Polynomial Regression Band

Polynomial Regression Band fits a degree-2 polynomial to the last 100 bars by least squares and offsets the fit by a multiple of its residual standard deviation, for a channel that curves with the trend rather than averaging it into a straight line. This definitive clean build of the polynomial regression band separates the tool's two versions: a causal end-point line keeping only the newest fitted value per bar — never redrawing, the series every alert measures against — and the current window's full curve, which openly re-fits on every bar.

How to Trade the Polynomial Regression Band?

• Slope turns: the end-point line flipping direction is the trend read; both turns fire alerts.
• Band pullbacks: a wick into the lower band while the fit rises (or the upper while it falls) is the pullback watched as curved dynamic support or resistance.
• Band closes and re-entries: a close beyond a band marks stretch — a context read, since strong trends can ride a band — and the close back inside frames mean reversion toward the fit.

Only the end-point series suits rule testing — a live alert matches its plotted history. The curve is context, deliberately not extended past the last bar: extrapolation is unreliable.

Polynomial Regression Band Settings

• Source (default close): series the polynomial is fitted to.
• Length (default 100): the rolling fit window — shorter hugs recent swings, longer smooths them.
• Degree (default 2): the fit's polynomial degree.
• Band Multiplier (default 2): half-width of the channel in residual standard deviations.
• End-Point Band (default enabled): the causal line and its bands.
• Fitted Curve (Repaints) (default enabled): the current window's curved channel.
• Gradient Fill (default enabled): shading from the fit to each band; remaining Style inputs set colors.

Frequently Asked Questions

How is this different from Linear Regression?

A straight-line fit like Linear Regression assumes one constant direction across the window, so it lags through rounded turns. From degree 2 upward the fit can curve inside the window — degree 1 would simply reproduce the linear case.

Why does the fitted curve move on every bar?

Each new bar re-fits the polynomial over the newest window, so the drawn segment shifts as data arrives — the nature of any windowed fit. Hence the end-point series, and why alerts are evaluated against it.

What degree should I use?

The default of 2, or 3, covers most charting uses: enough curvature for rounded trends without wild oscillation. Raise it only when the data clearly bends more — the cost is noise-fitting and unstable behavior near the window's edges.