epinnock/polymarket-nba

Polymarket NBA

Every NBA market Polymarket has listed, from its public APIs: 3,673 events, 66,056 markets (game lines, player props, futures, awards), 82.9 million price points and 10 million trades. Game-by-game markets start with the April 2024 playoffs. Snapshot taken 2026-10-08.

Table	Rows	One row is
events	3,673	a game or a futures/awards event (2,917 are games)
markets	66,056	one market (moneyline, spread, total, player prop, futures outcome...) with its result
prices	82.9M	one price point for a market's first outcome token
trades	10.05M	one fill, from Polymarket's Data API

Coverage

Season	Markets	Volume
2023-24 (playoffs only)	146	$4.7M
2024-25	8,458	$3.44B
2025-26	56,779	$4.50B
2026-27 (preseason to snapshot)	589	$6.0M

Market types: totals 11.8k, player points 9.6k, rebounds 9.6k, assists 9.2k, spreads 9.0k, futures/other 6.3k, moneyline 3.2k, first-half totals/spreads/moneyline 7.0k, and a few smaller types. Spreads, totals and props begin in the 2025-26 season; earlier seasons have moneylines only.

Tables

events: event_id, slug, title, is_game, series_id, tags, start_date, end_date, closed, closed_time, volume, liquidity, open_interest, neg_risk, n_markets, description.

markets: condition_id (key), market_id, event_id, event_slug, event_title, slug, question, group_item_title, market_type, player (props), line, outcomes, token_ids, final_prices, winner, closed, resolution_status, game_start_time, season, created_at, start_date, end_date, closed_time, volume, volume_clob, liquidity, neg_risk, description, trades_coverage.

prices: condition_id, token_id, fidelity (hourly | minute), timestamp, price. Prices are for the market's first outcome (token_ids[0], outcomes[0]); the other side is ≈ 1 − price.

• hourly: the market's whole life.
• minute: game markets only, from 24 hours before tipoff to 4 hours after.

trades: transaction_hash, condition_id, token_id, outcome, outcome_index, side, price, size, timestamp, wallet (proxy wallet address).

prices and trades are split into one file per season.

Known limitations

• Trades are incomplete for the busiest markets. Polymarket's Data API returns only the most recent ~2,000 trades per market. markets.trades_coverage says which markets are complete (45,876), capped (1,302) or had no trades (18,878). The capped markets are the big ones (median volume $2.3M) and hold 73% of all dollar volume, so do not compute per-wallet profit or total flow from this table. Use on-chain data (Polygon OrderFilled events) for complete trade history.
• Prices are sampled points, not every trade. Hourly points can be up to an hour before tipoff.
• Season comes from the game's tipoff date, or from the market's start date for futures; seasons run August to July. 1,553 markets never traded and have no price history.
• Results are Polymarket's settlement (final_prices, winner) and were not checked against box scores.
• Two Polymarket series hold the game markets: series 2 (through 2024-25) and 10345 (2025-26 on). Everything tagged NBA (tag 745) is included, so events also covers futures, awards, draft and free-agency markets.

How it was built

build.py in this repository rebuilds the dataset from scratch (about 2.5 hours, ~30 GB of staging):

python build.py meta      # Gamma API: events + markets, paged in monthly windows
python build.py prices    # CLOB /prices-history, 14-day windows (the API maximum)
python build.py trades    # Data API /trades
python build.py export    # Parquet under data/

Sources: gamma-api.polymarket.com, clob.polymarket.com, data-api.polymarket.com. Not affiliated with Polymarket or the NBA.

Example

import duckdb
duckdb.sql("""
  SELECT m.event_title, m.game_start_time, m.outcomes[1] AS team, p.price AS pretip_price, m.winner
  FROM 'data/markets.parquet' m
  JOIN 'data/prices/*.parquet' p USING (condition_id)
  WHERE m.market_type = 'moneyline' AND p.fidelity = 'minute'
    AND p.timestamp <= m.game_start_time::TIMESTAMPTZ
  QUALIFY row_number() OVER (PARTITION BY m.condition_id ORDER BY p.timestamp DESC) = 1
  LIMIT 10
""")