Prototype on the stream
Use server-side filters to sample high-premium or short-dated flow without drowning notebooks in every print.
Layered: flow × quant
Options Flow API for Quant Research
Design flow features on the live WebSocket, then prove them on the same field schema in historical SQL — avoiding train/serve skew between research and production.
Who this is for: quant researchers who need both a live options tape (~10M/day) and a SQL warehouse (2.8B+ trades from Feb 2025+) with consistent Greeks/IV/premium semantics.
Live feature → historical proof → production
Keep one definition of premium, delta bands, and DTE filters across stream and SQL.
Log feature vectors
Persist minute bars or event features (premium, |delta|, IV) with the same formulas you will run live.
Backtest in ClickHouse
Replay multi-week windows via historical SQL with row/timeout guardrails appropriate to research jobs.
Deploy unchanged thresholds
Promote filters to production WebSocket params and monitor drift against the SQL baseline.
Quant-specific flow concerns
- Train/serve consistency between SQL and WebSocket fields
- AGGREGATED vs RAW choice affects sweep statistics
- Trial SQL row caps — use full plan for serious backtests
- Chain snapshots for surface context at event time
- OPRA-licensed prints for professional research use under plan terms
- See also: options data API for quant hub
Example: validate a live premium/sentiment feature in SQL
-- Promote a live flow feature to historical validation
SELECT
toStartOfMinute(timestamp) AS minute,
sumIf(size * price * 100, sentiment = 'BULLISH') AS bull_prem,
sumIf(size * price * 100, sentiment = 'BEARISH') AS bear_prem,
avg(abs(delta)) AS avg_abs_delta
FROM RawOptionTrades
WHERE date = today() AND symbol = 'SPY' AND premium >= 50000
GROUP BY minute
ORDER BY minute;Quant stack links
Layer flow, quant hub, and UOA-for-quant paths.
Options flow for quants FAQ
Can I research only on historical SQL without the live feed?
Yes. Many studies start in SQL. The layered value is promoting the same features to WebSocket without redefining fields.
Does AGGREGATED mode change research results?
Yes. AGGREGATED merges same-contract same-instant prints; RAW keeps every exchange print. Pick one mode and keep it consistent from backtest to prod.
How far back does history go?
OptionData stores option trades from February 2025 onward — 2.8B+ rows at time of writing — queryable via ClickHouse SQL over REST.
Are Greeks model-based?
Greeks and IV are analytics estimates under model assumptions. Use them as features, not as guaranteed P&L.