Parsing SEC EDGAR: From Raw HTML to Structured Data

Parsing SEC EDGAR: From Raw HTML to Structured Data

Manual parsing of a 400-page 10-K takes an analyst half a day. Errors in extracting Revenue from tables—every fifth record. We built a system that pulls all key metrics, risks, and trends from PDFs into structured JSON in 5 minutes. And does it at scale—500 reports per hour. Unlike conventional parsers, our pipeline uses an LLM for adaptive extraction, handling reports with unstable markup. For instance, SEC EDGAR changes its HTML structure every quarter, but the model finds the needed sections by context.

Problems We Solve

Rate limiting and unstable structure. EDGAR allows no more than 10 requests per second, and HTML markup changes quarterly. Traditional BeautifulSoup parsers break at the first class-to-id rename. Our pipeline uses an LLM for adaptive extraction—the model determines where Revenue lies by context, even if the table is renamed.

Extracting unstructured sections. MD&A, Risk Factors—these are 20–30 pages of prose. Simply taking the text is not enough. We need to separate trends from facts, find numbers in natural language ("revenue increased by 12% to $5.2B"), and compute sentiment. We employ chain-of-thought prompts with few-shot examples from benchmark reports.

Year-over-year comparison. A risk that was third last year and first today signals trouble. The system automatically builds deltas: extracts Item 1A for current and prior year, computes semantic similarity of paragraphs, and marks what's new, intensified, or diminished. The output is a list of red flags with explanations.

How We Do It: Stack and Case Study

We use a hybrid architecture: Python + LangChain + OpenAI GPT-4 (or local vLLM with LLaMA 3). Vector storage—Qdrant for semantic search over historical reports. Deployment via Kubeflow on a GPU cluster (A10G).

from langchain.chains import create_extraction_chain
from langchain.chat_models import ChatOpenAI

llm = ChatOpenAI(model="gpt-4", temperature=0.1)
schema = {
    "properties": {
        "revenue": {"type": "number"},
        "net_income": {"type": "number"},
        "risk_factors_new": {"type": "array", "items": {"type": "string"}}
    }
}
chain = create_extraction_chain(llm, schema)
result = chain.run(filing_text)

Case study: parsed 5,000 annual reports overnight. A hedge fund client wanted to track retail rotation in 10-Ks. Our system processed all 5,000 documents in 8 hours, extracting the top 10 companies with a sharp rise in mentions of "inflation" and "supply chain." Manual verification showed 97% accuracy on risks. Comparison: three analysts would have spent a week on the same task. Our solution is 10x faster and more accurate. Contact us to discuss a similar project. Get a consultation on your tasks.

How the System Handles EDGAR's Rate Limit?

The SEC limits requests to 10 per second. We implemented adaptive throttling with a queue and retries on 429 errors. For bulk parsing, we use a distributed architecture on Ray—load is spread across instances, enabling up to 500 reports per hour without blocking.

Why Extraction Accuracy Reaches 98.5%

We combine XPath rules for tabular data and a fine-tuned LLM for textual sections. Cross-validation: numeric data is checked against prior periods, text via semantic search against benchmark reports. For financial figures, key metric accuracy is 98.5%.

Approach Comparison: Manual vs. Automated Parsing

Economies of scale: processing 1,000 reports per month cuts costs by 10x, recouping the investment in 2–3 months.

Work Stages

1. Analytics—discuss which reports, metrics, and signals are needed. Collect benchmark reports for cross-validation.
2. Design—select models (GPT-4 / LLaMA 3 / Mistral), design output schema and augmentation pipeline.
3. Implementation—build EDGAR adapter, parsers, LangChain chains, delta module, and dashboard.
4. Testing—run on 100 random reports, compare with manual parsing. Achieve recall >95% on key fields.
5. Deployment—deploy in your cluster or as SaaS. Set up monitoring (uptime, latency p99, report count).

