Structural Break TGARCH (Threshold GARCH with Structural Breaks)
Structural Break Threshold GARCH · Also known as: SB-TGARCH, threshold GARCH with structural breaks, GJR-GARCH with structural breaks, break-adjusted TGARCH
Structural Break TGARCH extends the Threshold GARCH (GJR-GARCH) model to accommodate discrete, permanent shifts in the volatility process. By detecting structural breaks and incorporating them — either as regime-specific intercepts or dummy variables — the model separates genuine volatility persistence from spurious persistence induced by ignored regime changes, and preserves the asymmetric leverage effect that characterises equity and financial return data.
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 Structural Break TGARCH when modelling financial return volatility over long samples that span identifiable structural events — financial crises, monetary regime changes, or major policy shifts. It is especially appropriate when a standard TGARCH yields implausibly high persistence (alpha + beta + gamma/2 close to 1) and ICSS or Bai-Perron tests reject variance homogeneity. It is also the right tool when the leverage effect is theoretically expected (equity returns) and must be preserved after correcting for breaks. Avoid it when the sample is short (fewer than ~500 observations), when no structural break is detected, or when continuous regime change is more plausible — in that case Markov-Switching GARCH is preferable.
Strengths & limitations
- Eliminates spurious volatility persistence caused by unmodelled structural breaks, yielding unbiased GARCH parameter estimates.
- Retains the asymmetric leverage effect of TGARCH/GJR-GARCH, which is empirically important for equity and currency data.
- Improves out-of-sample volatility forecasting by using regime-specific intercepts rather than a single, distorted long-run variance.
- Compatible with formal breakpoint detection algorithms (ICSS, Bai-Perron), making break selection data-driven rather than ad hoc.
- QMLE with sandwich standard errors ensures valid inference under non-Gaussian innovations.
- Break detection methods can disagree on the number and location of break dates, making the final model specification sensitive to the chosen detection procedure.
- Estimating regime-specific parameters requires sufficient observations in each sub-period; very short regimes lead to poorly identified GARCH coefficients.
- The model assumes abrupt, permanent breaks; gradual structural change or smooth transitions are better handled by time-varying parameter or Markov-Switching GARCH models.
- With many detected breaks, the model becomes heavily parameterised, risking overfitting and poor parsimony.
Frequently asked
How is TGARCH different from EGARCH when structural breaks are present?
TGARCH (GJR-GARCH) models asymmetry through an indicator on negative lagged shocks in the variance equation; EGARCH models it on the log-variance. Both can be augmented with structural break dummies, but TGARCH is more commonly used in break-adjusted frameworks because its parameters have direct variance-level interpretations, making regime comparisons straightforward.
How many observations are needed per regime for reliable estimation?
A practical minimum is roughly 100-200 observations per regime so that the ARCH(1), GARCH(1), and asymmetry (gamma) parameters are well-identified. Very short regimes — fewer than 50 observations — typically yield unstable estimates and should be merged with adjacent regimes or treated as outlier episodes.
Can I use calendar-based breaks (e.g., the 2008 financial crisis) instead of statistical break tests?
Yes, but statistical confirmation is strongly advisable. A calendar date chosen a priori may not align with the actual shift in the variance process. Run an ICSS or Bai-Perron test first; if the algorithm recovers a date near the event, the calendar break is corroborated. If it does not, trust the data-driven date or investigate further.
Does correcting for structural breaks always reduce GARCH persistence?
Almost always, yes. Lamoureux and Lastrapes (1990) showed empirically that break dummies sharply reduce estimated persistence. If persistence remains near 1 after break correction, it may indicate true long memory in volatility, warranting a FIGARCH or HYGARCH specification instead.
Which software implements Structural Break TGARCH?
R packages rugarch and MSwM support TGARCH estimation with break dummies; the strucchange and BreakoutDetection packages handle break detection. In Python, arch provides GJR-GARCH and accepts exogenous regressors (break dummies) in the variance equation. Stata users can combine vcemway robust standard errors with custom GARCH code via the garch command.
Sources
- Lamoureux, C. G., & Lastrapes, W. D. (1990). Persistence in variance, structural change, and the GARCH model. Journal of Business & Economic Statistics, 8(2), 225-234. DOI: 10.1080/07350015.1990.10509794 ↗
- Glosten, L. R., Jagannathan, R., & Runkle, D. E. (1993). On the relation between the expected value and the volatility of the nominal excess return on stocks. Journal of Finance, 48(5), 1779-1801. DOI: 10.1111/j.1540-6261.1993.tb05128.x ↗
How to cite this page
ScholarGate. (2026, June 3). Structural Break Threshold GARCH. ScholarGate. https://scholargate.app/en/econometrics/structural-break-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.
- EGARCH modelEconometrics↔ compare
- GARCH ModelEconometrics↔ compare
- TGARCH modelEconometrics↔ compare