Playbook
FDA catalysts: how drug approvals move biotech stocks
Single-drug FDA decisions are among the cleanest binary catalysts in public markets — a single ruling can move a stock ±30-80% in a day. But the run-up often matters more than the print, an approval is not automatically bullish, and measuring the move correctly requires a sector benchmark, not the S&P 500. Here is the full framework for studying FDA events with an agent.
The short version
- Binary drug decisions move single-name biotech stocks roughly ±30-80% in a day — one of the largest single-day move profiles in equities.
- The run-up can matter more than the print: stocks typically run 20-40% over the four to eight weeks before a decision, so much of the upside is priced before the announcement.
- Approval is not automatically bullish on the day it prints. A fully priced-in positive outcome often triggers a sell-the-news reaction.
- Benchmark biotech catalysts against XLV (SPDR Health Care), not the S&P 500 — then report sample size, mean/median CAR, and a t-stat before drawing any conclusions.
Why biotech binaries are different
Most equity events — earnings, guidance, macro prints — move stocks in a continuous distribution around a consensus estimate. An FDA decision on a single-drug application is structurally different: the outcome is discrete (approve or reject), the ruling date is often pre-announced (the PDUFA date, set by statute), and the entire value of an early-stage biotech can hinge on one product. That combination produces some of the largest single-day absolute moves available in public markets.
Class I recalls — the highest-severity category, involving products that could cause serious harm — are the mirror image: a surprise recall can erase a large fraction of a company’s market cap in a single session. Both approval and recall are public-domain data, sourced from the openFDA API (US FDA, CC0 public domain). Approval dates in the structured data are derived from submission status dates in the system; coverage of sponsor-to-company links is best-effort fuzzy resolution, because the FDA does not publish financial identifiers alongside drug records. Partial company mapping is expected and documented — a company with no resolved CIK simply has no financial-markets data attached.
The data surface is structurally clean, which is unusual. Most catalyst research requires you to scrape press releases or parse unstructured PDFs. FDA decisions are logged in a structured government database, timestamped, and refreshed on a defined cadence. That makes event-study methodology tractable in a way that, say, analyst rating changes or M&A rumors are not.
The run-up vs the print
The most common analytical mistake in biotech catalyst trading is treating the announcement day as the primary event. It frequently is not. Because the PDUFA date is known weeks in advance, speculative positioning accumulates in the run-up window. Stocks typically appreciate 20-40% over the four to eight weeks before a decision as probability-weighted expectation of approval gets priced in. By the time the ruling arrives, much of the upside is already in the price.
The consequence is sell-the-news dynamics: a positive ruling on a stock that has already priced in a high probability of approval can produce a flat or even negative announcement-day return. The approval is genuinely good news in an absolute sense — it just wasn’t new information to a market that had already moved. The analysis changes sharply if the drug was a long shot or if the market had priced in rejection.
Do not assume that “approval = buy.” The announcement-day return depends on the run-up, the implied probability priced in before the ruling, and the competitive context. Measuring the full event window — before and after — is the only way to see what actually happened.
This also means that studying only announcement-day returns gives you an incomplete picture. The economically interesting window for a researcher is the combined run-up plus the post-announcement drift — how much total return did an informed position generate from the time the drug’s candidacy was known to the point where the market had fully adjusted? That requires a longer event window and a cumulative abnormal return calculation, not just a day-of snapshot.
How to study the move properly
The standard tool for FDA event analysis is a market-adjusted cumulative abnormal return (CAR) study. The mechanics: for each event (approval or recall), compute the stock’s daily return minus the benchmark return for each day in the window (e.g., days −20 to +5 relative to the event date). Sum those residuals across the window to get the CAR for that event. Average across a sample of events, then test whether the mean is statistically different from zero.
The benchmark choice is not cosmetic. If you benchmark against the S&P 500, macro co-movements and general risk-on / risk-off days contaminate your residual. But even that is less important than the sector effect: healthcare stocks move together on FDA policy news, sector fund flows, and macro-medical themes. A biotech that happens to get its approval ruling on the same day as an adverse FDA press conference will look like a negative-CAR event even if the drug news was perfectly positive. The right benchmark is XLV, the SPDR Health Care Select Sector ETF — subtracting XLV’s return removes sector-level noise and isolates the company-specific drug news in the abnormal return estimate.
Reporting discipline matters as much as the benchmark choice. An event study is only informative if you report:
- n — the number of events in the sample. Single-digit n is anecdote, not evidence.
- Mean and median CAR — both, because biotech return distributions are highly skewed. The median is more robust to outlier events.
