PaperSep 23, 2026Free abstract
A Falsificationist Cycle Analysis of Cryptocurrencies
Luigi Garone
For more than half a century, the cycle analysis of markets has produced rich conceptual frameworks that have rarely been subjected to empirical falsification. The three historical schools — classical American (Hurst), Italian (battleplan and inverse cycle) and quantitative DSP (Ehlers) — are adopted as alternative doctrines, almost never tested jointly on a statistically relevant sample. This work formalizes a Popperian protocol that puts each founding claim of the three schools to the test on a multi-asset dataset — Bitcoin and Ethereum since 2017, Solana since 2020, extended to thirty-seven instruments in six classes and to six stock indices with up to ninety-nine years of daily history — over several thousand phased hierarchical cycles. The methodology is causal by construction and deterministic, and both properties are formalized in automated tests. The non-repainting of the module that detects Bayer tongues, tested in walk-forward mode over the whole series, holds — with a maturity margin derived from the measured latency — on every asset/timeframe combination that can be tested, but one of the five cannot be tested yet: it is therefore not proven, not demonstrated. To this scaffolding the protocol adds four requirements, which are the part of the work that outlives the individual results: the null value of a measure must be measured, not assumed, with surrogate series that preserve the power spectrum and destroy the temporal structure; a conditional rate must be accompanied by its own marginal, that is, by the same condition counted on the cases the event does not select; a negative outcome does not count until it has been shown that the test would recognize the phenomenon if it were there; and every level must be measured on the timeframe that resolves it, because the daily timeframe does not separate the shortest levels of the hierarchy and measuring them there changes the very numbers under discussion. The results refute the most exposed canonical assertions: Hurst’s fixed harmonic nesting and the principle of harmonicity — the latter with a test whose power is verified on a true synthetic scale, where on the daily ruler none of the markets with enough peaks turns out to be closer to the rungs than a set of periods drawn at random from the same grid —, the sequential mean-reversion of durations, Bayer’s canonical threshold for connecting cycles, the peak hold, and — by tautology — two claims of the Italian school: the inverse cycle, whose operational definition is not falsifiable ex ante, and the unconditional swing, whose success condition is true almost always even without the swing — in all the parent cycles of the daily, and in at least nine out of ten on every ruler. They confirm, with statistical significance and cross-asset convergence, the cyclic commonality of durations — strong at the short levels and, on long history, equally strong at the high levels —, cyclical translation as a predictor of polarity, the violation of the previous peak and, in a rewritten form that eliminates a set-theoretic inclusion, the two bearish signals of the Italian corpus. The same yardstick applies to the regularities the work could claim as its own, and there it is more severe. The short-cycle tail, measured against one hundred phase-randomized surrogates and one hundred AAFT per market with each level on the ruler that resolves it, leaves the cloud in one asset/level combination out of thirty-three, below the falsification threshold: it is what the detector produces, not what the market does. The same holds for the deviation from the nominal period and for the position of the peak within the cycle. And the work’s starting axiom, according to which a quaternary structure would be a binary of binaries, cannot be decided with this instrument: the trough that would separate the two binaries is the lowest of the three internal ones in 1.3% of the 557 measured cycles, within the band that the same detector produces on noise; and on synthetic series in which the binary of binaries is implanted, the detector never recovers it in numbers sufficient to vote. The contribution is threefold: a falsifiable and reproducible protocol applied uniformly to the whole corpus, regardless of the origin of the assertion; the verification of causality and determinism of the central analytical component, with non-repainting measured over the whole series instead of on a few points; and the systematic distinction between what the market produces and what the detector produces, which on several occasions changes the sign of a conclusion — to the point of removing, among the regularities this work could have claimed as its own, those that seemed most solid.
v1 · 10.2139/ssrn.7534678
For members:full text · PDF · dataset · replication code