Asclepius
07-bayesian-persuasion-disclosure.md

07-bayesian-persuasion-disclosure

The framework's Game-Theory Adversary agent treats biotech sponsor disclosures not as exogenous information but as strategically chosen signals. This document gives that lens its empirical grounding — the same discipline the reflexivity adjustment gets from 06-signaling-equilibrium.md.

Why this matters

Every disclosure decision a biotech sponsor makes — which subgroup analyses to report, which secondary endpoints to feature, when to issue press releases relative to fund-raising — is a choice over what signal structure to reveal. Kamenica and Gentzkow (2011) showed that a rational sender with a known signal structure can move a receiver's posterior beliefs in any direction the sender prefers, subject only to Bayes-plausibility. The sponsor is the sender; the investor is the receiver; the signal structure is the protocol-defined-but-discretionary disclosure pattern.

This frame is not cynical. Sponsors have legitimate reasons to disclose strategically — protecting trial integrity, sequencing communications, avoiding selective leakage. But from the investor's read, the framework's recommendation should incorporate the equilibrium-optimal signal- extraction behavior, not naive face-value reading of the press release.

Four empirically-grounded patterns

The Adversary agent's persuasion lens pattern-matches against four disclosure behaviors documented in peer-reviewed work. Each is detectable from public information about an asset.

1. Competitor-approval-triggered disclosure suppression

Kao (2024, Management Science 71(7):5948-5970) applies a difference-in-differences design to FDAAA-mandated trial-results disclosures and shows that a competitor's drug receiving FDA priority review reduces the focal firm's probability of disclosing its own trial results within two years by approximately 13 percentage points. Pharma firms are roughly 25 percentage points worse than public institutions on mandated disclosure compliance overall; the cross-firm cost imposed by a competitor's approval is dominant.

The mechanism is information-economic: when a competitor's approval shifts the receiver's prior about category viability, the focal firm's posterior — conditional on no disclosure — drifts upward (Akerlof's absence-of-bad-news read). Disclosure becomes costlier and is delayed strategically. The FDA has not collected any of the $37B+ in potential fines for FDAAA non-compliance, so the equilibrium is sustained.

Detectable signal: cross-reference the asset's named competitors' priority-review/approval dates against the sponsor's last clinical-results press release on the asset. Flag any sponsor whose competitor was approved >18 months ago without a corresponding focal-asset disclosure.

2. Pre-announcement information leakage

Rothenstein, Tomlinson, Tannock and Bouchard (2011, JNCI 103(20):1507) analyzed 23 randomized cancer trials whose Phase 3 readouts moved company stock prices. They found that price drift begins in the direction of the eventual result well before the formal announcement, with sustained post-announcement movements of ~9.4% for positive Phase 3 readouts. Negative readouts show asymmetric, larger drawdowns (PLOS ONE follow-up, PMC3737210). The drift is most plausibly explained by information leakage to a restricted set of informed traders before the press release.

Detectable signal: if the sponsor's stock has drifted meaningfully — say, >5% over the 30 trading days before a disclosed readout — the press release is confirming an already-leaked private signal rather than delivering new information. A framework's PoS estimate that has not been explicitly contemporaneous (i.e., that reads current sell-side consensus) has already double-counted the leakage. Flag and discount the informational content of the disclosure.

3. Post-hoc subgroup disclosure as signal partition

The publication of post-hoc subgroup analyses in oncology trials grew from 384 (2000-2014) to 695 (2019-2022) — a near-doubling on a per-trial basis (Yi et al. 2024, Therapeutic Advances in Medical Oncology, PMC11025370). Roughly 95% of these analyses do not apply multiplicity adjustment. Under Kamenica-Gentzkow's framework, this is the textbook persuasion- optimal behavior: the sponsor chooses which subgroup to publish after observing the data, and chooses the one whose posterior most favors the sponsor's preferred receiver action.

Detectable signal: any press release whose headline names a subgroup not pre-specified in the SAP — especially when the all-comers/ITT result is omitted from the same release — is a textbook persuasion- optimal signal partition. The Adversary should flag this as moderate- to-critical, depending on whether the omitted ITT result is recoverable from the underlying registration data.

4. Endpoint substitution and hierarchy omission

Documented commentary in the practitioner literature — most accessibly Derek Lowe in Science — flags the pattern of biotech press releases that do not state the prespecified primary endpoint's place in the testing hierarchy, or that foreground secondary endpoints when the primary did not separate. The SAP hierarchy is statistical preregistration; a release that buries it is choosing a signal structure that obscures the analytical pre-commitment.

Detectable signal: a readout release that reports a P-value on a secondary endpoint but not on the primary, or that does not name the hierarchy position of the primary, is an equilibrium-optimal persuasion move. Flag and downgrade.

What this lens does not claim

This is a frame for investor reading, not a claim about sponsor intent. A sponsor who engages in any of the four patterns above may be acting in the best interests of the asset's development; the lens does not infer motive. What it does do is treat the disclosure choice itself as endogenous to the sponsor's information set, so the framework's posterior on PoS is not contaminated by face-value reading of optimized signals.

This is the analytical move that distinguishes investor-grade reading of clinical disclosures from press-release-driven reading. The Adversary's persuasion lens operationalizes it.

Cross-references

Canonical citations