Three layers of signal, three different explanations

The event-based analysis produced too many results. Dozens of astrological features survived the strictest multiple-comparison corrections across S&P 500, VIX, and earthquake data. We flagged the obvious problem — slow-planet aliasing — and built two controls to kill it. The slow-planet signal died. But underneath it, two other signals were waiting.

One of them is calendar seasonality wearing an astrological costume. The other one we can’t explain yet.

Quick recap

We loaded four event datasets: 19,182 S&P 500 trading days (960 extreme), 9,128 VIX days (1,022 extreme), 10,424 M6+ earthquakes, and 62 wars. Computed natal charts for each event — planetary positions at that moment. Tested every astrological feature for enrichment among extreme events versus a stratified null distribution.

The initial results were absurd. 25 features survived FDR correction for S&P 500. 63 for VIX. The top hit — Saturn-Uranus opposition — had an odds ratio of 5.08 in the VIX data. We immediately suspected temporal clustering aliasing, confirmed it by checking when extreme days actually occur (75 of them in 2008 alone, when Saturn happened to be in Virgo), and set about building controls.

Two controls, applied in sequence.

Control 1: fast planets only

Saturn takes 29.5 years to orbit the Sun. Jupiter takes 12. Uranus takes 84. These planets barely move during a financial crisis. If extreme market days cluster in 2008 and Saturn was in Virgo all of 2008, every enrichment test involving Saturn in Virgo is just detecting “the 2008 crisis happened.” The null model — non-extreme days from the same decade — can’t control for this because extreme days cluster within specific years while null days spread across the entire decade.

The fix: restrict the analysis to fast-moving planets. The Sun completes its cycle in a year. The Moon in a month. Mercury, Venus, and Mars in 1–3 years. Their positions change fast enough that they’re decorrelated from multi-year crisis clustering. If Mars is in Capricorn for 6 weeks every 2 years, and extreme days happen to fall in those 6 weeks more often than expected, that can’t be explained by “the 2008 crisis.”

Results with fast planets only (Sun, Moon, Mercury, Venus, Mars):

VIX: 5 features survive FDR correction. Mars in Capricorn leads with an odds ratio of 2.28 and a p-value that rounds to zero. Mars in Leo: OR 1.73. Mercury in Libra: OR 1.58. Venus-Mars trine: OR 1.64. Mars in Sagittarius: OR 1.40.

S&P 500: nothing survives FDR, but Mars in Capricorn is the #1 fast-planet hit with a permutation p-value of 0.001. Twelve raw hits at p < 0.05 versus six expected — suggestive but not conclusive on its own.

Earthquakes: 4 features survive FDR. Moon in Leo (OR 1.15, permutation p = 0.0001) and Sun in Aries (OR 1.13). Tiny effect sizes but enormous sample.

Wars: nothing. 62 events is not enough.

The slow-planet signal is dead. But the fast-planet signal is not zero.

Control 2: block-shuffle null model

The fast-planet restriction eliminates aliasing from planets that move on multi-year timescales. But there’s a subtler version of the same problem: do extreme days cluster within specific weeks or months in ways that correlate with faster planetary cycles?

The block-shuffle null model addresses this directly. Instead of comparing extreme days to non-extreme days drawn from the whole decade, it groups extreme days into crisis episodes — clusters of extreme days within 60 days of each other — and shuffles labels within each episode’s time window. For each permutation, the same number of days in the same time window are randomly labeled “extreme.” This preserves all temporal structure: if 75 extreme days fall in October-November 2008, the permutation draws 75 random days from that same October-November 2008 window.

Any feature that survives this test cannot be explained by temporal clustering at any timescale.

S&P 500 block-shuffle results (fast planets, 10,000 permutations):

9 of 60 fast-planet features at p < 0.05 (expected ~3 by chance).

VIX block-shuffle results (fast planets, 10,000 permutations):

13 of 60 fast-planet features at p < 0.05 (expected ~3).

These features survive the strongest null model we can construct. Now we need to explain them.

Layer 2: the September–October effect

The Libra cluster jumps out: Sun in Libra, Mercury in Libra, Venus in Libra all survive block-shuffling in both datasets.

The Sun is in Libra from approximately September 23 to October 22. Mercury and Venus are always within a few signs of the Sun, so they’re frequently in Libra during the same period.

This is the well-documented September–October effect in financial markets. October is historically the most volatile month. The 1929 crash was in October. The 1987 crash was in October. The peak of 2008 volatility was in October. Even after controlling for crisis clustering with block-shuffling, October days within a crisis window are more volatile than, say, March days within the same crisis window.

