Kalshi is the largest regulated event-contract exchange in the US, and its markets — CPI prints, Fed decisions, elections, weather — are exactly the kind of thing quants want to backtest. But anyone who has tried quickly hits the same wall: Kalshi's API tells you what the orderbook looks like right now, and nothing about what it looked like before.
Why historical orderbook data is different
Trade history is an accounting record — exchanges keep it and usually share it. The orderbook is different: it's ephemeral state. Every placement, amendment, and cancel overwrites what came before, and unless someone is recording those events as they stream, the history is simply gone.
That matters because the book, not the trade tape, is where most strategy questions live:
- What was the spread on the 3.4% CPI strike in the hour before the release?
- How deep was the book when that 50k-contract order swept it?
- Where would my limit order have sat in the queue, and would it have filled?
- How did liquidity migrate across strikes as new information arrived?
A backtest built on sampled prices answers none of these. A backtest built on the full event stream answers all of them.
What Kalshi's own API provides
Kalshi's trading API is good at what it's for — trading:
- Markets and events — full catalog with settlement details
- Trades — historical executions per market
- Candlesticks — OHLC price history
- Orderbook — current depth only
That last line is the gap. There is no endpoint for "the book at 8:29:59 AM on CPI day." If you didn't record it, you can't get it.
How Predexon captures it
Predexon runs always-on capture infrastructure against Kalshi, recording every orderbook event tick by tick since January 2026: placements, updates, cancels, and fills, at full depth, with millisecond timestamps. The stream is finalized daily (T-1, around 06:00 UTC) and stored in the same normalized schema we use for Polymarket — so a research pipeline written for one venue runs on the other unchanged.
The result is Kalshi historical orderbook data you can actually download:
- Raw ticks — the event stream itself, for exact book reconstruction
- Aggregated tiers — 100ms, 1s, 1m, 1h snapshots for lighter workloads
- Compressed Parquet — columnar files that load straight into pandas, Polars, DuckDB, or ClickHouse
- Quote-first pricing — see the exact size and cost of any slice for free before paying; $50 free credits to start
# Quote a Kalshi slice — free
curl "https://api.predexon.com/v2/data/ticks/quote?venue=kalshi&granularity=raw&from=2025-01-01&to=2025-03-31" \
-H "x-api-key: <your-api-key>"What people build with it
Event-release backtests. Model how the book behaves in the seconds around CPI, NFP, and FOMC — where liquidity pulls, how fast the touch moves, what actually would have filled.
Market-making research. Spread capture, quote aggressiveness, and adverse-selection analysis need the full event stream, not snapshots.
Cross-venue studies. Because the schema matches our Polymarket orderbook history, you can compare how a regulated CLOB and a crypto-native CLOB price the same real-world event.
Academic work. Regulated event markets are a young, under-studied asset class — and citable, millisecond-resolution book data is the foundation a microstructure paper needs.
Getting started
- Sign up at dashboard.predexon.com — free tier, no credit card
- Quote a slice via
GET /v2/data/ticks/quote(free, shows exact size and price) - Download Parquet via a signed URL, and re-download free for 24 hours
Full details in the docs, or explore everything included in tick history.