Modeling Asymmetric Volatility with EGARCH
Abstract
This post presents an accessible introduction to the Exponential GARCH (EGARCH) model—a widely used tool in financial econometrics for modeling time-varying volatility in asset returns. Unlike standard GARCH models, EGARCH captures both volatility clustering and the leverage effect, whereby negative shocks tend to increase future volatility more than positive shocks of equal magnitude. The post demonstrates how EGARCH achieves this using a simplified form, with an emphasis on the asymmetry function's role in shaping variance dynamics. The dynamic visualization of EGARCH volatility asymmetry presented in this post was created by the author using Desmos.
Introduction
To illustrate the leverage mechanism, we intentionally compress the full econometric treatment and focus on isolating this single effect. Although the GARCH family includes many models, we use EGARCH to illustrate how such models effectively capture features of financial return series—particularly volatility clustering and the leverage effect. Empirical observations confirm that asset returns often exhibit these effects. Below, we explore the EGARCH model with a focus on the leverage effect.
Structure of the EGARCH Model: EGARCH (p,q)
The \(\mathrm{EGARCH(p},\mathrm{q)}\) can be written as:
\[\mathrm{y}_{\mathrm{t}}=\mathrm{\mu}_{\mathrm{t}}+\mathrm{\varepsilon}_{\mathrm{t}},\]
\[\mathrm{ \varepsilon}_{\mathrm{t}}=\mathrm{\sigma}_{\mathrm{t}}\mathrm{z}_{\mathrm{t}},\mathrm{ z}_{\mathrm{t}}\thicksim \mathrm{i}.\mathrm{i}.\mathrm{d}.\mathrm{ (}0,1),\]
\[\ln \left( \mathrm{\sigma}_{\mathrm{t}}^{2} \right) =\mathrm{\omega}+\Sigma _{\mathrm{i}=1}^{\mathrm{q}}\mathrm{ \beta}_{\mathrm{i}}\ln \left( \mathrm{\sigma}_{\mathrm{t}-\mathrm{i}}^{2} \right) +\Sigma _{\mathrm{j}=1}^{\mathrm{p}}\mathrm{ \alpha}_{\mathrm{j}}\left( \left| \mathrm{z}_{\mathrm{t}-\mathrm{j}} \right|+\mathrm{\phi}_{\mathrm{j}}\mathrm{z}_{\mathrm{t}-\mathrm{j}} \right) .\]
To optimally isolate the leverage effect, we adopt the EGARCH representation of Linton (2019), abstracting from alternative parametrizations and decompositions of asymmetry present in the canonical specification. EGARCH does not require special constraints on parameters, \(\ln \left( \mathrm{\sigma}_{\mathrm{t}}^{2} \right)\) can take any real value. The terms involving \(\ln \left( \mathrm{\sigma}_{\mathrm{t}-\mathrm{i}}^{2} \right)\) capture volatility clustering. In contrast, the components \(\mathrm{\alpha}_{\mathrm{j}}\left( \left| \mathrm{z}_{\mathrm{t}-\mathrm{j}} \right|+\mathrm{\phi}_{\mathrm{j}}\mathrm{z}_{\mathrm{t}-\mathrm{j}} \right)\) model the leverage effect —the asymmetric response of volatility to past shocks.
