Inconclusive · Filters · 2026-10-11
Do harmonics fail when a stock's recent daily returns have been trending (positive autocorrelation)?
What this means
This fits the owner's intuition that harmonics work when the stock is behaving in a mean-reverting way. The size is modest: skipping the top tercile would lift the edge by ~0.03-0.09R per setup and cost a third of the setups.
Explanation checked against the data.
- Hypothesis
- Harmonic entries are a mean-reversion bet, so they should beat random entries less when the stock's prior-60-day lag-1 return autocorrelation is high.
- Method
- Part of hidden-variable-hunt (pre-registered). ac1_60 is measured at the day before entry. Terciles with cuts of about -0.08 to -0.10 and +0.01 to +0.03. Edge vs fixed control (d), state-matched randoms, and the original control.
- Result
- Edge low / mid / high tercile: 1d +0.08 / +0.08 / -0.02; 1h +0.08 / +0.09 / -0.02; 4h +0.12 / +0.07 / -0.15. Harmonic R falls across terciles (4h +0.25 -> -0.14) while random R barely moves. Delta top-bottom: 4h q=0.034; 1h and 1d p~0.09 (q 0.21). Same sign in both halves, in both variants and with the original control.
- How much to trust it
- Low: Consistent in sign everywhere, but it clears multiple-testing correction on 4h only (n=735); the longer-window measures (ac1_250, variance ratio) are null
- Did it change the live rule?
- candidate filter for the ~2027-04 forward review: skip setups in the top ac1_60 tercile
- Notes
- Needs an out-of-sample check on new trades before any rule change.
Scripts and data for this study will be linked here when the research code is published on GitHub.
Education and research only. Not financial advice. Backtests use today's index members (survivorship bias).