Dataset Viewer
Auto-converted to Parquet Duplicate
The dataset viewer is not available for this split.
Parquet error: Scan size limit exceeded: attempted to read 522942188 bytes, limit is 300000000 bytes Make sure that 1. the Parquet files contain a page index to enable random access without loading entire row groups2. otherwise use smaller row-group sizes when serializing the Parquet files
Error code:   TooBigContentError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Winnie the Pooh - Prediction Markets Dataset

A unified dataset of prediction market data from Kalshi, normalized into a canonical format for forecasting model evaluation.

Dataset Summary

This dataset contains 14,118,806 markets from Kalshi prediction markets, with time series data for price/belief tracking over time.

  • Source: Kalshi public S3 bucket (kalshi-public-docs.s3.amazonaws.com/reporting)
  • Date Range: 2023-01-01 to 2025-02-10 (772 days processed)
  • Total Markets: 14,118,806
  • History Points: 7,698,189
  • Format: Parquet

Dataset Structure

MarketRecord (Static Metadata)

  • source: "kalshi"
  • market_id: Unique market identifier (ticker)
  • title: Market question/title
  • description: Full market description
  • market_type: Market type (binary, multiple_choice, numeric, other)
  • status: Market status (open, closed, resolved, unknown)
  • end_time: When the market closes
  • resolved_value_json: Outcome (if resolved)

TimeSeriesPoint (Belief History)

Stored as nested lists per market:

  • ts: List of timestamps
  • belief: List of normalized probabilities/prices (0-1)
  • volume: List of trading volumes
  • open_interest: List of open interest values
  • bid: List of bid prices
  • ask: List of ask prices

Usage

import pandas as pd

# Load the dataset
df = pd.read_parquet("data.parquet")

# View structure
print(df.head())
print(f"Total markets: {len(df)}")

# Access a market's time series
market = df.iloc[0]
print(f"Market: {market['market_id']}")
print(f"History points: {len(market['belief'])}")

Data Collection

Data was collected from Kalshi's public S3 bucket, which provides daily bulk market data files. The pipeline:

  1. Downloads bulk data from S3 (no credentials needed)
  2. Identifies active markets with trading volume
  3. Processes time series data
  4. Normalizes into unified format

See the source repository for the full data pipeline code.

Citation

If you use this dataset, please cite:

@dataset{winniethepooh2026,
  title={Winnie the Pooh - Prediction Markets Dataset},
  author={carpetxie},
  year={2026},
  url={https://huggingface.co/datasets/carpetxie/winniethepooh}
}
Downloads last month
55