Pre-Registration — Solar Phase → Sovereign-Default Clustering, year-clustered test

Registered: 2026-09-17, before any analysis was run. Why this document exists: the first registered test (PREREG_SOLAR_PHASE_CRISIS_TEST_2026-08-16.md, run the same day, result in SOLAR_PHASE_CRISIS_TEST_RESULT_2026-09-17.md) did not support crisis clustering as a whole. Banking crises showed nothing. Sovereign defaults showed a 12–16 % excess in the declining phase that was significant when every defaulting country counted as an event (p = 0.020) and fell to p = 0.051–0.066 when each onset-year counted once. That gap is the question: is the tilt a property of default years, or an artefact of many countries defaulting in the same year? This registration asks that one question, with a test that treats the year as the unit and the country count as its weight, so it cannot be answered by double-counting a wave.

Status: RUN 2026-09-17 — H1 not supported on the primary test (p = 0.053). Result: SOLAR_PHASE_SOVEREIGN_DEFAULT_TEST_RESULT_2026-09-17.md. Nothing below was revised. Revisions require a new dated document and both remain public. This test is a follow-up and is reported with the first, never instead of it.

1. Hypothesis, one direction only

H1: Years in which at least one sovereign external default begins fall in the declining phase of the solar cycle (maximum to following minimum) at a rate above the phase's share of time.

H0: Default-onset years are distributed across phase in proportion to phase length.

Banking crises are not part of this test. They were tested under the first registration and showed nothing; re-testing them here would be a second look at a null result.

2. Data — fixed, identical to the first registration

• Solar phases: SILSO published minima and maxima, cycles complete at run time (cycle 25 excluded). Declining = [maximum, next minimum). Same table, same parser, same file.
• Defaults: Reinhart & Rogoff sovereign external default series (HBS Global Crises panel, column "SOVEREIGN EXTERNAL DEBT 1"), 70 countries, 1800–2016, as published. Onset = the first year a country's flag turns from 0 to 1. Dated to mid-year. No additions, exclusions or re-dating.
• Scope: 1800 onward (source coverage). The pre-1850 exclusion is reported alongside as in the first test.

3. Unit of analysis and test — fixed now

• Unit: the onset-year. Each calendar year with ≥ 1 onset is one observation, carrying a weight equal to the number of countries that defaulted that year.

Sovereign-default onset years vs solar phase — RESULT of the follow-up registration

Registration: PREREG_SOLAR_PHASE_SOVEREIGN_DEFAULT_TEST_2026-09-17.md, written and fixed before the run. Run: 2026-09-17, same day, same data as the first test. Script: scripts/research/prereg_sovereign_default_phase_test.py. Verdict under §3–4: H1 NOT supported on the primary test. The tilt is there in direction and size (about 14 % more default-onset years in the declining phase than its share of time predicts) and it does not reach the declared threshold (p = 0.053).

Reading it

• Primary (unweighted onset-years, 1800–2015): 66 of 97 default-onset years fall in the declining phase against 57.8 expected; ratio 1.14; one-sided p = 0.053. Below the line, on the wrong side of it.
• Secondary (weighted circular-shift permutation): p = 0.121. Shifting the solar calendar at random produces a weighted declining-phase share at least this large about one time in eight. The timing of the large default waves relative to the solar phase is not unusual.
• From 1850: primary p = 0.069; the permutation p = 0.021 is the one figure under 0.05 in either test, and it is a secondary statistic on a subset. It is reported here and is not promoted (§3).