Polymarket data for quantitative research

What makes the dataset useful for research

• Full archive, no retention cliff. Multi-month strategy validation, longitudinal studies, and event studies on dated events (elections, FOMC, sports playoffs, hurricanes) all need data older than 31 days. Resolved Markets keeps it.
• Microstructure-grade fields. Each snapshot has full bids and asks arrays as Array(Tuple(price, size)), plus best bid/ask, mid, spread, top-5 cumulative depth, sequence number, event timestamp, capture timestamp, and the paired crypto spot price with its staleness in milliseconds.
• Sampling is measurable, not guessed. The sequence_number on every snapshot counts the orderbook events we applied to that token's book, so the delta between two stored rows tells you exactly how many book updates the sample skipped — the median pair on BTC 5m spans ~28 events over 109ms. When you need what happened inside that window, the trade tape is captured on a separate, unthrottled stream.
• Cross-category coverage. Crypto, sports, economics, weather, and Hyperliquid perpetual futures through one key. Useful for cross-asset and cross-event studies that competitor APIs can't answer.
• ClickHouse backend. Aggregations across billions of rows return in seconds. Enterprise customers get direct query access; everyone else gets the same shape via REST + the rm-api download CLI, which writes CSV, JSON or JSONL locally and keeps an incremental SQLite cache so repeat pulls only fetch what is new.

Snapshot schema (what every record contains)

{
  "condition_id":     "0x...",            // Polymarket market id
  "token_id":         "1234...",          // UP or DOWN token
  "side":             "UP",               // UP | DOWN
  "event_timestamp":  "2026-05-04T14:23:00.123Z",   // Polymarket emitted
  "capture_timestamp":"2026-05-04T14:23:00.131Z",   // we processed
  "sequence_number":  82731,
  "best_bid":         0.6231,
  "best_ask":         0.6244,
  "mid":              0.62375,
  "spread":           0.0013,
  "bids":             [[0.6231, 412.5], [0.6225, 800.0], ...],
  "asks":             [[0.6244, 350.0], [0.6250, 612.0], ...],
  "depth5_bid":       4123.7,             // cum size, top 5 levels
  "depth5_ask":       3998.4,
  "spot_crypto_usd":  62418.50,           // Binance ref at capture
  "spot_crypto_age_ms": 84
}

Patterns for common quant tasks

Backfill a strategy in vectorbt or Backtrader

from rm_api import Client
import pandas as pd

c = Client(api_key=os.environ["RM_API_KEY"])
snaps = c.snapshots(condition_id, frm="2026-01-01", to="2026-04-01", limit=500)
df = pd.DataFrame(snaps).set_index("capture_timestamp")
# Now feed df["mid"] into vectorbt / Backtrader as a price series.

Bulk export via CLI

rm-api download \
  --category crypto --subcategory BTC \
  --from 2026-01-01 --to 2026-05-01 \
  --format csv --output ./btc_snapshots.csv

Detect gaps before computing realized spread

SELECT condition_id, token_id,
       max(sequence_number) - min(sequence_number) AS span,
       count() AS rows,
       (max(sequence_number) - min(sequence_number) + 1) - count() AS missing
FROM polymarket.snapshots_hf
WHERE capture_timestamp >= now() - INTERVAL 1 DAY
GROUP BY condition_id, token_id
HAVING missing > 0
ORDER BY missing DESC;

Academic and research access

University researchers with a .edu email can request extended Enterprise access in exchange for a citation in published work. Email [email protected] with a short description of the project. The dataset has already been used for studies on prediction-market liquidity, microstructure of binary outcomes, and cross-asset event studies.