ERS Point-Optimal Unit-Root Test
Also known as: ERS P-test, Point-Optimal Unit-Root Test, ERS PT statistic, ERS Nokta-Optimal Birim Kök Testi
The Elliott-Rothenberg-Stock (ERS) Point-Optimal test, introduced in their landmark 1996 Econometrica paper, is a near-efficient parametric procedure for testing whether a univariate time series contains a unit root. By first applying GLS detrending at a carefully chosen local-to-unity value and then computing a likelihood-ratio-type statistic, it achieves power close to the Gaussian power envelope—making it one of the most powerful unit-root tests available to applied econometricians.
Key highlights
- Near-Gaussian-optimal power: achieves power close to the theoretical upper bound for invariant unit-root tests.
- Simple to compute: requires only OLS on quasi-differenced data and a closed-form \(P_T\) formula.
- Well-tabulated critical values for both intercept-only and intercept-plus-trend specifications.
- Complements DF-GLS, sharing the same GLS detrending step and enabling joint interpretation.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use the ERS Point-Optimal test when you require maximum power against local alternatives to a unit root in a univariate series, especially with moderate sample sizes (T = 50–500) where ADF power is known to be poor. It assumes linear data-generating processes, homoskedastic or weakly heteroskedastic innovations, and correct specification of the deterministic component (intercept or intercept-plus-trend). It is not suited to structural breaks, seasonal unit roots, or panel settings. When breaks are suspected, combine with Zivot-Andrews or DF-GLS diagnostics; for panels, use IPS or Pesaran panel tests instead.
Strengths & limitations
- Near-Gaussian-optimal power: achieves power close to the theoretical upper bound for invariant unit-root tests.
- Simple to compute: requires only OLS on quasi-differenced data and a closed-form \(P_T\) formula.
- Well-tabulated critical values for both intercept-only and intercept-plus-trend specifications.
- Complements DF-GLS, sharing the same GLS detrending step and enabling joint interpretation.
- Power is optimized only at the chosen local-to-unity point \(\bar{c}\); against alternatives farther from unity power may be suboptimal.
- Assumes correct specification of the deterministic component; misspecification degrades size and power.
- Not robust to structural breaks or non-linear dynamics in the data-generating process.
- Long-run variance estimation introduces an additional nuisance that can distort finite-sample inference if chosen carelessly.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
How does the ERS P-test differ from DF-GLS?
Both use GLS detrending with the same \(\bar{c}\) value, but they test in different ways. DF-GLS applies an ADF-type t-test to the detrended series and rejects for large negative t-values. The ERS \(P_T\) statistic is a likelihood-ratio-type quantity and rejects for small positive values. Elliott et al. (1996) show that \(P_T\) achieves the Gaussian power bound exactly at \(\bar{\alpha}\), whereas DF-GLS is near-efficient but not point-optimal.
How should I choose the lag length for long-run variance estimation?
A common approach is to estimate \(s^2\) using an AR spectral estimator with lag length selected by AIC or BIC on the GLS-detrended residuals, following the same procedure recommended for DF-GLS. Elliott et al. (1996) discuss consistent long-run variance estimation as a prerequisite; software implementations (e.g., Stata's erstest, R's urca) typically offer automatic or user-specified bandwidth/lag selection.
Can the ERS P-test handle structural breaks?
The standard ERS \(P_T\) test does not account for structural breaks; a break under the null can cause the test to over-reject, and a break under the alternative can reduce power. Extensions such as those by Perron (1997) or break-robust variants of GLS detrending exist in the literature, but the baseline ERS procedure should be supplemented with break diagnostics before drawing conclusions when breaks are suspected.
Sources
- 1.Elliott, G., Rothenberg, T. J., & Stock, J. H. (1996). Efficient tests for an autoregressive unit root. Econometrica, 64(4), 813–836.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 2). ERS Point-Optimal Test. ScholarGate. https://scholargate.app/econometrics/ers-point-optimal-test