Timeline and Cost

Timeline: 4 to 12 weeks depending on complexity (basic parsing vs. custom business rules). Cost is calculated individually—depends on report volume, required models, and need for a GPU cluster. On average, investment pays back in 2–3 months by reducing analyst time by 80%.

What's Included

Typical Mistakes and How to Avoid Them

• Ignoring XBRL—many 10-K filings include iXBRL tags. If you skip them, you lose accuracy. We use arelle to extract axis/concept.
• Overly aggressive rate limiting—IP block for a day. We set time.sleep(0.15) between requests and monitor the X-RateLimit-Remaining header.
• Not checking encoding—SEC stores documents in Windows-1252. Without conversion, UTF-8 breaks. We apply chardet and automatically convert.

Our experience—7+ years in NLP, 30+ projects in financial analytics. We provide a guarantee on extraction accuracy (98%+ by contract). Order a pilot parsing of 50 documents—result in 2 days. Get a consultation on your tasks. Learn more about the EDGAR format on Wikipedia.

NLP Development: Text Classification, NER, Embeddings, and Information Extraction

We often receive a task: process 50,000 support tickets — currently all manual. Dataset — 3,000 labeled examples, 12 categories, imbalance: one category occupies 40% of the sample, three at 1-2% each. Baseline accuracy — 78%. Sounds decent until you look at recall for rare classes: 0.31, 0.44, 0.28. These classes — complaints and churn threats — are most important to the business.

This is a typical NLP development project. The problem is not the algorithm but that accuracy is the wrong metric. Our experience across 30+ projects shows: we start by analyzing business metrics and only then choose the model.

Why accuracy is not the right metric for rare classes?

Accuracy ignores imbalance. If the "churn" class appears in 2% of cases, the model can predict "all good" and get 98% accuracy — but the business loses clients. Solution: F1 macro (averaged over all classes) or weighted F1. For NER — strict entity F1 (exact matches only). We guarantee: after choosing the correct metric, model quality becomes measurable and predictable.

Text Classification: From BERT to Distillation

BERT-like models are the standard for classification. ruBERT-base or ruBERT-large from DeepPavlov for Russian. multilingual-e5-large — for multiple languages in one pipeline. XLM-RoBERTa-large — a strong multilingual backbone.

Fine-tuning for classification: add a classification head on top of the [CLS] token, train for 3-5 epochs with lr=2e-5, weight decay=0.01. For imbalance — weighted CrossEntropyLoss or focal loss with gamma=2.0. Contact us — we will show a code snippet.

Imbalance case study. Dataset — 3,000 examples, imbalance 1:20. Solution: class_weight via sklearn + CrossEntropyLoss. Additionally — augmentation of rare classes via backtranslation (ru→en→ru through MarianMT). Recall for rare classes rose from 0.31 to 0.67 with a slight drop in accuracy (76%→74%). Full NLP development end-to-end took 3 weeks.

Distillation for production. BERT-large gives F1 0.89, but inference on CPU — 180ms. Distillation into DistilBERT or ruBERT-tiny2 reduces latency to 25ms with F1 0.84. Export to ONNX Runtime provides an additional 1.5-2x speedup. DistilBERT achieves 7x lower latency than BERT-large with only a 5% drop in macro F1 – a typical production trade-off.

How to choose between BERT and LLM for your task?

For most classification and extraction tasks, BERT-sized models offer the best trade-off between cost and performance. Shift to LLMs only when the task demands generation, complex reasoning, or zero-shot generalization.

NER: Named Entity Recognition

NER — extracting persons, organizations, locations, dates, amounts, document numbers. For general categories (PER, ORG, LOC), pre-trained models work well. For specialized ones (medical terms, legal concepts) — fine-tuning is needed.

Data annotation. The main cost of an NER project. For a quality model — 500-2,000 labeled sentences per entity type. Tools: Label Studio (open source) or Prodigy (by spaCy creators). IOB2 format — standard.

Architecture. Token classification on top of BERT: each token gets a label (B-PER, I-PER, O). spaCy 3.x with transformer pipeline — a convenient production choice.

