Screening & Bulk Dataset
Cross-sectional screeners and full-universe bulk exports, from a ranked idea list to a whole-market data feed.
What data it carries
Two ways to work at universe scale: rank it, or pull all of it.
- Quant Factor Screener · filter the universe by factor z-scores, risk scores, index membership and momentum regime
- Ownership Screener · screen by ownership %, holder count, top-10 share and Herfindahl (HHI) concentration
- Bulk statements · income, balance-sheet and cash-flow for every company, cursor-paginated
- Async exports · submit a job and download CSV or JSON for pipeline ingestion (Institutional plans)
Best real-life usage
- Idea generation: run a factor or ownership screen to get a ranked candidate list, then deep-dive each name.
- Data pipeline / feature store: bulk-export the full statements universe nightly into a warehouse.
- Systematic strategy build: screen → rank → backtest on a consistent, survivorship-safe universe.
Combine with other datasets
Each analysis below pairs this dataset with another to answer a question neither can alone.
Screen to a shortlist, then pull the full profile and statements for every hit.
Screen to a shortlist, then pull the holder list and insider activity for every hit.
Screen on factors, then retrieve the complete factor and risk panel for each surviving name.
Screen within an industry, then benchmark the shortlist against that industry's medians.
Over MCP with AI agents
Connect an AI agent to the CSIMarket MCP server (mcp.csimarket.com) and it can call these endpoints as tools, no glue code. Ask in plain language; the agent picks the endpoints, fetches the data, and composes the answer.
- “Find the 20 cheapest high-quality companies with rising institutional ownership.”, the agent runs the factor and ownership screeners and returns a ranked list.
- Agent-driven ETL: schedule an agent to submit the nightly bulk export and load the file when it's ready.
Available for AI training
AI / LLM trainingThe Screening & Bulk Dataset is licensable for AI and machine-learning work: training or fine-tuning a model on it, generating embeddings, indexing it as a retrieval corpus for RAG, or building a derivative model whose weights carry learned representations of it.
It is a separate licence from an API or MCP subscription. A subscription covers querying the data and working with the answer. It does not cover training a model on the data or persisting it into a durable AI knowledge base. Four licence types do: model training (LLM and ML), derivative model, RAG and vector index, and enterprise unlimited.
Delivered for pipeline use: CSV in long or wide form, Parquet for ML pipelines, JSON and NDJSON, and API streaming for automated workloads.