Do George Bayer’s Astrological Trading Rules Work? We Tested 55 of Them on 764 Markets

George Bayer and his handbook

Bayer was a contemporary of W. D. Gann who published planet-based market rules in unusual detail. His 1940 book is usually catalogued as Stock and Commodity Traders’ Hand-Book of Trend Determination. It has 48 numbered rules for wheat, 11 for hides that he says “stock traders” can use by substituting “stocks” for “hides”, one for cotton and one for lard. He traded wheat himself, and most of his examples are Chicago wheat from 1934 to 1940.

The rules are strikingly specific. Rule 2 expects three days of decline whenever Mercury’s daily motion is exactly 59′ faster than Mars’s. Rule 13 watches the seventh decimal of the logarithm of Neptune’s distance from the Earth. Rule 23 steps through the daily change in Saturn’s heliocentric latitude in fifths of an arc-second, which he admits is “not even as big as the twinkle of the eye-lid.” Others project from the Sun’s position at a market top, from the date Mercury turned retrograde a year earlier, or from where Mercury stood 80 years before.

Some rules name a direction (“a down trend of about three days”), some name tops or bottoms, but most promise a “change of trend” and leave the direction to the chart. Bayer also insists the rules are best used together: when two or three agree, “a change in the market MUST occur.” That is a set of claims precise enough to test.

Has anyone tested this?

We found no published test of the rules against market data. They circulate as scans of the book and are discussed one at a time, for example in Planetary Trader’s rule-by-rule series, usually with chart examples. Bookseller listings repeat the 48-rules-for-wheat description; none report how often the rules worked.

Two of our earlier studies came close. In Mercury retrograde midpoints (Bayer’s rule 44) and perihelion and aphelion dates we found that the S&P 500 turned no more often than on random days. This study runs the same kind of test on the whole book.

Step one: turning his rules into dates

Bayer read positions from the American Nautical Almanac and Raphael’s Ephemeris. We computed the same quantities with the Swiss Ephemeris at six-hour steps from 1810 to 2028: geocentric longitudes and speeds, heliocentric latitudes and distances, declinations, right ascensions, and the Almanac’s “reduction to orbit.” Where he quotes an Almanac number we can check it. Saturn’s heliocentric latitude on March 29, 1937: −2°12′29.0″ in his Almanac, −2°12′28.5″ in ours. Mars’s right ascension on January 10, 1938: 23h04m20s in both.

Then we compared the computed dates with his examples, requiring the event identity where we annotated one from his text, such as a particular latitude crossing or direction. That identity check covers 108 of the 333 examples. These comparisons help choose interpretations; they are not independent evidence of a market effect.

His examples help distinguish some readings. Rules 30, 31 and 6A say “heliocentric declination”, but the extremes he quotes (up to 27°19′) are impossible heliocentrically and only the Almanac’s geocentric declination reproduces his dates, 7 of 7 against 0 of 7 for rule 6A. Some mismatches appear to be slips in the book: on March 16, 1940, the example for rule 27, Mercury was retrograde, so its speed could not have crossed 59′ a day; rule 7’s list gives “June 17” for three different years.

Six rules can’t be made mechanical and were left out: three built on a celluloid ellipse that he says only gives results on Keuffel & Esser chart paper (rules 11, 12 and 18), a rule that reads a human figure in the chart (24), one that counts moves in fixed numbers of points from 1930s price levels (25), and one that needs a move’s end to find its midpoint (45). Rule 1A is the same as rule 39 and rule 5A the same as rule 44, by Bayer’s own account.

How we tested them

1. Markets. Every symbol in our production database: 481 US stocks, 23 US ETFs, 160 UK, German and Indian stocks, 38 futures including Chicago and Kansas City wheat, 31 indices, 16 currencies and 15 cryptocurrencies. That is 6.46 million daily bars, with the Dow Jones back to May 1896.
2. Dates. The rules produced 56,988 dates between 1896 and October 2026. A date falling on a weekend counts on the next trading day. Every date was scored on every market trading at the time: 11.18 million date–market pairs.
3. The claim, as Bayer makes it. A “change of trend” is scored as a swing high or swing low (the highest high or lowest low of the 11 days around it) within a day of the date, since he says selling “within that day” is doing fine. A “top” or “bottom” must be that kind of swing. A direction is scored as the price move over the number of days he names, from the close before the date, so the move on the day itself counts. We use log returns, capped at ±50% before scaling. In the 56 primary statistical tests, directional returns are divided by each market’s recent volatility over that horizon, so a more volatile market does not dominate. We also show the unscaled log-return percentages separately.
4. Against random dates. Each result is compared with the same market on random dates within a year either side. Swings were more common before 2001, when US prices moved in eighths and often tied; the local comparison takes that out.
5. Placebos. Each rule’s own pattern of dates was slid along the calendar by random amounts, and moved by whole years so that seasonal effects cancel. Rules that count from a market’s own tops and bottoms were also run with made-up numbers in place of Bayer’s.

