— —
The Moon
Rendered from JPL DE440s. The terminator is not drawn, the light sits along the true Moon-to-Sun vector and the camera along the true Moon-to-Earth vector, so the phase is a consequence of the geometry rather than a parameter. Libration is applied as a real rotation, which is why the limb detail shifts over a month. The surface is NASA LROC imagery over the LOLA elevation model.
The six figures above follow the date slider. The next new and full moon timestamps do not, they are always the next ones from today, since a “next full moon” relative to a date two years ago is a fact about the past dressed up as a forecast.
Distance sits at the 23th percentile of the perigee–apogee range (356,500–406,700 km). The next new and full moons are found by searching for the elongation crossing, not by adding 29.53 days to the last one, that accumulates error.
Scrubbing back
The slider under the render moves through four years of daily geometry, two back and two forward. Every frame is an ephemeris lookup rather than an interpolation, so a date last spring shows the Moon as it actually was, including the libration that had rolled a different stretch of the limb into view. Run a month to watch a full lunation, and the wobble is obvious: the face is not fixed.
Why this page exists
Lunar phase is a real, published hypothesis. Yuan and Zheng's Are Investors Moonstruck? reported lower equity returns around the full moon across a large sample of countries, and it is a paper this study replicates against rather than dismisses. It sits in Tier 2 of the hypothesis list: cyclical, well-documented, and with no mechanism anyone has been able to state.
The rendering above is careful for the same reason the statistics below will be. A page that got the astronomy wrong while claiming to test it rigorously would be making the opposite point to the intended one.
The test
Not yet run
The harness has not been built, and until it is there is nothing honest to put here. What will appear, once it runs: returns bucketed by the eight lunar phases, the effect size for each bucket against its complement, a bootstrap confidence interval, the raw p-value, and the same value after correction across the full family of roughly thirty hypotheses.
Beside it, the number that usually matters more, the minimum detectable effect at 80% power given the observations that exist. If detecting a plausible lunar effect would need more market history than has ever been recorded. That is the finding, and it gets stated as plainly as any rejection would.
Inputs are already live: 14,011 daily index observations from 1971, and the phase for every one of those days computed from the same ephemeris driving the render above.
If a result does appear above the corrected threshold, read the Bangladesh line before believing it.