Radar Perene / Archive / science

A contrarian signal, tested and refuted: what happened when our own thesis did not hold

◦ Index methodology v2.2 (working papers with DOI). See the methodology.

Science

Every research house has a pet thesis. Ours was among the most respectable: the idea that extreme market pessimism carries, inside it, the announcement of the turn — that fear, taken to its limit, becomes a buy signal for whoever has the stomach. It is the contrarian thesis, and it has pedigree: it appears in the memoirs of celebrated investors, in sentiment handbooks, in half of all year-end interviews. We tested it against our own sentiment indicator. It did not hold.

When a house's own thesis fails the test, the protocol of a research operation allows a single honest outcome: publish the failure with the same formality that would have been reserved for success. That is what happened. The study went out as a working paper, with the negative conclusion in the text itself, deposited on Zenodo under DOI 10.5281/zenodo.21327608 — citable, dated, versioned. This article is about that decision, not about the test.

A thesis that deserved the test

Contrarian folklore survives because it is partly plausible. Major market bottoms have, on occasion, coincided with extreme pessimism readings — and selective memory does the rest, keeping the coincidences and discarding the extremes that led nowhere. The honest question was never "has extreme fear ever preceded a turn?" (it has), but rather: do extreme readings of the indicator offer a systematic edge, defined before the test, that survives scrutiny?

That formulation carries a cost the trading-desk conversation never pays: it can fail. And there is a respectable theoretical reason to expect it to. If prices incorporate information with any efficiency — the central argument of the efficient market hypothesis — an edge as public as "buy when everyone is afraid" should be among the first to disappear. Testing the contrarian thesis is, at bottom, asking whether the market leaves money on the table in the best-lit corner of the room.

What the public outcome says — and what stays on the bench

What we can show is what any reader can verify: the published conclusion records that the tactical edge tested does not hold up. The current version of the study sits in the repository with the previous version preserved alongside it, as scientific versioning practice requires. How the windows were defined, which variations were pre-specified, what the exact test design was — that is bench work, and the bench is not the place of this text.

The asymmetry between what gets published and what gets detailed is not a badly kept secret; it is the normal division of labor between a public result and the craft that produced it. The outcome is verifiable by anyone; the full method lives where it always has, in the house's internal routine.

Why publish one's own refutation

The most informative shelf in a research library is the one holding the results that came to nothing — and it is precisely the shelf almost no one displays. The incentives of publishing and marketing point the other way: the indicator that "worked" gets shown, the one that failed gets quietly archived, and the public forms an image of the house made only of hits. The technical name for this is publication bias; the practical name is advertising.

Publishing the refutation with a DOI inverts the incentive on purpose. First, because it creates a record the house itself cannot rewrite later: the negative conclusion is dated and deposited beyond our editorial reach. Second, because it changes what the reader can infer from our positive results — a house that publishes its own failures hands the public a ruler for calibrating trust in its successes. Third, because a thesis refuted with method is worth more than a thesis intact for lack of testing: the space of hypotheses has shrunk, and shrinking the space of hypotheses is what research does.

None of this turns the failure into a trophy. The protocol does not celebrate the error; it prevents the error from becoming a secret. The distinction matters: what got published was not "we were wrong" — it was "we tested it, it did not confirm, and the record is here".

What a refutation does not settle

A negative conclusion is about the formulation tested, not about every possible formulation. The study does not show that extreme sentiment is useless as information — it shows that the specific edge tested, under conditions defined before the test, did not hold. Future versions of the question, with new data or a different design, remain legitimate; that is how a research line advances, by refinement rather than stubbornness. What publication guarantees is that any new version will have to measure itself against a public record, not against a convenient memory.

Frequently asked questions

Doesn't publishing a negative result weaken the house's credibility?

Experience suggests the opposite. A track record made only of hits is statistically implausible, and attentive readers know it. The public record of a self-refutation is one of the few proofs of methodological honesty that cannot be fabricated after the fact.

Does the negative result mean the sentiment indicator is useless?

No. It means the specific tactical edge tested did not confirm. An indicator can describe the environment well without offering an operational edge — description and edge are different claims, tested in different ways.

Why deposit the refutation with a DOI instead of just mentioning it on the site?

Because a site text can be edited or removed; a DOI deposit cannot. The DOI transfers custody of the record to an independent repository, which makes the commitment to the outcome irreversible.

Is the contrarian thesis dead in the literature?

No — it remains debated, with mixed results that vary by market, period and definition of "extreme". What the house's study adds is a specific record from the Brazilian market, with a negative outcome, available to anyone who wants to confront it.

---

Continue the trail: How many windows fit a signal: why seven and not one →

House reading: what the sentiment indicator records each session is in the Diário; the archive's extreme episodes, in the Atlas.

Putting a specific "edge" through the same sieve that refuted ours is the kind of exercise the house's bench knows from the inside — and it yields more in conversation than in text.