What it means
Ordinary statistical models assume that the noise around a forecast has a constant size. Financial markets mock that assumption, because calm weeks cluster together and so do violent ones.
ARCH, introduced by Robert Engle in 1982, was built to capture exactly this pattern of volatility clustering. The name unpacks the mechanism.
The model is autoregressive because today's variance depends on past squared errors, conditional because it updates as new information arrives, and heteroskedastic because the variance is allowed to differ from period to period. A large surprise yesterday means a wide band of uncertainty today.
This idea earned Engle the 2003 Nobel Prize in Economic Sciences, shared with Clive Granger. The Nobel committee highlighted how the model reshaped risk measurement, because it gave analysts a way to forecast not just where a price was heading but how rough the ride would be.
In practice, ARCH estimates the current variance as a weighted average of recent squared residuals. When markets have been quiet, the forecast variance shrinks, and after a shock it jumps, then decays as calm returns.
Risk limits, margin requirements and option prices all adjust accordingly. The basic model was soon extended.
Generalised ARCH, or GARCH, adds lagged variances alongside lagged squared errors, which captures the smooth persistence of volatility with fewer parameters, and most applied work today uses GARCH. Asymmetric versions let bad news raise volatility more than good news, matching the leverage effect seen in equities, while long-memory versions capture slow-decaying shocks, each variant existing because a real desk found the plain version wanting.
ARCH-style models also explain why returns show fat tails: even if each period's shock is normal conditional on current volatility, the shifting variance makes the unconditional distribution heavy-tailed, so extreme days become far more common than a constant-variance model predicts. Estimation follows the same discipline as any time-series work.
The analyst first fits a mean model, often a simple autoregression, then models its squared residuals with the ARCH variance equation, and software handles the likelihood maximisation while the analyst judges whether the pattern is stable enough to trust. The model does not explain why volatility clusters, since news arrival, leverage effects and market structure are candidate explanations while ARCH itself is agnostic about causes, and for managers outside trading floors the lesson generalises: any process whose risk comes in bursts, such as supply disruptions or credit losses, deserves a model that lets uncertainty breathe, because planning around a single average error size understates risk exactly when it is largest.
In practice
Real-world examples.
Example
A risk officer forecasts tomorrow's value-at-risk using a volatility estimate that jumped after yesterday's market plunge. The wider estimate lifts the capital the desk must hold against its positions.
Example
An options desk raises its volatility inputs after a cluster of large daily moves, lifting option prices. Customers buying protection pay more because the model expects turbulence to persist.
Example
A pension fund scales down equity exposure automatically when estimated variance crosses a preset ceiling. The rule cuts risk during turbulence and restores it as the estimate decays.
Formula
Calculation
The ARCH(1) variance equation is sigma2(t) = omega + alpha e2(t-1), where e(t-1) is yesterday's error.
Example: with omega = 0.1 and alpha = 0.85, a squared error of 4 yesterday implies a variance today of 0.1 + 0.85 x 4 = 3.5, a standard deviation of about 1.87%. If the next day's squared error is only 1, the variance falls to 0.1 + 0.85 x 1 = 0.95, a standard deviation of about 0.97%. On a $1,000,000 position, a daily standard deviation of 1.87% is roughly $18,700 of expected swing, against roughly $9,700 after calm returns.Case study
Seen in the real world.
This is a fictional, illustrative example. A treasury team at an importer fits an ARCH model to daily exchange-rate moves. After a week of large swings, the model doubles the forecast variance, so the team widens its hedging band before the next currency purchase instead of being caught flat. In this illustrative story, the team re-estimates the model each quarter, because volatility regimes shift with market structure. When the swings fade, the forecast variance decays and the hedging band narrows again, which keeps the cost of hedging in line with the actual risk.
Watch out
Common mistakes.
- Assuming constant volatility when measuring risk over long windows. Periods of calm and turmoil alternate, so a single average variance misprices both.
- Reading an ARCH variance forecast as a direction forecast. The model says nothing about whether prices will rise or fall, only how wide the range of outcomes is.
- Treating the estimated parameters as permanent. Volatility regimes shift with market structure, so models need periodic re-estimation rather than set-and-forget use.
Questions
People also ask.
Who created ARCH?
Robert Engle introduced the model in 1982, and he shared the 2003 Nobel Prize in Economic Sciences largely for this work on time-varying volatility.
What is the difference between ARCH and GARCH?
GARCH adds lagged variance terms to the equation, capturing persistent volatility with fewer parameters than a long ARCH model.
Does ARCH predict price direction?
No. It forecasts the size of likely fluctuations, not their sign, which is why it is used for risk measurement rather than market timing.
From the founder's library
Take it further with the book.
Build your financial confidence beyond this definition. Shihan's full-length guide, Accounting Fundamentals, takes the same plain-English approach and turns it into a complete, practical playbook for non-finance managers, business owners and students - with chapter-end quiz answers and presentation slides included.
25% off with code MMHQ25, applied at checkout. Priced in USD - checkout may show the equivalent in your local currency.
View the book and save 25%