Insights
Unusual Options Activity Research: What Serious Sources Test
Alphanume Team · September 4, 2026
An unusual options print is an observation. Turning it into a signal requires far more work than drawing an arrow beside volume.
Options activity feels informative because the tape contains visible urgency. A large sweep, a volume spike, or a block above the offer looks like a trader expressing a view. The same print can also be a hedge, one leg of a spread, a position close, dealer inventory transfer, or delayed reporting.
Serious research does not ask whether a trade looks bullish. It asks whether a precisely defined event predicts a precisely defined outcome after controlling for the ordinary ways options volume becomes large.
Start with the economic object, not the alert
The term unusual options activity combines several different objects: unusual contract volume, unusual premium, unusual aggressor direction, unusual open-interest change, and unusual concentration near a catalyst. Each implies a different hypothesis.
- Volume anomaly: today's contract volume is large relative to its own history or to open interest.
- Directional pressure: trades appear to demand liquidity at the ask or bid after quote alignment.
- Position formation: next-day open interest suggests that volume opened rather than closed exposure.
- Volatility demand: implied volatility moves after controlling for the underlying, surface shape, and market volatility.
- Catalyst concentration: activity clusters in expiries that span a scheduled or inferred event.
Choose one object. Combining them after seeing returns creates a flexible screen that is difficult to falsify and easy to overfit.
Classify the trade before assigning intent
A trade classification model must align executions to the prevailing bid and ask, account for quote latency, and handle prints reported in batches. Even a correct aggressor label does not reveal the customer's portfolio. Buying calls can hedge short stock. Selling puts can be one leg of a volatility position.
Spread detection is therefore central. Look for contracts in the same underlying with nearby timestamps, matched size, related expirations, and offsetting deltas or vegas. A single leg extracted from a multi-leg order can reverse the inferred direction of the actual package.
The options-chain guide is a useful starting point for contract fields. Researchers also need corporate-action adjustments, exercise conventions, and a policy for trades that cannot be classified without invention.
Use open interest carefully
Volume greater than yesterday's open interest does not prove a new position. Contracts can change hands multiple times in one session, and open interest is generally observed after clearing rather than at execution time. Next-day changes help, though they still aggregate opening, closing, exercises, assignments, and corrections.
A defensible study treats open-interest change as supporting evidence. It may define high-confidence opening activity when volume, next-day open-interest growth, spread classification, and trade direction agree. Ambiguous events remain ambiguous instead of being forced into a clean story.
Contract adjustments deserve their own audit. Splits, special dividends, mergers, and symbol changes can create contracts whose strikes, multipliers, or deliverables no longer resemble standard options. A naive volume baseline can flag these series as extraordinary even when the anomaly is administrative. Either normalize adjusted contracts with documented rules or exclude them consistently from the eligible universe.
This is one reason a historical alert feed must be point-in-time. Reconstructed rankings that use revised open interest or end-of-day volatility can leak information unavailable when the alert would have fired. See what point-in-time data means before evaluating any vendor backfill.
Build control groups that answer the actual claim
Raw post-alert returns are not evidence. High-volume options often occur in volatile names, around earnings, after sharp stock moves, and in securities with strong retail attention. Those conditions have their own return distributions.
- Underlying control: match on market capitalization, liquidity, recent return, realized volatility, and short interest where available.
- Calendar control: compare events at similar times of day and distances to earnings, expiration, and known announcements.
- Contract control: match moneyness, tenor, delta, spread width, and baseline option volume.
- Regime control: condition on index volatility, market direction, and broad option-volume regime.
- Selection control: replay the exact alert rule across every eligible contract, including alerts that look boring in hindsight.
The estimand should be explicit. You might test next-day delta-hedged option returns, underlying returns through the next catalyst, implied-volatility change, or realized-versus-implied spread. Those outcomes are not interchangeable.
A minimum viable event-study specification
for each option trade at timestamp t:
join the last eligible quote before t
classify aggressor side with an ambiguity bucket
group probable multi-leg executions
compute volume surprise using history available at t
attach catalyst distance and next-day open-interest change
apply the frozen alert rule
assign a matched non-alert control
measure underlying, option, and delta-hedged outcomes
subtract fees, spread cost, and hedge turnover
retain every event, including overlapping and failed signalsFreeze the specification before reading subgroup results. Report event counts, cross-sectional concentration, median outcomes, tails, and clustered uncertainty. A thousand alerts in five meme stocks do not provide a thousand independent observations.
The strongest work also separates discovery from confirmation. Develop thresholds in one period, lock them, and evaluate later data. If the signal depends on a percentile, compute that percentile only from history available at the event timestamp.
Report abstention as a model output. When quotes are crossed, timestamps cannot be aligned, the trade is probably multi-leg, or the contract history is too short, the correct classification may be unknown. Coverage will fall, yet the surviving sample will have a meaning that can be defended.
Execution can erase a visually strong result
Options are expensive instruments for weak signals. Displayed spreads can be wide, fills are path-dependent, and a backtest that marks at the midpoint may claim liquidity no trader could obtain. Contract selection adds another decision layer after the alert fires.
Model at least three fill assumptions: optimistic midpoint, executable quote with a measured improvement rule, and conservative touch plus slippage. Add commissions, exchange fees, assignment risk, exercise handling, and the cost of delta hedging if the thesis is about volatility rather than direction.
Capacity needs contract-level treatment. A strategy that buys a fixed fraction of displayed size will behave differently from one that assumes unlimited fills. The options data provider guide lists the fields worth checking before choosing a historical source.
What a serious source should disclose
- The exact alert rule and when every input became observable.
- How trades, spreads, cancellations, and quote conditions were classified.
- Whether open interest is contemporaneous, next-day, revised, or absent.
- The universe, delisting policy, corporate-action treatment, and catalyst calendar.
- The return object, hedge rule, exit rule, and overlapping-position policy.
- Costs, fill assumptions, capacity limits, and results under worse execution.
- Every tested subgroup or a correction for the search across them.
Our mispriced-options study shows the kind of question that becomes interesting only after the mechanism is specified. The order-flow research guide shows why trade classification and market mechanics can matter more than an exciting label.
The practical conclusion
Unusual activity can be a useful feature. It is rarely a complete strategy. The research advantage comes from reconstructing packages, controlling for catalysts and volatility, choosing a tradeable outcome, and preserving the alerts that failed.
Begin with a narrow replication rather than a universal scanner. One underlying universe, one contract class, and one holding period make timestamp and execution errors easier to see. Expand coverage only after the first result survives an untouched period and a deliberately worse fill assumption.
Be skeptical of evidence built from screenshots, selected winners, or a vendor's current leaderboard. Ask for a full point-in-time event file and replay it under your own execution model. If you cannot reproduce the signal definition, treat the feed as an idea generator.
For more mechanism-first examples, browse the Alphanume Research archive. The restrained next step is simple: define one anomaly, one control group, and one outcome before collecting more alerts.