Radar Perene / Archive / science

The signal registry that also knows when to quit

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

Science

Market signals almost never die in public. They are announced with fanfare, cited while they work, and then simply vanish from conversation — no obituary, no date, no explanation. Months later, when someone asks about the indicator that "never missed", the answer comes in a lowered voice: it stopped working, nobody quite knows when. That unrecorded death is not a detail: it is the hole through which any chance of learning something from the signal escapes.

A complete signal registry needs two rules, not one: the rule that defines when a signal is born, and the rule that defines when it is retired — declared just as far in advance, applied just as coldly. In the house's registry, the second rule has existed since the design stage. The exact expiry criteria are bench detail; the reason they exist is the subject of this article, and it is public by nature.

Why signals expire

A market signal is not a law of physics; it is a regularity conditioned on an environment. Let the conditions change — the interest-rate regime, the market's composition, the behavior of participants whom the signal itself helped inform — and the regularity can dissolve without warning. The finance literature documents this decay with regularity of its own: published effects tend to shrink once known, and patterns tied to a specific regime rarely cross the border into the next one, a phenomenon adjacent to what we describe in regime switching.

If signals expire, a registry that only knows how to admit them is structurally blind to half the story. It accumulates entries and never exits — and a ledger that only grows is not a portrait of what works; it is a museum of what once seemed to work.

The retiring rule is harder than the admitting rule

Admitting a signal is psychologically cheap: there is enthusiasm, novelty, the promise of usefulness. Retiring is expensive, for three reasons any researcher will recognize. First, attachment: a signal that has produced correct readings becomes emotional property, and abandoning it feels like ingratitude. Second, ambiguity: signals rarely fail cleanly — they weaken, oscillate, show occasional flickers of life, and there is always an argument for waiting one more cycle. Third, the narrative cost: every retirement is a public admission that something once admitted no longer deserves its place.

That is exactly why the expiry criterion must be defined beforehand, when no one is attached to anything. A quitting rule written on admission day does not negotiate with the affection of two years later. The parallel with admission is symmetric: just as a signal is only born after maturing across pre-defined fixed windows, it only remains while it keeps clearing criteria that were not written to save it.

What a recorded retirement teaches

The difference between a signal that vanishes and a signal retired with a date is the difference between forgetting and learning. A recorded retirement preserves three pieces of information that silent disappearance destroys.

The first is duration: how long the signal lived between admission and expiry — the beginning of an empirical answer to "how long does a pattern last in this market?". The second is the mode of death: did it expire by gradual weakening or by abrupt rupture? The two shapes tell different stories about the environment. The third is the house's own rate: what fraction of admitted signals reaches retirement, and how fast — the kind of internal statistic that disciplines enthusiasm for each new candidate. None of these lessons requires the signal to have succeeded; all of them require its death to have been written down.

Quitting as method, not defeat

There is a misreading this article wants to undo: the idea that retiring signals is the sad side of the operation. In the house's design it is the opposite — the quitting rule is what gives permanence its value. In a registry without expiry, "being in the registry" carries no information, because nothing ever leaves; in a registry where expiry is enforced as a matter of course, every active signal carries the information that it has cleared, so far, the same criteria that have already retired others. Selection only means something when exit is possible.

This is the same logic that governs the rest of the house's output: the hypothesis that can be refuted, the finding that can fall in a retest, the signal that can expire. What cannot die cannot mean anything either.

Frequently asked questions

Can a retired signal come back?

It can be readmitted — but as a new candidate, going through full maturation again, with no credit for its previous life. The old biography is information; it is not a passport.

Is expiry automatic, or does it involve judgment?

The design favors criteria defined in advance, precisely to shrink the room for judgment at the moment when attachment is strongest. The exact calibration between rule and judgment is a bench decision.

Does retiring a signal amount to saying it never worked?

No. It means the signal stopped clearing the permanence criteria. A signal may have described well an environment that existed and ended — the retirement dates the end of the environment; it does not disown the past.

How many house signals have been retired?

The registry's contents — how many signals, which ones, on what dates — are operational and not public. What is public is the architecture: admission by maturation, permanence by criterion, exit on the record.

---

Continue the trail: From the signal registry to the published index: the road not every signal travels →

House reading: the daily behavior of the living indicators is in the Diário; the regimes that ended, with dates, in the Atlas.

Reviewing a strategy's forgotten signals — the ones nobody formally retired — tends to be the most revealing part of an audit, and it is a conversation the house knows.