wehub-resource-syncPrediction Markets
Event-outcome probability data from Kalshi (CFTC-regulated) and Polymarket (on-chain). Used as an alternative asset class alongside equities, crypto, and futures.
Datasets
| Venue | Frequency | Coverage | Loader |
|---|---|---|---|
| Kalshi | Daily OHLCV | 2021–present | load_kalshi |
| Polymarket | Daily OHLCV | 2020–present | load_polymarket |
Download
# Both venues (~2 minutes, no API key)
uv run python data/prediction_markets/download.py
# Individual
uv run python data/prediction_markets/download.py --provider kalshi
uv run python data/prediction_markets/download.py --provider polymarket
Coverage levers
The defaults live in config.yaml. Override on a per-run
basis without touching the file:
| Flag | Applies to | Effect |
|---|---|---|
--start-date YYYY-MM-DD |
both | Override start window |
--end-date YYYY-MM-DD |
both | Override end window |
--max-markets-per-category N |
Polymarket | Cap on markets retained per category |
--search-results-per-query N |
Polymarket | Search-result depth per query before the per-category cap |
--categories crypto,economics |
Polymarket | Replace the category list (each must have a search_terms entry) |
# Wider window, deeper search, more markets per bucket
uv run python data/prediction_markets/download.py \
--provider polymarket \
--start-date 2025-01-01 --end-date 2026-12-31 \
--max-markets-per-category 10 \
--search-results-per-query 25
The resolved values are printed at the top of every run. --dry-run
shows what would be downloaded without hitting the API.
Files Produced
$ML4T_DATA_PATH/prediction_markets/
├── kalshi_events.parquet # Event-level OHLCV
├── kalshi_events_profile.json # Schema + summary stats
├── polymarket_events.parquet
├── polymarket_events_profile.json
└── prediction_markets_metadata.parquet # Combined event metadata
Loading
from data import load_kalshi, load_polymarket
kalshi = load_kalshi(start_date="2025-10-01")
polymarket = load_polymarket(symbols=["WILL-BITCOIN-HIT-150K-BY-JUNE-2026:YES"])
Both loaders accept symbols, start_date, and end_date filters and
return canonical (timestamp, symbol, open, high, low, close, volume)
schema. Polymarket also surfaces a category column (provider label).
Consumers
- Chapter 2 — cross-asset EDA (demonstrates events-as-assets framing)
- Chapter 6 — prediction-market case study (funding-rate-style alpha on event resolution)
License / Cost / Size
- Kalshi: public REST endpoint; no API key required for historical candles. Kalshi's developer terms permit personal research use (https://kalshi.com/developers/terms). Cite "Kalshi" when publishing results.
- Polymarket: public Gamma/CLOB endpoints; no API key for historical candles. Polymarket's ToS permit read-only research use (https://polymarket.com/terms-of-service). Cite "Polymarket" when publishing.
- Disk: under 2 MB total for both venues (historical OHLCV only — no order-book snapshots).
- Runtime: ~2 minutes for a full refresh.
Notes
Kalshi is CFTC-regulated and cash-settled; Polymarket settles on-chain in stablecoin and is blocked from US users. The downloader does not authenticate as a trader — it only pulls historical candles.