The astrological encoding is rediscovering calendar seasonality. Sun in Libra is not a celestial influence — it’s a calendar label. “The Sun is in Libra” means “it’s late September to late October,” and late September to late October is genuinely the most volatile period of the year. Mercury and Venus follow the Sun, amplifying the seasonal pattern through correlated astrological features.

This is methodologically interesting — the pipeline correctly detected a real pattern in market data — but it’s not astrologically interesting. Any seasonal decomposition would find the same thing.

Layer 3: Mars in Capricorn

Mars in Capricorn survives everything.

• Survives fast-planet restriction (Mars completes its cycle in ~2 years, not correlated with crisis clustering)
• Survives block-shuffle permutation in both S&P 500 (p = 0.033) and VIX (p < 0.0001)
• Appears as the #1 fast-planet hit in the standard null analysis for both datasets
• Odds ratio of 2.28 in VIX (extreme days are 2.28× more likely to occur when Mars is in Capricorn)
• Mars spends approximately 6 weeks in Capricorn every 2.1 years — this is not a seasonal pattern

Mars in Capricorn cannot be explained by:

• Slow-planet aliasing: Mars moves too fast (eliminated by fast-planet restriction)
• Calendar seasonality: Mars in Capricorn falls in different calendar months across different years
• Temporal clustering: block-shuffling preserves the crisis structure and the signal persists

What Mars in Capricorn could be:

• A genuine signal where Mars’s orbital position correlates with market volatility through some unknown mechanism
• A subtler form of period aliasing that our controls haven’t captured — Mars’s 2.1-year cycle could interact with some economic cycle of similar period
• A coincidence that happens to appear in two related datasets (S&P 500 and VIX are highly correlated — VIX literally measures S&P 500 implied volatility) and would not replicate on independent data

The honest assessment: we don’t know which one it is. The statistical evidence is strong. The prior probability of any astrological influence on markets is very low. The right move is more testing.

The earthquake signals

Moon in Leo (OR 1.15, Bonferroni p = 0.004) and Sun in Aries (OR 1.13, Bonferroni p = 0.03) survive FDR correction in the fast-planet-only earthquake analysis.

The effect sizes are tiny — a 15% and 13% increase in earthquake likelihood. But with 10,924 events, these are statistically robust. And the Sun and Moon are the two bodies with a plausible physical mechanism for influencing seismic activity: tidal forces. The Earth’s crust experiences tidal stress from the gravitational pull of the Sun and Moon, and several geophysics papers have reported weak correlations between tidal loading and earthquake triggering.

We note this without strong claims. The astrological encoding (sign placement) is a very crude proxy for tidal force, which depends on declination and distance, not ecliptic longitude. A proper test would use tidal stress calculations, not zodiac bins. But the fact that the two bodies with the strongest tidal influence are the two that show signal is at least consistent with the mechanism.

What’s next

The Mars-in-Capricorn signal needs three things before we take it seriously:

Replication on independent data. The S&P 500 and VIX are not independent — VIX is derived from S&P 500 options. We need a genuinely independent market: the Nikkei 225, the FTSE 100, or commodity futures. If Mars in Capricorn shows up in Japanese market volatility, the coincidence explanation gets much harder.

Out-of-sample testing. Split the S&P 500 data into a training period (1950–2000) and a test period (2000–2026). If the signal is only present in one era, it’s less likely to be real.

Mechanism investigation. Mars’s 2.1-year synodic cycle — is there any known economic or market cycle with a similar period? If Mars in Capricorn is a proxy for some ~2-year periodicity in market structure, that would explain the signal without invoking celestial influence.

The methodology

Fast-planet analysis. Restricted to Sun (index 0), Moon (1), Mercury (2), Venus (3), Mars (4). 60 planet-sign tests + up to 60 aspect tests among fast-planet pairs + element/modality counts using only fast planets. ~127 total tests per dataset.

Block-shuffle permutation. Crisis episodes defined as clusters of extreme days with no gap > 60 days. S&P 500: 108 episodes. VIX: 57 episodes. For each episode, the permutation window extends 30 days before and after the episode boundaries. Within each window, the same number of days are randomly labeled “extreme.” 10,000 permutations, seed 42.

Independence note. S&P 500 and VIX share 7,753 overlapping trading days. The VIX extreme days are correlated with (but not identical to) S&P 500 extreme days — VIX measures implied volatility, not realized returns. Treat the cross-dataset consistency as suggestive, not as independent replication.