Third-Party Research & Methodology Only

This section shares summaries of third-party academic research and descriptions of quantitative models. The content represents the findings of the original researchers, not the opinions or recommendations of Foxholm Financial. Foxholm Financial does not publish hypothetical or backtested performance metrics on its quantitative research pages. All content is restricted to methodology, signal construction, factor logic, and risk architecture. SEC rules require that investment advisers not present misleading performance data, and our methodology-only approach reflects that standard and the firm's fiduciary obligations.

GARCH

Econometric Model Volatility Risk Academic Finance

GARCH, short for Generalized Autoregressive Conditional Heteroskedasticity, is a statistical model that forecasts how the size of price swings changes over time. Its central insight is that volatility is not constant: calm periods and turbulent periods tend to cluster together, and today's expected variation depends on both recent shocks and recent variation.

The name describes the mechanism. "Conditional" means the forecast depends on what just happened rather than on a fixed long-run average. "Heteroskedasticity" is the technical term for changing variance, meaning the spread of returns is not the same from one period to the next. "Autoregressive" and "generalized" describe how the model lets current variance depend on its own recent values. Together, these features let GARCH capture a pattern that simpler models miss, namely that large moves tend to follow large moves and quiet stretches tend to follow quiet stretches.

Definition

A GARCH model expresses today's variance (a measure of how spread out returns are) as a weighted combination of three pieces: a baseline long-run level, the size of the most recent return shock, and the variance estimated for the previous period. By feeding recent variance back into the current estimate, the model produces forecasts that rise after turbulent periods and fade gradually as markets calm. This feedback is what allows GARCH to track volatility clustering directly.

The most common version, GARCH(1,1), uses one recent shock term and one recent variance term, which is often enough to describe the behavior of financial returns. The model treats volatility (the degree of price swings) as something that drifts and persists rather than something fixed. That single change from a constant-variance assumption to a time-varying one is the source of most of the model's value.

Key Principle

GARCH works because financial volatility is persistent: a shock today raises the expected size of swings tomorrow, and that elevated level decays over time rather than disappearing at once. This persistence is why constant-variance models tend to understate risk right when it matters most, during turbulent stretches. By letting recent shocks and recent variance shape the current forecast, GARCH produces a standard deviation estimate that responds to changing conditions instead of averaging them away.

How It Works

The model is fit to a history of returns by estimating the weights on each of its three components. A larger weight on the recent shock term means volatility reacts quickly to new surprises, while a larger weight on the recent variance term means volatility is slow to fade once it rises. The sum of these two weights describes how persistent volatility is: a sum close to one implies shocks linger for a long time, while a smaller sum implies they die out quickly.

Once fit, the model can forecast variance one or more periods ahead, and those forecasts feed naturally into other risk tools. A GARCH estimate of near-term volatility can serve as an input to value-at-risk calculations, giving a loss threshold that tightens during turbulent periods and loosens during calm ones. This responsiveness is the practical payoff: risk estimates that adjust to conditions rather than lagging behind them.

The ARCH and GARCH Family

Model	Core Idea	What It Adds
ARCH	Variance depends on recent squared shocks	First model of time-varying variance
GARCH	Variance depends on recent shocks and recent variance	Captures persistence with fewer terms
Asymmetric variants	Down moves and up moves can affect variance differently	Reflects that losses often raise volatility more

ARCH, the original model, captured the idea that variance depends on recent shocks. GARCH generalized it by adding the recent variance term, which let the model describe long-lasting volatility with far fewer parameters. Later extensions, including asymmetric variants, allow falling and rising prices to affect future volatility by different amounts, reflecting the common observation that sharp declines tend to stir up larger swings than gains of the same size.

Applications

GARCH is widely used to forecast near-term volatility for risk management, option valuation, and portfolio construction. Because the model produces a forward-looking variance estimate, it can feed value-at-risk systems, inform hedging decisions, and help size positions in proportion to expected market turbulence. It also provides a useful point of comparison with implied volatility (the volatility the options market expects), since the two represent different routes to the same forward-looking question.

The model's appeal lies in capturing a stylized fact that plain historical averages cannot. Returns in many markets show heavier extremes than a simple bell curve predicts, a pattern related to fat tails. Time-varying volatility is one source of that pattern, and a model that tracks it can produce risk estimates that better reflect how markets actually behave during stressed periods.

Known Limitations

Limitations to Keep in Mind

Backward-looking by design. GARCH forecasts future volatility from past returns, so it reacts to shocks rather than anticipating them. A sudden, unprecedented event can move markets before the model has any data reflecting it.
Sensitive to model choice. Results depend on the specific variant, the number of terms, and the assumed return distribution. Different reasonable choices can produce different forecasts for the same data, so the output reflects the modeler's assumptions as well as the market.
Estimation can be unstable. Fitting the model requires enough data and can be sensitive to outliers or to structural changes in the market. A regime shift can leave the estimated weights poorly suited to current conditions.
Basic versions miss asymmetry. A standard GARCH model treats upward and downward shocks the same way, even though declines often raise volatility more than gains. Capturing this requires more elaborate variants, which add complexity.
Forecasts revert to an average. Longer-horizon GARCH forecasts pull back toward the model's long-run variance level. This makes the model less informative for far-ahead horizons than for the near term it was built to describe.

Academic Origin

The approach began with Robert Engle, who introduced the autoregressive conditional heteroskedasticity (ARCH) model in 1982. Engle showed that the variance of economic time series could be modeled as depending on its own recent history, capturing the clustering of large and small movements that earlier constant-variance methods ignored. This work earned him a share of the 2003 Nobel Memorial Prize in Economic Sciences.

Tim Bollerslev generalized the model in 1986, adding the lagged variance term that defines GARCH. The generalization let the model describe persistent volatility with far fewer parameters than ARCH required, which made it practical for everyday use. The GARCH framework has since become a foundation of financial econometrics and spawned a large family of related models.