Radar Perene / Archive / science

How the Atlas tests whether a precedent is a stylized fact or a coincidence

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

Science

A coin lands heads five times in a row. Two explanations compete: the coin is loaded, or chance did what chance does. Both hypotheses produce the same sequence — which is exactly why looking at the sequence is not enough. Every historical market archive lives this dispute at industrial scale: patterns that repeated a few times compete for the title of rule against patterns that merely happened. Telling one from the other is, perhaps, the central task of anyone who studies the past without wanting to be fooled by it.

A stylized fact is an empirical regularity that repeats with enough consistency, across different periods and contexts, to be treated as a trait of the system — not an accident of one particular sample window. The term entered economics with Nicholas Kaldor, in 1961, to name the broad patterns a serious theory must explain before proposing anything.

The Atlas — this house's archive of historical episodes — exists on top of that border. Each essay records an episode; the permanent temptation is to promote the episode to law. The house's method is a set of questions designed to resist that temptation.

The first question: how many times, in how many worlds

Repetition, by itself, impresses more than it informs. Five heads in a row happen, on average, once in every 32 honest sequences — whoever looks at many sequences will find many perfectly normal "loaded coins." That is why the Atlas's first question is not "how many times did the pattern appear?" but "in how many different environments did it appear?".

A pattern that only repeats inside the same regime — the same rate cycle, the same market mood, the same decade — may be a trait of the regime, not of the system. The archive treats that distinction as its own subject: changing level is not changing regime documents why the house separates the two before counting repetitions. Five occurrences scattered across distinct environments are worth more, as evidence, than ten occurrences clustered in a single chapter of history.

The second question: does the pattern survive outside the shop window?

Every pattern is discovered in a sample window — and the window of discovery is always the most flattering one, because that is where the pattern drew attention. The honest test begins when one looks for the pattern where it was not discovered: other periods, other assets of the same family, shifted windows. The literature calls this validation outside the original window; the house calls it taking the pattern out of the shop window.

It is the stage where most candidates die. And the house can say so with public backing, because it has run its own collection through that sieve. The working paper Brazilian Intramarket Relationships started from a set of relations between sectors of the Brazilian market — pairs that folklore treats as reliable gears — and retested them under the same ruler. In numbers: of the relations examined, a single one survived the retest, the one linking the utilities sector to the market's defensive moves; the study is public on Zenodo, with an active DOI, and its second version openly revises a finding the first one upheld. The rest of the folklore did not pass. Publishing the mortality — not just the survivor — is part of the method.

The third question: do the precedents agree with each other?

A genuine stylized fact produces similar outcomes when conditions look similar. A coincidental pattern produces whatever chance dictates. That is why the Atlas confronts its own episodes with one another: when three similar precedents point in three directions, the archive records the divergence as information — the subject has its own essay in when precedents disagree. Disagreement among precedents is not the archive failing; it is the archive saying that there is no rule there, only repertoire.

That is, perhaps, the most visible difference between the house's method and the common use of historical analogies. The common use picks the precedent that confirms the thesis and presents it alone. The method forces the precedent to show up with its siblings — including the ones that tell another story.

What the Atlas does with the verdict

The test's outcome is not binary, and the archive's vocabulary reflects that. On the top step sit the regularities that crossed environments, survived outside the shop window and keep their precedents in reasonable agreement — those are treated as stylized facts and cited as reliable context. On the middle step, promising patterns with short track records: recorded, monitored, never promoted in advance. On the bottom step, what failed — and what failed remains in the archive, dated, because the list of what did not work is the part of the method that protects the reader most.

None of this closes the conversation for good. A stylized fact is a revocable title: the system changes, and regularities of decades have dissolved in economic history without asking permission. The archive treats every promotion as provisional by construction — the ruler stays applied, including to the approved.

Frequently asked questions

Is a stylized fact the same as an economic law?

No. A law claims to hold always; a stylized fact is a dominant regularity that admits exceptions and an expiration date. Economics has almost no laws; it has well-documented stylized facts — and the term's modesty is part of its value.

How many repetitions are enough to promote a pattern?

There is no magic number, and distrusting whoever offers one is good hygiene. The honest answer combines three things: how many occurrences, how scattered across distinct environments, and how consistent in their outcomes. Ten cases in the same regime can be worth less than five across different regimes.

What happens when a stylized fact in the archive fails?

The failure is recorded with a date and the pattern steps down — publicly. The retest of the intersector relations cited above is the example with a DOI: findings of the first version that did not hold are revised in the second version's own text.

Is a discovered coincidence always useless?

No — as long as it is labeled as what it is. An honestly recorded coincidence becomes a hypothesis under observation. The damage never comes from the coincidence; it comes from the wrong label.

The same ruler the Atlas applies to the past can be pointed at a question that has no study yet — the subject of research on demand: when research does not depend on a department →

House readings: today's note, in the Daily · the precedents on file, in the Atlas.

Running a specific precedent through this method's three questions is an exercise the house performs in conversation, outside the public article.