ImpliedData/prediction-markets

Prediction Markets Dataset

Historical market metadata and 1-hour OHLCV candle data from two prediction market platforms: Polymarket and Manifold Markets. Collected and maintained by ImpliedData.

Files

File	Description	Rows
markets_metadata.parquet	Full market catalog — titles, categories, statuses, dates	~8.2M
polymarket_ohlcv_1h_sample.parquet	1h OHLCV, top 70 Polymarket markets by volume, last 6 months	~80K
manifold_ohlcv_1h_sample.parquet	1h OHLCV, top 60 Manifold markets by volume, last 6 months	~50K

Note: markets_metadata.parquet includes Kalshi market titles and categories (publicly listed on their website) for research purposes. No Kalshi price or trade data is included in this release.

Schema

markets_metadata.parquet

All ~8.2M markets collected across platforms. No price data — catalog only.

Column	Type	Description
market_id	string	Platform-native identifier
title	string	Market question text
platform	string	polymarket, kalshi, or manifold
category	string	Topic category (politics, crypto, sports, etc.)
status	string	open, resolved, closed, or cancelled
resolution	string	Outcome if resolved — YES/NO or numeric
opened	date	Date the market was created
closes	date	Resolution date, if known
trader_count	int	Unique traders (null for a portion of markets)

*_ohlcv_1h_sample.parquet

One row per market per hour.

Column	Type	Description
timestamp	datetime (UTC)	Candle open time
market_id	int	Internal market ID — join with metadata via platform
platform_id	int	1 = Polymarket, 3 = Manifold
open	float	Opening probability (0–1)
high	float	Highest probability in the hour
low	float	Lowest probability in the hour
close	float	Closing probability (0–1)
volume	float	Total traded volume in the hour
trade_count	int	Number of trades in the hour

Prices are implied probabilities, not dollar amounts. A value of 0.73 means the market was pricing the event at 73% probability at candle open.

Manifold: Uses play money, not real money. Prices reflect crowd probability estimates but the incentive structure differs from real-money markets. Volume is in Mana (MANA), not USD.

Platform coverage

Platform	Markets	Trade records	Date range
Polymarket	~120K	14.8M	Oct 2020 – present
Manifold	~111K	5.0M	Feb 2023 – present

Loading the data

import pandas as pd

# Market catalog
markets = pd.read_parquet("markets_metadata.parquet")
print(f"{len(markets):,} markets")
print(markets.groupby("platform")["status"].value_counts())

# Polymarket OHLCV candles
candles = pd.read_parquet("polymarket_ohlcv_1h_sample.parquet")
candles["timestamp"] = pd.to_datetime(candles["timestamp"])

# Most active market in the sample
top_id = candles["market_id"].value_counts().index[0]
m = candles[candles["market_id"] == top_id].sort_values("timestamp")
print(m[["timestamp", "close", "volume"]].tail(20))

# Load via HuggingFace datasets library
from datasets import load_dataset

ds = load_dataset("ImpliedData/prediction-markets", data_files="polymarket_ohlcv_1h_sample.parquet")

Known limitations

• OHLCV files cover ~130 markets total (top by trade volume). Low-volume markets are not in this release.
• OHLCV date range is limited to the past 6 months. Full histories are available in the ImpliedData paid dataset.
• trader_count is null for a significant share of markets.
• Manifold data before February 2023 is sparse.

Full dataset

The complete dataset — all platforms, full trade history, intraday (1m/5m/15m) candles, and continuous updates — is available at implieddata.com.

Citation

@dataset{implieddata2026,
  title     = {ImpliedData Prediction Markets Dataset},
  author    = {ImpliedData},
  year      = {2026},
  url       = {https://huggingface.co/datasets/ImpliedData/prediction-markets},
  note      = {Historical trade data from Polymarket and Manifold Markets}
}

License

Underlying market data originates from Polymarket and Manifold Markets and is subject to each platform's terms of service. The schema, documentation, and normalisation methodology are the work of ImpliedData (implieddata.com). This dataset is made available for research purposes only.