Move Beyond Tables: Simulating Volatility in ARCH–GARCH Models

🔍 The Problem

Autoregressive conditional heteroskedasticity (ARCH) and generalized ARCH (GARCH) models estimate both the conditional mean and the conditional error variance of a time series. While simulation methods already allow disaggregation of short- and long-run effects on the conditional mean, inferences about the conditional error variance remain largely limited to static, tabular interpretation. That limitation obscures how volatility responds to shocks and can produce incomplete or incorrect inferences.

🧭 Main Contribution

Introduces a simulation-based approach to interpreting conditional error variance in ARCH–GARCH processes. The paper demonstrates that changes in these processes are conditional on starting values, other covariates, and model dynamics, and it offers three bootstrapping techniques to simulate conditional error variance outcomes. The usefulness of each technique is illustrated through replication of prominent empirical studies.

Key points:

• Conditional error variance responses depend on starting values, covariates, and dynamic structure—factors that can undermine simple tabular conclusions.
• Three distinct bootstrapping procedures are developed to simulate and visualize conditional error variance from fitted ARCH–GARCH models.
• Replications of influential studies show how each technique clarifies statistical and substantive claims about volatility.

🛠️ How the Techniques Work

Each bootstrapping technique generates simulated paths of the conditional error variance given the fitted ARCH–GARCH specification, producing prediction intervals and dynamic visualizations. This approach moves interpretation from single-point tables to simulated distributions that capture both short- and long-run variance effects and their uncertainty.

📌 Why It Matters

Simulation and prediction are essential for drawing reliable statistical and substantive inferences about volatility in dynamic time series. The proposed methods enable researchers to visualize uncertainty, separate short- versus long-run variance impacts, and avoid misleading conclusions that arise from relying solely on tabular summaries.