kalshi-trades / README.md
TrevorJS's picture
Upload README.md with huggingface_hub
5c622ba verified
metadata
license: cc-by-4.0
task_categories:
  - tabular-classification
  - time-series-forecasting
tags:
  - prediction-markets
  - kalshi
  - finance
  - trading
  - event-contracts
size_categories:
  - 100M<n<1B

Kalshi Prediction Market Trades & Markets

Complete public trade-level and market-level data from Kalshi, the first CFTC-regulated prediction market exchange in the United States.

Dataset Summary

Count Timespan
Trades 154,505,005 June 2021 – January 2026
Markets 17,464,713 June 2021 – January 2026
Events 4,721,915
Total contracts traded 34.6 billion

Data Files

trades/

16 parquet shards (~10M rows each), sorted by created_time.

Column Type Description
trade_id string Unique trade identifier
ticker string Market ticker (joins to markets.ticker)
count int64 Number of contracts in the trade
yes_price int64 Price of YES contracts (1–99 cents)
no_price int64 Price of NO contracts (1–99 cents)
taker_side string Side the taker bought: yes or no
created_time timestamp[us, UTC] When the trade occurred

markets/

4 parquet shards (~5M rows each).

Column Type Description
ticker string Unique market ticker
event_ticker string Parent event ticker (groups related markets)
market_type string Market type (e.g. binary)
title string Human-readable market title
yes_sub_title string Label for the YES side
no_sub_title string Label for the NO side
status string Market status: settled, closed, active, etc.
yes_bid int64 Best YES bid at time of fetch
yes_ask int64 Best YES ask at time of fetch
no_bid int64 Best NO bid at time of fetch
no_ask int64 Best NO ask at time of fetch
last_price int64 Last traded price
volume int64 Lifetime volume (contracts)
volume_24h int64 24-hour volume
open_interest int64 Open interest
result string Resolution: yes, no, or empty if unresolved
created_time timestamp When the market was created
open_time timestamp When the market opened for trading
close_time timestamp When the market closed

Usage

from datasets import load_dataset

ds = load_dataset("TrevorJS/kalshi-trades")

# or with DuckDB
import duckdb
con = duckdb.connect()
con.execute("SELECT * FROM 'hf://datasets/TrevorJS/kalshi-trades/trades/*.parquet' LIMIT 10")

Data Quality

  • 0 duplicate trades (21,950 duplicates removed during cleaning)
  • 0 null trade IDs or prices
  • 100% trade-to-market join coverage (64 orphan tickers with no API-available market record were removed; 1,258 trades dropped, <0.001% of dataset)
  • All timestamps are UTC

Provenance

Trade data was collected from the Kalshi public API. Market metadata was backfilled using the /markets batch ticker endpoint. The dataset combines data originally collected by Andrew Becker (June 2021 – November 2025) with an incremental update through January 2026.

Acknowledgments

  • Andrew Becker for the original kalshi_trades dataset and collection infrastructure
  • Kalshi for providing a public API

License

CC-BY-4.0