The Volatility Model the Whole Industry Actually Uses: GARCH

Sometimes the most important paper isn't the one with the big new idea, it's the one that makes the big idea practical. Robert Engle's ARCH model (see our explainer on it) proved that market jumpiness clusters and can be forecast. But the original recipe was clunky. Four years later, a graduate student named Tim Bollerslev added one small ingredient that made the whole thing lightweight, elegant, and easy to use. The result, called GARCH, became the model that risk desks, options traders, and academics reach for to this day.

This is the story of a tweak that beat the original.

The problem GARCH was fixing

Quick recap of the setup. Volatility, how much prices bounce around, isn't constant. It clusters: stormy periods follow stormy periods, calm follows calm. Engle's ARCH model captured this by forecasting tomorrow's jumpiness from the size of recent surprise moves.

The catch: volatility in real markets can stay elevated for a long time after a shock. To capture that long memory, ARCH had to explicitly look back over many, many past days and give each one its own knob to tune. That's a lot of knobs. Fitting them was fiddly, and the model got unwieldy.

Imagine trying to remember a long phone number by reciting every digit out loud, over and over. Exhausting. There had to be a shortcut.

The one clever tweak

Bollerslev's insight is almost embarrassingly simple once you see it.

Instead of forecasting tomorrow's volatility from all the recent raw shocks, he said: also feed in yesterday's volatility forecast itself.

That's it. That's the whole move. And it's powerful because yesterday's forecast already quietly contains a summary of everything that came before it. Yesterday's number was built from the day before, which was built from the day before that, and so on. So instead of dragging around a long list of past days, you carry one tidy running summary that updates each day.

Here's an analogy. Suppose you want to track your average spending. One way: every day, re-add up every purchase you've ever made. Painful. The smarter way: keep a running average, and each day just nudge it a little toward today's spending. Same information, almost none of the bookkeeping. GARCH is that running-average trick, applied to market risk.

The recipe, in plain words

Every day, GARCH builds its forecast of tomorrow's jumpiness by blending three things:

• A baseline, the market's long-run "normal" level of volatility, a kind of gravity it always drifts back toward.
• Yesterday's surprise, how big was today's actual move? A shock pushes the forecast up.
• Yesterday's forecast, how jumpy did we already think things were? This carries the momentum forward.

Mix those three with the right weights and you get a forecast that spikes quickly when a shock hits and then decays smoothly back toward normal as calm returns. It behaves just like real market volatility, and it does it with a handful of numbers instead of dozens.

Why "mean-reverting volatility" is such a useful idea

Baked into GARCH is a feature traders love: volatility gets pulled back toward its long-run average. After a crisis, GARCH doesn't expect chaos forever, it expects the storm to fade. During a dead-calm stretch, it doesn't expect glass-smooth markets forever, it expects some bumpiness to return.

This gives you a genuinely useful forecast of the path of risk, not just today's snapshot. Ask GARCH "how jumpy will things be over the next month?" and it can answer, because it knows how fast today's storm should fade toward normal. That single feature is why GARCH sits under so much of practical finance.

Where it shows up in the real world

GARCH isn't a museum piece, it's a daily tool:

• Options and derivatives. The price of an option hinges on expected future volatility. GARCH-style forecasts help traders judge whether options look cheap or expensive.
• Risk limits and capital. Banks and funds size their safety cushions off volatility forecasts. When GARCH says storms are coming, prudent desks pull in their horns.
• Position sizing. Many systematic strategies bet smaller when forecast volatility is high and larger when it's low, aiming for a steady level of risk. GARCH is a natural engine for that.
• A whole family of cousins. GARCH spawned dozens of variants (EGARCH, GJR-GARCH, and more) that fix specific weaknesses, proof of how solid the base design was.

The honest limitations

GARCH is a workhorse, not a crystal ball. Keep these in mind:

• It's always looking backward. GARCH forecasts future storms from past storms. A shock that comes out of nowhere, a surprise announcement, a sudden default, blindsides it just like everyone else. It reacts fast, but it can't pre-see.
• The plain version treats good and bad news alike. A big rally and a big crash of the same size bump the forecast equally. In reality, crashes usually spike volatility more, fear moves faster than greed. The GJR and EGARCH variants were invented specifically to fix this.
• It assumes tomorrow rhymes with the past. GARCH is estimated from history and quietly assumes the market keeps behaving the way it did. When the market's whole character shifts, new regime, new rules, new players, the old settings can mislead.

The one-line takeaway

Bollerslev took Engle's brilliant-but-bulky idea and made it travel light: carry a running summary of how jumpy the market has been, nudge it with each day's surprise, and let it drift back toward normal. That simple recipe turned volatility forecasting from an academic exercise into a tool the whole industry still runs on.