Inconclusive · Entries · 2026-10-11 (updated with the price-blind reference)
Does a harmonic pattern add anything beyond buying a 15% pullback?
What this means
On daily, a dumb 'buy a 15% dip' rule earns the same R per trade as harmonics and beats a price-blind random entry by +0.16R, while harmonics beat it by nothing. So on daily the pattern adds nothing beyond buying pullbacks, and the plain dip rule does the pullback job better. On 1h it is the other way round: the dip rule is no better than random, and harmonics are +0.06R better than random and +0.09R better than the dip rule. On 1h the pattern seems to time the bounce rather than just find the dip. 4h is too small to tell.
Explanation checked against the data.
- Hypothesis
- Pre-registered: a mechanical dip-buy (first close >=15% below the 60-day high after a reset to within 5%) with the same 2R exit earns as much as harmonic entries, and harmonics show no edge outside pullbacks.
- Method
- Same tickers and periods as the harmonic samples. Dip entry at the next open, median harmonic stop % per TF (1d 8.15%, 1h 3.54%, 4h 6.00%), 2R target, full exit, sim_many; intraday windows scaled (60 days = 420 1h / 120 4h bars). Head-to-head mean R with ticker-clustered CI; each vs a no-active-setup control (g) (+-250 days, excluding dip and harmonic spans). Harmonic trades split by coincidence with a dip trigger (+-5 days) and by being in a >=15% pullback at entry. Post-hoc: in-pullback harmonics vs random entries in the same pullback state. Update: both legs also compared against price-blind randoms (h): same ticker, same calendar year as the entry, no exclusions, the leg's own stop %/2R; (h2) same quarter as a sensitivity.
- Result
- Head-to-head harmonic vs dip R: 1d +0.256 vs +0.231, diff +0.03 (-0.04,+0.09); 1h +0.144 vs +0.058, +0.09 (+0.01,+0.16); 4h +0.078 vs +0.099, -0.02 (-0.14,+0.09). PRIMARY edge vs (h): harmonic -0.01 (-0.07,+0.05) / +0.06 (+0.01,+0.11) / -0.04 (-0.14,+0.05); dip +0.16 (+0.13,+0.19) / +0.00 (-0.05,+0.06) / +0.06 (-0.01,+0.12). Sensitivities: vs (h2), dip +0.36 / +0.14 / +0.23 (flattered: dips cluster in crash quarters); vs (g), harmonic -0.06 / +0.02 / -0.09 and dip -0.09 / -0.26 / -0.26. Harmonic in a >=15% pullback at entry vs same-state randoms (post-hoc): 1d -0.13 (-0.33,+0.09), 1h +0.11 (-0.00,+0.23), 4h +0.16 (-0.08,+0.44).
- How much to trust it
- Medium: Pre-registered, one fixed dip rule, clustered CIs; the pre-registered +-5-day coincidence split leaked look-ahead (dip triggers after the fill = harmonic stopped out), and the cleaner pullback-state splits are partly post-hoc
- Did it change the live rule?
- none; 'in a >=15% pullback from the 60-day high at entry' added as a candidate intraday filter for the ~2027-04 forward review
- Notes
- The pre-registered decision rule passes on 1h only. The first 15% close is a falling knife: the dip rule loses to the no-pattern control by 0.09-0.26R. The earlier 'dip proxy beats the +-60 control by +0.28-0.38R' (survivorship-pit-check) compared against a window that contains the decline, so it is not comparable.
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).