Structural Break NARDL
Structural Break Nonlinear Autoregressive Distributed Lag Model · Also known as: SB-NARDL, NARDL with structural breaks, nonlinear ARDL with break, asymmetric ARDL structural break
Structural Break NARDL extends the Nonlinear Autoregressive Distributed Lag (NARDL) bounds-testing framework by explicitly accommodating one or more structural breaks in the long-run relationship. It separates positive and negative changes in the regressor, tests for cointegration, and allows regime shifts, providing a richer picture of asymmetric and break-sensitive dynamics between variables.
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 NARDL when your time series is long enough to credibly estimate multiple regimes (typically at least 60–100 observations), when theory or visual inspection suggests that the relationship between variables has shifted at one or more historical episodes, and when economic reasoning implies that positive and negative changes in the regressor may not have symmetric effects. Common applications include energy-growth nexus studies, exchange rate pass-through, and financial development-growth links spanning policy change periods. Do not use this approach when the sample is short (fewer than ~50 observations after break trimming), when there is no theoretical basis for expecting a break, or when the regressors are I(2) — the ARDL bounds framework requires I(0) or I(1) variables only.
Strengths & limitations
- Jointly captures nonlinearity (asymmetry) and parameter instability (structural change), two common features of macroeconomic and financial time series.
- The bounds testing approach remains valid when regressors are a mix of I(0) and I(1), avoiding the need to pre-classify integration orders.
- Endogenous break detection reduces pretesting bias relative to imposing breaks at known dates.
- Dynamic multipliers provide an intuitive, graphical summary of the adjustment path and speed of convergence by regime and sign of shock.
- More robust long-run estimates than standard ARDL or NARDL when structural breaks are genuinely present and ignored.
- Requires a sufficiently long time series; estimating breaks with fewer than 50 usable observations per regime is unreliable.
- The number and location of breaks must be chosen carefully; over-specifying breaks leads to fragmented regimes and loss of degrees of freedom.
- No widely accepted break-adjusted critical value tables exist for NARDL bounds tests; practitioners often rely on standard Pesaran et al. tables, which may not be appropriate.
- Computationally demanding when combining multiple breaks with automated lag selection and asymmetry testing, increasing the risk of specification search bias.
- I(2) variables invalidate the bounds testing procedure; unit root pre-testing is essential.
Frequently asked
How do I choose the number of structural breaks?
Use formal tests such as Bai and Perron (2003) to determine the number of breaks sequentially, or apply information criteria (BIC/LWZ). Limit the number of breaks to what the sample size can credibly support — as a rough guide, each regime should contain at least 20–30 observations.
Can I use Structural Break NARDL if my series is I(2)?
No. The ARDL bounds-testing approach requires all variables to be at most I(1). If any variable is I(2), first-differencing to achieve I(1) or switching to a different cointegration framework is necessary.
What critical values should I use for the bounds test with break dummies?
Most applied papers use the standard Pesaran, Shin & Smith (2001) or Narayan (2005) critical values as a benchmark, but note that break dummies alter the distribution of the F-statistic. Some studies use bootstrap critical values for greater accuracy.
How does Structural Break NARDL differ from plain NARDL?
Standard NARDL assumes parameter stability throughout the sample. Structural Break NARDL augments the model with break dummies at one or more estimated dates, allowing the intercept, trend, or slope of the long-run relationship to shift, thereby avoiding biased and inconsistent estimates when regime changes are present.
Is it appropriate to impose the break date rather than estimate it?
Imposing a known break date (e.g., the 2008 financial crisis) is acceptable when there is strong prior justification and avoids pretesting. Endogenous search is preferable when the break timing is uncertain, but it introduces additional uncertainty in the estimated break location and should be reported transparently.
Sources
- Shin, Y., Yu, B., & Greenwood-Nimmo, M. (2014). Modelling asymmetric cointegration and dynamic multipliers in a nonlinear ARDL framework. In W. C. Horrace & R. C. Sickles (Eds.), Festschrift in Honor of Peter Schmidt (pp. 281–314). Springer. DOI: 10.1007/978-1-4899-8008-3_9 ↗
- Pesaran, M. H., Shin, Y., & Smith, R. J. (2001). Bounds testing approaches to the analysis of level relationships. Journal of Applied Econometrics, 16(3), 289–326. DOI: 10.1002/jae.616 ↗
How to cite this page
ScholarGate. (2026, June 3). Structural Break Nonlinear Autoregressive Distributed Lag Model. ScholarGate. https://scholargate.app/en/econometrics/structural-break-nardl
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.
- ARIMA modelEconometrics↔ compare
- Engle-Granger Cointegration TestEconometrics↔ compare
- Nonlinear ARDLEconometrics↔ compare
- Structural Break ARDL Bounds TestEconometrics↔ compare
- Vector Error Correction ModelEconometrics↔ compare
- Zivot-Andrews Structural Break TestEconometrics↔ compare