Prediction Market Price History — Kalshi & Polymarket Time Series for Backtesting

Historical price series for any Kalshi or Polymarket market, both exchanges normalized to one schema. OHLC candles where the exchange has them, tick prices where it does not, plus a per-market summary row so you can find the markets that actually moved before paying for their full series.

Find markets by keyword — you do not need to know Kalshi's ticker syntax.

The problem this solves

Backtesting a prediction-market strategy means answering "what was this trading at three hours before it resolved?" Getting there yourself means: learn that a Kalshi candlestick lives at /series/{series}/markets/{ticker}/candlesticks and that the series is the first dash-segment of the ticker; learn that Polymarket history is keyed by a CLOB token id you have to dig out of clobTokenIds, and that the Yes leg is not reliably index 0; then reconcile cents against dollars and OHLC against single prices.

That is an afternoon before you have written a line of strategy. This actor is that afternoon, already done.

Output

price_point

{
  "type": "price_point",
  "exchange": "kalshi",
  "market_id": "KXHIGHNY-26SEP07-B79.5",
  "question": "Will the maximum temperature be 79-80° on Sep 7, 2026?",
  "timestamp": 1788652800,
  "time_utc": "2026-09-06T00:00:00+00:00",
  "price": 0.47, "open": 0.42, "high": 0.51, "low": 0.40, "close": 0.47,
  "yes_bid": 0.46, "yes_ask": 0.48,
  "volume": 1200, "open_interest": 8800
}

Polymarket rows carry the same fields with price/close filled and OHLC null, so a single dataframe holds both exchanges without special-casing.

market_summary

One row per market, sorted by widest range first:

{
  "type": "market_summary",
  "market_id": "KXHIGHNY-26SEP07-B79.5",
  "points": 168, "first_price": 0.20, "last_price": 0.44,
  "min_price": 0.10, "max_price": 0.65, "range": 0.55, "change": 0.24,
  "first_time_utc": "2026-09-01T00:00:00+00:00",
  "last_time_utc": "2026-09-07T23:00:00+00:00"
}

Turn includePricePoints off and you get only these rows — a cheap scan to find which of 200 markets are worth pulling in full. Then re-run with points on for just those.

Example input

Everything the New York temperature series did last month, hourly:

{ "kalshiSeriesTicker": "KXHIGHNY", "status": "settled",
  "historyDays": 30, "granularityMinutes": 60, "maxMarkets": 50 }

Cheap scan first — summaries only, no per-tick billing:

{ "search": ["fed", "cpi", "rates"], "maxMarkets": 200,
  "includePricePoints": false, "includeSummary": true }

Exact markets, daily granularity, both exchanges:

{ "kalshiTickers": ["KXHIGHNY-26SEP07-B79.5"],
  "polymarketTokenIds": ["32338220190071351435772801779725302244575775216413325951443816017994629993401"],
  "granularityMinutes": 1440, "historyDays": 365 }

Pricing

Record	Price
price_point	$0.0003
market_summary	$0.002

A year of daily prices for 50 markets is about 18,000 points ≈ $5.50. A summary-only scan of 200 markets is $0.40.

Pairs with

• Kalshi & Polymarket Market Data + Cross-Exchange Price Gaps — live prices and arbitrage gaps.
• Weather Data for Prediction Markets — the NWS settlement values these temperature markets resolve against. Join price history to the actual outcome and you have a labelled dataset.

Notes and limits

• Kalshi has shipped both a nested (price.open) and a flat candlestick shape, and both cents and dollars. Everything is normalized to dollars in 0–1 whichever arrives.
• Polymarket returns a single price per timestamp, not OHLC.
• A market with no trading in the requested window returns no rows and is logged, not treated as an error.
• Polymarket's price API only accepts a window of about a week per request, so longer ranges are walked in week-sized chunks and stitched together for you. You can ask for a year and it just works.
• Kalshi has no title search and tens of thousands of open markets, so a keyword may not reach the ones you want. Targeting a series (kalshiSeriesTicker, e.g. KXHIGHNY) or exact tickers is the reliable path; the run warns you when a keyword finds nothing.
• Every cost is bounded by maxMarkets × maxPointsPerMarket, so a wide date range cannot surprise you.
• Discovery scanning is not billed — only the rows you receive.

Support

Open an issue on the actor page. Requests for other exchanges or extra fields are welcome.