Nested entities. Standard IOB models cannot handle nested entities (organization inside an address). For such tasks — span-based NER: SpanBERT or SpERT. More complex but correct.

Post-processing is mandatory. The model predicts tokens — normalized entities are needed. Date — dateparser. Amounts — regex + validation. Names — deduplication via rapidfuzz. Included in our standard delivery.

Sentiment Analysis and Opinion Mining

Binary classification positive/negative works out of the box with BERT. Complexity — aspect-based sentiment analysis (ABSA): "the restaurant has good food but terrible service." For ABSA: aspect extraction (NER) + sentiment per aspect. Joint models BERT-for-ABSA — quality on Russian data is lower due to dataset scarcity. RuSentiment, SentiRuEval — main resources.

For production with simple positive/negative/neutral: distil models are enough. Three classes, balanced dataset, 2,000+ examples — F1 macro 0.82-0.87 in 1-2 days.

Text Summarization

Extractive summarization (select sentences) — TextRank or BM25 without training. Fast, no hallucinations. Good for long documents.

Abstractive (generates new text) — seq2seq: mT5, mBART, FRED-T5, ruT5-large. For production via LLM API (GPT-4, Claude) — often the best cost/quality/speed trade-off.

Embeddings: Vector Representations of Text

Embeddings are the foundation of semantic search, deduplication, clustering, RAG. Quality critically affects downstream tasks.

Models. E5-large-v2, BGE-M3, multilingual-e5-large — strong multilingual embedders. sentence-transformers/paraphrase-multilingual-mpnet-base-v2 — fast option. For Russian: ru-en-RoSBERTa (Skoltech) performs well on semantic textual similarity.

Embedding quality evaluation uses the MTEB benchmark as standard. But top results on MTEB don't guarantee success on a domain dataset — we build domain-specific eval.

Fine-tuning embeddings. If standard models don't give the required Recall@k — contrastive learning on domain pairs with MultipleNegativesRankingLoss. How to perform this for domain data:

1. Collect 500–2,000 semantically similar pairs from your domain.
2. Apply MultipleNegativesRankingLoss with a batch size of 32–64.
3. Train for 1–3 epochs using AdamW (lr=2e-5).
4. Evaluate Recall@k on a held-out domain test set.

This approach yields a 5–15% improvement in Recall@k in practice.

Dimensionality and storage. E5-large: 1024 dim, float32 — 4KB per vector. For 10M documents — 40GB. Quantization int8 reduces to 10GB. FAISS IVF_PQ — more compact but with losses. Included in our deployment recommendations.

Information Extraction

Structured extraction is a frequent task. Examples: key contract terms, technical characteristics, dates and amounts from invoices.

1. Regex + rule-based. For INN, OGRN, amounts, dates — more reliable than neural networks. No data required.
2. NER + post-processing. For variable formats.
3. LLM with structured output. GPT‑4 / Claude with JSON schema — for complex documents. Cost: minimal per document. For 10k+ documents/day — we calculate the economics.

We guarantee a hybrid: regex/NER for typical fields + LLM for edge cases. Our guarantee is backed by years of production experience and more than 30 projects.

Work Stages

Prototype on existing data — 1-3 weeks. Production system with CI/CD — 1.5–2.5 months. Cost is calculated individually — get a consultation for a project estimate.

What's Included

• Model and pipeline architecture documentation
• Access to the model via REST API (FastAPI + ONNX)
• Client team training (2-hour webinar + Q&A)
• Accuracy guarantee on the agreed test set
• Months of post-delivery support (bug fixes, adaptation to new data)

Our Experience

Years of NLP projects from classification to RAG systems. The team includes ML engineers experienced with Hugging Face, spaCy, LangChain, MLOps. We use vLLM, Kubeflow, Weights & Biases — a production stack, not toys. Contact us to evaluate your NLP project within two days — request a free consultation on your text processing pipeline.