Radar Perene / Archive / science

How research is born: from question to hypothesis

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

Science

No research project we know from the inside ever began with data. It began with an irritation: a sentence spoken with too much conviction, a chart circulating as proof of something it did not prove, a rule of thumb repeated by so many people that no one remembered where it came from. Irritation is cheap and abundant. What separates the researcher from the mere dissenter is what gets done with it in the days that follow.

Research is born when a loose question is converted into a testable hypothesis: a specific claim, with a declared direction and a criterion for being wrong, that the available data can confirm or refute. Everything that comes after — method, testing, publication — depends on that conversion. Most questions never make it across.

The question is not the hypothesis

"Does the market's fear get it wrong?" is a legitimate and useless question. Legitimate because it points at something real; useless because no test can answer it — "fear", "market" and "wrong" are words without a ruler. The conversion demands three decisions the original question hides: which indicator measures the phenomenon, over what window it will be observed, and what counts as an answer.

The converted version sounds less poetic and more accountable: "when indicator X reaches extreme range Y, do returns over the following Z months differ from the historical pattern?". It loses its dinner-party charm; it gains the only property that matters — it can be wrong. A claim that cannot be wrong is not a hypothesis; it is an opinion with borrowed vocabulary.

The silent funnel between the two

Between the irritation and the hypothesis lies work that almost never appears in the final texts: checking whether the question has already been answered, whether the data exists, whether the series is long enough to support any conclusion at all. It is the reading and exploration phase — legitimate, free, uncommitted. What this house learned to respect is the etiquette between phases: exploring is rummaging without promises; testing requires declaring beforehand. Whoever blends the two produces the vice the literature calls p-hacking, and the boundary between the phases is precisely what pre-analysis registration protects.

That funnel discards most candidates. Questions die because the data does not exist in public quality, because the answer is already in the literature, because the available series is too short to tell a finding from chance — or because, phrased with rigor, the question turns out to have asked nothing.

A house case, told lightly

One of the questions that made it through the funnel here was born of the market's most repeated folklore: the idea that extreme pessimism signals a turning point — that fear, at the limit, gets it wrong. That was the loose question. The converted version became a specific hypothesis about the house's sentiment index: that extreme readings of the indicator would carry an exploitable contrarian edge, over windows defined before the test.

What we can show of that process is the outcome, because it is public: the hypothesis was tested and did not hold. The study was published anyway, with the negative conclusion in the text itself, as a working paper on Zenodo under DOI 10.5281/zenodo.21327608. The middle of the road — how the window was chosen, which variations were pre-specified, the protocol applied — stays on the bench, where the craft lives. What this article can say is the essential part: the question became a hypothesis before it touched the data, and that is why the negative answer was a finding, not an embarrassment.

What makes a question worth the effort

Over time, three informal filters settled into the house routine. The first is decidability: is there public data capable of answering, within a human timeframe? The second is consequence: does the answer, whichever way it falls, change any reading — or is it drawer curiosity? The third is the most uncomfortable: does the question survive its own rigorous formulation? Many do not; phrasing them well already answers them, and that too is a research outcome, even if it never becomes a text.

None of this is a recipe — it is the honest summary of a process still under revision here. The full depth, from a reader's question to a registrable hypothesis, is bench work.

Frequently asked questions

Does every hypothesis need to be original?

No. Retesting a known hypothesis with new data, a new period or a different market is legitimate research — a good part of the finance literature is exactly that. The originality can live in the data, not in the question.

What if no data exists for the question?

The question waits. Part of the craft is keeping questions open until a public source reaches them — and acknowledging the limit is more honest than answering with bad data.

How many questions become research, in practice?

Few. In the funnel described above, most die at the data check, the literature check or the formulation itself. There is no universal number; what the house's experience suggests is that high mortality is a sign of a working filter, not of sterility.

Does intuition have a place in this process?

It does — at the origin. The initial irritation is almost always intuition. What research demands is that intuition agree to be converted into a testable claim and, if it comes to that, refuted.

---

Continue the trail: Levels of research: from curiosity to working paper, from preprint to peer review →

House reading: what the surviving hypotheses produce every day is in the Diário.

Turning a specific question into a registrable hypothesis is the kind of exercise that yields more in conversation than in text — and the house has that conversation, when sought out.

Characters: Method