Simplified EGARCH Expression: EGARCH (1,1)
To better understand the core mechanics of the EGARCH model, we consider its simplest form — the \(\mathrm{EGARCH(}1,1)\) specification:
\[\ln \!\:\left( \mathrm{\sigma}_{\mathrm{t}}^{2} \right) =\mathrm{\omega}+\mathrm{\beta}_1\ln \!\:\left( \mathrm{\sigma}_{\mathrm{t}-1}^{2} \right) +\mathrm{\alpha}_1\left( \left| \mathrm{z}_{\mathrm{t}-1} \right|+\mathrm{\phi}_1\mathrm{z}_{\mathrm{t}-1} \right) .\]
This can also be rewritten in its exponential form, which highlights the origin of the “E” in EGARCH:
\[\mathrm{\sigma}_{\mathrm{t}}^{2}=\mathrm{e}^{\mathrm{\omega}+\mathrm{\beta}_1\ln \!\:\left( \mathrm{\sigma}_{\mathrm{t}-1}^{2} \right) +\mathrm{\alpha}_1\left( \left| \mathrm{z}_{\mathrm{t}-1} \right|+\mathrm{\phi}_1\mathrm{z}_{\mathrm{t}-1} \right)}.\]
This formulation makes it clear that the conditional variance \(\mathrm{\sigma}_{\mathrm{t}}^{2}\) is an increasing exponential function of past log-variance and standardized shocks. The component
\[\mathrm{\alpha}_1\left( \left| \mathrm{z}_{\mathrm{t}-1} \right|+\mathrm{\phi}_1\mathrm{z}_{\mathrm{t}-1} \right) \]
plays a central role in shaping how the variance at time \(\mathrm{t}\) responds to the shock at time \(\mathrm{t}-1\). Here, \(\mathrm{z}_{\mathrm{t}-1}\) represents the standardized innovation from the previous period. The term \(\left| \mathrm{z}_{\mathrm{t}-1} \right|\) reflects the magnitude of the shock, regardless of its direction — this accounts for the size effect. The term \(\mathrm{\phi}_1\mathrm{z}_{\mathrm{t}-1}\) captures the directional component of the shock, with \(\mathrm{\phi}_1\) controlling the strength of asymmetry. This allows the model to differentiate the impact of negative and positive shocks on volatility. The parameter \(\mathrm{\alpha}_1\) governs the combined effect of the standardized shock’s magnitude, direction, and asymmetry on the log-variance. This formulation allows the model to capture not only how large a shock was, but also whether it was positive or negative, and whether its direction matters differently.
Visualizing the EGARCH Leverage Effect
To illustrate how the EGARCH model captures asymmetry in volatility responses, we present the following dynamic visualization. On the horizontal axis are values of the standardized shock \(\mathrm{z}_{\mathrm{t}-1}\), and on the vertical axis are the corresponding values of conditional variance \(\mathrm{\sigma}_{\mathrm{t}}^{2}\).
Let’s interpret what can be observed from the plot. Set \(c=\mathrm{\omega}+\mathrm{\beta}_1\ln \!\:\left( \mathrm{\sigma}_{\mathrm{t}-1}^{2} \right) =0.5\) and fix \(\mathrm{a}=\mathrm{\alpha}_1=1\). Now, explore the behavior of the model by varying the asymmetry parameter \(\mathrm{\phi}_1\).
- When \(\mathrm{b}=\mathrm{\phi}_1>1\) , the plot clearly shows that positive shocks \(\mathrm{z}_{\mathrm{t}-1}>0\) lead to a much larger increase in variance compared to negative shocks.
- When \(\mathrm{b}=\mathrm{\phi}_1<-1\) , the reverse effect is observed: negative shocks drive variance significantly more than positive ones.
This visualization highlights how EGARCH flexibly models different types of nonlinear asymmetry in volatility dynamics — a key advantage over symmetric models like standard GARCH.
EGARCH in R and Python
Implementations available on GitHub in both R and Python estimate an \(\mathrm{EGARCH(}1,1)\) model using return data stored in a CSV file and perform a one-step-ahead forecast of the conditional mean and volatility. While both platforms yield consistent parameter estimates, note that the parameter names and notations may differ slightly between their output formats—and from the notation used in this post.
Conclusion: EGARCH in Volatility Arbitrage
EGARCH’s ability to capture asymmetric volatility can be especially useful in volatility arbitrage. After a sharp negative return, the model anticipates increased future volatility due to its sensitivity to the direction of shocks. This allows traders to detect moments when forecasted volatility—derived from the model—exceeds implied volatility observed in option prices, signaling a potential arbitrage opportunity. By entering long-volatility positions, such as buying underpriced options before the market adjusts, traders can capitalize on this gap. Conversely, the model can also signal short-volatility opportunities when forecasted volatility falls below implied volatility. Moreover, the model’s forecasted volatility can help determine position parameters such as option strike selection, trade sizing, and hedge ratios, enabling a more structured and risk-aware approach to volatility exposure.
To put volatility modeling in a broader trading strategy context, see: Foundations of Market Timing and Directional Change Trading.
Resources
- Linton, O. (2019). Financial econometrics: Models and methods. University of Cambridge.
- EGARCH Volatility Modeling Scripts (R + Python)
- Synthetic Data in Inferential Backtesting — textbook covering ARMA–GARCH–Copula simulation workflows
- Synthetic OHLC Simulation via DCC–GARCH — R and Python implementation with backtesting interface
