Dual Regression Channel

Every other fitted channel in this family computes a single trend line and then offsets it — by a standard deviation, by the extreme residuals, by an ATR multiple. All three are symmetric about one slope, which means their boundaries are always parallel. There is only one slope in the model, so a market whose highs and lows are moving at different rates cannot be represented at all.

That is exactly the market this tool exists for. Triangles, wedges and broadening formations are all defined by the upper and lower envelopes having different slopes. Fitting each boundary separately is the only way to measure that difference rather than assume it away, and the difference is the number worth having: the gap between the two slopes is how fast the range is closing or opening, per bar.

The construction is a straightforward application of ordinary least squares to two series instead of one. Fitting a line to swing highs and another to swing lows is what a chartist does by hand when drawing a wedge; doing it by least squares removes the discretion about which pivots to honour, and produces a slope for each envelope rather than a drawn impression of one. Classical pattern analysis — Schabacker, Edwards and Magee,Bulkowski — catalogues the resulting shapes extensively; what it does not supply is a rate of convergence, which is what this measures.

The tool takes 2 clicks, which select a time range. It then fits one least-squares line through the highs of the bars in range and another through the lows, and draws each as a boundary. The centre line, when shown, is the midline between them.

The minimum sample is five bars, higher than the three the single-fit channels require. Two lines are being fitted here, and a channel whose slopes differ because four bars happened to fall that way is measuring noise rather than a structure. Below five bars the tool draws a dashed placement guide instead of a channel.

Highs and lows are not configurable. There is no price-source control, because the tool measures the range envelope and fitting both boundaries to the same series would collapse them onto one line. This is the one fitted channel in the family without a Source setting, and the omission is deliberate.

The anchors select a sample; they do not shape the channel. Only the time of each anchor matters, hence the horizontal resize cursor on both grips. Dragging one changes which bars are fitted, not where the lines go. There is no width grip, because both boundaries come from the data.

The centre line is taken as the midpoint between the two boundaries rather than as a third fit through the closes. A fit to closes would be a different line that merely happens to sit nearby, and the midline is what actually bisects the channel when the boundaries have different slopes.

Wedge and triangle breakout: a converging channel is compression, and compression resolves. Trade the break of whichever boundary gives way, with the stop on the far side of the channel and a measured-move target from the channel’s width at its widest point. TheWidth readout gives that starting height directly.

Timing from the apex estimate: the apex figure is the number of bars until the boundaries meet at the current rates. Structures usually resolve well before reaching the apex — commonly somewhere in the last third of the way there — so the estimate is most useful as a window rather than a date. A compression with a hundred bars to its apex is not yet urgent; one with eight is.

Diverging channels (broadening formations): when the width is increasing, volatility is expanding and both boundaries are moving away from price. These are poor mean-reversion environments because the target keeps receding, and they usually signal indecision rather than trend. Reduce size rather than trying to trade the edges.

Reading the two correlations separately: this is the tool’s most distinctive output. A strong r highs with a weak r lows means resistance is behaving in an orderly, trendable way while support is erratic — so the upper boundary is tradeable and the lower is not. Acting on the strong side only is a more defensible use of the drawing than treating both edges as equivalent.

1. 1
   Select a single structure

   Click the start and end of one consolidation or trend. Only the times matter. Spanning two structures produces two meaningless fits.
2. 2
   Give it enough bars

   Five is the minimum; a few dozen is where the slopes become trustworthy. If nothing draws, the selection is too short.
3. 3
   Read the shape

   Check whether the readout says converging, diverging or parallel. That is the tool’s primary answer, and it is often not the one your eye gave.
4. 4
   Check both correlations

   Weak on both means there is no structure to measure. Strong on one and weak on the other means only one boundary is worth trading.
5. 5
   Extend right

   Carry the boundaries forward to see the convergence, and read the apex estimate for a rough resolution window.

• The readout only calls a channel converging or diverging once the width change over the sample is material relative to the current width. Small slope differences on noisy data are reported as parallel, which is the honest answer.
• A converging channel with two strong correlations is a textbook compression and the highest-quality signal this tool produces.
• Convergence with weak correlations is a coincidence of the sample, not a structure. Extend or move the anchors and it will usually vanish.
• The apex estimate is a linear extrapolation of two fitted slopes and nothing more. It describes the sample, not the future.
• The boundaries are fitted to every bar’s high and low, not to swing pivots. So a fitted boundary usually sits inside the extreme wicks rather than touching them — unlike a hand-drawn wedge, and unlike the Raff channel, which is built to touch.
• Upper and lower styling follows screen position, so if a converging channel is extended past its apex the boundaries cross and the two styles swap.

Against a Raff channel on the same bars: Raff’s boundaries touch the two most extreme bars and are always parallel; these are fitted to all the highs and all the lows and need not be. Drawing both shows how much of the apparent range is set by outliers — if the Raff channel is far wider, one or two bars are doing the work and the convergence measured here is the more representative figure.

Confirming a pattern call: if you believe you are looking at an ascending triangle, the upper slope should be near zero and the lower clearly positive. The two slopes turn a visual pattern claim into something checkable, and frequently disconfirm it — many “triangles” turn out to have two sloping boundaries and be wedges instead.

Rolling the sample forward: draw the channel over one window, then move both anchors forward together by a fixed amount. A convergence rate that is increasing means compression is accelerating and resolution is closer than the raw apex figure suggests.

Volume alongside convergence: genuine compression is usually accompanied by contracting volume, and the break by expanding volume. A channel that converges while volume stays flat or rises is more likely two unrelated trends being fitted than a real compression.

• 1.Encyclopedia of Chart Patterns — Thomas N. Bulkowski
• 2.Technical Analysis of Stock Trends — Robert D. Edwards & John Magee
• 3.Technical Analysis and Stock Market Profits — Richard Schabacker (1932)
• 4.Linear Regression — Investopedia
• 5.Triangle Patterns — A Complete Guide — Investopedia
• 6.Trading Systems and Methods — Perry J. Kaufman