Why It Matters
Staying ahead of market-moving events requires access to timely, accurate information. A wealth of this insight is hidden within the free-form text of SEC filings. Traditionally, extracting this data meant either manually reading and annotating documents—a time-consuming, labor-intensive process—or using basic Natural Language Processing (NLP) techniques that often produced high error rates and limited insights.
Our solution changes the game. By leveraging a Large Language Model (LLM), we can now efficiently extract critical information from SEC data with greater accuracy and speed.
What We Focus On: SEC Form 8-K
SEC Form 8-K is a mandatory filing that companies must submit within four business days to report significant, or “material,” events. These events can include acquisitions, bankruptcies, and board member resignations. Each event is listed under specific items, providing some context, but to grasp the full details, you still need to read the actual text.
For example, if you want to identify companies that entered into new lease agreements last month, you’d look at filings under Item 1.01 (Entry into a Material Definitive Agreement). While EDGAR’s metadata indicates the form type and reported items, it doesn’t specify the exact nature of the agreements. Item 1.01 could include leases, mergers, or licensing agreements.
Our Solution
We developed a tool that goes beyond surface-level metadata. It reads the actual text of each item to identify specific events, like new lease agreements, and tags them accurately. Here’s how it works:
- Text Extraction: We extract the specific item text from each 8-K filing. The consistent structure of these documents allows us to apply simple NLP rules to isolate the relevant sections.
- Contextual Tagging: Once the text is extracted, we use item-specific LLM prompts to analyze the content. For example, when processing Item 1.01 or Item 1.02 (Termination of a Material Definitive Agreement), the model answers questions like, “Does this agreement involve a lease, merger, or licensing deal?” Based on these answers, we create detailed, context-specific tags.
Bringing It All Together
Our tool integrates EDGAR metadata (form and item data) to create an organized Item Table. Each item’s text is extracted using NLP techniques and then processed by an LLM to answer targeted yes-or-no questions. The results are tagged accordingly, making it easy to filter and identify filings related to specific events.
This approach significantly reduces manual effort, improves accuracy, and provides actionable insights from SEC data that were previously buried in complex documents.