Generalized Autoregressive Conditional Heteroskedasticity
Read the paperopens doi.org in a new tab
What they found
Bollerslev extended ARCH by letting today's variance depend on yesterday's variance as well as yesterday's squared shock, which produces smooth, persistent volatility with far fewer parameters. The GARCH(1,1) model became the workhorse of financial volatility modeling: it captures volatility clustering and mean reversion, produces reasonable multi-day forecasts, and is simple to estimate. Almost every risk system, from value-at-risk to options desks' realized-volatility forecasts, uses GARCH or a descendant.
What you can use
- GARCH(1,1) captures the two key facts of market volatility: it clusters, and it mean-reverts to a long-run level.
- After a volatility spike, expect volatility to decay gradually, not snap back; the model's persistence parameter tells you how slowly.
- A simple exponentially weighted volatility estimate is a special case of GARCH and is good enough for most position sizing.
Caveats
Symmetric: the basic GARCH ignores that down moves raise volatility more than up moves (see Bekaert-Wu). Econometrics paper with no trading application.
Tags: volatility, garch, econometrics, forecasting
Summaries are our own reading of the paper, not the authors' words. Educational only, not advice. Discuss it in Book Club.