Market Myth Audit #0: Does Options Flow Front-Run Biotech Catalysts?

The myth

“Smart money knows before biotech catalysts, and you can see it in the options flow.”

You have seen the screenshots. A biotech gets acquired or a drug readout hits, and within hours someone posts the options tape from days earlier: aggressive buying in short-dated, far out-of-the-money calls that now look prescient. The implied playbook writes itself: watch the flow into FDA decision dates and data readouts, follow the confident-looking positioning, and ride the outcome someone else already knows.

The trigger for this study was one of those cases: a biotech buyout announced in early July 2026 where, in hindsight, a burst of cheap short-dated call buying appeared several sessions before the announcement. The question we set out to answer was whether that story generalizes. Not “did it happen once,” but “does pre-event positioning predict catalyst outcomes often enough to matter?”

How we tested it

Data. Three sources, combined: a professional FDA catalyst calendar (5,619 deduplicated events from 2021 to 2026, filtered to 3,548 precise-dated catalysts on 599 optionable biotech tickers); roughly two years of daily per-ticker institutional options-flow data (call and put volume, premium splits, open interest, and each stock’s own rolling baselines); and split-adjusted daily prices from our data providers, including delisted and acquired names to limit survivorship bias. The study window was 2024-08-15 to 2026-07-07, about 23 months, bounded by the depth of the options-volume history.

Features. For every ticker-day we computed positioning features using only data available on that day, aligned to the day before the event, with no lookahead: call-volume elevation versus the stock’s own baseline, call-to-put volume ratio, bullish premium share, net call premium scaled to the stock’s own history, open-interest tilt, and market-adjusted price behavior versus the biotech sector. Everything self-normalized, so a busy mega-cap and a quiet small-cap are judged against their own norms.

Three tests, fixed in advance

• •Prediction: do the ten trading days before a scheduled catalyst predict the event reaction, in direction and in size? “Material” was defined up front as a market-adjusted move of at least 8% across the event’s two-day window.
• •Pop pre-detection: for every single-day market-adjusted pop of at least 25% in the biotech universe (buyouts and surprise readouts live here), would the same features have flagged the stock in the prior ten sessions at better than chance?
• •Base-rate honesty check: run the composite bullish alert across all 242,605 biotech ticker-days in the window and measure how often it fires and what follows, so any “hit rate” from tests 1 and 2 has a denominator.

What we found

Direction: nothing, leaning contrarian. Across 3,441 scheduled catalysts (932 of them material), 50.8% of material outcomes were up: a coin flip. The bullish positioning features scored an AUC of 0.41 to 0.46 for separating up-moves from down-moves, and the deviation below 0.50 was statistically significant (z = minus 2.9). Read that carefully: in our sample, elevated pre-event call buying was associated with slightly worse odds that the catalyst resolved upward. The crowd in the tape before biotech events was not smart money. It was just a crowd.

Big pops: mostly unreadable, and barely flaggable where readable. Of 2,775 single-day pops of at least 25% in the window, 86% happened in names with no usable options market at all: thin microcaps where there is simply no flow to read. Restricted to windows where options data existed, the features flagged 61.8% of pre-pop windows versus 52.4% of matched random windows, a lift of about 9 percentage points over chance, with a median lead of about 7 trading days. Real, but weak. Also telling: only 7.5% of those pops coincided with a same-week calendar catalyst. The giant-pop tail is mostly acquisitions, financings, and microcap spikes, not scheduled events.

Base rates keep everyone honest. Across all 242,605 biotech ticker-days, the composite bullish alert fired on 8.3% of days. Following an alert, the probability of a material up-move within ten days was 38.0%, versus a 31.3% base rate: a 1.22x lift. Tightening the alert raised the lift to 1.33x while cutting frequency. Monotonic, so the signal is not pure noise, but a 1.2x to 1.3x lift on a noisy base rate is not a tradable directional edge and we do not present it as one.

The one thing that did show up: directionless intensity. On high-impact scheduled catalysts, when pre-event activity ran hot against the stock’s own baseline, the events moved more, with no reliable information about which way. In our sample of high-impact catalysts:

Intensity discrimination for material-versus-not reached AUC 0.66 to 0.67 on high-impact events, and it held for self-normalized features, so it is not just a big-stock artifact. To be explicit about what this is: activity clustering before scheduled events coincided with larger moves in either direction. It says “this readout may be loud,” not “this readout will go up.” We treat it as a descriptive volatility observation, full stop.

What we did not find

• •No directional edge. The headline myth, that flow tells you the outcome, failed in a sample of 3,441 events. Worse than chance, not better.
• •No acquisition front-running system. Most of the pops that make the screenshots happen in names with no options market to read, and most are not on any catalyst calendar in the first place.
• •No “prescient print” fingerprint. As a follow-up, we tried to reverse-engineer the motivating case: a filter built around the exact profile of that famous pre-buyout print (short-dated, far out of the money, fresh open interest, volume several multiples of the stock’s own baseline). Over 83 sessions of scanning it fired 72 times. The apparent payoff lived entirely in moves of 50% or more, where the lift looked like 4.26x. That bucket contained exactly two events, and one of them was the case the filter was fitted on. Remove it and the result is 1 hit in 53 firings: not significant. This is the textbook overfitting trap, and the base rate of 50% moves is so low (0.65% of firings’ windows) that a properly powered test would take over a decade of scanning. We classify this question as likely unanswerable with available data, which is its own kind of answer.

Limitations

Stated plainly, because they bound every claim above:

• •Event timing is fuzzy. The calendar gives precise dates but not whether news lands before or after the close; we mitigated with a two-day reaction window, which blurs some measurements.
• •Survivorship is only partly handled. Prices cover delisted and acquired names, but options-volume history ends at delisting, and the current-universe construction misses some dead tickers.
• •We tested volume, not option prices. Without deep options price history we cannot simulate a volatility strategy’s profit and loss, so the intensity pattern is shown as move-size probability, not as a strategy result.
• •Microcap data artifacts pollute the extreme-pop set, including one print so extreme it can only be a split-adjustment data glitch; they do not affect the optionable-universe conclusions.
• •Dark-pool positioning is untested at scale. Per-ticker dark-pool history only became available to us around March 2026, roughly four months of it, which is far too short to backtest. The block-trade version of this myth remains open, not refuted.

What would change our mind

We hold the verdict, not a grudge. Any of the following would reopen the question: deeper options history that extends the sample beyond 23 months and one market regime; dark-pool block data accumulating past March 2026 until a real backtest is possible, at which point we intend to run it; different event classes, since earnings, macro prints, or index events may behave differently than FDA catalysts; or a forward, pre-registered replication of the intensity observation on out-of-sample events. If someone publishes a well-constructed study finding a directional edge we missed, we will audit that too.

What's next for the series

Audit #1 will be pre-registered: question, data, tests, and cutoffs published before a single number is computed, then the results, whatever they are. If there is a piece of market folklore you want put under the same lamp, send it in; reader submissions are how the queue gets built. In the meantime, our live market pressure index shows how we present aggregate data descriptively, the momentum compression pillar covers the framework this research program grew out of, and the learning hub has the plain-English background on every concept used here.