Distribution-Free Price Channels with IRLS Quantile Regression
Summary
This article builds a price channel from independently fitted conditional quantile lines rather than a mean and standard deviation. It fits lower, median and upper quantiles with pinball loss, using iteratively reweighted least squares to solve for each line. The resulting channel can be asymmetric and is less sensitive to extreme residuals than a squared-error fit. The author describes implementation details, including residual-weight safeguards, relative convergence checks, and indicators that display the lines and channel-width measures.
Validation examines whether observed prices fall within the stated quantile proportions, compares channel width with Bollinger, ATR and regression-based widths, and tests a deliberately simple trading system across instruments. The article reports that the edges should not be read as forecasts or entry signals: a naive strategy lost in most tested configurations, and quantile coverage does not translate into reliable bounds for the next bar. Width behaves as a volatility feature, but horizon-matched ATR often performs better. The author also notes that separately fitted edges may cross, and that convergence and coverage can vary by instrument and period.
Key ideas
- Pinball loss fits a chosen conditional quantile without assuming normally distributed residuals.
- IRLS approximates the quantile fit through repeated weighted least-squares solves.
- Independent upper and lower fits allow asymmetric channels but can also cross.
- The article tests quantile coverage and compares normalized channel width with other volatility measures.
- Its reported results argue against using channel edges as entry triggers or probabilistic bounds on the next bar.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.