Robust TGARCH — Threshold GARCH with Robust Estimation
Robust Threshold Generalized Autoregressive Conditional Heteroscedasticity Model · Also known as: robust GJR-GARCH, robust threshold GARCH, heavy-tail TGARCH, outlier-robust TGARCH
Robust TGARCH extends the Threshold GARCH model by replacing the conventional maximum likelihood objective with an estimator that is resistant to heavy-tailed innovations and outlying observations. It captures asymmetric volatility responses — where negative shocks amplify variance more than positive shocks — while remaining reliable when the return distribution deviates strongly from normality.
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 Robust TGARCH when modelling financial return volatility in series that show both asymmetric responses to positive and negative shocks and frequent extreme observations or heavy tails — for example, individual stock returns, cryptocurrency prices, or emerging-market exchange rates. It is also appropriate when diagnostic checks on a standard TGARCH suggest that the standardized residuals have very high excess kurtosis or when model estimates are unstable across subsamples due to occasional crash events. Do not use it when returns appear nearly Gaussian and outlier-free; standard TGARCH or EGARCH suffices then. Also avoid this approach for very short series (fewer than ~200 observations) where robust estimation may converge poorly.
Strengths & limitations
- Captures the leverage effect: negative shocks increase conditional variance by more than equally sized positive shocks.
- Resistant to extreme observations and heavy-tailed innovations that would distort standard TGARCH estimates.
- Produces valid inference through sandwich standard errors even under distributional misspecification.
- Forecasts of conditional volatility are less susceptible to single crisis events inflating or deflating parameter estimates.
- Compatible with value-at-risk and expected shortfall calculations in risk management contexts.
- More computationally intensive than standard TGARCH due to iterative robust weighting or heavy-tail likelihood evaluation.
- Choice of the robust loss function (t-QMLE vs. M-estimator vs. Laplace) is not fully standardised and can affect results.
- Requires a reasonably long series (200+ observations) to reliably estimate the asymmetric parameter and robust weights jointly.
- Interpretation of robustness properties depends on the specific estimator chosen, complicating comparison across studies.
Frequently asked
What distinguishes Robust TGARCH from standard TGARCH?
Standard TGARCH is typically estimated by Gaussian quasi-maximum likelihood, which is efficient under normality but sensitive to outliers. Robust TGARCH replaces or augments this with an estimator — such as Student-t QMLE or an M-estimator — that down-weights extreme observations, preserving the asymmetric volatility structure without letting large shocks dominate the parameter estimates.
How is the threshold effect tested in a Robust TGARCH?
The asymmetry parameter γ is tested with a t-statistic based on the sandwich (Huber-White) covariance matrix. A significantly positive γ confirms that negative shocks raise volatility more than positive shocks. The test is robust to distributional misspecification because the sandwich estimator is used.
Is Robust TGARCH the same as GJR-GARCH with a robust estimator?
Essentially yes. The GJR-GARCH model of Glosten, Jagannathan, and Runkle (1993) uses the same threshold indicator structure as Zakoian's TGARCH — both capture asymmetric volatility. Applying a robust estimator to either yields what is commonly labelled Robust TGARCH or robust GJR-GARCH.
What sample size is needed for Robust TGARCH?
As a practical guideline, at least 200 observations are recommended for GARCH-class models in general; robust estimation adds additional complexity, so 300–500 or more observations improve convergence and the reliability of asymptotic inference.
Does robust estimation eliminate the need to check for unit-root or structural breaks?
No. Robustness addresses the sensitivity to outlying innovations within a stationary model, but non-stationarity or structural breaks require separate tests (ADF, KPSS, Zivot-Andrews). A Robust TGARCH fit to a non-stationary series will still produce unreliable forecasts.
Sources
- Zakoian, J.-M. (1994). Threshold heteroskedastic models. Journal of Economic Dynamics and Control, 18(5), 931–955. DOI: 10.1016/0165-1889(94)90039-6 ↗
- Preminger, A., & Storti, G. (2017). Least squares estimation for GARCH (1,1) model with heavy tailed errors. The Econometrics Journal, 20(1), 221–258. link ↗
How to cite this page
ScholarGate. (2026, June 3). Robust Threshold Generalized Autoregressive Conditional Heteroscedasticity Model. ScholarGate. https://scholargate.app/en/econometrics/robust-tgarch
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.
- ARCH modelEconometrics↔ compare
- DCC-GARCH modelEconometrics↔ compare
- EGARCH modelEconometrics↔ compare
- Robust ARCH modelEconometrics↔ compare
- Robust GARCH modelEconometrics↔ compare
- TGARCH modelEconometrics↔ compare