Raff Regression Channel

The channel is named for Gilbert Raff, who set out the construction in his work on trend analysis in the early 1990s. It starts from the same least-squares line as an ordinary regression channel, then departs from it in one decisive way: instead of offsetting the fit by a measure of average dispersion, each boundary is pushed out to thesingle furthest bar on that side.

The consequence is that the channel is guaranteed to contain every bar in the sample, and that its two boundaries are almost never equidistant from the fit. That asymmetry is not a flaw to be corrected — it is information. A channel whose upper boundary sits much further from the fit than its lower one is describing a trend whose excursions have been predominantly upward, which is a different market from one with symmetric excursions.

Because the boundaries are defined by extremes rather than by an average, the channel makes a different kind of claim than a standard-deviation channel does. A ±2σ channel says “most bars fall in here”; a Raff channel says “all bars in this sample fell in here, and here is exactly how far the worst one on each side went.” The first is a statistical statement about a distribution, the second a factual statement about a specific stretch of price.

The tool takes 2 clicks, which select a time range. It fits a least-squares line through the chosen price source over the bars in that range, finds the largest residual above the line and the largest below, and places a boundary at each. A minimum of three bars is needed before a fit exists — two points define a line exactly, so their residuals are zero and there is no channel.

The anchors select a sample; they do not shape the channel. Only the time of each anchor matters, which is why both grips use a horizontal resize cursor. Dragging one changes which bars are fitted, not where the lines go — the lines go where the data puts them. There is no width grip, because the width is a property of the residuals and there is nothing for a drag to write.

Internally this is the same module as the Regression Trend, differing only in how the band distance is measured — Raff uses max deviation, where the standard channel uses standard deviation. The Band basis control in the properties panel exposes this, so you can switch a Raff channel to either of the other measures without redrawing it. It is a separate palette entry rather than a setting because the asymmetric, all-bars-enclosed construction is a distinct analytical object, and one users should not have to know to go looking for.

The sample is resolved by time, so panning and zooming cannot change which bars are included. Each boundary end is converted from its own price independently, so the offset stays a constant distance in price rather than in pixels — which matters on a logarithmic axis.

Boundary rejection: because each boundary touched exactly one bar in the sample, it is by definition a level price reached and then retreated from. A return to it is a test of the most extreme behaviour the trend has shown so far, which makes it a defensible entry for a trade back toward the fit — with a stop just beyond, since a break means the trend is doing something it has not done in the sample.

Breakout confirmation: the enclosure property gives the break a clean meaning. Price closing outside a Raff channel is not merely unusual; it is unprecedented within the selected sample. That is a stronger signal than exceeding a standard-deviation band, which bars do routinely.

Reading the asymmetry: compare the two offsets. A much larger upper offset in a rising channel means the advances have overshot the trend more than the pullbacks have undershot it — buyers pressing hard, with shallow corrections. The reverse pattern in a rising channel — a large lower offset — is a warning that the pullbacks are getting deeper than the rallies are strong.

Trend quality filter: use Pearson r to decide whether to trade the channel at all. Above roughly 0.9 the trend is orderly and the boundaries are meaningful. Below about 0.6 the line is fitting noise, and a channel drawn around noise will contain every bar without saying anything useful.

1. 1
   Pick one trend

   Click the start and end of a single directional move. Only the times matter. Spanning a reversal produces a flat fit with huge deviations and an r near zero.
2. 2
   Check r before anything else

   If Pearson r is weak, the channel is describing noise. Adjust the anchors to isolate a cleaner stretch rather than trying to interpret the boundaries.
3. 3
   Read the asymmetry

   Look at where each boundary touches. Those two bars are the trend’s most extreme moments, and the difference in their distances is the channel’s main insight.
4. 4
   Leave Deviation at 1

   That is the Raff construction. Raising it pads the channel and breaks the enclosure property.
5. 5
   Extend to project

   Set Extend to Right to carry the boundaries forward as levels for the next test, remembering they are extrapolated from the selected bars only.

• Every bar inside the channel is guaranteed, not observed — it is how the boundaries were constructed. So containment is not evidence the channel is working.
• The bar that touches each boundary is the informative part. Check when it printed: an extreme set at the start of the sample describes a trend that has since settled, one set at the end describes a trend still pushing.
• A very wide channel with a strong r means an orderly trend with large swings around it; a narrow one with strong r means a tight, disciplined trend.
• Adding bars to the sample can only widen the channel or leave it unchanged, never narrow it, because a new extreme replaces the old one and a lesser bar does not. Widening on extension is therefore expected and not a signal.
• A single outlier bar — a gap or a spike — sets a boundary the rest of the sample never approached. This construction is more sensitive to outliers than any other channel here, by design.

Raff against standard deviation on the same sample: draw both over identical anchors. The gap between the Raff boundary and the 2σ boundary measures how much of the channel’s width comes from outliers rather than from sustained dispersion. A large gap says one or two bars are doing all the work.

Switching Band basis to standard error: standard error divides by the square root of the sample size, so it tightens as the sample grows. It describes confidence in thetrend line rather than the spread of price around it, which is the right measure when the question is “how well do I know this slope?” rather than “how far does price stray?”

Sequential channels: draw Raff channels over consecutive equal-length stretches of the same trend. Rising r values mean the trend is becoming more orderly; falling ones mean it is deteriorating, often before the price action makes that obvious.

• 1.Trading the Regression Channel — Gilbert Raff (1991)
• 2.Raff Regression Channel — ChartSchool — StockCharts
• 3.Regression trend drawing tool — TradingView Help
• 4.Linear Regression — Investopedia
• 5.Technical Analysis of the Financial Markets — John J. Murphy
• 6.Statistics for Traders — Perry J. Kaufman, Trading Systems and Methods