Why volatility clusters
Financial volatility is driven by information arrival, and information does not arrive at constant rates — major news creates uncertainty that triggers buying and selling, which itself creates further price moves and further reaction. Leverage also amplifies clustering: falling asset prices can force margin calls and deleveraging, creating additional selling pressure independent of new information. Uncertainty itself persists — when a crisis begins, uncertainty about the magnitude and duration keeps volatility elevated until resolution. Herd behaviour causes cascades. The practical result: conditional volatility (today’s expected volatility given recent history) is predictable in a way that unconditional (average) volatility is not.
The GARCH model
The GARCH(1,1) model is the workhorse specification. It models conditional variance σ²ₜ (today’s variance given available information) as: σ²ₜ = ω + αϵ²ₜ−₁ + βσ²ₜ−₁. In plain English: today’s expected variance equals a constant (ω, the long-run variance), plus α times yesterday’s squared return (the "news" component — large shocks drive up future volatility), plus β times yesterday’s variance (the "persistence" component — high volatility tends to stay high). In practice, α + β typically sums to 0.97–0.99 for equity markets, implying high persistence — volatility shocks decay slowly toward the long-run mean over weeks to months. GARCH forecasts are used extensively in options pricing (dynamic volatility models), risk management (VaR calculation), and portfolio construction.
“Volatility, like weather, is not predictable in its timing — but the persistence of volatility regimes is among the most reliable patterns in all of financial data.”
What this means for you
Volatility clustering has immediate practical implications. Option pricing: models assuming constant volatility (Black-Scholes) systematically misprice options during volatile regimes. Risk management: VaR models that assume i.i.d. returns understate risk after large shocks — when you most need risk models to be conservative, historical-simulation VaR calibrated to recent calm history will be worst. Practical rule of thumb: when market volatility has been elevated recently, expect it to remain elevated — and size positions accordingly. The VIX (implied volatility index) in options markets reflects this clustering: when VIX spikes, mean reversion suggests it will eventually fall, but the timing is uncertain and the spike may persist for months.