The results: 56 statistical tests

We made 59 comparisons: 56 statistical tests and three descriptive analog comparisons. The chart puts the statistical results on one scale: the difference from random dates, in standard errors. The orange marks show a standard-normal reference; the rules share events and markets, so the shape alone does not establish that every rule is noise.

The turning-point rules, 31 of them, sit almost exactly on the random-date rate of 38.6%: the dates of rule 13 (Neptune) were turns 38.6% of the time, rule 40 (Venus’s latitude) 38.6%, rule 1 (Mercury’s speed) 38.4%. Two in five days are within a day of a swing point, which is why examples are so easy to find. The rules that name a top or a bottom were within a few points of random as well.

The rules that name a direction did no better. Rule 2’s three days of decline after Mercury runs 59′ ahead of Mars turned into an average rise of 0.24% on our markets, against 0.11% after random dates. Rule 42 expects two days up; prices fell 0.06%. Rule 31 expects at least a few days of rising prices; they fell 0.30% over five days, against a 0.19% rise on random dates. These are unscaled log-return percentages; seven of nine underperformed their unscaled-return random-date comparison. None of the nine primary directional tests, which scale each market by its volatility, cleared the bar.

Only rule 30, Venus’s declination crossing 23°26′51″, cleared p < 0.05. Adjusted for running 56 statistical tests (Benjamini–Hochberg), its adjusted q-value is 0.82. Every result is listed at the end of the article.

The rules Bayer trusted most

Bayer graded his rules. Some he called small, some “not quite as good” as others. A few he praised in terms no modern backtest report would use:

Rule 6 is his most famous call. Mars stood at 16°35′ Virgo on July 9–10, 1931; a year later, he notes, came “the enormous bear market low of stocks within a day,” the Dow’s low of July 8, 1932. The rule then steps back one sign each time Mars comes round, and Bayer reports it had “no miss whatever” from 1932 on.

Run backwards and forwards on the Dow, the record looks different. The rule’s 20 dates from 1897 to 1930 include none within a day of a swing low, where random dates would give about four. Of its five dates in 1932–1940, the years it was found, two sit near major lows. Of the 50 dates since the book was published, 9 were within a day of a swing low, where random dates give 8.9. The 1931 starting point was chosen because the 1932 low came a year later; after that, the rule is a random calendar.

Wheat, and the years he wrote in

Bayer warned that each market “moves its own way” and that his wheat rules shouldn’t be expected to work on hides. So the fairest test of the wheat rules is wheat. Our wheat data are Chicago (ZW) and Kansas City (KE) futures from 2000 to 2026. Across all 56 statistical tests we could score on wheat, 26 came out positive and one cleared p < 0.05: rule 1C, his rule for lard, at +4.1 points. Chance alone gives about 28 positive results and three below 0.05. Rule 6 picked lows in wheat less often than random dates (6.7% against 18.7%).

The hides rules, which he offers to stock traders, did no better on stocks: none of those we scored on the 481 US stocks cleared the bar there. And on the Dow during his own research years, 1932 to mid-1940, 22 of 49 date rules were positive and two cleared p < 0.05 (rules 2 and 13), about what luck gives. Neither held up after 1940.

The one rule that cleared the bar

Rule 30 says trend changes come when Venus’s declination passes 23°26′51″, the Sun’s greatest declination. Across all markets its dates were within a day of a turn 41.8% of the time, against 38.7% for random dates. Moving its dates by whole years, to the same calendar days in other years, never did as well in 69 tries, so it isn’t a seasonal effect. But the result doesn’t hold together:

The whole difference comes from 2000 onward. Before 2000 the rule did slightly worse than random dates, on the Dow before 1932 it did worse, in wheat it did nothing, and it shows no tendency to mark the larger turns Bayer was after. The estimated effect does not replicate across these periods and outcomes. With many rules, an isolated result needs independent confirmation.

Counting from tops and bottoms

Five rules count from a market’s own turning points. Rule 32 takes the Sun’s position at a major top or bottom, adds 126°30′ (“I dug out of the Hebrew Talmud”) and waits for Mercury to reach that point, months later; then the move that followed the top or bottom should repeat. We ran these from every major top and bottom of every market (the highest high or lowest low of 41 days), counting a projection only once the top or bottom was confirmed, and repeated each rule with made-up numbers.

The 126°30′ result is +0.7% of a typical 10-day move, about four hundredths of a percent for a stock that moves 2% a day, and two made-up angles did better. In wheat the rule went the wrong way. The other four rules sit within half a point of random dates, and Bayer’s numbers rank in the middle of the made-up ones.

