GARCH-family Clustering

GARCH-family Clustering runs GARCH(1,1), GJR-GARCH(1,1) or EGARCH(1,1) in a pane, plotting conditional volatility with its long-run anchor, a mean-reverting forecast path and the active coefficients. The recursions, variance-targeting anchor and quasi-MLE grid fit follow the published GARCH-family clustering models; GJR and EGARCH add the leverage effect where sell-offs expand volatility more than rallies.

How to Trade the GARCH-family Clustering?

• Stressed regime: conditional volatility crossing above the long-run anchor (alertable) opens a phase where elevated volatility tends to persist — widen stops, trim size.
• Calm regime: decay back below the anchor restores compression conditions.
• Volatility shocks: a return beyond the sigma threshold marks the surprises whose aftermath the persistence term stretches out.
• Forecast path: the dotted projection decays toward the anchor at the persistence rate — a horizon read for sizing decisions.

GARCH-family Clustering Settings

• Model (default GARCH(1,1)): the symmetric workhorse, or GJR / EGARCH for leverage effects.
• Source (default close) and Window Length (default 250): the sample behind the anchor and the fit.
• Annualize (default enabled) with Bars Per Year (default 252): comparability across instruments.
• Parameters (default Fitted (rolling quasi-MLE grid)) with Refit Every (bars) (500); Alpha - News (0.10), Beta - Persistence (0.85) and Gamma - Asymmetry (0.05) drive Manual mode and seed the recursion until the first fit.
• Show Forecast (on) with Horizon (20); EWMA Benchmark (off) with Lambda (0.94); Shock Threshold (sigmas) (2.0) and Mark Shocks (off).
• Show Dashboard (on); Regime Shading (default Stressed), colors and Gradient Fill are style options.

Frequently Asked Questions

How is this different from EWMA volatility?

EWMA Volatility is the limiting case — persistence pinned at one, with no anchor to revert to — so shocks never fully decay. Overlay the built-in benchmark and watch how fast the GARCH line peels away after a shock: the quicker the separation, the stronger the anchor's pull.

Should I use fitted or manual parameters?

Fitted re-estimates the coefficients on a coarse likelihood grid at each refit interval. Manual replicates published estimates or stress-tests persistence, and it also runs the recursion until the first fit completes.

What do persistence and half-life mean in the dashboard?

Persistence is the combined weight of news and memory; the closer to one, the longer volatility episodes last. Half-life converts that into bars until half a shock has decayed.