Time-Varying Parameter EGARCH Model
Time-Varying Parameter Exponential GARCH Model · Also known as: TVP-EGARCH, time-varying EGARCH, EGARCH with time-varying parameters, dynamic parameter EGARCH
The TVP-EGARCH model extends Nelson's (1991) Exponential GARCH by allowing the volatility equation's parameters — including the leverage effect coefficient — to drift continuously over time. This makes it possible to capture structural change and regime evolution in financial return volatility without imposing a fixed break date.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use TVP-EGARCH when you model financial return volatility over long samples that span different economic regimes, crises, or regulatory shifts, and you suspect that leverage effects or persistence parameters have not remained constant. It is especially valuable in equity, currency, and commodity markets during prolonged structural change. Do not use it when the sample is short (fewer than roughly 500 observations), because the filter needs sufficient data to identify gradual parameter drift from noise. Also avoid it if a simple regime-switching model with two or three regimes fits the data adequately — TVP-EGARCH has many hyperparameters and can overfit if the sample is modest or the volatility dynamics are actually stable.
Strengths & limitations
- Captures continuously evolving leverage and persistence effects without requiring a pre-specified break date.
- Inherits EGARCH's natural log-variance specification, ensuring positivity without inequality constraints on parameters.
- Accommodates asymmetric volatility responses to positive versus negative shocks with a time-varying leverage coefficient.
- Flexible state-space formulation allows straightforward incorporation of additional predictors or exogenous variables into the parameter dynamics.
- Produces smooth parameter trajectories that are easy to interpret and plot alongside economic events.
- Requires long high-frequency time series (typically 500 or more observations) to reliably identify parameter drift.
- Estimation via nonlinear Kalman filtering is computationally intensive and sensitive to starting values and the specification of Q.
- The number of hyperparameters grows with the dimension of Q, risking over-parameterisation in moderate samples.
- Inference on parameter paths is approximate — filter-based uncertainty bands do not account for estimation error in the hyperparameters.
Frequently asked
How does TVP-EGARCH differ from a Markov-switching EGARCH?
Markov-switching EGARCH allows parameters to jump between a small number of discrete regimes with estimated transition probabilities. TVP-EGARCH instead allows parameters to drift continuously over time as random walks. TVP is more appropriate when change is gradual and regime boundaries are unclear; Markov-switching is better when the data show abrupt, recurring regime shifts.
What estimation software can fit TVP-EGARCH?
The model can be implemented in R (using the dlm or KFAS packages for the state-space structure combined with custom EGARCH log-likelihood code), in Python with filterpy or pykalman, and in MATLAB. Ox/SsfPack is also used in the academic literature for score-driven variants.
What is a score-driven alternative to TVP-EGARCH?
Harvey and Chakravarty (2008) and Creal, Koopman and Lucas (2013) proposed observation-driven 'score-driven' or GAS (Generalised Autoregressive Score) models where parameters are updated by the scaled score of the log-likelihood. GAS-EGARCH is a practical alternative that avoids the nonlinear filter, is easier to estimate by maximum likelihood, and has well-understood large-sample properties.
Is TVP-EGARCH suitable for daily versus high-frequency data?
It was developed primarily for daily return series. At higher frequencies (tick or minute data) realised-volatility-based models often outperform GARCH-type specifications. At lower frequencies (weekly or monthly) the sample size may be insufficient to identify parameter drift.
How do I choose the process noise covariance Q?
Q is typically estimated jointly with the other hyperparameters by maximising the log-likelihood. A common simplification is to set Q proportional to a scalar variance parameter, reducing dimensionality. Cross-validation on a hold-out period or forecast-evaluation criteria (e.g., mean squared forecast error of variance) can also guide the choice.
Sources
- Nelson, D. B. (1991). Conditional heteroskedasticity in asset returns: A new approach. Econometrica, 59(2), 347–370. DOI: 10.2307/2938260 ↗
- Harvey, A. C. (2013). Dynamic Models for Volatility and Heavy Tails: With Applications to Financial and Economic Time Series. Cambridge University Press. ISBN: 9781107034723
How to cite this page
ScholarGate. (2026, June 3). Time-Varying Parameter Exponential GARCH Model. ScholarGate. https://scholargate.app/en/econometrics/time-varying-parameter-egarch-model
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- EGARCH modelEconometrics↔ compare
- GARCH ModelEconometrics↔ compare
- Stochastic Volatility ModelFinance↔ compare