Volatility Clustering: Why Calm Follows Calm, Chaos Follows Chaos

Key Takeaways

• Absolute or squared daily returns autocorrelate at 0.15 to 0.30 at lag 1 for liquid equities, decaying over dozens of lags.
• GARCH(1,1) with alpha near 0.08 and beta near 0.90 produces an alpha plus beta sum of 0.98, giving slow multi-week decay of shocks.
• Using a static 250-day rolling standard deviation for VaR underestimates risk at crisis onset and overestimates it long after calm returns.
• Volatility forecasts from GARCH guide options pricing, position sizing, and dynamic stop-loss levels across all institutional strategies.

What It Is

Volatility clustering describes a specific pattern in financial time series: the absolute size of returns shows strong positive autocorrelation over many lags, while the raw returns themselves show almost none at short horizons. A 3 percent drop today makes another large move tomorrow more likely, but it does not reliably predict whether that move will be up or down.

Benoit Mandelbrot flagged the pattern in 1963 when studying cotton prices, writing that "large changes tend to be followed by large changes, of either sign, and small changes tend to be followed by small changes." Rama Cont later listed it among the universal stylized facts of asset returns, alongside fat tails and leverage effects.

The Intuition

Markets digest information unevenly. When uncertainty spikes, whether from an earnings surprise, a central bank announcement, or a liquidity shock, the flow of news does not resolve in a single tick. Traders reposition over hours and days, hedgers adjust deltas, and risk limits force further unwinds. The result is that a high-volatility state persists for a while before reverting to normal.

The same logic runs in reverse. When nothing is happening, spreads are tight, dealers are comfortable, and the feedback loops that amplify price moves are dormant. That calm tends to last until something breaks it.

How It Works

You can measure volatility clustering with the autocorrelation function (ACF) of squared or absolute returns:

ACF(k) = Corr( |r_t| , |r_{t-k}| )

For liquid equities, ACF(1) on |r_t| is typically 0.15 to 0.30 and decays slowly over dozens of lags. The ACF of raw returns r_t at the same horizon is usually statistically indistinguishable from zero.

The workhorse model for this pattern is GARCH(1,1), introduced by Engle in 1982 and generalized by Bollerslev in 1986. It writes conditional variance as a weighted blend of a long-run level, yesterday's variance, and yesterday's squared shock:

sigma^2_t = omega + alpha * r^2_{t-1} + beta * sigma^2_{t-1}

Where omega is a constant, alpha is the weight on the most recent squared return, and beta is the weight on the prior variance. Typical equity estimates put alpha near 0.08 and beta near 0.90, so alpha + beta is close to 1. That near-unit sum is why shocks take weeks to decay.

Worked Example

Suppose the S&P 500 has an unconditional daily standard deviation of about 1 percent, implying a variance of 0.0001. After a one-day move of minus 4 percent, the squared return is 0.0016.

Using GARCH(1,1) with omega = 0.000002, alpha = 0.08, beta = 0.90, yesterday's variance 0.0001:

sigma^2_today = 0.000002 + 0.08 * 0.0016 + 0.90 * 0.0001 = 0.000220

The new one-day volatility forecast is the square root, about 1.48 percent. The single 4 percent shock roughly 1.5x's the model's next-day volatility, and because beta is large, the forecast stays elevated for many sessions before reverting to 1 percent. That persistence is the clustering.

Common Mistakes

• Confusing clustering with predictable returns. Volatility is forecastable; direction is not. Clustering does not mean you can time the next rally or crash. It means you can size positions, set stops, and price options knowing risk is currently high or low.

• Using constant-volatility models in risk systems. Value-at-Risk built on a rolling 250-day standard deviation reacts too slowly. It underestimates risk at the start of a stress regime and overestimates it well after calm returns. GARCH, EWMA, or realized-volatility estimators react faster.

• Ignoring asymmetry. Equity volatility responds more to negative shocks than to positive ones of equal size, known as the leverage effect. Symmetric GARCH misses this. GJR-GARCH and EGARCH allow the asymmetric term.

• Treating intraday and daily clustering as the same process. Intraday volatility has its own U-shaped pattern around the open and close. Blindly applying a daily GARCH to 5-minute bars produces misleading forecasts. Deseasonalize first.

• Fitting GARCH and then forgetting about it. Parameters drift. Periodic refits, rolling windows, or Bayesian updating are standard practice. A 10-year-old estimate is not a current estimate.

Frequently Asked Questions

Q: What is volatility clustering in simple terms? It is the observation that the size of market moves tends to persist: a day of large moves makes the next day's moves likely to be large too, while quiet periods tend to stay quiet until some event breaks the calm.

Q: How does volatility clustering affect investment decisions? It means current realized volatility is the best short-term predictor of near-future volatility, so risk systems that use GARCH or EWMA volatility estimates can size positions and set stop losses more accurately than systems using a long rolling average.

Q: What is a real-world example of volatility clustering? After the S&P 500 fell 4 percent in a single day in early 2020, GARCH(1,1) would have raised the next-day volatility forecast from 1 percent to roughly 1.5 percent, staying elevated for several weeks before decaying back toward the long-run level.

Q: How can investors use volatility clustering? Use a GARCH or EWMA model to estimate current conditional volatility, then scale position sizes inversely so that the dollar risk per trade remains constant regardless of whether the market is in a calm or stressed regime.

Q: How is volatility clustering different from serial correlation in returns? Serial correlation in returns means the direction of returns predicts future direction. Volatility clustering means only the magnitude is predictable, not the direction. The two are largely independent: daily returns have near-zero serial correlation but absolute returns are strongly autocorrelated.

Sources

1. Mandelbrot, B. (1963). "The Variation of Certain Speculative Prices." Journal of Business, 36(4), 394-419. https://www.jstor.org/stable/2350970
2. Cont, R. (2001). "Empirical properties of asset returns: stylized facts and statistical issues." Quantitative Finance, 1(2), 223-236. https://www.tandfonline.com/doi/abs/10.1080/713665670
3. Engle, R. (1982). "Autoregressive Conditional Heteroskedasticity with Estimates of the Variance of UK Inflation." Econometrica, 50(4), 987-1007. https://www.jstor.org/stable/1912773
4. Bollerslev, T. (1986). "Generalized Autoregressive Conditional Heteroskedasticity." Journal of Econometrics, 31(3), 307-327. https://doi.org/10.1016/0304-4076(86)90063-1

Disclaimer

This article is educational content only and is not financial advice. Nothing here is a recommendation to buy, sell, or hold any security. Consult a licensed advisor before making investment decisions.