Robust Engle-Granger Cointegration Test
Also known as: robust EG cointegration, outlier-robust cointegration test, robust two-step cointegration, robust EG test
The Robust Engle-Granger cointegration test adapts the classic two-step Engle-Granger procedure to withstand outliers, heavy-tailed error distributions, and additive noise that can severely distort standard residual-based cointegration inference. By substituting robust regression and robust unit-root testing for classical OLS and ADF steps, it yields reliable conclusions about long-run equilibrium relationships even when the data contain anomalous observations.
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 the Robust Engle-Granger test when you have two or more I(1) time series and suspect that outliers, data errors, or heavy-tailed innovations may compromise standard cointegration inference. It is particularly appropriate in financial or macroeconomic data prone to crises, policy shocks, or measurement errors. Prefer the Johansen approach when you have three or more variables and need to test for the number of cointegrating vectors simultaneously. The standard Engle-Granger test is adequate when the series are clean and errors are approximately normal; the robust version is warranted when diagnostic checks reveal influential observations or non-normal residuals. Do not use this method when variables are not all integrated of the same order.
Strengths & limitations
- Resistant to the distorting effects of outliers and heavy-tailed errors that can invalidate standard cointegration tests.
- Preserves the intuitive two-step structure of the original Engle-Granger procedure, making results straightforward to communicate.
- Provides more reliable estimates of the cointegrating vector when the data contain anomalous observations.
- Compatible with subsequent robust error-correction modelling for short-run dynamics.
- Applicable in practice using existing robust regression packages combined with ADF tests on residuals.
- Still limited to bivariate or small-system cointegration; the Johansen approach is preferred for systems with multiple cointegrating relationships.
- The choice of robust estimator (MM, LTS, etc.) and tuning constants affects results and requires justification.
- Engle-Granger critical values must be used for the residual unit-root test, and robustified critical values may differ; small-sample performance depends on the specific robust method chosen.
- Cannot determine the number of cointegrating vectors when more than two variables are analysed.
- Requires that all variables are confirmed I(1) beforehand; pre-testing uncertainty compounds inference uncertainty.
Frequently asked
How does the robust version differ from the standard Engle-Granger test?
The standard test uses OLS for the cointegrating regression and a classical ADF test on the residuals. The robust version replaces OLS with an outlier-resistant estimator such as MM-regression or least trimmed squares, and may also apply a robust unit-root test, so that a small number of extreme observations cannot dominate the inference.
Which robust estimator should I choose for the first step?
MM-estimation is a common choice because it combines high breakdown point (resistance to many outliers) with high efficiency under normality. Least trimmed squares is another option. The choice should be guided by the expected contamination level and available software; results should be checked for sensitivity to the estimator.
Should I use standard ADF critical values or Engle-Granger critical values?
Always use Engle-Granger critical values (from MacKinnon's response surfaces or simulation) when testing residuals from a cointegrating regression, because the residuals are estimated quantities. Standard ADF critical values are too lenient and will over-reject the null of no cointegration.
When should I prefer the Johansen test over this approach?
The Johansen test is preferred when you have three or more variables and need to identify multiple cointegrating vectors simultaneously. The Engle-Granger framework, including its robust variant, is most natural for bivariate relationships or when you have a clear theoretical a priori dependent variable.
What do I do after confirming cointegration with this test?
Estimate a robust error-correction model (ECM) that includes the lagged residual from the robust cointegrating regression as the error-correction term. Using robust estimation in the ECM ensures that the short-run dynamics are also protected from outlier distortion.
Sources
- Engle, R. F., & Granger, C. W. J. (1987). Co-integration and error correction: Representation, estimation, and testing. Econometrica, 55(2), 251–276. DOI: 10.2307/1913236 ↗
- Hao, K., & Shaffer, A. (2021). Robust cointegration testing in the presence of outliers. Journal of Statistical Computation and Simulation, 91(10), 2137–2154. link ↗
How to cite this page
ScholarGate. (2026, June 3). Robust Engle-Granger Cointegration Test. ScholarGate. https://scholargate.app/en/econometrics/robust-engle-granger-cointegration
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.
- Engle-Granger Cointegration TestEconometrics↔ compare
- Fourier Engle-Granger cointegrationEconometrics↔ compare
- Robust OLSEconometrics↔ compare
- Robust VECMEconometrics↔ compare
- Structural break Engle-Granger cointegrationEconometrics↔ compare