Volatility · 7 min read
GARCH Volatility Forecasting, Explained Simply
Markets do not deliver risk evenly: quiet stretches and violent ones cluster together. GARCH is the workhorse model that turns that clustering into a forward volatility estimate.
Markets do not hand out risk in even doses. Calm stretches are followed by more calm; a violent day tends to be followed by more violent days. This pattern is called volatility clustering, and GARCH is the model built to capture it.
GARCH stands for Generalised Autoregressive Conditional Heteroskedasticity — a mouthful that simply means "a model where today’s expected volatility depends on recent shocks and recent volatility." The everyday version: turbulence is sticky.
Why a single volatility number is not enough
The naive approach is to take the standard deviation of the last year of returns and call that your risk. The problem: it weights a sleepy day twelve months ago exactly the same as yesterday’s crash. It is slow to react when conditions change and slow to relax when they calm down.
GARCH fixes this by letting recent information matter more. After a big move, its volatility estimate rises quickly; through a quiet stretch, it decays back down. The result is a forecast that tracks the market’s actual mood rather than a stale average.
How GARCH(1,1) thinks
The workhorse version, GARCH(1,1), builds tomorrow’s variance from three pieces: a long-run baseline the market drifts back toward, the size of the most recent shock, and yesterday’s volatility level. The balance between those last two controls how fast volatility spikes and how slowly it fades.
What investors get from it
- A forward-looking volatility estimate instead of a backward-looking average.
- Sharper risk measures — feeding GARCH volatility into VaR and CVaR makes them react to regime shifts.
- Earlier warning when a calm market starts to destabilise, before the headline numbers move.
GARCH is not a crystal ball. It cannot predict the direction of returns, only the magnitude of expected swings — and it can be caught out by a shock with no recent precedent. But as a forward volatility estimate, it is dramatically better than a flat historical average.
Key takeaways
- Volatility clusters: calm follows calm, chaos follows chaos.
- A flat historical standard deviation reacts too slowly to regime changes.
- GARCH(1,1) weights recent shocks and recent volatility to forecast the next move’s size.
- It sharpens VaR and CVaR but forecasts magnitude, not direction.
See GARCH-driven volatility on your own holdings — run the free diagnostic.