Robust Augmented Dickey-Fuller Unit Root Test
Also known as: robust ADF test, HAC-corrected ADF, heteroscedasticity-robust unit root test, GLS-detrended ADF
The Robust ADF unit root test extends the classical ADF procedure with improvements that correct for size distortions arising from heteroscedastic or serially correlated errors, and from poor lag-length selection. Drawing on GLS detrending (Elliott, Rothenberg, and Stock 1996) and modified information criteria (Ng and Perron 2001), it delivers reliable size and power in the presence of non-standard error processes common in macroeconomic and financial time series.
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 ADF test when you suspect the error term is conditionally heteroscedastic (e.g., financial returns data with ARCH/GARCH dynamics), when preliminary diagnostics suggest serially correlated or non-normal errors, or when the standard ADF is known to over-reject in your type of data. It is appropriate for moderately long time series (T >= 50) where information criteria can reliably select lag length. Do not use it as a substitute for structural-break-aware tests when regime shifts are the primary concern — in that case prefer Zivot-Andrews or Perron break tests. Also avoid applying it to very short series (T < 40) where the GLS efficiency gain cannot materialise.
Strengths & limitations
- Controls empirical size near the nominal level even under conditional heteroscedasticity and ARCH errors.
- Modified information criteria (MAIC/MSIC) dramatically reduce over-rejection caused by poor lag selection.
- GLS detrending raises power against near-integrated alternatives compared with standard OLS-demeaned ADF.
- Applicable to both constant-only and constant-plus-trend specifications without changing the core procedure.
- Well-established critical value tables and widely implemented in econometric software (EViews, Stata, R urca/tseries).
- Requires sufficient sample size (T >= 50) for GLS detrending and MIC lag selection to work reliably.
- Assumes a single, stable linear autoregressive structure; does not accommodate structural breaks or nonlinear mean-reversion without additional modifications.
- Critical values differ from the standard ADF and must be obtained from simulation tables, which can vary by software implementation.
- Heteroscedasticity-robust wild bootstrap extensions add computational cost and require careful specification of the bootstrap scheme.
Frequently asked
How does the Robust ADF differ from the standard ADF?
The standard ADF uses OLS detrending and conventional AIC/BIC for lag selection, which causes size distortions under heteroscedastic or strongly autocorrelated errors. The robust version uses GLS detrending and a modified information criterion (MAIC/MSIC) that corrects these distortions, and may add HAC or wild bootstrap adjustments for conditional heteroscedasticity.
What are the MZ-alpha and MZ-t statistics?
They are the Ng-Perron (2001) modified versions of the classical Phillips-Perron Z-alpha and Z-t statistics, computed on GLS-detrended data with lag length chosen by MAIC. They have much better size and power properties than their OLS-based counterparts, especially in small and medium samples.
Should I use Robust ADF or DF-GLS?
DF-GLS (Elliott et al. 1996) is the t-statistic on the GLS-detrended lagged level with lags chosen by standard BIC. Ng-Perron's MZ-t improves on it by using MAIC for lag selection and a normalisation that controls size better under MA errors. If ARCH effects are present, a wild-bootstrap version of either is preferable.
What lag selection criterion should I use?
Ng and Perron (2001) recommend their modified AIC (MAIC), which penalises under-fitting more aggressively than standard AIC. In practice, bound the search between 0 and floor(12*(T/100)^0.25) lags and report the chosen order alongside the test statistic.
Can I apply Robust ADF to panel data?
The single-series robust ADF applies to one time series at a time. For panel settings, use panel unit root tests such as Im-Pesaran-Shin (IPS) or the Pesaran CIPS test, which account for cross-sectional dependence and can incorporate heteroscedasticity-robust inference within each cross-section.
Sources
- Ng, S., and Perron, P. (2001). Lag length selection and the construction of unit root tests with good size and power. Econometrica, 69(6), 1519-1554. DOI: 10.1111/1468-0262.00256 ↗
- Elliott, G., Rothenberg, T. J., and Stock, J. H. (1996). Efficient tests for an autoregressive unit root. Econometrica, 64(4), 813-836. DOI: 10.2307/2171846 ↗
How to cite this page
ScholarGate. (2026, June 3). Robust Augmented Dickey-Fuller Unit Root Test. ScholarGate. https://scholargate.app/en/econometrics/robust-adf-unit-root-test
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.
- Augmented Dickey-Fuller unit root testEconometrics↔ compare
- KPSS TestEconometrics↔ compare
- Nonlinear ADF Unit Root TestEconometrics↔ compare
- Panel ADF Unit Root TestEconometrics↔ compare
- Phillips-Perron unit root testEconometrics↔ compare
- Zivot-Andrews Structural Break TestEconometrics↔ compare