- t-statistic — to determine whether the mean CAR is distinguishable from zero given the cross-event variance. A small sample with high idiosyncratic variance will produce an insignificant t-stat even on a large mean.
A single-name event study is always descriptive. If you are studying one company’s history of FDA decisions, the n is too small to draw causal conclusions. It tells you what happened; it does not tell you what will happen. Cross-company samples with meaningful n and a significant t-stat are the threshold for an “edge.”
The insider × FDA cross-signal
One of the more unusual research angles available when you have both SEC filings and FDA records in the same system is a direct join: did insiders make open-market purchases or sales in the weeks before a drug decision, and what was the realized post-event move? This cross-signal uses non-derivative open-market transactions from SEC Form 4 filings — the cleanest proxy for voluntary insider conviction — keyed to the same company-event that triggered an FDA action.
You can compute both the raw post-event return and the XLV-adjusted abnormal return for each insider-trade-to-FDA-event pair. The output is a list of coincidences: cases where an insider bought or sold before a known catalyst, and what the stock did afterward.
This is descriptive only. The sample sizes in any individual company’s history are tiny, and a coincidence between an insider purchase and a positive FDA outcome is not evidence of informed trading, a predictive edge, or anything actionable. It is a reading list — a set of events that warrant further investigation, not a conclusion. State this clearly whenever you surface the output. The value is as a starting point for due diligence, not as a signal in its own right.
How to track catalysts with an agent
With a connected agent and a markets MCP server, the full workflow runs in plain English. The tools map directly onto the analytical steps above:
| Step | Tool | What it does |
|---|---|---|
| Get the event feed | list_drug_approvals | Structured FDA approval records with sponsor, drug name, application number, and approval date |
| Get recall events | list_drug_recalls | FDA recall records by severity class, company, and product; Class I is the market-moving tier |
| Per-company history | fda_company_events | Full approval and recall history for a single issuer, with resolved company CIK where available |
| Measure the move | fda_event_study | Market-adjusted CAR vs XLV over a configurable event window; returns mean, median, n, and t-stat |
| Cross the insider signal | insider_activity_before_fda | Form 4 non-derivative trades in the pre-event window joined to FDA outcomes; descriptive, small n |
A plain-English version you can paste to a connected agent:
“Pull the FDA approval events for the past six months. For any company where the CIK resolved, run an event study: CAR vs XLV over a window from 30 days before to 10 days after the approval date. Report the mean and median CAR, the sample size, and the t-stat. Also show me the run-up in the 30-day window before each announcement so I can see how much was already priced. Flag any cases with insider open-market purchases in the 60 days before the ruling, and note that those are descriptive coincidences, not signals.”
You can also set up a persistent FDA watchlist so you are notified when new approvals or Class I recalls are filed for any company in your watchlist. The fda watch type anchors on approval date and recall initiation date, and fires an alert each time a new event is logged. That replaces the need to manually poll the approval feed and lets you focus on the analysis rather than the monitoring.
What not to conclude
- Approval is not automatically bullish on the day it prints. If the run-up has already priced in a high probability of success, the announcement-day return can be flat or negative even on a genuine positive ruling.
- Partial company-mapping coverage is by design. The FDA publishes no financial identifiers alongside drug records. Company-to-CIK resolution is best-effort fuzzy matching; a missing CIK does not indicate a data error — it means the sponsor could not be unambiguously matched to a public company.
- The insider × FDA join is descriptive with tiny n. Coincidences between pre-event insider trading and positive outcomes are not evidence of a predictive edge. Treat the output as a reading list, not a signal.
- Binary catalysts in companies with weak balance sheets are gambles, not value plays. A company that burns cash and has no approved products outside the drug under review is effectively a single-event option. Position sizing should reflect that, regardless of how good the event-study statistics look in aggregate.
- Always report n and t-stat. An event study on three or four cases from one company’s history has no statistical weight. The honest conclusion for small samples is “unconfirmed” — surface the numbers and let the reader evaluate.
FDA catalyst analysis is one of the strongest event-driven research workflows available to a systematic researcher, because the data is public, structured, and timestamped. The discipline is in understanding what the data can and cannot tell you: the announcement-day return depends on the run-up, the right benchmark removes sector noise, and any insider-activity cross-signal is descriptive only. Get those three things right, and “there’s an FDA catalyst coming” stops being a headline and starts being an analysis.
For more on insider open-market conviction, see how to find insider buying that actually predicts returns — the same open-market-only and cluster-buy filters apply when interpreting the pre-FDA insider cross-signal.