Historical Options Flow Replay: Trade Tape, Minute Volume & the OI Simulator

The Historical API serves two kinds of data. The minute quote-and-greeks store carries bid, ask, sizes, IV and greeks for every contract, and it is what the exposure and summary endpoints replay from. Options volume lives somewhere else entirely: in the raw trade tape - 4.66 billion real prints since January 2017 - which has its own set of replay endpoints. This article is about that tape, the analytics built on it, and the intraday OI simulator that falls out of it.

The Problem With End-of-Day Open Interest

Every gamma-exposure model multiplies per-strike open interest by per-contract gamma. Open interest is published once per session, by the OCC, overnight. So when you compute GEX at 11:00 ET on Tuesday, you are using Monday's settled positions and Tuesday's greeks.

For a slow book that is fine. For 0DTE, or any session where a large position is opened and closed inside the day, it is badly wrong. The strikes that matter most for intraday hedging are frequently the ones that did not exist in size at last night's settlement.

The fix is to reconstruct what happened between settlement and now, from the prints. That is exactly what the flow layer does.

What Is Actually Stored

Two tape tables, both partitioned by day, both replayed point-in-time.

The side tag is the same buy / sell / mid classification the live ingest pipeline applies, derived from where the print landed against the prevailing bid and ask. Replay runs the identical classifier, so a historical bucket and a live bucket are computed the same way rather than approximated.

Per-symbol print counts on the options tape, the largest names in the snapshot:

4.66 billion prints in total, headed by SPY at 1.30 billion and SPXW at 815 million, and 52 symbols replay the full flow suite. /v1/tickers returns the live inventory.

The Endpoints

All of these take the same required at parameter as the rest of the Historical API, and all return the same JSON shape as their live counterparts.

A Real Minute

15:30 ET on 15 June 2020, five minutes of SPY option tape:

curl -H "X-Api-Key: YOUR_API_KEY" \
  "https://historical.flashalpha.com/v1/flow/options/SPY/history?at=2020-06-15T15:30:00&minutes=5"

{"symbol":"SPY","minutes":5,"count":5,"buckets":[
 {"ts":"2020-06-15T19:25:00","buyVolume":3288,"sellVolume":5991,"midVolume":2409,"netVolume":-2703,"tradeCount":1154,"biggestTrade":833,"vwap":2.27},
 {"ts":"2020-06-15T19:26:00","buyVolume":5752,"sellVolume":5427,"midVolume":2865,"netVolume":325,"tradeCount":1275,"biggestTrade":449,"vwap":2.80},
 {"ts":"2020-06-15T19:27:00","buyVolume":8533,"sellVolume":8549,"midVolume":3893,"netVolume":-16,"tradeCount":1900,"biggestTrade":600,"vwap":2.16},
 {"ts":"2020-06-15T19:28:00","buyVolume":9763,"sellVolume":5432,"midVolume":2124,"netVolume":4331,"tradeCount":1493,"biggestTrade":500,"vwap":2.00},
 {"ts":"2020-06-15T19:29:00","buyVolume":6319,"sellVolume":6897,"midVolume":2297,"netVolume":-578,"tradeCount":1361,"biggestTrade":400,"vwap":2.19}]}

Roughly 1,400 prints per minute, signed. The day-to-date summary at the same instant: 580,620 trades across 5,496 distinct contracts, 2,367,003 bought against 2,397,798 sold, 1,286,736 unclassifiable at the mid, net −30,795 contracts, largest single print 9,632 lots.

The at parameter is ET wall-clock; tape bucket timestamps come back in UTC, so a 15:30 ET request returns buckets labelled 19:25 to 19:29, while the analytics endpoints echo as_of in ET.

The OI Simulator

The tape is the input; the simulator is what most people actually want. It answers "what is open interest right now, mid-session" rather than "what was it at last night's settlement".

curl -H "X-Api-Key: YOUR_API_KEY" \
  "https://historical.flashalpha.com/v1/flow/oi/SPY?at=2020-06-15T15:30:00"

{"symbol":"SPY","as_of":"2020-06-15T19:30:00Z","expiry":null,
 "official_oi":26562722,"simulated_oi":26549508,"intraday_oi_delta":-13214,
 "oi_delta_confidence":0.43,"effective_oi":26562952,
 "contracts_total":10390,"contracts_with_flow":3453}

Mechanically: a per-minute snapshot is materialised by replaying the tape through the same side classifier and OI-delta estimator the live ingest uses, then any residual sub-minute prints between the minute floor and your exact at are applied on top. Replay and live therefore agree at the sub-minute boundary rather than drifting apart.

Feed that into /v1/flow/summary and you get the dealer-facing read at the same instant:

{"symbol":"SPY","underlying_price":307.025,"flow_direction":"amplifying",
 "intraday_oi_delta":-13214,"contracts_with_flow":3453,
 "live_gex":85103261.19,"flow_gex_pct_shift":1.37}

flow_gex_pct_shift is the number to watch: how far the day's flow has moved gamma exposure away from the settled-OI picture that every EOD-based GEX model is still showing.

Why This Is Leak-Free

Flow data is where lookahead bias is easiest to introduce and hardest to spot, so the controls are worth stating explicitly.

1. Every tape and OI read is bounded ts <= at. There is no branch in the query layer that can see past the requested instant.
2. Settled OI is start-of-day. The open interest stamped on a date is the interest from the previous session's settlement, so using it at 10:00 that morning is exactly what a live desk had.
3. Sub-minute residuals are replayed, not interpolated. Asking for 10:30:17 gives you the 10:30 snapshot plus the prints up to second 17, not the whole 10:30 bar.
4. No wall-clock reads exist in the replay code. A source-level test fails the build if any file in the replay namespace references the system clock. The only clock is injected, which is what makes a replay reproducible.

What You Can Build With It

• Effective-OI gamma models. Recompute GEX against effective_oi instead of settled OI and measure whether the intraday correction improves your hedging-pressure signal. flow_gex_pct_shift tells you when the correction is large enough to matter.
• Signed-volume momentum. Per-minute netVolume on the option tape against forward underlying returns, conditioned on the gamma regime from the exposure endpoints.
• Block detection with survivorship intact. /blocks at a size threshold, replayed across years, gives you a clean study of what large prints preceded - without the survivorship bias of a curated "unusual activity" feed.
• Cross-asset lead-lag. The underlying tape and the option tape share a clock, so you can measure whether option flow leads or follows the stock at minute resolution.
• Cross-sectional heat. /v1/flow/options/leaderboard replayed daily gives a point-in-time ranking of which underlyings were absorbing the most directional option flow.

Related Reading

• Backtest Flow Signals with Minute-Level Replay - a 502-session worked study on this data, artifacts included
• Dealer Hedge Flow in the August 2024 Carry Unwind - the same tape stress-tested as a pre-registered 5-minute event study
• Historical Options Analytics API: Replay GEX, VRP & Dealer Positioning
• Effective Open Interest: Live GEX From Flow
• Point-in-Time Greeks and Lookahead Bias
• Day-Over-Day OI Change: Detecting New Positioning

Upgrade to Alpha API Spec

The tape runs from January 2017 and the replay endpoints are live today. If your research needs intraday open interest, signed option volume, or dealer positioning that reflects what traded today rather than what settled last night, that is the flow layer, and it replays at minute resolution with the same no-lookahead guarantees as the rest of the Historical API.