Does the Market Turn at the Midpoint of Mercury Retrograde? We Tested All 241 Since 1950

The Conclusion

Everything below is how we got there — the definitions, the base rates, the controls, and an interactive table of all 241 retrogrades since 1950 where you can open the chart for every midpoint and check us.

The Claim — and Its Astronomy Checks Out

Three times a year, Mercury appears to reverse course and drift backwards through the zodiac for about three weeks. The current retrograde runs from June 29 to July 23, 2026, and its halfway date — the midpoint between the two stations — falls on the night of July 11–12. So the claim's astronomy is accurate. It is also not arbitrary: the midpoint of a retrograde coincides almost exactly with Mercury's inferior conjunction, the moment it passes directly between Earth and the Sun. In traditional astrology this is the “cazimi” — the heart of the retrograde — and it is a natural candidate for a turning point if Mercury cycles drove markets at all.

That gives us a clean, mechanical test. We computed geocentric Mercury positions from a Swiss Ephemeris for every day from 1950 to 2026, detected all retrograde periods, took each one's time midpoint, and asked one question: did the S&P 500 change direction on or near that day more often than it does on any random day?

Test 1: The Claim as Stated

“Direction change” has to be defined before looking, or it will stretch to fit whatever the chart shows. We used a standard swing pivot: the highest intraday high or lowest intraday low of a ±5 trading-day window. Intraday extremes rather than closes, deliberately — a V-shaped reversal that bottoms on a long wick counts as a turn. A midpoint scores a hit if such a pivot lands within two trading days of it. That is a generous definition — a five-day window either side, and a two-day tolerance on top.

Checked by eye over the recent past, the claim looks alive: 18 of the 33 midpoints since 2016 (54.5%) had a pivot nearby, including some spectacular ones — the January 24, 2022 V-bottom fell exactly on a midpoint. Scroll enough charts and you will walk away a believer. The crucial step — the one chart-scrollers never run — is the control: how often does a randomly chosen trading day pass the exact same test?

The Full 76-Year Record

Extending the test across every tolerance from the exact day to ±3 trading days gives the midpoints four chances to prove themselves. They don't take any of them.

Nothing here comes close to significance (p-values are one-sided, testing whether the midpoints beat the base rate — values near 1 mean they score below it). We repeated the grid with looser (±3-day) and stricter (±8-day) swing definitions: twelve combinations in all, and not one beat its base rate. At the widest windows the midpoints score below random — most plausibly noise from the way market turns cluster, not evidence that Mercury suppresses reversals.

Check Every Midpoint Yourself: All 241 Since 1950

Don't take the aggregates on faith. Below is every Mercury retrograde since 1950 with its midpoint, whether the S&P 500 printed a swing pivot within two trading days of it, and a chart for each one — click any row to see what the market actually did around that date. Compare the hit rate chip against the random-day base rate next to it: that gap, not the raw count of hits, is the entire question. The hits and the misses look exactly alike until you know which is which.

Test 2: Require the Reversal to Mean Something

A reversal nobody could trade is not a reversal worth predicting. So we tightened the definition: a day counts as a reversal only if price moved at least 1.5% against the direction of the prior five days, within the next five trading days. No requirement on the size of the move into the day — just a tradeable counter-move out of it. On the S&P 500, 28.4% of all trading days qualify (5,462 of 19,242).

With a definition that actually filters, the midpoint doesn't just fail to shine — it slips further behind the random-day deck:

Below the base rate in all four windows. If anything, the wider windows flirt with significance in the wrong direction (±2 days: 53.9% vs. 59.8%, p ≈ 0.04 for “fewer reversals than random”) — which we read as noise, not as Mercury calming markets. Either way, there is nothing here to trade.

The last decade — the window that makes the claim feel alive — looks like this on the exact day:

What About Trading Only When the Cycle Is Hot?

Fair question — our platform applies exactly this idea to seasonal patterns: fit a cycle to the strategy's equity curve, call the rising half of the sine wave HOT and the falling half COLD, and only trade the hot phases. So we ran the midpoint idea through the same engine (Lomb-Scargle spectral analysis on the HP-detrended equity curve — the identical code that powers the app).

First the idea needs a testable rule. The natural one from this study: at every midpoint, the backtest fades the prior five-day move — long when the market fell into the date, short when it rose — exiting after five trading days. That produces 240 simulated trades since 1950. Baseline result: a 44.6% win rate and −55.1% total return over 76 years. For scale, buy-and-hold returned roughly +44,000% over the same period. There is nothing here — but let's filter it anyway. The fitted cycle has a dominant period of about 49 events (≈16 years), and the regime comes from the slope of the fitted sine wave.

That looks like a rescue: the hot phases turn a catastrophic loser into a modest winner and the cold phases hold all the damage. (For the record, the fitted wave currently sits in negative territory — COLD by the app's classification, at under 20% confidence — with its slope just turning upward.) But this table was produced by fitting the sine wave to the entire equity curve and then grading each trade with hindsight — the fit already knows where the good and bad stretches were. Even with that advantage, the hot/cold win-rate split is plain noise (46.2% vs. 43.4%, p = 0.34). The honest version is walk-forward: refit the cycle before every trade using only the data available at the time, counting a trade only when the forecast said HOT.

The magic doesn't just evaporate — it inverts. Without hindsight, the forecast-HOT trades lose more per trade than the forecast-COLD ones (−0.40% vs. −0.17%), the win-rate gap is comfortably inside noise (p = 0.41), and every row still loses money. The in-sample table above was the sine wave memorizing history, not Mercury telling the future. Here are all three equity curves side by side — taking every trade, taking only the hindsight-HOT trades, and taking only the forecast-HOT trades:

All figures on this page are historical backtest statistics published for research purposes only — nothing here is trading advice or a recommendation to buy or sell any security.

Slicing Deeper: Combustion, Signs, and the Leo Cell

Maybe the midpoints only work under the right conditions. So we sliced the 240 trades by every astrological variable a practitioner might reach for: the sidereal sign and element Mercury occupied during the retrograde, the Moon's phase at the midpoint, the retrograde's duration, Mercury's distance from Earth at the conjunction, and Jupiter's and Saturn's elements. Across eight families of splits, not one separates winners from losers at any defensible threshold (the best family p-value is 0.13, before correcting for the eight ways we looked).

A note on combustion, since Mercury retrograde plus combust is a popular combination. Mercury stations 15–22° from the Sun and is combust (within the traditional 8.5° orb) only for about the middle third of each retrograde — a window that is, by construction, centered on the midpoint. So for midpoint trades the condition is not a filter at all: the midpoint is the inferior conjunction, and Mercury is combust at every single one (widest separation in all 240 events: 7.3°). Grading by how close the conjunction is doesn't help either: the tighter half of conjunctions won 40.8% vs. 48.3% for the looser half (the wrong direction for the theory), and the three near-cazimi midpoints (under 1°) all lost.

One cell in the grid does glitter: midpoints with Mercury in sidereal Leo went 16 of 22 (72.7%), averaging +1.11% per trade — and the record is eerily symmetric, 8 of 11 in each half of the 76 years. Here is every one of them, so you can inspect the evidence yourself:

Why the Claim Feels So True

The Checklist for Any “Always Reverses On” Claim

This study took an afternoon, and the recipe generalizes to every date-based market claim you will ever hear:

We build and test astronomical timing models for a living — some survive this kind of scrutiny and become strategies on our platform. This one doesn't. That distinction, between patterns that survive a base-rate test and patterns that only survive eyeballing, is the entire game.

Frequently Asked Questions