Gaps, the 80-year cycle and the Mars “pictures”

• Gaps (rule 26). “A gap up is therefore bullish and NOT bearish.” Across 399,060 gaps, where a day traded entirely above or below the day before, the next day continued in the gap’s direction by 2% of a typical day’s move. Over the next five days the estimate was −0.5% of a typical five-day move, range −2.9% to +2.3%. For rule 9A, we read Bayer’s “a few days” as the crossing’s mapped trading session through three trading sessions later, mapping closed-market Mars full-hour crossings to the next session. Those gaps gave +2.8% of a typical five-day move [−3.2%, +10.9%], p 0.22. Daily currency bars from our data provider show “gaps” on 5–10% of days in a market that trades around the clock, and those keep going; we read that as a feature of how the bars are built and left it out.
• The 80-year cycle (rule 16). Bayer copied old prices from the Financial Chronicle in the New York library, expecting wheat in 1940 to move like wheat in 1860. The Dow is the one market we hold with 80 years to spare. Its monthly returns since 1977 correlate +0.05 with those of 80 years earlier, but 12 other gaps between 40 and 110 years correlate more strongly; the months moved the same way 53.2% of the time, against 52.0% for the typical gap.
• The Mars “pictures” (rules 7A and 8A). The market should repeat what it did when Mars stood 19h20m earlier in right ascension. Bayer calls 7A “the best of the lot,” but requires both dates to be prograde; 8A uses 11h20m as a fallback when 7A cannot be used. We compare five-session log returns. Across 761 markets, 7A’s correlation was −0.0001. The conditional 8A fallback gave −0.0287 across 613 markets; the combined procedure gave −0.0026 across 761.

  Mars can pass the same right ascension several times during a retrograde loop, and the book does not uniquely specify which earlier crossing to use or whether 8A permits a retrograde current date. We chose the latest earlier crossing and allowed current retrograde motion for 8A, requiring a prograde historical date. Skipping current retrograde dates instead gave −0.0001; requiring a unique historical crossing gave −0.0027. These are descriptive correlations, excluded from the 56 statistical tests. Parameter ranks are also descriptive comparisons, rather than null p-values.

Why the rules looked so convincing

Many of Bayer’s astronomical examples can be reconstructed. The trouble is how many dates there are. In 1939 alone his rules produce 417 dates on 230 different days.

Allow a day either side, as Bayer did, and 343 of the 365 days in 1939 have at least one rule behind them. Since 1962 the share is 97.4%, and 87.4% of days have two or more. That leaves many opportunities to match turning points after the fact. That is how a book of 61 rules can be full of good examples and still mark turns no more often than a calendar picked at random.

Bayer’s answer was confluence: trust the days when several rules agree. We counted, for every day since 1962, how many rules fall within a day of it, from none to 12. The 528 days with 8 to 12 rules were major turning points (the extreme of the 41 days around them) 0.5 points more often than random dates. Days with 10 or more rules did worse than their random-date comparison. Turn frequency did not rise steadily with the number of rules. These descriptive bins do not have an uncertainty test attached.

What this test can’t tell you

• It isn’t a test of Bayer the trader. He combined the rules with his ellipse, gaps, the shape of the chart and his own judgment, and he told readers to wait a day or two to see which way “the cat jumps.” Those choices can’t be coded. What we tested is whether the dates themselves carry information.
• Our wheat starts in 2000. The rules were built on 1930s wheat, which we don’t have, and our wheat series is a front-month continuous contract with jumps at each roll. The Dow covers 1896 onward, including Bayer’s own years.
• Some readings are ours. The examples helped choose mechanical readings, but rule 19 still fits poorly and the Mars “pictures” permit different interpretations during retrograde loops. We report those sensitivities; a different reading remains a different test.
• Daily bars, today’s symbols. Turning points are measured on daily highs and lows. The markets are those in our database today, so companies that failed or merged away are missing.

So do George Bayer’s rules work?

Most of Bayer’s printed dates can be reconstructed, with some uncertain readings. Across 764 markets and 130 years, including wheat and the years he wrote in, we found no consistent advantage over random dates in turning-point frequency or directional returns. The one rule that cleared the unadjusted bar did not survive the correction for 56 statistical tests.

What the book shows instead is how convincing a set of rules can look when it marks nearly every day of the year. Testing against random dates, not against remembered examples, is the part that was missing in 1940.

Every result

All 764 markets, every date from 1896 to October 2026. Turning points: within a day of a swing high or low (or, for the top and bottom rules, of that kind of swing). Directions: average move the way Bayer says over his number of days; negative means the other way. The primary directional tests express moves as a percentage of each market’s typical move over that horizon. A separate descriptive table shows the same dates in unscaled log-return percentages, with its own confidence ranges and p-values.

Common questions