Inconsistent
Moon phases: the sign flipped; not confirmed
The idea
Two published papers report higher stock returns around new moons than around full moons (Dichev and Janes, Journal of Private Equity, 2003; Yuan, Zheng and Zhu, Journal of Empirical Finance, 2006). The test: are average SPY returns higher around new moons than around full moons? We computed phases from Meeus, Astronomical Algorithms (2nd ed., 1998).
What happened
| Average daily excess return | New moon | Full moon | Result |
|---|---|---|---|
| 2017–2021 | 0.063% | 0.069% | Fail (wrong sign) |
| 2022–2023 | 0.045% | −0.025% | Pass |
An executable version (hold SPY at 75% on the favorable days) returned a Sharpe of 0.42 in 2017–2021 and 0.17 in 2022–2023.
The official runs
This idea was judged on a sign test, not on a book: is SPY's average daily excess return (over T-bills) higher in the 15-day new-moon windows than in the 15-day full-moon windows? It is one pre-registered test per window, not a search. Here are the numbers behind it, counting trading days in each phase window. Phases are computed from Meeus, as on the rest of this page.
| Average daily excess return, SPY | New moon | Full moon | Result |
|---|---|---|---|
| 2005–2016 (build window, context only) | 0.043% (n=1,541) | 0.026% (n=1,446) | Context only |
| 2017–2021 | 0.063% (n=628) | 0.069% (n=600) | Fail (wrong sign) |
| 2022–2023 | 0.045% (n=246) | −0.025% (n=230) | Pass |
The 2005–2016 row is context only; no pass or fail was attached to it. The test passed in 2022–2023 and failed in 2017–2021, and a rule with no real effect will pass any single window about half the time. The handful of "other" sessions that fall in neither window (5 to 33 per window) are not shown.
The executable version is reported alongside, but it was not the pass test. Idle cash credits the 3-month T-bill rate, worst drop is from the simulated book, and the same-days SPY row is frictionless.
| 2017–2021 | Sharpe | Yearly return | Worst drop | Time invested |
|---|---|---|---|---|
| Executable book: SPY at 75% in new-moon windows Reported, not the pass test | 0.42 | 4.52% | 16.6% | 38% |
| SPY, same invested days Reference | 0.61 | 6.28% | 15.8% | 38% |
| SPY, fully invested Reference | 0.92 | 18.28% | 33.7% | 100% |
| 2022–2023 | Sharpe | Yearly return | Worst drop | Time invested |
|---|---|---|---|---|
| Executable book: SPY at 75% in new-moon windows Reported, not the pass test | 0.17 | 5.03% | 10.8% | 38% |
| SPY, same invested days Reference | −0.17 | 1.48% | 11.0% | 38% |
| SPY, fully invested Reference | 0.12 | 4.26% | 22.1% | 100% |
We did not chart an equity curve for this one. The pass test was the sign comparison above, and a growth line for the 75% book would invite a reading the test never made.
Why it failed
The sign flipped between windows. It was one pre-registration test per window, not a search, and a rule with no real effect will pass any given window about half the time. We parked lunar, Gann and Bradley-style timing as a family.
Receipts: the test was pre-registered before the run.