The Stanford EDGAR Filings Dataset: Reconstructing U.S. Corporate and Financial Disclosures into Layout-Faithful and Token-Efficient Pretraining Data

2026-06-16 · Source: Artificial Intelligence · Field: Finance & Economics — Corporate Finance & Treasury, FinTech & Digital Financial Services, Capital Markets & Investment Management · Depth: Expert, quick

Summary

The Stanford EDGAR Filings Dataset (SEFD) is an open reconstruction of U.S. SEC filings, designed to address the scarcity of high-quality, long-context training data for large language models (LLMs). It transforms audited financial statements, risk disclosures, ownership reports, accounting notes, and market-moving event filings into layout-faithful MultiMarkdown. This corpus is token-efficient, model-ready, and exhibits less than 0.1% overlap with Common Crawl-derived corpora. SEFD-v1, an initial public snapshot, comprises 152 billion tokens, with a larger archive estimated at 550 billion tokens across 18.5 million filings. The dataset supports financial language modeling, evaluation, reasoning, forecasting, compliance, and document understanding. Additionally, SEFD introduces two benchmarks: EDGAR-Forecast for numerical forecasting and EDGAR-OCR for complex financial table transcription.

Key takeaway

For machine learning engineers and AI scientists developing LLMs for financial applications, the Stanford EDGAR Filings Dataset offers a critical, open-source solution to data scarcity. You can utilize SEFD-v1's 152 billion tokens to pretrain models on authentic, layout-faithful financial disclosures, significantly improving performance on tasks like forecasting and compliance. Consider integrating the EDGAR-Forecast and EDGAR-OCR benchmarks into your evaluation pipelines to rigorously test model capabilities.

Key insights

The Stanford EDGAR Filings Dataset provides a novel, open source of layout-faithful financial data for LLM pretraining and evaluation.

Principles

• High-quality long-context data is scarce.
• Layout-faithful reconstruction enhances utility.
• Open datasets can address proprietary data gaps.

Method

SEC filings are reconstructed into layout-faithful MultiMarkdown, making them token-efficient and model-ready for LLM pretraining and downstream tasks.

In practice

• Train LLMs on audited financial statements.
• Evaluate models on financial forecasting.
• Develop tools for compliance and document understanding.

Topics

• Stanford EDGAR Filings Dataset
• SEC Filings
• Large Language Models
• Financial Language Modeling
• Pretraining Data
• Financial Forecasting
• Document Understanding

Best for: Research Scientist, AI Engineer, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• Predicting Overnight Stock Changes Post-Earnings: Tomorrow’s Price Hidden in Today’s Language — Earnings call language, processed with domain-specific NLP, predicts overnight stock movements.
• Benchmarking Open-Source Layout Detection Models for Data Snapshot Extraction from Institutional Documents — Current open-source layout detection models fail to generalize for "data snapshot extraction" from complex institutional documents.
• Building SEC EDGAR Financial Analytics With CocoIndex and Apache Doris — CocoIndex and Apache Doris enable auditable, real-time hybrid search across diverse unstructured financial data.
• Fin-RATE: A Real-world Financial Analytics and Tracking Evaluation Benchmark for LLMs on SEC Filings — LLMs struggle with multi-document, multi-entity, and temporal financial reasoning, revealing critical gaps in current evaluation.
• FENCE: A Financial and Multimodal Jailbreak Detection Dataset — Multimodal jailbreaking poses significant risks to VLMs, especially in finance, necessitating specialized detection datasets.
• Build an LLM from Scratch 2: Working with text data — Text data preparation for LLMs involves tokenization, converting tokens to IDs, and generating contextual embeddings.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.