Options flow

Definition

Options-flow analysis reads the tape of options transactions — volume by strike/expiry/side, sweep orders, premium size, open- interest changes — as a positioning and information signal about the underlying. The premise: informed traders use options for leverage, so unusual options activity may front-run stock moves. The evidence gives the premise qualified support with a decisive catch: the informative signal in the academic record requires knowing trade DIRECTION (buyer- vs seller-initiated), which public feeds only approximate.

How it works / structure

• Raw materials: per-contract volume vs open interest (volume >> OI = new positioning), trade-size buckets, sweep detection (marketable orders split across exchanges — urgency), premium spent, and next-day OI confirmation (ms-option-chain).
• Direction inference: trades at/above ask read as buys, at/below bid as sells — an approximation that degrades in tight/fast markets and for spread legs (a “call buy” print may be one leg of a neutral spread; multi-leg decomposition is the hard part).
• Parameters (engine-executable): unusualness definition (volume vs trailing baseline — rel_volume for options), direction-inference rule (pinned), aggregation window, event-proximity filter (flow before known catalysts is hedging-contaminated), and the confirmation rule (OI change).
• Aggregate cousin: put/call ratios are options flow collapsed to one number (opt-put-call-ratio).

When it applies

Hypothesis generation on single names (unusual flow → investigate the calendar and filings — lens-event-catalyst); sentiment/ positioning context (sent-cot-reports); dealer- positioning inputs at index level. Never standalone: flow without direction confidence and event context is noise with a narrative.

Risk profile & failure modes

• Direction misclassification: the bid/ask rule mislabels a material share of prints; spread decomposition failures turn neutral structures into phantom directional signals.
• Hedging contamination: much large flow is hedging (corporate, dealer, fund) with no directional information — the flow’s WHY is unobservable.
• Retail noise floor: post-2020 retail options volume (including 0DTE — opt-0dte-mechanics) buried informed flow deeper in the tape; historical signal estimates may not transfer.
• Marketing folklore: “follow the smart-money sweeps” services claim unaudited hit rates — folklore, labeled.

Evidence & limits

Pan-Poteshman (2006), using data WITH known trade direction, found put/call volume ratios predicted individual stock returns over days — the canonical evidence that informed trading occurs in options. Their signal used non-public direction data; public approximations are strictly weaker. Platform stance: flow is evidence-grounded as a phenomenon, weak as a public signal; every flow convention is replayed against direction-inference error.

Falsifiable-thesis examples

Illustrations only, not signals:

• “Names in universe U with top-decile inferred call-buy flow will outperform the universe over the next 5 sessions this quarter in replay” — falsified by the bucket returns.
• “This flagged flow event will be followed by a disclosed catalyst (filing, announcement) within 10 sessions” — falsified by the calendar.

Cross-references

• Collapsed aggregate: opt-put-call-ratio; positioning frame: sent-cot-reports
• Mechanics: ms-option-chain, opt-0dte-mechanics
• Discipline: lens-sentiment (crowd vs informed), lens-event-catalyst (calendar context)