporal analysis of corporate risk language.

## Languages

The text is English (`en`) and comes from U.S. SEC filings.

## What Item 1A Means

SEC Form 10-K is the annual report that many U.S. public companies file with
the U.S. Securities and Exchange Commission. Item 1A is the Risk Factors
section of that filing. In this section, companies describe material
uncertainties that could affect their business, financial condition or future
results.

A risk factor is company-written disclosure about something that could go
wrong. It can cover regulation, litigation, commodity prices, climate exposure,
operations, supply chains, financing, cybersecurity, competition or broader
macroeconomic conditions. The text is useful for studying how firms describe
risk over time, but it is not evidence that a risk has already happened.

## Dataset Structure

### Data Instances

Each example contains a document identifier and the extracted Item 1A text.

```json
{
  "file_name": "AEE_2019.txt",
  "text": "<full Item 1A Risk Factors text>"
}
```

### Data Fields

The default config contains a two-column Parquet corpus:

| Field | Type | Description |
|---|---|---|
| `file_name` | string | Source raw file name in `TICKER_YEAR.txt` format. |
| `text` | string | Full raw Item 1A Risk Factors text. |

### Data Splits

| Split | Rows |
|---|---:|
| train | 1,088 |

The dataset has one split because it is a raw document corpus rather than a
train, validation and test benchmark.

## Dataset Statistics

| Field | Value |
|---|---:|
| Raw text rows | 1,088 |
| Companies identified from file names | 138 |
| Fiscal years identified from file names | 2015-2024 |
| Total raw words | 10,482,592 |
| Mean raw words per row | 9,635 |
| Total raw characters | 71,329,285 |

## Dataset Creation

### Source Data

| Source | Role in this dataset |
|---|---|
| SEC EDGAR Form 10-K filings | Original public annual reports filed by U.S. public companies. |
| Item 1A Risk Factors | The extracted section that contains company-written descriptions of material business risks. |
| `edgartools` | Retrieval and extraction helper used to access 10-K filings and resolve their Item 1A sections. |

The original language producers are public companies filing annual reports
with the U.S. Securities and Exchange Commission. The source records are
public Form 10-K filings available through SEC EDGAR.

### Curation Rationale

The corpus was built to support analysis of how corporate risk-factor language
changes over time. Keeping the release at the raw Item 1A document level makes
the source text usable for different NLP tasks without imposing predefined
labels or scores.

### Collection And Processing

Form 10-K annual filings were retrieved from SEC EDGAR. Item 1A sections were
resolved with `edgartools`, then retained as document-level observations when
the extracted risk-factor section was usable. The released text is not
whitespace-normalized, sentence-split, keyword filtered, labeled or scored.
Document identifiers preserve the `TICKER_YEAR.txt` naming convention used
during extraction.

### Annotations

The dataset contains no human or machine annotations.

## Loading

```python
from datasets import load_dataset

docs = load_dataset("MichaelDG/esg-commitment-verifiability", split="train")
first_doc = docs[0]
print(first_doc["file_name"])
print(first_doc["text"][:500])
```

## Intended Use And Limitations

Use this dataset for NLP research on SEC risk-factor language. It can support
document representation learning, retrieval, topic modeling and temporal text
analysis.

The corpus is not representative of all public companies or all SEC filings.
It covers curated Item 1A documents from identified company-years in the
2015-2024 fiscal-year window. Item 1A is broader than ESG disclosure.
Environmental language may be absent from some documents.

Risk-factor text is company-written disclosure about possible risks. It should
not be treated as evidence that a risk occurred, legal advice, investment
advice or an environmental performance score.

## Personal And Sensitive Information

The dataset consists of public company filings. It is not designed to identify
private individuals, although source filings may contain names or legal matter
details when companies include them in public disclosures.

## Licensing Information

The source texts come from public SEC filings. Users are responsible for
complying with SEC EDGAR access terms and any rights that may apply to the
source filing text.

## Citation Information

Campbell, J. L., Chen, H., Dhaliwal, D. S., Lu, H. and Steele, L. B. (2014).
The information content of mandatory risk factor disclosures in corporate
filings. *Review of Accounting Studies*, 19(1), 396-455.
[https://doi.org/10.1007/s11142-013-9258-3](https://doi.org/10.1007/s11142-013-9258-3)

Loughran, T. and McDonald, B. (2011). When is a liability not a liability?
Textual analysis, dictionaries and 10-Ks. *Journal of Finance*, 66(1), 35-65.
[https://doi.org/10.1111/j.1540-6261.2010.01625.x](https://doi.org/10.1111/j.1540-6261.2010.01625.x)