Not supported · Market regime · 2026-10-11
Is there a hidden state that makes harmonics work at some times and not others?
What this means
No hidden switch turns harmonics from no-edge into strong-edge. The apparent 'harmonics work in down-trending stocks' effect comes from how the time-local control is built. The one pattern-specific signal is small: harmonics underperform when a stock's recent daily returns have been positively autocorrelated (trending).
Explanation checked against the data.
- Hypothesis
- Pre-registered: harmonic edge over matched random entries differs by stock mean-reversion tendency, trend context, volatility state, market state, sector or size, all measured at entry.
- Method
- 14 state variables at the last daily bar before entry: ac1_60, ac1_250, vr5_250, ma50_dist, above_ma200, mom6, mom12, rv_rel, rv_expanding, spy_above_ma200, spy_rv20 (VIX not cached), spy_er20, sector, 20-day median dollar volume. Terciles or binary splits. Delta = edge in top bucket minus bottom bucket, using the fixed control (d) and state-matched randoms. 81 tests, ticker-clustered bootstrap, BH q-values. A lead needs q<0.10, the same sign in both halves, and support on another TF.
- Result
- 21 tests had q<0.10. The trend variables (ma50_dist, above_ma200, mom6) met the formal rule on all TFs under control (d), with q 0.003-0.034. Per bucket, though, harmonic R is flat and only the random leg moves (e.g. 1d below MA50: random -0.04; furthest above: +0.50). State-matched, daily flips sign (+0.27) and 4h is ~0, so this is a control artefact. ac1_60 (60-day lag-1 autocorrelation) is the one genuine lead: 4h Delta -0.27 (-0.48,-0.06), q=0.034; 1h -0.10 (p 0.085), 1d -0.10 (p 0.089); same sign in both halves, in both variants and with the original control. Sector, size, volatility and market state: nothing.
- How much to trust it
- Medium: Pre-registered, BH-corrected, clustered CIs; the main lead clears BH on 4h only
- Did it change the live rule?
- none; ac1_60 added to the April 2027 forward-review list
- Notes
- 1h spy_rv20 (higher market vol -> better edge, +0.19, q=0.003) had no support on other TFs. 1h harmonics did better in weak-momentum stocks even state-matched (-0.16), but daily is opposite.
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).