For each instrument, fit four log-regression channels at horizons
63 / 126 / 252 / 504 trading days. Each channel's lower band
— placed at K = 2 residual standard deviations below the fitted line —
is a candidate support level. Lower bands within
2% of each other merge into one confluence point.
Two or more horizons agreeing on the same price = high-probability support;
four horizons agreeing = a structural floor.
2026-05-12as of
12instruments
6strong (≥ 3-horizon)
2weak (single-horizon)
The rule
A log-linear regression fits a straight line to log(close) over a
rolling window. The lower channel boundary is the fitted line minus
K standard deviations of the in-sample residuals, then exponentiated
back to price space. Practically: the level below which roughly 2.5%
of bars in the window have closed.
Running this at four horizons — 63, 126, 252, 504 bars,
approximately 3 months / 6 months / 1 year / 2 years — produces four
candidate supports. When two or more of those land on top of each other
(within 2% of each other), the read is that the price
level is significant across multiple time scales. That is the
confluence concept: structural levels show up on multiple lenses.
Output sort: instruments where four horizons converge at one level
appear first; instruments with only a single channel band visible
rank last. Distance from current price tells you how far the floor sits below today.
The four lower bands and their distance from current price, per instrument.
Where multiple rows on the same instrument share the same price, those are
the cluster members. Where they're spread out, the instrument has been
trending hard enough that the longer-horizon channel sits well below the
shorter ones — the regression hasn't kept up.
Support Confluence is a static read — it tells you where bands sit today.
Combined with Sector RS (which sectors are leading vs. lagging) and the
Closelook Pattern Engine, it sharpens the entry-zone question: if a
leading sector pulls back, where would a multi-horizon support cluster
catch it?